[LU-920] include xfstests into regular testing runs

• Type: Improvement
• Resolution: Unresolved
• Priority: Minor
• Labels: test_script_improvements tests
• Story Points: 3
• Assignee: Pankaj Gugale (Inactive)
• Reporter: Andreas Dilger

Description

In http://review.whamcloud.com/1708 xfstests was used to test the ldiskfs large xattr support.

There is a fair amount of useful testing that could be gotten from xfstests by running it against ldiskfs and also against Lustre. This will ensure that a number of standard Linux-specific APIs are being tested, in addition to normal and abnormal IO tests.

As a first pass, it would be possible to run xfstests as a single standalone test script with a Lustre test-framework.sh wrapper to output test results to Maloo, as well as allow Lustre formatting and remote setup.

A more advanced deployment might hook into xfstests to allow per-subtest results from xfstests to be output to Maloo instead of having a single global pass/fail result. This could possibly be done by hooking test-framework.sh directly into the xfstests harness (hopefully with very minimal code changes to xfstests), or alternately if that is too complex as a post-processing step to convert the xfstests output into a format suitable for Maloo.

We don't want to modify xfstests in any significant way, so that it is possible to update it regularly from the upstream version. It would also be beneficial to submit upstream patches to indicate which tests can run on a mounted Lustre filesystem so that we don't have to maintain this continually.

Issue Links

• is related to: LU-8865 integrate fs_test test into lustre/tests
• is related to: LU-6007 FIEMAP fails xfstests's fiemap-tester

Comments

[Andreas Dilger]

Hi Pankaj,
I'm wondering if you have a resource that could look into adding the xfstests (now named "fstests") into the Lustre Autotest framework, so that these tests are run regularly as part of the lustre-reviews testing? They would improve test coverage for Lustre in new ways beyond our current regression test cases (though there would definitely be some overlap as well).

This would require some bash scripting skills so that xfstests can reformat the Lustre filesystem using the test-framework.sh functionality, and the test PASS/FAIL results from the individual xfstests are reported to Maloo properly.

Then, running these tests a number of times to determine which of them are failing on Lustre, and some analysis to determine if they are irrelevant due to missing features and should be skipped, or if the failure is exposing a legitimate bug in the Lustre code. In case of a defect, or if uncertain, a Jira ticket can be filed to track the issue.

This is not a critical issue, but now that our QE team has grown significantly, it would be great if we can make progress on the testing backlog. Feel free to reassign this ticket to a holding alias if there is one for QE issues.