You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Auron's current metric design is tailored for batch processing scenarios, where metrics are updated only after a task has completed.
However, in Flink environments, tasks are long-running, so we need to update metrics periodically.
For more details, see rt.rs
Activity
qingfei1994 commentedon Aug 18, 2026
I'd like to work on this issue. Is it still available?
Tartarus0zm commentedon Aug 19, 2026
hi @qingfei1994 assigned to you
Tartarus0zm commentedon Aug 31, 2026
hi @qingfei1994 is anything update for this issue?
[AURON apache#2189] Periodically publish native metrics for long-runn…
qingfei1994 commentedon Aug 31, 2026
Yes, will create a draft pull request for this.
Summary is as below:
MetricNode.adddoes not double-count.auron.metrics.update.interval.ms(default 1000; 0 = finalize-only).FlinkMetricNode.fromPlanso Kafka (including fusedProject[Filter[KafkaScan]]) and Calc share a plan-shaped metric tree forgetChild(i).[AURON apache#2189] Periodically publish native metrics for long-runn…
[AURON apache#2189] introduce auron.metrics.update.enabled to only en…
[AURON apache#2189] Hold metric_state lock across getMetrics to seria…