From ae7a132f38404e9f654ab1b7c5dd84ba6a3efda6 Mon Sep 17 00:00:00 2001
From: Erik Johnston <erik@matrix.org>
Date: Tue, 31 Jan 2017 13:40:09 +0000
Subject: [PATCH] Better handle 404 response for federation /send/

---
 synapse/federation/transaction_queue.py |  1 +
 synapse/util/retryutils.py              | 15 +++++++++++++--
 2 files changed, 14 insertions(+), 2 deletions(-)

diff --git a/synapse/federation/transaction_queue.py b/synapse/federation/transaction_queue.py
index d18f6b6cfd..cb106c6a1b 100644
--- a/synapse/federation/transaction_queue.py
+++ b/synapse/federation/transaction_queue.py
@@ -319,6 +319,7 @@ class TransactionQueue(object):
                     destination,
                     self.clock,
                     self.store,
+                    backoff_on_404=True,  # If we get a 404 the other side has gone
                 )
 
                 device_message_edus, device_stream_id, dev_list_id = (
diff --git a/synapse/util/retryutils.py b/synapse/util/retryutils.py
index e2de7fce91..cc88a0b532 100644
--- a/synapse/util/retryutils.py
+++ b/synapse/util/retryutils.py
@@ -88,7 +88,7 @@ class RetryDestinationLimiter(object):
     def __init__(self, destination, clock, store, retry_interval,
                  min_retry_interval=10 * 60 * 1000,
                  max_retry_interval=24 * 60 * 60 * 1000,
-                 multiplier_retry_interval=5,):
+                 multiplier_retry_interval=5, backoff_on_404=False):
         """Marks the destination as "down" if an exception is thrown in the
         context, except for CodeMessageException with code < 500.
 
@@ -107,6 +107,7 @@ class RetryDestinationLimiter(object):
                 a failed request, in milliseconds.
             multiplier_retry_interval (int): The multiplier to use to increase
                 the retry interval after a failed request.
+            backoff_on_404 (bool): Back off if we get a 404
         """
         self.clock = clock
         self.store = store
@@ -116,6 +117,7 @@ class RetryDestinationLimiter(object):
         self.min_retry_interval = min_retry_interval
         self.max_retry_interval = max_retry_interval
         self.multiplier_retry_interval = multiplier_retry_interval
+        self.backoff_on_404 = backoff_on_404
 
     def __enter__(self):
         pass
@@ -123,7 +125,16 @@ class RetryDestinationLimiter(object):
     def __exit__(self, exc_type, exc_val, exc_tb):
         valid_err_code = False
         if exc_type is not None and issubclass(exc_type, CodeMessageException):
-            valid_err_code = exc_val.code != 429 and 0 <= exc_val.code < 500
+            if exc_val.code < 400:
+                valid_err_code = True
+            elif exc_val.code == 404 and self.backoff_on_404:
+                valid_err_code = False
+            elif exc_val.code == 429:
+                valid_err_code = False
+            elif exc_val.code < 500:
+                valid_err_code = True
+            else:
+                valid_err_code = False
 
         if exc_type is None or valid_err_code:
             # We connected successfully.