Analyze user_ips before running deduplication

Due to the table locks taken out by the naive upsert, the table
statistics may be out of date. During deduplication it is important that
the correct index is used as otherwise a full table scan may be
incorrectly used, which can end up thrashing the database badly.

1 files changed, 24 insertions, 0 deletions

diff --git a/synapse/storage/client_ips.py b/synapse/storage/client_ips.py
index 091d7116c5..6f81406269 100644
--- a/synapse/storage/client_ips.py
+++ b/synapse/storage/client_ips.py
@@ -66,6 +66,11 @@ class ClientIpStore(background_updates.BackgroundUpdateStore):
         )
 
         self.register_background_update_handler(
+            "user_ips_analyze",
+            self._analyze_user_ip,
+        )
+
+        self.register_background_update_handler(
             "user_ips_remove_dupes",
             self._remove_user_ip_dupes,
         )
@@ -109,6 +114,25 @@ class ClientIpStore(background_updates.BackgroundUpdateStore):
         defer.returnValue(1)
 
     @defer.inlineCallbacks
+    def _analyze_user_ip(self, progress, batch_size):
+        # Background update to analyze user_ips table before we run the
+        # deduplication background update. The table may not have been analyzed
+        # for ages due to the table locks.
+        #
+        # This will lock out the naive upserts to user_ips while it happens, but
+        # the analyze should be quick (28GB table takes ~10s)
+        def user_ips_analyze(txn):
+            txn.execute("ANALYZE user_ips")
+
+        end_last_seen = yield self.runInteraction(
+            "user_ips_analyze", user_ips_analyze
+        )
+
+        yield self._end_background_update("user_ips_analyze")
+
+        defer.returnValue(1)
+
+    @defer.inlineCallbacks
     def _remove_user_ip_dupes(self, progress, batch_size):
         # This works function works by scanning the user_ips table in batches
         # based on `last_seen`. For each row in a batch it searches the rest of