Troubleshooting
Use the symptom to identify the failing layer, apply a targeted fix, and check the result.
On this page
The console does not open
- On the server, check
systemctl status graphlakeandjournalctl -u graphlake -n 80 --no-pager. - Check
/healthzfor the running build and/readyzfor readiness. If the service is ready locally, check SSH forwarding or the reverse proxy on the client path. - For an HTTPS deployment, confirm the real origin matches
--public-urland the proxy preservesHostandOrigin.
See Monitor and troubleshoot for cluster and ingress checks.
A model is missing or cannot be published
- No published revision. Open the model, validate its draft, then publish. Clients discover published models within their role.
- Source or column error. Browse source fields from the server. Check the process’s environment variable names, file paths, and read permissions.
- Duplicate key or invalid endpoint. Compare both ends of the mapping, including types, tenant columns, and component order. Run the sample if you need a known-good baseline.
- Revision conflict. Reload the current draft or active revision and review the changes before saving or querying again.
A query is empty, slow, or incomplete
- Empty result. Check the active revision, account and tenant filters, time scope, and direction. Reduce the query to one known key and one hop.
503 source_unavailable. Check the source in Runtime Settings → Health Checks, then verify connectivity and credentials from the GraphLake host.- Timeout or budget error. Start with a selective key, bound the hop count, and return only required fields. Use
EXPLAINto inspect the plan before increasing a budget. - Preview is truncated. Use the graph continuation or result export path for your client. Check the full-result permission for Arrow export.
- Source changed or neighbor session expired. Rerun the original query to create a fresh result and expand again.
See Query recipes for known results in all three dialects.
A local copy or CSR is stale
Check the last successful load, its watermark, and whether the model actually uses the local copy. Physical deletes and rows below the published watermark can require a full reconciliation. If a CSR generation is older than the needed data, rebuild it or query the appropriate live model.
Rerun a known-result query after the repair and compare the revision and source state. See Local tables and Graph algorithms.