1 Commits
Author SHA1 Message Date
jcoffey-dev fde43774b4 PostgreSQL search GIN indexes without a pending list
ci / fork-checks (pull_request) Successful in 30s
ci / build (pull_request) Successful in 7m26s
A three-node rehearsal on PostgreSQL saw searches take about 185 ms
with 80 to 260 pages in the full-text indexes' pending lists, 2 to 6 ms
right after gin_clean_pending_list() or VACUUM, then creep back up as
mail came in. The search tables' GIN indexes were created with the
default fastupdate=on: new entries wait in an unindexed pending list
that every search scans in full until VACUUM (or 4 MB of backlog)
merges it, and autovacuum only visits an insert-only table after
thousands of inserts.

The search GIN indexes are now created WITH (fastupdate = off), so an
insert pays its index update at once. The schema step runs at every
startup (create_search_tables, via SearchStore::create_indexes), so
indexes made before this change are switched there: when an index's
reloptions don't already turn fastupdate off, ALTER INDEX ... SET
(fastupdate = off) and one gin_clean_pending_list() merge its backlog.
The ALTER takes a SHARE UPDATE EXCLUSIVE lock, which blocks neither
reads nor writes; after the first startup the step is one catalog read
per index. A failure is logged and startup goes on (search still
works, only slower).

Per-table autovacuum settings for the search tables are left alone.
The pending list was the only reason the insert threshold mattered for
search; dead tuples and freezing are served by the defaults, and table
settings would override whatever tuning the DBA has done.

MySQL is unaffected: InnoDB FULLTEXT keeps new entries in an in-memory
cache that queries read directly, with no setting like fastupdate.

