race condition and missing results in aggregation #732
Conversation
This file contains hidden or 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
timja
approved these changes
Sep 17, 2025
jtnord
commented
Sep 17, 2025
src/main/java/hudson/tasks/test/AggregatedTestResultPublisher.java
Outdated
Show resolved
Hide resolved
In attempting to diagnose an ATH failure I noticed what looks like a potential race condition that could lead to results not being correctly updated. For example in rare cases (but maybe not so rare in our ATH) if 2 builds finish at T1, a web request comes in a T2 and then another build fishes at T3 followed by a web request at T4. When T1 arrives the timestamp is set to T1, and then in T2 the current value is checked against the timestamp and set to T1+1 When T3 arrives the timestamp is set to T3. THe issue occurs if you have multiple CPUs and updating the code is fast and T1 = T2 = T3. When the request at T4 arrives the code thinks everything is up-to-date when it is not.
The dependencies may not be calculated when we create the action. The dependency graph is updated asynchronously by Jenkins, so we may not have the full graph when the TestResultAction is created This caused the job calculation to be fixed in stone, and potentially incorrect if the test results where obtained before the graph calculation had completed. By deferring obtaining the downstream jobs we reduce the risk of partial results being shown, and they will be eventually complete (when some run completes). We are not computing the graph so this should have little overhead on the action.
jtnord
force-pushed
the
junit-aggregation
branch
from
0c60cce to
54d4976
Compare
jtnord
changed the title
jtnord
commented
Sep 18, 2025
|
|
||
| private transient List<AbstractProject> noFingerprints; | ||
|
|
||
| @SuppressWarnings("deprecation") // calls getProject in constructor, so needs owner immediately |
There was a problem hiding this comment.
no longer true as calculation is defered, so the constructor was cleaned up (left as deprecated as it was public)
Comment on lines
163
to
165
| if (jobs == null) { | ||
| return getProject().getTransitiveDownstreamProjects(); | ||
| } |
There was a problem hiding this comment.
this fixes missing results for downstream projects if the dependency graph has not yet been rebuilt.
It may still miss some things, but we will at least be eventually consistent as Runs complete. If you have an Jenkins that quiet this could last for some time, but there is not DependencyGraphListener or another way to monitor it for changes that I am aware of.
Comment on lines
-152
to
-162
| if (jobs == null) { | ||
| // resolve null as the transitive downstream jobs | ||
| StringBuilder buf = new StringBuilder(); | ||
| for (AbstractProject p : getProject().getTransitiveDownstreamProjects()) { | ||
| if (buf.length() > 0) { | ||
| buf.append(','); | ||
| } | ||
| buf.append(p.getFullName()); | ||
| } | ||
| jobs = buf.toString(); | ||
| } |
There was a problem hiding this comment.
moved to getJobs() so this is updated as the dependencyGraph is recomputed.
Comment on lines
166
to
172
| Set<AbstractProject> projects = getProject().getTransitiveDownstreamProjects(); | ||
| return projects.stream() | ||
| .filter(AbstractProject.class::isInstance) | ||
| .map(AbstractProject.class::cast) | ||
| .filter(p -> p.hasPermission(Item.READ)) | ||
| .collect(Collectors.toSet()); | ||
| } |
There was a problem hiding this comment.
previously the code would set the string list of jobs which we would then recover below with access control checks. this puts the access control here with Item.READ.
jtnord
commented
Sep 18, 2025
| * Gets the jobs to be monitored. | ||
| */ | ||
| public Collection<AbstractProject> getJobs() { | ||
| if (jobs == null) { |
There was a problem hiding this comment.
this was the main fix for the failing internal ATH.
timja
reviewed
Sep 18, 2025
src/main/java/hudson/tasks/test/AggregatedTestResultPublisher.java
Outdated
Show resolved
Hide resolved
…java Co-authored-by: Tim Jacomb <[email protected]>
jtnord
commented
Sep 18, 2025
src/main/java/hudson/tasks/test/AggregatedTestResultPublisher.java
Outdated
Show resolved
Hide resolved
timja
approved these changes
Sep 18, 2025
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.
In attempting to diagnose an ATH failure I noticed what looks like a potential race condition that could lead to results not being correctly updated.
There is also issue where if jobs are not specified and are to be inferre,d then these are obtained from the dependency graph at action creation time, but the graph may not be correct at this point in time as it is built asynchronously.
There is also an other timing issue.
For example in rare cases (but maybe not so rare in our ATH) if 2 builds finish at T1, a web request comes in a T2 and then another build fishes at T3 followed by a web request at T4.
When T1 arrives the timestamp is set to T1, and then in T2 the current value is checked against the timestamp and set to T1+1
When T3 arrives the timestamp is set to T3.
THe issue occurs if you have multiple CPUs and updating the code is fast and T1 = T2 = T3.
When the request at T4 arrives the code thinks everything is up-to-date when it is not.
Testing done
internal ATH consistently failed before these changes, and passes after.
Submitter checklist