[ClusterLabs] CCIB migration from Pacemaker 1.x to 2.x

[Andrei Borzenkov]

23.01.2021 19:10, Sharma, Jaikumar пишет:
> Hi guys,
> 
> I'm newbie to high availability clusters, pls excuse me - learning tools stack (corosync & pacemaker).
> 
> In fact, our high availability solution is based on Debian 9.x (pacemaker 1.x and corosync 2.x) - which worked as expected.
> 
> This  cluster application uses postgresql 9.6 and Wildfly 10.x
> 
> I've tried to migrate this solution to Debian 10 (pacemaker 2.x & corosync 3.x), but unfortunately we have got hiccups (described below).
> 
> Debian 10 based cluster using Postgresql 11 & Wildfly (10.x) and we have got data sync issues between two nodes.
> If I reboot master node after it comes back it joins the cluster in a usual way and databases of both nodes are immediately gets in sync (As many times I do, it works as expected).
> 
> But if plug out the network cable of the master node

Which cable? There is corosync network, there is application network,
there may be backend network, and they may use different adapters.

> then other node becomes master and application is operable and as soon as I plugin in the network cable back to the master node then database sync issues occur.
> 
> Have tried to find the reason but no success so far...
> Attached is my template.cib .
> 
> Would appreciate any pointers as what may be the reason behind this!
> 

Assuming "network cable" means corosync network - you have

	stonith-enabled="false" \
	no-quorum-policy="ignore" \

so when connection between nodes is interrupted, both become master and
database goes out of sync.

> Thank you in advance!
> 
> Regards,
> Jaikumar
> 
> 
> _______________________________________________
> Manage your subscription:
> https://lists.clusterlabs.org/mailman/listinfo/users
> 
> ClusterLabs home: https://www.clusterlabs.org/
>