Skip to main content
Applies to: runtime changes in 6.1.0 and the one-click port-binding change in 6.1.1. Back to the 6.1 checklist.

Bundled TEI log level

The Helm chart now defaults hub.embeddings.env.RUST_LOG to warn because TEI 1.9 includes its API key in INFO-level startup arguments. If you need a different log level, set hub.embeddings.env.RUST_LOG explicitly and use it only with a runtime that redacts credentials from its logs.

Database transaction limits

Formbricks now sets its database transaction limits explicitly instead of relying on library defaults. Under load, the old 5-second limit was cancelling writes that were about to succeed: user identification, response updates and survey resets. The new limits are at least as permissive as the old ones, so they need no upgrade step of their own.
  • Transactions may run for up to 20 seconds.
  • Waiting for a free database connection is bounded by connect_timeout in DATABASE_URL (5 seconds by default). With connect_timeout=0, transactions give up waiting after 6 seconds; other queries wait until a connection is free.

Idle transactions (opt-in)

PostgreSQL can end any session left idle inside an open transaction, which releases the locks that session holds and frees its connection. Formbricks does not ask for this by default: it travels as a startup parameter, and a server or pooler that does not accept the parameter refuses the whole connection instead of ignoring it. Turn it on by adding it to DATABASE_URL in milliseconds:
Leaving the parameter out, or setting it to 0, sends nothing and keeps your server’s own setting. An invalid value (anything but digits, or past PostgreSQL’s 32-bit limit) sends nothing and logs a warning. It applies to every process that connects with that DATABASE_URL — the migration runner uses MIGRATE_DATABASE_URL instead when that is set, so it follows that one. Keep the value well above the longest a migration leaves a transaction idle, or leave the parameter out.
Do not set it if Formbricks reaches PostgreSQL through PgBouncer, Supavisor, RDS Proxy or another pooler: PgBouncer answers a connection carrying a startup parameter it does not know with FATAL: unsupported startup parameter: idle_in_transaction_session_timeout. Listing the parameter in PgBouncer’s ignore_startup_parameters lets the connection through, but PgBouncer then drops it, so the safeguard does not apply either way. If DATABASE_URL carries both the parameter and pgbouncer=true, Formbricks sends it and logs a warning.

6.1.1: one-click port binding

The one-click installer now publishes the Formbricks container’s port 3000 on 127.0.0.1 instead of every interface, and ./formbricks.sh update changes an existing one-click installation the same way, in docker-compose.yml and in your Compose override file if you have one, keeping each original as <file>.before-loopback-port. Traefik keeps serving Formbricks on ports 80 and 443, so the app is only reachable from outside through Traefik and its TLS and request limits. For that to hold against other machines on the same local network, Docker Engine must be 28.0 or later and not 28.2.0 to 28.3.2; the installer and the update warn on those versions. If you reach port 3000 from other machines on purpose, publish it on an explicit address after updating, such as - 0.0.0.0:3000:3000 (add - "[::]:3000:3000" for IPv6). The update only rewrites the bare - 3000:3000 mapping, so it leaves an address you chose alone; putting the bare mapping back would be narrowed again next time. Compose setups that don’t use the one-click script are not changed. If yours publishes 3000:3000 and a reverse proxy on the same host is the intended way in, bind it to 127.0.0.1:3000:3000 as well.

Hub worker for Docker

Fresh Docker and one-click installations add hub-worker in 6.1. Earlier v5/v6.0 Compose files do not include it, although Helm already has a worker. The worker consumes every PostgreSQL-backed asynchronous Hub job, including webhook deliveries and any configured enrichment, so an installation that runs the Hub API without it leaves those jobs queued. An existing installation keeps its local docker-compose.yml during formbricks.sh update, so the updater cannot add the service for you. Adding it is recommended rather than required in this release: the updater prints a warning and continues when the resolved Compose project does not contain hub-worker. Keep the worker enabled so asynchronous processing completes; do not infer a future enforcement date from this warning.
1

Download the release configuration

In the directory containing your existing docker-compose.yml, download the exact target release. That step sets target_version and saves docker-compose.release.yml. Back up your working file:
2

