<div dir="ltr"><div><div><div><div><div><div><div><div>Hi team!<br><br></div></div>Last week we were working on notification problem in ceilometer during tempest tests creation. Tests for notification passed successfully on Postgres but failed on MySQL. This made us start investigations and this email contains some results. <br>
<br></div>As it turned out, tempest as it is is something like performance-testing for Ceilometer. It contains 2057 tests. Almost in all test OpenStack resources are being created and deleted: images, instances, volumes. E.g. during instance creation nova sends 9 notifications. And all the tests are running in parallel for about 40 minutes. <br>
</div>From ceilometer-collector logs we may found very useful message: <br><pre><font size="4"><span><a name="144b1ef6f6d7eb33_144b1c6a5bf5963f__2014-03-10_09_42_41_356" href="http://logs.openstack.org/36/64136/20/check/check-tempest-dsvm-full/e361520/logs/screen-ceilometer-collector.txt.gz#_2014-03-10_09_42_41_356" target="_blank">2014-03-10 09:42:41.356</a> 22845 DEBUG ceilometer.dispatcher.database [req-16ea95c5-6454-407a-9c64-94d5ef900c9e - - - - -] metering data storage.objects.outgoing.bytes for b7a490322e65422cb1129b13b49020e6 @ 2014-03-10T09:34:31.090107: </span></font></pre>
So collector starts to process_metering_data in dispatcher only in 9:42 but nova sent it in 9:34. To look at whole picture please take look at picture [1]. It illustrates time difference based on this message in logs. <br>
</div>
Besides, I decided to take a look on difference between the RPC-publisher sends the message and the collector receives the message. To create this plot I've parsed the lines like below from anotifications log:<br></div>
<br><pre><font size="4"><span><a name="144b1ef6f6d7eb33_144b1c6a5bf5963f__2014-03-10_09_25_49_333" href="http://logs.openstack.org/36/64136/20/check/check-tempest-dsvm-full/e361520/logs/screen-ceilometer-anotification.txt.gz#_2014-03-10_09_25_49_333" target="_blank">2014-03-10 09:25:49.333</a> 22833 DEBUG ceilometer.openstack.common.rpc.amqp [-] UNIQUE_ID is <span style="background-color:rgb(182,215,168)">683dd3f130534b9fbb5606aef862b8</span><span style="background-color:rgb(182,215,168)">3d</span>.<br>
</span></font></pre><pre><font size="4"><span><br></span></font></pre></div>After that I found the corresponding id in collector log:<br><pre><font size="4"><span></span><span><a name="144b1ef6f6d7eb33__2014-03-10_09_25_49_352" href="http://logs.openstack.org/36/64136/20/check/check-tempest-dsvm-full/e361520/logs/screen-ceilometer-collector.txt.gz#_2014-03-10_09_25_49_352" target="_blank">2014-03-10 09:25:49.352</a> 22845 DEBUG ceilometer.openstack.common.rpc.amqp [-] received {u'_context_domain': None, u'_context_request_id': u'req-0a5fafe6-e097-4f90-a68a-a91da1cff22c', <br>
u'args': {u'data': [...,<br> u'message_id': u'f7ad63fc-a835-11e3-8223-bc764e205385', u'counter_type': u'gauge'}]}, u'_context_read_only': False, u'_unique_id': u'<span style="background-color:rgb(182,215,168)">683dd3f130534b9fbb5606aef862b83d</span>', <br>
u'_context_user_identity': u'- - - - -', u'_context_instance_uuid': None, u'_context_show_deleted': False, u'_context_tenant': None, u'_context_auth_token': '<SANITIZED>', <br>
....} _safe_log /opt/stack/new/ceilometer/ceilometer/openstack/common/rpc/common.py:280<br><br></span></font></pre>So in the example above we see time-difference only in 20 milliseconds. But it grows very quickly :( To see it please take a look on picture [2]. <br>
<br></div><div>To summarize pictures:<br></div><div>1. Picture 1: Axis Y: amount of seconds between nova creates notification and the collector retrieves the message. Axis X: timestamp <br></div><div>2. Picture 2: Axis Y: amount of seconds between the publisher publishes the message and the collector retrieves the message. Axis X: timestamp<br>
<br></div><div>These pictures are almost the same and it makes me think that collector cannot manage with big amount of messages. What do you think about it? Do you agree or you need more evidences, e.g. amount of messages in rabbit or amth else?</div>
<div style>Let's discuss that in [Ceilometer] topic first, I will create a new thread about testing strategy in tempest later. Because in this circumstances we forced to refuse from created notification tests and cannot reduce time for polling because it will make everything even worst.</div>
<div style><br></div><div>[1]: <a href="http://postimg.org/image/r4501bdyb/" target="_blank">http://postimg.org/image/r4501bdyb/</a><br>
[2]: <a href="http://postimg.org/image/yy5a1ste1/" target="_blank">http://postimg.org/image/yy5a1ste1/</a><br></div><div><br></div><div style>Thanks for your attention,</div><div style>Nadya</div></div>