Managed Polygres: graph build jobs stuck in 'queued' forever; readiness reports ready but /graph/* returns GRAPH_NOT_READY

[clickmediapropy]

Summary

On a managed Polygres project, graph builds never execute: every polygres graph build enqueues a job that stays queued indefinitely (no worker ever picks it up). Meanwhile all readiness surfaces report ready, but every Runtime API graph query returns GRAPH_NOT_READY, which makes the inconsistency hard to diagnose from the outside.

Environment

• Managed Polygres, project pe63aae9e1dcc642022d6e7a (org "Structured Owl"), tier tier_nano (beta billing)
• Host: shared-lower-005 (pool + direct)
• CLI: polygres-cli 0.1.2, SDK polygres-sdk 0.1.0
• Graph config: 3 small tables (repos, catalog_items, join table repo_items), 2 FK relationships proposed by polygres graph discover and applied verbatim (pruned to those 3 tables). ~15/214/218 rows. invalid_reason is NULL.

Steps to reproduce

1. polygres graph discover → polygres graph config apply --file graph-config.json → polygres graph build
2. CLI prints Graph building
3. Wait 30+ minutes; repeat graph build twice more

Expected

Build runs; /graph/related answers.

Actual

• polygres graph status → {"build_status": "ready", "needs_rebuild": false, "ready": true}
• polygres ready and GET /retrieval/readiness → graph: {ready: true, status: ready, management_state: managed}
• POST /graph/related (valid start ref) → {"error": {"code": "GRAPH_NOT_READY", "message": "Graph retrieval is not ready."}} — persists 30+ min after the last build request

Inspecting the project DB over DIRECT_URL:

• graph._build_jobs: 3 rows, all status='queued', started_at NULL, worker_pid NULL, progress_message='queued for background build' (created 00:27, 00:36, and later, 2026-08-01 UTC). Latest build_id: d3292470-1def-4a44-b6f3-ae91d59a49c7.
• polygres_runtime.graph_configurations: build_status='building', needs_rebuild=t, invalid_reason NULL.
• graph._sync_log: CDC rows flowing normally for the registered tables.
• polygres_runtime.runtime_audit_events: graph_config_save and graph_activation_verification events present; no build execution events.

Impact / asks

1. Background build worker for this shared host/tier appears to not be draining the queue — can you kick it / confirm worker health for shared-lower-005?
2. Control-plane graph status / ready reporting ready: true while jobs have never run and needs_rebuild=t seems like a reconciliation bug — the query-time gate and the readiness surfaces disagree, so clients can't tell what state they're really in.

Happy to provide request IDs (e.g. req_i0RMX2VBHgPrIk2q1ZcKAg for a failing /graph/related) or run further diagnostics.

[evokoa-admin]

Hey! Yeah our queue system started having issues because of a surge of traffic. Patching now. Thank you for letting us know, and for the detailed bug report.