Sorry for the very big commit that may lead to git log issues and merge
conflicts, but it's a major step forward:
* Server can be faster at startup because imports() are async and we can
easily lazy import big modules
* Angular doesn't seem to support ES import (with .js extension), so we
had to correctly organize peertube into a monorepo:
* Use yarn workspace feature
* Use typescript reference projects for dependencies
* Shared projects have been moved into "packages", each one is now a
node module (with a dedicated package.json/tsconfig.json)
* server/tools have been moved into apps/ and is now a dedicated app
bundled and published on NPM so users don't have to build peertube
cli tools manually
* server/tests have been moved into packages/ so we don't compile
them every time we want to run the server
* Use isolatedModule option:
* Had to move from const enum to const
(https://www.typescriptlang.org/docs/handbook/enums.html#objects-vs-enums)
* Had to explictely specify "type" imports when used in decorators
* Prefer tsx (that uses esbuild under the hood) instead of ts-node to
load typescript files (tests with mocha or scripts):
* To reduce test complexity as esbuild doesn't support decorator
metadata, we only test server files that do not import server
models
* We still build tests files into js files for a faster CI
* Remove unmaintained peertube CLI import script
* Removed some barrels to speed up execution (less imports)
* Add support for saving video files to object storage
* Add support for custom url generation on s3 stored files
Uses two config keys to support url generation that doesn't directly go
to (compatible s3). Can be used to generate urls to any cache server or
CDN.
* Upload files to s3 concurrently and delete originals afterwards
* Only publish after move to object storage is complete
* Use base url instead of url template
* Fix mistyped config field
* Add rudenmentary way to download before transcode
* Implement Chocobozzz suggestions
https://github.com/Chocobozzz/PeerTube/pull/4290#issuecomment-891670478
The remarks in question:
Try to use objectStorage prefix instead of s3 prefix for your function/variables/config names
Prefer to use a tree for the config: s3.streaming_playlists_bucket -> object_storage.streaming_playlists.bucket
Use uppercase for config: S3.STREAMING_PLAYLISTS_BUCKETINFO.bucket -> OBJECT_STORAGE.STREAMING_PLAYLISTS.BUCKET (maybe BUCKET_NAME instead of BUCKET)
I suggest to rename moveJobsRunning to pendingMovingJobs (or better, create a dedicated videoJobInfo table with a pendingMove & videoId columns so we could also use this table to track pending transcoding jobs)
https://github.com/Chocobozzz/PeerTube/pull/4290/files#diff-3e26d41ca4bda1de8e1747af70ca2af642abcc1e9e0bfb94239ff2165acfbde5R19 uses a string instead of an integer
I think we should store the origin object storage URL in fileUrl, without base_url injection. Instead, inject the base_url at "runtime" so admins can easily change this configuration without running a script to update DB URLs
* Import correct function
* Support multipart upload
* Remove import of node 15.0 module stream/promises
* Extend maximum upload job length
Using the same value as for redundancy downloading seems logical
* Use dynamic part size for really large uploads
Also adds very small part size for local testing
* Fix decreasePendingMove query
* Resolve various PR comments
* Move to object storage after optimize
* Make upload size configurable and increase default
* Prune webtorrent files that are stored in object storage
* Move files after transcoding jobs
* Fix federation
* Add video path manager
* Support move to external storage job in client
* Fix live object storage tests
Co-authored-by: Chocobozzz <me@florianbigard.com>
Chocobozzz
Yehuda Deutsch
Jelle Besseling