public__vip doesn't failover if depends on ping_vip__public (ping checker)

[Roman Sokolkov]

I have MOS 9.1 cluster and while performing HA tests i've found that public__vip doesnt't failover properly, if ping checker enabled.

Steps to reproduce:
1. Ping public vip from outside
2. Hard poweroff controller with public__vip

Expected result:
public vip is pingable

Actual result:
publiv vip is not pingable

Workaround:
Disable ping checker before deployment in UI.

or

"crm configure edit" and remove

location loc_ping_vip__public vip__public \
        rule -inf: not_defined pingd or pingd lte 0

BTW, i've found guilty location in pacemaker
location clone_ping_vip__public-on-node-5.vpn.elits.com clone_ping_vip__public 100: node-5.vpn.elits.com

Can anyone comment why it presented?

[Vladimir Kuklin (vkuklin)]

Roman, could you please share logs or details whether the specified gateway is pingable from other controllers? It is expected that we run public VIP on the nodes that have actual access to the outer world. I would mark this mark as Invalid unless the gateway is pingable from those other controllers.

[Vladimir Kuklin (vkuklin)]

I looked into the environment and noticed that nodes 1 and 3 have replaced deployment info which sets run_ping_checker attribute on those nodes to false, which in turn disables pinger service on those nodes, and thus public__vip is not running on the nodes. It is almost surely a flaw in the upgrade procedure which the cluster undertook - we need to flush particular replaced deployment info if we want regular serializer to take effect or we need to feed additional data through extensions pipeline instead of replacing it. Thus I will mark this bug as Invalid.