Bugfixes in RFS work coordination update queries to claim leases #769
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
The easy one which probably wasn't ever causing issues is to convert 409 conflict responses immediately to a VERSION_CONFLICT result rather than waiting to parse the response body (& possibly be surprised). The other was to set the operation to noop for the update_by_query result when the conditional checks (like whether the new expiration is > the old expiration). That removes a big (& only?) pressure where the subsequent call to get the assigned document would return 0 results - which made sense since no document was actually updated, even though the update_by_query reported that one was. Signed-off-by: Greg Schohn <[email protected]>
Codecov ReportAttention: Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## main #769 +/- ##
============================================
+ Coverage 68.51% 68.53% +0.01%
+ Complexity 1578 1577 -1
============================================
Files 275 275
Lines 11596 11660 +64
Branches 735 737 +2
============================================
+ Hits 7945 7991 +46
- Misses 3248 3263 +15
- Partials 403 406 +3
Flags with carried forward coverage won't be shown. Click here to find out more. ☔ View full report in Codecov by Sentry. |
gregschohn
requested review from
AndreKurait,
chelma,
lewijacn,
mikaylathompson,
okhasawn,
peternied and
sumobrian
as code owners
AndreKurait
reviewed
Jun 26, 2024
| @@ -400,6 +402,9 @@ UpdateResult assignOneWorkItem(long expirationWindowSeconds) throws IOException | |||
|
|
|||
| var response = httpClient.makeJsonRequest(POST_METHOD, INDEX_NAME + "/_update_by_query?refresh=true&max_docs=1", | |||
| null, body); | |||
| if (response.getStatusCode() == 409) { | |||
| return UpdateResult.VERSION_CONFLICT; | |||
There was a problem hiding this comment.
I think we need to check for resultTree['noop'] > 0 on the below elseIf [line 413]
There was a problem hiding this comment.
Thanks - addressed in the latest commit (if I understood the concern)
… the server-side/peer clocks don't agree that an item has been expired. That results in no update happening and used to just throw. Now it throws and the client backs off a bit before retrying again. The FullTest now puts some clock jitter into each of its clients and it's no longer seeing an IllegalStateException (or the newly subclassed PotentialClockDriftDetectedException) with the retries in place. Signed-off-by: Greg Schohn <[email protected]>
…hohn/opensearch-migrations into FixFalseUpdatesForRFSCoordination
gregschohn
changed the title
…ething else altogether. Drift is retryable, other issues aren't. We should probably get more careful about other response codes (or timeouts), but that's a bigger and more sweeping change. Signed-off-by: Greg Schohn <[email protected]>
AndreKurait
approved these changes
Jun 26, 2024
peternied
reviewed
Jun 26, 2024
| assert numUpdated <= 1; | ||
| if (numUpdated > 0) { | ||
| return UpdateResult.SUCCESSFUL_ACQUISITION; | ||
| } else if (resultTree.path(VERSION_CONFLICTS_FIELD_NAME).longValue() > 0) { | ||
| return UpdateResult.VERSION_CONFLICT; | ||
| } else if (resultTree.path("total").longValue() == 0) { | ||
| return UpdateResult.NOTHING_TO_ACQUIRE; | ||
| } else if (noops > 0) { | ||
| throw new PotentialClockDriftDetectedException("Found " + noops + |
There was a problem hiding this comment.
VERSION_CONFLICT isn't being handled with an exception, what do you think about adding UpdateResult.CLOCK_DRIFT and log the number of noops found here?
This cleans up the logic around acquireNextWorkItem(...) to be more consistent.
Nothing is calling that code that is changed, can we add some validation since there are some clear off-by-one errors possible? |
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.
The easy one which probably wasn't ever causing issues is to convert 409 conflict responses immediately to a VERSION_CONFLICT result rather than waiting to parse the response body (& possibly be surprised). The other was to set the operation to noop for the update_by_query result when the conditional checks (like whether the new expiration is > the old expiration). That removes a big (& only?) pressure where the subsequent call to get the assigned document would return 0 results - which made sense since no document was actually updated, even though the update_by_query reported that one was.
Description
Issues Resolved
https://opensearch.atlassian.net/jira/software/c/projects/MIGRATIONS/boards/1?assignee=63d1918f69c7ae3958d1f57f&selectedIssue=MIGRATIONS-1810
Testing
Ran through the FullTest, observing none of the "The query for the assigned work document returned 0" errors. The test still doesn't assert that that shouldn't happen though.
Check List
By submitting this pull request, I confirm that my contribution is made under the terms of the Apache 2.0 license.
For more information on following Developer Certificate of Origin and signing off your commits, please check here.