diff --git a/backend/migrations/20260405150554_cleanup_delete_indexes.down.sql b/backend/migrations/20260405150554_cleanup_delete_indexes.down.sql
new file mode 100644
index 0000000000..9f3c14d3a7
--- /dev/null
+++ b/backend/migrations/20260405150554_cleanup_delete_indexes.down.sql
@@ -0,0 +1,2 @@
+DROP INDEX IF EXISTS concurrency_key_ended_at_only_idx;
+DROP INDEX IF EXISTS autoscaling_event_applied_at_idx;
diff --git a/backend/migrations/20260405150554_cleanup_delete_indexes.up.sql b/backend/migrations/20260405150554_cleanup_delete_indexes.up.sql
new file mode 100644
index 0000000000..4d638c30b0
--- /dev/null
+++ b/backend/migrations/20260405150554_cleanup_delete_indexes.up.sql
@@ -0,0 +1,7 @@
+-- Add indexes on the time column alone to speed up the periodic cleanup DELETEs in monitor.rs:
+--   DELETE FROM concurrency_key WHERE ended_at <= now() - ($1::bigint::text || ' s')::interval
+--   DELETE FROM autoscaling_event WHERE applied_at <= now() - ($1::bigint::text || ' s')::interval
+-- The existing composite indexes (key, ended_at DESC) and (worker_group, applied_at) cannot be
+-- used efficiently when filtering only on the time column, so Postgres falls back to a seq scan.
+CREATE INDEX IF NOT EXISTS concurrency_key_ended_at_only_idx ON concurrency_key (ended_at);
+CREATE INDEX IF NOT EXISTS autoscaling_event_applied_at_idx ON autoscaling_event (applied_at);