[ML] Close results stream before data frame analytics job stops #67828
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
Investigating the failures in elastic#67581 it looked like after restarting the regression job the process was started but no data was loaded. So the process was getting stuck waiting for data. Looking into the code it looks like this can be explained by the fact that `AnalyticsResultProcessor` counts down its completion latch before it closes the results stream. This means the job may go to `stopped` state while the out stream is still alive, which on windows results to the directory with the named pipes staying around. Then when the job is started again, which the test does immediately, the old pipes are used and thus the data is not sent to the the new process. This commit fixes this while refactoring ML processes to consume and close the out stream in their close method so that calling code is not responsible for doing that. Fixes elastic#67581
|
Pinging @elastic/ml-core (:ml) |
| @@ -117,7 +117,6 @@ public void process(AnalyticsProcess<AnalyticsResult> process) { | |||
| completeResultsProgress(); | |||
| } | |||
| completionLatch.countDown(); | |||
| process.consumeAndCloseOutputStream(); | |||
There was a problem hiding this comment.
Here is the offending code piece. Latch down was counted down before we close output stream.
|
@elasticmachine update branch |
|
It turns out this fix is not correct. We cannot consume and close the output stream in the process close method. The first thing close does it to close the processes input stream. This signals the process termination. However, the process might be writing out results as a result of terminating. Thus, we need the result processors to keep being able to read results and only after those results are all written out close the output stream. I'll come back with a different fix. |
dimitris-athanasiou
deleted the
close-results-stream-before-dfa-job-stops
branch
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.
Investigating the failures in #67581 it looked like after restarting
the regression job the process was started but no data was loaded.
So the process was getting stuck waiting for data.
Looking into the code it looks like this can be explained by the
fact that
AnalyticsResultProcessorcounts down its completionlatch before it closes the results stream. This means the job
may go to
stoppedstate while the out stream is still alive,which on windows results to the directory with the named pipes
staying around. Then when the job is started again, which the
test does immediately, the old pipes are used and thus the
data is not sent to the the new process.
This commit fixes this while refactoring ML processes to consume
and close the out stream in their close method so that calling
code is not responsible for doing that.
Fixes #67581