go-ethereum/p2p/simulations/network.go

1091 lines
29 KiB
Go
Raw Normal View History

// Copyright 2017 The go-ethereum Authors
// This file is part of the go-ethereum library.
//
// The go-ethereum library is free software: you can redistribute it and/or modify
// it under the terms of the GNU Lesser General Public License as published by
// the Free Software Foundation, either version 3 of the License, or
// (at your option) any later version.
//
// The go-ethereum library is distributed in the hope that it will be useful,
// but WITHOUT ANY WARRANTY; without even the implied warranty of
// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
// GNU Lesser General Public License for more details.
//
// You should have received a copy of the GNU Lesser General Public License
// along with the go-ethereum library. If not, see <http://www.gnu.org/licenses/>.
package simulations
import (
"bytes"
"context"
"encoding/json"
"errors"
"fmt"
"math/rand"
"sync"
"time"
"github.com/ethereum/go-ethereum/event"
"github.com/ethereum/go-ethereum/log"
"github.com/ethereum/go-ethereum/p2p"
all: new p2p node representation (#17643) Package p2p/enode provides a generalized representation of p2p nodes which can contain arbitrary information in key/value pairs. It is also the new home for the node database. The "v4" identity scheme is also moved here from p2p/enr to remove the dependency on Ethereum crypto from that package. Record signature handling is changed significantly. The identity scheme registry is removed and acceptable schemes must be passed to any method that needs identity. This means records must now be validated explicitly after decoding. The enode API is designed to make signature handling easy and safe: most APIs around the codebase work with enode.Node, which is a wrapper around a valid record. Going from enr.Record to enode.Node requires a valid signature. * p2p/discover: port to p2p/enode This ports the discovery code to the new node representation in p2p/enode. The wire protocol is unchanged, this can be considered a refactoring change. The Kademlia table can now deal with nodes using an arbitrary identity scheme. This requires a few incompatible API changes: - Table.Lookup is not available anymore. It used to take a public key as argument because v4 protocol requires one. Its replacement is LookupRandom. - Table.Resolve takes *enode.Node instead of NodeID. This is also for v4 protocol compatibility because nodes cannot be looked up by ID alone. - Types Node and NodeID are gone. Further commits in the series will be fixes all over the the codebase to deal with those removals. * p2p: port to p2p/enode and discovery changes This adapts package p2p to the changes in p2p/discover. All uses of discover.Node and discover.NodeID are replaced by their equivalents from p2p/enode. New API is added to retrieve the enode.Node instance of a peer. The behavior of Server.Self with discovery disabled is improved. It now tries much harder to report a working IP address, falling back to 127.0.0.1 if no suitable address can be determined through other means. These changes were needed for tests of other packages later in the series. * p2p/simulations, p2p/testing: port to p2p/enode No surprises here, mostly replacements of discover.Node, discover.NodeID with their new equivalents. The 'interesting' API changes are: - testing.ProtocolSession tracks complete nodes, not just their IDs. - adapters.NodeConfig has a new method to create a complete node. These changes were needed to make swarm tests work. Note that the NodeID change makes the code incompatible with old simulation snapshots. * whisper/whisperv5, whisper/whisperv6: port to p2p/enode This port was easy because whisper uses []byte for node IDs and URL strings in the API. * eth: port to p2p/enode Again, easy to port because eth uses strings for node IDs and doesn't care about node information in any way. * les: port to p2p/enode Apart from replacing discover.NodeID with enode.ID, most changes are in the server pool code. It now deals with complete nodes instead of (Pubkey, IP, Port) triples. The database format is unchanged for now, but we should probably change it to use the node database later. * node: port to p2p/enode This change simply replaces discover.Node and discover.NodeID with their new equivalents. * swarm/network: port to p2p/enode Swarm has its own node address representation, BzzAddr, containing both an overlay address (the hash of a secp256k1 public key) and an underlay address (enode:// URL). There are no changes to the BzzAddr format in this commit, but certain operations such as creating a BzzAddr from a node ID are now impossible because node IDs aren't public keys anymore. Most swarm-related changes in the series remove uses of NewAddrFromNodeID, replacing it with NewAddr which takes a complete node as argument. ToOverlayAddr is removed because we can just use the node ID directly.
2018-09-24 17:59:00 -05:00
"github.com/ethereum/go-ethereum/p2p/enode"
"github.com/ethereum/go-ethereum/p2p/simulations/adapters"
)
2018-06-20 07:06:27 -05:00
var DialBanTimeout = 200 * time.Millisecond
// NetworkConfig defines configuration options for starting a Network
type NetworkConfig struct {
ID string `json:"id"`
DefaultService string `json:"default_service,omitempty"`
}
// Network models a p2p simulation network which consists of a collection of
// simulated nodes and the connections which exist between them.
//
// The Network has a single NodeAdapter which is responsible for actually
// starting nodes and connecting them together.
//
// The Network emits events when nodes are started and stopped, when they are
// connected and disconnected, and also when messages are sent between nodes.
type Network struct {
NetworkConfig
Nodes []*Node `json:"nodes"`
all: new p2p node representation (#17643) Package p2p/enode provides a generalized representation of p2p nodes which can contain arbitrary information in key/value pairs. It is also the new home for the node database. The "v4" identity scheme is also moved here from p2p/enr to remove the dependency on Ethereum crypto from that package. Record signature handling is changed significantly. The identity scheme registry is removed and acceptable schemes must be passed to any method that needs identity. This means records must now be validated explicitly after decoding. The enode API is designed to make signature handling easy and safe: most APIs around the codebase work with enode.Node, which is a wrapper around a valid record. Going from enr.Record to enode.Node requires a valid signature. * p2p/discover: port to p2p/enode This ports the discovery code to the new node representation in p2p/enode. The wire protocol is unchanged, this can be considered a refactoring change. The Kademlia table can now deal with nodes using an arbitrary identity scheme. This requires a few incompatible API changes: - Table.Lookup is not available anymore. It used to take a public key as argument because v4 protocol requires one. Its replacement is LookupRandom. - Table.Resolve takes *enode.Node instead of NodeID. This is also for v4 protocol compatibility because nodes cannot be looked up by ID alone. - Types Node and NodeID are gone. Further commits in the series will be fixes all over the the codebase to deal with those removals. * p2p: port to p2p/enode and discovery changes This adapts package p2p to the changes in p2p/discover. All uses of discover.Node and discover.NodeID are replaced by their equivalents from p2p/enode. New API is added to retrieve the enode.Node instance of a peer. The behavior of Server.Self with discovery disabled is improved. It now tries much harder to report a working IP address, falling back to 127.0.0.1 if no suitable address can be determined through other means. These changes were needed for tests of other packages later in the series. * p2p/simulations, p2p/testing: port to p2p/enode No surprises here, mostly replacements of discover.Node, discover.NodeID with their new equivalents. The 'interesting' API changes are: - testing.ProtocolSession tracks complete nodes, not just their IDs. - adapters.NodeConfig has a new method to create a complete node. These changes were needed to make swarm tests work. Note that the NodeID change makes the code incompatible with old simulation snapshots. * whisper/whisperv5, whisper/whisperv6: port to p2p/enode This port was easy because whisper uses []byte for node IDs and URL strings in the API. * eth: port to p2p/enode Again, easy to port because eth uses strings for node IDs and doesn't care about node information in any way. * les: port to p2p/enode Apart from replacing discover.NodeID with enode.ID, most changes are in the server pool code. It now deals with complete nodes instead of (Pubkey, IP, Port) triples. The database format is unchanged for now, but we should probably change it to use the node database later. * node: port to p2p/enode This change simply replaces discover.Node and discover.NodeID with their new equivalents. * swarm/network: port to p2p/enode Swarm has its own node address representation, BzzAddr, containing both an overlay address (the hash of a secp256k1 public key) and an underlay address (enode:// URL). There are no changes to the BzzAddr format in this commit, but certain operations such as creating a BzzAddr from a node ID are now impossible because node IDs aren't public keys anymore. Most swarm-related changes in the series remove uses of NewAddrFromNodeID, replacing it with NewAddr which takes a complete node as argument. ToOverlayAddr is removed because we can just use the node ID directly.
2018-09-24 17:59:00 -05:00
nodeMap map[enode.ID]int
// Maps a node property string to node indexes of all nodes that hold this property
propertyMap map[string][]int
Conns []*Conn `json:"conns"`
connMap map[string]int
nodeAdapter adapters.NodeAdapter
events event.Feed
lock sync.RWMutex
quitc chan struct{}
}
// NewNetwork returns a Network which uses the given NodeAdapter and NetworkConfig
func NewNetwork(nodeAdapter adapters.NodeAdapter, conf *NetworkConfig) *Network {
return &Network{
NetworkConfig: *conf,
nodeAdapter: nodeAdapter,
all: new p2p node representation (#17643) Package p2p/enode provides a generalized representation of p2p nodes which can contain arbitrary information in key/value pairs. It is also the new home for the node database. The "v4" identity scheme is also moved here from p2p/enr to remove the dependency on Ethereum crypto from that package. Record signature handling is changed significantly. The identity scheme registry is removed and acceptable schemes must be passed to any method that needs identity. This means records must now be validated explicitly after decoding. The enode API is designed to make signature handling easy and safe: most APIs around the codebase work with enode.Node, which is a wrapper around a valid record. Going from enr.Record to enode.Node requires a valid signature. * p2p/discover: port to p2p/enode This ports the discovery code to the new node representation in p2p/enode. The wire protocol is unchanged, this can be considered a refactoring change. The Kademlia table can now deal with nodes using an arbitrary identity scheme. This requires a few incompatible API changes: - Table.Lookup is not available anymore. It used to take a public key as argument because v4 protocol requires one. Its replacement is LookupRandom. - Table.Resolve takes *enode.Node instead of NodeID. This is also for v4 protocol compatibility because nodes cannot be looked up by ID alone. - Types Node and NodeID are gone. Further commits in the series will be fixes all over the the codebase to deal with those removals. * p2p: port to p2p/enode and discovery changes This adapts package p2p to the changes in p2p/discover. All uses of discover.Node and discover.NodeID are replaced by their equivalents from p2p/enode. New API is added to retrieve the enode.Node instance of a peer. The behavior of Server.Self with discovery disabled is improved. It now tries much harder to report a working IP address, falling back to 127.0.0.1 if no suitable address can be determined through other means. These changes were needed for tests of other packages later in the series. * p2p/simulations, p2p/testing: port to p2p/enode No surprises here, mostly replacements of discover.Node, discover.NodeID with their new equivalents. The 'interesting' API changes are: - testing.ProtocolSession tracks complete nodes, not just their IDs. - adapters.NodeConfig has a new method to create a complete node. These changes were needed to make swarm tests work. Note that the NodeID change makes the code incompatible with old simulation snapshots. * whisper/whisperv5, whisper/whisperv6: port to p2p/enode This port was easy because whisper uses []byte for node IDs and URL strings in the API. * eth: port to p2p/enode Again, easy to port because eth uses strings for node IDs and doesn't care about node information in any way. * les: port to p2p/enode Apart from replacing discover.NodeID with enode.ID, most changes are in the server pool code. It now deals with complete nodes instead of (Pubkey, IP, Port) triples. The database format is unchanged for now, but we should probably change it to use the node database later. * node: port to p2p/enode This change simply replaces discover.Node and discover.NodeID with their new equivalents. * swarm/network: port to p2p/enode Swarm has its own node address representation, BzzAddr, containing both an overlay address (the hash of a secp256k1 public key) and an underlay address (enode:// URL). There are no changes to the BzzAddr format in this commit, but certain operations such as creating a BzzAddr from a node ID are now impossible because node IDs aren't public keys anymore. Most swarm-related changes in the series remove uses of NewAddrFromNodeID, replacing it with NewAddr which takes a complete node as argument. ToOverlayAddr is removed because we can just use the node ID directly.
2018-09-24 17:59:00 -05:00
nodeMap: make(map[enode.ID]int),
propertyMap: make(map[string][]int),
connMap: make(map[string]int),
quitc: make(chan struct{}),
}
}
// Events returns the output event feed of the Network.
2018-05-08 06:08:43 -05:00
func (net *Network) Events() *event.Feed {
return &net.events
}
// NewNodeWithConfig adds a new node to the network with the given config,
// returning an error if a node with the same ID or name already exists
2018-05-08 06:08:43 -05:00
func (net *Network) NewNodeWithConfig(conf *adapters.NodeConfig) (*Node, error) {
net.lock.Lock()
defer net.lock.Unlock()
if conf.Reachable == nil {
all: new p2p node representation (#17643) Package p2p/enode provides a generalized representation of p2p nodes which can contain arbitrary information in key/value pairs. It is also the new home for the node database. The "v4" identity scheme is also moved here from p2p/enr to remove the dependency on Ethereum crypto from that package. Record signature handling is changed significantly. The identity scheme registry is removed and acceptable schemes must be passed to any method that needs identity. This means records must now be validated explicitly after decoding. The enode API is designed to make signature handling easy and safe: most APIs around the codebase work with enode.Node, which is a wrapper around a valid record. Going from enr.Record to enode.Node requires a valid signature. * p2p/discover: port to p2p/enode This ports the discovery code to the new node representation in p2p/enode. The wire protocol is unchanged, this can be considered a refactoring change. The Kademlia table can now deal with nodes using an arbitrary identity scheme. This requires a few incompatible API changes: - Table.Lookup is not available anymore. It used to take a public key as argument because v4 protocol requires one. Its replacement is LookupRandom. - Table.Resolve takes *enode.Node instead of NodeID. This is also for v4 protocol compatibility because nodes cannot be looked up by ID alone. - Types Node and NodeID are gone. Further commits in the series will be fixes all over the the codebase to deal with those removals. * p2p: port to p2p/enode and discovery changes This adapts package p2p to the changes in p2p/discover. All uses of discover.Node and discover.NodeID are replaced by their equivalents from p2p/enode. New API is added to retrieve the enode.Node instance of a peer. The behavior of Server.Self with discovery disabled is improved. It now tries much harder to report a working IP address, falling back to 127.0.0.1 if no suitable address can be determined through other means. These changes were needed for tests of other packages later in the series. * p2p/simulations, p2p/testing: port to p2p/enode No surprises here, mostly replacements of discover.Node, discover.NodeID with their new equivalents. The 'interesting' API changes are: - testing.ProtocolSession tracks complete nodes, not just their IDs. - adapters.NodeConfig has a new method to create a complete node. These changes were needed to make swarm tests work. Note that the NodeID change makes the code incompatible with old simulation snapshots. * whisper/whisperv5, whisper/whisperv6: port to p2p/enode This port was easy because whisper uses []byte for node IDs and URL strings in the API. * eth: port to p2p/enode Again, easy to port because eth uses strings for node IDs and doesn't care about node information in any way. * les: port to p2p/enode Apart from replacing discover.NodeID with enode.ID, most changes are in the server pool code. It now deals with complete nodes instead of (Pubkey, IP, Port) triples. The database format is unchanged for now, but we should probably change it to use the node database later. * node: port to p2p/enode This change simply replaces discover.Node and discover.NodeID with their new equivalents. * swarm/network: port to p2p/enode Swarm has its own node address representation, BzzAddr, containing both an overlay address (the hash of a secp256k1 public key) and an underlay address (enode:// URL). There are no changes to the BzzAddr format in this commit, but certain operations such as creating a BzzAddr from a node ID are now impossible because node IDs aren't public keys anymore. Most swarm-related changes in the series remove uses of NewAddrFromNodeID, replacing it with NewAddr which takes a complete node as argument. ToOverlayAddr is removed because we can just use the node ID directly.
2018-09-24 17:59:00 -05:00
conf.Reachable = func(otherID enode.ID) bool {
2018-05-08 06:08:43 -05:00
_, err := net.InitConn(conf.ID, otherID)
2018-06-20 07:06:27 -05:00
if err != nil && bytes.Compare(conf.ID.Bytes(), otherID.Bytes()) < 0 {
return false
}
return true
}
}
// check the node doesn't already exist
2018-06-20 07:06:27 -05:00
if node := net.getNode(conf.ID); node != nil {
return nil, fmt.Errorf("node with ID %q already exists", conf.ID)
}
2018-05-08 06:08:43 -05:00
if node := net.getNodeByName(conf.Name); node != nil {
return nil, fmt.Errorf("node with name %q already exists", conf.Name)
}
// if no services are configured, use the default service
if len(conf.Lifecycles) == 0 {
conf.Lifecycles = []string{net.DefaultService}
}
// use the NodeAdapter to create the node
2018-05-08 06:08:43 -05:00
adapterNode, err := net.nodeAdapter.NewNode(conf)
if err != nil {
return nil, err
}
node := newNode(adapterNode, conf, false)
log.Trace("Node created", "id", conf.ID)
nodeIndex := len(net.Nodes)
net.nodeMap[conf.ID] = nodeIndex
2018-05-08 06:08:43 -05:00
net.Nodes = append(net.Nodes, node)
// Register any node properties with the network-level propertyMap
for _, property := range conf.Properties {
net.propertyMap[property] = append(net.propertyMap[property], nodeIndex)
}
// emit a "control" event
2018-05-08 06:08:43 -05:00
net.events.Send(ControlEvent(node))
return node, nil
}
// Config returns the network configuration
2018-05-08 06:08:43 -05:00
func (net *Network) Config() *NetworkConfig {
return &net.NetworkConfig
}
// StartAll starts all nodes in the network
2018-05-08 06:08:43 -05:00
func (net *Network) StartAll() error {
for _, node := range net.Nodes {
p2p, swarm: fix node up races by granular locking (#18976) * swarm/network: DRY out repeated giga comment I not necessarily agree with the way we wait for event propagation. But I truly disagree with having duplicated giga comments. * p2p/simulations: encapsulate Node.Up field so we avoid data races The Node.Up field was accessed concurrently without "proper" locking. There was a lock on Network and that was used sometimes to access the field. Other times the locking was missed and we had a data race. For example: https://github.com/ethereum/go-ethereum/pull/18464 The case above was solved, but there were still intermittent/hard to reproduce races. So let's solve the issue permanently. resolves: ethersphere/go-ethereum#1146 * p2p/simulations: fix unmarshal of simulations.Node Making Node.Up field private in 13292ee897e345045fbfab3bda23a77589a271c1 broke TestHTTPNetwork and TestHTTPSnapshot. Because the default UnmarshalJSON does not handle unexported fields. Important: The fix is partial and not proper to my taste. But I cut scope as I think the fix may require a change to the current serialization format. New ticket: https://github.com/ethersphere/go-ethereum/issues/1177 * p2p/simulations: Add a sanity test case for Node.Config UnmarshalJSON * p2p/simulations: revert back to defer Unlock() pattern for Network It's a good patten to call `defer Unlock()` right after `Lock()` so (new) error cases won't miss to unlock. Let's get back to that pattern. The patten was abandoned in 85a79b3ad3c5863f8612d25c246bcfad339f36b7, while fixing a data race. That data race does not exist anymore, since the Node.Up field got hidden behind its own lock. * p2p/simulations: consistent naming for test providers Node.UnmarshalJSON * p2p/simulations: remove JSON annotation from private fields of Node As unexported fields are not serialized. * p2p/simulations: fix deadlock in Network.GetRandomDownNode() Problem: GetRandomDownNode() locks -> getDownNodeIDs() -> GetNodes() tries to lock -> deadlock On Network type, unexported functions must assume that `net.lock` is already acquired and should not call exported functions which might try to lock again. * p2p/simulations: ensure method conformity for Network Connect* methods were moved to p2p/simulations.Network from swarm/network/simulation. However these new methods did not follow the pattern of Network methods, i.e., all exported method locks the whole Network either for read or write. * p2p/simulations: fix deadlock during network shutdown `TestDiscoveryPersistenceSimulationSimAdapter` often got into deadlock. The execution was stuck on two locks, i.e, `Kademlia.lock` and `p2p/simulations.Network.lock`. Usually the test got stuck once in each 20 executions with high confidence. `Kademlia` was stuck in `Kademlia.EachAddr()` and `Network` in `Network.Stop()`. Solution: in `Network.Stop()` `net.lock` must be released before calling `node.Stop()` as stopping a node (somehow - I did not find the exact code path) causes `Network.InitConn()` to be called from `Kademlia.SuggestPeer()` and that blocks on `net.lock`. Related ticket: https://github.com/ethersphere/go-ethereum/issues/1223 * swarm/state: simplify if statement in DBStore.Put() * p2p/simulations: remove faulty godoc from private function The comment started with the wrong method name. The method is simple and self explanatory. Also, it's private. => Let's just remove the comment.
2019-02-18 00:38:14 -06:00
if node.Up() {
continue
}
2018-05-08 06:08:43 -05:00
if err := net.Start(node.ID()); err != nil {
return err
}
}
return nil
}
// StopAll stops all nodes in the network
2018-05-08 06:08:43 -05:00
func (net *Network) StopAll() error {
for _, node := range net.Nodes {
p2p, swarm: fix node up races by granular locking (#18976) * swarm/network: DRY out repeated giga comment I not necessarily agree with the way we wait for event propagation. But I truly disagree with having duplicated giga comments. * p2p/simulations: encapsulate Node.Up field so we avoid data races The Node.Up field was accessed concurrently without "proper" locking. There was a lock on Network and that was used sometimes to access the field. Other times the locking was missed and we had a data race. For example: https://github.com/ethereum/go-ethereum/pull/18464 The case above was solved, but there were still intermittent/hard to reproduce races. So let's solve the issue permanently. resolves: ethersphere/go-ethereum#1146 * p2p/simulations: fix unmarshal of simulations.Node Making Node.Up field private in 13292ee897e345045fbfab3bda23a77589a271c1 broke TestHTTPNetwork and TestHTTPSnapshot. Because the default UnmarshalJSON does not handle unexported fields. Important: The fix is partial and not proper to my taste. But I cut scope as I think the fix may require a change to the current serialization format. New ticket: https://github.com/ethersphere/go-ethereum/issues/1177 * p2p/simulations: Add a sanity test case for Node.Config UnmarshalJSON * p2p/simulations: revert back to defer Unlock() pattern for Network It's a good patten to call `defer Unlock()` right after `Lock()` so (new) error cases won't miss to unlock. Let's get back to that pattern. The patten was abandoned in 85a79b3ad3c5863f8612d25c246bcfad339f36b7, while fixing a data race. That data race does not exist anymore, since the Node.Up field got hidden behind its own lock. * p2p/simulations: consistent naming for test providers Node.UnmarshalJSON * p2p/simulations: remove JSON annotation from private fields of Node As unexported fields are not serialized. * p2p/simulations: fix deadlock in Network.GetRandomDownNode() Problem: GetRandomDownNode() locks -> getDownNodeIDs() -> GetNodes() tries to lock -> deadlock On Network type, unexported functions must assume that `net.lock` is already acquired and should not call exported functions which might try to lock again. * p2p/simulations: ensure method conformity for Network Connect* methods were moved to p2p/simulations.Network from swarm/network/simulation. However these new methods did not follow the pattern of Network methods, i.e., all exported method locks the whole Network either for read or write. * p2p/simulations: fix deadlock during network shutdown `TestDiscoveryPersistenceSimulationSimAdapter` often got into deadlock. The execution was stuck on two locks, i.e, `Kademlia.lock` and `p2p/simulations.Network.lock`. Usually the test got stuck once in each 20 executions with high confidence. `Kademlia` was stuck in `Kademlia.EachAddr()` and `Network` in `Network.Stop()`. Solution: in `Network.Stop()` `net.lock` must be released before calling `node.Stop()` as stopping a node (somehow - I did not find the exact code path) causes `Network.InitConn()` to be called from `Kademlia.SuggestPeer()` and that blocks on `net.lock`. Related ticket: https://github.com/ethersphere/go-ethereum/issues/1223 * swarm/state: simplify if statement in DBStore.Put() * p2p/simulations: remove faulty godoc from private function The comment started with the wrong method name. The method is simple and self explanatory. Also, it's private. => Let's just remove the comment.
2019-02-18 00:38:14 -06:00
if !node.Up() {
continue
}
2018-05-08 06:08:43 -05:00
if err := net.Stop(node.ID()); err != nil {
return err
}
}
return nil
}
// Start starts the node with the given ID
all: new p2p node representation (#17643) Package p2p/enode provides a generalized representation of p2p nodes which can contain arbitrary information in key/value pairs. It is also the new home for the node database. The "v4" identity scheme is also moved here from p2p/enr to remove the dependency on Ethereum crypto from that package. Record signature handling is changed significantly. The identity scheme registry is removed and acceptable schemes must be passed to any method that needs identity. This means records must now be validated explicitly after decoding. The enode API is designed to make signature handling easy and safe: most APIs around the codebase work with enode.Node, which is a wrapper around a valid record. Going from enr.Record to enode.Node requires a valid signature. * p2p/discover: port to p2p/enode This ports the discovery code to the new node representation in p2p/enode. The wire protocol is unchanged, this can be considered a refactoring change. The Kademlia table can now deal with nodes using an arbitrary identity scheme. This requires a few incompatible API changes: - Table.Lookup is not available anymore. It used to take a public key as argument because v4 protocol requires one. Its replacement is LookupRandom. - Table.Resolve takes *enode.Node instead of NodeID. This is also for v4 protocol compatibility because nodes cannot be looked up by ID alone. - Types Node and NodeID are gone. Further commits in the series will be fixes all over the the codebase to deal with those removals. * p2p: port to p2p/enode and discovery changes This adapts package p2p to the changes in p2p/discover. All uses of discover.Node and discover.NodeID are replaced by their equivalents from p2p/enode. New API is added to retrieve the enode.Node instance of a peer. The behavior of Server.Self with discovery disabled is improved. It now tries much harder to report a working IP address, falling back to 127.0.0.1 if no suitable address can be determined through other means. These changes were needed for tests of other packages later in the series. * p2p/simulations, p2p/testing: port to p2p/enode No surprises here, mostly replacements of discover.Node, discover.NodeID with their new equivalents. The 'interesting' API changes are: - testing.ProtocolSession tracks complete nodes, not just their IDs. - adapters.NodeConfig has a new method to create a complete node. These changes were needed to make swarm tests work. Note that the NodeID change makes the code incompatible with old simulation snapshots. * whisper/whisperv5, whisper/whisperv6: port to p2p/enode This port was easy because whisper uses []byte for node IDs and URL strings in the API. * eth: port to p2p/enode Again, easy to port because eth uses strings for node IDs and doesn't care about node information in any way. * les: port to p2p/enode Apart from replacing discover.NodeID with enode.ID, most changes are in the server pool code. It now deals with complete nodes instead of (Pubkey, IP, Port) triples. The database format is unchanged for now, but we should probably change it to use the node database later. * node: port to p2p/enode This change simply replaces discover.Node and discover.NodeID with their new equivalents. * swarm/network: port to p2p/enode Swarm has its own node address representation, BzzAddr, containing both an overlay address (the hash of a secp256k1 public key) and an underlay address (enode:// URL). There are no changes to the BzzAddr format in this commit, but certain operations such as creating a BzzAddr from a node ID are now impossible because node IDs aren't public keys anymore. Most swarm-related changes in the series remove uses of NewAddrFromNodeID, replacing it with NewAddr which takes a complete node as argument. ToOverlayAddr is removed because we can just use the node ID directly.
