Implement an API call to disable recovery for a single cluster

[pythian-ciciliani]

On scenarios where an Orchestrator cluster is managing hundredths of MySQL pods, and you need to be able to disable recovery for a specific one, you need to either keep a list of all pods in RecoverMasterClusterFilters and add/remove items (with the corresponding reload of all orchestrator processes) or downtime each pod member individually.

It would be nice to be able to disable recovery for a single pod through the API

$ orchestrator-client -c disable-cluster-recovery -a mycluster1

$ orchestrator-client -c enable-cluster-recovery -a mycluster1

As discussed with @shlomi-noach, it would also be interesting to include a visual aid on the GUI and in the ascii topology output to remind users that recovery for that cluster is disabled

[shlomi-noach]

Things to consider:

• if disable/enable are supported, then show or get should also be added
• should there be a timeout to disable?
• what if a human triggers a manual failover/takeover? In line with existing behavior it should be allowed.
• would such failover/takeover re-enable recoveries on a disabled cluster?
• should failure detection report on failed disabled clusters?

[yangeagle]

If we set the master of a mysql cluster begin-downtime, can this achieve the goal?

[shlomi-noach]

yes, a being-downtime disables recoveries; however, notice that begin-downtime applies to a single host. So you'l need to make sure to always apply it on the master of the topology (if the master changes then apply it to the new master), or just apply it onto all cluster instances.