You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
* `testGlobalOutstandingRequestThrottlingWithRequestThrottlerDisabled`: Only resume after looking at metrics
Resuming the pipeline before looking at the metrics triggers a race between the test and other server threads, causing us to sometimes observe a larger `prep_processor_request_queued` than expected.
The corresponding failures look like this:
RequestThrottlerTest.testGlobalOutstandingRequestThrottlingWithRequestThrottlerDisabled:340 expected: <3> but was: <7>
The metric we look at is updated at the front of the pipeline, before the portion blocked by `resumeProcess`; we don't have to wait for the latter to observe the effect of the global outstanding limit.
* `testLargeRequestThrottling`: Allow more time for request ingestion
The 5s timeout used before this patch was sometimes not sufficient for the server to process the five requests of interest. The corresponding failures look like this:
RequestThrottlerTest.testLargeRequestThrottling:297 expected: <2> but was: <0>
or this:
RequestThrottlerTest.testLargeRequestThrottling:299 expected: <5> but was: <4>
We can use a higher timeout value as the `await` normally quickly terminates with the countdown reaching zero (as opposed to timing out); the increase in duration is only paid on loaded (and slow) executors.
Author: Damien Diederen <[email protected]>
Reviewers: Enrico Olivelli <[email protected]>, Norbert Kalmar <[email protected]>
Closesapache#1739 from ztzg/ZOOKEEPER-4327-flaky-request-throttler-test
0 commit comments