[Gluster-devel] Tests with timing assumptions

Jeff Darcy jdarcy at redhat.com
Fri Dec 23 17:22:29 UTC 2016


Just for "fun" I ran an experiment where I added a three-second sleep to the daemon startup code, to see which tests would fail because of timing assumptions.  A three-second delay is something that could happen *at any time*, especially on a virtualized system (such as our test systems are) with a noisy neighbor.  Tests which trip over such a delay are pretty likely to fail randomly.  I ran each test four times, twice with the sleep and twice without.  If the sleep made no difference, there'd be only a 6% chance of having both normal runs succeed and both "sleepy" runs fail.  Here's the list of tests that seem suspect on that basis.

./bugs/glusterd/bug-1242543-replace-brick.t
./bugs/glusterd/859927/repl.t
./bugs/replicate/bug-1180545.t
./bugs/replicate/bug-1341650.t
./bugs/replicate/bug-821056.t
./bugs/replicate/bug-1134691-afr-lookup-metadata-heal.t
./bugs/replicate/bug-1402730.t
./bugs/replicate/bug-1266876-allow-reset-brick-for-same-path.t
./bugs/replicate/bug-913051.t
./bugs/replicate/bug-1290965-detect-bitrotten-objects.t
./bugs/replicate/bug-865825.t
./bugs/replicate/bug-1058797.t
./bugs/replicate/bug-1046624.t
./bugs/replicate/bug-1238398-split-brain-resolution.t
./bugs/replicate/bug-918437-sh-mtime.t
./bugs/replicate/bug-1139230.t
./bugs/replicate/bug-957877.t
./bugs/replicate/bug-986905.t
./bugs/replicate/886998/strict-readdir.t
./bugs/replicate/bug-1365455.t
./bugs/replicate/bug-1190069-afr-stale-index-entries.t
./bugs/replicate/bug-977797.t
./bugs/replicate/bug-1363721.t
./bugs/replicate/bug-1238508-self-heal.t
./bugs/replicate/bug-765564.t
./bugs/replicate/bug-1297695.t
./bugs/replicate/bug-1325792.t
./bugs/replicate/bug-802417.t
./bugs/replicate/bug-1305031-block-reads-on-metadata-sbrain.t
./bugs/replicate/bug-1130892.t
./bugs/replicate/bug-1292379.t
./bugs/replicate/bug-1386188-sbrain-fav-child.t
./bugs/replicate/bug-1037501.t
./basic/afr/split-brain-favorite-child-policy.t



More information about the Gluster-devel mailing list