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

List:       mesos-dev
Subject:    Re: Design Document for IPv6 support
From:       Evers Benno <bennoe () yandex-team ! ru>
Date:       2016-09-30 12:38:53
Message-ID: d3b80a39-380b-efff-9fd0-77b5147d4189 () yandex-team ! ru
[Download RAW message or body]

Hi Avinash,

thanks for your comments, they raised some really good points about the
isolator.

Coincidentally, I'm currently re-reading this part of the mesos source
in depth, so hopefully I'll be able to address them in-depth in the
coming days, along with some more detail on the dual-stack part.

Best regards,
Benno



On 27.09.2016 03:13, Avinash Sridharan wrote:
> Hi Benno,
>  Thanks for taking the pains and putting in the effort to put up the
> proposal. I think the proposal in general looks very reasonable for a
> standalone IPv6 mesos stack. I had a few comments on the "network
> isolation" section, namely adding support to `network/cni` isolator but
> that can be done in parallel to the effort of enabling IPv6 in the Mesos
> control-plane.
> 
> The only other suggestion I have is that we should probably flush out the
> requirements/path towards a dual stack a bit more. In terms of the effort,
> I think you have already quantified it with the main requirement being UPID
> being able to support multiple sockets (on different address families).
> However, giving more details of what command line options will be available
> for dual stack Mesos (possibly --ip and --ip6 both would be allowed), and
> more importantly what would be the interaction between Mesos v6 masters,
> with Mesos v4 clients or vice versa, would be really helpful.
> 
> On Fri, Sep 23, 2016 at 7:41 AM, Evers Benno <bennoe@yandex-team.ru> wrote:
> 
>> Hi,
>>
>> some of you may faintly remember my announcement that we're planning to
>> add IPv6 support for Mesos roughly half a year ago.
>>
>> While the idea was met favorably, and there seems to be a general
>> consensus that it is a good feature to have, there was also a feeling
>> that a more detailed design document would be helpful to start the
>> discussion.
>>
>> On the other hand, writing code is an integral part of the design
>> process since a purely theoretical design document probably wouldn't
>> survive first contact with the real world, so... long story short, we
>> have been running a small mesos cluster in an IPv6-only production
>> environment for the last few months, and I managed to write up a
>> somewhat more formal proposal at
>> https://docs.google.com/document/d/1jC3CpIROq5Lib40u9o1TUnXyUsKiX
>> T0DQOiwFJa0PJk
>>
>> If you prefer the look of latex typesetting, there is also a pdf
>> version available at
>> https://github.com/lava/mesos/blob/bennoe/ipv6/docs/mesos-ipv6.pdf
>>
>> For those who remember my previous mails, this proposal is actually much
>> more streamlined in comparison, looking to provide minimal core IPv6
>> connectivity which can be expanded in the future if desired.
>>
>> I'm looking forward to receive any comments, thoughts, questions, etc.
>>
>> Best regards,
>> Benno
>>
> 
> 
> 
[prev in list] [next in list] [prev in thread] [next in thread]