On Thu, Sep 7, 2017 at 11:55 AM, Eryu Guan <eguan@xxxxxxxxxx> wrote: >> >> +NUM_FILES=4 >> >> +NUM_OPS=200 >> >> +FSX_OPTS="-N $NUM_OPS -d -P $SANITY_DIR -i $LOGWRITES_DMDEV" >> >> +# Set random seeds for fsx runs (0 for timestamp + pid) >> >> +seeds=(0 0 0 0) >> > >> > Hmm, is this array necessary? Why not pass 0 to -S directly in the >> > commandline? And generic/502 follows the same pattern. >> > >> >> The reason for the array is trying to follow your guidelines for running >> the test: >> - Normally test runs with random seed (for which the array is not needed) >> and every fsx process prints out the seed to the full log >> - When test detects a failure, you want to retry the test with same seed >> numbers, but those are different seed numbers for every fsx process >> - So the array is here to make setting those presets possible > > Yeah, an array makes debug easier. Then I think a comment about the > debug usage of the array would be sufficient. Thanks! > FYI, I removed the array from generic/502, because it was never useful during debugging. I guess the randomness stems mostly from timing of forking the fsx processes in that test. I reduced NUM_OPS in cloned files test to 10 in v4, because after a few holes and writes the clone is no longer a clone anyway. I already found a seed value to reproduce a failure of generic/502 on xfs failures are still random, but on spinning disk, I get them every 2 test runs. Amir.
