[LU-7346] Reintroduce tests from SLOW list back to review process

• Status: Resolved
• Project: Lustre
• Fix Version/s: Lustre 2.9.0
• Type: Improvement
• Priority: Minor
• Reporter: Saurabh Tandan (Inactive)
• Assignee: Saurabh Tandan (Inactive)
• Resolution: Fixed
• Severity: 3

Description

There is no real definition of what constitutes a SLOW test, the general rule of thumb is, if a test runs for five minutes or more, it is eligible to be marked as SLOW. Attached is the list of tests which can be reintroduced to the normal review process as they are not running for five minutes or more. The results for these tests can also be found at https://wiki.hpdd.intel.com/display/QE/Lustre+Test+Suites+-+SLOW+Test+List . The tests were run multiple times to have more accurate time values.

Issue Links

• is related to: LU-4537 Track slow tests in review-dne

Comments

[Andreas Dilger]

Saurabh, thanks for running these tests.

The next step is for you to make a patch to remove the recommended tests from the EXCEPT_SLOW= list. It would also be useful to annotate the remaining EXCEPT_SLOW tests with their test times, as is done in recovery-small, so that the information you captured here is available in the future.

In terms of tests to enable, I'm generally in favour of your recommendations. For recovery-small test_51 it makes sense to enable this for ZFS as well, since it isn't much over 5 minutes, and it keeps the testing coverage consistent. It doesn't make sense to enable sanity test_51b without enabling test_51ba, since that would leave 70000 files in the filesystem and may cause later tests to run out of inodes.

In addition to figuring out which tests take a long time, it might be a worthwhile exercise to understand why the tests are taking so long (if that is clear from reading the test script, don't spend hours on it), and seeing if there is some way to actually reduce the test time rather than just skipping it.

For sanity 24D it might make sense to make that test run as part of test_51b (or immediately thereafter, e.g. test_51b0 similar) since test_51b has already created the required number of files and it would reduce runtime when SLOW=yes is selected. I agree test_24D takes too long to run for every test, so it shouldn't be enabled by default.

Also, have you looked at currently-running tests to see if any of them are taking too long and should be skipped (like LU-4537)?

[Saurabh Tandan (Inactive)]

Andreas, Thanks a lot for your valuable input and directing my attention to currently running tests.

I will go ahead and start working on the next step to enable them on the EXCEPT_SLOW list.

I totally agree to your suggestion of enabling recovery_small test_51 as it's just little over 5 minutes. But regarding sanity test_51b it makes sense to me now, why was it disabled previously. So this is a point to be looked upon. I will have more detailed look into sanity test_51b, test_51ba and also test_24D if that needs to be included in the enabling list. If sanity test_24D is combined with test_51b or run just immediately after test_51b, I think we are good to do that. But I would like to spend more time looking upon if it makes sense to enable sanity test_51b and test_51ba because the timing for test_51ba alone is around 13 minutes or so.

[Gerrit Updater]

Saurabh Tandan ([e-mail]) uploaded a new patch: http://review.whamcloud.com/17143
Subject: LU-7346 tests: Reintroduce SLOW tests to review process
Project: fs/lustre-release
Branch: master
Current Patch Set: 1
Commit: 35fc6ceef4c29e5da456e8674a99306f28fd0254

[Gerrit Updater]

Oleg Drokin ([e-mail]) merged in patch http://review.whamcloud.com/17143/
Subject: LU-7346 tests: Reintroduce SLOW tests to review process
Project: fs/lustre-release
Branch: master
Current Patch Set:
Commit: 28497055cdac33293122b73e62c00555a32a5f88

[Joseph Gmitter (Inactive)]

reopening to correct fields to show up in 2.9 changelog