On Tue, Jul 25, 2023 at 06:56:45PM -0700, Darrick J. Wong wrote:
> From: Darrick J. Wong <djwong@xxxxxxxxxx>
>
> Create a "-smoketest" parameter to check that will run generic
> filesystem smoke testing for five minutes apiece. Since there are only
> five smoke tests, this is effectively a 16min super-quick test.
The code is setting SOAK_DURATION to 4 minutes, not 5 minutes.
However, when I ran the moral equivalent:
kvm-xfstests --soak-duration 4m --fail-loop-count 0 -c ext4/4k \
generic/475 generic/476 generic/521 generic/522 generic/642
It overall took 17 minutes to run, with just under a minute of test
infrastructure overhead (in the check script and my wrapper scripts),
with the actual test time as follows:
ext4/4k: 5 tests, 975 seconds
generic/475 Pass 242s
generic/476 Pass 244s
generic/521 Pass 241s
generic/522 Pass 241s
generic/642 Pass 7s
Totals: 5 tests, 0 skipped, 0 failures, 0 errors, 975s
The time which generic/642 ran was surprising so I took a closer look.
It does claim to be in group "soak", and it even tries to canonicalize
SOAK_DURATION (I'm not sure why, since the check script does this
already). But generic/642 doesn't seem to use SOAK_DURATION. It does
caculate a DURATION value, but it doesn't actually use SOAK_DURATION.
So that sounds like a bug in the generic/642 test?
There was also a bug xfstests's "make install" in that it doesn't
actually install src/soak_duration.awk, but I'll send that a patch
fixing that under separate cover.
Darrick -- suppose changed the SOAK_DURATION down to 2 minutes? How
much do you think that would materially affect the code coverage
metrics, and the overall effectiveness of the smoke test? If we get
generci/642 to honor SOAK_DURATION, using an overall 2 minute soak for
each test would translate to the smoke test taking about 13 minutes,
which would be great from a drive-by patch submitter perspective.
- Ted
> With gcov enabled, running these tests yields about ~75% coverage for
> iomap and ~60% for xfs; or ~50% for ext4 and ~75% for ext4; and ~45% for
> btrfs. Coverage was about ~65% for the pagecache.
