pbacsko commented on PR #1045:
URL: https://github.com/apache/yunikorn-k8shim/pull/1045#issuecomment-4854526296
@adityadtu5 thanks for working on this. I know that in its current stage
it's still Draft, but I guess already has its main shape.
To me this patch is concerning, it seems more complex than necessary. It's
almost like we're overcomplicating the retry business logic: we're solving one
problem and introducing a new one.
The classification feels overwhelmingly complex:
1. Taxonomy is hard to understand (action, scope, durability)
2. "Shadow mode" - what is this?
3. Heavy error wrapping
4. Two separate retry mechanism: `AssumePod()`: 30 steps, 1s cap 30s;
`Bind()` in `task.go`: 4 steps, 100ms, cap 2s
5. Node suppression in a complex data structure
(`map[taskID]map[nodeID]time.Time`)
At a minimum, we need to use a much simpler classification mechanism like:
```
func classifyBindError(err error) (retrySame, retryDifferent, success, fail
bool) {
if err == nil { return }
if apierrors.IsAlreadyExists(err) { success = true; return }
if isTransient(err) { retrySame = true; return }
if isNodeScoped(err) { retryDifferent = true; return }
fail = true
}
```
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