2018-09-24 17:59:00 -05:00
func (net *Network) Start(id enode.ID) error {
2018-05-08 06:08:43 -05:00
return net.startWithSnapshots(id, nil)
}
// startWithSnapshots starts the node with the given ID using the give
// snapshots
all: new p2p node representation (#17643) Package p2p/enode provides a generalized representation of p2p nodes which can contain arbitrary information in key/value pairs. It is also the new home for the node database. The "v4" identity scheme is also moved here from p2p/enr to remove the dependency on Ethereum crypto from that package. Record signature handling is changed significantly. The identity scheme registry is removed and acceptable schemes must be passed to any method that needs identity. This means records must now be validated explicitly after decoding. The enode API is designed to make signature handling easy and safe: most APIs around the codebase work with enode.Node, which is a wrapper around a valid record. Going from enr.Record to enode.Node requires a valid signature. * p2p/discover: port to p2p/enode This ports the discovery code to the new node representation in p2p/enode. The wire protocol is unchanged, this can be considered a refactoring change. The Kademlia table can now deal with nodes using an arbitrary identity scheme. This requires a few incompatible API changes: - Table.Lookup is not available anymore. It used to take a public key as argument because v4 protocol requires one. Its replacement is LookupRandom. - Table.Resolve takes *enode.Node instead of NodeID. This is also for v4 protocol compatibility because nodes cannot be looked up by ID alone. - Types Node and NodeID are gone. Further commits in the series will be fixes all over the the codebase to deal with those removals. * p2p: port to p2p/enode and discovery changes This adapts package p2p to the changes in p2p/discover. All uses of discover.Node and discover.NodeID are replaced by their equivalents from p2p/enode. New API is added to retrieve the enode.Node instance of a peer. The behavior of Server.Self with discovery disabled is improved. It now tries much harder to report a working IP address, falling back to 127.0.0.1 if no suitable address can be determined through other means. These changes were needed for tests of other packages later in the series. * p2p/simulations, p2p/testing: port to p2p/enode No surprises here, mostly replacements of discover.Node, discover.NodeID with their new equivalents. The 'interesting' API changes are: - testing.ProtocolSession tracks complete nodes, not just their IDs. - adapters.NodeConfig has a new method to create a complete node. These changes were needed to make swarm tests work. Note that the NodeID change makes the code incompatible with old simulation snapshots. * whisper/whisperv5, whisper/whisperv6: port to p2p/enode This port was easy because whisper uses []byte for node IDs and URL strings in the API. * eth: port to p2p/enode Again, easy to port because eth uses strings for node IDs and doesn't care about node information in any way. * les: port to p2p/enode Apart from replacing discover.NodeID with enode.ID, most changes are in the server pool code. It now deals with complete nodes instead of (Pubkey, IP, Port) triples. The database format is unchanged for now, but we should probably change it to use the node database later. * node: port to p2p/enode This change simply replaces discover.Node and discover.NodeID with their new equivalents. * swarm/network: port to p2p/enode Swarm has its own node address representation, BzzAddr, containing both an overlay address (the hash of a secp256k1 public key) and an underlay address (enode:// URL). There are no changes to the BzzAddr format in this commit, but certain operations such as creating a BzzAddr from a node ID are now impossible because node IDs aren't public keys anymore. Most swarm-related changes in the series remove uses of NewAddrFromNodeID, replacing it with NewAddr which takes a complete node as argument. ToOverlayAddr is removed because we can just use the node ID directly.
2018-09-24 17:59:00 -05:00
func (net *Network) startWithSnapshots(id enode.ID, snapshots map[string][]byte) error {
2018-06-20 07:06:27 -05:00
net.lock.Lock()
p2p, swarm: fix node up races by granular locking (#18976) * swarm/network: DRY out repeated giga comment I not necessarily agree with the way we wait for event propagation. But I truly disagree with having duplicated giga comments. * p2p/simulations: encapsulate Node.Up field so we avoid data races The Node.Up field was accessed concurrently without "proper" locking. There was a lock on Network and that was used sometimes to access the field. Other times the locking was missed and we had a data race. For example: https://github.com/ethereum/go-ethereum/pull/18464 The case above was solved, but there were still intermittent/hard to reproduce races. So let's solve the issue permanently. resolves: ethersphere/go-ethereum#1146 * p2p/simulations: fix unmarshal of simulations.Node Making Node.Up field private in 13292ee897e345045fbfab3bda23a77589a271c1 broke TestHTTPNetwork and TestHTTPSnapshot. Because the default UnmarshalJSON does not handle unexported fields. Important: The fix is partial and not proper to my taste. But I cut scope as I think the fix may require a change to the current serialization format. New ticket: https://github.com/ethersphere/go-ethereum/issues/1177 * p2p/simulations: Add a sanity test case for Node.Config UnmarshalJSON * p2p/simulations: revert back to defer Unlock() pattern for Network It's a good patten to call `defer Unlock()` right after `Lock()` so (new) error cases won't miss to unlock. Let's get back to that pattern. The patten was abandoned in 85a79b3ad3c5863f8612d25c246bcfad339f36b7, while fixing a data race. That data race does not exist anymore, since the Node.Up field got hidden behind its own lock. * p2p/simulations: consistent naming for test providers Node.UnmarshalJSON * p2p/simulations: remove JSON annotation from private fields of Node As unexported fields are not serialized. * p2p/simulations: fix deadlock in Network.GetRandomDownNode() Problem: GetRandomDownNode() locks -> getDownNodeIDs() -> GetNodes() tries to lock -> deadlock On Network type, unexported functions must assume that `net.lock` is already acquired and should not call exported functions which might try to lock again. * p2p/simulations: ensure method conformity for Network Connect* methods were moved to p2p/simulations.Network from swarm/network/simulation. However these new methods did not follow the pattern of Network methods, i.e., all exported method locks the whole Network either for read or write. * p2p/simulations: fix deadlock during network shutdown `TestDiscoveryPersistenceSimulationSimAdapter` often got into deadlock. The execution was stuck on two locks, i.e, `Kademlia.lock` and `p2p/simulations.Network.lock`. Usually the test got stuck once in each 20 executions with high confidence. `Kademlia` was stuck in `Kademlia.EachAddr()` and `Network` in `Network.Stop()`. Solution: in `Network.Stop()` `net.lock` must be released before calling `node.Stop()` as stopping a node (somehow - I did not find the exact code path) causes `Network.InitConn()` to be called from `Kademlia.SuggestPeer()` and that blocks on `net.lock`. Related ticket: https://github.com/ethersphere/go-ethereum/issues/1223 * swarm/state: simplify if statement in DBStore.Put() * p2p/simulations: remove faulty godoc from private function The comment started with the wrong method name. The method is simple and self explanatory. Also, it's private. => Let's just remove the comment.
2019-02-18 00:38:14 -06:00
defer net.lock.Unlock()
2018-06-20 07:06:27 -05:00
node := net.getNode(id)
if node == nil {
return fmt.Errorf("node %v does not exist", id)
}
p2p, swarm: fix node up races by granular locking (#18976) * swarm/network: DRY out repeated giga comment I not necessarily agree with the way we wait for event propagation. But I truly disagree with having duplicated giga comments. * p2p/simulations: encapsulate Node.Up field so we avoid data races The Node.Up field was accessed concurrently without "proper" locking. There was a lock on Network and that was used sometimes to access the field. Other times the locking was missed and we had a data race. For example: https://github.com/ethereum/go-ethereum/pull/18464 The case above was solved, but there were still intermittent/hard to reproduce races. So let's solve the issue permanently. resolves: ethersphere/go-ethereum#1146 * p2p/simulations: fix unmarshal of simulations.Node Making Node.Up field private in 13292ee897e345045fbfab3bda23a77589a271c1 broke TestHTTPNetwork and TestHTTPSnapshot. Because the default UnmarshalJSON does not handle unexported fields. Important: The fix is partial and not proper to my taste. But I cut scope as I think the fix may require a change to the current serialization format. New ticket: https://github.com/ethersphere/go-ethereum/issues/1177 * p2p/simulations: Add a sanity test case for Node.Config UnmarshalJSON * p2p/simulations: revert back to defer Unlock() pattern for Network It's a good patten to call `defer Unlock()` right after `Lock()` so (new) error cases won't miss to unlock. Let's get back to that pattern. The patten was abandoned in 85a79b3ad3c5863f8612d25c246bcfad339f36b7, while fixing a data race. That data race does not exist anymore, since the Node.Up field got hidden behind its own lock. * p2p/simulations: consistent naming for test providers Node.UnmarshalJSON * p2p/simulations: remove JSON annotation from private fields of Node As unexported fields are not serialized. * p2p/simulations: fix deadlock in Network.GetRandomDownNode() Problem: GetRandomDownNode() locks -> getDownNodeIDs() -> GetNodes() tries to lock -> deadlock On Network type, unexported functions must assume that `net.lock` is already acquired and should not call exported functions which might try to lock again. * p2p/simulations: ensure method conformity for Network Connect* methods were moved to p2p/simulations.Network from swarm/network/simulation. However these new methods did not follow the pattern of Network methods, i.e., all exported method locks the whole Network either for read or write. * p2p/simulations: fix deadlock during network shutdown `TestDiscoveryPersistenceSimulationSimAdapter` often got into deadlock. The execution was stuck on two locks, i.e, `Kademlia.lock` and `p2p/simulations.Network.lock`. Usually the test got stuck once in each 20 executions with high confidence. `Kademlia` was stuck in `Kademlia.EachAddr()` and `Network` in `Network.Stop()`. Solution: in `Network.Stop()` `net.lock` must be released before calling `node.Stop()` as stopping a node (somehow - I did not find the exact code path) causes `Network.InitConn()` to be called from `Kademlia.SuggestPeer()` and that blocks on `net.lock`. Related ticket: https://github.com/ethersphere/go-ethereum/issues/1223 * swarm/state: simplify if statement in DBStore.Put() * p2p/simulations: remove faulty godoc from private function The comment started with the wrong method name. The method is simple and self explanatory. Also, it's private. => Let's just remove the comment.
2019-02-18 00:38:14 -06:00
if node.Up() {
return fmt.Errorf("node %v already up", id)
}
log.Trace("Starting node", "id", id, "adapter", net.nodeAdapter.Name())
if err := node.Start(snapshots); err != nil {
log.Warn("Node startup failed", "id", id, "err", err)
return err
}
p2p, swarm: fix node up races by granular locking (#18976) * swarm/network: DRY out repeated giga comment I not necessarily agree with the way we wait for event propagation. But I truly disagree with having duplicated giga comments. * p2p/simulations: encapsulate Node.Up field so we avoid data races The Node.Up field was accessed concurrently without "proper" locking. There was a lock on Network and that was used sometimes to access the field. Other times the locking was missed and we had a data race. For example: https://github.com/ethereum/go-ethereum/pull/18464 The case above was solved, but there were still intermittent/hard to reproduce races. So let's solve the issue permanently. resolves: ethersphere/go-ethereum#1146 * p2p/simulations: fix unmarshal of simulations.Node Making Node.Up field private in 13292ee897e345045fbfab3bda23a77589a271c1 broke TestHTTPNetwork and TestHTTPSnapshot. Because the default UnmarshalJSON does not handle unexported fields. Important: The fix is partial and not proper to my taste. But I cut scope as I think the fix may require a change to the current serialization format. New ticket: https://github.com/ethersphere/go-ethereum/issues/1177 * p2p/simulations: Add a sanity test case for Node.Config UnmarshalJSON * p2p/simulations: revert back to defer Unlock() pattern for Network It's a good patten to call `defer Unlock()` right after `Lock()` so (new) error cases won't miss to unlock. Let's get back to that pattern. The patten was abandoned in 85a79b3ad3c5863f8612d25c246bcfad339f36b7, while fixing a data race. That data race does not exist anymore, since the Node.Up field got hidden behind its own lock. * p2p/simulations: consistent naming for test providers Node.UnmarshalJSON * p2p/simulations: remove JSON annotation from private fields of Node As unexported fields are not serialized. * p2p/simulations: fix deadlock in Network.GetRandomDownNode() Problem: GetRandomDownNode() locks -> getDownNodeIDs() -> GetNodes() tries to lock -> deadlock On Network type, unexported functions must assume that `net.lock` is already acquired and should not call exported functions which might try to lock again. * p2p/simulations: ensure method conformity for Network Connect* methods were moved to p2p/simulations.Network from swarm/network/simulation. However these new methods did not follow the pattern of Network methods, i.e., all exported method locks the whole Network either for read or write. * p2p/simulations: fix deadlock during network shutdown `TestDiscoveryPersistenceSimulationSimAdapter` often got into deadlock. The execution was stuck on two locks, i.e, `Kademlia.lock` and `p2p/simulations.Network.lock`. Usually the test got stuck once in each 20 executions with high confidence. `Kademlia` was stuck in `Kademlia.EachAddr()` and `Network` in `Network.Stop()`. Solution: in `Network.Stop()` `net.lock` must be released before calling `node.Stop()` as stopping a node (somehow - I did not find the exact code path) causes `Network.InitConn()` to be called from `Kademlia.SuggestPeer()` and that blocks on `net.lock`. Related ticket: https://github.com/ethersphere/go-ethereum/issues/1223 * swarm/state: simplify if statement in DBStore.Put() * p2p/simulations: remove faulty godoc from private function The comment started with the wrong method name. The method is simple and self explanatory. Also, it's private. => Let's just remove the comment.
2019-02-18 00:38:14 -06:00
node.SetUp(true)
log.Info("Started node", "id", id)
ev := NewEvent(node)
net.events.Send(ev)
// subscribe to peer events
client, err := node.Client()
if err != nil {
return fmt.Errorf("error getting rpc client for node %v: %s", id, err)
}
events := make(chan *p2p.PeerEvent)
sub, err := client.Subscribe(context.Background(), "admin", events, "peerEvents")
if err != nil {
return fmt.Errorf("error getting peer events for node %v: %s", id, err)
}
2018-05-08 06:08:43 -05:00
go net.watchPeerEvents(id, events, sub)
return nil
}
// watchPeerEvents reads peer events from the given channel and emits
// corresponding network events
all: new p2p node representation (#17643) Package p2p/enode provides a generalized representation of p2p nodes which can contain arbitrary information in key/value pairs. It is also the new home for the node database. The "v4" identity scheme is also moved here from p2p/enr to remove the dependency on Ethereum crypto from that package. Record signature handling is changed significantly. The identity scheme registry is removed and acceptable schemes must be passed to any method that needs identity. This means records must now be validated explicitly after decoding. The enode API is designed to make signature handling easy and safe: most APIs around the codebase work with enode.Node, which is a wrapper around a valid record. Going from enr.Record to enode.Node requires a valid signature. * p2p/discover: port to p2p/enode This ports the discovery code to the new node representation in p2p/enode. The wire protocol is unchanged, this can be considered a refactoring change. The Kademlia table can now deal with nodes using an arbitrary identity scheme. This requires a few incompatible API changes: - Table.Lookup is not available anymore. It used to take a public key as argument because v4 protocol requires one. Its replacement is LookupRandom. - Table.Resolve takes *enode.Node instead of NodeID. This is also for v4 protocol compatibility because nodes cannot be looked up by ID alone. - Types Node and NodeID are gone. Further commits in the series will be fixes all over the the codebase to deal with those removals. * p2p: port to p2p/enode and discovery changes This adapts package p2p to the changes in p2p/discover. All uses of discover.Node and discover.NodeID are replaced by their equivalents from p2p/enode. New API is added to retrieve the enode.Node instance of a peer. The behavior of Server.Self with discovery disabled is improved. It now tries much harder to report a working IP address, falling back to 127.0.0.1 if no suitable address can be determined through other means. These changes were needed for tests of other packages later in the series. * p2p/simulations, p2p/testing: port to p2p/enode No surprises here, mostly replacements of discover.Node, discover.NodeID with their new equivalents. The 'interesting' API changes are: - testing.ProtocolSession tracks complete nodes, not just their IDs. - adapters.NodeConfig has a new method to create a complete node. These changes were needed to make swarm tests work. Note that the NodeID change makes the code incompatible with old simulation snapshots. * whisper/whisperv5, whisper/whisperv6: port to p2p/enode This port was easy because whisper uses []byte for node IDs and URL strings in the API. * eth: port to p2p/enode Again, easy to port because eth uses strings for node IDs and doesn't care about node information in any way. * les: port to p2p/enode Apart from replacing discover.NodeID with enode.ID, most changes are in the server pool code. It now deals with complete nodes instead of (Pubkey, IP, Port) triples. The database format is unchanged for now, but we should probably change it to use the node database later. * node: port to p2p/enode This change simply replaces discover.Node and discover.NodeID with their new equivalents. * swarm/network: port to p2p/enode Swarm has its own node address representation, BzzAddr, containing both an overlay address (the hash of a secp256k1 public key) and an underlay address (enode:// URL). There are no changes to the BzzAddr format in this commit, but certain operations such as creating a BzzAddr from a node ID are now impossible because node IDs aren't public keys anymore. Most swarm-related changes in the series remove uses of NewAddrFromNodeID, replacing it with NewAddr which takes a complete node as argument. ToOverlayAddr is removed because we can just use the node ID directly.
2018-09-24 17:59:00 -05:00
func (net *Network) watchPeerEvents(id enode.ID, events chan *p2p.PeerEvent, sub event.Subscription) {
defer func() {
sub.Unsubscribe()
// assume the node is now down
2018-05-08 06:08:43 -05:00
net.lock.Lock()
2018-06-20 07:06:27 -05:00
defer net.lock.Unlock()
2018-05-08 06:08:43 -05:00
node := net.getNode(id)
2018-06-20 07:06:27 -05:00
if node == nil {
return
}
p2p, swarm: fix node up races by granular locking (#18976) * swarm/network: DRY out repeated giga comment I not necessarily agree with the way we wait for event propagation. But I truly disagree with having duplicated giga comments. * p2p/simulations: encapsulate Node.Up field so we avoid data races The Node.Up field was accessed concurrently without "proper" locking. There was a lock on Network and that was used sometimes to access the field. Other times the locking was missed and we had a data race. For example: https://github.com/ethereum/go-ethereum/pull/18464 The case above was solved, but there were still intermittent/hard to reproduce races. So let's solve the issue permanently. resolves: ethersphere/go-ethereum#1146 * p2p/simulations: fix unmarshal of simulations.Node Making Node.Up field private in 13292ee897e345045fbfab3bda23a77589a271c1 broke TestHTTPNetwork and TestHTTPSnapshot. Because the default UnmarshalJSON does not handle unexported fields. Important: The fix is partial and not proper to my taste. But I cut scope as I think the fix may require a change to the current serialization format. New ticket: https://github.com/ethersphere/go-ethereum/issues/1177 * p2p/simulations: Add a sanity test case for Node.Config UnmarshalJSON * p2p/simulations: revert back to defer Unlock() pattern for Network It's a good patten to call `defer Unlock()` right after `Lock()` so (new) error cases won't miss to unlock. Let's get back to that pattern. The patten was abandoned in 85a79b3ad3c5863f8612d25c246bcfad339f36b7, while fixing a data race. That data race does not exist anymore, since the Node.Up field got hidden behind its own lock. * p2p/simulations: consistent naming for test providers Node.UnmarshalJSON * p2p/simulations: remove JSON annotation from private fields of Node As unexported fields are not serialized. * p2p/simulations: fix deadlock in Network.GetRandomDownNode() Problem: GetRandomDownNode() locks -> getDownNodeIDs() -> GetNodes() tries to lock -> deadlock On Network type, unexported functions must assume that `net.lock` is already acquired and should not call exported functions which might try to lock again. * p2p/simulations: ensure method conformity for Network Connect* methods were moved to p2p/simulations.Network from swarm/network/simulation. However these new methods did not follow the pattern of Network methods, i.e., all exported method locks the whole Network either for read or write. * p2p/simulations: fix deadlock during network shutdown `TestDiscoveryPersistenceSimulationSimAdapter` often got into deadlock. The execution was stuck on two locks, i.e, `Kademlia.lock` and `p2p/simulations.Network.lock`. Usually the test got stuck once in each 20 executions with high confidence. `Kademlia` was stuck in `Kademlia.EachAddr()` and `Network` in `Network.Stop()`. Solution: in `Network.Stop()` `net.lock` must be released before calling `node.Stop()` as stopping a node (somehow - I did not find the exact code path) causes `Network.InitConn()` to be called from `Kademlia.SuggestPeer()` and that blocks on `net.lock`. Related ticket: https://github.com/ethersphere/go-ethereum/issues/1223 * swarm/state: simplify if statement in DBStore.Put() * p2p/simulations: remove faulty godoc from private function The comment started with the wrong method name. The method is simple and self explanatory. Also, it's private. => Let's just remove the comment.
2019-02-18 00:38:14 -06:00
node.SetUp(false)
ev := NewEvent(node)
net.events.Send(ev)
}()
for {
select {
case event, ok := <-events:
if !ok {
return
}
peer := event.Peer
switch event.Type {
case p2p.PeerEventTypeAdd:
2018-05-08 06:08:43 -05:00
net.DidConnect(id, peer)
case p2p.PeerEventTypeDrop:
2018-05-08 06:08:43 -05:00
net.DidDisconnect(id, peer)
case p2p.PeerEventTypeMsgSend:
2018-05-08 06:08:43 -05:00
net.DidSend(id, peer, event.Protocol, *event.MsgCode)
case p2p.PeerEventTypeMsgRecv:
2018-05-08 06:08:43 -05:00
net.DidReceive(peer, id, event.Protocol, *event.MsgCode)
}
case err := <-sub.Err():
if err != nil {
log.Error("Error in peer event subscription", "id", id, "err", err)
}
return
}
}
}
// Stop stops the node with the given ID
all: new p2p node representation (#17643) Package p2p/enode provides a generalized representation of p2p nodes which can contain arbitrary information in key/value pairs. It is also the new home for the node database. The "v4" identity scheme is also moved here from p2p/enr to remove the dependency on Ethereum crypto from that package. Record signature handling is changed significantly. The identity scheme registry is removed and acceptable schemes must be passed to any method that needs identity. This means records must now be validated explicitly after decoding. The enode API is designed to make signature handling easy and safe: most APIs around the codebase work with enode.Node, which is a wrapper around a valid record. Going from enr.Record to enode.Node requires a valid signature. * p2p/discover: port to p2p/enode This ports the discovery code to the new node representation in p2p/enode. The wire protocol is unchanged, this can be considered a refactoring change. The Kademlia table can now deal with nodes using an arbitrary identity scheme. This requires a few incompatible API changes: - Table.Lookup is not available anymore. It used to take a public key as argument because v4 protocol requires one. Its replacement is LookupRandom. - Table.Resolve takes *enode.Node instead of NodeID. This is also for v4 protocol compatibility because nodes cannot be looked up by ID alone. - Types Node and NodeID are gone. Further commits in the series will be fixes all over the the codebase to deal with those removals. * p2p: port to p2p/enode and discovery changes This adapts package p2p to the changes in p2p/discover. All uses of discover.Node and discover.NodeID are replaced by their equivalents from p2p/enode. New API is added to retrieve the enode.Node instance of a peer. The behavior of Server.Self with discovery disabled is improved. It now tries much harder to report a working IP address, falling back to 127.0.0.1 if no suitable address can be determined through other means. These changes were needed for tests of other packages later in the series. * p2p/simulations, p2p/testing: port to p2p/enode No surprises here, mostly replacements of discover.Node, discover.NodeID with their new equivalents. The 'interesting' API changes are: - testing.ProtocolSession tracks complete nodes, not just their IDs. - adapters.NodeConfig has a new method to create a complete node. These changes were needed to make swarm tests work. Note that the NodeID change makes the code incompatible with old simulation snapshots. * whisper/whisperv5, whisper/whisperv6: port to p2p/enode This port was easy because whisper uses []byte for node IDs and URL strings in the API. * eth: port to p2p/enode Again, easy to port because eth uses strings for node IDs and doesn't care about node information in any way. * les: port to p2p/enode Apart from replacing discover.NodeID with enode.ID, most changes are in the server pool code. It now deals with complete nodes instead of (Pubkey, IP, Port) triples. The database format is unchanged for now, but we should probably change it to use the node database later. * node: port to p2p/enode This change simply replaces discover.Node and discover.NodeID with their new equivalents. * swarm/network: port to p2p/enode Swarm has its own node address representation, BzzAddr, containing both an overlay address (the hash of a secp256k1 public key) and an underlay address (enode:// URL). There are no changes to the BzzAddr format in this commit, but certain operations such as creating a BzzAddr from a node ID are now impossible because node IDs aren't public keys anymore. Most swarm-related changes in the series remove uses of NewAddrFromNodeID, replacing it with NewAddr which takes a complete node as argument. ToOverlayAddr is removed because we can just use the node ID directly.
2018-09-24 17:59:00 -05:00
func (net *Network) Stop(id enode.ID) error {
p2p, swarm: fix node up races by granular locking (#18976) * swarm/network: DRY out repeated giga comment I not necessarily agree with the way we wait for event propagation. But I truly disagree with having duplicated giga comments. * p2p/simulations: encapsulate Node.Up field so we avoid data races The Node.Up field was accessed concurrently without "proper" locking. There was a lock on Network and that was used sometimes to access the field. Other times the locking was missed and we had a data race. For example: https://github.com/ethereum/go-ethereum/pull/18464 The case above was solved, but there were still intermittent/hard to reproduce races. So let's solve the issue permanently. resolves: ethersphere/go-ethereum#1146 * p2p/simulations: fix unmarshal of simulations.Node Making Node.Up field private in 13292ee897e345045fbfab3bda23a77589a271c1 broke TestHTTPNetwork and TestHTTPSnapshot. Because the default UnmarshalJSON does not handle unexported fields. Important: The fix is partial and not proper to my taste. But I cut scope as I think the fix may require a change to the current serialization format. New ticket: https://github.com/ethersphere/go-ethereum/issues/1177 * p2p/simulations: Add a sanity test case for Node.Config UnmarshalJSON * p2p/simulations: revert back to defer Unlock() pattern for Network It's a good patten to call `defer Unlock()` right after `Lock()` so (new) error cases won't miss to unlock. Let's get back to that pattern. The patten was abandoned in 85a79b3ad3c5863f8612d25c246bcfad339f36b7, while fixing a data race. That data race does not exist anymore, since the Node.Up field got hidden behind its own lock. * p2p/simulations: consistent naming for test providers Node.UnmarshalJSON * p2p/simulations: remove JSON annotation from private fields of Node As unexported fields are not serialized. * p2p/simulations: fix deadlock in Network.GetRandomDownNode() Problem: GetRandomDownNode() locks -> getDownNodeIDs() -> GetNodes() tries to lock -> deadlock On Network type, unexported functions must assume that `net.lock` is already acquired and should not call exported functions which might try to lock again. * p2p/simulations: ensure method conformity for Network Connect* methods were moved to p2p/simulations.Network from swarm/network/simulation. However these new methods did not follow the pattern of Network methods, i.e., all exported method locks the whole Network either for read or write. * p2p/simulations: fix deadlock during network shutdown `TestDiscoveryPersistenceSimulationSimAdapter` often got into deadlock. The execution was stuck on two locks, i.e, `Kademlia.lock` and `p2p/simulations.Network.lock`. Usually the test got stuck once in each 20 executions with high confidence. `Kademlia` was stuck in `Kademlia.EachAddr()` and `Network` in `Network.Stop()`. Solution: in `Network.Stop()` `net.lock` must be released before calling `node.Stop()` as stopping a node (somehow - I did not find the exact code path) causes `Network.InitConn()` to be called from `Kademlia.SuggestPeer()` and that blocks on `net.lock`. Related ticket: https://github.com/ethersphere/go-ethereum/issues/1223 * swarm/state: simplify if statement in DBStore.Put() * p2p/simulations: remove faulty godoc from private function The comment started with the wrong method name. The method is simple and self explanatory. Also, it's private. => Let's just remove the comment.
2019-02-18 00:38:14 -06:00
// IMPORTANT: node.Stop() must NOT be called under net.lock as
// node.Reachable() closure has a reference to the network and
// calls net.InitConn() what also locks the network. => DEADLOCK
// That holds until the following ticket is not resolved:
var err error
node, err := func() (*Node, error) {
net.lock.Lock()
defer net.lock.Unlock()
node := net.getNode(id)
if node == nil {
return nil, fmt.Errorf("node %v does not exist", id)
}
if !node.Up() {
return nil, fmt.Errorf("node %v already down", id)
}
node.SetUp(false)
return node, nil
}()
if err != nil {
return err
}
p2p, swarm: fix node up races by granular locking (#18976) * swarm/network: DRY out repeated giga comment I not necessarily agree with the way we wait for event propagation. But I truly disagree with having duplicated giga comments. * p2p/simulations: encapsulate Node.Up field so we avoid data races The Node.Up field was accessed concurrently without "proper" locking. There was a lock on Network and that was used sometimes to access the field. Other times the locking was missed and we had a data race. For example: https://github.com/ethereum/go-ethereum/pull/18464 The case above was solved, but there were still intermittent/hard to reproduce races. So let's solve the issue permanently. resolves: ethersphere/go-ethereum#1146 * p2p/simulations: fix unmarshal of simulations.Node Making Node.Up field private in 13292ee897e345045fbfab3bda23a77589a271c1 broke TestHTTPNetwork and TestHTTPSnapshot. Because the default UnmarshalJSON does not handle unexported fields. Important: The fix is partial and not proper to my taste. But I cut scope as I think the fix may require a change to the current serialization format. New ticket: https://github.com/ethersphere/go-ethereum/issues/1177 * p2p/simulations: Add a sanity test case for Node.Config UnmarshalJSON * p2p/simulations: revert back to defer Unlock() pattern for Network It's a good patten to call `defer Unlock()` right after `Lock()` so (new) error cases won't miss to unlock. Let's get back to that pattern. The patten was abandoned in 85a79b3ad3c5863f8612d25c246bcfad339f36b7, while fixing a data race. That data race does not exist anymore, since the Node.Up field got hidden behind its own lock. * p2p/simulations: consistent naming for test providers Node.UnmarshalJSON * p2p/simulations: remove JSON annotation from private fields of Node As unexported fields are not serialized. * p2p/simulations: fix deadlock in Network.GetRandomDownNode() Problem: GetRandomDownNode() locks -> getDownNodeIDs() -> GetNodes() tries to lock -> deadlock On Network type, unexported functions must assume that `net.lock` is already acquired and should not call exported functions which might try to lock again. * p2p/simulations: ensure method conformity for Network Connect* methods were moved to p2p/simulations.Network from swarm/network/simulation. However these new methods did not follow the pattern of Network methods, i.e., all exported method locks the whole Network either for read or write. * p2p/simulations: fix deadlock during network shutdown `TestDiscoveryPersistenceSimulationSimAdapter` often got into deadlock. The execution was stuck on two locks, i.e, `Kademlia.lock` and `p2p/simulations.Network.lock`. Usually the test got stuck once in each 20 executions with high confidence. `Kademlia` was stuck in `Kademlia.EachAddr()` and `Network` in `Network.Stop()`. Solution: in `Network.Stop()` `net.lock` must be released before calling `node.Stop()` as stopping a node (somehow - I did not find the exact code path) causes `Network.InitConn()` to be called from `Kademlia.SuggestPeer()` and that blocks on `net.lock`. Related ticket: https://github.com/ethersphere/go-ethereum/issues/1223 * swarm/state: simplify if statement in DBStore.Put() * p2p/simulations: remove faulty godoc from private function The comment started with the wrong method name. The method is simple and self explanatory. Also, it's private. => Let's just remove the comment.
2019-02-18 00:38:14 -06:00
err = node.Stop() // must be called without net.lock
net.lock.Lock()
defer net.lock.Unlock()
if err != nil {
p2p, swarm: fix node up races by granular locking (#18976) * swarm/network: DRY out repeated giga comment I not necessarily agree with the way we wait for event propagation. But I truly disagree with having duplicated giga comments. * p2p/simulations: encapsulate Node.Up field so we avoid data races The Node.Up field was accessed concurrently without "proper" locking. There was a lock on Network and that was used sometimes to access the field. Other times the locking was missed and we had a data race. For example: https://github.com/ethereum/go-ethereum/pull/18464 The case above was solved, but there were still intermittent/hard to reproduce races. So let's solve the issue permanently. resolves: ethersphere/go-ethereum#1146 * p2p/simulations: fix unmarshal of simulations.Node Making Node.Up field private in 13292ee897e345045fbfab3bda23a77589a271c1 broke TestHTTPNetwork and TestHTTPSnapshot. Because the default UnmarshalJSON does not handle unexported fields. Important: The fix is partial and not proper to my taste. But I cut scope as I think the fix may require a change to the current serialization format. New ticket: https://github.com/ethersphere/go-ethereum/issues/1177 * p2p/simulations: Add a sanity test case for Node.Config UnmarshalJSON * p2p/simulations: revert back to defer Unlock() pattern for Network It's a good patten to call `defer Unlock()` right after `Lock()` so (new) error cases won't miss to unlock. Let's get back to that pattern. The patten was abandoned in 85a79b3ad3c5863f8612d25c246bcfad339f36b7, while fixing a data race. That data race does not exist anymore, since the Node.Up field got hidden behind its own lock. * p2p/simulations: consistent naming for test providers Node.UnmarshalJSON * p2p/simulations: remove JSON annotation from private fields of Node As unexported fields are not serialized. * p2p/simulations: fix deadlock in Network.GetRandomDownNode() Problem: GetRandomDownNode() locks -> getDownNodeIDs() -> GetNodes() tries to lock -> deadlock On Network type, unexported functions must assume that `net.lock` is already acquired and should not call exported functions which might try to lock again. * p2p/simulations: ensure method conformity for Network Connect* methods were moved to p2p/simulations.Network from swarm/network/simulation. However these new methods did not follow the pattern of Network methods, i.e., all exported method locks the whole Network either for read or write. * p2p/simulations: fix deadlock during network shutdown `TestDiscoveryPersistenceSimulationSimAdapter` often got into deadlock. The execution was stuck on two locks, i.e, `Kademlia.lock` and `p2p/simulations.Network.lock`. Usually the test got stuck once in each 20 executions with high confidence. `Kademlia` was stuck in `Kademlia.EachAddr()` and `Network` in `Network.Stop()`. Solution: in `Network.Stop()` `net.lock` must be released before calling `node.Stop()` as stopping a node (somehow - I did not find the exact code path) causes `Network.InitConn()` to be called from `Kademlia.SuggestPeer()` and that blocks on `net.lock`. Related ticket: https://github.com/ethersphere/go-ethereum/issues/1223 * swarm/state: simplify if statement in DBStore.Put() * p2p/simulations: remove faulty godoc from private function The comment started with the wrong method name. The method is simple and self explanatory. Also, it's private. => Let's just remove the comment.
2019-02-18 00:38:14 -06:00
node.SetUp(true)
return err
}
log.Info("Stopped node", "id", id, "err", err)
ev := ControlEvent(node)
net.events.Send(ev)
return nil
}
// Connect connects two nodes together by calling the "admin_addPeer" RPC
// method on the "one" node so that it connects to the "other" node
all: new p2p node representation (#17643) Package p2p/enode provides a generalized representation of p2p nodes which can contain arbitrary information in key/value pairs. It is also the new home for the node database. The "v4" identity scheme is also moved here from p2p/enr to remove the dependency on Ethereum crypto from that package. Record signature handling is changed significantly. The identity scheme registry is removed and acceptable schemes must be passed to any method that needs identity. This means records must now be validated explicitly after decoding. The enode API is designed to make signature handling easy and safe: most APIs around the codebase work with enode.Node, which is a wrapper around a valid record. Going from enr.Record to enode.Node requires a valid signature. * p2p/discover: port to p2p/enode This ports the discovery code to the new node representation in p2p/enode. The wire protocol is unchanged, this can be considered a refactoring change. The Kademlia table can now deal with nodes using an arbitrary identity scheme. This requires a few incompatible API changes: - Table.Lookup is not available anymore. It used to take a public key as argument because v4 protocol requires one. Its replacement is LookupRandom. - Table.Resolve takes *enode.Node instead of NodeID. This is also for v4 protocol compatibility because nodes cannot be looked up by ID alone. - Types Node and NodeID are gone. Further commits in the series will be fixes all over the the codebase to deal with those removals. * p2p: port to p2p/enode and discovery changes This adapts package p2p to the changes in p2p/discover. All uses of discover.Node and discover.NodeID are replaced by their equivalents from p2p/enode. New API is added to retrieve the enode.Node instance of a peer. The behavior of Server.Self with discovery disabled is improved. It now tries much harder to report a working IP address, falling back to 127.0.0.1 if no suitable address can be determined through other means. These changes were needed for tests of other packages later in the series. * p2p/simulations, p2p/testing: port to p2p/enode No surprises here, mostly replacements of discover.Node, discover.NodeID with their new equivalents. The 'interesting' API changes are: - testing.ProtocolSession tracks complete nodes, not just their IDs. - adapters.NodeConfig has a new method to create a complete node. These changes were needed to make swarm tests work. Note that the NodeID change makes the code incompatible with old simulation snapshots. * whisper/whisperv5, whisper/whisperv6: port to p2p/enode This port was easy because whisper uses []byte for node IDs and URL strings in the API. * eth: port to p2p/enode Again, easy to port because eth uses strings for node IDs and doesn't care about node information in any way. * les: port to p2p/enode Apart from replacing discover.NodeID with enode.ID, most changes are in the server pool code. It now deals with complete nodes instead of (Pubkey, IP, Port) triples. The database format is unchanged for now, but we should probably change it to use the node database later. * node: port to p2p/enode This change simply replaces discover.Node and discover.NodeID with their new equivalents. * swarm/network: port to p2p/enode Swarm has its own node address representation, BzzAddr, containing both an overlay address (the hash of a secp256k1 public key) and an underlay address (enode:// URL). There are no changes to the BzzAddr format in this commit, but certain operations such as creating a BzzAddr from a node ID are now impossible because node IDs aren't public keys anymore. Most swarm-related changes in the series remove uses of NewAddrFromNodeID, replacing it with NewAddr which takes a complete node as argument. ToOverlayAddr is removed because we can just use the node ID directly.
2018-09-24 17:59:00 -05:00
func (net *Network) Connect(oneID, otherID enode.ID) error {
p2p, swarm: fix node up races by granular locking (#18976) * swarm/network: DRY out repeated giga comment I not necessarily agree with the way we wait for event propagation. But I truly disagree with having duplicated giga comments. * p2p/simulations: encapsulate Node.Up field so we avoid data races The Node.Up field was accessed concurrently without "proper" locking. There was a lock on Network and that was used sometimes to access the field. Other times the locking was missed and we had a data race. For example: https://github.com/ethereum/go-ethereum/pull/18464 The case above was solved, but there were still intermittent/hard to reproduce races. So let's solve the issue permanently. resolves: ethersphere/go-ethereum#1146 * p2p/simulations: fix unmarshal of simulations.Node Making Node.Up field private in 13292ee897e345045fbfab3bda23a77589a271c1 broke TestHTTPNetwork and TestHTTPSnapshot. Because the default UnmarshalJSON does not handle unexported fields. Important: The fix is partial and not proper to my taste. But I cut scope as I think the fix may require a change to the current serialization format. New ticket: https://github.com/ethersphere/go-ethereum/issues/1177 * p2p/simulations: Add a sanity test case for Node.Config UnmarshalJSON * p2p/simulations: revert back to defer Unlock() pattern for Network It's a good patten to call `defer Unlock()` right after `Lock()` so (new) error cases won't miss to unlock. Let's get back to that pattern. The patten was abandoned in 85a79b3ad3c5863f8612d25c246bcfad339f36b7, while fixing a data race. That data race does not exist anymore, since the Node.Up field got hidden behind its own lock. * p2p/simulations: consistent naming for test providers Node.UnmarshalJSON * p2p/simulations: remove JSON annotation from private fields of Node As unexported fields are not serialized. * p2p/simulations: fix deadlock in Network.GetRandomDownNode() Problem: GetRandomDownNode() locks -> getDownNodeIDs() -> GetNodes() tries to lock -> deadlock On Network type, unexported functions must assume that `net.lock` is already acquired and should not call exported functions which might try to lock again. * p2p/simulations: ensure method conformity for Network Connect* methods were moved to p2p/simulations.Network from swarm/network/simulation. However these new methods did not follow the pattern of Network methods, i.e., all exported method locks the whole Network either for read or write. * p2p/simulations: fix deadlock during network shutdown `TestDiscoveryPersistenceSimulationSimAdapter` often got into deadlock. The execution was stuck on two locks, i.e, `Kademlia.lock` and `p2p/simulations.Network.lock`. Usually the test got stuck once in each 20 executions with high confidence. `Kademlia` was stuck in `Kademlia.EachAddr()` and `Network` in `Network.Stop()`. Solution: in `Network.Stop()` `net.lock` must be released before calling `node.Stop()` as stopping a node (somehow - I did not find the exact code path) causes `Network.InitConn()` to be called from `Kademlia.SuggestPeer()` and that blocks on `net.lock`. Related ticket: https://github.com/ethersphere/go-ethereum/issues/1223 * swarm/state: simplify if statement in DBStore.Put() * p2p/simulations: remove faulty godoc from private function The comment started with the wrong method name. The method is simple and self explanatory. Also, it's private. => Let's just remove the comment.
2019-02-18 00:38:14 -06:00
net.lock.Lock()
defer net.lock.Unlock()
return net.connect(oneID, otherID)
}
func (net *Network) connect(oneID, otherID enode.ID) error {
log.Debug("Connecting nodes with addPeer", "id", oneID, "other", otherID)
p2p, swarm: fix node up races by granular locking (#18976) * swarm/network: DRY out repeated giga comment I not necessarily agree with the way we wait for event propagation. But I truly disagree with having duplicated giga comments. * p2p/simulations: encapsulate Node.Up field so we avoid data races The Node.Up field was accessed concurrently without "proper" locking. There was a lock on Network and that was used sometimes to access the field. Other times the locking was missed and we had a data race. For example: https://github.com/ethereum/go-ethereum/pull/18464 The case above was solved, but there were still intermittent/hard to reproduce races. So let's solve the issue permanently. resolves: ethersphere/go-ethereum#1146 * p2p/simulations: fix unmarshal of simulations.Node Making Node.Up field private in 13292ee897e345045fbfab3bda23a77589a271c1 broke TestHTTPNetwork and TestHTTPSnapshot. Because the default UnmarshalJSON does not handle unexported fields. Important: The fix is partial and not proper to my taste. But I cut scope as I think the fix may require a change to the current serialization format. New ticket: https://github.com/ethersphere/go-ethereum/issues/1177 * p2p/simulations: Add a sanity test case for Node.Config UnmarshalJSON * p2p/simulations: revert back to defer Unlock() pattern for Network It's a good patten to call `defer Unlock()` right after `Lock()` so (new) error cases won't miss to unlock. Let's get back to that pattern. The patten was abandoned in 85a79b3ad3c5863f8612d25c246bcfad339f36b7, while fixing a data race. That data race does not exist anymore, since the Node.Up field got hidden behind its own lock. * p2p/simulations: consistent naming for test providers Node.UnmarshalJSON * p2p/simulations: remove JSON annotation from private fields of Node As unexported fields are not serialized. * p2p/simulations: fix deadlock in Network.GetRandomDownNode() Problem: GetRandomDownNode() locks -> getDownNodeIDs() -> GetNodes() tries to lock -> deadlock On Network type, unexported functions must assume that `net.lock` is already acquired and should not call exported functions which might try to lock again. * p2p/simulations: ensure method conformity for Network Connect* methods were moved to p2p/simulations.Network from swarm/network/simulation. However these new methods did not follow the pattern of Network methods, i.e., all exported method locks the whole Network either for read or write. * p2p/simulations: fix deadlock during network shutdown `TestDiscoveryPersistenceSimulationSimAdapter` often got into deadlock. The execution was stuck on two locks, i.e, `Kademlia.lock` and `p2p/simulations.Network.lock`. Usually the test got stuck once in each 20 executions with high confidence. `Kademlia` was stuck in `Kademlia.EachAddr()` and `Network` in `Network.Stop()`. Solution: in `Network.Stop()` `net.lock` must be released before calling `node.Stop()` as stopping a node (somehow - I did not find the exact code path) causes `Network.InitConn()` to be called from `Kademlia.SuggestPeer()` and that blocks on `net.lock`. Related ticket: https://github.com/ethersphere/go-ethereum/issues/1223 * swarm/state: simplify if statement in DBStore.Put() * p2p/simulations: remove faulty godoc from private function The comment started with the wrong method name. The method is simple and self explanatory. Also, it's private. => Let's just remove the comment.
2019-02-18 00:38:14 -06:00
conn, err := net.initConn(oneID, otherID)
if err != nil {
return err
}
client, err := conn.one.Client()
if err != nil {
return err
}
2018-05-08 06:08:43 -05:00
net.events.Send(ControlEvent(conn))
return client.Call(nil, "admin_addPeer", string(conn.other.Addr()))
}
// Disconnect disconnects two nodes by calling the "admin_removePeer" RPC
// method on the "one" node so that it disconnects from the "other" node
all: new p2p node representation (#17643) Package p2p/enode provides a generalized representation of p2p nodes which can contain arbitrary information in key/value pairs. It is also the new home for the node database. The "v4" identity scheme is also moved here from p2p/enr to remove the dependency on Ethereum crypto from that package. Record signature handling is changed significantly. The identity scheme registry is removed and acceptable schemes must be passed to any method that needs identity. This means records must now be validated explicitly after decoding. The enode API is designed to make signature handling easy and safe: most APIs around the codebase work with enode.Node, which is a wrapper around a valid record. Going from enr.Record to enode.Node requires a valid signature. * p2p/discover: port to p2p/enode This ports the discovery code to the new node representation in p2p/enode. The wire protocol is unchanged, this can be considered a refactoring change. The Kademlia table can now deal with nodes using an arbitrary identity scheme. This requires a few incompatible API changes: - Table.Lookup is not available anymore. It used to take a public key as argument because v4 protocol requires one. Its replacement is LookupRandom. - Table.Resolve takes *enode.Node instead of NodeID. This is also for v4 protocol compatibility because nodes cannot be looked up by ID alone. - Types Node and NodeID are gone. Further commits in the series will be fixes all over the the codebase to deal with those removals. * p2p: port to p2p/enode and discovery changes This adapts package p2p to the changes in p2p/discover. All uses of discover.Node and discover.NodeID are replaced by their equivalents from p2p/enode. New API is added to retrieve the enode.Node instance of a peer. The behavior of Server.Self with discovery disabled is improved. It now tries much harder to report a working IP address, falling back to 127.0.0.1 if no suitable address can be determined through other means. These changes were needed for tests of other packages later in the series. * p2p/simulations, p2p/testing: port to p2p/enode No surprises here, mostly replacements of discover.Node, discover.NodeID with their new equivalents. The 'interesting' API changes are: - testing.ProtocolSession tracks complete nodes, not just their IDs. - adapters.NodeConfig has a new method to create a complete node. These changes were needed to make swarm tests work. Note that the NodeID change makes the code incompatible with old simulation snapshots. * whisper/whisperv5, whisper/whisperv6: port to p2p/enode This port was easy because whisper uses []byte for node IDs and URL strings in the API. * eth: port to p2p/enode Again, easy to port because eth uses strings for node IDs and doesn't care about node information in any way. * les: port to p2p/enode Apart from replacing discover.NodeID with enode.ID, most changes are in the server pool code. It now deals with complete nodes instead of (Pubkey, IP, Port) triples. The database format is unchanged for now, but we should probably change it to use the node database later. * node: port to p2p/enode This change simply replaces discover.Node and discover.NodeID with their new equivalents. * swarm/network: port to p2p/enode Swarm has its own node address representation, BzzAddr, containing both an overlay address (the hash of a secp256k1 public key) and an underlay address (enode:// URL). There are no changes to the BzzAddr format in this commit, but certain operations such as creating a BzzAddr from a node ID are now impossible because node IDs aren't public keys anymore. Most swarm-related changes in the series remove uses of NewAddrFromNodeID, replacing it with NewAddr which takes a complete node as argument. ToOverlayAddr is removed because we can just use the node ID directly.
2018-09-24 17:59:00 -05:00
func (net *Network) Disconnect(oneID, otherID enode.ID) error {
2018-05-08 06:08:43 -05:00
conn := net.GetConn(oneID, otherID)
if conn == nil {
return fmt.Errorf("connection between %v and %v does not exist", oneID, otherID)
}
if !conn.Up {
return fmt.Errorf("%v and %v already disconnected", oneID, otherID)
}
client, err := conn.one.Client()
if err != nil {
return err
}
2018-05-08 06:08:43 -05:00
net.events.Send(ControlEvent(conn))
return client.Call(nil, "admin_removePeer", string(conn.other.Addr()))
}
// DidConnect tracks the fact that the "one" node connected to the "other" node
all: new p2p node representation (#17643) Package p2p/enode provides a generalized representation of p2p nodes which can contain arbitrary information in key/value pairs. It is also the new home for the node database. The "v4" identity scheme is also moved here from p2p/enr to remove the dependency on Ethereum crypto from that package. Record signature handling is changed significantly. The identity scheme registry is removed and acceptable schemes must be passed to any method that needs identity. This means records must now be validated explicitly after decoding. The enode API is designed to make signature handling easy and safe: most APIs around the codebase work with enode.Node, which is a wrapper around a valid record. Going from enr.Record to enode.Node requires a valid signature. * p2p/discover: port to p2p/enode This ports the discovery code to the new node representation in p2p/enode. The wire protocol is unchanged, this can be considered a refactoring change. The Kademlia table can now deal with nodes using an arbitrary identity scheme. This requires a few incompatible API changes: - Table.Lookup is not available anymore. It used to take a public key as argument because v4 protocol requires one. Its replacement is LookupRandom. - Table.Resolve takes *enode.Node instead of NodeID. This is also for v4 protocol compatibility because nodes cannot be looked up by ID alone. - Types Node and NodeID are gone. Further commits in the series will be fixes all over the the codebase to deal with those removals. * p2p: port to p2p/enode and discovery changes This adapts package p2p to the changes in p2p/discover. All uses of discover.Node and discover.NodeID are replaced by their equivalents from p2p/enode. New API is added to retrieve the enode.Node instance of a peer. The behavior of Server.Self with discovery disabled is improved. It now tries much harder to report a working IP address, falling back to 127.0.0.1 if no suitable address can be determined through other means. These changes were needed for tests of other packages later in the series. * p2p/simulations, p2p/testing: port to p2p/enode No surprises here, mostly replacements of discover.Node, discover.NodeID with their new equivalents. The 'interesting' API changes are: - testing.ProtocolSession tracks complete nodes, not just their IDs. - adapters.NodeConfig has a new method to create a complete node. These changes were needed to make swarm tests work. Note that the NodeID change makes the code incompatible with old simulation snapshots. * whisper/whisperv5, whisper/whisperv6: port to p2p/enode This port was easy because whisper uses []byte for node IDs and URL strings in the API. * eth: port to p2p/enode Again, easy to port because eth uses strings for node IDs and doesn't care about node information in any way. * les: port to p2p/enode Apart from replacing discover.NodeID with enode.ID, most changes are in the server pool code. It now deals with complete nodes instead of (Pubkey, IP, Port) triples. The database format is unchanged for now, but we should probably change it to use the node database later. * node: port to p2p/enode This change simply replaces discover.Node and discover.NodeID with their new equivalents. * swarm/network: port to p2p/enode Swarm has its own node address representation, BzzAddr, containing both an overlay address (the hash of a secp256k1 public key) and an underlay address (enode:// URL). There are no changes to the BzzAddr format in this commit, but certain operations such as creating a BzzAddr from a node ID are now impossible because node IDs aren't public keys anymore. Most swarm-related changes in the series remove uses of NewAddrFromNodeID, replacing it with NewAddr which takes a complete node as argument. ToOverlayAddr is removed because we can just use the node ID directly.
2018-09-24 17:59:00 -05:00
func (net *Network) DidConnect(one, other enode.ID) error {
2018-06-20 07:06:27 -05:00
net.lock.Lock()
defer net.lock.Unlock()
conn, err := net.getOrCreateConn(one, other)
if err != nil {
return fmt.Errorf("connection between %v and %v does not exist", one, other)
}
if conn.Up {
return fmt.Errorf("%v and %v already connected", one, other)
}
conn.Up = true
2018-05-08 06:08:43 -05:00
net.events.Send(NewEvent(conn))
return nil
}
// DidDisconnect tracks the fact that the "one" node disconnected from the
// "other" node
all: new p2p node representation (#17643) Package p2p/enode provides a generalized representation of p2p nodes which can contain arbitrary information in key/value pairs. It is also the new home for the node database. The "v4" identity scheme is also moved here from p2p/enr to remove the dependency on Ethereum crypto from that package. Record signature handling is changed significantly. The identity scheme registry is removed and acceptable schemes must be passed to any method that needs identity. This means records must now be validated explicitly after decoding. The enode API is designed to make signature handling easy and safe: most APIs around the codebase work with enode.Node, which is a wrapper around a valid record. Going from enr.Record to enode.Node requires a valid signature. * p2p/discover: port to p2p/enode This ports the discovery code to the new node representation in p2p/enode. The wire protocol is unchanged, this can be considered a refactoring change. The Kademlia table can now deal with nodes using an arbitrary identity scheme. This requires a few incompatible API changes: - Table.Lookup is not available anymore. It used to take a public key as argument because v4 protocol requires one. Its replacement is LookupRandom. - Table.Resolve takes *enode.Node instead of NodeID. This is also for v4 protocol compatibility because nodes cannot be looked up by ID alone. - Types Node and NodeID are gone. Further commits in the series will be fixes all over the the codebase to deal with those removals. * p2p: port to p2p/enode and discovery changes This adapts package p2p to the changes in p2p/discover. All uses of discover.Node and discover.NodeID are replaced by their equivalents from p2p/enode. New API is added to retrieve the enode.Node instance of a peer. The behavior of Server.Self with discovery disabled is improved. It now tries much harder to report a working IP address, falling back to 127.0.0.1 if no suitable address can be determined through other means. These changes were needed for tests of other packages later in the series. * p2p/simulations, p2p/testing: port to p2p/enode No surprises here, mostly replacements of discover.Node, discover.NodeID with their new equivalents. The 'interesting' API changes are: - testing.ProtocolSession tracks complete nodes, not just their IDs. - adapters.NodeConfig has a new method to create a complete node. These changes were needed to make swarm tests work. Note that the NodeID change makes the code incompatible with old simulation snapshots. * whisper/whisperv5, whisper/whisperv6: port to p2p/enode This port was easy because whisper uses []byte for node IDs and URL strings in the API. * eth: port to p2p/enode Again, easy to port because eth uses strings for node IDs and doesn't care about node information in any way. * les: port to p2p/enode Apart from replacing discover.NodeID with enode.ID, most changes are in the server pool code. It now deals with complete nodes instead of (Pubkey, IP, Port) triples. The database format is unchanged for now, but we should probably change it to use the node database later. * node: port to p2p/enode This change simply replaces discover.Node and discover.NodeID with their new equivalents. * swarm/network: port to p2p/enode Swarm has its own node address representation, BzzAddr, containing both an overlay address (the hash of a secp256k1 public key) and an underlay address (enode:// URL). There are no changes to the BzzAddr format in this commit, but certain operations such as creating a BzzAddr from a node ID are now impossible because node IDs aren't public keys anymore. Most swarm-related changes in the series remove uses of NewAddrFromNodeID, replacing it with NewAddr which takes a complete node as argument. ToOverlayAddr is removed because we can just use the node ID directly.
