[openstack-dev] [Neutron] Ironing out outstanding issues in time for RC1

Ihar Hrachyshka ihrachys at redhat.com
Sun Mar 13 22:14:18 UTC 2016

Armando M. <armamig at gmail.com> wrote:

> Neutrinos,
> We have about ~20 outstanding bugs marked Medium/High/Critical, and we  
> have only one or two days left to have a chance to get them in the gate  
> queue [1]. Can I trouble you to go and make sure patches are up to date  
> and well reviewed?
> Many thanks,
> Armando
> [1] https://launchpad.net/neutron/+milestone/mitaka-rc1

Hi Armando and all,

[Full disclosure: I am really interested in getting stable/mitaka cut off  
asap due to the code sprint starting on Mon where we would like to land a  
number of N bits to master.]

Currently I see 25 unreleased bugs targeted for RC1. I believe the list is  
too broad and does not represent actual team priorities as of right now; I  
suggest we go thru the list and postpone those bugs that either won’t land  
on Mon, or aren't really critical for the release; then cut-off  
stable/mitaka to unblock master.

If you think the list is fine, remember that at this point we should land  
only safe fixes, or those that make release impossible [aka ‘something base  
to the cloud operation is totally broken in Mitaka comparing to Liberty’].  
If you like, you can compare Neutron with its 25 targeted bugs to other  
projects (hint: nova - 2 bugs; glance - 2 bugs; cinder - 5 bugs; horizon -  
12 bugs). If we would start landing all cool stuff that we happen to  
produce in the last week, we would be undermining freeze and release  
process, also raising chances of releasing a pile of broken code.

With that in mind, I went through the target to see what is really critical  
for the release.


We have 18 bugs that are High+. Below is each of those bugs, with [*] mark  
where I think RC target is justified at this point.

https://bugs.launchpad.net/neutron/+bug/1523031: linuxbridge + l2pop + l3ha  
^ while it’s unfortunate, I don’t see how it stands for a release critical  
bug since it affects setup that is not really that common. Also, looking at  
the bug state, I don’t see any work started on it. I would prefer we drop  
it out of RC1 target.

^ fullstack (non-voting) job sometimes fails for native ovsdb interface. No  
idea why it’s critical for the release. Suggesting to postpone to N.

https://bugs.launchpad.net/bugs/1513765 [*]
^ conntrack calls block ovs agent; patch in review optimizes for some use  
cases, but does not tackle the general issue of the agent being blocked;  
patch opens some rolling upgrade scenario concerns too since it touches RPC  
version and method signatures; that said, we indeed want to try to tackle  
it in RC;

https://bugs.launchpad.net/neutron/+bug/1556139 [*]
^ create/delete race when returning new resource body in ml2; the patch is  
up for review and ready for merge; good to go;

^ port ready event sent to nova before dhcp is ready; not a regression: was  
always the case; the proper fix would require a lot of work; definitely  
won’t happen in Mitaka; I believe should be dropped from RC target;

https://bugs.launchpad.net/neutron/+bug/1456073 [*]
^ dvr not ready yet when live migration triggered; targeted for RC on both  
nova and neutron sides; neutron patch in review, has needed +2s; good to go;

^ dvr returns fip targeted pings with fixed ips; patch up for review [WIP];  
not sure whether it will make it; I suggest we don’t wait for that one;

^ physical network aware dhcp scheduler; honestly, that one is rather  
feature-ish; I would untarget it from Mitaka on that ground;

^ ipset race; this one seems abandoned right now; no patches for review;  
long standing issue; I would move it to N;

^ allow to delete other’s ports from your network; while obviously a bug,  
the fix could be also considered rather feature-ish, since it change API  
behaviour; there seems to be no active patches in gerrit for that; I would  
postpone it till N when we’ll have more time to look into the proper fix;

https://bugs.launchpad.net/bugs/1501206 [*]
^ dhcp ports reply to requests from other subnets; potential security hole;  
patch up for review; could indeed be worth looking at for RC;

https://bugs.launchpad.net/bugs/1514056 [*]
^ vlan external connectivity disrupted on ovs agent restart; patch up for  
review [tiny one, need respin]; not a regression; would be fine to see it  
fixed in RC but not a tragedy if it’s skipped;

^ bulk rpc calls time out for l2 agent with default rpc timeout; honestly,  
I don’t believe it’s valid to have it High (there is a tunable for the  
timeout provided by oslo.messaging library); no proper fix up for review; I  
would say, we should postpone to N;

^ exception error fix; breaks i18n freeze; I don’t believe it justifies RC  
target at all since nothing is broken;

^ fullstack database teardown failing; workaround already merged; proper  
fix would probably require changes in db backend; should be moved to N;

^ retry exception in ipam code; no patch in gerrit; bug comments suggest  
that a fix would involve alembic, which is not allowed anymore for M;  
should be postponed to N;

^ agents are too aggressive fetching; base patch in gerrit, though agents  
are not touched yet; honestly, seems a bit late to do such a change since  
we would not have time to stabilize on it; I suggest we defer to N;

^ neutron imports from tempest private code; while nice to have, I don’t  
believe it should be tracked for RC.

(Other bugs are Medium and lower.)


So from the list of 18 High+ bugs, only 5 of them are somewhat critical for  
release success [and even that list could be trimmed with no tragical  

I suggest we focus on those five and cut off stable/mitaka branch when most  
of them land master [or we know that it won’t happen in a day]. For other  
stuff, I prefer we take best effort approach and backport to stable/mitaka  
before final release if fixes are in time for master, AND they really  
justify High priority, AND if they are really proved to be safe.



More information about the OpenStack-dev mailing list