v2.4.25

Patch release v2.4.25

CHANGELOG:

• connection-manager: SFU WebSocket connection IDs are now UUIDs (v4) c6f4b27
  • Should be future proof for absurdly long-lived deploys where the connection manager process is isolated
• logging, code quality: improvements to redis-wrapper, bbb-gw, process-manager, connection-manager eb5d4a1 5bc8832 ce0f084 57ab49a 09ce412
• logging: add option to disable stdout logging 74a0321
  • More fine-grained control to avoid logging to two places at once
  • Flags: default.yml's log.stdout, LOG_STDOUT_ENABLED environment flag (boolean)
• connection-manager: fix a rare edge case with error handling when setting up inbound Redis clients in the main process 09ce412
• process-manager: add process respawn interval 57ab49a
  • Avoids a process forking spiral when server gets OOM'ed to hell or something botched one of the children processes
• kurento: add flag to toggle ICE state change tracking 86d2720
  • Enabling it mainly generates logs about peer's ICE state transitions in the server
  • Disabled by default while we assess any performance impact due to enabling it. If it works OK, we'll use it in the future to harden derelict endpoint cleanups
  • default.yml's kurentoTrackIceStateChanges. KURENTO_TRACK_ICE_STATE_CHANGES environment flag (boolean)
• kurento: disable mDNS candidate processing by default d14bb22
  • Commit 377435c (v2.4.23) enabled mDNS processing by default. This was a decision made after testing it in a simulated intranet AND load testing it in regular setups.
  • Seeing that it worked with no apparent drawbacks, it was enabled. It seems, though, that enabling it causes stalls in Kurento
    processing pipelines earlier than expected in some specific providers. That effectively reduces Kurento's capacity by a fair margin.
  • We've only seen that happen in one specific cloud provider. It's not generalized. But we're reverting it to disabled by default until we
    figure out what's up.