2018-09-24 17:59:00 -05:00
func (net *Network) DidDisconnect(one, other enode.ID) error {
2018-06-20 07:06:27 -05:00
net.lock.Lock()
defer net.lock.Unlock()
conn := net.getConn(one, other)
if conn == nil {
return fmt.Errorf("connection between %v and %v does not exist", one, other)
}
if !conn.Up {
return fmt.Errorf("%v and %v already disconnected", one, other)
}
conn.Up = false
2018-06-20 07:06:27 -05:00
conn.initiated = time.Now().Add(-DialBanTimeout)
2018-05-08 06:08:43 -05:00
net.events.Send(NewEvent(conn))
return nil
}
// DidSend tracks the fact that "sender" sent a message to "receiver"
all: new p2p node representation (#17643) Package p2p/enode provides a generalized representation of p2p nodes which can contain arbitrary information in key/value pairs. It is also the new home for the node database. The "v4" identity scheme is also moved here from p2p/enr to remove the dependency on Ethereum crypto from that package. Record signature handling is changed significantly. The identity scheme registry is removed and acceptable schemes must be passed to any method that needs identity. This means records must now be validated explicitly after decoding. The enode API is designed to make signature handling easy and safe: most APIs around the codebase work with enode.Node, which is a wrapper around a valid record. Going from enr.Record to enode.Node requires a valid signature. * p2p/discover: port to p2p/enode This ports the discovery code to the new node representation in p2p/enode. The wire protocol is unchanged, this can be considered a refactoring change. The Kademlia table can now deal with nodes using an arbitrary identity scheme. This requires a few incompatible API changes: - Table.Lookup is not available anymore. It used to take a public key as argument because v4 protocol requires one. Its replacement is LookupRandom. - Table.Resolve takes *enode.Node instead of NodeID. This is also for v4 protocol compatibility because nodes cannot be looked up by ID alone. - Types Node and NodeID are gone. Further commits in the series will be fixes all over the the codebase to deal with those removals. * p2p: port to p2p/enode and discovery changes This adapts package p2p to the changes in p2p/discover. All uses of discover.Node and discover.NodeID are replaced by their equivalents from p2p/enode. New API is added to retrieve the enode.Node instance of a peer. The behavior of Server.Self with discovery disabled is improved. It now tries much harder to report a working IP address, falling back to 127.0.0.1 if no suitable address can be determined through other means. These changes were needed for tests of other packages later in the series. * p2p/simulations, p2p/testing: port to p2p/enode No surprises here, mostly replacements of discover.Node, discover.NodeID with their new equivalents. The 'interesting' API changes are: - testing.ProtocolSession tracks complete nodes, not just their IDs. - adapters.NodeConfig has a new method to create a complete node. These changes were needed to make swarm tests work. Note that the NodeID change makes the code incompatible with old simulation snapshots. * whisper/whisperv5, whisper/whisperv6: port to p2p/enode This port was easy because whisper uses []byte for node IDs and URL strings in the API. * eth: port to p2p/enode Again, easy to port because eth uses strings for node IDs and doesn't care about node information in any way. * les: port to p2p/enode Apart from replacing discover.NodeID with enode.ID, most changes are in the server pool code. It now deals with complete nodes instead of (Pubkey, IP, Port) triples. The database format is unchanged for now, but we should probably change it to use the node database later. * node: port to p2p/enode This change simply replaces discover.Node and discover.NodeID with their new equivalents. * swarm/network: port to p2p/enode Swarm has its own node address representation, BzzAddr, containing both an overlay address (the hash of a secp256k1 public key) and an underlay address (enode:// URL). There are no changes to the BzzAddr format in this commit, but certain operations such as creating a BzzAddr from a node ID are now impossible because node IDs aren't public keys anymore. Most swarm-related changes in the series remove uses of NewAddrFromNodeID, replacing it with NewAddr which takes a complete node as argument. ToOverlayAddr is removed because we can just use the node ID directly.
2018-09-24 17:59:00 -05:00
func (net *Network) DidSend(sender, receiver enode.ID, proto string, code uint64) error {
msg := &Msg{
One: sender,
Other: receiver,
Protocol: proto,
Code: code,
Received: false,
}
2018-05-08 06:08:43 -05:00
net.events.Send(NewEvent(msg))
return nil
}
// DidReceive tracks the fact that "receiver" received a message from "sender"
all: new p2p node representation (#17643) Package p2p/enode provides a generalized representation of p2p nodes which can contain arbitrary information in key/value pairs. It is also the new home for the node database. The "v4" identity scheme is also moved here from p2p/enr to remove the dependency on Ethereum crypto from that package. Record signature handling is changed significantly. The identity scheme registry is removed and acceptable schemes must be passed to any method that needs identity. This means records must now be validated explicitly after decoding. The enode API is designed to make signature handling easy and safe: most APIs around the codebase work with enode.Node, which is a wrapper around a valid record. Going from enr.Record to enode.Node requires a valid signature. * p2p/discover: port to p2p/enode This ports the discovery code to the new node representation in p2p/enode. The wire protocol is unchanged, this can be considered a refactoring change. The Kademlia table can now deal with nodes using an arbitrary identity scheme. This requires a few incompatible API changes: - Table.Lookup is not available anymore. It used to take a public key as argument because v4 protocol requires one. Its replacement is LookupRandom. - Table.Resolve takes *enode.Node instead of NodeID. This is also for v4 protocol compatibility because nodes cannot be looked up by ID alone. - Types Node and NodeID are gone. Further commits in the series will be fixes all over the the codebase to deal with those removals. * p2p: port to p2p/enode and discovery changes This adapts package p2p to the changes in p2p/discover. All uses of discover.Node and discover.NodeID are replaced by their equivalents from p2p/enode. New API is added to retrieve the enode.Node instance of a peer. The behavior of Server.Self with discovery disabled is improved. It now tries much harder to report a working IP address, falling back to 127.0.0.1 if no suitable address can be determined through other means. These changes were needed for tests of other packages later in the series. * p2p/simulations, p2p/testing: port to p2p/enode No surprises here, mostly replacements of discover.Node, discover.NodeID with their new equivalents. The 'interesting' API changes are: - testing.ProtocolSession tracks complete nodes, not just their IDs. - adapters.NodeConfig has a new method to create a complete node. These changes were needed to make swarm tests work. Note that the NodeID change makes the code incompatible with old simulation snapshots. * whisper/whisperv5, whisper/whisperv6: port to p2p/enode This port was easy because whisper uses []byte for node IDs and URL strings in the API. * eth: port to p2p/enode Again, easy to port because eth uses strings for node IDs and doesn't care about node information in any way. * les: port to p2p/enode Apart from replacing discover.NodeID with enode.ID, most changes are in the server pool code. It now deals with complete nodes instead of (Pubkey, IP, Port) triples. The database format is unchanged for now, but we should probably change it to use the node database later. * node: port to p2p/enode This change simply replaces discover.Node and discover.NodeID with their new equivalents. * swarm/network: port to p2p/enode Swarm has its own node address representation, BzzAddr, containing both an overlay address (the hash of a secp256k1 public key) and an underlay address (enode:// URL). There are no changes to the BzzAddr format in this commit, but certain operations such as creating a BzzAddr from a node ID are now impossible because node IDs aren't public keys anymore. Most swarm-related changes in the series remove uses of NewAddrFromNodeID, replacing it with NewAddr which takes a complete node as argument. ToOverlayAddr is removed because we can just use the node ID directly.
2018-09-24 17:59:00 -05:00
func (net *Network) DidReceive(sender, receiver enode.ID, proto string, code uint64) error {
msg := &Msg{
One: sender,
Other: receiver,
Protocol: proto,
Code: code,
Received: true,
}
2018-05-08 06:08:43 -05:00
net.events.Send(NewEvent(msg))
return nil
}
// GetNode gets the node with the given ID, returning nil if the node does not
// exist
all: new p2p node representation (#17643) Package p2p/enode provides a generalized representation of p2p nodes which can contain arbitrary information in key/value pairs. It is also the new home for the node database. The "v4" identity scheme is also moved here from p2p/enr to remove the dependency on Ethereum crypto from that package. Record signature handling is changed significantly. The identity scheme registry is removed and acceptable schemes must be passed to any method that needs identity. This means records must now be validated explicitly after decoding. The enode API is designed to make signature handling easy and safe: most APIs around the codebase work with enode.Node, which is a wrapper around a valid record. Going from enr.Record to enode.Node requires a valid signature. * p2p/discover: port to p2p/enode This ports the discovery code to the new node representation in p2p/enode. The wire protocol is unchanged, this can be considered a refactoring change. The Kademlia table can now deal with nodes using an arbitrary identity scheme. This requires a few incompatible API changes: - Table.Lookup is not available anymore. It used to take a public key as argument because v4 protocol requires one. Its replacement is LookupRandom. - Table.Resolve takes *enode.Node instead of NodeID. This is also for v4 protocol compatibility because nodes cannot be looked up by ID alone. - Types Node and NodeID are gone. Further commits in the series will be fixes all over the the codebase to deal with those removals. * p2p: port to p2p/enode and discovery changes This adapts package p2p to the changes in p2p/discover. All uses of discover.Node and discover.NodeID are replaced by their equivalents from p2p/enode. New API is added to retrieve the enode.Node instance of a peer. The behavior of Server.Self with discovery disabled is improved. It now tries much harder to report a working IP address, falling back to 127.0.0.1 if no suitable address can be determined through other means. These changes were needed for tests of other packages later in the series. * p2p/simulations, p2p/testing: port to p2p/enode No surprises here, mostly replacements of discover.Node, discover.NodeID with their new equivalents. The 'interesting' API changes are: - testing.ProtocolSession tracks complete nodes, not just their IDs. - adapters.NodeConfig has a new method to create a complete node. These changes were needed to make swarm tests work. Note that the NodeID change makes the code incompatible with old simulation snapshots. * whisper/whisperv5, whisper/whisperv6: port to p2p/enode This port was easy because whisper uses []byte for node IDs and URL strings in the API. * eth: port to p2p/enode Again, easy to port because eth uses strings for node IDs and doesn't care about node information in any way. * les: port to p2p/enode Apart from replacing discover.NodeID with enode.ID, most changes are in the server pool code. It now deals with complete nodes instead of (Pubkey, IP, Port) triples. The database format is unchanged for now, but we should probably change it to use the node database later. * node: port to p2p/enode This change simply replaces discover.Node and discover.NodeID with their new equivalents. * swarm/network: port to p2p/enode Swarm has its own node address representation, BzzAddr, containing both an overlay address (the hash of a secp256k1 public key) and an underlay address (enode:// URL). There are no changes to the BzzAddr format in this commit, but certain operations such as creating a BzzAddr from a node ID are now impossible because node IDs aren't public keys anymore. Most swarm-related changes in the series remove uses of NewAddrFromNodeID, replacing it with NewAddr which takes a complete node as argument. ToOverlayAddr is removed because we can just use the node ID directly.
2018-09-24 17:59:00 -05:00
func (net *Network) GetNode(id enode.ID) *Node {
net.lock.RLock()
defer net.lock.RUnlock()
2018-05-08 06:08:43 -05:00
return net.getNode(id)
}
p2p, swarm: fix node up races by granular locking (#18976) * swarm/network: DRY out repeated giga comment I not necessarily agree with the way we wait for event propagation. But I truly disagree with having duplicated giga comments. * p2p/simulations: encapsulate Node.Up field so we avoid data races The Node.Up field was accessed concurrently without "proper" locking. There was a lock on Network and that was used sometimes to access the field. Other times the locking was missed and we had a data race. For example: https://github.com/ethereum/go-ethereum/pull/18464 The case above was solved, but there were still intermittent/hard to reproduce races. So let's solve the issue permanently. resolves: ethersphere/go-ethereum#1146 * p2p/simulations: fix unmarshal of simulations.Node Making Node.Up field private in 13292ee897e345045fbfab3bda23a77589a271c1 broke TestHTTPNetwork and TestHTTPSnapshot. Because the default UnmarshalJSON does not handle unexported fields. Important: The fix is partial and not proper to my taste. But I cut scope as I think the fix may require a change to the current serialization format. New ticket: https://github.com/ethersphere/go-ethereum/issues/1177 * p2p/simulations: Add a sanity test case for Node.Config UnmarshalJSON * p2p/simulations: revert back to defer Unlock() pattern for Network It's a good patten to call `defer Unlock()` right after `Lock()` so (new) error cases won't miss to unlock. Let's get back to that pattern. The patten was abandoned in 85a79b3ad3c5863f8612d25c246bcfad339f36b7, while fixing a data race. That data race does not exist anymore, since the Node.Up field got hidden behind its own lock. * p2p/simulations: consistent naming for test providers Node.UnmarshalJSON * p2p/simulations: remove JSON annotation from private fields of Node As unexported fields are not serialized. * p2p/simulations: fix deadlock in Network.GetRandomDownNode() Problem: GetRandomDownNode() locks -> getDownNodeIDs() -> GetNodes() tries to lock -> deadlock On Network type, unexported functions must assume that `net.lock` is already acquired and should not call exported functions which might try to lock again. * p2p/simulations: ensure method conformity for Network Connect* methods were moved to p2p/simulations.Network from swarm/network/simulation. However these new methods did not follow the pattern of Network methods, i.e., all exported method locks the whole Network either for read or write. * p2p/simulations: fix deadlock during network shutdown `TestDiscoveryPersistenceSimulationSimAdapter` often got into deadlock. The execution was stuck on two locks, i.e, `Kademlia.lock` and `p2p/simulations.Network.lock`. Usually the test got stuck once in each 20 executions with high confidence. `Kademlia` was stuck in `Kademlia.EachAddr()` and `Network` in `Network.Stop()`. Solution: in `Network.Stop()` `net.lock` must be released before calling `node.Stop()` as stopping a node (somehow - I did not find the exact code path) causes `Network.InitConn()` to be called from `Kademlia.SuggestPeer()` and that blocks on `net.lock`. Related ticket: https://github.com/ethersphere/go-ethereum/issues/1223 * swarm/state: simplify if statement in DBStore.Put() * p2p/simulations: remove faulty godoc from private function The comment started with the wrong method name. The method is simple and self explanatory. Also, it's private. => Let's just remove the comment.
2019-02-18 00:38:14 -06:00
func (net *Network) getNode(id enode.ID) *Node {
i, found := net.nodeMap[id]
if !found {
return nil
}
return net.Nodes[i]
}
// GetNodeByName gets the node with the given name, returning nil if the node does
// not exist
2018-05-08 06:08:43 -05:00
func (net *Network) GetNodeByName(name string) *Node {
net.lock.RLock()
defer net.lock.RUnlock()
2018-05-08 06:08:43 -05:00
return net.getNodeByName(name)
}
func (net *Network) getNodeByName(name string) *Node {
for _, node := range net.Nodes {
if node.Config.Name == name {
return node
}
}
return nil
}
// GetNodeIDs returns the IDs of all existing nodes
// Nodes can optionally be excluded by specifying their enode.ID.
func (net *Network) GetNodeIDs(excludeIDs ...enode.ID) []enode.ID {
net.lock.RLock()
defer net.lock.RUnlock()
return net.getNodeIDs(excludeIDs)
}
func (net *Network) getNodeIDs(excludeIDs []enode.ID) []enode.ID {
// Get all current nodeIDs
nodeIDs := make([]enode.ID, 0, len(net.nodeMap))
for id := range net.nodeMap {
nodeIDs = append(nodeIDs, id)
}
if len(excludeIDs) > 0 {
// Return the difference of nodeIDs and excludeIDs
return filterIDs(nodeIDs, excludeIDs)
}
return nodeIDs
}
// GetNodes returns the existing nodes.
// Nodes can optionally be excluded by specifying their enode.ID.
func (net *Network) GetNodes(excludeIDs ...enode.ID) []*Node {
net.lock.RLock()
defer net.lock.RUnlock()
return net.getNodes(excludeIDs)
}
func (net *Network) getNodes(excludeIDs []enode.ID) []*Node {
if len(excludeIDs) > 0 {
nodeIDs := net.getNodeIDs(excludeIDs)
return net.getNodesByID(nodeIDs)
}
return net.Nodes
}
// GetNodesByID returns existing nodes with the given enode.IDs.
// If a node doesn't exist with a given enode.ID, it is ignored.
func (net *Network) GetNodesByID(nodeIDs []enode.ID) []*Node {
net.lock.RLock()
defer net.lock.RUnlock()
return net.getNodesByID(nodeIDs)
}
func (net *Network) getNodesByID(nodeIDs []enode.ID) []*Node {
nodes := make([]*Node, 0, len(nodeIDs))
for _, id := range nodeIDs {
node := net.getNode(id)
if node != nil {
nodes = append(nodes, node)
}
}
return nodes
}
// GetNodesByProperty returns existing nodes that have the given property string registered in their NodeConfig
func (net *Network) GetNodesByProperty(property string) []*Node {
net.lock.RLock()
defer net.lock.RUnlock()
return net.getNodesByProperty(property)
}
func (net *Network) getNodesByProperty(property string) []*Node {
nodes := make([]*Node, 0, len(net.propertyMap[property]))
for _, nodeIndex := range net.propertyMap[property] {
nodes = append(nodes, net.Nodes[nodeIndex])
}
p2p, swarm: fix node up races by granular locking (#18976) * swarm/network: DRY out repeated giga comment I not necessarily agree with the way we wait for event propagation. But I truly disagree with having duplicated giga comments. * p2p/simulations: encapsulate Node.Up field so we avoid data races The Node.Up field was accessed concurrently without "proper" locking. There was a lock on Network and that was used sometimes to access the field. Other times the locking was missed and we had a data race. For example: https://github.com/ethereum/go-ethereum/pull/18464 The case above was solved, but there were still intermittent/hard to reproduce races. So let's solve the issue permanently. resolves: ethersphere/go-ethereum#1146 * p2p/simulations: fix unmarshal of simulations.Node Making Node.Up field private in 13292ee897e345045fbfab3bda23a77589a271c1 broke TestHTTPNetwork and TestHTTPSnapshot. Because the default UnmarshalJSON does not handle unexported fields. Important: The fix is partial and not proper to my taste. But I cut scope as I think the fix may require a change to the current serialization format. New ticket: https://github.com/ethersphere/go-ethereum/issues/1177 * p2p/simulations: Add a sanity test case for Node.Config UnmarshalJSON * p2p/simulations: revert back to defer Unlock() pattern for Network It's a good patten to call `defer Unlock()` right after `Lock()` so (new) error cases won't miss to unlock. Let's get back to that pattern. The patten was abandoned in 85a79b3ad3c5863f8612d25c246bcfad339f36b7, while fixing a data race. That data race does not exist anymore, since the Node.Up field got hidden behind its own lock. * p2p/simulations: consistent naming for test providers Node.UnmarshalJSON * p2p/simulations: remove JSON annotation from private fields of Node As unexported fields are not serialized. * p2p/simulations: fix deadlock in Network.GetRandomDownNode() Problem: GetRandomDownNode() locks -> getDownNodeIDs() -> GetNodes() tries to lock -> deadlock On Network type, unexported functions must assume that `net.lock` is already acquired and should not call exported functions which might try to lock again. * p2p/simulations: ensure method conformity for Network Connect* methods were moved to p2p/simulations.Network from swarm/network/simulation. However these new methods did not follow the pattern of Network methods, i.e., all exported method locks the whole Network either for read or write. * p2p/simulations: fix deadlock during network shutdown `TestDiscoveryPersistenceSimulationSimAdapter` often got into deadlock. The execution was stuck on two locks, i.e, `Kademlia.lock` and `p2p/simulations.Network.lock`. Usually the test got stuck once in each 20 executions with high confidence. `Kademlia` was stuck in `Kademlia.EachAddr()` and `Network` in `Network.Stop()`. Solution: in `Network.Stop()` `net.lock` must be released before calling `node.Stop()` as stopping a node (somehow - I did not find the exact code path) causes `Network.InitConn()` to be called from `Kademlia.SuggestPeer()` and that blocks on `net.lock`. Related ticket: https://github.com/ethersphere/go-ethereum/issues/1223 * swarm/state: simplify if statement in DBStore.Put() * p2p/simulations: remove faulty godoc from private function The comment started with the wrong method name. The method is simple and self explanatory. Also, it's private. => Let's just remove the comment.
2019-02-18 00:38:14 -06:00
return nodes
}
// GetNodeIDsByProperty returns existing node's enode IDs that have the given property string registered in the NodeConfig
func (net *Network) GetNodeIDsByProperty(property string) []enode.ID {
net.lock.RLock()
defer net.lock.RUnlock()
return net.getNodeIDsByProperty(property)
}
func (net *Network) getNodeIDsByProperty(property string) []enode.ID {
nodeIDs := make([]enode.ID, 0, len(net.propertyMap[property]))
for _, nodeIndex := range net.propertyMap[property] {
node := net.Nodes[nodeIndex]
nodeIDs = append(nodeIDs, node.ID())
}
return nodeIDs
}
// GetRandomUpNode returns a random node on the network, which is running.
func (net *Network) GetRandomUpNode(excludeIDs ...enode.ID) *Node {
net.lock.RLock()
defer net.lock.RUnlock()
p2p, swarm: fix node up races by granular locking (#18976) * swarm/network: DRY out repeated giga comment I not necessarily agree with the way we wait for event propagation. But I truly disagree with having duplicated giga comments. * p2p/simulations: encapsulate Node.Up field so we avoid data races The Node.Up field was accessed concurrently without "proper" locking. There was a lock on Network and that was used sometimes to access the field. Other times the locking was missed and we had a data race. For example: https://github.com/ethereum/go-ethereum/pull/18464 The case above was solved, but there were still intermittent/hard to reproduce races. So let's solve the issue permanently. resolves: ethersphere/go-ethereum#1146 * p2p/simulations: fix unmarshal of simulations.Node Making Node.Up field private in 13292ee897e345045fbfab3bda23a77589a271c1 broke TestHTTPNetwork and TestHTTPSnapshot. Because the default UnmarshalJSON does not handle unexported fields. Important: The fix is partial and not proper to my taste. But I cut scope as I think the fix may require a change to the current serialization format. New ticket: https://github.com/ethersphere/go-ethereum/issues/1177 * p2p/simulations: Add a sanity test case for Node.Config UnmarshalJSON * p2p/simulations: revert back to defer Unlock() pattern for Network It's a good patten to call `defer Unlock()` right after `Lock()` so (new) error cases won't miss to unlock. Let's get back to that pattern. The patten was abandoned in 85a79b3ad3c5863f8612d25c246bcfad339f36b7, while fixing a data race. That data race does not exist anymore, since the Node.Up field got hidden behind its own lock. * p2p/simulations: consistent naming for test providers Node.UnmarshalJSON * p2p/simulations: remove JSON annotation from private fields of Node As unexported fields are not serialized. * p2p/simulations: fix deadlock in Network.GetRandomDownNode() Problem: GetRandomDownNode() locks -> getDownNodeIDs() -> GetNodes() tries to lock -> deadlock On Network type, unexported functions must assume that `net.lock` is already acquired and should not call exported functions which might try to lock again. * p2p/simulations: ensure method conformity for Network Connect* methods were moved to p2p/simulations.Network from swarm/network/simulation. However these new methods did not follow the pattern of Network methods, i.e., all exported method locks the whole Network either for read or write. * p2p/simulations: fix deadlock during network shutdown `TestDiscoveryPersistenceSimulationSimAdapter` often got into deadlock. The execution was stuck on two locks, i.e, `Kademlia.lock` and `p2p/simulations.Network.lock`. Usually the test got stuck once in each 20 executions with high confidence. `Kademlia` was stuck in `Kademlia.EachAddr()` and `Network` in `Network.Stop()`. Solution: in `Network.Stop()` `net.lock` must be released before calling `node.Stop()` as stopping a node (somehow - I did not find the exact code path) causes `Network.InitConn()` to be called from `Kademlia.SuggestPeer()` and that blocks on `net.lock`. Related ticket: https://github.com/ethersphere/go-ethereum/issues/1223 * swarm/state: simplify if statement in DBStore.Put() * p2p/simulations: remove faulty godoc from private function The comment started with the wrong method name. The method is simple and self explanatory. Also, it's private. => Let's just remove the comment.
2019-02-18 00:38:14 -06:00
return net.getRandomUpNode(excludeIDs...)
