[Openstack-operators] scaling gnocchi metricd

gordon chung gord at live.ca
Wed Mar 29 15:46:26 UTC 2017



On 29/03/17 02:10 AM, Ionut Biru - Fleio wrote:
> I'm not using influxdb, just basic configuration generated by openstack
> ansible, which enables file storage by default.
>
>
> The reason for bumping those values was to process a lot of measures and
> 400 seems a high number at that time.
>
>
> I did use the below values without any impact
>
> metric_processing_delay = 0

as Alex mentioned, you actually don't want to have these values set too 
low. specifically, setting metric_processing_delay very low, may cause 
already schedule metrics to be schedule again. you'll need to play 
around with this value. one tip to setting this value for gnocchi < 
4.0.0, is to consider long your system takes to process TASKS_PER_WORKER 
metrics... this may be close to ~2seconds (by default, which is still 
very low). in gnocchi 4.0, the scheduling works differently, so above is 
not relevant

>
> metric_reporting_delay = 1

this shouldn't be low, it is only used to trigger debugging logs. this 
should be ~60s (or longer depending on what you're doing with these 
logs). just realise building the report may create significant load in 
your environment since it must consider entire backlog.

>
> metric_cleanul_delay = 10
>
>

this probably shouldn't be done that frequently as generally metrics 
aren't deleted all that often. even then, when metrics are delete, they 
are immediately inaccessible, the cleanup is just to remove stuff that 
is inaccessible.


-- 
gord



More information about the OpenStack-operators mailing list