Add the worker and shared settings

Copy the reference file’s x-hub-runtime-environment block to the top level of your Compose file. Replace hub and add hub-worker under services:, including their shared environment and dependencies, so the API and worker receive the same database and taxonomy configuration. Preserve local proxy, storage, database, and image-pin customizations. hub, hub-worker, and hub-migrate must resolve to the same Hub image reference.
3

Validate the configuration

Validate the result before changing the running stack:
4

Apply migrations in maintenance

During the maintenance procedure, keep app/Hub writers stopped and complete application migrations first. For a one-click installation, rerun ./formbricks.sh update — the warning disappears once the resolved project contains hub-worker. For a manually managed Compose installation, apply Hub migrations before starting the API and worker from the new image:
5

Verify the worker

Verify that the worker stays running and that the existing River backlog decreases:
The worker has no HTTP health endpoint. A stable process and decreasing queue are the readiness signals. If it restarts, verify PostgreSQL connectivity and every configured enrichment provider credential. Jobs already in River remain stored in PostgreSQL and begin draining when the worker starts; no repair or re-enqueue is required. To roll back only this change, remove the worker container before restoring the backed-up Compose file so its restart: always policy cannot start it again:
Completed jobs remain committed and pending jobs remain queued, but all Hub asynchronous processing pauses again. A fully custom deployment that runs Hub workers outside this Compose project can ignore the warning: the update is not blocked.

Cube model files for Docker

6.1 adds the npsCount measure and the valueBand dimension to the Cube model. The NPS breakdown and CSAT breakdown chart presets use them. On Docker the model is two files next to your Compose file, cube/cube.js and cube/schema/FeedbackRecords.js. Until they are refreshed, those presets fail with 'npsCount' not found.
  • One-click: ./formbricks.sh update, run with the target release’s updater as described under one-click upgrade, refreshes both files while the stack is stopped. It takes them from the release your Formbricks image runs, and keeps the previous copies in cube/previous/. If either file cannot be refreshed, both stay as they were, and the update finishes and prints the command to retry. If the image is pinned to a digest or a custom tag, the updater leaves both files alone; use the Docker Compose step below with your release’s tag.
  • Docker Compose: use the Cube model download and installation steps from the directory containing your docker-compose.yml. If you refresh outside an update, recreate Cube afterwards with docker compose up -d --no-deps --force-recreate cube.
  • Helm: the chart ships the model, so helm upgrade replaces it. No action is needed.

Envoy Redis images

Chart 6.1.0’s bundled Envoy Redis images are unavailable upstream. Configure a dedicated external Redis/Valkey backend and keep envoyRedis.enabled=false; see Envoy Gateway. Keep the rate-limit backend separate from the application’s Redis traffic.

Helm security contexts and taxonomy timeouts

  • Helm: the web container is locked down. It now runs with a read-only root filesystem, as non-root user 1001, with all Linux capabilities dropped. This only affects custom images or sidecars that write inside the container: mount a writable path with deployment.extraVolumes and deployment.extraVolumeMounts, or set deployment.containerSecurityContext.readOnlyRootFilesystem: false. If you set deployment.securityContext.runAsUser, move it to deployment.containerSecurityContext.runAsUser, which now takes precedence (#9450)
  • Helm: AuthZed hooks run as user 1001. If your custom image needs another user, set authzed.initialization.securityContext.runAsUser (#9516)
  • Helm: taxonomy timeout overrides are validated. With taxonomy.enabled=true, rendering fails unless taxonomy.terminationGracePeriodSeconds is at least taxonomy.runDeadlineSeconds + 30, and taxonomy.hubStaleRunTimeoutSeconds is greater than taxonomy.runDeadlineSeconds. The defaults (900, 930 and 1800 seconds) pass (#9451)

Auth environment aliases

New installations must set BETTER_AUTH_URL and BETTER_AUTH_SECRET. The 6.1 runtime resolves the auth URL from BETTER_AUTH_URL, then NEXTAUTH_URL; it does not fall back to WEBAPP_URL. Existing NEXTAUTH_URL and NEXTAUTH_SECRET remain supported; no rename is needed. If renaming, copy the existing values. A different secret invalidates sessions and outstanding authentication links. Keep a configured secret in test environments too; the current runtime rejects a missing value.