[Gluster-devel] Approaching regression-test gridlock

[Jeff Darcy]

We're getting close to that point again where nothing can pass,
including patches to fix regression tests, because even those can't get
past the gauntlet of spurious failures on other tests.  Poornima
recently sent out a very helpful list of failures requiring further
investigation:

        http://www.gluster.org/pipermail/gluster-devel/2016-April/049016.html

I'll repeat some of the most pertinent info here:

        ./tests/basic/tier/tier.t 
        ./tests/bugs/quick-read/bug-846240.t 
        ./tests/basic/afr/self-heal.t 
        ./tests/basic/quota-untar.t 
        ./tests/bugs/disperse/bug-1161621.t 
        ./tests/bugs/quota/bug-1293601.t 
        ./tests/bugs/protocol/bug-808400-repl.t 
        ./tests/bugs/replicate/bug-1297695.t 
        ./tests/basic/uss.t 
        ./tests/basic/ec/ec-background-heals.t 
        ./tests/basic/afr/gfid-self-heal.t 
        ./tests/basic/nsr/nsr.t 
        ./tests/bugs/tier/bug-1279376-rename-demoted-file.t 
        ./tests/basic/quota-anon-fd-nfs.t 
        ./tests/bugs/snapshot/bug-1109889.t 
        ./tests/bugs/glusterd/bug-1209329_daemon-svcs-on-reset-volume.t 

If you are responsible for any of these tests, please take a moment to
check and see if there's either a real bug to be fixed or something that
you can do to make the test more reliable.

Also, I'd like to make a suggestion.  Could we make it so that patches
which only change a test only run that test, instead of going through a
full run of every other test that might randomly fail?  Seems like it
would save a lot of time and aggravation.