On May 27, 2024 / 07:49, Shinichiro Kawasaki wrote: > On May 22, 2024 / 07:02, Shinichiro Kawasaki wrote: > > On May 21, 2024 / 13:24, Josef Bacik wrote: > > > Because NBD has the old style configure the device directly config we > > > sometimes have spurious failures where the device wasn't quite ready > > > before the rest of the test continued. > > > > > > nbd/002 had been using _wait_for_nbd_connect, however this helper waits > > > for the recv task to show up, which actually happens slightly before the > > > size is set and we're actually ready to be read from. This means we > > > would sometimes fail nbd/002 because the device wasn't quite right. > > > > > > Additionally nbd/001 has a similar issue where we weren't waiting for > > > the device to be ready before going ahead with the test, which would > > > cause spurious failures. > > > > > > Fix this by adjusting _wait_for_nbd_connect to only exit once the size > > > for the device is being reported properly, meaning that it's ready to be > > > read from. > > > > > > Then add this call to nbd/001 to eliminate the random failures we would > > > see with this test. > > > > > > Signed-off-by: Josef Bacik <josef@xxxxxxxxxxxxxx> > > > > Josef, thank you very much. I've been seeing nbd/001 on my test system and this > > patch avoids it :) The change looks good to me. I will leave it for several > > days before applying it, in case anyone has some more comments. > > I've applied it. Thanks! To: Yi Zhang Yi, I guess this fix may avoid the nbd/002 failure that CKI reports recently [*]. To apply the fix, could you rebase the blktests for CKI runs? [*] https://datawarehouse.cki-project.org/kcidb/tests/12631448
