[
https://issues.apache.org/jira/browse/FLINK-3018?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15008728#comment-15008728
]
Stephan Ewen commented on FLINK-3018:
-------------------------------------
There are multiple execution attempts in cases of failures, they are all
remembered, but the UI shows only the latest one.
For batch jobs, that makes sense. For streaming jobs, I am unsure. To get this
right, we would need to include the metrics in the checkpointing, which is a
major change. The metrics are currently communicates asynchronously, as part of
heartbeat-like messages, they interfere with nothing.
I disagree that this is critical, though. Accumulated numbers mean little in
many continuous streaming, settings, where metrics like throughput and latency
mean much more, and look at the recent stats anyways.
> Job state discarded from web interface for restarting jobs
> ----------------------------------------------------------
>
> Key: FLINK-3018
> URL: https://issues.apache.org/jira/browse/FLINK-3018
> Project: Flink
> Issue Type: Improvement
> Components: Webfrontend
> Reporter: Gyula Fora
> Priority: Critical
>
> When a streaming job goes into the restarting status and recovers, the web ui
> information is totally lost (number of records/bytes processed etc).
> This is very misleading and should not happen.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
