On 14/06/16 18:00, "Matt Riedemann" <mrie...@linux.vnet.ibm.com> wrote:
>On 6/14/2016 10:14 AM, Kris G. Lindgren wrote: >> Cern is running ceilometer at scale with many thousands of compute >> nodes. I think their blog goes into some detail about it [1], but I >> don’t have a direct link to it. >> >> >> [1] - http://openstack-in-production.blogspot.com/ >> ___________________________________________________________________ >> Kris Lindgren >> Senior Linux Systems Engineer >> GoDaddy >> >> From: Bill Jones <bill.jo...@sungardas.com >> <mailto:bill.jo...@sungardas.com>> >> Date: Tuesday, June 14, 2016 at 9:03 AM >> To: "openstack-oper." <openstack-operators@lists.openstack.org >> <mailto:openstack-operators@lists.openstack.org>> >> Subject: [Openstack-operators] Scaling Ceilometer compute agent? >> >> Has anyone had any experience with scaling ceilometer compute agents? >> >> We're starting to see messages like this in logs for some of our compute >> agents: >> >> WARNING ceilometer.openstack.common.loopingcall [-] task <function >> interval_task at 0x2092cf8> run outlasted interval by 293.25 sec >> >> This is an indication that the compute agent failed to execute its >> pipeline processing within the allotted interval (in our case 10 min). >> The result of this is that less instance samples are generated per hour >> than expected, and this causes billing issues for us due to the way we >> calculate usage. >> >> It looks like we have three options for addressing this: make the >> pipeline run faster, increase the interval time, or scale the compute >> agents. I'm investigating the latter. >> >> I think I read in the ceilometer architecture docs that the agents are >> designed to scale, but I don't see anything in the docs on how to >> facilitate that. Any pointers would be appreciated. >> >> Thanks, >> Bill >> >> >> _______________________________________________ >> OpenStack-operators mailing list >> OpenStack-operators@lists.openstack.org >> http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-operators >> > >This is the specific blog post I think: > >http://openstack-in-production.blogspot.com/2014/03/cern-cloud-architecture-update-for.html > To be complete, we are running ceilometer but do struggle, both with data volumes being recorded and with the time to extract reports. Using a dedicate keystone helps avoid impact on other cloud activities, but I don’t think this is a good architecture to replicate. With the rework on the metering architecture, there are developments around projects such as gnocchi and aodh but we have not yet got to a production state (RPM packaging, Puppet). We’ll look again with Mitaka once the upgrade is done for the CERN cloud and will produce a blog once we’ve got there. Tim >-- > >Thanks, > >Matt Riedemann > > >_______________________________________________ >OpenStack-operators mailing list >OpenStack-operators@lists.openstack.org >http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-operators _______________________________________________ OpenStack-operators mailing list OpenStack-operators@lists.openstack.org http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-operators
