Merge lp:~brian-murray/daisy/remove-sandbox-dir-after-retrace into lp:daisy

[Brian Murray]

This will help reduce the amount of storage that the retracer processes require by cleaning up the sandbox dir in $release-name/cache-$tmpdir/ after every retrace. The cache folder could probably use cleaning up on a regular basis (for the development release at least) as new package versions are a common thing.

[Evan (ev)]

This was originally done for efficiency, if memory serves. In fact, leaving off the --sandbox-dir option would give you much the same effect in a cleaner way.

Could we not instead have a process watching the disk usage on the retracer? When it gets to a charm-defined percentage limit, this process could send SIGINT to the retracer process, which could be taught to finish up the current retrace and then terminate. The watchdog process would then clean up the retracer cache dirs and restart the retracer.

Alternatively, you could have the retracer do the cleanup on behalf of the watchdog, or just flat out kill the retracer process and put our ability to deal with retracers that die mid-retrace to the test. :)

It *should* throw the work back on the queue, since we didn't ack the message in that case, but I've never thoroughly tested it.

review: Needs Information

[Evan (ev)]

Alternatively, if you're keen on the more expedient solution, I'd suggest just dropping the --sandbox-dir option, knowing that we can horizontally scale out the retracers now with juju add-unit.

If you're going down that road, +1.

review: Approve

[Brian Murray (brian-murray)]

I think removing --sandbox-dir does more than we actually want. I'm suggesting that we stop using apport-retrace's "--sandbox-dir" option but continuing to use "--cache".

"--sandbox-dir" for retracer.py is a combination of both of these as setup_cache returns the directory for "--sandbox-dir" for apport-retrace and the directory for "--cache". So by not using retracer.py's --sandbox-dir we'll lose the benefit of cacheing deb files.

[Evan (ev)]

Could we not teach retracer.py to just use apport's --cache and not
--sandbox-dir? It seems like a bug that --sandbox-dir is an optional
argument in retracer.py despite it very much not being an optional
thing. :)

Either way you want to implement it, though.

On 21 March 2014 13:33, Brian Murray <email address hidden> wrote:
> I think removing --sandbox-dir does more than we actually want. I'm suggesting that we stop using apport-retrace's "--sandbox-dir" option but continuing to use "--cache".
>
> "--sandbox-dir" for retracer.py is a combination of both of these as setup_cache returns the directory for "--sandbox-dir" for apport-retrace and the directory for "--cache". So by not using retracer.py's --sandbox-dir we'll lose the benefit of cacheing deb files.
> --
> https://code.launchpad.net/~brian-murray/daisy/remove-sandbox-dir-after-retrace/+merge/212187
> You are reviewing the proposed merge of lp:~brian-murray/daisy/remove-sandbox-dir-after-retrace into lp:daisy.

[Evan (ev)]

review: Approve

[Brian Murray (brian-murray)]

A variation of this was merged:

------------------------------------------------------------
revno: 594
committer: Brian Murray <email address hidden>
branch nick: trunk
timestamp: Tue 2014-12-16 14:40:34 -0800
message:
  daisy/retracer.py: add options to have sandbox directory and ddebs cache wiped after retracing.

=== modified file 'daisy/retracer.py'
--- daisy/retracer.py	2014-03-13 22:53:52 +0000
+++ daisy/retracer.py	2014-03-21 16:10:29 +0000
@@ -563,6 +563,11 @@
             log(traceback.format_exc())
             raise
         finally:
+            if sandbox:
+                # remove the sandbox as it consumes the most space
+                # but recreate the dir as setup_cache does
+                shutil.rmtree(sandbox)
+                os.mkdir(sandbox)
             rm_eff(report_path)
 
         try: