Filter instances by IP when listing is not efficient enough

[Zhenyu Zheng]

Filtering instance by IP when listing is done by:
http://git.openstack.org/cgit/openstack/nova/tree/nova/compute/api.py#n2457

this is due to that IP address is one part of the instance.network_info JSON,
we have to iterate one by one to find the instance that matches the request.

This is actually not efficient enough, on our public cloud deployment, one
Enterprise user could own over 1000 instances and when they filter by
IP, HTTP 500 timeout raised.

We might consider improve the implementation.

[Zhenyu Zheng (zhengzhenyu)]

Actually, our test is done using Horizon, and it has a limit added automatically
when listing instances. But for IP filters, the limit is removed first to get a
full instance list, then match the IP and then apply limit again.

[Matt Riedemann (mriedem)]

Marking as confirmed since this is a long-standing known issue.

[Matt Riedemann (mriedem)]

Discussed this at the Queens PTG, notes are here:

https://etherpad.openstack.org/p/nova-ptg-queens

We have a few ideas, but need to start working on proof of concept solutions.

[jichenjc (jichenjc)]

I remember this bug was changed to a bp, should close it now?

[Zhenyu Zheng (zhengzhenyu)]

yeah, we should close this now, but how to do it? should we link the bp here?

[jichenjc (jichenjc)]

guess just add bp and patch link might be enough? anyway, let's close it :)

[Zhenyu Zheng (zhengzhenyu)]

The bug is solved with blueprint:
https://blueprints.launchpad.net/nova/+spec/improve-filter-instances-by-ip-performance