[devel at bb.net] Multi-master with 3 masters

[Pierre Tardy]

Hi aakash,

Indeed, the triggereable schedulers are exception to that, you need to put
them on all the masters where the corresponding trigger step is configured.

Pierre

On Wed, Sep 19, 2018 at 6:56 PM Aakash Jain <aj355 at cornell.edu> wrote:

> On Thu, Sep 13, 2018 at 1:16 PM Pierre Tardy <tardyp at gmail.com> wrote:
>
>> Hi Aakash,
>>
>> This sounds like a good plan.
>>
>> What you can have is a haproxy distributing the workerload over your two
>> masters, this helps avoid to specifically configure workers.
>> haproxy can work with roundrobin of tcp sockets that what you should
>> configure for your worker master.
>>
>> Also, you should put your schedulers and changesource in your UI master.
>>
>
> I tried to put schedulers only on UI master, however "triggers" doesn't
> work on other master. It fails with the error: "unknown triggered
> scheduler: u'scheduler-name'"
>
>
>> There are some bugs in the way buildbot do master election, and you
>> shouldn't count on it unless prepared to fix some bugs
>>
>> You can restart the masters independently without issues afaik.
>>
>> Regards
>>
>> Pierre
>>
>> On Wed, Sep 12, 2018 at 7:43 AM Aakash Jain <aj355 at cornell.edu> wrote:
>>
>>> Hello,
>>>
>>> I have a multi-master Buildbot configuration with following
>>> configuration:
>>> master1: handling web-server,
>>> master2: handling everything else (including workers communication,
>>> build-request-distributor etc.)
>>>
>>> Buildbot version 1.1.1
>>> Twisted: 17.5.0
>>> 16-core cpu
>>>
>>>
>>> However, master2 is probably heavily loaded and I observe large delay in
>>> scheduling builds (even when workers are idle)
>>> https://github.com/buildbot/buildbot/issues/3661
>>>
>>> I am planning to split master2 into two masters (master2 and master3),
>>> one running on pb port 9989 and another one on pb port 9990. Some of the
>>> workers would be configured to connect to master2 and rest to master3.
>>>
>>> All 3 masters would load same configuration. There wouldn't be any load
>>> balancer, since all masters are running on different ports (workers decide
>>> which master to connect to based on how they are configured). I am hoping
>>> that this would help with the delays, since both master2 and master3 would
>>> be running the build-request-distributor (activity_loop).
>>>
>>> - Does this 3-master configuration looks correct?
>>> - Is this a supported configuration?
>>> - Can I restart one master independent of other?
>>> - Any feedback/comment/consideration?
>>>
>>>
>>> Thanks
>>> Aakash
>>>
>> --
>>
>>
>> --
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.buildbot.net/pipermail/devel/attachments/20180920/3cc91905/attachment.html>