diff options
| -rw-r--r-- | changelog.d/10359.bugfix | 1 | ||||
| -rw-r--r-- | synapse/storage/schema/state/delta/61/02state_groups_state_n_distinct.sql.postgres | 34 |
2 files changed, 35 insertions, 0 deletions
diff --git a/changelog.d/10359.bugfix b/changelog.d/10359.bugfix new file mode 100644 index 0000000000..d318f8fa08 --- /dev/null +++ b/changelog.d/10359.bugfix @@ -0,0 +1 @@ +Fix PostgreSQL sometimes using table scans for queries against `state_groups_state` table, taking a long time and a large amount of IO. diff --git a/synapse/storage/schema/state/delta/61/02state_groups_state_n_distinct.sql.postgres b/synapse/storage/schema/state/delta/61/02state_groups_state_n_distinct.sql.postgres new file mode 100644 index 0000000000..35a153da7b --- /dev/null +++ b/synapse/storage/schema/state/delta/61/02state_groups_state_n_distinct.sql.postgres @@ -0,0 +1,34 @@ +/* Copyright 2021 The Matrix.org Foundation C.I.C + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + + +-- By default the postgres statistics collector massively underestimates the +-- number of distinct state groups are in the `state_groups_state`, which can +-- cause postgres to use table scans for queries for multiple state groups. +-- +-- To work around this we can manually tell postgres the number of distinct state +-- groups there are by setting `n_distinct` (a negative value here is the number +-- of distinct values divided by the number of rows, so -0.02 means on average +-- there are 50 rows per distinct value). We don't need a particularly +-- accurate number here, as a) we just want it to always use index scans and b) +-- our estimate is going to be better than the one made by the statistics +-- collector. + +ALTER TABLE state_groups_state ALTER COLUMN state_group SET (n_distinct = -0.02); + +-- Ideally we'd do an `ANALYZE state_groups_state (state_group)` here so that +-- the above gets picked up immediately, but that can take a bit of time so we +-- rely on the autovacuum eventually getting run and doing that in the +-- background for us. |