diff options
| author | zsxwing <zsxwing@gmail.com> | 2015-07-31 12:10:55 -0700 |
|---|---|---|
| committer | Tathagata Das <tathagata.das1565@gmail.com> | 2015-07-31 12:10:55 -0700 |
| commit | d04634701413410938a133358fe1d9fbc077645e (patch) | |
| tree | 476d642665d6cb74613d6dc0c29033dfbefa178f /mllib/src/test | |
| parent | 3afc1de89cb4de9f8ea74003dd1e6b5b006d06f0 (diff) | |
| download | spark-d04634701413410938a133358fe1d9fbc077645e.tar.gz spark-d04634701413410938a133358fe1d9fbc077645e.tar.bz2 spark-d04634701413410938a133358fe1d9fbc077645e.zip | |
[SPARK-9504] [STREAMING] [TESTS] Use eventually to fix the flaky test
The previous code uses `ssc.awaitTerminationOrTimeout(500)`. Since nobody will stop it during `awaitTerminationOrTimeout`, it's just like `sleep(500)`. In a super overloaded Jenkins worker, the receiver may be not able to start in 500 milliseconds. Verified this in the log of https://amplab.cs.berkeley.edu/jenkins/job/SparkPullRequestBuilder/39149/ There is no log about starting the receiver before this failure. That's why `assert(runningCount > 0)` failed.
This PR replaces `awaitTerminationOrTimeout` with `eventually` which should be more reliable.
Author: zsxwing <zsxwing@gmail.com>
Closes #7823 from zsxwing/SPARK-9504 and squashes the following commits:
7af66a6 [zsxwing] Remove wrong assertion
5ba2c99 [zsxwing] Use eventually to fix the flaky test
Diffstat (limited to 'mllib/src/test')
0 files changed, 0 insertions, 0 deletions