[ https://issues.apache.org/jira/browse/IMPALA-8179?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16765094#comment-16765094 ]
Tim Armstrong commented on IMPALA-8179: --------------------------------------- Providing a mechanism to seed them is a good idea. I disagree with seeding them by default - I don't think a test is flaky just because its randomised. If tests are randomised they need to be written carefully so that they're only testing actual invariants of the code, but we have a bunch of tests that fit that category. If we have a product bug or a poorly-written test we'd rather know than masking it by derandomising the test. > Backend tests are flaky: time-based srand > ----------------------------------------- > > Key: IMPALA-8179 > URL: https://issues.apache.org/jira/browse/IMPALA-8179 > Project: IMPALA > Issue Type: Bug > Components: Backend > Affects Versions: Impala 3.2.0 > Reporter: Jim Apple > Assignee: Joe McDonnell > Priority: Major > > https://gerrit.cloudera.org/#/c/12124/9/be/src/service/unified-betest-main.cc > added > {noformat} > uint32_t seed = time(NULL); > cout << "seed = " << seed << endl; > srand(seed); > {noformat} > There is no way to force the seed, and backend tests using {{rand()}} are > then flaky by default. We should make them unflaky by default (using a > specific random seed) and we should also provide a way to override this by > setting it by time or to a specific seed for just one run, to help with > debugging. > If we want the coverage that comes from running with a different seed each > time, we could run a test multiple times. -- This message was sent by Atlassian JIRA (v7.6.3#76005) --------------------------------------------------------------------- To unsubscribe, e-mail: issues-all-unsubscr...@impala.apache.org For additional commands, e-mail: issues-all-h...@impala.apache.org