[SPARK-35215][SQL] Update custom metric per certain rows and at the end of the task #32330
Conversation
|
Kubernetes integration test starting |
|
Kubernetes integration test status failure |
|
Test build #137908 has finished for PR 32330 at commit
|
|
cc @cloud-fan |
| s"${metric.name()}") | ||
| customMetrics(metric.name()).set(metric.value()) | ||
| if (numRow % CustomMetrics.numRowsPerUpdate == 0) { | ||
| reader.currentMetricsValues.foreach { metric => |
There was a problem hiding this comment.
can we move it into a method to reuse code?
| object CustomMetrics { | ||
| private[spark] val V2_CUSTOM = "v2Custom" | ||
|
|
||
| private[spark] val numRowsPerUpdate = 100L |
There was a problem hiding this comment.
does it need to be a long?
There was a problem hiding this comment.
numRow is a long, I guess this can be just int.
| customMetrics(metric.name()).set(metric.value()) | ||
| if (numRow % CustomMetrics.numRowsPerUpdate == 0) { | ||
| partitionReader.currentMetricsValues.foreach { metric => | ||
| assert(customMetrics.contains(metric.name()), |
There was a problem hiding this comment.
I'm not sure how useful is the assert here. It's for internal error only and customMetrics(metric.name()) will fail too.
There was a problem hiding this comment.
I can remove it. I also thought it is not necessary but just added for a comment before.
|
@dongjoon-hyun Yes, I will rebase this PR. Thanks. |
|
Rebased and updated for the comments. @cloud-fan @dongjoon-hyun |
|
Kubernetes integration test starting |
|
Kubernetes integration test status failure |
|
Test build #138161 has finished for PR 32330 at commit
|
| object CustomMetrics { | ||
| private[spark] val V2_CUSTOM = "v2Custom" | ||
|
|
||
| private[spark] val numRowsPerUpdate = 100 |
There was a problem hiding this comment.
nit: NUM_ROWS_PER_UPDATE since it's a constant?
|
Kubernetes integration test starting |
|
Kubernetes integration test status failure |
|
Test build #138198 has finished for PR 32330 at commit
|
|
thanks, merging to master! |
4 participants
What changes were proposed in this pull request?
This patch changes custom metric updating to update per certain rows (currently 100), instead of per row.
Why are the changes needed?
Based on previous discussion #31451 (comment), we should only update custom metrics per certain (e.g. 100) rows and also at the end of the task. Updating per row doesn't make too much benefit.
Does this PR introduce any user-facing change?
No
How was this patch tested?
Existing unit test.