https://wiki.openstack.org/w/index.php?action=history&feed=atom&title=User%3AGibi%2FHorizonInstanceStatistics 2026-07-10T07:17:20Z Revision history for this page on the wiki MediaWiki 1.28.2 https://wiki.openstack.org/w/index.php?title=User:Gibi/HorizonInstanceStatistics&diff=48940&oldid=prev 2014-04-16T16:03:46Z

<p>Created page with &quot; = Horizon instance statistic query = [http://paste.openstack.org/show/54676/ pastebin] [http://people.redhat.com/~lsurette/OpenStack/Horizon%20Overview%20Pages_v1.0.pdf horiz...&quot;</p> <p><b>New page</b></p><div><br /> = Horizon instance statistic query =<br /> [http://paste.openstack.org/show/54676/ pastebin]<br /> [http://people.redhat.com/~lsurette/OpenStack/Horizon%20Overview%20Pages_v1.0.pdf horizon pdf]<br /> <br /> <br /> * fact1: &quot;instance&quot; meter reports 1 if the instance is _allocated_ regardless of the state of the instance. Except if the VM is in error state, in that case &quot;instance&quot; meter does not report sample.<br /> * fact2: ceilometer does not have samples from the &quot;instance&quot; meter with counter_volume 0 for deleted instances obviously<br /> * fact3: the state of the instance is not part of the metadata of the &quot;instance&quot; sample<br /> <br /> == Problem 1 ==<br /> So first you want to calculate avg(counter_volume) grouped by resource_id then make a sum(counter_volume) grouped by project_id. <br /> The avg(counter_volume) will produce invalid result for example if we have 6 measurement in an hour but the instance was running only half of the time, as we will have only 3 samples instead of 6 so the avg will be 1 instead of 1/2 due to fact2. If it is not a problem to get 1 instead of 1/2 for this case then you can use distinct instead of avg. The other possibility is to make the measurement period available on the API and just get sum(counter_volume) from the statistics API and then on the client side calculate the avg for the known period.<br /> <br /> == Problem 2 ==<br /> Your query with two group bys cannot be expressed in one SELECT statement. We have to translate the first avg (or distinct) step to a subquery of the second step. Our original idea for complex query statistic will allow multiple group bys but we haven't planned to allow such group bys that needs subqueries.<br /> <br /> == Problem with DISTINCT ==<br /> SELECT count(*), project_id FROM ( SELECT DISTINCT resource_id, project_id FROM Meter) GROUP BY project_id<br /> <br /> SELECT count(*), project_id FROM (SELECT avg(counter_volume), resource_id, project_id FROM Meter GROUP BY resource_id, project_id) GROUP BY project_id<br /> <br /> ==&gt;&gt;<br /> <br /> SELECT count(DISTINCT(resource_id, project_id)) FROM Meter GROUP BY project_id<br /> * Only the count function is meaningful<br /> * Only the instance meter can be aggregated this way<br /> * Only the resource_id, project_id seems a valid DISTINCT parameter ???<br /> <br /> = TODO = <br /> Gauge VM state counter https://blueprints.launchpad.net/ceilometer/+spec/state-meter<br /> Propose separate sub meter like instance:active, instance:stopped for clarity</div>

Gibi