[OpenStack-Infra] Debugging nodepool over-allocation with history tracking (109132)

Jeremy Stanley fungi at yuggoth.org
Tue Aug 26 13:43:13 UTC 2014


On 2014-08-26 16:04:39 +1000 (+1000), Ian Wienand wrote:
[...]
> So really it didn't launch these nodes ("No current image").  Thus on
> the next loop
[...]
> I see it make the grants again (doing what it is supposed to do,
> notice it allocates some fedora nodes there).
> 
> Is it possible this is what was ascribed to the double-quota problem?

I don't think so, no. In this case we restarted with a full
compliment of images already available. Also, graphite stats showed
us spiking a building count multiples of our aggregate quota and I
seem to remember also seeing many of them go into error at the
providers (it appeared that quota enforcement lagged in nova so we
were initially allowed to request many more instances than we
actually had quota to support).

> It would be great to get the logs to see what happened after the
> launch calculation to see if there is some way we can double-count.
[...]

July 23 nodepool logs have already been rotated out, but perhaps
James saved a copy in a more durable location.
-- 
Jeremy Stanley



More information about the OpenStack-Infra mailing list