On 2018-03-06 14:01, Alberto Garcia wrote: > This patch tweaks TestParallelOps in iotest 030 so it allocates data > in smaller regions (256KB/512KB instead of 512KB/1MB) and the > block-stream job in test_stream_commit() only needs to copy data that > is at the very end of the image. > > This way when the block-stream job is awakened it will finish right > away without any chance of being stopped by block_job_sleep_ns(). This > triggers the bug that was fixed by 3d5d319e1221082974711af1d09d82f and > 1a63a907507fbbcfaee3f622907ec24 and is therefore a more useful test > case for parallel block jobs. > > After this patch the aforementiond bug can also be reproduced with the > test_stream_parallel() test case. > > Since with this change the stream job in test_stream_commit() finishes > early, this patch introduces a similar test case where both jobs are > slowed down so they can actually run in parallel. > > Signed-off-by: Alberto Garcia <be...@igalia.com> > Cc: John Snow <js...@redhat.com> > --- > > v4: Mention that commit 1a63a907507fbbcfaee3f622907ec24 also > contributes to solve the original bug (both commits need to > reverted in order to reproduce this bug reliably). > > Rewrite the loop that writes data into the images to make it more > readable.
Thanks! Applied to my block tree: https://github.com/XanClic/qemu/commits/block (Still took me a couple of attempts to get it to fail both commits reverted, though...) Max
Description: OpenPGP digital signature