On 12/4/2019 8:03 PM, 양유석 wrote:

Hi, stackers.

I wonder how other system deal with the problem we encountered. Any advice would be appreciated.

He had a over 1000 hypervisors and hit some connection bottleneck in a DB. The DB instance handle all Openstack services (nova, neutron, cinder, glance ...)

Also, we have some additional systems (kubernetes, consul...) which periodically request openstack API.

Since most of request from additional systems is READ, I am trying to find cache approach for Openstack service. Although some services like keystone has their own caching layer, I could not find general solutions.

Is there any project to solve this problem (reduce DB overhead in a lot of read overhead) I found oslo.cache only provide decorator for a function, so wonder there is any service level approach.

Thanks

It's not a cache but if you're looking for help with READ load you could setup a read-only copy of the nova database(s) and configure a slave connection: https://docs.openstack.org/nova/latest/configuration/config.html#api_databas... Like the oslo.cache usage in nova, that's also on a per-function/query basis and it's been awhile since there was a concentrated effort on making more queries able to read from the slave connection, but if there are more that could be identified as beneficial to use that read-only connection then I'm sure patches would be welcome. -- Thanks, Matt