Try to only back off if we think we failed to connect to the remote

synapse/federation/transaction_queue.py

@@ -167,15 +167,6 @@ class TransactionQueue(object):
             logger.info("TX [%s] Nothing to send", destination)
             return
 
-        logger.debug(
-            "TX [%s] Attempting new transaction"
-            " (pdus: %d, edus: %d, failures: %d)",
-            destination,
-            len(pending_pdus),
-            len(pending_edus),
-            len(pending_failures)
-        )
-
         # Sort based on the order field
         pending_pdus.sort(key=lambda t: t[2])
 
@@ -194,7 +185,15 @@ class TransactionQueue(object):
                 self.store,
             )
 
-            with limiter:
+            logger.debug(
+                "TX [%s] Attempting new transaction"
+                " (pdus: %d, edus: %d, failures: %d)",
+                destination,
+                len(pending_pdus),
+                len(pending_edus),
+                len(pending_failures)
+            )
+
             self.pending_transactions[destination] = 1
 
             logger.debug("TX [%s] Persisting transaction...", destination)
@@ -220,6 +219,7 @@ class TransactionQueue(object):
                 transaction.transaction_id,
             )
 
+            with limiter:
                 # Actually send the transaction
 
                 # FIXME (erikj): This is a bit of a hack to make the Pdu age

synapse/util/retryutils.py

@@ -15,6 +15,8 @@
 
 from twisted.internet import defer
 
+from synapse.api.errors import CodeMessageException
+
 import logging
 
 
@@ -67,7 +69,7 @@ def get_retry_limiter(destination, clock, store, **kwargs):
 class RetryDestinationLimiter(object):
     def __init__(self, destination, clock, store, retry_interval,
                  min_retry_interval=20000, max_retry_interval=60 * 60 * 1000,
-                 multiplier_retry_interval=2):
+                 multiplier_retry_interval=2,):
         self.clock = clock
         self.store = store
         self.destination = destination
@@ -87,7 +89,11 @@ class RetryDestinationLimiter(object):
                 failure.value
             )
 
-        if exc_type is None and exc_val is None and exc_tb is None:
+        valid_err_code = False
+        if exc_type is CodeMessageException:
+            valid_err_code = 0 <= exc_val.code < 500
+
+        if exc_type is None or valid_err_code:
             # We connected successfully.
             if not self.retry_interval:
                 return