- 12.0.1: Temporal failing to start on hubs with bundled PostgreSQL (details below).
- 12.0.2: Guardian chat failing with “Sorry, I encountered an error while processing your request.” on hubs served over plain HTTP. Browsers only provide the
crypto.randomUUIDAPI on HTTPS pages, and the Guardian UI relied on it. The fix is in theonprem-12.0.2apiserver image; the other application images are byte-identical to 12.0.0.
12.0.0, so monitored clusters need no upgrade.
PostgreSQL connections (12.0.1)
On 12.0.0 the bundled PostgreSQL kept its defaultmax_connections of 100. Temporal’s 12 server pods hold up to 20 pooled connections each on the temporal database plus their visibility pools, so the last Temporal pods to start never got a connection and crashlooped with no usable database connection found. Alerts, CRD sync and Guardian workflows depend on Temporal.
12.0.1 and later:
- raise the bundled PostgreSQL
max_connectionsto 400 through the newpostgres.bundled.cluster.parametersvalue, and - size the Temporal pools explicitly: 20 connections for the default store and 5 for visibility per pod, a worst case of 300 with the default replica counts.
postgres.bundled.cluster.parameters.max_connections to match. If you use external PostgreSQL, make sure it allows at least 400 connections for Metoro.
Hub commands assume the Helm release is named
metoro and runs in the metoro-hub namespace. Adjust them if your installation uses different values.Upgrade The Hub
Start from the values file used for the existing hub release. If you do not have it locally, export the user-supplied Helm values:12.0.2:
max_connections with a rolling restart of the PostgreSQL instances, including a switchover of the primary. Wait for it to finish, then restart Temporal so any crashlooping pods reconnect immediately:
Check The Rollout
- The
metoro-postgresqlcluster reportsCluster in healthy stateandSHOW max_connectionsreturns400. - Every Temporal pod is
Runningand ready. - Existing PostgreSQL, ClickHouse, Temporal, and telemetry data remain in place.
