Graceful shutdown of workers
- Dominant language
- Java
- Stars
- 200
- Forks
- 73
- PR merge metrics
- No merged PRs in 30d
Description
Currently. workers are stopped by being killed. This doesn't allow any guarantees to be made as to the completion of transactions and/or observers that are currently running on each worker.
From conversations in #252, there are 3 main stop options that we should provide. The first option reflects the current behavior.
- Stop immediately (kill)
- Stop after in-progress transactions complete
- Stop after all notifications have been processed
Contributor guide
Research direction
Start by tracing how workers are currently stopped and where in-progress transactions and observers are handled. Compare the three requested stop modes—immediate kill, after transactions complete, and after notifications are processed—and identify the worker-control entry points involved. Done means each mode provides the stated completion guarantee.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- java
- Domain
- distributed-systems
- Issue type
- Feature
- Difficulty
- 5/5
- Estimated time
- Over a week
- Activity status
- Stale
- Clarity
- Mostly clear
- Newbie friendliness
- 35/100