In case you care: GitHub confirms that this race condition can definitely happen. They don't think it's a regression -- they think that a change they made a few weeks ago simply made this behavior observable (at least in some cases).
They're talking through it internally, but apparently this is not going to be something easy to fix. They advised us to put in a delay/retry mechanism (which is what I already did). > On Mar 28, 2018, at 7:35 AM, Jeff Squyres (jsquyres) <jsquy...@cisco.com> > wrote: > > SHORT VERSION > > I have confirmed that there is a fun GitHub race condition that is sometimes > causing the commit email checker and the signed-off-by checker to sometimes > appear to hang (i.e., be stuck in the yellow dot). I have pushed up a fix; > hopefully this will address the issue. > > MORE DETAIL > > I finally caught GitHub in the act: sometimes there's a race condition. > > 1. Github issues the webhook to our aws.open-mpi.org server and sends us a > JSON blob. > 2. PHP on aws.open-mpi.org turns around and tries to download the commit URL > from the JSON blob. > 3. Sometimes GitHub issues a "Not found" error when we try to download that > commit URL. > > Meaning: GitHub issued the webhook callback to us before the commit URL (that > they sent to us in the webhook) was ready. > > Now that I know what to look for, it's easy to see when this happens. > > The solution is simple: wait a short while and then try again. Then the > commit URL will be valid. > > So I just put in a few retries when this case happens: > > https://github.com/jsquyres/github-webhooks/pull/15 > > We'll see if this fixes the problem. > > -- > Jeff Squyres > jsquy...@cisco.com > -- Jeff Squyres jsquy...@cisco.com _______________________________________________ devel mailing list devel@lists.open-mpi.org https://lists.open-mpi.org/mailman/listinfo/devel
