potential memory leak in compute agent

[gordon chung]

when setting pipeline.yaml interval to high frequency, it appears that the compute agent is not properly freeing up memory.

in my environment (devstack with neutron, a single server instance started, and 5 sec pollster interval), in the span of 3hrs, the resident memory consumed by compute agent went from ~52mb to ~65mb... a roughly 20% jump in memory.

[gordon chung (chungg)]

i can't recreate this against master currently.

[Fabrizio Soppelsa (fsoppelsa)]

Observed also in IceHouse 2014.1.3.

The value of RSS have increased by nearly 500% from 51996 at start of the test to 303788 when the test was ended for the ceilometer-agent-compute process.

This arose after some Rally benchmarking scenarios (Nova and Cinder - create/destroy instances/storages) ran for several days.

[ZhiQiang Fan (aji-zqfan)]

please look at this bug in novaclient: https://bugs.launchpad.net/python-novaclient/+bug/1433491
which will definitely cause memory leak

https://github.com/openstack/ceilometer/blob/stable/icehouse/requirements.txt#L26
https://github.com/openstack/python-novaclient/blob/2.20.0/novaclient/client.py#L440

[Luo Gangyi (luogangyi)]

Found this problem too, and thansk to ZhiQiang gave the hint.

After I upgrade the python-novaclient, It seems OK now.