Limit back_off value to avoid very long retry interval

src/context.c

@@ -679,6 +679,7 @@ upstreams_create(getdns_context *context, size_t size)
 	r->count = 0;
 	r->current_udp = 0;
 	r->current_stateful = 0;
+    r->max_backoff_value = context->max_backoff_value;
 	r->tls_backoff_time = context->tls_backoff_time;
 	r->tls_connection_retries = context->tls_connection_retries;
 	r->log = context->log;
@@ -1664,6 +1665,7 @@ getdns_context_create_with_extended_memory_functions(
 	result->tls_backoff_time = 3600;
 	result->tls_connection_retries = 2;
 	result->limit_outstanding_queries = 0;
+    result->max_backoff_value = 1000;
 
 	/* unbound context is initialized here */
 	/* Unbound needs SSL to be init'ed this early when TLS is used. However we
@@ -2362,6 +2364,22 @@ getdns_context_set_round_robin_upstreams(getdns_context *context, uint8_t value)
     return GETDNS_RETURN_GOOD;
 }               /* getdns_context_set_round_robin_upstreams */
 
+/*
+ * getdns_context_set_max_backoff_value
+ *
+ */
+getdns_return_t
+getdns_context_set_max_backoff_value(getdns_context *context, uint16_t value)
+{
+    RETURN_IF_NULL(context, GETDNS_RETURN_INVALID_PARAMETER);
+
+    context->max_backoff_value = value;
+
+    dispatch_updated(context, GETDNS_CONTEXT_CODE_MAX_BACKOFF_VALUE);
+
+    return GETDNS_RETURN_GOOD;
+}               /* getdns_context_set_max_backoff_value */
+
 /*
  * getdns_context_set_tls_backoff_time
  *
@@ -3931,6 +3949,8 @@ _get_context_settings(getdns_context* context)
 	                           context->tls_auth)
 	    || getdns_dict_set_int(result, "round_robin_upstreams",
 	                           context->round_robin_upstreams)
+	    || getdns_dict_set_int(result, "max_backoff_value",
+	                           context->max_backoff_value)
 	    || getdns_dict_set_int(result, "tls_backoff_time",
 	                           context->tls_backoff_time)
 	    || getdns_dict_set_int(result, "tls_connection_retries",
@@ -4378,6 +4398,15 @@ getdns_context_get_round_robin_upstreams(getdns_context *context,
     return GETDNS_RETURN_GOOD;
 }
 
+getdns_return_t
+getdns_context_get_max_backoff_value(getdns_context *context,
+    uint16_t* value) {
+    RETURN_IF_NULL(context, GETDNS_RETURN_INVALID_PARAMETER);
+    RETURN_IF_NULL(value, GETDNS_RETURN_INVALID_PARAMETER);
+    *value = context->max_backoff_value;
+    return GETDNS_RETURN_GOOD;
+}
+
 getdns_return_t
 getdns_context_get_tls_backoff_time(getdns_context *context,
     uint16_t* value) {

src/context.h

@@ -263,6 +263,7 @@ typedef struct getdns_upstreams {
 	size_t count;
 	size_t current_udp;
 	size_t current_stateful;
+    uint16_t max_backoff_value;
 	uint16_t tls_backoff_time;
 	uint16_t tls_connection_retries;
 	getdns_log_config log;
@@ -357,6 +358,7 @@ struct getdns_context {
 	getdns_tls_authentication_t  tls_auth;  /* What user requested for TLS*/
 	getdns_tls_authentication_t  tls_auth_min; /* Derived minimum auth allowed*/
 	uint8_t              round_robin_upstreams;
+    uint16_t             max_backoff_value;
 	uint16_t             tls_backoff_time;
 	uint16_t             tls_connection_retries;
 

src/getdns/getdns_extra.h.in

@@ -103,6 +103,9 @@ extern "C" {
 #define GETDNS_CONTEXT_CODE_TLS_CURVES_LIST 634
 #define GETDNS_CONTEXT_CODE_TLS_CURVES_LIST_TEXT "Change related to getdns_context_set_tls_curves_list"
 
