Uh oh!
There was an error while loading. Please reload this page.
[SPARK-3984] [SPARK-3983] Fix incorrect scheduler delay and display task deserialization time in UI - #2832
[SPARK-3984] [SPARK-3983] Fix incorrect scheduler delay and display task deserialization time in UI#2832kayousterhout wants to merge 5 commits into
Conversation
SparkQA
commented
Oct 17, 2014
QA tests have started for PR 2832 at commit
|
SparkQA
commented
Oct 17, 2014
QA tests have finished for PR 2832 at commit
|
AmplabJenkins
commented
Oct 17, 2014
Test FAILed. |
andrewor14
commented
Oct 20, 2014
retest this please |
SparkQA
commented
Oct 20, 2014
QA tests have started for PR 2832 at commit
|
SparkQA
commented
Oct 20, 2014
QA tests have finished for PR 2832 at commit
|
AmplabJenkins
commented
Oct 20, 2014
Test FAILed. |
shivaram
commented
Oct 20, 2014
@kayousterhout - This is failing scalastyle checks -- Could you run style check locally ? |
kayousterhout
commented
Oct 20, 2014
I'm holding off on this until I finish https://issues.apache.org/jira/browse/SPARK-4016 due to the concern that otherwise these new metrics will add confusion for the average user. |
JoshRosen
commented
Oct 31, 2014
Now that I've merged #2867, this should be unblocked. |
kayousterhout
commented
Oct 31, 2014
Thanks @JoshRosen -- fixing this up now! |
JoshRosen
commented
Oct 31, 2014
Just a head's up: I merged #3031, the "hide accumulators column when empty" patch, so it's likely to cause conflicts here; you might want to merge / rebase. |
This commit fixes the scheduler delay in the UI (which previously included things that are not scheduler delay, like time to deserialize the task and serialize the result), and also adds finer-grained information to the summary table for each stage about task launch overhead (which is useful for debugging performance of short jobs, where the overhead is not-insignificant).
81fb86b to
335be4bComparekayousterhout
commented
Nov 1, 2014
Updated this so the two new metrics are hideable! |
SparkQA
commented
Nov 1, 2014
Test build #22666 has started for PR 2832 at commit
|
SparkQA
commented
Nov 1, 2014
Test build #22666 has finished for PR 2832 at commit
|
AmplabJenkins
commented
Nov 1, 2014
Test PASSed. |
kayousterhout
commented
Nov 1, 2014
@JoshRosen@andrewor14 does one of you have time to take a look at this? |
SparkQA
commented
Nov 1, 2014
Test build #22710 has started for PR 2832 at commit
|
SparkQA
commented
Nov 2, 2014
Test build #22710 has finished for PR 2832 at commit
|
AmplabJenkins
commented
Nov 2, 2014
Test PASSed. |
SparkQA
commented
Nov 5, 2014
Test build #22920 has started for PR 2832 at commit
|
SparkQA
commented
Nov 5, 2014
Test build #22920 has finished for PR 2832 at commit
|
AmplabJenkins
commented
Nov 5, 2014
Test PASSed. |
kayousterhout
commented
Nov 5, 2014
@pwendell as per our discussion, I changed this to eliminate the additional metric about task launch, so now this change just fixes the scheduler delay to be correct, and show task deserialization time in the UI. Does this look OK? |
pwendell
commented
Nov 5, 2014
LGTM! |
SparkQA
commented
Nov 5, 2014
Test build #22946 has started for PR 2832 at commit
|
SparkQA
commented
Nov 5, 2014
Test build #22946 has finished for PR 2832 at commit
|
AmplabJenkins
commented
Nov 5, 2014
Test FAILed. |
kayousterhout
commented
Nov 5, 2014
Jenkins, retest this please |
SparkQA
commented
Nov 5, 2014
Test build #22949 has started for PR 2832 at commit
|
SparkQA
commented
Nov 5, 2014
Test build #22949 has finished for PR 2832 at commit
|
AmplabJenkins
commented
Nov 5, 2014
Test PASSed. |
andrewor14
commented
Nov 5, 2014
LGTM too |
…ask deserialization time in UI This commit fixes the scheduler delay in the UI (which previously included things that are not scheduler delay, like time to deserialize the task and serialize the result), and also adds information about time to deserialize tasks to the optional additional metrics. Time to deserialize the task can be large relative to task time for short jobs, and understanding when it is high can help developers realize that they should try to reduce closure size (e.g, by including less data in the task description). cc shivaram etrain Author: Kay Ousterhout <kayousterhout@gmail.com> Closes#2832 from kayousterhout/SPARK-3983 and squashes the following commits: 0c1398e [Kay Ousterhout] Fixed ordering 531575d [Kay Ousterhout] Removed executor launch time 1f13afe [Kay Ousterhout] Minor spacing fixes 335be4b [Kay Ousterhout] Made metrics hideable 5bc3cba [Kay Ousterhout] [SPARK-3984] [SPARK-3983] Improve UI task metrics. (cherry picked from commit a46497e) Signed-off-by: Kay Ousterhout <kayousterhout@gmail.com>
kayousterhout
commented
Nov 5, 2014
Thanks for looking at this @andrewor14 and @pwendell ! I've merged it into master and 1.2. |
This commit fixes the scheduler delay in the UI (which previously
included things that are not scheduler delay, like time to
deserialize the task and serialize the result), and also
adds information about time to deserialize tasks to the optional
additional metrics. Time to deserialize the task can be large relative
to task time for short jobs, and understanding when it is high can help
developers realize that they should try to reduce closure size (e.g, by including
less data in the task description).
cc @shivaram@etrain