}
// GetRandomUpNode returns a random node on the network, which is running.
func (net *Network) getRandomUpNode(excludeIDs ...enode.ID) *Node {
return net.getRandomNode(net.getUpNodeIDs(), excludeIDs)
}
func (net *Network) getUpNodeIDs() (ids []enode.ID) {
2018-05-08 06:08:43 -05:00
for _, node := range net.Nodes {
p2p, swarm: fix node up races by granular locking (#18976) * swarm/network: DRY out repeated giga comment I not necessarily agree with the way we wait for event propagation. But I truly disagree with having duplicated giga comments. * p2p/simulations: encapsulate Node.Up field so we avoid data races The Node.Up field was accessed concurrently without "proper" locking. There was a lock on Network and that was used sometimes to access the field. Other times the locking was missed and we had a data race. For example: https://github.com/ethereum/go-ethereum/pull/18464 The case above was solved, but there were still intermittent/hard to reproduce races. So let's solve the issue permanently. resolves: ethersphere/go-ethereum#1146 * p2p/simulations: fix unmarshal of simulations.Node Making Node.Up field private in 13292ee897e345045fbfab3bda23a77589a271c1 broke TestHTTPNetwork and TestHTTPSnapshot. Because the default UnmarshalJSON does not handle unexported fields. Important: The fix is partial and not proper to my taste. But I cut scope as I think the fix may require a change to the current serialization format. New ticket: https://github.com/ethersphere/go-ethereum/issues/1177 * p2p/simulations: Add a sanity test case for Node.Config UnmarshalJSON * p2p/simulations: revert back to defer Unlock() pattern for Network It's a good patten to call `defer Unlock()` right after `Lock()` so (new) error cases won't miss to unlock. Let's get back to that pattern. The patten was abandoned in 85a79b3ad3c5863f8612d25c246bcfad339f36b7, while fixing a data race. That data race does not exist anymore, since the Node.Up field got hidden behind its own lock. * p2p/simulations: consistent naming for test providers Node.UnmarshalJSON * p2p/simulations: remove JSON annotation from private fields of Node As unexported fields are not serialized. * p2p/simulations: fix deadlock in Network.GetRandomDownNode() Problem: GetRandomDownNode() locks -> getDownNodeIDs() -> GetNodes() tries to lock -> deadlock On Network type, unexported functions must assume that `net.lock` is already acquired and should not call exported functions which might try to lock again. * p2p/simulations: ensure method conformity for Network Connect* methods were moved to p2p/simulations.Network from swarm/network/simulation. However these new methods did not follow the pattern of Network methods, i.e., all exported method locks the whole Network either for read or write. * p2p/simulations: fix deadlock during network shutdown `TestDiscoveryPersistenceSimulationSimAdapter` often got into deadlock. The execution was stuck on two locks, i.e, `Kademlia.lock` and `p2p/simulations.Network.lock`. Usually the test got stuck once in each 20 executions with high confidence. `Kademlia` was stuck in `Kademlia.EachAddr()` and `Network` in `Network.Stop()`. Solution: in `Network.Stop()` `net.lock` must be released before calling `node.Stop()` as stopping a node (somehow - I did not find the exact code path) causes `Network.InitConn()` to be called from `Kademlia.SuggestPeer()` and that blocks on `net.lock`. Related ticket: https://github.com/ethersphere/go-ethereum/issues/1223 * swarm/state: simplify if statement in DBStore.Put() * p2p/simulations: remove faulty godoc from private function The comment started with the wrong method name. The method is simple and self explanatory. Also, it's private. => Let's just remove the comment.
2019-02-18 00:38:14 -06:00
if node.Up() {
ids = append(ids, node.ID())
}
}
return ids
}
// GetRandomDownNode returns a random node on the network, which is stopped.
func (net *Network) GetRandomDownNode(excludeIDs ...enode.ID) *Node {
net.lock.RLock()
defer net.lock.RUnlock()
return net.getRandomNode(net.getDownNodeIDs(), excludeIDs)
}
func (net *Network) getDownNodeIDs() (ids []enode.ID) {
for _, node := range net.Nodes {
p2p, swarm: fix node up races by granular locking (#18976) * swarm/network: DRY out repeated giga comment I not necessarily agree with the way we wait for event propagation. But I truly disagree with having duplicated giga comments. * p2p/simulations: encapsulate Node.Up field so we avoid data races The Node.Up field was accessed concurrently without "proper" locking. There was a lock on Network and that was used sometimes to access the field. Other times the locking was missed and we had a data race. For example: https://github.com/ethereum/go-ethereum/pull/18464 The case above was solved, but there were still intermittent/hard to reproduce races. So let's solve the issue permanently. resolves: ethersphere/go-ethereum#1146 * p2p/simulations: fix unmarshal of simulations.Node Making Node.Up field private in 13292ee897e345045fbfab3bda23a77589a271c1 broke TestHTTPNetwork and TestHTTPSnapshot. Because the default UnmarshalJSON does not handle unexported fields. Important: The fix is partial and not proper to my taste. But I cut scope as I think the fix may require a change to the current serialization format. New ticket: https://github.com/ethersphere/go-ethereum/issues/1177 * p2p/simulations: Add a sanity test case for Node.Config UnmarshalJSON * p2p/simulations: revert back to defer Unlock() pattern for Network It's a good patten to call `defer Unlock()` right after `Lock()` so (new) error cases won't miss to unlock. Let's get back to that pattern. The patten was abandoned in 85a79b3ad3c5863f8612d25c246bcfad339f36b7, while fixing a data race. That data race does not exist anymore, since the Node.Up field got hidden behind its own lock. * p2p/simulations: consistent naming for test providers Node.UnmarshalJSON * p2p/simulations: remove JSON annotation from private fields of Node As unexported fields are not serialized. * p2p/simulations: fix deadlock in Network.GetRandomDownNode() Problem: GetRandomDownNode() locks -> getDownNodeIDs() -> GetNodes() tries to lock -> deadlock On Network type, unexported functions must assume that `net.lock` is already acquired and should not call exported functions which might try to lock again. * p2p/simulations: ensure method conformity for Network Connect* methods were moved to p2p/simulations.Network from swarm/network/simulation. However these new methods did not follow the pattern of Network methods, i.e., all exported method locks the whole Network either for read or write. * p2p/simulations: fix deadlock during network shutdown `TestDiscoveryPersistenceSimulationSimAdapter` often got into deadlock. The execution was stuck on two locks, i.e, `Kademlia.lock` and `p2p/simulations.Network.lock`. Usually the test got stuck once in each 20 executions with high confidence. `Kademlia` was stuck in `Kademlia.EachAddr()` and `Network` in `Network.Stop()`. Solution: in `Network.Stop()` `net.lock` must be released before calling `node.Stop()` as stopping a node (somehow - I did not find the exact code path) causes `Network.InitConn()` to be called from `Kademlia.SuggestPeer()` and that blocks on `net.lock`. Related ticket: https://github.com/ethersphere/go-ethereum/issues/1223 * swarm/state: simplify if statement in DBStore.Put() * p2p/simulations: remove faulty godoc from private function The comment started with the wrong method name. The method is simple and self explanatory. Also, it's private. => Let's just remove the comment.
2019-02-18 00:38:14 -06:00
if !node.Up() {
ids = append(ids, node.ID())
}
}
return ids
}
// GetRandomNode returns a random node on the network, regardless of whether it is running or not
func (net *Network) GetRandomNode(excludeIDs ...enode.ID) *Node {
net.lock.RLock()
defer net.lock.RUnlock()
return net.getRandomNode(net.getNodeIDs(nil), excludeIDs) // no need to exclude twice
}
func (net *Network) getRandomNode(ids []enode.ID, excludeIDs []enode.ID) *Node {
filtered := filterIDs(ids, excludeIDs)
l := len(filtered)
if l == 0 {
return nil
}
swarm: fix network/stream data races (#19051) * swarm/network/stream: newStreamerTester cleanup only if err is nil * swarm/network/stream: raise newStreamerTester waitForPeers timeout * swarm/network/stream: fix data races in GetPeerSubscriptions * swarm/storage: prevent data race on LDBStore.batchesC https://github.com/ethersphere/go-ethereum/issues/1198#issuecomment-461775049 * swarm/network/stream: fix TestGetSubscriptionsRPC data race https://github.com/ethersphere/go-ethereum/issues/1198#issuecomment-461768477 * swarm/network/stream: correctly use Simulation.Run callback https://github.com/ethersphere/go-ethereum/issues/1198#issuecomment-461783804 * swarm/network: protect addrCountC in Kademlia.AddrCountC function https://github.com/ethersphere/go-ethereum/issues/1198#issuecomment-462273444 * p2p/simulations: fix a deadlock calling getRandomNode with lock https://github.com/ethersphere/go-ethereum/issues/1198#issuecomment-462317407 * swarm/network/stream: terminate disconnect goruotines in tests * swarm/network/stream: reduce memory consumption when testing data races * swarm/network/stream: add watchDisconnections helper function * swarm/network/stream: add concurrent counter for tests * swarm/network/stream: rename race/norace test files and use const * swarm/network/stream: remove watchSim and its panic * swarm/network/stream: pass context in watchDisconnections * swarm/network/stream: add concurrent safe bool for watchDisconnections * swarm/storage: fix LDBStore.batchesC data race by not closing it
2019-02-13 06:03:23 -06:00
return net.getNode(filtered[rand.Intn(l)])
}
func filterIDs(ids []enode.ID, excludeIDs []enode.ID) []enode.ID {
exclude := make(map[enode.ID]bool)
for _, id := range excludeIDs {
exclude[id] = true
}
var filtered []enode.ID
for _, id := range ids {
if _, found := exclude[id]; !found {
filtered = append(filtered, id)
}
}
return filtered
}
// GetConn returns the connection which exists between "one" and "other"
// regardless of which node initiated the connection
all: new p2p node representation (#17643) Package p2p/enode provides a generalized representation of p2p nodes which can contain arbitrary information in key/value pairs. It is also the new home for the node database. The "v4" identity scheme is also moved here from p2p/enr to remove the dependency on Ethereum crypto from that package. Record signature handling is changed significantly. The identity scheme registry is removed and acceptable schemes must be passed to any method that needs identity. This means records must now be validated explicitly after decoding. The enode API is designed to make signature handling easy and safe: most APIs around the codebase work with enode.Node, which is a wrapper around a valid record. Going from enr.Record to enode.Node requires a valid signature. * p2p/discover: port to p2p/enode This ports the discovery code to the new node representation in p2p/enode. The wire protocol is unchanged, this can be considered a refactoring change. The Kademlia table can now deal with nodes using an arbitrary identity scheme. This requires a few incompatible API changes: - Table.Lookup is not available anymore. It used to take a public key as argument because v4 protocol requires one. Its replacement is LookupRandom. - Table.Resolve takes *enode.Node instead of NodeID. This is also for v4 protocol compatibility because nodes cannot be looked up by ID alone. - Types Node and NodeID are gone. Further commits in the series will be fixes all over the the codebase to deal with those removals. * p2p: port to p2p/enode and discovery changes This adapts package p2p to the changes in p2p/discover. All uses of discover.Node and discover.NodeID are replaced by their equivalents from p2p/enode. New API is added to retrieve the enode.Node instance of a peer. The behavior of Server.Self with discovery disabled is improved. It now tries much harder to report a working IP address, falling back to 127.0.0.1 if no suitable address can be determined through other means. These changes were needed for tests of other packages later in the series. * p2p/simulations, p2p/testing: port to p2p/enode No surprises here, mostly replacements of discover.Node, discover.NodeID with their new equivalents. The 'interesting' API changes are: - testing.ProtocolSession tracks complete nodes, not just their IDs. - adapters.NodeConfig has a new method to create a complete node. These changes were needed to make swarm tests work. Note that the NodeID change makes the code incompatible with old simulation snapshots. * whisper/whisperv5, whisper/whisperv6: port to p2p/enode This port was easy because whisper uses []byte for node IDs and URL strings in the API. * eth: port to p2p/enode Again, easy to port because eth uses strings for node IDs and doesn't care about node information in any way. * les: port to p2p/enode Apart from replacing discover.NodeID with enode.ID, most changes are in the server pool code. It now deals with complete nodes instead of (Pubkey, IP, Port) triples. The database format is unchanged for now, but we should probably change it to use the node database later. * node: port to p2p/enode This change simply replaces discover.Node and discover.NodeID with their new equivalents. * swarm/network: port to p2p/enode Swarm has its own node address representation, BzzAddr, containing both an overlay address (the hash of a secp256k1 public key) and an underlay address (enode:// URL). There are no changes to the BzzAddr format in this commit, but certain operations such as creating a BzzAddr from a node ID are now impossible because node IDs aren't public keys anymore. Most swarm-related changes in the series remove uses of NewAddrFromNodeID, replacing it with NewAddr which takes a complete node as argument. ToOverlayAddr is removed because we can just use the node ID directly.
2018-09-24 17:59:00 -05:00
func (net *Network) GetConn(oneID, otherID enode.ID) *Conn {
net.lock.RLock()
defer net.lock.RUnlock()
2018-05-08 06:08:43 -05:00
return net.getConn(oneID, otherID)
}
// GetOrCreateConn is like GetConn but creates the connection if it doesn't
// already exist
all: new p2p node representation (#17643) Package p2p/enode provides a generalized representation of p2p nodes which can contain arbitrary information in key/value pairs. It is also the new home for the node database. The "v4" identity scheme is also moved here from p2p/enr to remove the dependency on Ethereum crypto from that package. Record signature handling is changed significantly. The identity scheme registry is removed and acceptable schemes must be passed to any method that needs identity. This means records must now be validated explicitly after decoding. The enode API is designed to make signature handling easy and safe: most APIs around the codebase work with enode.Node, which is a wrapper around a valid record. Going from enr.Record to enode.Node requires a valid signature. * p2p/discover: port to p2p/enode This ports the discovery code to the new node representation in p2p/enode. The wire protocol is unchanged, this can be considered a refactoring change. The Kademlia table can now deal with nodes using an arbitrary identity scheme. This requires a few incompatible API changes: - Table.Lookup is not available anymore. It used to take a public key as argument because v4 protocol requires one. Its replacement is LookupRandom. - Table.Resolve takes *enode.Node instead of NodeID. This is also for v4 protocol compatibility because nodes cannot be looked up by ID alone. - Types Node and NodeID are gone. Further commits in the series will be fixes all over the the codebase to deal with those removals. * p2p: port to p2p/enode and discovery changes This adapts package p2p to the changes in p2p/discover. All uses of discover.Node and discover.NodeID are replaced by their equivalents from p2p/enode. New API is added to retrieve the enode.Node instance of a peer. The behavior of Server.Self with discovery disabled is improved. It now tries much harder to report a working IP address, falling back to 127.0.0.1 if no suitable address can be determined through other means. These changes were needed for tests of other packages later in the series. * p2p/simulations, p2p/testing: port to p2p/enode No surprises here, mostly replacements of discover.Node, discover.NodeID with their new equivalents. The 'interesting' API changes are: - testing.ProtocolSession tracks complete nodes, not just their IDs. - adapters.NodeConfig has a new method to create a complete node. These changes were needed to make swarm tests work. Note that the NodeID change makes the code incompatible with old simulation snapshots. * whisper/whisperv5, whisper/whisperv6: port to p2p/enode This port was easy because whisper uses []byte for node IDs and URL strings in the API. * eth: port to p2p/enode Again, easy to port because eth uses strings for node IDs and doesn't care about node information in any way. * les: port to p2p/enode Apart from replacing discover.NodeID with enode.ID, most changes are in the server pool code. It now deals with complete nodes instead of (Pubkey, IP, Port) triples. The database format is unchanged for now, but we should probably change it to use the node database later. * node: port to p2p/enode This change simply replaces discover.Node and discover.NodeID with their new equivalents. * swarm/network: port to p2p/enode Swarm has its own node address representation, BzzAddr, containing both an overlay address (the hash of a secp256k1 public key) and an underlay address (enode:// URL). There are no changes to the BzzAddr format in this commit, but certain operations such as creating a BzzAddr from a node ID are now impossible because node IDs aren't public keys anymore. Most swarm-related changes in the series remove uses of NewAddrFromNodeID, replacing it with NewAddr which takes a complete node as argument. ToOverlayAddr is removed because we can just use the node ID directly.
2018-09-24 17:59:00 -05:00
func (net *Network) GetOrCreateConn(oneID, otherID enode.ID) (*Conn, error) {
2018-05-08 06:08:43 -05:00
net.lock.Lock()
defer net.lock.Unlock()
return net.getOrCreateConn(oneID, otherID)
}
all: new p2p node representation (#17643) Package p2p/enode provides a generalized representation of p2p nodes which can contain arbitrary information in key/value pairs. It is also the new home for the node database. The "v4" identity scheme is also moved here from p2p/enr to remove the dependency on Ethereum crypto from that package. Record signature handling is changed significantly. The identity scheme registry is removed and acceptable schemes must be passed to any method that needs identity. This means records must now be validated explicitly after decoding. The enode API is designed to make signature handling easy and safe: most APIs around the codebase work with enode.Node, which is a wrapper around a valid record. Going from enr.Record to enode.Node requires a valid signature. * p2p/discover: port to p2p/enode This ports the discovery code to the new node representation in p2p/enode. The wire protocol is unchanged, this can be considered a refactoring change. The Kademlia table can now deal with nodes using an arbitrary identity scheme. This requires a few incompatible API changes: - Table.Lookup is not available anymore. It used to take a public key as argument because v4 protocol requires one. Its replacement is LookupRandom. - Table.Resolve takes *enode.Node instead of NodeID. This is also for v4 protocol compatibility because nodes cannot be looked up by ID alone. - Types Node and NodeID are gone. Further commits in the series will be fixes all over the the codebase to deal with those removals. * p2p: port to p2p/enode and discovery changes This adapts package p2p to the changes in p2p/discover. All uses of discover.Node and discover.NodeID are replaced by their equivalents from p2p/enode. New API is added to retrieve the enode.Node instance of a peer. The behavior of Server.Self with discovery disabled is improved. It now tries much harder to report a working IP address, falling back to 127.0.0.1 if no suitable address can be determined through other means. These changes were needed for tests of other packages later in the series. * p2p/simulations, p2p/testing: port to p2p/enode No surprises here, mostly replacements of discover.Node, discover.NodeID with their new equivalents. The 'interesting' API changes are: - testing.ProtocolSession tracks complete nodes, not just their IDs. - adapters.NodeConfig has a new method to create a complete node. These changes were needed to make swarm tests work. Note that the NodeID change makes the code incompatible with old simulation snapshots. * whisper/whisperv5, whisper/whisperv6: port to p2p/enode This port was easy because whisper uses []byte for node IDs and URL strings in the API. * eth: port to p2p/enode Again, easy to port because eth uses strings for node IDs and doesn't care about node information in any way. * les: port to p2p/enode Apart from replacing discover.NodeID with enode.ID, most changes are in the server pool code. It now deals with complete nodes instead of (Pubkey, IP, Port) triples. The database format is unchanged for now, but we should probably change it to use the node database later. * node: port to p2p/enode This change simply replaces discover.Node and discover.NodeID with their new equivalents. * swarm/network: port to p2p/enode Swarm has its own node address representation, BzzAddr, containing both an overlay address (the hash of a secp256k1 public key) and an underlay address (enode:// URL). There are no changes to the BzzAddr format in this commit, but certain operations such as creating a BzzAddr from a node ID are now impossible because node IDs aren't public keys anymore. Most swarm-related changes in the series remove uses of NewAddrFromNodeID, replacing it with NewAddr which takes a complete node as argument. ToOverlayAddr is removed because we can just use the node ID directly.
2018-09-24 17:59:00 -05:00
func (net *Network) getOrCreateConn(oneID, otherID enode.ID) (*Conn, error) {
2018-05-08 06:08:43 -05:00
if conn := net.getConn(oneID, otherID); conn != nil {
return conn, nil
}
2018-05-08 06:08:43 -05:00
one := net.getNode(oneID)
if one == nil {
return nil, fmt.Errorf("node %v does not exist", oneID)
}
2018-05-08 06:08:43 -05:00
other := net.getNode(otherID)
if other == nil {
return nil, fmt.Errorf("node %v does not exist", otherID)
}
conn := &Conn{
One: oneID,
Other: otherID,
one: one,
other: other,
}
label := ConnLabel(oneID, otherID)
2018-05-08 06:08:43 -05:00
net.connMap[label] = len(net.Conns)
net.Conns = append(net.Conns, conn)
return conn, nil
}
all: new p2p node representation (#17643) Package p2p/enode provides a generalized representation of p2p nodes which can contain arbitrary information in key/value pairs. It is also the new home for the node database. The "v4" identity scheme is also moved here from p2p/enr to remove the dependency on Ethereum crypto from that package. Record signature handling is changed significantly. The identity scheme registry is removed and acceptable schemes must be passed to any method that needs identity. This means records must now be validated explicitly after decoding. The enode API is designed to make signature handling easy and safe: most APIs around the codebase work with enode.Node, which is a wrapper around a valid record. Going from enr.Record to enode.Node requires a valid signature. * p2p/discover: port to p2p/enode This ports the discovery code to the new node representation in p2p/enode. The wire protocol is unchanged, this can be considered a refactoring change. The Kademlia table can now deal with nodes using an arbitrary identity scheme. This requires a few incompatible API changes: - Table.Lookup is not available anymore. It used to take a public key as argument because v4 protocol requires one. Its replacement is LookupRandom. - Table.Resolve takes *enode.Node instead of NodeID. This is also for v4 protocol compatibility because nodes cannot be looked up by ID alone. - Types Node and NodeID are gone. Further commits in the series will be fixes all over the the codebase to deal with those removals. * p2p: port to p2p/enode and discovery changes This adapts package p2p to the changes in p2p/discover. All uses of discover.Node and discover.NodeID are replaced by their equivalents from p2p/enode. New API is added to retrieve the enode.Node instance of a peer. The behavior of Server.Self with discovery disabled is improved. It now tries much harder to report a working IP address, falling back to 127.0.0.1 if no suitable address can be determined through other means. These changes were needed for tests of other packages later in the series. * p2p/simulations, p2p/testing: port to p2p/enode No surprises here, mostly replacements of discover.Node, discover.NodeID with their new equivalents. The 'interesting' API changes are: - testing.ProtocolSession tracks complete nodes, not just their IDs. - adapters.NodeConfig has a new method to create a complete node. These changes were needed to make swarm tests work. Note that the NodeID change makes the code incompatible with old simulation snapshots. * whisper/whisperv5, whisper/whisperv6: port to p2p/enode This port was easy because whisper uses []byte for node IDs and URL strings in the API. * eth: port to p2p/enode Again, easy to port because eth uses strings for node IDs and doesn't care about node information in any way. * les: port to p2p/enode Apart from replacing discover.NodeID with enode.ID, most changes are in the server pool code. It now deals with complete nodes instead of (Pubkey, IP, Port) triples. The database format is unchanged for now, but we should probably change it to use the node database later. * node: port to p2p/enode This change simply replaces discover.Node and discover.NodeID with their new equivalents. * swarm/network: port to p2p/enode Swarm has its own node address representation, BzzAddr, containing both an overlay address (the hash of a secp256k1 public key) and an underlay address (enode:// URL). There are no changes to the BzzAddr format in this commit, but certain operations such as creating a BzzAddr from a node ID are now impossible because node IDs aren't public keys anymore. Most swarm-related changes in the series remove uses of NewAddrFromNodeID, replacing it with NewAddr which takes a complete node as argument. ToOverlayAddr is removed because we can just use the node ID directly.
2018-09-24 17:59:00 -05:00
func (net *Network) getConn(oneID, otherID enode.ID) *Conn {
label := ConnLabel(oneID, otherID)
2018-05-08 06:08:43 -05:00
i, found := net.connMap[label]
if !found {
return nil
}
2018-05-08 06:08:43 -05:00
return net.Conns[i]
}
// InitConn(one, other) retrieves the connection model for the connection between
// peers one and other, or creates a new one if it does not exist
// the order of nodes does not matter, i.e., Conn(i,j) == Conn(j, i)
// it checks if the connection is already up, and if the nodes are running
// NOTE:
// it also checks whether there has been recent attempt to connect the peers
// this is cheating as the simulation is used as an oracle and know about
// remote peers attempt to connect to a node which will then not initiate the connection
all: new p2p node representation (#17643) Package p2p/enode provides a generalized representation of p2p nodes which can contain arbitrary information in key/value pairs. It is also the new home for the node database. The "v4" identity scheme is also moved here from p2p/enr to remove the dependency on Ethereum crypto from that package. Record signature handling is changed significantly. The identity scheme registry is removed and acceptable schemes must be passed to any method that needs identity. This means records must now be validated explicitly after decoding. The enode API is designed to make signature handling easy and safe: most APIs around the codebase work with enode.Node, which is a wrapper around a valid record. Going from enr.Record to enode.Node requires a valid signature. * p2p/discover: port to p2p/enode This ports the discovery code to the new node representation in p2p/enode. The wire protocol is unchanged, this can be considered a refactoring change. The Kademlia table can now deal with nodes using an arbitrary identity scheme. This requires a few incompatible API changes: - Table.Lookup is not available anymore. It used to take a public key as argument because v4 protocol requires one. Its replacement is LookupRandom. - Table.Resolve takes *enode.Node instead of NodeID. This is also for v4 protocol compatibility because nodes cannot be looked up by ID alone. - Types Node and NodeID are gone. Further commits in the series will be fixes all over the the codebase to deal with those removals. * p2p: port to p2p/enode and discovery changes This adapts package p2p to the changes in p2p/discover. All uses of discover.Node and discover.NodeID are replaced by their equivalents from p2p/enode. New API is added to retrieve the enode.Node instance of a peer. The behavior of Server.Self with discovery disabled is improved. It now tries much harder to report a working IP address, falling back to 127.0.0.1 if no suitable address can be determined through other means. These changes were needed for tests of other packages later in the series. * p2p/simulations, p2p/testing: port to p2p/enode No surprises here, mostly replacements of discover.Node, discover.NodeID with their new equivalents. The 'interesting' API changes are: - testing.ProtocolSession tracks complete nodes, not just their IDs. - adapters.NodeConfig has a new method to create a complete node. These changes were needed to make swarm tests work. Note that the NodeID change makes the code incompatible with old simulation snapshots. * whisper/whisperv5, whisper/whisperv6: port to p2p/enode This port was easy because whisper uses []byte for node IDs and URL strings in the API. * eth: port to p2p/enode Again, easy to port because eth uses strings for node IDs and doesn't care about node information in any way. * les: port to p2p/enode Apart from replacing discover.NodeID with enode.ID, most changes are in the server pool code. It now deals with complete nodes instead of (Pubkey, IP, Port) triples. The database format is unchanged for now, but we should probably change it to use the node database later. * node: port to p2p/enode This change simply replaces discover.Node and discover.NodeID with their new equivalents. * swarm/network: port to p2p/enode Swarm has its own node address representation, BzzAddr, containing both an overlay address (the hash of a secp256k1 public key) and an underlay address (enode:// URL). There are no changes to the BzzAddr format in this commit, but certain operations such as creating a BzzAddr from a node ID are now impossible because node IDs aren't public keys anymore. Most swarm-related changes in the series remove uses of NewAddrFromNodeID, replacing it with NewAddr which takes a complete node as argument. ToOverlayAddr is removed because we can just use the node ID directly.
