Root cause: When mpc-service restarts, the old Kafka consumer doesn't
properly disconnect, causing the broker to wait for sessionTimeout
(~5 minutes) before completing rebalance. This blocks app startup.
Solution: Enable NestJS shutdown hooks with app.enableShutdownHooks().
This ensures onModuleDestroy() is called on SIGTERM/SIGINT, which calls
consumer.disconnect() and allows immediate rebalance on next startup.
Also reverted the "don't await consumer.run()" workaround since the
proper fix is graceful shutdown.
Sources:
- https://github.com/tulios/kafkajs/issues/807
- https://kafka.js.org/docs/consuming🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Remove Address field from PartyEndpoint (parties connect to router themselves)
- Update K8s Discovery to only manage PartyID and Role labels
- Add Party registration and SessionEvent protobuf definitions
- Implement PartyRegistry and SessionEventBroadcaster domain logic
- Add RegisterParty and SubscribeSessionEvents gRPC handlers
- Prepare infrastructure for party-driven MPC coordination
This is the first phase of migrating from coordinator-driven to party-driven
architecture following international MPC system design patterns.
hailin
Developer