[
https://issues.apache.org/jira/browse/FLINK-14206?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16937665#comment-16937665
]
Zhu Zhu edited comment on FLINK-14206 at 9/25/19 12:02 PM:
-----------------------------------------------------------
[~gjy] I think most users are using {{fullRestart}} for failure detection. What
they care is failovers, not matter it's a full or a partial restart.
I agree a new meter {{numberOfRestarts}} is better, as discussed in
FLINK-14164. But I'm not sure whether it's a good idea to introduce a new
metric into Flink 1.9.1. And [~trohrmann] prefers not to do that in 1.9
([ref|https://issues.apache.org/jira/browse/FLINK-14164?focusedCommentId=16937503&page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#comment-16937503]).
So changing {{fullRestart}} is a workaround if we'd like to support Flink 1.9
to measure failovers when fine grained recovery is enabled.
However, whether to keep {{fullRestart}} to measure only global restarts is
under discussion in FLINK-14164. I will keep an eye on it.
was (Author: zhuzh):
[~gjy] I think most users are using {{fullRestart}} for failure detection. What
they care is failovers, not matter it's a full or a partial restart.
I agree a new meter {{numberOfRestarts}} is better, as discussed in
FLINK-14164. But I'm not sure whether it's a good idea to introduce it into
Flink 1.9. And [~trohrmann] prefers not to do that in 1.9
([ref|https://issues.apache.org/jira/browse/FLINK-14164?focusedCommentId=16937503&page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#comment-16937503]).
So changing {{fullRestart}} is a workaround if we'd like to support Flink 1.9
to measure failovers when fine grained recovery is enabled.
However, whether to keep {{fullRestart}} to measure only global restarts is
under discussion in FLINK-14164. I will keep an eye on it.
> Let fullRestart metric count fine grained restarts as well
> ----------------------------------------------------------
>
> Key: FLINK-14206
> URL: https://issues.apache.org/jira/browse/FLINK-14206
> Project: Flink
> Issue Type: Improvement
> Components: Runtime / Coordination
> Affects Versions: 1.9.0
> Reporter: Zhu Zhu
> Priority: Major
> Fix For: 1.9.1
>
>
> With fine grained recovery introduced in 1.9.0, the {{fullRestart}} metric
> only counts how many times the entire graph has been restarted, not including
> the number of fine grained failure restarts.
> As many users leverage this metric for failure detecting monitoring and
> alerting, I'd propose to make it also count fine grained failure restarts.
> The concrete proposal is:
> 1. Add a counter {{numberOfRestartCounter}} in ExecutionGraph to count all
> restarts. The counter is not to be registered to metric groups.
> 2. Let {{fullRestart}} query the value of the counter, instead of
> {{ExecutionGraph#globalModVersion}}
> 3. increment {{numberOfRestartCounter}} in {{ExecutionGraph#failGlobal}}
> 4. increment {{numberOfRestartCounter}} in
> {{ExecutionGraph#notifyExecutionChange}} where notifying the failover
> strategy, or maybe in {{AdaptedRestartPipelinedRegionStrategyNG}} to only
> count failovers really happened
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
