Hi all, We've still had lingering false positive failure reports on some of these generic dmlogwrites tests on XFS due to metadata ordering issues. The logwrites mechanism relies on discard to provide zeroing behavior to avoid this, but if that is not available, this can result in subtle failures that take time to diagnose. This series updates the remaining generic dmlogwrites tests to use the same scheme we used in generic/482 to address this problem, which is to explicitly use a thin volume for predictable discard support. It also adds a discard zeroing behavior check as a backstop against future tests. The thought crossed my mind of pushing much of this code down into the common dmlogwrites code to reduce duplication, but I didn't want to get too deep into the weeds of reworking the common code to address this problem in a handful of tests. Thoughts, reviews, flames appreciated. Brian Brian Foster (4): generic: require discard zero behavior for dmlogwrites on XFS generic/455: use thin volume for dmlogwrites target device generic/457: use thin volume for dmlogwrites target device generic/470: use thin volume for dmlogwrites target device common/dmlogwrites | 10 ++++++++-- common/rc | 14 ++++++++++++++ tests/generic/455 | 36 ++++++++++++++++++++++-------------- tests/generic/457 | 33 +++++++++++++++++++++------------ tests/generic/470 | 24 ++++++++++++++++++------ 5 files changed, 83 insertions(+), 34 deletions(-) -- 2.25.4
