[RFE] Caching instance id in the metadata agent to have less RPC messages sent to server

[Slawek Kaplonski]

Problem like that was mentioned in Vancouvers's "Challenges & lessons from operating OpenStack at Scale" forum session (https://etherpad.opendev.org/p/challenges-largescale)
In some cases, neutron metadata agent may put a lot of load on the RPC bus and neutron server as every time VM is asking for some of the metadata, this agent needs to find out what the id of the instance is. Because of that it sends RPC message to get port details every time.
It could be potentially avoided by some simple caching of the instance id in the agent's side.

[Sahid Orentino (sahid-ferdjaoui)]

I may missing something but don't we want to just enable the cache using memcache for the metadata agent?

This should be added in metadata_agent.ini

[cache]
backend = oslo_cache.memcache_pool
enabled = True
memcache_servers = ...
expiration_time = 300
memcache_pool_maxsize = 20

We ran into such issue and this clearly fixed/improved the issue.

[Rodolfo Alonso (rodolfo-alonso-hernandez)]

This RFE was approved during the drivers meeting celebrated today [1]. No spec is required.

The developer will check first the proposal in c#1, using the oslo_cache library. Once tested, it will be documented in the Neutron repository.

Thanks for this proposal.

[1]https://meetings.opendev.org/meetings/neutron_drivers/2023/neutron_drivers.2023-06-30-14.00.log.html

[Lajos Katona (lajos-katona)]

Related patch:
https://review.opendev.org/c/openstack/neutron/+/887776

[OpenStack Infra (hudson-openstack)]

This issue was fixed in the openstack/neutron 23.0.0.0b3 development milestone.