Speed up the SQS test suite - #1679
Merged
Merged
Conversation
Every integration class paid the production defaults of a twenty second listener shutdown and a twenty second acknowledgement shutdown when its context closed. Set both to zero on the factories the tests already declare, so the setting is visible where it applies and a test can opt out, rather than through a global customizer that would override every class silently. The SQS module goes from about 2:05 to about 1:22 across the three JDKs. Acknowledgement batching is left alone. Zeroing the interval and threshold as well was measured and bought nothing: it turns every acknowledgement into its own call and the added load failed the interceptor and back pressure tests in both runs it was tried in. SqsIntegrationTests keeps its default factory untouched so its listeners still exercise the production defaults.
The container start ran inside a synchronized @BeforeAll, so the first classes scheduled held their worker threads for its whole duration. A plain blocking wait does not make the pool compensate with another thread, so the module spent its first twenty seconds with one or two tests in flight while five hundred unit tests that need no container waited behind it. Start it when discovery orders the classes, and order the classes that need no container first so they fill that window. The start is asked for only when a class in the run needs it, so a run of unit tests alone still starts nothing. The test phase goes from 74.7 to 65.9 seconds, and the first ten seconds go from one or two tests in flight to five to seven.
The class is the tail of the module and its longest methods are spent waiting out real backoff: linear ran 2+4+6 seconds and the jittered exponential 2+4+8 before the fourth delivery. The listeners derive their expected elapsed time from the same constants and assert with '>=', so the contract that backoff delays redelivery is unchanged, and the arithmetic itself is covered by JitterStrategiesTest and JitterContextTest. awspring#1676 removed the assertion floors; this reduces the wait those assertions were measuring. The test phase goes from 57.9 to 50.5 seconds.
These have no container and no IO, so the time they spend is simulated rather than waited on. CompositeBackPressureHandlerTest configured a five second no-permits timeout for a test that asserts the wait lasted at least one second, and slept a fixed 200ms to let a thread reach a blocking state. Use a two second timeout, which still proves the assertion, and observe the thread's state instead of guessing how long it takes to get there. MessageGroupingSinkTests and BatchingAcknowledgementProcessorTests sleep for a random interval to vary completion order. The variation is the point, not its magnitude, so shorten the range. The sleep whose length decides whether acknowledgements are still in flight when the processor stops is left alone, since the assertion depends on it.
tomazfernandes
requested review from
MatejNedic and
maciejwalkowiak
as code owners
August 9, 2026 18:03
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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
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.
Changes
Test code only. No test is removed or weakened.
Give the factories explicit shutdown timeouts. Every integration class paid the 20 second listener and 20 second acknowledgement shutdown defaults when its context closed. Set both to zero on the factories the tests already declare.
SqsIntegrationTestskeeps its default factory untouched, so its listeners still exercise the production defaults.Start Localstack off the worker threads. The start ran inside a synchronized
@BeforeAlland held its worker thread throughout, so the module spent its first twenty seconds with one or two tests in flight. Start it during discovery instead, and order the classes that need no container first to fill that window.Shrink the error handler backoff windows. Linear ran 2+4+6 seconds and the jittered exponential 2+4+8 before the fourth delivery. The listeners compute their expected elapsed time from the same constants and assert with
>=, so the contract is unchanged. #1675 covers the arithmetic.Stop three unit tests waiting on real time.
CompositeBackPressureHandlerTestwaited five seconds to assert a one second minimum, and slept 200ms for a thread to block. A two second timeout still proves the assertion, and the thread's state is observable directly.MessageGroupingSinkTestsandBatchingAcknowledgementProcessorTestssleep a random interval for ordering variety, so the range is shorter.Result
Two runs of each on CI against current main, launched together.
Green on all six jobs. Test counts are unchanged.