2018-09-24 17:59:00 -05:00
func (net *Network) InitConn(oneID, otherID enode.ID) (*Conn, error) {
2018-05-08 06:08:43 -05:00
net.lock.Lock()
defer net.lock.Unlock()
p2p, swarm: fix node up races by granular locking (#18976) * swarm/network: DRY out repeated giga comment I not necessarily agree with the way we wait for event propagation. But I truly disagree with having duplicated giga comments. * p2p/simulations: encapsulate Node.Up field so we avoid data races The Node.Up field was accessed concurrently without "proper" locking. There was a lock on Network and that was used sometimes to access the field. Other times the locking was missed and we had a data race. For example: https://github.com/ethereum/go-ethereum/pull/18464 The case above was solved, but there were still intermittent/hard to reproduce races. So let's solve the issue permanently. resolves: ethersphere/go-ethereum#1146 * p2p/simulations: fix unmarshal of simulations.Node Making Node.Up field private in 13292ee897e345045fbfab3bda23a77589a271c1 broke TestHTTPNetwork and TestHTTPSnapshot. Because the default UnmarshalJSON does not handle unexported fields. Important: The fix is partial and not proper to my taste. But I cut scope as I think the fix may require a change to the current serialization format. New ticket: https://github.com/ethersphere/go-ethereum/issues/1177 * p2p/simulations: Add a sanity test case for Node.Config UnmarshalJSON * p2p/simulations: revert back to defer Unlock() pattern for Network It's a good patten to call `defer Unlock()` right after `Lock()` so (new) error cases won't miss to unlock. Let's get back to that pattern. The patten was abandoned in 85a79b3ad3c5863f8612d25c246bcfad339f36b7, while fixing a data race. That data race does not exist anymore, since the Node.Up field got hidden behind its own lock. * p2p/simulations: consistent naming for test providers Node.UnmarshalJSON * p2p/simulations: remove JSON annotation from private fields of Node As unexported fields are not serialized. * p2p/simulations: fix deadlock in Network.GetRandomDownNode() Problem: GetRandomDownNode() locks -> getDownNodeIDs() -> GetNodes() tries to lock -> deadlock On Network type, unexported functions must assume that `net.lock` is already acquired and should not call exported functions which might try to lock again. * p2p/simulations: ensure method conformity for Network Connect* methods were moved to p2p/simulations.Network from swarm/network/simulation. However these new methods did not follow the pattern of Network methods, i.e., all exported method locks the whole Network either for read or write. * p2p/simulations: fix deadlock during network shutdown `TestDiscoveryPersistenceSimulationSimAdapter` often got into deadlock. The execution was stuck on two locks, i.e, `Kademlia.lock` and `p2p/simulations.Network.lock`. Usually the test got stuck once in each 20 executions with high confidence. `Kademlia` was stuck in `Kademlia.EachAddr()` and `Network` in `Network.Stop()`. Solution: in `Network.Stop()` `net.lock` must be released before calling `node.Stop()` as stopping a node (somehow - I did not find the exact code path) causes `Network.InitConn()` to be called from `Kademlia.SuggestPeer()` and that blocks on `net.lock`. Related ticket: https://github.com/ethersphere/go-ethereum/issues/1223 * swarm/state: simplify if statement in DBStore.Put() * p2p/simulations: remove faulty godoc from private function The comment started with the wrong method name. The method is simple and self explanatory. Also, it's private. => Let's just remove the comment.
2019-02-18 00:38:14 -06:00
return net.initConn(oneID, otherID)
}
func (net *Network) initConn(oneID, otherID enode.ID) (*Conn, error) {
if oneID == otherID {
return nil, fmt.Errorf("refusing to connect to self %v", oneID)
}
2018-05-08 06:08:43 -05:00
conn, err := net.getOrCreateConn(oneID, otherID)
if err != nil {
return nil, err
}
if conn.Up {
return nil, fmt.Errorf("%v and %v already connected", oneID, otherID)
}
2018-06-20 07:06:27 -05:00
if time.Since(conn.initiated) < DialBanTimeout {
return nil, fmt.Errorf("connection between %v and %v recently attempted", oneID, otherID)
}
err = conn.nodesUp()
if err != nil {
log.Trace("Nodes not up", "err", err)
return nil, fmt.Errorf("nodes not up: %v", err)
}
log.Debug("Connection initiated", "id", oneID, "other", otherID)
conn.initiated = time.Now()
return conn, nil
}
// Shutdown stops all nodes in the network and closes the quit channel
2018-05-08 06:08:43 -05:00
func (net *Network) Shutdown() {
for _, node := range net.Nodes {
log.Debug("Stopping node", "id", node.ID())
if err := node.Stop(); err != nil {
log.Warn("Can't stop node", "id", node.ID(), "err", err)
}
}
2018-05-08 06:08:43 -05:00
close(net.quitc)
}
// Reset resets all network properties:
// empties the nodes and the connection list
2018-05-08 06:08:43 -05:00
func (net *Network) Reset() {
net.lock.Lock()
defer net.lock.Unlock()
//re-initialize the maps
2018-05-08 06:08:43 -05:00
net.connMap = make(map[string]int)
all: new p2p node representation (#17643) Package p2p/enode provides a generalized representation of p2p nodes which can contain arbitrary information in key/value pairs. It is also the new home for the node database. The "v4" identity scheme is also moved here from p2p/enr to remove the dependency on Ethereum crypto from that package. Record signature handling is changed significantly. The identity scheme registry is removed and acceptable schemes must be passed to any method that needs identity. This means records must now be validated explicitly after decoding. The enode API is designed to make signature handling easy and safe: most APIs around the codebase work with enode.Node, which is a wrapper around a valid record. Going from enr.Record to enode.Node requires a valid signature. * p2p/discover: port to p2p/enode This ports the discovery code to the new node representation in p2p/enode. The wire protocol is unchanged, this can be considered a refactoring change. The Kademlia table can now deal with nodes using an arbitrary identity scheme. This requires a few incompatible API changes: - Table.Lookup is not available anymore. It used to take a public key as argument because v4 protocol requires one. Its replacement is LookupRandom. - Table.Resolve takes *enode.Node instead of NodeID. This is also for v4 protocol compatibility because nodes cannot be looked up by ID alone. - Types Node and NodeID are gone. Further commits in the series will be fixes all over the the codebase to deal with those removals. * p2p: port to p2p/enode and discovery changes This adapts package p2p to the changes in p2p/discover. All uses of discover.Node and discover.NodeID are replaced by their equivalents from p2p/enode. New API is added to retrieve the enode.Node instance of a peer. The behavior of Server.Self with discovery disabled is improved. It now tries much harder to report a working IP address, falling back to 127.0.0.1 if no suitable address can be determined through other means. These changes were needed for tests of other packages later in the series. * p2p/simulations, p2p/testing: port to p2p/enode No surprises here, mostly replacements of discover.Node, discover.NodeID with their new equivalents. The 'interesting' API changes are: - testing.ProtocolSession tracks complete nodes, not just their IDs. - adapters.NodeConfig has a new method to create a complete node. These changes were needed to make swarm tests work. Note that the NodeID change makes the code incompatible with old simulation snapshots. * whisper/whisperv5, whisper/whisperv6: port to p2p/enode This port was easy because whisper uses []byte for node IDs and URL strings in the API. * eth: port to p2p/enode Again, easy to port because eth uses strings for node IDs and doesn't care about node information in any way. * les: port to p2p/enode Apart from replacing discover.NodeID with enode.ID, most changes are in the server pool code. It now deals with complete nodes instead of (Pubkey, IP, Port) triples. The database format is unchanged for now, but we should probably change it to use the node database later. * node: port to p2p/enode This change simply replaces discover.Node and discover.NodeID with their new equivalents. * swarm/network: port to p2p/enode Swarm has its own node address representation, BzzAddr, containing both an overlay address (the hash of a secp256k1 public key) and an underlay address (enode:// URL). There are no changes to the BzzAddr format in this commit, but certain operations such as creating a BzzAddr from a node ID are now impossible because node IDs aren't public keys anymore. Most swarm-related changes in the series remove uses of NewAddrFromNodeID, replacing it with NewAddr which takes a complete node as argument. ToOverlayAddr is removed because we can just use the node ID directly.
2018-09-24 17:59:00 -05:00
net.nodeMap = make(map[enode.ID]int)
net.propertyMap = make(map[string][]int)
2018-05-08 06:08:43 -05:00
net.Nodes = nil
net.Conns = nil
}
// Node is a wrapper around adapters.Node which is used to track the status
// of a node in the network
type Node struct {
adapters.Node `json:"-"`
// Config if the config used to created the node
Config *adapters.NodeConfig `json:"config"`
p2p, swarm: fix node up races by granular locking (#18976) * swarm/network: DRY out repeated giga comment I not necessarily agree with the way we wait for event propagation. But I truly disagree with having duplicated giga comments. * p2p/simulations: encapsulate Node.Up field so we avoid data races The Node.Up field was accessed concurrently without "proper" locking. There was a lock on Network and that was used sometimes to access the field. Other times the locking was missed and we had a data race. For example: https://github.com/ethereum/go-ethereum/pull/18464 The case above was solved, but there were still intermittent/hard to reproduce races. So let's solve the issue permanently. resolves: ethersphere/go-ethereum#1146 * p2p/simulations: fix unmarshal of simulations.Node Making Node.Up field private in 13292ee897e345045fbfab3bda23a77589a271c1 broke TestHTTPNetwork and TestHTTPSnapshot. Because the default UnmarshalJSON does not handle unexported fields. Important: The fix is partial and not proper to my taste. But I cut scope as I think the fix may require a change to the current serialization format. New ticket: https://github.com/ethersphere/go-ethereum/issues/1177 * p2p/simulations: Add a sanity test case for Node.Config UnmarshalJSON * p2p/simulations: revert back to defer Unlock() pattern for Network It's a good patten to call `defer Unlock()` right after `Lock()` so (new) error cases won't miss to unlock. Let's get back to that pattern. The patten was abandoned in 85a79b3ad3c5863f8612d25c246bcfad339f36b7, while fixing a data race. That data race does not exist anymore, since the Node.Up field got hidden behind its own lock. * p2p/simulations: consistent naming for test providers Node.UnmarshalJSON * p2p/simulations: remove JSON annotation from private fields of Node As unexported fields are not serialized. * p2p/simulations: fix deadlock in Network.GetRandomDownNode() Problem: GetRandomDownNode() locks -> getDownNodeIDs() -> GetNodes() tries to lock -> deadlock On Network type, unexported functions must assume that `net.lock` is already acquired and should not call exported functions which might try to lock again. * p2p/simulations: ensure method conformity for Network Connect* methods were moved to p2p/simulations.Network from swarm/network/simulation. However these new methods did not follow the pattern of Network methods, i.e., all exported method locks the whole Network either for read or write. * p2p/simulations: fix deadlock during network shutdown `TestDiscoveryPersistenceSimulationSimAdapter` often got into deadlock. The execution was stuck on two locks, i.e, `Kademlia.lock` and `p2p/simulations.Network.lock`. Usually the test got stuck once in each 20 executions with high confidence. `Kademlia` was stuck in `Kademlia.EachAddr()` and `Network` in `Network.Stop()`. Solution: in `Network.Stop()` `net.lock` must be released before calling `node.Stop()` as stopping a node (somehow - I did not find the exact code path) causes `Network.InitConn()` to be called from `Kademlia.SuggestPeer()` and that blocks on `net.lock`. Related ticket: https://github.com/ethersphere/go-ethereum/issues/1223 * swarm/state: simplify if statement in DBStore.Put() * p2p/simulations: remove faulty godoc from private function The comment started with the wrong method name. The method is simple and self explanatory. Also, it's private. => Let's just remove the comment.
2019-02-18 00:38:14 -06:00
// up tracks whether or not the node is running
up bool
upMu *sync.RWMutex
}
func newNode(an adapters.Node, ac *adapters.NodeConfig, up bool) *Node {
return &Node{Node: an, Config: ac, up: up, upMu: new(sync.RWMutex)}
}
func (n *Node) copy() *Node {
configCpy := *n.Config
return newNode(n.Node, &configCpy, n.Up())
p2p, swarm: fix node up races by granular locking (#18976) * swarm/network: DRY out repeated giga comment I not necessarily agree with the way we wait for event propagation. But I truly disagree with having duplicated giga comments. * p2p/simulations: encapsulate Node.Up field so we avoid data races The Node.Up field was accessed concurrently without "proper" locking. There was a lock on Network and that was used sometimes to access the field. Other times the locking was missed and we had a data race. For example: https://github.com/ethereum/go-ethereum/pull/18464 The case above was solved, but there were still intermittent/hard to reproduce races. So let's solve the issue permanently. resolves: ethersphere/go-ethereum#1146 * p2p/simulations: fix unmarshal of simulations.Node Making Node.Up field private in 13292ee897e345045fbfab3bda23a77589a271c1 broke TestHTTPNetwork and TestHTTPSnapshot. Because the default UnmarshalJSON does not handle unexported fields. Important: The fix is partial and not proper to my taste. But I cut scope as I think the fix may require a change to the current serialization format. New ticket: https://github.com/ethersphere/go-ethereum/issues/1177 * p2p/simulations: Add a sanity test case for Node.Config UnmarshalJSON * p2p/simulations: revert back to defer Unlock() pattern for Network It's a good patten to call `defer Unlock()` right after `Lock()` so (new) error cases won't miss to unlock. Let's get back to that pattern. The patten was abandoned in 85a79b3ad3c5863f8612d25c246bcfad339f36b7, while fixing a data race. That data race does not exist anymore, since the Node.Up field got hidden behind its own lock. * p2p/simulations: consistent naming for test providers Node.UnmarshalJSON * p2p/simulations: remove JSON annotation from private fields of Node As unexported fields are not serialized. * p2p/simulations: fix deadlock in Network.GetRandomDownNode() Problem: GetRandomDownNode() locks -> getDownNodeIDs() -> GetNodes() tries to lock -> deadlock On Network type, unexported functions must assume that `net.lock` is already acquired and should not call exported functions which might try to lock again. * p2p/simulations: ensure method conformity for Network Connect* methods were moved to p2p/simulations.Network from swarm/network/simulation. However these new methods did not follow the pattern of Network methods, i.e., all exported method locks the whole Network either for read or write. * p2p/simulations: fix deadlock during network shutdown `TestDiscoveryPersistenceSimulationSimAdapter` often got into deadlock. The execution was stuck on two locks, i.e, `Kademlia.lock` and `p2p/simulations.Network.lock`. Usually the test got stuck once in each 20 executions with high confidence. `Kademlia` was stuck in `Kademlia.EachAddr()` and `Network` in `Network.Stop()`. Solution: in `Network.Stop()` `net.lock` must be released before calling `node.Stop()` as stopping a node (somehow - I did not find the exact code path) causes `Network.InitConn()` to be called from `Kademlia.SuggestPeer()` and that blocks on `net.lock`. Related ticket: https://github.com/ethersphere/go-ethereum/issues/1223 * swarm/state: simplify if statement in DBStore.Put() * p2p/simulations: remove faulty godoc from private function The comment started with the wrong method name. The method is simple and self explanatory. Also, it's private. => Let's just remove the comment.
2019-02-18 00:38:14 -06:00
}
// Up returns whether the node is currently up (online)
p2p, swarm: fix node up races by granular locking (#18976) * swarm/network: DRY out repeated giga comment I not necessarily agree with the way we wait for event propagation. But I truly disagree with having duplicated giga comments. * p2p/simulations: encapsulate Node.Up field so we avoid data races The Node.Up field was accessed concurrently without "proper" locking. There was a lock on Network and that was used sometimes to access the field. Other times the locking was missed and we had a data race. For example: https://github.com/ethereum/go-ethereum/pull/18464 The case above was solved, but there were still intermittent/hard to reproduce races. So let's solve the issue permanently. resolves: ethersphere/go-ethereum#1146 * p2p/simulations: fix unmarshal of simulations.Node Making Node.Up field private in 13292ee897e345045fbfab3bda23a77589a271c1 broke TestHTTPNetwork and TestHTTPSnapshot. Because the default UnmarshalJSON does not handle unexported fields. Important: The fix is partial and not proper to my taste. But I cut scope as I think the fix may require a change to the current serialization format. New ticket: https://github.com/ethersphere/go-ethereum/issues/1177 * p2p/simulations: Add a sanity test case for Node.Config UnmarshalJSON * p2p/simulations: revert back to defer Unlock() pattern for Network It's a good patten to call `defer Unlock()` right after `Lock()` so (new) error cases won't miss to unlock. Let's get back to that pattern. The patten was abandoned in 85a79b3ad3c5863f8612d25c246bcfad339f36b7, while fixing a data race. That data race does not exist anymore, since the Node.Up field got hidden behind its own lock. * p2p/simulations: consistent naming for test providers Node.UnmarshalJSON * p2p/simulations: remove JSON annotation from private fields of Node As unexported fields are not serialized. * p2p/simulations: fix deadlock in Network.GetRandomDownNode() Problem: GetRandomDownNode() locks -> getDownNodeIDs() -> GetNodes() tries to lock -> deadlock On Network type, unexported functions must assume that `net.lock` is already acquired and should not call exported functions which might try to lock again. * p2p/simulations: ensure method conformity for Network Connect* methods were moved to p2p/simulations.Network from swarm/network/simulation. However these new methods did not follow the pattern of Network methods, i.e., all exported method locks the whole Network either for read or write. * p2p/simulations: fix deadlock during network shutdown `TestDiscoveryPersistenceSimulationSimAdapter` often got into deadlock. The execution was stuck on two locks, i.e, `Kademlia.lock` and `p2p/simulations.Network.lock`. Usually the test got stuck once in each 20 executions with high confidence. `Kademlia` was stuck in `Kademlia.EachAddr()` and `Network` in `Network.Stop()`. Solution: in `Network.Stop()` `net.lock` must be released before calling `node.Stop()` as stopping a node (somehow - I did not find the exact code path) causes `Network.InitConn()` to be called from `Kademlia.SuggestPeer()` and that blocks on `net.lock`. Related ticket: https://github.com/ethersphere/go-ethereum/issues/1223 * swarm/state: simplify if statement in DBStore.Put() * p2p/simulations: remove faulty godoc from private function The comment started with the wrong method name. The method is simple and self explanatory. Also, it's private. => Let's just remove the comment.
2019-02-18 00:38:14 -06:00
func (n *Node) Up() bool {
n.upMu.RLock()
defer n.upMu.RUnlock()
return n.up
}
// SetUp sets the up (online) status of the nodes with the given value
p2p, swarm: fix node up races by granular locking (#18976) * swarm/network: DRY out repeated giga comment I not necessarily agree with the way we wait for event propagation. But I truly disagree with having duplicated giga comments. * p2p/simulations: encapsulate Node.Up field so we avoid data races The Node.Up field was accessed concurrently without "proper" locking. There was a lock on Network and that was used sometimes to access the field. Other times the locking was missed and we had a data race. For example: https://github.com/ethereum/go-ethereum/pull/18464 The case above was solved, but there were still intermittent/hard to reproduce races. So let's solve the issue permanently. resolves: ethersphere/go-ethereum#1146 * p2p/simulations: fix unmarshal of simulations.Node Making Node.Up field private in 13292ee897e345045fbfab3bda23a77589a271c1 broke TestHTTPNetwork and TestHTTPSnapshot. Because the default UnmarshalJSON does not handle unexported fields. Important: The fix is partial and not proper to my taste. But I cut scope as I think the fix may require a change to the current serialization format. New ticket: https://github.com/ethersphere/go-ethereum/issues/1177 * p2p/simulations: Add a sanity test case for Node.Config UnmarshalJSON * p2p/simulations: revert back to defer Unlock() pattern for Network It's a good patten to call `defer Unlock()` right after `Lock()` so (new) error cases won't miss to unlock. Let's get back to that pattern. The patten was abandoned in 85a79b3ad3c5863f8612d25c246bcfad339f36b7, while fixing a data race. That data race does not exist anymore, since the Node.Up field got hidden behind its own lock. * p2p/simulations: consistent naming for test providers Node.UnmarshalJSON * p2p/simulations: remove JSON annotation from private fields of Node As unexported fields are not serialized. * p2p/simulations: fix deadlock in Network.GetRandomDownNode() Problem: GetRandomDownNode() locks -> getDownNodeIDs() -> GetNodes() tries to lock -> deadlock On Network type, unexported functions must assume that `net.lock` is already acquired and should not call exported functions which might try to lock again. * p2p/simulations: ensure method conformity for Network Connect* methods were moved to p2p/simulations.Network from swarm/network/simulation. However these new methods did not follow the pattern of Network methods, i.e., all exported method locks the whole Network either for read or write. * p2p/simulations: fix deadlock during network shutdown `TestDiscoveryPersistenceSimulationSimAdapter` often got into deadlock. The execution was stuck on two locks, i.e, `Kademlia.lock` and `p2p/simulations.Network.lock`. Usually the test got stuck once in each 20 executions with high confidence. `Kademlia` was stuck in `Kademlia.EachAddr()` and `Network` in `Network.Stop()`. Solution: in `Network.Stop()` `net.lock` must be released before calling `node.Stop()` as stopping a node (somehow - I did not find the exact code path) causes `Network.InitConn()` to be called from `Kademlia.SuggestPeer()` and that blocks on `net.lock`. Related ticket: https://github.com/ethersphere/go-ethereum/issues/1223 * swarm/state: simplify if statement in DBStore.Put() * p2p/simulations: remove faulty godoc from private function The comment started with the wrong method name. The method is simple and self explanatory. Also, it's private. => Let's just remove the comment.
2019-02-18 00:38:14 -06:00
func (n *Node) SetUp(up bool) {
n.upMu.Lock()
defer n.upMu.Unlock()
n.up = up
}
// ID returns the ID of the node
all: new p2p node representation (#17643) Package p2p/enode provides a generalized representation of p2p nodes which can contain arbitrary information in key/value pairs. It is also the new home for the node database. The "v4" identity scheme is also moved here from p2p/enr to remove the dependency on Ethereum crypto from that package. Record signature handling is changed significantly. The identity scheme registry is removed and acceptable schemes must be passed to any method that needs identity. This means records must now be validated explicitly after decoding. The enode API is designed to make signature handling easy and safe: most APIs around the codebase work with enode.Node, which is a wrapper around a valid record. Going from enr.Record to enode.Node requires a valid signature. * p2p/discover: port to p2p/enode This ports the discovery code to the new node representation in p2p/enode. The wire protocol is unchanged, this can be considered a refactoring change. The Kademlia table can now deal with nodes using an arbitrary identity scheme. This requires a few incompatible API changes: - Table.Lookup is not available anymore. It used to take a public key as argument because v4 protocol requires one. Its replacement is LookupRandom. - Table.Resolve takes *enode.Node instead of NodeID. This is also for v4 protocol compatibility because nodes cannot be looked up by ID alone. - Types Node and NodeID are gone. Further commits in the series will be fixes all over the the codebase to deal with those removals. * p2p: port to p2p/enode and discovery changes This adapts package p2p to the changes in p2p/discover. All uses of discover.Node and discover.NodeID are replaced by their equivalents from p2p/enode. New API is added to retrieve the enode.Node instance of a peer. The behavior of Server.Self with discovery disabled is improved. It now tries much harder to report a working IP address, falling back to 127.0.0.1 if no suitable address can be determined through other means. These changes were needed for tests of other packages later in the series. * p2p/simulations, p2p/testing: port to p2p/enode No surprises here, mostly replacements of discover.Node, discover.NodeID with their new equivalents. The 'interesting' API changes are: - testing.ProtocolSession tracks complete nodes, not just their IDs. - adapters.NodeConfig has a new method to create a complete node. These changes were needed to make swarm tests work. Note that the NodeID change makes the code incompatible with old simulation snapshots. * whisper/whisperv5, whisper/whisperv6: port to p2p/enode This port was easy because whisper uses []byte for node IDs and URL strings in the API. * eth: port to p2p/enode Again, easy to port because eth uses strings for node IDs and doesn't care about node information in any way. * les: port to p2p/enode Apart from replacing discover.NodeID with enode.ID, most changes are in the server pool code. It now deals with complete nodes instead of (Pubkey, IP, Port) triples. The database format is unchanged for now, but we should probably change it to use the node database later. * node: port to p2p/enode This change simply replaces discover.Node and discover.NodeID with their new equivalents. * swarm/network: port to p2p/enode Swarm has its own node address representation, BzzAddr, containing both an overlay address (the hash of a secp256k1 public key) and an underlay address (enode:// URL). There are no changes to the BzzAddr format in this commit, but certain operations such as creating a BzzAddr from a node ID are now impossible because node IDs aren't public keys anymore. Most swarm-related changes in the series remove uses of NewAddrFromNodeID, replacing it with NewAddr which takes a complete node as argument. ToOverlayAddr is removed because we can just use the node ID directly.
2018-09-24 17:59:00 -05:00
func (n *Node) ID() enode.ID {
2018-05-08 06:08:43 -05:00
return n.Config.ID
}
// String returns a log-friendly string
2018-05-08 06:08:43 -05:00
func (n *Node) String() string {
return fmt.Sprintf("Node %v", n.ID().TerminalString())
}
// NodeInfo returns information about the node
2018-05-08 06:08:43 -05:00
func (n *Node) NodeInfo() *p2p.NodeInfo {
// avoid a panic if the node is not started yet
2018-05-08 06:08:43 -05:00
if n.Node == nil {
return nil
}
2018-05-08 06:08:43 -05:00
info := n.Node.NodeInfo()
info.Name = n.Config.Name
return info
}
// MarshalJSON implements the json.Marshaler interface so that the encoded
// JSON includes the NodeInfo
2018-05-08 06:08:43 -05:00
func (n *Node) MarshalJSON() ([]byte, error) {
return json.Marshal(struct {
Info *p2p.NodeInfo `json:"info,omitempty"`
Config *adapters.NodeConfig `json:"config,omitempty"`
Up bool `json:"up"`
}{
2018-05-08 06:08:43 -05:00
Info: n.NodeInfo(),
Config: n.Config,
p2p, swarm: fix node up races by granular locking (#18976) * swarm/network: DRY out repeated giga comment I not necessarily agree with the way we wait for event propagation. But I truly disagree with having duplicated giga comments. * p2p/simulations: encapsulate Node.Up field so we avoid data races The Node.Up field was accessed concurrently without "proper" locking. There was a lock on Network and that was used sometimes to access the field. Other times the locking was missed and we had a data race. For example: https://github.com/ethereum/go-ethereum/pull/18464 The case above was solved, but there were still intermittent/hard to reproduce races. So let's solve the issue permanently. resolves: ethersphere/go-ethereum#1146 * p2p/simulations: fix unmarshal of simulations.Node Making Node.Up field private in 13292ee897e345045fbfab3bda23a77589a271c1 broke TestHTTPNetwork and TestHTTPSnapshot. Because the default UnmarshalJSON does not handle unexported fields. Important: The fix is partial and not proper to my taste. But I cut scope as I think the fix may require a change to the current serialization format. New ticket: https://github.com/ethersphere/go-ethereum/issues/1177 * p2p/simulations: Add a sanity test case for Node.Config UnmarshalJSON * p2p/simulations: revert back to defer Unlock() pattern for Network It's a good patten to call `defer Unlock()` right after `Lock()` so (new) error cases won't miss to unlock. Let's get back to that pattern. The patten was abandoned in 85a79b3ad3c5863f8612d25c246bcfad339f36b7, while fixing a data race. That data race does not exist anymore, since the Node.Up field got hidden behind its own lock. * p2p/simulations: consistent naming for test providers Node.UnmarshalJSON * p2p/simulations: remove JSON annotation from private fields of Node As unexported fields are not serialized. * p2p/simulations: fix deadlock in Network.GetRandomDownNode() Problem: GetRandomDownNode() locks -> getDownNodeIDs() -> GetNodes() tries to lock -> deadlock On Network type, unexported functions must assume that `net.lock` is already acquired and should not call exported functions which might try to lock again. * p2p/simulations: ensure method conformity for Network Connect* methods were moved to p2p/simulations.Network from swarm/network/simulation. However these new methods did not follow the pattern of Network methods, i.e., all exported method locks the whole Network either for read or write. * p2p/simulations: fix deadlock during network shutdown `TestDiscoveryPersistenceSimulationSimAdapter` often got into deadlock. The execution was stuck on two locks, i.e, `Kademlia.lock` and `p2p/simulations.Network.lock`. Usually the test got stuck once in each 20 executions with high confidence. `Kademlia` was stuck in `Kademlia.EachAddr()` and `Network` in `Network.Stop()`. Solution: in `Network.Stop()` `net.lock` must be released before calling `node.Stop()` as stopping a node (somehow - I did not find the exact code path) causes `Network.InitConn()` to be called from `Kademlia.SuggestPeer()` and that blocks on `net.lock`. Related ticket: https://github.com/ethersphere/go-ethereum/issues/1223 * swarm/state: simplify if statement in DBStore.Put() * p2p/simulations: remove faulty godoc from private function The comment started with the wrong method name. The method is simple and self explanatory. Also, it's private. => Let's just remove the comment.
