Scheduler not robust to worker dns changes
Open
networking
- Dominant language
- Python
- Stars
- 1.7k
- Forks
- 778
- Avg merge
- 2h 50m
- Merged PRs (30d)
- 3
Description
If a worker connects to the scheduler and gives a name rather than an IP address, and then subsequently the worker's IP address changes, the scheduler still tries to connect using the old IP address.
The scheduler appears to resolve the name and then cache the looked up address, leading to this issue.
Perhaps the scheduler should invalidate the cache when a worker registers with the scheduler?
Contributor guide
Assessment
This issue has not been assessed yet.