Galera options for WAN (Cross DC replication)

[Oleksandr Drach]

Hello,

I have searched through numerous documents and presentations in Internet to found optimal Galera parameters and here what I've got:

 wsrep_provider_options="gmcast.listen_addr = tcp://8.8.8.8:4567; gcs.fc_limit = 128; evs.send_window=512; evs.user_send_window=512; evs.keepalive_period = PT3S;  evs.inactive_check_period = PT10S; evs.suspect_timeout = PT30S; evs.inactive_timeout = PT1M; evs.consensus_timeout = PT1M;"

Can you suggest and additional parameters to add or improve my existing once?

Thanks in advance!

[Alex Yurchenko]

1) evs.send_window should be greater than evs.user_send_window,
preferably by a factor of 2
2) there seems to be a bug associated with non-default evs.send_window*
settings. So it is no longer recommended.

On 2013-02-26 17:46, Oleksandr Drach wrote:
> Hello,
>
> I have searched through numerous documents and presentations in
> Internet to
> found optimal Galera parameters and here what I've got:

> * wsrep_provider_options="gmcast.listen_addr = tcp://8.8.8.8:4567;

> gcs.fc_limit = 128; evs.send_window=512; evs.user_send_window=512;
> evs.keepalive_period = PT3S; evs.inactive_check_period = PT10S;
> evs.suspect_timeout = PT30S; evs.inactive_timeout = PT1M;

> evs.consensus_timeout = PT1M;"*

>
> Can you suggest and additional parameters to add or improve my
> existing
> once?
>
> Thanks in advance!

--
Alexey Yurchenko,
Codership Oy, www.codership.com
Skype: alexey.yurchenko, Phone: +358-400-516-011

[Oleksandr Drach]

Thanks, Alex!

But your answer is a bit confusing to me :)

On Tuesday, February 26, 2013 6:19:13 PM UTC+2, Alexey Yurchenko wrote:

1) evs.send_window should be greater than evs.user_send_window,
preferably by a factor of 2

So I have to set  "evs.send_window=1024; evs.user_send_window=512;", right?

2) there seems to be a bug associated with non-default evs.send_window*
settings. So it is no longer recommended.

 

If I do understand you correctly you tell that setting custom settings for both evs.send_window and  evs.user_send_window may (probably) raise an exception. Please correct me if I am wrong.

So to sum up:

1. It is not recommended to use custom "evs.send_" settings because there may be a bug.

2. If someone insists on using custom "evs.send_" settings then he has to set evs.send_window=evs.user_send_window*2.

Hope my conclusions were right :)

[Alex Yurchenko]

On 2013-02-27 11:10, Oleksandr Drach wrote:
> Thanks, Alex!
> But your answer is a bit confusing to me :)
>
> On Tuesday, February 26, 2013 6:19:13 PM UTC+2, Alexey Yurchenko
> wrote:
>>
>> 1) evs.send_window should be greater than evs.user_send_window,
>> preferably by a factor of 2
>>
>

> So I have to set "*evs.send_window=1024*;

> evs.user_send_window=512;",
> right?
>
> 2) there seems to be a bug associated with non-default
> evs.send_window*
>> settings. So it is no longer recommended.
>>
>
> If I do understand you correctly you tell that setting custom
> settings for

> both *evs.send_window* and *evs.user_send_window* may (probably)

> raise an
> exception. Please correct me if I am wrong.
>
> So to sum up:

> 1. It is not recommended to use custom *"evs.send_" *settings because

> there
> may be a bug.

> 2. If someone insists on using custom *"evs.send_" *settings then he
> has to
> set *evs.send_window=evs.user_send_window*2.*

>
> Hope my conclusions were right :)

yes, you're correct. Also these are performance optimization settings.
To know if you will ever benefit of those, you need to test.