[prev in list] [next in list] [prev in thread] [next in thread] 

List:       openvswitch-dev
Subject:    [ovs-dev] OVN - Pluggable Distributed DB Infrastructure for OVSDB
From:       gal.sagie () gmail ! com (Gal Sagie)
Date:       2015-06-30 21:10:30
Message-ID: CAG9LJa5pWeL11HUU20F3j8ebO0cpQuBTubh+-z39V8AZjAiB8g () mail ! gmail ! com
[Download RAW message or body]

Thanks for the inputs, i cant say i agree because:

1) This is not just a scale issue, this is also a Single point of
failure/HA/Redundancy problem

2) As the code size increase it will be harder to do this abstraction and
it will be more error prone

3) I wasn't suggesting to actually replace the DB now, just make the API
pluggable now and plug it to ovsdb-server
    current implementation, so in the future we would have an easier and
safer way to plug other solutions
    because we gave this point some thought at this stage in time.

But since i am the only one thinking like that, i respect everyone else
opinion :)

On Tue, Jun 30, 2015 at 11:42 PM, Kyle Mestery <mestery at mestery.com> wrote:

> On Tue, Jun 30, 2015 at 2:35 PM, Russell Bryant <rbryant at redhat.com>
> wrote:
>
>> On 06/30/2015 04:42 AM, Gal Sagie wrote:
>> > Hi Justin,
>> >
>> > The idea was not to choose one database implementation over the other,
>> > but design things in such a way
>> > that its pluggable, so in terms of OVN, switching between
>> > implementations is transparent (as much as possible)
>> >
>> > I think that different setups can have different scale criteria and
>> > might need different DB solutions.
>> >
>> > I think the IDL library (the python and C implementations) will be
>> > rather easy to adjust as OVSDB have clear API
>> >
>> > So doing this work now, we can easily change and experiments with
>> > different DB solutions
>>
>> Ben and Justin have made a pretty good case that this is a premature
>> since we really haven't done much real testing to see how ovsdb-server
>> scales.  This could potentially be a pretty big distraction and I'm not
>> sure it's worth it.  I don't think adding this later will be
>> significantly more difficult than it is today if it's needed.
>>
>> If someone wants to do some experimentation in the short term, it's
>> certainly possible.  There is an API generated from the ovsdb schema
>> that you could re-implement.
>>
>> It might be better time spent to think about OVN testing and how we can
>> do good, reproducible scale tests.  Ben has done some awesome work here,
>> but it seems we may need to a multi-node environment since at some point
>> it sounds like the number of processes on a single host becomes a
>> problem in the simulated test env.
>>
>>
> +1
>
> Thanks for taking the time to answer this one everyone, and I agree with
> Russell that it makes more sense to do this later down the road rather than
> distract things at the moment.
>
> Thanks,
> Kyle
>
>
>> --
>> Russell Bryant
>> _______________________________________________
>> dev mailing list
>> dev at openvswitch.org
>> http://openvswitch.org/mailman/listinfo/dev
>>
>
>


-- 
Best Regards ,

The G.

[prev in list] [next in list] [prev in thread] [next in thread]