[osprofiler][rally][openstack] Using osprofiler/rally directly on production Openstack system
Is there any concern with using osprofiler/rally directly on the production Openstack system?
Hi!
OSProfiler enables tracing only requests with a special header. The header is not embedded in each request (even if you configure osprofiler in your system), you need to use a special CLI argument to set it. So even if the tracing of one particular request slows done the flow (which should not happen), it should give zero impact on the performance of the whole system.
As for Rally (in particular, the task component), it creates resources with a special naming format that allows to filter out only these resources during the cleanup process. We are using Rally in production as a part of the monitoring for ~5 years or so.
пн, 6 мар. 2023 г. в 23:34, hai wu haiwu.us@gmail.com:
Is there any concern with using osprofiler/rally directly on the production Openstack system?
Thanks Andriy! That sounds very promising. I just tried to install rally, and hit one known bug: https://bugs.launchpad.net/rally/+bug/2004022. I downgraded SQLAlchemy and was able to get rally installed. But I could not find any sample task to run per this document url: https://rally.readthedocs.io/en/latest/quick_start/tutorial/step_1_setting_u.... It seems most of the sample tasks have been deleted already per its github history.
Which sample task I could use to try this out (for example, list all openstack instances..)? It seems its documentation is out of date..
Thanks, Hai
On Mon, Mar 6, 2023 at 5:10 PM Andriy Kurilin andr.kurilin@gmail.com wrote:
Hi!
OSProfiler enables tracing only requests with a special header. The header is not embedded in each request (even if you configure osprofiler in your system), you need to use a special CLI argument to set it. So even if the tracing of one particular request slows done the flow (which should not happen), it should give zero impact on the performance of the whole system.
As for Rally (in particular, the task component), it creates resources with a special naming format that allows to filter out only these resources during the cleanup process. We are using Rally in production as a part of the monitoring for ~5 years or so.
пн, 6 мар. 2023 г. в 23:34, hai wu haiwu.us@gmail.com:
Is there any concern with using osprofiler/rally directly on the production Openstack system?
-- Best regards, Andrey Kurilin.
It seems they might have been moved here? https://github.com/openstack/rally-openstack/tree/master/samples/tasks/scena....
If so, the rally documentation needs to be updated ..
On Mon, Mar 6, 2023 at 7:36 PM hai wu haiwu.us@gmail.com wrote:
Thanks Andriy! That sounds very promising. I just tried to install rally, and hit one known bug: https://bugs.launchpad.net/rally/+bug/2004022. I downgraded SQLAlchemy and was able to get rally installed. But I could not find any sample task to run per this document url: https://rally.readthedocs.io/en/latest/quick_start/tutorial/step_1_setting_u.... It seems most of the sample tasks have been deleted already per its github history.
Which sample task I could use to try this out (for example, list all openstack instances..)? It seems its documentation is out of date..
Thanks, Hai
On Mon, Mar 6, 2023 at 5:10 PM Andriy Kurilin andr.kurilin@gmail.com wrote:
Hi!
OSProfiler enables tracing only requests with a special header. The header is not embedded in each request (even if you configure osprofiler in your system), you need to use a special CLI argument to set it. So even if the tracing of one particular request slows done the flow (which should not happen), it should give zero impact on the performance of the whole system.
As for Rally (in particular, the task component), it creates resources with a special naming format that allows to filter out only these resources during the cleanup process. We are using Rally in production as a part of the monitoring for ~5 years or so.
пн, 6 мар. 2023 г. в 23:34, hai wu haiwu.us@gmail.com:
Is there any concern with using osprofiler/rally directly on the production Openstack system?
-- Best regards, Andrey Kurilin.
participants (2)
-
Andriy Kurilin
-
hai wu