missing patches

[Brian Norris]

Hi Damien,

On Fri, Oct 16, 2015 at 01:43:04PM +0100, Damien Lespiau wrote:
> On Mon, Oct 12, 2015 at 01:54:24PM -0700, Brian Norris wrote:
> > Occasionally, I'll run across patches from the linux-mtd mailing list
> > that never show up on the ozlabs.org Patchwork instance. I don't
> > really have much visibility into this problem, so I wanted to reach
> > out here.
> > 
> > For one example: the patch shows up at the mailing list [1], but not
> > here on Patchwork [2]. The patchwork parser helper [3] seems to parse
> > the email OK locally, and it recognizes it as a patch.
> > 
> > The problem is actually the same for the entire series of which [1] is
> > a part. I think I've ran across a few others of these over time, but I
> > usually just ignore it and continue on. After all, patchwork only
> > augments my workflow -- it's not central.
> > 
> > Any tips?
> 
> I do also see this happening on fdo.org. Two theories right now:
> 
>   - mailman fails us. I have seen occurences where the patch shows up in
>     the mailman archive but isn't received by some people (some of them
>     can be patchwork). This would need to be debugged more seriously
>     looking at what mailman does.

Hmm, that's an interesting theory. I'd hope that's not the problem, as
that affects much more than just patchwork :(

Personally, my experience with dropped mails is due to spam filters, so
nobody gets delivery from mailman. I've never *noticed* mail being
delivered only to some people. But then, that may just be because I
haven't observed it, not because it didn't happen.

>   - mails are received by postfix (or sendmail, ...) and piped over to
>     patchwork's parsemail.sh through /etc/aliases. It's not entirely
>     clear reading the documentation and I haven't gone through postfix
>     source code, but it's apparent that there's no serialization: there
>     can be several mail parsing processes running at the same, racing
>     for db accesses. It may or may not a problem with the patches only
>     patchwork, but my series work does make that issue obvious. I have a
>     patch in testing that takes a file lock to ensure the mail insertion
>     is atomic, it may well solve other issues (hand waves, I haven't
>     really authored parsemail.py looking for such races).

Db races sounds interesting too.

For either explanation, it's odd that an entire 11-patch series got
dropped. The latter explanation really sounds like it would likely
result in only some of the patches getting dropped, not all of them. I
guess the former is unknown, since we don't really know what the root
cause of such dropping could be.

> HTH,

Helps a bit, though I'm not about to try to debug mailman. Thanks for
your thoughts!

Regards,
Brian