2019-02-18 00:38:14 -06:00
Up: n.Up(),
})
}
// UnmarshalJSON implements json.Unmarshaler interface so that we don't lose Node.up
// status. IMPORTANT: The implementation is incomplete; we lose p2p.NodeInfo.
p2p, swarm: fix node up races by granular locking (#18976) * swarm/network: DRY out repeated giga comment I not necessarily agree with the way we wait for event propagation. But I truly disagree with having duplicated giga comments. * p2p/simulations: encapsulate Node.Up field so we avoid data races The Node.Up field was accessed concurrently without "proper" locking. There was a lock on Network and that was used sometimes to access the field. Other times the locking was missed and we had a data race. For example: https://github.com/ethereum/go-ethereum/pull/18464 The case above was solved, but there were still intermittent/hard to reproduce races. So let's solve the issue permanently. resolves: ethersphere/go-ethereum#1146 * p2p/simulations: fix unmarshal of simulations.Node Making Node.Up field private in 13292ee897e345045fbfab3bda23a77589a271c1 broke TestHTTPNetwork and TestHTTPSnapshot. Because the default UnmarshalJSON does not handle unexported fields. Important: The fix is partial and not proper to my taste. But I cut scope as I think the fix may require a change to the current serialization format. New ticket: https://github.com/ethersphere/go-ethereum/issues/1177 * p2p/simulations: Add a sanity test case for Node.Config UnmarshalJSON * p2p/simulations: revert back to defer Unlock() pattern for Network It's a good patten to call `defer Unlock()` right after `Lock()` so (new) error cases won't miss to unlock. Let's get back to that pattern. The patten was abandoned in 85a79b3ad3c5863f8612d25c246bcfad339f36b7, while fixing a data race. That data race does not exist anymore, since the Node.Up field got hidden behind its own lock. * p2p/simulations: consistent naming for test providers Node.UnmarshalJSON * p2p/simulations: remove JSON annotation from private fields of Node As unexported fields are not serialized. * p2p/simulations: fix deadlock in Network.GetRandomDownNode() Problem: GetRandomDownNode() locks -> getDownNodeIDs() -> GetNodes() tries to lock -> deadlock On Network type, unexported functions must assume that `net.lock` is already acquired and should not call exported functions which might try to lock again. * p2p/simulations: ensure method conformity for Network Connect* methods were moved to p2p/simulations.Network from swarm/network/simulation. However these new methods did not follow the pattern of Network methods, i.e., all exported method locks the whole Network either for read or write. * p2p/simulations: fix deadlock during network shutdown `TestDiscoveryPersistenceSimulationSimAdapter` often got into deadlock. The execution was stuck on two locks, i.e, `Kademlia.lock` and `p2p/simulations.Network.lock`. Usually the test got stuck once in each 20 executions with high confidence. `Kademlia` was stuck in `Kademlia.EachAddr()` and `Network` in `Network.Stop()`. Solution: in `Network.Stop()` `net.lock` must be released before calling `node.Stop()` as stopping a node (somehow - I did not find the exact code path) causes `Network.InitConn()` to be called from `Kademlia.SuggestPeer()` and that blocks on `net.lock`. Related ticket: https://github.com/ethersphere/go-ethereum/issues/1223 * swarm/state: simplify if statement in DBStore.Put() * p2p/simulations: remove faulty godoc from private function The comment started with the wrong method name. The method is simple and self explanatory. Also, it's private. => Let's just remove the comment.
2019-02-18 00:38:14 -06:00
func (n *Node) UnmarshalJSON(raw []byte) error {
// TODO: How should we turn back NodeInfo into n.Node?
// Ticket: https://github.com/ethersphere/go-ethereum/issues/1177
var node struct {
p2p, swarm: fix node up races by granular locking (#18976) * swarm/network: DRY out repeated giga comment I not necessarily agree with the way we wait for event propagation. But I truly disagree with having duplicated giga comments. * p2p/simulations: encapsulate Node.Up field so we avoid data races The Node.Up field was accessed concurrently without "proper" locking. There was a lock on Network and that was used sometimes to access the field. Other times the locking was missed and we had a data race. For example: https://github.com/ethereum/go-ethereum/pull/18464 The case above was solved, but there were still intermittent/hard to reproduce races. So let's solve the issue permanently. resolves: ethersphere/go-ethereum#1146 * p2p/simulations: fix unmarshal of simulations.Node Making Node.Up field private in 13292ee897e345045fbfab3bda23a77589a271c1 broke TestHTTPNetwork and TestHTTPSnapshot. Because the default UnmarshalJSON does not handle unexported fields. Important: The fix is partial and not proper to my taste. But I cut scope as I think the fix may require a change to the current serialization format. New ticket: https://github.com/ethersphere/go-ethereum/issues/1177 * p2p/simulations: Add a sanity test case for Node.Config UnmarshalJSON * p2p/simulations: revert back to defer Unlock() pattern for Network It's a good patten to call `defer Unlock()` right after `Lock()` so (new) error cases won't miss to unlock. Let's get back to that pattern. The patten was abandoned in 85a79b3ad3c5863f8612d25c246bcfad339f36b7, while fixing a data race. That data race does not exist anymore, since the Node.Up field got hidden behind its own lock. * p2p/simulations: consistent naming for test providers Node.UnmarshalJSON * p2p/simulations: remove JSON annotation from private fields of Node As unexported fields are not serialized. * p2p/simulations: fix deadlock in Network.GetRandomDownNode() Problem: GetRandomDownNode() locks -> getDownNodeIDs() -> GetNodes() tries to lock -> deadlock On Network type, unexported functions must assume that `net.lock` is already acquired and should not call exported functions which might try to lock again. * p2p/simulations: ensure method conformity for Network Connect* methods were moved to p2p/simulations.Network from swarm/network/simulation. However these new methods did not follow the pattern of Network methods, i.e., all exported method locks the whole Network either for read or write. * p2p/simulations: fix deadlock during network shutdown `TestDiscoveryPersistenceSimulationSimAdapter` often got into deadlock. The execution was stuck on two locks, i.e, `Kademlia.lock` and `p2p/simulations.Network.lock`. Usually the test got stuck once in each 20 executions with high confidence. `Kademlia` was stuck in `Kademlia.EachAddr()` and `Network` in `Network.Stop()`. Solution: in `Network.Stop()` `net.lock` must be released before calling `node.Stop()` as stopping a node (somehow - I did not find the exact code path) causes `Network.InitConn()` to be called from `Kademlia.SuggestPeer()` and that blocks on `net.lock`. Related ticket: https://github.com/ethersphere/go-ethereum/issues/1223 * swarm/state: simplify if statement in DBStore.Put() * p2p/simulations: remove faulty godoc from private function The comment started with the wrong method name. The method is simple and self explanatory. Also, it's private. => Let's just remove the comment.
2019-02-18 00:38:14 -06:00
Config *adapters.NodeConfig `json:"config,omitempty"`
Up bool `json:"up"`
}
p2p, swarm: fix node up races by granular locking (#18976) * swarm/network: DRY out repeated giga comment I not necessarily agree with the way we wait for event propagation. But I truly disagree with having duplicated giga comments. * p2p/simulations: encapsulate Node.Up field so we avoid data races The Node.Up field was accessed concurrently without "proper" locking. There was a lock on Network and that was used sometimes to access the field. Other times the locking was missed and we had a data race. For example: https://github.com/ethereum/go-ethereum/pull/18464 The case above was solved, but there were still intermittent/hard to reproduce races. So let's solve the issue permanently. resolves: ethersphere/go-ethereum#1146 * p2p/simulations: fix unmarshal of simulations.Node Making Node.Up field private in 13292ee897e345045fbfab3bda23a77589a271c1 broke TestHTTPNetwork and TestHTTPSnapshot. Because the default UnmarshalJSON does not handle unexported fields. Important: The fix is partial and not proper to my taste. But I cut scope as I think the fix may require a change to the current serialization format. New ticket: https://github.com/ethersphere/go-ethereum/issues/1177 * p2p/simulations: Add a sanity test case for Node.Config UnmarshalJSON * p2p/simulations: revert back to defer Unlock() pattern for Network It's a good patten to call `defer Unlock()` right after `Lock()` so (new) error cases won't miss to unlock. Let's get back to that pattern. The patten was abandoned in 85a79b3ad3c5863f8612d25c246bcfad339f36b7, while fixing a data race. That data race does not exist anymore, since the Node.Up field got hidden behind its own lock. * p2p/simulations: consistent naming for test providers Node.UnmarshalJSON * p2p/simulations: remove JSON annotation from private fields of Node As unexported fields are not serialized. * p2p/simulations: fix deadlock in Network.GetRandomDownNode() Problem: GetRandomDownNode() locks -> getDownNodeIDs() -> GetNodes() tries to lock -> deadlock On Network type, unexported functions must assume that `net.lock` is already acquired and should not call exported functions which might try to lock again. * p2p/simulations: ensure method conformity for Network Connect* methods were moved to p2p/simulations.Network from swarm/network/simulation. However these new methods did not follow the pattern of Network methods, i.e., all exported method locks the whole Network either for read or write. * p2p/simulations: fix deadlock during network shutdown `TestDiscoveryPersistenceSimulationSimAdapter` often got into deadlock. The execution was stuck on two locks, i.e, `Kademlia.lock` and `p2p/simulations.Network.lock`. Usually the test got stuck once in each 20 executions with high confidence. `Kademlia` was stuck in `Kademlia.EachAddr()` and `Network` in `Network.Stop()`. Solution: in `Network.Stop()` `net.lock` must be released before calling `node.Stop()` as stopping a node (somehow - I did not find the exact code path) causes `Network.InitConn()` to be called from `Kademlia.SuggestPeer()` and that blocks on `net.lock`. Related ticket: https://github.com/ethersphere/go-ethereum/issues/1223 * swarm/state: simplify if statement in DBStore.Put() * p2p/simulations: remove faulty godoc from private function The comment started with the wrong method name. The method is simple and self explanatory. Also, it's private. => Let's just remove the comment.
2019-02-18 00:38:14 -06:00
if err := json.Unmarshal(raw, &node); err != nil {
return err
}
*n = *newNode(nil, node.Config, node.Up)
p2p, swarm: fix node up races by granular locking (#18976) * swarm/network: DRY out repeated giga comment I not necessarily agree with the way we wait for event propagation. But I truly disagree with having duplicated giga comments. * p2p/simulations: encapsulate Node.Up field so we avoid data races The Node.Up field was accessed concurrently without "proper" locking. There was a lock on Network and that was used sometimes to access the field. Other times the locking was missed and we had a data race. For example: https://github.com/ethereum/go-ethereum/pull/18464 The case above was solved, but there were still intermittent/hard to reproduce races. So let's solve the issue permanently. resolves: ethersphere/go-ethereum#1146 * p2p/simulations: fix unmarshal of simulations.Node Making Node.Up field private in 13292ee897e345045fbfab3bda23a77589a271c1 broke TestHTTPNetwork and TestHTTPSnapshot. Because the default UnmarshalJSON does not handle unexported fields. Important: The fix is partial and not proper to my taste. But I cut scope as I think the fix may require a change to the current serialization format. New ticket: https://github.com/ethersphere/go-ethereum/issues/1177 * p2p/simulations: Add a sanity test case for Node.Config UnmarshalJSON * p2p/simulations: revert back to defer Unlock() pattern for Network It's a good patten to call `defer Unlock()` right after `Lock()` so (new) error cases won't miss to unlock. Let's get back to that pattern. The patten was abandoned in 85a79b3ad3c5863f8612d25c246bcfad339f36b7, while fixing a data race. That data race does not exist anymore, since the Node.Up field got hidden behind its own lock. * p2p/simulations: consistent naming for test providers Node.UnmarshalJSON * p2p/simulations: remove JSON annotation from private fields of Node As unexported fields are not serialized. * p2p/simulations: fix deadlock in Network.GetRandomDownNode() Problem: GetRandomDownNode() locks -> getDownNodeIDs() -> GetNodes() tries to lock -> deadlock On Network type, unexported functions must assume that `net.lock` is already acquired and should not call exported functions which might try to lock again. * p2p/simulations: ensure method conformity for Network Connect* methods were moved to p2p/simulations.Network from swarm/network/simulation. However these new methods did not follow the pattern of Network methods, i.e., all exported method locks the whole Network either for read or write. * p2p/simulations: fix deadlock during network shutdown `TestDiscoveryPersistenceSimulationSimAdapter` often got into deadlock. The execution was stuck on two locks, i.e, `Kademlia.lock` and `p2p/simulations.Network.lock`. Usually the test got stuck once in each 20 executions with high confidence. `Kademlia` was stuck in `Kademlia.EachAddr()` and `Network` in `Network.Stop()`. Solution: in `Network.Stop()` `net.lock` must be released before calling `node.Stop()` as stopping a node (somehow - I did not find the exact code path) causes `Network.InitConn()` to be called from `Kademlia.SuggestPeer()` and that blocks on `net.lock`. Related ticket: https://github.com/ethersphere/go-ethereum/issues/1223 * swarm/state: simplify if statement in DBStore.Put() * p2p/simulations: remove faulty godoc from private function The comment started with the wrong method name. The method is simple and self explanatory. Also, it's private. => Let's just remove the comment.
2019-02-18 00:38:14 -06:00
return nil
}
// Conn represents a connection between two nodes in the network
type Conn struct {
// One is the node which initiated the connection
all: new p2p node representation (#17643) Package p2p/enode provides a generalized representation of p2p nodes which can contain arbitrary information in key/value pairs. It is also the new home for the node database. The "v4" identity scheme is also moved here from p2p/enr to remove the dependency on Ethereum crypto from that package. Record signature handling is changed significantly. The identity scheme registry is removed and acceptable schemes must be passed to any method that needs identity. This means records must now be validated explicitly after decoding. The enode API is designed to make signature handling easy and safe: most APIs around the codebase work with enode.Node, which is a wrapper around a valid record. Going from enr.Record to enode.Node requires a valid signature. * p2p/discover: port to p2p/enode This ports the discovery code to the new node representation in p2p/enode. The wire protocol is unchanged, this can be considered a refactoring change. The Kademlia table can now deal with nodes using an arbitrary identity scheme. This requires a few incompatible API changes: - Table.Lookup is not available anymore. It used to take a public key as argument because v4 protocol requires one. Its replacement is LookupRandom. - Table.Resolve takes *enode.Node instead of NodeID. This is also for v4 protocol compatibility because nodes cannot be looked up by ID alone. - Types Node and NodeID are gone. Further commits in the series will be fixes all over the the codebase to deal with those removals. * p2p: port to p2p/enode and discovery changes This adapts package p2p to the changes in p2p/discover. All uses of discover.Node and discover.NodeID are replaced by their equivalents from p2p/enode. New API is added to retrieve the enode.Node instance of a peer. The behavior of Server.Self with discovery disabled is improved. It now tries much harder to report a working IP address, falling back to 127.0.0.1 if no suitable address can be determined through other means. These changes were needed for tests of other packages later in the series. * p2p/simulations, p2p/testing: port to p2p/enode No surprises here, mostly replacements of discover.Node, discover.NodeID with their new equivalents. The 'interesting' API changes are: - testing.ProtocolSession tracks complete nodes, not just their IDs. - adapters.NodeConfig has a new method to create a complete node. These changes were needed to make swarm tests work. Note that the NodeID change makes the code incompatible with old simulation snapshots. * whisper/whisperv5, whisper/whisperv6: port to p2p/enode This port was easy because whisper uses []byte for node IDs and URL strings in the API. * eth: port to p2p/enode Again, easy to port because eth uses strings for node IDs and doesn't care about node information in any way. * les: port to p2p/enode Apart from replacing discover.NodeID with enode.ID, most changes are in the server pool code. It now deals with complete nodes instead of (Pubkey, IP, Port) triples. The database format is unchanged for now, but we should probably change it to use the node database later. * node: port to p2p/enode This change simply replaces discover.Node and discover.NodeID with their new equivalents. * swarm/network: port to p2p/enode Swarm has its own node address representation, BzzAddr, containing both an overlay address (the hash of a secp256k1 public key) and an underlay address (enode:// URL). There are no changes to the BzzAddr format in this commit, but certain operations such as creating a BzzAddr from a node ID are now impossible because node IDs aren't public keys anymore. Most swarm-related changes in the series remove uses of NewAddrFromNodeID, replacing it with NewAddr which takes a complete node as argument. ToOverlayAddr is removed because we can just use the node ID directly.
2018-09-24 17:59:00 -05:00
One enode.ID `json:"one"`
// Other is the node which the connection was made to
all: new p2p node representation (#17643) Package p2p/enode provides a generalized representation of p2p nodes which can contain arbitrary information in key/value pairs. It is also the new home for the node database. The "v4" identity scheme is also moved here from p2p/enr to remove the dependency on Ethereum crypto from that package. Record signature handling is changed significantly. The identity scheme registry is removed and acceptable schemes must be passed to any method that needs identity. This means records must now be validated explicitly after decoding. The enode API is designed to make signature handling easy and safe: most APIs around the codebase work with enode.Node, which is a wrapper around a valid record. Going from enr.Record to enode.Node requires a valid signature. * p2p/discover: port to p2p/enode This ports the discovery code to the new node representation in p2p/enode. The wire protocol is unchanged, this can be considered a refactoring change. The Kademlia table can now deal with nodes using an arbitrary identity scheme. This requires a few incompatible API changes: - Table.Lookup is not available anymore. It used to take a public key as argument because v4 protocol requires one. Its replacement is LookupRandom. - Table.Resolve takes *enode.Node instead of NodeID. This is also for v4 protocol compatibility because nodes cannot be looked up by ID alone. - Types Node and NodeID are gone. Further commits in the series will be fixes all over the the codebase to deal with those removals. * p2p: port to p2p/enode and discovery changes This adapts package p2p to the changes in p2p/discover. All uses of discover.Node and discover.NodeID are replaced by their equivalents from p2p/enode. New API is added to retrieve the enode.Node instance of a peer. The behavior of Server.Self with discovery disabled is improved. It now tries much harder to report a working IP address, falling back to 127.0.0.1 if no suitable address can be determined through other means. These changes were needed for tests of other packages later in the series. * p2p/simulations, p2p/testing: port to p2p/enode No surprises here, mostly replacements of discover.Node, discover.NodeID with their new equivalents. The 'interesting' API changes are: - testing.ProtocolSession tracks complete nodes, not just their IDs. - adapters.NodeConfig has a new method to create a complete node. These changes were needed to make swarm tests work. Note that the NodeID change makes the code incompatible with old simulation snapshots. * whisper/whisperv5, whisper/whisperv6: port to p2p/enode This port was easy because whisper uses []byte for node IDs and URL strings in the API. * eth: port to p2p/enode Again, easy to port because eth uses strings for node IDs and doesn't care about node information in any way. * les: port to p2p/enode Apart from replacing discover.NodeID with enode.ID, most changes are in the server pool code. It now deals with complete nodes instead of (Pubkey, IP, Port) triples. The database format is unchanged for now, but we should probably change it to use the node database later. * node: port to p2p/enode This change simply replaces discover.Node and discover.NodeID with their new equivalents. * swarm/network: port to p2p/enode Swarm has its own node address representation, BzzAddr, containing both an overlay address (the hash of a secp256k1 public key) and an underlay address (enode:// URL). There are no changes to the BzzAddr format in this commit, but certain operations such as creating a BzzAddr from a node ID are now impossible because node IDs aren't public keys anymore. Most swarm-related changes in the series remove uses of NewAddrFromNodeID, replacing it with NewAddr which takes a complete node as argument. ToOverlayAddr is removed because we can just use the node ID directly.
2018-09-24 17:59:00 -05:00
Other enode.ID `json:"other"`
// Up tracks whether or not the connection is active
Up bool `json:"up"`
// Registers when the connection was grabbed to dial
initiated time.Time
one *Node
other *Node
}
// nodesUp returns whether both nodes are currently up
2018-05-08 06:08:43 -05:00
func (c *Conn) nodesUp() error {
p2p, swarm: fix node up races by granular locking (#18976) * swarm/network: DRY out repeated giga comment I not necessarily agree with the way we wait for event propagation. But I truly disagree with having duplicated giga comments. * p2p/simulations: encapsulate Node.Up field so we avoid data races The Node.Up field was accessed concurrently without "proper" locking. There was a lock on Network and that was used sometimes to access the field. Other times the locking was missed and we had a data race. For example: https://github.com/ethereum/go-ethereum/pull/18464 The case above was solved, but there were still intermittent/hard to reproduce races. So let's solve the issue permanently. resolves: ethersphere/go-ethereum#1146 * p2p/simulations: fix unmarshal of simulations.Node Making Node.Up field private in 13292ee897e345045fbfab3bda23a77589a271c1 broke TestHTTPNetwork and TestHTTPSnapshot. Because the default UnmarshalJSON does not handle unexported fields. Important: The fix is partial and not proper to my taste. But I cut scope as I think the fix may require a change to the current serialization format. New ticket: https://github.com/ethersphere/go-ethereum/issues/1177 * p2p/simulations: Add a sanity test case for Node.Config UnmarshalJSON * p2p/simulations: revert back to defer Unlock() pattern for Network It's a good patten to call `defer Unlock()` right after `Lock()` so (new) error cases won't miss to unlock. Let's get back to that pattern. The patten was abandoned in 85a79b3ad3c5863f8612d25c246bcfad339f36b7, while fixing a data race. That data race does not exist anymore, since the Node.Up field got hidden behind its own lock. * p2p/simulations: consistent naming for test providers Node.UnmarshalJSON * p2p/simulations: remove JSON annotation from private fields of Node As unexported fields are not serialized. * p2p/simulations: fix deadlock in Network.GetRandomDownNode() Problem: GetRandomDownNode() locks -> getDownNodeIDs() -> GetNodes() tries to lock -> deadlock On Network type, unexported functions must assume that `net.lock` is already acquired and should not call exported functions which might try to lock again. * p2p/simulations: ensure method conformity for Network Connect* methods were moved to p2p/simulations.Network from swarm/network/simulation. However these new methods did not follow the pattern of Network methods, i.e., all exported method locks the whole Network either for read or write. * p2p/simulations: fix deadlock during network shutdown `TestDiscoveryPersistenceSimulationSimAdapter` often got into deadlock. The execution was stuck on two locks, i.e, `Kademlia.lock` and `p2p/simulations.Network.lock`. Usually the test got stuck once in each 20 executions with high confidence. `Kademlia` was stuck in `Kademlia.EachAddr()` and `Network` in `Network.Stop()`. Solution: in `Network.Stop()` `net.lock` must be released before calling `node.Stop()` as stopping a node (somehow - I did not find the exact code path) causes `Network.InitConn()` to be called from `Kademlia.SuggestPeer()` and that blocks on `net.lock`. Related ticket: https://github.com/ethersphere/go-ethereum/issues/1223 * swarm/state: simplify if statement in DBStore.Put() * p2p/simulations: remove faulty godoc from private function The comment started with the wrong method name. The method is simple and self explanatory. Also, it's private. => Let's just remove the comment.
2019-02-18 00:38:14 -06:00
if !c.one.Up() {
2018-05-08 06:08:43 -05:00
return fmt.Errorf("one %v is not up", c.One)
}
p2p, swarm: fix node up races by granular locking (#18976) * swarm/network: DRY out repeated giga comment I not necessarily agree with the way we wait for event propagation. But I truly disagree with having duplicated giga comments. * p2p/simulations: encapsulate Node.Up field so we avoid data races The Node.Up field was accessed concurrently without "proper" locking. There was a lock on Network and that was used sometimes to access the field. Other times the locking was missed and we had a data race. For example: https://github.com/ethereum/go-ethereum/pull/18464 The case above was solved, but there were still intermittent/hard to reproduce races. So let's solve the issue permanently. resolves: ethersphere/go-ethereum#1146 * p2p/simulations: fix unmarshal of simulations.Node Making Node.Up field private in 13292ee897e345045fbfab3bda23a77589a271c1 broke TestHTTPNetwork and TestHTTPSnapshot. Because the default UnmarshalJSON does not handle unexported fields. Important: The fix is partial and not proper to my taste. But I cut scope as I think the fix may require a change to the current serialization format. New ticket: https://github.com/ethersphere/go-ethereum/issues/1177 * p2p/simulations: Add a sanity test case for Node.Config UnmarshalJSON * p2p/simulations: revert back to defer Unlock() pattern for Network It's a good patten to call `defer Unlock()` right after `Lock()` so (new) error cases won't miss to unlock. Let's get back to that pattern. The patten was abandoned in 85a79b3ad3c5863f8612d25c246bcfad339f36b7, while fixing a data race. That data race does not exist anymore, since the Node.Up field got hidden behind its own lock. * p2p/simulations: consistent naming for test providers Node.UnmarshalJSON * p2p/simulations: remove JSON annotation from private fields of Node As unexported fields are not serialized. * p2p/simulations: fix deadlock in Network.GetRandomDownNode() Problem: GetRandomDownNode() locks -> getDownNodeIDs() -> GetNodes() tries to lock -> deadlock On Network type, unexported functions must assume that `net.lock` is already acquired and should not call exported functions which might try to lock again. * p2p/simulations: ensure method conformity for Network Connect* methods were moved to p2p/simulations.Network from swarm/network/simulation. However these new methods did not follow the pattern of Network methods, i.e., all exported method locks the whole Network either for read or write. * p2p/simulations: fix deadlock during network shutdown `TestDiscoveryPersistenceSimulationSimAdapter` often got into deadlock. The execution was stuck on two locks, i.e, `Kademlia.lock` and `p2p/simulations.Network.lock`. Usually the test got stuck once in each 20 executions with high confidence. `Kademlia` was stuck in `Kademlia.EachAddr()` and `Network` in `Network.Stop()`. Solution: in `Network.Stop()` `net.lock` must be released before calling `node.Stop()` as stopping a node (somehow - I did not find the exact code path) causes `Network.InitConn()` to be called from `Kademlia.SuggestPeer()` and that blocks on `net.lock`. Related ticket: https://github.com/ethersphere/go-ethereum/issues/1223 * swarm/state: simplify if statement in DBStore.Put() * p2p/simulations: remove faulty godoc from private function The comment started with the wrong method name. The method is simple and self explanatory. Also, it's private. => Let's just remove the comment.