store::search_gin::postgres_gin_fastupdate (new, PostgreSQL) builds
the search schema in a schema of its own and checks pg_class.reloptions:
fastupdate=off on every GIN index of a fresh schema; then, with the
option reset to the default and 500 rows pending, one startup turns it
off everywhere and leaves no pending tuples (pgstatginindex); a second
startup changes nothing. On main it fails at the first check.
2026-09-24 15:51:06 -07:00
3 changed files with 234 additions and 1 deletions
+73 -1
View File
@@ -265,12 +265,21 @@ async fn create_search_tables<T: SearchableField + PsqlSearchField + 'static>(
for field in T::all_fields() {
if field.is_text() || field.is_json() {
let column_name = field.column();
// inbuxa: with GIN's default fastupdate=on, new entries wait in
// an unindexed pending list that every search scans in full
// until a VACUUM (or 4 MB of backlog) merges it. On a mailbox
// taking steady mail that list never drains and searches slow
// from milliseconds to hundreds of them. Pay the index update
// at insert time instead.
let index_name = format!("gin_{table_name}_{column_name}");
let create_index_query = format!(
"CREATE INDEX IF NOT EXISTS gin_{table_name}_{column_name} ON {table_name} USING GIN({column_name})",
"CREATE INDEX IF NOT EXISTS {index_name} ON {table_name} USING GIN({column_name}) WITH (fastupdate = off)",
);
conn.execute(&create_index_query, &[])
.await
.map_err(into_error)?;
// Indexes made before this change keep fastupdate=on
disable_gin_fastupdate(conn, &index_name).await;
}
if field.is_indexed() {
@@ -287,6 +296,69 @@ async fn create_search_tables<T: SearchableField + PsqlSearchField + 'static>(
Ok(())
}
/// inbuxa: turns fastupdate off on a GIN index made with the default and
/// merges the pending list it has built up. Idempotent: an index that already
/// has the option is left alone, so this costs one catalog read per index at
/// startup. A failure is logged and startup goes on, since search still works,
/// only slower.
async fn disable_gin_fastupdate(conn: &Object, index_name: &str) {
if let Err(err) = try_disable_gin_fastupdate(conn, index_name).await {
trc::event!(
Store(trc::StoreEvent::PostgresqlError),
Details = format!("Failed to turn off fastupdate on search index {index_name}"),
Reason = err.to_string(),
);
}
}
async fn try_disable_gin_fastupdate(conn: &Object, index_name: &str) -> trc::Result<()> {
let options = conn
.query_opt(
"SELECT COALESCE(reloptions, '{}')::text[] FROM pg_class WHERE oid = to_regclass($1)",
&[&index_name],
)
.await
.map_err(into_error)?
.map(|row| row.try_get::<_, Vec<String>>(0))
.transpose()
.map_err(into_error)?;
let Some(options) = options else {
return Ok(());
};
if gin_fastupdate_is_off(&options) {
return Ok(());
}
// SET (fastupdate) takes a SHARE UPDATE EXCLUSIVE lock, which doesn't
// block reads or writes. Turning it off stops new entries going to the
// pending list but doesn't flush the entries already there.
conn.execute(
&format!("ALTER INDEX {index_name} SET (fastupdate = off)"),
&[],
)
.await
.map_err(into_error)?;
conn.query_one(
"SELECT gin_clean_pending_list($1::text::regclass)",
&[&index_name],
)
.await
.map_err(into_error)?;
Ok(())
}
/// Whether a relation's reloptions turn GIN's fastupdate off.
fn gin_fastupdate_is_off(options: &[String]) -> bool {
options.iter().any(|option| {
option.split_once('=').is_some_and(|(name, value)| {
name.trim().eq_ignore_ascii_case("fastupdate")
&& matches!(
value.trim().to_ascii_lowercase().as_str(),
"off" | "false" | "no" | "0" | "f" | "n"
)
})
})
}
async fn discover_ts_configs(pool: &Pool) -> AHashSet<&'static str> {
let mut ts_configs = AHashSet::from_iter([PG_FALLBACK_LANG, PG_UNSTEMMED_LANG]);
+2
View File
@@ -21,6 +21,8 @@ pub mod replica_mysql; // inbuxa: read replicas on MySQL
#[cfg(all(feature = "postgres", feature = "redis"))]
pub mod replica_cluster; // inbuxa: read replicas across nodes
pub mod scaleout; // inbuxa: scale-out storage
#[cfg(feature = "postgres")]
pub mod search_gin; // inbuxa: GIN indexes without a pending list
#[cfg(any(feature = "postgres", feature = "mysql"))]
pub mod sql_timeout;
pub mod task_locks; // inbuxa: task locks across nodes
+159
View File
@@ -0,0 +1,159 @@
/*
* SPDX-FileCopyrightText: 2026 Coffey Labs
*
* SPDX-License-Identifier: AGPL-3.0-only
*/
//! PostgreSQL full-text GIN indexes are built with fastupdate off, and an
//! index made earlier with the default is switched over at startup. With
//! fastupdate on, new entries wait in a pending list that every search scans
//! in full until VACUUM merges it.
use crate::utils::storage::build_data_store;
use registry::schema::structs::DataStore;
use store::{Rows, SearchStore, Store};
const SCHEMA: &str = "gin_fastupdate_test";
#[tokio::test(flavor = "multi_thread")]
pub async fn postgres_gin_fastupdate() {
println!("Running PostgreSQL GIN fastupdate test...");
// Work in a schema of our own so the shared search tables are untouched
let admin = Store::build(build_data_store("PostgreSql", "").await)
.await
.expect("Failed to connect to PostgreSQL");
for query in [
format!("DROP SCHEMA IF EXISTS {SCHEMA} CASCADE"),
format!("CREATE SCHEMA {SCHEMA}"),
] {
admin.sql_query::<usize>(&query, vec![]).await.unwrap();
}
let DataStore::PostgreSql(mut config) = build_data_store("PostgreSql", "").await else {
unreachable!()
};
config.options = Some(format!("-c search_path={SCHEMA}"));
let store = Store::build(DataStore::PostgreSql(config))
.await
.expect("Failed to connect to PostgreSQL");
let search = SearchStore::Store(store.clone());
// A fresh schema
search.create_indexes().await.unwrap();
let indexes = gin_indexes(&admin).await;
assert!(
indexes.len() >= 4,
"expected the search GIN indexes, found {indexes:?}"
);
for (name, options) in &indexes {
assert!(
options.contains("fastupdate=off"),
"fresh index {name} has options {options:?}"
);
}
// A schema from before the change: the same indexes, made with the
// default fastupdate=on, and a pending list with something in it
for (name, _) in &indexes {
admin
.sql_query::<usize>(
&format!("ALTER INDEX {SCHEMA}.{name} RESET (fastupdate)"),
vec![],
)
.await
.unwrap();
}
for (name, options) in gin_indexes(&admin).await {
assert!(
!options.contains("fastupdate"),
"index {name} still has options {options:?}"
);
}
admin
.sql_query::<usize>(
&format!(
"INSERT INTO {SCHEMA}.s_email (accid, docid, subj, body) \
SELECT 1, n, to_tsvector('simple', 'pending subject ' || n), \
to_tsvector('simple', 'pending body text ' || n) \
FROM generate_series(1, 500) n"
),
vec![],
)
.await
.unwrap();
assert!(
pending_tuples(&admin, "gin_s_email_body").await > 0,
"no pending list to merge"
);
// Startup on the existing schema switches every index over and merges
// what was pending
search.create_indexes().await.unwrap();
for (name, options) in gin_indexes(&admin).await {
assert!(
options.contains("fastupdate=off"),
"existing index {name} has options {options:?} after startup"
);
}
assert_eq!(pending_tuples(&admin, "gin_s_email_body").await, 0);
// And a second startup changes nothing
search.create_indexes().await.unwrap();
for (name, options) in gin_indexes(&admin).await {
assert!(options.contains("fastupdate=off"), "{name}: {options:?}");
}
admin
.sql_query::<usize>(&format!("DROP SCHEMA {SCHEMA} CASCADE"), vec![])
.await
.unwrap();
}
/// The GIN indexes in the test schema with their reloptions.
async fn gin_indexes(admin: &Store) -> Vec<(String, String)> {
admin
.sql_query::<Rows>(
&format!(
"SELECT c.relname::text, COALESCE(array_to_string(c.reloptions, ','), '') \
FROM pg_class c JOIN pg_namespace n ON n.oid = c.relnamespace \
JOIN pg_am a ON a.oid = c.relam \
WHERE n.nspname = '{SCHEMA}' AND c.relkind = 'i' AND a.amname = 'gin' \
ORDER BY 1"
),
vec![],
)
.await
.unwrap()
.rows
.into_iter()
.map(|row| {
let mut values = row.values.into_iter();
(
values.next().unwrap().to_str().into_owned(),
values.next().unwrap().to_str().into_owned(),
)
})
.collect()
}
/// Tuples waiting in a GIN index's pending list (pgstattuple is a contrib
/// extension the test database has).
async fn pending_tuples(admin: &Store, index: &str) -> i64 {
admin
.sql_query::<usize>("CREATE EXTENSION IF NOT EXISTS pgstattuple", vec![])
.await
.unwrap();
admin
.sql_query::<Rows>(
&format!("SELECT pending_tuples FROM pgstatginindex('{SCHEMA}.{index}'::regclass)"),
vec![],
)
.await
.unwrap()
.rows
.into_iter()
.next()
.and_then(|row| row.values.into_iter().next())
.map(|value| value.to_str().parse::<i64>().unwrap())
.unwrap()
}