pnowojski commented on pull request #14526:
URL: https://github.com/apache/flink/pull/14526#issuecomment-754068307
> On startup, values are Nan and sometimes all boxes are red
NaNs I think come from the WebUI, when it hasn't yet managed to update those
values. I think it's better to display them as `NaN` as oppose of picking some
unknown values.
Boxes are red until the first update of metric views. Until then, `idleTime`
and `backPressuredTime` are zero, so by definition `busyTime` is `1000ms/s`.
> "Busy (max)" label - this shounds to me like max in time, not across
subtasks
How would you rename this label?
> For sources, Busy time is always NaN% (can we remove volatile
busyTimeEnabled?)
That's by design. I thought it's better to display `NaN%` to clearly
indicate value couldn't not be calculated instead of calculating `0`.
> With DataStreammAllRoundTestJob, mapper is almost as busy as it's upstream
sliding window (76% vs 84%) - this seems strange to me
If not for the load spikes from `WindowOperator`, this would be probably
100% busy for upstream and ~90% busy for the downstream operator. And that I
think would be expected if downstream operator is barely keeping up.
If you add the load spikes from `WindowOperator` to the mix, I think the
explanation is similar. Downstream operator is barely keeping up overall (on
average), however during the load spikes downstream is the bottleneck,
back-pressuring the the `WindowOperator`. This I think explains that mapper is
idling ~24% of time, with ~76% busy, while `WindowOperator` is idling 0% of the
time, with ~17% backpressured and ~83% busy.
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
