Open Stack

Le 2016-11-03 19:27, Joshua Harlow a écrit :
> *Next-gen oslo.service replacement*
> 
> This one may require a little more of a plan on how to make it work,
> but the gist is that medhi (and others) has created
> https://github.com/sileht/cotyledon which is a nice replacement for
> oslo.service that ceilometer is using (and others?) and the idea was
> to start to figure out how to move away from (or replace with?)
> olso.service with that library.

If people are interested, the doc is here [0]

> I'd like to see a spec with some of the details/thoughts on how that
> could be possible, what changes would still be needed. I think from
> that session that the following questions were raised:
>
> - Can multiprocessing (or subprocess?) be used (instead of os.fork)
> - What to do about windows?

I have already solved those two by using multiprocessing, disable 
SIGHUP and write down the limitation [1]

> - Is it possible to create a oslo.service compat layer that preserves 
> the oslo.service API but uses cotyledon under the covers to smooth the 
> transition/adoption of other projects to cotyledon

Not sure it's easy to do, cotyledon API encourages user to not create 
any python objects before the process is forked to ensure we didn't have 
any rpc/mysql/... connections open and unused, FDs open, lock acquired, 
(put any thing here that can result in race when using os.fork()). While 
oslo.service forces the user to create objects before the fork occurs.

> - Perhaps in general we should write how an adoption could happen for a 
> consuming project (maybe just writing down how ceilometer made the 
> switch would be a good start, what issues were encountered, how they 
> were resolved...)

This is avialable here [2]

[0] http://cotyledon.readthedocs.io/en/latest/
[1] http://cotyledon.readthedocs.io/en/latest/non-posix-support.html
[2] http://cotyledon.readthedocs.io/en/latest/oslo-service-migration.html
-- 
Mehdi Abaakouk
mail: sileht at sileht.net
:w
irc: sileht