[SPARK-34015][R] Fixing input timing in gapply #31021
Closed
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
HyukjinKwon
reviewed
Jan 5, 2021
There was a problem hiding this comment.
Do you mind:
- keeping the PR description template https://github.com/apache/spark/blob/master/.github/PULL_REQUEST_TEMPLATE
- filing a JIRA, see http://spark.apache.org/contributing.html
- show how you tested this logic. Maybe show the logs before / after this change
WamBamBoozle
changed the title
|
@HyukjinKwon -- please have another look |
|
ok to test |
|
Test build #133695 has finished for PR 31021 at commit
|
|
Kubernetes integration test starting |
|
Kubernetes integration test status success |
HyukjinKwon
changed the title
HyukjinKwon
pushed a commit
that referenced
this pull request
Jan 6, 2021
### What changes were proposed in this pull request? When sparkR is run at log level INFO, a summary of how the worker spent its time processing the partition is printed. There is a logic error where it is over-reporting the time inputting rows. In detail: the variable inputElap in a wider context is used to mark the end of reading rows, but in the part changed here it was used as a local variable for measuring the beginning of compute time in a loop over the groups in the partition. Thus, the error is not observable if there is only one group per partition, which is what you get in unit tests. For our application, here's what a log entry looks like before these changes were applied: `20/10/09 04:08:58 INFO RRunner: Times: boot = 0.013 s, init = 0.005 s, broadcast = 0.000 s, read-input = 529.471 s, compute = 492.037 s, write-output = 0.020 s, total = 1021.546 s` this indicates that we're spending more time reading rows than operating on the rows. After these changes, it looks like this: `20/12/15 06:43:29 INFO RRunner: Times: boot = 0.013 s, init = 0.010 s, broadcast = 0.000 s, read-input = 120.275 s, compute = 1680.161 s, write-output = 0.045 s, total = 1812.553 s ` ### Why are the changes needed? Metrics shouldn't mislead? ### Does this PR introduce _any_ user-facing change? Aside from no longer misleading, no ### How was this patch tested? unit tests passed. Field test results seem plausible Closes #31021 from WamBamBoozle/input_timing. Authored-by: Tom.Howland <Tom.Howland@target.com> Signed-off-by: HyukjinKwon <gurwls223@apache.org> (cherry picked from commit 3d8ee49) Signed-off-by: HyukjinKwon <gurwls223@apache.org>
|
Merged to master and branch-3.1 |
|
Thanks, @WamBamBoozle. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
What changes were proposed in this pull request?
When sparkR is run at log level INFO, a summary of how the worker spent its time processing the partition is printed. There is a logic error where it is over-reporting the time inputting rows.
In detail: the variable inputElap in a wider context is used to mark the end of reading rows, but in the part changed here it was used as a local variable for measuring the beginning of compute time in a loop over the groups in the partition. Thus, the error is not observable if there is only one group per partition, which is what you get in unit tests.
For our application, here's what a log entry looks like before these changes were applied:
20/10/09 04:08:58 INFO RRunner: Times: boot = 0.013 s, init = 0.005 s, broadcast = 0.000 s, read-input = 529.471 s, compute = 492.037 s, write-output = 0.020 s, total = 1021.546 sthis indicates that we're spending more time reading rows than operating on the rows.
After these changes, it looks like this:
20/12/15 06:43:29 INFO RRunner: Times: boot = 0.013 s, init = 0.010 s, broadcast = 0.000 s, read-input = 120.275 s, compute = 1680.161 s, write-output = 0.045 s, total = 1812.553 sWhy are the changes needed?
Metrics shouldn't mislead?
Does this PR introduce any user-facing change?
Aside from no longer misleading, no
How was this patch tested?
unit tests passed. Field test results seem plausible