HPCC-32933 Fix execute timings for loop activity #19264
Conversation
Track execute timings are incurred by CATCH_NEXTROW. And remove execute timings in getNextRow as it is executed by the CNextRowLoader thread. Signed-off-by: Shamser Ahmed <[email protected]>
|
Jira Issue: https://hpccsystems.atlassian.net//browse/HPCC-32933 Jirabot Action Result: |
| @@ -439,6 +438,7 @@ class CLoopSlaveActivity : public CLoopSlaveActivityBase | |||
| } | |||
| CATCH_NEXTROW() | |||
| { | |||
| ActivityTimer t(slaveTimerStats, timeActivities); | |||
There was a problem hiding this comment.
I think this change looks correct in principle, but in practice, unless the lookahead time that is being performed by the feeder, the timing calculation may well be more inaccurate.
Signed-off-by: Shamser Ahmed <[email protected]>
| { | ||
| ret.setown(curInput->nextRow()); // more cope with groups somehow.... | ||
| LookAheadTimer t(slaveTimerStats, timeActivities); |
There was a problem hiding this comment.
hm, I think it's a bit more nuanced than this.
curInput is only the inputStream initially.. it then changes, the 2nd iteration of the loop, is not processing the input, but the result of the previous iteration. As it stands if it's doing many iterations, it will count it all against look ahead time.
I think you need to set a flag and only do until curInput changes.
…he purposes of lookahead Signed-off-by: Shamser Ahmed <[email protected]>
Track execute timings are incurred by CATCH_NEXTROW. And remove execute timings in getNextRow as it is executed by the CNextRowLoader thread. Track lookahead time.
Type of change:
Checklist:
Smoketest:
Testing: