[Lustre-devel] [Bug 11554] sanity.sh dependencies missing from release tarball
scj...@clusterfs.com
https://bugzilla.lustre.org/show_bug.cgi?id=11554
What |Removed |Added
----------------------------------------------------------------------------
Summary|sanity.sh dependencies |sanity.sh dependencies
|missing from release |missing from release tarball
I think Jim is really referring to what we ship in the tarball. The 1.4.8
tarball includes the sanity.sh script but not any of the other scripts Jim
refers to. So aside from the idea of creating a lustre-tests RPM, we need to
fix the tarball. I have updated the summary accordingly.
_______________________________________________
Lustre-devel mailing list
Lustre...@clusterfs.com
https://mail.clusterfs.com/mailman/listinfo/lustre-devel
nat...@clusterfs.com
https://bugzilla.lustre.org/show_bug.cgi?id=11554
(In reply to comment #3)
> I think Jim is really referring to what we ship in the tarball. The 1.4.8
> tarball includes the sanity.sh script but not any of the other scripts Jim
> refers to. So aside from the idea of creating a lustre-tests RPM, we need to
> fix the tarball. I have updated the summary accordingly.
I just noticed myself that b1_5 has the same issue. sanity.sh is included, but
e.g. oos.sh which it calls is not.
I think probably none of the test dir should be included. We could include a
simple hardcoded script that starts a local loopback mdt, ost, and client in the
utils directory maybe for a simple "does lustre work" check.
gar...@llnl.gov
https://bugzilla.lustre.org/show_bug.cgi?id=11554
(In reply to comment #4)
Please don't remove the tests - we are putting them to good use.
I had hoped that CFS would consider implementing a 'make check' target in the
top level Makefile as we have done. Its good practice and generally easy to run
sanity (and likely other single node tests that we haven't tried yet) before
committing changes.
If its not useful to CFS, its at least useful to us. We generally carry a very
large patch stack, and although we're pretty good at running IOR and such on
clusters against candidate release, we've not been good about basic sanity
testing. Having the ability to detect some regressions with minimal effort helps.
scj...@clusterfs.com
https://bugzilla.lustre.org/show_bug.cgi?id=11554
Hi Jim,
(In reply to comment #5)
> I had hoped that CFS would consider implementing a 'make check' target in the
> top level Makefile as we have done. Its good practice and generally easy to run
> sanity (and likely other single node tests that we haven't tried yet) before
> committing changes.
Part of our development process (for any patch) involves running
acceptance-small.sh, which includes sanity. In my case, I build on a different
machine than I test on (my desktop vs. a UML instance) so 'make check' would not
be useful. But I undestand that it would be useful to others, and wouldn't
object to seeing one appear at some point :)