+#define GETDNS_CONTEXT_CODE_MAX_BACKOFF_VALUE 635
+#define GETDNS_CONTEXT_CODE_MAX_BACKOFF_VALUE_TEXT "Change related to getdns_context_set_max_backoff_value"
+
 /** @}
   */
 
@@ -768,6 +771,19 @@ getdns_return_t
 getdns_context_set_tls_curves_list(
     getdns_context *context, const char *curves_list);
 
+/**
+ * Set the maximum number of messages that can be sent to other upstreams
+ * before the upstream which has previously timed out will be tried again.
+ * @see getdns_context_get_max_backoff_value
+ * @param[in] context  The context to configure
+ * @param[in[ value    Number of messages sent to other upstreams before 
+ *                     retrying the upstream which had timed out.
+ * @return GETDNS_RETURN_GOOD on success
+ * @return GETDNS_RETURN_INVALID_PARAMETER if context is null.
+ */
+getdns_return_t
+getdns_context_set_max_backoff_value(getdns_context *context, uint16_t value);
+
 
 /**
  * Get the current resolution type setting from this context.
@@ -1076,7 +1092,7 @@ getdns_context_get_tls_authentication(getdns_context *context,
 /**
  * Get whether the context is configured to round robin queries over the available
  * upstreams.
- * @see getdns_context_get_round_robin_upstreams
+ * @see getdns_context_set_round_robin_upstreams
  * @param[in]  context The context from which to get the setting
  * @param[out] value 1 if the setting is on, 0 otherwise
  * @return GETDNS_RETURN_GOOD when successful
@@ -1303,6 +1319,19 @@ getdns_return_t
 getdns_context_get_tls_curves_list(
     getdns_context *context, const char **curves_list);
 
+/**
+ * Get the maximum number of messages that can be sent to other upstreams
+ * before the upstream which has previously timed out will be tried again.
+ * @see getdns_context_set_max_backoff_value
+ * @param[in]  context  The context from which to get the setting
+ * @param[out] value    Number of messages sent to other upstreams before 
+ *                      retrying the upstream which had timed out.
+ * @return GETDNS_RETURN_GOOD on success
+ * @return GETDNS_RETURN_INVALID_PARAMETER if context is null.
+ */
+getdns_return_t
+getdns_context_get_max_backoff_value(getdns_context *context,
+    uint16_t* value);
 
 /** @}
  */

src/libgetdns.symbols

@@ -29,6 +29,7 @@ getdns_context_get_num_pending_requests
 getdns_context_get_resolution_type
 getdns_context_get_resolvconf
 getdns_context_get_round_robin_upstreams
+getdns_context_get_max_backoff_value
 getdns_context_get_suffix
 getdns_context_get_timeout
 getdns_context_get_tls_authentication
@@ -73,6 +74,7 @@ getdns_context_set_resolution_type
 getdns_context_set_resolvconf
 getdns_context_set_return_dnssec_status
 getdns_context_set_round_robin_upstreams
+getdns_context_set_max_backoff_value
 getdns_context_set_suffix
 getdns_context_set_timeout
 getdns_context_set_tls_authentication

src/stub.c

@@ -460,8 +460,13 @@ stub_next_upstream(getdns_network_req *netreq)
 {
 	getdns_dns_req *dnsreq = netreq->owner;
 
-	if (! --netreq->upstream->to_retry) 
+	if (! --netreq->upstream->to_retry) {
-		netreq->upstream->to_retry = -(netreq->upstream->back_off *= 2);
+        /* Limit back_off value to configured maximum */
+        if (netreq->upstream->back_off * 2 > dnsreq->context->max_backoff_value)
+            netreq->upstream->to_retry = -(dnsreq->context->max_backoff_value);
+        else
+            netreq->upstream->to_retry = -(netreq->upstream->back_off *= 2);
+    }
 
 	dnsreq->upstreams->current_udp+=GETDNS_UPSTREAM_TRANSPORTS;
 	if (dnsreq->upstreams->current_udp >= dnsreq->upstreams->count)