[openstack-dev] [ceilometer]support direct alarm_evaluator db access
Nobuyoshi Nihongi
nihongi.nobuyoshi at lab.ntt.co.jp
Mon Apr 21 10:01:54 UTC 2014
Hi Liu sheng,
We're also investigating alarm_evaluator performance.
We observed that alarm_evaluator spends half of the time for calling
ceilometerclient while evaluating alarms.
Allowing alarm_evaluator directly access db will greatly improve the
performance if you have many alarms.
Best regards,
Nihongi
>From: liusheng <liusheng1175 at 126.com>
>To: openstack-dev at lists.openstack.org
>Date: Tue, 15 Apr 2014 15:26:58 +0800
>Subject: [openstack-dev] [ceilometer]support direct alarm_evaluator db access
>
>Hi there,
>Currently,alarm_evaluator invoke ceilometerclient to get all assigned
>alarms. and then, evaluate per alarm by get statistics, which will also
>call ceilometerclient. this process is:
>evaluator-->ceilometerclient-->ceilometer API-->db.
>If we use default option of evaluation_interval (60s), and if we have
>many alarms, alarm_evaluator will frequently invoke ceilometerclient and
>will produce many http requests per minute.
>That is inefficient,it affect the performance of ceilometer(data
>collection, data query, e.g.). The better way is allowing
>alarm_evaluator access db directely.
>
>Should the related codes of alarm-evaluator need a refactor?
>
>Can you provide your thoughts about this?
>
>I have registered a related blueprint:
>https://blueprints.launchpad.net/ceilometer/+spec/direct-alarm-evaluator-db-
>access
>
>Best regards
>Liu sheng
>
>
>_______________________________________________
>OpenStack-dev mailing list
>OpenStack-dev at lists.openstack.org
>http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
--
Nobuyoshi Nihongi
NTT Software Innovation Center
Phone: +81 422 59 4880
E-mail: nihongi.nobuyoshi at lab.ntt.co.jp
More information about the OpenStack-dev
mailing list