Skip to main content
Applies to: installations crossing into 6.1.1, 6.0.4 or 5.4.7 on the corresponding release line. They supplement the standard upgrade procedure. If you also cross a minor or major release, complete its upgrade notes.

Before

Redis private certificates

If you use rediss:// with a private CA and previously relied on NODE_TLS_REJECT_UNAUTHORIZED=0, configure REDIS_TLS_CA with your PEM CA bundle before upgrading. Redis now verifies certificates even when that global flag is set. Keep verification enabled. Plain Redis connections and publicly trusted certificates need no change for this fix. See environment variables for CA configuration.

MCP clients

Split legacy JSON-RPC batches larger than 100 messages. Individual MCP calls are unaffected.

Hub API and worker

For Helm targets 6.0.4 and 5.4.7, Hub migrations run in the pre-upgrade hub-migration Job. For Compose, run hub-migrate after app migrations and before starting Hub. Older Compose files do not include hub-worker; use the worker guidance when adding a worker for asynchronous processing, preserving the selected Hub version. Open a PostgreSQL client connected to Hub’s database (HUB_DATABASE_URL if configured, otherwise the application database). Run this read-only estimate; the result is an approximate row count, not a backfill:
A result of -1 means the estimate is unknown, not zero. Ask the database owner to run ANALYZE feedback_records; and retry, or use SELECT count(*) FROM feedback_records; for an exact count (which scans the table). Do not choose a shorter timeout from an unknown estimate. The release notes recommend allowing more time above roughly five million rows. Add --timeout 30m --set hub.migration.activeDeadlineSeconds=3600 to your normal Helm upgrade command. Rehearse against your data: the estimate is not a runtime guarantee. Hub also backfills missing sentiment, emotions and translations, which consumes AI-provider calls. To opt out, set ENRICHMENT_RECONCILE_ENABLED=false on the Hub services before starting the target release; with Helm, use:

One-click port binding in 6.1.1

The updater binds port 3000 to 127.0.0.1. Public access continues through Traefik on ports 80 and 443. If another host connects directly to port 3000, change that route or follow the explicit binding instructions. This installer change is documented for 6.1.1; do not assume the same behavior on the older release lines.

During

Use the procedure for your target: 4.x/5.x, or v6 Docker / Kubernetes. Keep traffic closed and writers stopped until app and Hub migrations finish; do not start a worker on the new image against an unmigrated Hub database.

After

  • Confirm Redis TLS connections succeed without disabling certificate verification.
  • Verify both Hub services use the selected image, migrations complete and background jobs resume.
  • Check sign-in, SAML if configured, MCP clients and public survey access through your proxy.
Run the deployment check when a matching check image is available. If a required check fails, keep maintenance active and follow recovery.

Recovery

Restore coordinated state and matching app/Hub images if repair cannot complete. The absence of an app schema migration does not establish that a Hub downgrade is safe. Release notes: 6.1.1, 6.0.4 and 5.4.7. These are release-source checks, not an upgrade rehearsal of your instance.