[pgpool-general: 6471] Re: trusted_servers

[Muhammad Usama]

On Wed, Mar 20, 2019 at 7:09 PM Alexander Dorogensky <
amazinglifetime at gmail.com> wrote:
>
> Thanks for a very detailed explanation!
>
> How long can it take pgpool to detect that the trusted server (one) is
not responsive?


Pgpool-II issues the ping command "ping -c3 trusted_server" for each server
listed in trusted_servers list after every wd_interval seconds, and whenever
ping result for every server in the trusted_servers list comes out
negative, Pgpool node consider itself as isolated and removes itself from
cluster.
So effectively it takes maximum wd_interval number of seconds plus 3 ping
attempts to detect the trusted_server is not responsive.

>
> We're using pgpool 3.6.9 and we have 2 node (primary/standby postgres
cluster with streaming replication) configuration and each node has pgpool.
>
> We had to set trusted_servers because otherwise pgpool wouldn't detect
network partition events such as when
> network interface goes down on one of the nodes. So I believe in this
configuration we wouldn't have been able to use
> the pgpool's split-brain syndrome auto detection?
>

Pgpool-II should still works very well in case of 2 node cluster ( although
recommended is min 3). And in this case even if the split brain
happens it takes just few seconds for watchdog to detect and recover from
it. So you can still disable  trusted_servers in your setup.

Also to protect against network interface going down you can use the
interface monitoring in watchdog wd_monitoring_interfaces_list
(see *wd_monitoring_interfaces_list* in section 5.14.7 of
http://www.pgpool.net/docs/latest/en/html/runtime-watchdog-config.html#CONFIG-WATCHDOG-SERVER-CONFIGURATION
)


Thanks
Best Regards
Muhammad Usama


>
> Thank you Muhammad!
>
>
>
>
>
> On Wed, Mar 20, 2019 at 8:23 AM Muhammad Usama <m.usama at gmail.com> wrote:
>>
>> Hi
>>
>> On Wed, Mar 13, 2019 at 6:28 PM Alexander Dorogensky <
amazinglifetime at gmail.com> wrote:
>>>
>>> Hi everyone,
>>>
>>> We generally set this parameter to one host to remedy split brain
situations but some customers have unreliable networks and therefore ping
to that host would fail and pgpool would shut down.
>>>
>>> My understanding was that trusted server(s) connection will only be
tested after degenerated state was detected (before triggering failover),
is it correct?
>>>
>>> If not.. how do I configure trusted servers in a flaky network?
>>
>>
>> No thats not correct, If the trusted_server(s) are configured than
Pgpool-II try to ping the trusted_servers after every life-check interval
(wd_interval),
>>
>> Basically the trusted_servers checking was introduced in the pgpool
watchdog, during the early days of watchdog, and as you mentioned its
purpose was to
>> mitigate the split-brain syndrome, by making sure that the Pgpool-II
node which gets isolated from the network because of network partitioning or
>> some other network issue removes itself from the Pgpool-II cluster.
>>
>> But since Pgpool-II 3.5 there have been lots of enhancements made to the
Pgpool-II watchdog specifically in the area of guarding against the
split-brain and
>> other similar scenarios, and the latest versions of Pgpool-II does not
require trusted_servers configuration to detect and avoid split-brain.
>>
>> So if you are using the Pgpool-II > V 3.5 and the Pgpool-II cluster is
configured with odd number of nodes (>=3) than you can safely disable the
trusted_server
>> configuration, and watchdog will take care of split-brain issue itself.
>>
>> Thanks
>> Best Regards
>> Muhammad Usama
>>
>>
>>>
>>> Thanks
>>> _______________________________________________
>>> pgpool-general mailing list
>>> pgpool-general at pgpool.net
>>> http://www.pgpool.net/mailman/listinfo/pgpool-general
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.sraoss.jp/pipermail/pgpool-general/attachments/20190320/8a2802b3/attachment-0001.html>