On Mon, Sep 29, 2014 at 08:51:27PM +0200, Thomas Graf wrote:
> We are hitting spurious CI failures. I assume it's due to the VM
> running relatively slow and thus the current ovs-appctl time/warp not
> being sufficient. I'm not familiar with all the tests cases, can we
> just +50% on the current values to account for the slowness? Better
> fix?
We don't generally have trouble with spurious test failures due to
fixed limits, etc., because few of our tests (maybe none of them?) use
fixed time-based fudge factors. Usually, it's due to races, either
races that haven't been tracked down yet or races that are just hard
to fix.
Usually, we work around these by replacing
make check
by
make check || make check TESTSUITEFLAGS=--recheck
which simply re-runs each of the failing tests to give them another
chance. However, there's no such simple solution for "distcheck".
I'm not sure of the best thing to do here. Of course, we could track
down and fix all of the offending tests, and that is the best solution
in the long run. But new tests will continue to be added, and it is
difficult to make sophisticated tests perfectly reliable, and it is
often difficult to reproduce the failures on personal machines. We
could switch from "make distcheck" to "make check" for travis; that
might be reasonable.
_______________________________________________
dev mailing list
dev@openvswitch.org
http://openvswitch.org/mailman/listinfo/dev
