Odd scaling behavior with low targetMessagesPerWorker

[practo]

Odd scaling behavior with low targetMessagesPerWorker

[alok87]

@nvelaborja
--resync-period=1 is very low a value. You should use default or set it higher. This sets the kubernetes events sync to 1 second. https://godoc.org/k8s.io/client-go/informers#NewSharedInformerFactory Why do you need a 1 second sync?
--sqs-short-poll-interval=1 Short poll interval is the interval at which AWS SQS is checked for queue length. With 1 second interval you may get rate limited if your queues are many by AWS. Again why do you really need it to be 1 second?
--sqs-long-poll-interval=1 If you are using long polling you should make use of the maximum number of seconds you can long poll which is 20 seconds allowed by AWS.

So unless required we should be using the default values.

TargetMessagesPerWorker is the target metric which will scale your workers from min to max with every tick if the current value is above the target. (just like CPU target scaling happens)
Target messages per worker name is given because of the calculation used to determine the number https://github.com/practo/k8s-worker-pod-autoscaler/blob/master/pkg/controller/controller.go#L355-L397

Please use the default values for the above flags and it should work.

Also it is recommended to use the latest release https://github.com/practo/k8s-worker-pod-autoscaler/releases/tag/v0.2.0 (as the above issue could also be because of the issue which was solved in #27 which came in v0.2.0)

[alok87]

Closing this as there is no odd scaling behaviour. Parameters for WPA needs to be configured properly or use defaults.
Please reopen this if you still have any concern.