[
https://issues.apache.org/jira/browse/FLINK-31867?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18089115#comment-18089115
]
Dennis-Mircea Ciupitu commented on FLINK-31867:
-----------------------------------------------
I think this is largely addressed now and could be closed.
- {{JobAutoScalerImpl}} enforces a minimum of two observations before
evaluating: {{if (collectedMetrics.getMetricHistory().size() < 2) return;}}
(added in FLINK-34266/FLINK-34213, after this ticket was filed).
- FLINK-39826 closed the related misconfiguration: {{DefaultValidator}} now
rejects {{job.autoscaler.metrics.window <
kubernetes.operator.reconcile.interval}}, since otherwise the window holds a
single sample and the two-sample minimum is never reached.
This could still be relevant if the intent is to make the minimum configurable
rather than always defaulting to 2. That said, the floor of 2 is what
guarantees the correctness of the subsequent autoscaler calculations, so it
shouldn't go below that. If users want more observations within a window, the
right lever is already there: size the metric window relative to the reconcile
interval so the window holds as many samples as they need. (cc [~gyfora])
> Enforce a minimum number of observations within a metric window
> ---------------------------------------------------------------
>
> Key: FLINK-31867
> URL: https://issues.apache.org/jira/browse/FLINK-31867
> Project: Flink
> Issue Type: Improvement
> Components: Autoscaler, Kubernetes Operator
> Reporter: Maximilian Michels
> Priority: Major
> Labels: pull-request-available, stale-assigned
>
> The metric window is currently only time-based. We should make sure we see a
> minimum number of observations to ensure we don't decide based on too few
> observations.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
