[LU-4556] speed up sanity-lfsck and sanity-scrub tests

• Type: Technical task
• Resolution: Fixed
• Priority: Major
• Fix Version/s: Lustre 2.6.0
• Affects Version/s: Lustre 2.6.0
• Labels: improvement
• Assignee: nasf (Inactive)
• Reporter: Andreas Dilger

Description

The test time of sanity-scrub and sanity-lfsck is much longer than it needs to be, especially in DNE mode, because it is reformatting all of the OSTs and MDTs for every test, and then repopulating the filesystem with test files.

It would be better to avoid reformatting the filesystems at all, and instead just compare the "new" results with the "old" results.

There may be other potential areas of improvement also.

Issue Links

• duplicates: LU-4686 Try to NOT reformat the device in LFSCK tests
• is related to: LU-4537 Track slow tests in review-dne
• is related to: LU-5112 Interop 2.5 server with higher version client failure on test suite sanity-lfsck test_5: Fail to cleanup the env
• is related to: LU-4763 sanity-scrub test 11: FAIL: (8) Expect 0 objects skipped on mds1, but got 14
• is related to: LU-4202 Test failure sanity-lfsck test_8: Expect 'scanning-phase2', but got 'scanning-phase1'

Comments

[nasf (Inactive)]

The patch has been landed to master.

[Andreas Dilger]

While this is useful to fix, it would not block the release of 2.6.0 so reducing the priority. I'm happy that some good improvements have been made with the current patch, and would prefer that effort be focussed on fixing defects for 2.6.

[nasf (Inactive)]

It is very important to reduce the Maloo tests run time.

[nasf (Inactive)]

Sometimes, create on Maloo is very slow, less than 50 objects/sec, no logs, very strange.
sanity-scrub in my local environment (with 2 MDT and 2 OST) takes about 770 second.

[nasf (Inactive)]

Here is the patch:

http://review.whamcloud.com/9704

[nasf (Inactive)]

New version sanity-lfsck takes about 750 seconds under DNE mode in my local VM environment. New version sanity-scrub is still in testing.