| Message ID | aa3f3c927b62d1da51166efafa856e18d01cc1ac.1692861033.git.anand.jain@oracle.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | btrfs-progs: tests: fix random mkfs.btrfs failure due to loopdev cache | expand |
On Thu, Aug 24, 2023 at 03:13:04PM +0800, Anand Jain wrote: > Sometimes, I randomly see failures like below. > > [TEST/fsck] 013-extent-tree-rebuild > failed: /Volumes/ws/btrfs-progs/mkfs.btrfs -f /Volumes/ws/btrfs-progs/tests/test.img > test failed for case 013-extent-tree-rebuild > make: *** [Makefile:484: test-fsck] Error 1 > > Looks like losetup -D failed because the device busy, however if ran > again it would successeed, possible that loop device is still writing > to the backing store. > > Using losetup directio option as below it never reproduced so far. That's interesting that it makes it work because in the command of test 013-extent-tree-rebuild the helper prepare_loopdevs() is not used at all and mkfs works on the file directly. The default TEST_DEV is a plain file and transparently mounted as loop device by run_check_mount_test_dev by -o loop, which uses defaults for losetup. And the direct io is off by default, there's no mount option for that. I'm not sure how to fix that, I want to let the test suite run without the need for a block device. A fix could be on the test level to force the TEST_DEV to be a loop device, but handling it transparently inside the mount/umount helpers may be fragile. We could add helpers for tests that need a block device, which would mean the prepare and cleanup calls. Alternatively the on-exit function that can be done via signal traps can be used but then it's for test case updated anyway.
diff --git a/tests/common b/tests/common index 602a4122f8bd..72ea8c688ec5 100644 --- a/tests/common +++ b/tests/common @@ -834,7 +834,7 @@ prepare_loopdevs() chmod a+rw "$loopdev_prefix$i" truncate -s0 "$loopdev_prefix$i" truncate -s2g "$loopdev_prefix$i" - loopdevs[$i]=`run_check_stdout $SUDO_HELPER losetup --find --show "$loopdev_prefix$i"` + loopdevs[$i]=`run_check_stdout $SUDO_HELPER losetup --direct-io=on --find --show "$loopdev_prefix$i"` done }
Sometimes, I randomly see failures like below. [TEST/fsck] 013-extent-tree-rebuild failed: /Volumes/ws/btrfs-progs/mkfs.btrfs -f /Volumes/ws/btrfs-progs/tests/test.img test failed for case 013-extent-tree-rebuild make: *** [Makefile:484: test-fsck] Error 1 Looks like losetup -D failed because the device busy, however if ran again it would successeed, possible that loop device is still writing to the backing store. Using losetup directio option as below it never reproduced so far. Signed-off-by: Anand Jain <anand.jain@oracle.com> --- tests/common | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)