The new launcher [mode] will run a background thread, listening to Kubernetes events and updating an internal Pod state table. The refresh operation then will be able to skip the update if status did not change.
There are many gotchas in following the event stream. But we can eventually solve them.
Potential issues:
- We get les sup to date info about Pods staying in same state (e.g. Pending)
- Event stream can get stuck or crash
- Need to handle both Pods and Jobs
- Race conditions in multi-threading
The new launcher [mode] will run a background thread, listening to Kubernetes events and updating an internal Pod state table. The refresh operation then will be able to skip the update if status did not change.
There are many gotchas in following the event stream. But we can eventually solve them.
Potential issues: