Re: Concurrency issue in TestKgioUnixConnect?

[Eric Wong <normalperson@yhbt.net>]

Hleb Valoshka <375gnu@gmail.com> wrote:
> On 9/1/13, Eric Wong <normalperson@yhbt.net> wrote:
> >> http://anonscm.debian.org/gitweb/?p=pkg-ruby-extras/ruby-kgio.git;a=blob;f=debian/patches/0002-Change-prefix-of-temporary-sockets-to-prevent-races.patch
> > Btw, do you want me to take that patch into the upstream repo?
> 
> As you wish, your tests are also affected by this issue, cause they
> run in parallel too.

I just signed-off and pushed the following out:

Hleb Valoshka (2):
      Change prefix of temporary sockets to prevent races
      Don't dump 20M in case of failure

I've never been hit by the issue even though I've always known it's a
potential issue on very busy systems.  It looks like Tempfile embeds the
PID in the temporary name, so maybe the buildds are recycling PIDs very
quickly?

I haven't looked closely at your fakeroot workaround.

> > It could be improved, too, to use a tmpdir for each test invocation.
> > That would allow multiple instances of the kgio test to run safely at
> > once (e.g. for different versions of ruby/rubinius).
> 
> Hmm, interesting idea, currently in Debian tests with different
> versions of ruby are run one-by-one, but who knows how such process
> may be implemented in other OSes.

I used to test some software in parallel under Ruby
1.8/Rubinius/1.9.1/1.9.2/1.9.3; but eventually stopped caring about
1.8, 1.9.{1,2} and Rubinius itself was taking too long to build.