permitio/opal 0.9.9-rc.3

on GitHub

What's Changed

• fix(opal-server): PR3 follow-ups from the staging campaign — fleet purge to client-less workers, boot metrics namespace, watcher flag gate, log volume, broadcaster TCP keepalive by @Zivxx in #945

Highlights:

• Broadcaster resilience (P9): permit-broadcaster 0.2.6 → 0.2.7 enables TCP keepalive on every Postgres backbone connection (on by default, 30/10/3). A broadcast DB whose address goes silent — e.g. an RDS Multi-AZ failover — is now detected within ~60 s and the existing reconnect+resync path recovers; previously the reader stayed connected-and-deaf indefinitely while health checks stayed green. Note: only add keepalives* parameters to OPAL_BROADCAST_URI once every server sharing it runs ≥ rc.3 (see the configuration docs); the defaults need no URI change.
• Fleet purge reaches client-less workers (P5): every scopes worker now holds the broadcast listening context for its lifetime. Ops note: budget the broadcast DB's max_connections for 10 × workers × pods during a rolling restart (asyncpg pool min_size).
• Boot-phase metrics are namespaced (P1), OPAL_REPO_WATCHER_ENABLED properly gates the scopes watcher (P3), sync-failure log volume is bounded (P6).

Full Changelog: 0.9.9-rc.2...0.9.9-rc.3