[Buildbot-devel] git step fails to remove old build directory, causing source to be copied to wrong directory

[Dustin J. Mitchell]

That's certainly weird behavior!

Note that you're using the master-side Git step.  That runs the
'rmdir' command on the slave, which invokes 'rm -rf' as a shell
command (slave/buildslave/commands/fs.py).

So the question is, is `rmdir` incorrectly returning success (a bug in
slave/buildslave/commands/fs.py) or is the git step incorrectly
handling a failure from the `rmdir` command (a bug in
master/buildbot/steps/source/git.py).

Adding some more tests for fs.py might be a good place to start -- its
test suite is pretty slim right now.

Dustin

On Fri, Sep 19, 2014 at 5:26 PM, Dan Kegel <dank at kegel.com> wrote:
> I'm using buildbot-slave-0.8.7 (out of inertia), with the git step like:
>
>         factory.addStep(Git(repourl=repourl, mode='full',
> submodules=True, method='copy', branch=sbranch,
> getDescription={'tags':True}))
>
> Sometimes a previous build leaves some whacky process running,
> so when the next git checkout tries to remove the old build directory,
> it fails with
>
> rm: cannot remove
> '/home/buildbot/slave-state/cygwin/g-speak-yobuild10-win7/mzreach-webrtc-win7-msvc2013-master/build/webrtc/trunk/chromium/_gclient_src_4cokd8/.git/objects/pack/tmp_pack_Khspxc':
> Device or resource busy
>
> Buildbot then does a chmod (in hopes of fixing the failure), but of
> course the 2nd rm still fails, since it wasn't a permission problem.
>
> That second failure should trigger a build failure, but doesn't.  As a result,
> the following cp creates a source subdirectory in the old build directory (!)
> which is rather confusing.
>
> It kind of looks like the source tries to abort if the 2nd rm fails;
> not sure why
> that doesn't happen.
>
> I should probably have the bot reboot after each build to avoid the problem.
>
> But on the buildbot side, assuming this hasn't already been fixed:
> I guess the question is, how does one add a unit test for this?
> I imagine add one to buildslave/test/unit/test_commands_git.py
> that uses a mock /bin/rm that always fails?
> - Dan
>
> ------------------------------------------------------------------------------
> Slashdot TV.  Video for Nerds.  Stuff that Matters.
> http://pubads.g.doubleclick.net/gampad/clk?id=160591471&iu=/4140/ostg.clktrk
> _______________________________________________
> Buildbot-devel mailing list
> Buildbot-devel at lists.sourceforge.net
> https://lists.sourceforge.net/lists/listinfo/buildbot-devel