Re: Fedora rawhide compose report: changes
On Wed, 2018-01-31 at 10:06 -0800, Kevin Fenzi wrote:
On 01/28/2018 05:31 AM, Adam Williamson wrote:
> On Wed, 2018-01-24 at 20:22 +0000, Fedora Rawhide Report wrote:
> > OLD: Fedora-Rawhide-20180117.n.1
> > NEW: Fedora-Rawhide-20180123.n.1
>
> We've had a couple of these odd duplicate mails with the compose ID
> missing from the subject lately. I have to confess I do not know what's
> causing them to happen at all; I just can't see how it could be
> happening, or find out what compose script exactly is doing it. Going
> by datagrepper logs, these mails just don't seem to link up
> with...anything. It's kinda very strange.
I think these might be happening when we have a rawhide compose running,
another one starts then some bits hit OOM and are killed. One of the
composes finishes mostly normally, but the other one finishes... weirdly.
Ah. Interesting. So the block in nightly.sh that tries to bail if
pungi-koji doesn't exit 0 somehow gets skipped or is satisfied in this
case? It's odd that all the other stuff like sending out fedmsgs
doesn't seem to happen in this case, though...
We perhaps should look at wrapping this so we only run one compose at
a
time?
Seems viable, yeah, or try to find some stronger way to make sure the
script bails unless pungi actually works?
--
Adam Williamson
Fedora QA Community Monkey
IRC: adamw | Twitter: AdamW_Fedora | identi.ca: adamwfedora
http://www.happyassassin.net