From b0cc806422b11a4c8e7dace3a249e8d603133e72 Mon Sep 17 00:00:00 2001
From: JSONbored <49853598+JSONbored@users.noreply.github.com>
Date: Mon, 20 Jul 2026 04:01:49 -0700
Subject: [PATCH] fix(orb): index orb_relay_pending.created_at to stop full
 table scans (#7430)

pruneRelayPending's SELECT and DELETE filter only by created_at, but both
existing indexes on orb_relay_pending lead with installation_id -- neither
could serve that predicate, so both fell back to a full table scan, run
fleet-wide on every 30s orb-relay-drain pull, occasionally exceeding the
pull request's own 30s timeout budget (Sentry: orb_relay_drain TimeoutError,
recurring since 2026-07-15).

Adds the missing created_at index (migration 0167, mirrored in the Drizzle
schema) so both queries become index range scans.
---
 ...167_orb_relay_pending_created_at_index.sql |  5 +++
 src/db/schema.ts                              |  3 ++
 ...orb-relay-pending-created-at-index.test.ts | 37 +++++++++++++++++++
 3 files changed, 45 insertions(+)
 create mode 100644 migrations/0167_orb_relay_pending_created_at_index.sql
 create mode 100644 test/unit/orb-relay-pending-created-at-index.test.ts

diff --git a/migrations/0167_orb_relay_pending_created_at_index.sql b/migrations/0167_orb_relay_pending_created_at_index.sql
new file mode 100644
index 0000000000..f12072beb9
--- /dev/null
+++ b/migrations/0167_orb_relay_pending_created_at_index.sql
@@ -0,0 +1,5 @@
+-- pruneRelayPending (src/orb/relay.ts) runs a fleet-wide TTL sweep filtered ONLY by created_at (not scoped to
+-- one installation_id), on every orb-relay-drain pull (#7430). Both existing indexes on this table lead with
+-- installation_id, so that predicate couldn't use either one -- the SELECT and DELETE both fell back to a
+-- full table scan, occasionally exceeding the pull request's 30s timeout budget.
+CREATE INDEX IF NOT EXISTS idx_orb_relay_pending_created_at ON orb_relay_pending (created_at);
diff --git a/src/db/schema.ts b/src/db/schema.ts
index aa3cd1d232..bb8be66b9a 100644
--- a/src/db/schema.ts
+++ b/src/db/schema.ts
@@ -879,6 +879,9 @@ export const orbRelayPending = sqliteTable(
     coalesce: index("idx_orb_relay_pending_coalesce")
       .on(table.installationId, table.coalesceKey)
       .where(sql`coalesce_key IS NOT NULL`),
+    // pruneRelayPending (src/orb/relay.ts) filters/deletes by created_at alone (fleet-wide TTL sweep, not
+    // scoped to one installation) -- neither index above leads with created_at, so that scan was unindexed.
+    createdAt: index("idx_orb_relay_pending_created_at").on(table.createdAt),
   }),
 );
 
diff --git a/test/unit/orb-relay-pending-created-at-index.test.ts b/test/unit/orb-relay-pending-created-at-index.test.ts
new file mode 100644
index 0000000000..328958b33c
--- /dev/null
+++ b/test/unit/orb-relay-pending-created-at-index.test.ts
@@ -0,0 +1,37 @@
+import { describe, expect, it } from "vitest";
+import { createTestEnv } from "../helpers/d1";
+
+// Regression test for #7430: pruneRelayPending (src/orb/relay.ts) filters/deletes by created_at alone
+// (a fleet-wide TTL sweep, not scoped to one installation_id). Neither pre-existing index on
+// orb_relay_pending leads with created_at, so both queries fell back to a full table scan -- occasionally
+// exceeding the orb-relay-drain pull's 30s timeout budget under fleet load.
+describe("orb_relay_pending created_at index", () => {
+  it("creates the created_at index and pruneRelayPending's queries use it (SEARCH, not SCAN)", async () => {
+    const env = createTestEnv();
+
+    const idx = await env.DB.prepare("SELECT name FROM sqlite_master WHERE type='index' AND name = ?")
+      .bind("idx_orb_relay_pending_created_at")
+      .first<{ name: string }>();
+    expect(idx?.name).toBe("idx_orb_relay_pending_created_at");
+
+    // The drop-log sample SELECT: WHERE created_at < ? ORDER BY created_at, delivery_id LIMIT ?.
+    const selectPlan = await env.DB.prepare(
+      "EXPLAIN QUERY PLAN SELECT delivery_id, event_name, installation_id FROM orb_relay_pending WHERE created_at < datetime('now', '-' || ? || ' hours') ORDER BY created_at, delivery_id LIMIT ?",
+    )
+      .bind(24, 20)
+      .all<{ detail: string }>();
+    const selectDetail = (selectPlan.results ?? []).map((row) => row.detail).join(" ");
+    expect(selectDetail).toContain("idx_orb_relay_pending_created_at");
+    expect(selectDetail).not.toContain("SCAN orb_relay_pending");
+
+    // The prune DELETE: WHERE created_at < ?.
+    const deletePlan = await env.DB.prepare(
+      "EXPLAIN QUERY PLAN DELETE FROM orb_relay_pending WHERE created_at < datetime('now', '-' || ? || ' hours')",
+    )
+      .bind(24)
+      .all<{ detail: string }>();
+    const deleteDetail = (deletePlan.results ?? []).map((row) => row.detail).join(" ");
+    expect(deleteDetail).toContain("idx_orb_relay_pending_created_at");
+    expect(deleteDetail).not.toContain("SCAN orb_relay_pending");
+  });
+});