Hello Devs,

I'm starting this thread to discuss the issue we have with subnet
re-creation.
The step to reproduce are described in
https://bugs.launchpad.net/opencontrail/+bug/1722837

We are using OpenContrail from R3.1.1.x branch, but I believe this issue is
in all branches and is connected to haproxy setup and 'in-process caching'
of subnets in api, when the cached subnets are not properly synchronized
with other API processes on other nodes.

For now I did the workaround patch
https://review.opencontrail.org/#/c/36418/ But this patch only tries to
solve the subnet re-creation and I believe the design of in process caching
can cause other issues too, because the addr_mgmt is used in multiple other
cases. One other issue I see  with this cache is it is growing without any
limit, no LRU etc.

So the question is - Am I missing something or is this really design issue?

I think the proper way to fix is - not to use the 'in process cache'  - use
external cache (e.g. redis) or nothing.
The workaround for working cloud can be to setup api proxy to not to do
active-active (round robin or some other loadbalancing) but active-backup
(backup check).

Any thoughts?

Regards,

Petr Jediny
_______________________________________________
Dev mailing list
[email protected]
http://lists.opencontrail.org/mailman/listinfo/dev_lists.opencontrail.org

Reply via email to