2019-02-18 00:38:14 -06:00
if !c.other.Up() {
2018-05-08 06:08:43 -05:00
return fmt.Errorf("other %v is not up", c.Other)
}
return nil
}
// String returns a log-friendly string
2018-05-08 06:08:43 -05:00
func (c *Conn) String() string {
return fmt.Sprintf("Conn %v->%v", c.One.TerminalString(), c.Other.TerminalString())
}
// Msg represents a p2p message sent between two nodes in the network
type Msg struct {
all: new p2p node representation (#17643) Package p2p/enode provides a generalized representation of p2p nodes which can contain arbitrary information in key/value pairs. It is also the new home for the node database. The "v4" identity scheme is also moved here from p2p/enr to remove the dependency on Ethereum crypto from that package. Record signature handling is changed significantly. The identity scheme registry is removed and acceptable schemes must be passed to any method that needs identity. This means records must now be validated explicitly after decoding. The enode API is designed to make signature handling easy and safe: most APIs around the codebase work with enode.Node, which is a wrapper around a valid record. Going from enr.Record to enode.Node requires a valid signature. * p2p/discover: port to p2p/enode This ports the discovery code to the new node representation in p2p/enode. The wire protocol is unchanged, this can be considered a refactoring change. The Kademlia table can now deal with nodes using an arbitrary identity scheme. This requires a few incompatible API changes: - Table.Lookup is not available anymore. It used to take a public key as argument because v4 protocol requires one. Its replacement is LookupRandom. - Table.Resolve takes *enode.Node instead of NodeID. This is also for v4 protocol compatibility because nodes cannot be looked up by ID alone. - Types Node and NodeID are gone. Further commits in the series will be fixes all over the the codebase to deal with those removals. * p2p: port to p2p/enode and discovery changes This adapts package p2p to the changes in p2p/discover. All uses of discover.Node and discover.NodeID are replaced by their equivalents from p2p/enode. New API is added to retrieve the enode.Node instance of a peer. The behavior of Server.Self with discovery disabled is improved. It now tries much harder to report a working IP address, falling back to 127.0.0.1 if no suitable address can be determined through other means. These changes were needed for tests of other packages later in the series. * p2p/simulations, p2p/testing: port to p2p/enode No surprises here, mostly replacements of discover.Node, discover.NodeID with their new equivalents. The 'interesting' API changes are: - testing.ProtocolSession tracks complete nodes, not just their IDs. - adapters.NodeConfig has a new method to create a complete node. These changes were needed to make swarm tests work. Note that the NodeID change makes the code incompatible with old simulation snapshots. * whisper/whisperv5, whisper/whisperv6: port to p2p/enode This port was easy because whisper uses []byte for node IDs and URL strings in the API. * eth: port to p2p/enode Again, easy to port because eth uses strings for node IDs and doesn't care about node information in any way. * les: port to p2p/enode Apart from replacing discover.NodeID with enode.ID, most changes are in the server pool code. It now deals with complete nodes instead of (Pubkey, IP, Port) triples. The database format is unchanged for now, but we should probably change it to use the node database later. * node: port to p2p/enode This change simply replaces discover.Node and discover.NodeID with their new equivalents. * swarm/network: port to p2p/enode Swarm has its own node address representation, BzzAddr, containing both an overlay address (the hash of a secp256k1 public key) and an underlay address (enode:// URL). There are no changes to the BzzAddr format in this commit, but certain operations such as creating a BzzAddr from a node ID are now impossible because node IDs aren't public keys anymore. Most swarm-related changes in the series remove uses of NewAddrFromNodeID, replacing it with NewAddr which takes a complete node as argument. ToOverlayAddr is removed because we can just use the node ID directly.
2018-09-24 17:59:00 -05:00
One enode.ID `json:"one"`
Other enode.ID `json:"other"`
Protocol string `json:"protocol"`
Code uint64 `json:"code"`
Received bool `json:"received"`
}
// String returns a log-friendly string
2018-05-08 06:08:43 -05:00
func (m *Msg) String() string {
return fmt.Sprintf("Msg(%d) %v->%v", m.Code, m.One.TerminalString(), m.Other.TerminalString())
}
// ConnLabel generates a deterministic string which represents a connection
// between two nodes, used to compare if two connections are between the same
// nodes
all: new p2p node representation (#17643) Package p2p/enode provides a generalized representation of p2p nodes which can contain arbitrary information in key/value pairs. It is also the new home for the node database. The "v4" identity scheme is also moved here from p2p/enr to remove the dependency on Ethereum crypto from that package. Record signature handling is changed significantly. The identity scheme registry is removed and acceptable schemes must be passed to any method that needs identity. This means records must now be validated explicitly after decoding. The enode API is designed to make signature handling easy and safe: most APIs around the codebase work with enode.Node, which is a wrapper around a valid record. Going from enr.Record to enode.Node requires a valid signature. * p2p/discover: port to p2p/enode This ports the discovery code to the new node representation in p2p/enode. The wire protocol is unchanged, this can be considered a refactoring change. The Kademlia table can now deal with nodes using an arbitrary identity scheme. This requires a few incompatible API changes: - Table.Lookup is not available anymore. It used to take a public key as argument because v4 protocol requires one. Its replacement is LookupRandom. - Table.Resolve takes *enode.Node instead of NodeID. This is also for v4 protocol compatibility because nodes cannot be looked up by ID alone. - Types Node and NodeID are gone. Further commits in the series will be fixes all over the the codebase to deal with those removals. * p2p: port to p2p/enode and discovery changes This adapts package p2p to the changes in p2p/discover. All uses of discover.Node and discover.NodeID are replaced by their equivalents from p2p/enode. New API is added to retrieve the enode.Node instance of a peer. The behavior of Server.Self with discovery disabled is improved. It now tries much harder to report a working IP address, falling back to 127.0.0.1 if no suitable address can be determined through other means. These changes were needed for tests of other packages later in the series. * p2p/simulations, p2p/testing: port to p2p/enode No surprises here, mostly replacements of discover.Node, discover.NodeID with their new equivalents. The 'interesting' API changes are: - testing.ProtocolSession tracks complete nodes, not just their IDs. - adapters.NodeConfig has a new method to create a complete node. These changes were needed to make swarm tests work. Note that the NodeID change makes the code incompatible with old simulation snapshots. * whisper/whisperv5, whisper/whisperv6: port to p2p/enode This port was easy because whisper uses []byte for node IDs and URL strings in the API. * eth: port to p2p/enode Again, easy to port because eth uses strings for node IDs and doesn't care about node information in any way. * les: port to p2p/enode Apart from replacing discover.NodeID with enode.ID, most changes are in the server pool code. It now deals with complete nodes instead of (Pubkey, IP, Port) triples. The database format is unchanged for now, but we should probably change it to use the node database later. * node: port to p2p/enode This change simply replaces discover.Node and discover.NodeID with their new equivalents. * swarm/network: port to p2p/enode Swarm has its own node address representation, BzzAddr, containing both an overlay address (the hash of a secp256k1 public key) and an underlay address (enode:// URL). There are no changes to the BzzAddr format in this commit, but certain operations such as creating a BzzAddr from a node ID are now impossible because node IDs aren't public keys anymore. Most swarm-related changes in the series remove uses of NewAddrFromNodeID, replacing it with NewAddr which takes a complete node as argument. ToOverlayAddr is removed because we can just use the node ID directly.
2018-09-24 17:59:00 -05:00
func ConnLabel(source, target enode.ID) string {
var first, second enode.ID
if bytes.Compare(source.Bytes(), target.Bytes()) > 0 {
first = target
second = source
} else {
first = source
second = target
}
return fmt.Sprintf("%v-%v", first, second)
}
// Snapshot represents the state of a network at a single point in time and can
// be used to restore the state of a network
type Snapshot struct {
Nodes []NodeSnapshot `json:"nodes,omitempty"`
Conns []Conn `json:"conns,omitempty"`
}
// NodeSnapshot represents the state of a node in the network
type NodeSnapshot struct {
Node Node `json:"node,omitempty"`
// Snapshots is arbitrary data gathered from calling node.Snapshots()
Snapshots map[string][]byte `json:"snapshots,omitempty"`
}
// Snapshot creates a network snapshot
2018-05-08 06:08:43 -05:00
func (net *Network) Snapshot() (*Snapshot, error) {
return net.snapshot(nil, nil)
}
func (net *Network) SnapshotWithServices(addServices []string, removeServices []string) (*Snapshot, error) {
return net.snapshot(addServices, removeServices)
}
func (net *Network) snapshot(addServices []string, removeServices []string) (*Snapshot, error) {
2018-05-08 06:08:43 -05:00
net.lock.Lock()
defer net.lock.Unlock()
snap := &Snapshot{
2018-05-08 06:08:43 -05:00
Nodes: make([]NodeSnapshot, len(net.Nodes)),
}
2018-05-08 06:08:43 -05:00
for i, node := range net.Nodes {
snap.Nodes[i] = NodeSnapshot{Node: *node.copy()}
p2p, swarm: fix node up races by granular locking (#18976) * swarm/network: DRY out repeated giga comment I not necessarily agree with the way we wait for event propagation. But I truly disagree with having duplicated giga comments. * p2p/simulations: encapsulate Node.Up field so we avoid data races The Node.Up field was accessed concurrently without "proper" locking. There was a lock on Network and that was used sometimes to access the field. Other times the locking was missed and we had a data race. For example: https://github.com/ethereum/go-ethereum/pull/18464 The case above was solved, but there were still intermittent/hard to reproduce races. So let's solve the issue permanently. resolves: ethersphere/go-ethereum#1146 * p2p/simulations: fix unmarshal of simulations.Node Making Node.Up field private in 13292ee897e345045fbfab3bda23a77589a271c1 broke TestHTTPNetwork and TestHTTPSnapshot. Because the default UnmarshalJSON does not handle unexported fields. Important: The fix is partial and not proper to my taste. But I cut scope as I think the fix may require a change to the current serialization format. New ticket: https://github.com/ethersphere/go-ethereum/issues/1177 * p2p/simulations: Add a sanity test case for Node.Config UnmarshalJSON * p2p/simulations: revert back to defer Unlock() pattern for Network It's a good patten to call `defer Unlock()` right after `Lock()` so (new) error cases won't miss to unlock. Let's get back to that pattern. The patten was abandoned in 85a79b3ad3c5863f8612d25c246bcfad339f36b7, while fixing a data race. That data race does not exist anymore, since the Node.Up field got hidden behind its own lock. * p2p/simulations: consistent naming for test providers Node.UnmarshalJSON * p2p/simulations: remove JSON annotation from private fields of Node As unexported fields are not serialized. * p2p/simulations: fix deadlock in Network.GetRandomDownNode() Problem: GetRandomDownNode() locks -> getDownNodeIDs() -> GetNodes() tries to lock -> deadlock On Network type, unexported functions must assume that `net.lock` is already acquired and should not call exported functions which might try to lock again. * p2p/simulations: ensure method conformity for Network Connect* methods were moved to p2p/simulations.Network from swarm/network/simulation. However these new methods did not follow the pattern of Network methods, i.e., all exported method locks the whole Network either for read or write. * p2p/simulations: fix deadlock during network shutdown `TestDiscoveryPersistenceSimulationSimAdapter` often got into deadlock. The execution was stuck on two locks, i.e, `Kademlia.lock` and `p2p/simulations.Network.lock`. Usually the test got stuck once in each 20 executions with high confidence. `Kademlia` was stuck in `Kademlia.EachAddr()` and `Network` in `Network.Stop()`. Solution: in `Network.Stop()` `net.lock` must be released before calling `node.Stop()` as stopping a node (somehow - I did not find the exact code path) causes `Network.InitConn()` to be called from `Kademlia.SuggestPeer()` and that blocks on `net.lock`. Related ticket: https://github.com/ethersphere/go-ethereum/issues/1223 * swarm/state: simplify if statement in DBStore.Put() * p2p/simulations: remove faulty godoc from private function The comment started with the wrong method name. The method is simple and self explanatory. Also, it's private. => Let's just remove the comment.
2019-02-18 00:38:14 -06:00
if !node.Up() {
continue
}
snapshots, err := node.Snapshots()
if err != nil {
return nil, err
}
snap.Nodes[i].Snapshots = snapshots
for _, addSvc := range addServices {
haveSvc := false
for _, svc := range snap.Nodes[i].Node.Config.Lifecycles {
if svc == addSvc {
haveSvc = true
break
}
}
if !haveSvc {
snap.Nodes[i].Node.Config.Lifecycles = append(snap.Nodes[i].Node.Config.Lifecycles, addSvc)
}
}
if len(removeServices) > 0 {
var cleanedServices []string
for _, svc := range snap.Nodes[i].Node.Config.Lifecycles {
haveSvc := false
for _, rmSvc := range removeServices {
if rmSvc == svc {
haveSvc = true
break
}
}
if !haveSvc {
cleanedServices = append(cleanedServices, svc)
}
}
snap.Nodes[i].Node.Config.Lifecycles = cleanedServices
}
}
for _, conn := range net.Conns {
if conn.Up {
snap.Conns = append(snap.Conns, *conn)
}
}
return snap, nil
}
// longrunning tests may need a longer timeout
var snapshotLoadTimeout = 900 * time.Second
// Load loads a network snapshot
2018-05-08 06:08:43 -05:00
func (net *Network) Load(snap *Snapshot) error {
// Start nodes.
for _, n := range snap.Nodes {
2018-05-08 06:08:43 -05:00
if _, err := net.NewNodeWithConfig(n.Node.Config); err != nil {
return err
}
p2p, swarm: fix node up races by granular locking (#18976) * swarm/network: DRY out repeated giga comment I not necessarily agree with the way we wait for event propagation. But I truly disagree with having duplicated giga comments. * p2p/simulations: encapsulate Node.Up field so we avoid data races The Node.Up field was accessed concurrently without "proper" locking. There was a lock on Network and that was used sometimes to access the field. Other times the locking was missed and we had a data race. For example: https://github.com/ethereum/go-ethereum/pull/18464 The case above was solved, but there were still intermittent/hard to reproduce races. So let's solve the issue permanently. resolves: ethersphere/go-ethereum#1146 * p2p/simulations: fix unmarshal of simulations.Node Making Node.Up field private in 13292ee897e345045fbfab3bda23a77589a271c1 broke TestHTTPNetwork and TestHTTPSnapshot. Because the default UnmarshalJSON does not handle unexported fields. Important: The fix is partial and not proper to my taste. But I cut scope as I think the fix may require a change to the current serialization format. New ticket: https://github.com/ethersphere/go-ethereum/issues/1177 * p2p/simulations: Add a sanity test case for Node.Config UnmarshalJSON * p2p/simulations: revert back to defer Unlock() pattern for Network It's a good patten to call `defer Unlock()` right after `Lock()` so (new) error cases won't miss to unlock. Let's get back to that pattern. The patten was abandoned in 85a79b3ad3c5863f8612d25c246bcfad339f36b7, while fixing a data race. That data race does not exist anymore, since the Node.Up field got hidden behind its own lock. * p2p/simulations: consistent naming for test providers Node.UnmarshalJSON * p2p/simulations: remove JSON annotation from private fields of Node As unexported fields are not serialized. * p2p/simulations: fix deadlock in Network.GetRandomDownNode() Problem: GetRandomDownNode() locks -> getDownNodeIDs() -> GetNodes() tries to lock -> deadlock On Network type, unexported functions must assume that `net.lock` is already acquired and should not call exported functions which might try to lock again. * p2p/simulations: ensure method conformity for Network Connect* methods were moved to p2p/simulations.Network from swarm/network/simulation. However these new methods did not follow the pattern of Network methods, i.e., all exported method locks the whole Network either for read or write. * p2p/simulations: fix deadlock during network shutdown `TestDiscoveryPersistenceSimulationSimAdapter` often got into deadlock. The execution was stuck on two locks, i.e, `Kademlia.lock` and `p2p/simulations.Network.lock`. Usually the test got stuck once in each 20 executions with high confidence. `Kademlia` was stuck in `Kademlia.EachAddr()` and `Network` in `Network.Stop()`. Solution: in `Network.Stop()` `net.lock` must be released before calling `node.Stop()` as stopping a node (somehow - I did not find the exact code path) causes `Network.InitConn()` to be called from `Kademlia.SuggestPeer()` and that blocks on `net.lock`. Related ticket: https://github.com/ethersphere/go-ethereum/issues/1223 * swarm/state: simplify if statement in DBStore.Put() * p2p/simulations: remove faulty godoc from private function The comment started with the wrong method name. The method is simple and self explanatory. Also, it's private. => Let's just remove the comment.
2019-02-18 00:38:14 -06:00
if !n.Node.Up() {
continue
}
2018-05-08 06:08:43 -05:00
if err := net.startWithSnapshots(n.Node.Config.ID, n.Snapshots); err != nil {
return err
}
}
// Prepare connection events counter.
allConnected := make(chan struct{}) // closed when all connections are established
done := make(chan struct{}) // ensures that the event loop goroutine is terminated
defer close(done)
// Subscribe to event channel.
// It needs to be done outside of the event loop goroutine (created below)
// to ensure that the event channel is blocking before connect calls are made.
events := make(chan *Event)
sub := net.Events().Subscribe(events)
defer sub.Unsubscribe()
go func() {
// Expected number of connections.
total := len(snap.Conns)
// Set of all established connections from the snapshot, not other connections.
// Key array element 0 is the connection One field value, and element 1 connection Other field.
connections := make(map[[2]enode.ID]struct{}, total)
for {
select {
case e := <-events:
// Ignore control events as they do not represent
// connect or disconnect (Up) state change.
if e.Control {
continue
}
// Detect only connection events.
if e.Type != EventTypeConn {
continue
}
connection := [2]enode.ID{e.Conn.One, e.Conn.Other}
// Nodes are still not connected or have been disconnected.
if !e.Conn.Up {
// Delete the connection from the set of established connections.
// This will prevent false positive in case disconnections happen.
delete(connections, connection)
log.Warn("load snapshot: unexpected disconnection", "one", e.Conn.One, "other", e.Conn.Other)
continue
}
// Check that the connection is from the snapshot.
for _, conn := range snap.Conns {
if conn.One == e.Conn.One && conn.Other == e.Conn.Other {
// Add the connection to the set of established connections.
connections[connection] = struct{}{}
if len(connections) == total {
// Signal that all nodes are connected.
close(allConnected)
return
}
break
}
}
case <-done:
// Load function returned, terminate this goroutine.
return
}
}
}()
// Start connecting.
for _, conn := range snap.Conns {
p2p, swarm: fix node up races by granular locking (#18976) * swarm/network: DRY out repeated giga comment I not necessarily agree with the way we wait for event propagation. But I truly disagree with having duplicated giga comments. * p2p/simulations: encapsulate Node.Up field so we avoid data races The Node.Up field was accessed concurrently without "proper" locking. There was a lock on Network and that was used sometimes to access the field. Other times the locking was missed and we had a data race. For example: https://github.com/ethereum/go-ethereum/pull/18464 The case above was solved, but there were still intermittent/hard to reproduce races. So let's solve the issue permanently. resolves: ethersphere/go-ethereum#1146 * p2p/simulations: fix unmarshal of simulations.Node Making Node.Up field private in 13292ee897e345045fbfab3bda23a77589a271c1 broke TestHTTPNetwork and TestHTTPSnapshot. Because the default UnmarshalJSON does not handle unexported fields. Important: The fix is partial and not proper to my taste. But I cut scope as I think the fix may require a change to the current serialization format. New ticket: https://github.com/ethersphere/go-ethereum/issues/1177 * p2p/simulations: Add a sanity test case for Node.Config UnmarshalJSON * p2p/simulations: revert back to defer Unlock() pattern for Network It's a good patten to call `defer Unlock()` right after `Lock()` so (new) error cases won't miss to unlock. Let's get back to that pattern. The patten was abandoned in 85a79b3ad3c5863f8612d25c246bcfad339f36b7, while fixing a data race. That data race does not exist anymore, since the Node.Up field got hidden behind its own lock. * p2p/simulations: consistent naming for test providers Node.UnmarshalJSON * p2p/simulations: remove JSON annotation from private fields of Node As unexported fields are not serialized. * p2p/simulations: fix deadlock in Network.GetRandomDownNode() Problem: GetRandomDownNode() locks -> getDownNodeIDs() -> GetNodes() tries to lock -> deadlock On Network type, unexported functions must assume that `net.lock` is already acquired and should not call exported functions which might try to lock again. * p2p/simulations: ensure method conformity for Network Connect* methods were moved to p2p/simulations.Network from swarm/network/simulation. However these new methods did not follow the pattern of Network methods, i.e., all exported method locks the whole Network either for read or write. * p2p/simulations: fix deadlock during network shutdown `TestDiscoveryPersistenceSimulationSimAdapter` often got into deadlock. The execution was stuck on two locks, i.e, `Kademlia.lock` and `p2p/simulations.Network.lock`. Usually the test got stuck once in each 20 executions with high confidence. `Kademlia` was stuck in `Kademlia.EachAddr()` and `Network` in `Network.Stop()`. Solution: in `Network.Stop()` `net.lock` must be released before calling `node.Stop()` as stopping a node (somehow - I did not find the exact code path) causes `Network.InitConn()` to be called from `Kademlia.SuggestPeer()` and that blocks on `net.lock`. Related ticket: https://github.com/ethersphere/go-ethereum/issues/1223 * swarm/state: simplify if statement in DBStore.Put() * p2p/simulations: remove faulty godoc from private function The comment started with the wrong method name. The method is simple and self explanatory. Also, it's private. => Let's just remove the comment.
2019-02-18 00:38:14 -06:00
if !net.GetNode(conn.One).Up() || !net.GetNode(conn.Other).Up() {
//in this case, at least one of the nodes of a connection is not up,
//so it would result in the snapshot `Load` to fail
continue
}
2018-05-08 06:08:43 -05:00
if err := net.Connect(conn.One, conn.Other); err != nil {
return err
}
}
select {
// Wait until all connections from the snapshot are established.
case <-allConnected:
// Make sure that we do not wait forever.
case <-time.After(snapshotLoadTimeout):
return errors.New("snapshot connections not established")
}
return nil
}
// Subscribe reads control events from a channel and executes them
2018-05-08 06:08:43 -05:00
func (net *Network) Subscribe(events chan *Event) {
for {
select {
case event, ok := <-events:
if !ok {
return
}
if event.Control {
2018-05-08 06:08:43 -05:00
net.executeControlEvent(event)
}
2018-05-08 06:08:43 -05:00
case <-net.quitc:
return
}
}
}
2018-05-08 06:08:43 -05:00
func (net *Network) executeControlEvent(event *Event) {
log.Trace("Executing control event", "type", event.Type, "event", event)
switch event.Type {
case EventTypeNode:
2018-05-08 06:08:43 -05:00
if err := net.executeNodeEvent(event); err != nil {
log.Error("Error executing node event", "event", event, "err", err)
}
case EventTypeConn:
2018-05-08 06:08:43 -05:00
if err := net.executeConnEvent(event); err != nil {
log.Error("Error executing conn event", "event", event, "err", err)
}
case EventTypeMsg:
log.Warn("Ignoring control msg event")
}
}
2018-05-08 06:08:43 -05:00
func (net *Network) executeNodeEvent(e *Event) error {
p2p, swarm: fix node up races by granular locking (#18976) * swarm/network: DRY out repeated giga comment I not necessarily agree with the way we wait for event propagation. But I truly disagree with having duplicated giga comments. * p2p/simulations: encapsulate Node.Up field so we avoid data races The Node.Up field was accessed concurrently without "proper" locking. There was a lock on Network and that was used sometimes to access the field. Other times the locking was missed and we had a data race. For example: https://github.com/ethereum/go-ethereum/pull/18464 The case above was solved, but there were still intermittent/hard to reproduce races. So let's solve the issue permanently. resolves: ethersphere/go-ethereum#1146 * p2p/simulations: fix unmarshal of simulations.Node Making Node.Up field private in 13292ee897e345045fbfab3bda23a77589a271c1 broke TestHTTPNetwork and TestHTTPSnapshot. Because the default UnmarshalJSON does not handle unexported fields. Important: The fix is partial and not proper to my taste. But I cut scope as I think the fix may require a change to the current serialization format. New ticket: https://github.com/ethersphere/go-ethereum/issues/1177 * p2p/simulations: Add a sanity test case for Node.Config UnmarshalJSON * p2p/simulations: revert back to defer Unlock() pattern for Network It's a good patten to call `defer Unlock()` right after `Lock()` so (new) error cases won't miss to unlock. Let's get back to that pattern. The patten was abandoned in 85a79b3ad3c5863f8612d25c246bcfad339f36b7, while fixing a data race. That data race does not exist anymore, since the Node.Up field got hidden behind its own lock. * p2p/simulations: consistent naming for test providers Node.UnmarshalJSON * p2p/simulations: remove JSON annotation from private fields of Node As unexported fields are not serialized. * p2p/simulations: fix deadlock in Network.GetRandomDownNode() Problem: GetRandomDownNode() locks -> getDownNodeIDs() -> GetNodes() tries to lock -> deadlock On Network type, unexported functions must assume that `net.lock` is already acquired and should not call exported functions which might try to lock again. * p2p/simulations: ensure method conformity for Network Connect* methods were moved to p2p/simulations.Network from swarm/network/simulation. However these new methods did not follow the pattern of Network methods, i.e., all exported method locks the whole Network either for read or write. * p2p/simulations: fix deadlock during network shutdown `TestDiscoveryPersistenceSimulationSimAdapter` often got into deadlock. The execution was stuck on two locks, i.e, `Kademlia.lock` and `p2p/simulations.Network.lock`. Usually the test got stuck once in each 20 executions with high confidence. `Kademlia` was stuck in `Kademlia.EachAddr()` and `Network` in `Network.Stop()`. Solution: in `Network.Stop()` `net.lock` must be released before calling `node.Stop()` as stopping a node (somehow - I did not find the exact code path) causes `Network.InitConn()` to be called from `Kademlia.SuggestPeer()` and that blocks on `net.lock`. Related ticket: https://github.com/ethersphere/go-ethereum/issues/1223 * swarm/state: simplify if statement in DBStore.Put() * p2p/simulations: remove faulty godoc from private function The comment started with the wrong method name. The method is simple and self explanatory. Also, it's private. => Let's just remove the comment.
2019-02-18 00:38:14 -06:00
if !e.Node.Up() {
2018-05-08 06:08:43 -05:00
return net.Stop(e.Node.ID())
}
2018-05-08 06:08:43 -05:00
if _, err := net.NewNodeWithConfig(e.Node.Config); err != nil {
return err
}
2018-05-08 06:08:43 -05:00
return net.Start(e.Node.ID())
}
2018-05-08 06:08:43 -05:00
func (net *Network) executeConnEvent(e *Event) error {
if e.Conn.Up {
2018-05-08 06:08:43 -05:00
return net.Connect(e.Conn.One, e.Conn.Other)
}
return net.Disconnect(e.Conn.One, e.Conn.Other)
}