perf: count completed jobs on the v2_job_completed index with a timeout (#11211)

* perf: count completed jobs on the v2_job_completed index with a timeout

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>

* test: cover the count_jobs tags filter on completed jobs

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
This commit is contained in:
Ruben Fiszel
2026-09-18 08:22:59 +02:00
committed by GitHub
co-authored by Claude Opus 5
parent d440ebb520
commit d1a25360b0
2 changed files with 107 additions and 25 deletions
+60
View File
@@ -1199,3 +1199,63 @@ async fn test_wm_labels_from_result_merged_with_static_labels(
Ok(())
}
/// `tag` lives only on `v2_job`, which count_jobs joins only when `tags` is set.
#[sqlx::test(fixtures("base"))]
async fn test_count_completed_jobs_tags_filter(db: Pool<Postgres>) -> anyhow::Result<()> {
initialize_tracing().await;
let server = ApiServer::start(db.clone()).await?;
let port = server.addr.port();
let client = windmill_api_client::create_client(
&format!("http://localhost:{port}"),
"SECRET_TOKEN".to_string(),
);
for (ws, tag, status) in [
("test-workspace", "deno", "success"),
("test-workspace", "deno", "failure"),
("test-workspace", "python3", "success"),
("other-workspace", "deno", "success"),
] {
let id = uuid::Uuid::new_v4();
sqlx::query("INSERT INTO v2_job (id, workspace_id, tag) VALUES ($1, $2, $3)")
.bind(id)
.bind(ws)
.bind(tag)
.execute(&db)
.await?;
sqlx::query(
"INSERT INTO v2_job_completed (id, workspace_id, status, duration_ms) VALUES ($1, $2, $3::job_status, 0)",
)
.bind(id)
.bind(ws)
.bind(status)
.execute(&db)
.await?;
}
for (query, expected) in [
("", 3),
("tags=deno", 2),
("tags=deno&success=true", 1),
("tags=deno,python3&completed_after_s_ago=3600", 3),
] {
let response = client
.client()
.get(format!(
"{}/w/test-workspace/jobs/completed/count_jobs?{query}",
client.baseurl()
))
.send()
.await?;
assert!(
response.status().is_success(),
"{query}: {}",
response.text().await?
);
assert_eq!(response.json::<i64>().await?, expected, "{query}");
}
Ok(())
}
+47 -25
View File
@@ -4369,12 +4369,12 @@ async fn count_completed_jobs_detail(
Query(query): Query<CountCompletedJobsQuery>,
) -> error::JsonResult<i64> {
let mut sqlb = SqlBuilder::select_from("v2_job_completed");
//FOR RLS
sqlb.join("v2_job USING (id)");
sqlb.field("COUNT(*) as count");
// Filtering on v2_job.workspace_id instead would keep the planner off
// ix_job_workspace_id_completed_at_all and scan the whole retention window.
if !(w_id == "admins" && query.all_workspaces.unwrap_or(false)) {
sqlb.and_where_eq("v2_job.workspace_id", "?".bind(&w_id));
sqlb.and_where_eq("v2_job_completed.workspace_id", "?".bind(&w_id));
}
if let Some(after_s_ago) = query.completed_after_s_ago {
@@ -4393,6 +4393,7 @@ async fn count_completed_jobs_detail(
}
if let Some(tags) = query.tags {
sqlb.join("v2_job USING (id)");
sqlb.and_where_in(
"v2_job.tag",
&tags.split(",").map(|t| quote(t)).collect::<Vec<_>>(),
@@ -4400,7 +4401,19 @@ async fn count_completed_jobs_detail(
}
let sql = sqlb.sql()?;
let stats = sqlx::query_scalar::<_, i64>(&sql).fetch_one(&db).await?;
let mut tx = db.begin().await?;
set_list_jobs_statement_timeout(&mut tx).await?;
let stats = sqlx::query_scalar::<_, i64>(&sql)
.fetch_one(&mut *tx)
.await
.map_err(|e| {
list_jobs_timeout_error(
e,
"Counting completed jobs",
"Lower completed_after_s_ago or narrow the filters.",
)
})?;
tx.commit().await?;
Ok(Json(stats))
}
@@ -4429,6 +4442,33 @@ lazy_static::lazy_static! {
.unwrap_or(30);
}
/// A client that gives up does not cancel its query, so without this bound every retry of a
/// slow filter stacks another scan running until the connection-wide 5min timeout.
async fn set_list_jobs_statement_timeout(tx: &mut Transaction<'_, Postgres>) -> error::Result<()> {
let timeout_secs = *LIST_JOBS_STATEMENT_TIMEOUT_SECS;
if timeout_secs > 0 {
sqlx::query(&format!("SET LOCAL statement_timeout = '{timeout_secs}s'"))
.execute(&mut **tx)
.await?;
}
Ok(())
}
fn list_jobs_timeout_error(e: sqlx::Error, action: &str, hint: &str) -> Error {
let timeout_secs = *LIST_JOBS_STATEMENT_TIMEOUT_SECS;
match e {
sqlx::Error::Database(ref db_err)
if timeout_secs > 0 && db_err.code().as_deref() == Some("57014") =>
{
Error::Generic(
StatusCode::BAD_REQUEST,
format!("{action} took more than {timeout_secs}s and was stopped. {hint}"),
)
}
e => e.into(),
}
}
async fn list_jobs(
authed: ApiAuthed,
Extension(user_db): Extension<UserDB>,
@@ -4545,32 +4585,14 @@ async fn list_jobs(
};
// tracing::info!("sql: {}", &sql);
let mut tx: Transaction<'_, Postgres> = user_db.begin(&authed).await?;
// A client that gives up does not cancel its query, so without this bound every retry of a
// slow filter stacks another scan running until the connection-wide 5min timeout.
let timeout_secs = *LIST_JOBS_STATEMENT_TIMEOUT_SECS;
if timeout_secs > 0 {
sqlx::query(&format!("SET LOCAL statement_timeout = '{timeout_secs}s'"))
.execute(&mut *tx)
.await?;
}
set_list_jobs_statement_timeout(&mut tx).await?;
let jobs: Vec<UnifiedJob> = sqlx::query_as(&sql)
.fetch_all(&mut *tx)
.warn_after_seconds_with_sql(5, format!("list_jobs: {}", sql))
.await
.map_err(|e| match e {
sqlx::Error::Database(ref db_err)
if timeout_secs > 0 && db_err.code().as_deref() == Some("57014") =>
{
Error::Generic(
StatusCode::BAD_REQUEST,
format!(
"Listing jobs took more than {timeout_secs}s and was stopped. Set a start date or narrow the filters."
),
)
}
e => e.into(),
.map_err(|e| {
list_jobs_timeout_error(e, "Listing jobs", "Set a start date or narrow the filters.")
})?;
tx.commit().await?;