From 90743c9d8910679bb688070b3929f3005e106d00 Mon Sep 17 00:00:00 2001 From: Erik Johnston Date: Wed, 23 Jan 2019 08:45:18 +0000 Subject: [PATCH] Fixup removal of duplicate `user_ips` rows (#4432) * Remove unnecessary ORDER BY clause * Add logging * Newsfile --- changelog.d/4432.misc | 1 + synapse/storage/client_ips.py | 6 +++++- 2 files changed, 6 insertions(+), 1 deletion(-) create mode 100644 changelog.d/4432.misc diff --git a/changelog.d/4432.misc b/changelog.d/4432.misc new file mode 100644 index 000000000..047061ed3 --- /dev/null +++ b/changelog.d/4432.misc @@ -0,0 +1 @@ +Apply a unique index to the user_ips table, preventing duplicates. diff --git a/synapse/storage/client_ips.py b/synapse/storage/client_ips.py index 78721a941..b228a20ac 100644 --- a/synapse/storage/client_ips.py +++ b/synapse/storage/client_ips.py @@ -143,6 +143,11 @@ class ClientIpStore(background_updates.BackgroundUpdateStore): # If it returns None, then we're processing the last batch last = end_last_seen is None + logger.info( + "Scanning for duplicate 'user_ips' rows in range: %s <= last_seen < %s", + begin_last_seen, end_last_seen, + ) + def remove(txn): # This works by looking at all entries in the given time span, and # then for each (user_id, access_token, ip) tuple in that range @@ -170,7 +175,6 @@ class ClientIpStore(background_updates.BackgroundUpdateStore): SELECT user_id, access_token, ip FROM user_ips WHERE {} - ORDER BY last_seen ) c INNER JOIN user_ips USING (user_id, access_token, ip) GROUP BY user_id, access_token, ip