[SOLVED] Problem HA when lan corosync remains active

[robertozappaterreni]

We have a 5-node plus ceph installation, with separate networks for LAN - COROSYNC and STORAGE. We had a problem on the LAN card of a node but the HA did not intervene I assume because corosync continued to communicate between the nodes quietly having a separate network. Is it correct? Is there any way to migrate VMs autonomously even if only the LAN network dies?

 

[shanreich]

You could try adding the storage network as a fallback for the LAN network, then the cluster should operate via the Storage network while you sort out the problems on the LAN network.

 

[robertozappaterreni]

thank you for your reply,
Unfortunately, the storage network is a completely separate network and does not have access to the internet.
I wanted to understand, does the Corosync network have to go down in order for the HA to work?

 

[shanreich]

Do you have 3 networks? LAN, COROSYNC & STORAGE? In that case, if the LAN network is down and the COROSYNC network continues to work, then the cluster will not migrate VMs since it doesn't detect any problems.

If the COROSYNC network goes down, then HA will also not be able to do anything since there is no network available for migrating the HA VMs. HA is for cases when one node malfunctions and fails, then this will be detected by the other nodes in the cluster and the VMs will get moved to functioning nodes.

 

[robertozappaterreni]

Hai 3 reti? LAN, COROSYNC E ARCHIVIAZIONE? In tal caso, se la rete LAN è inattiva e la rete COROSYNC continua a funzionare, il cluster non eseguirà la migrazione delle VM poiché non rileva alcun problema.

Se la rete COROSYNC si interrompe, anche HA non sarà in grado di eseguire alcuna operazione poiché non è disponibile alcuna rete per la migrazione delle VM HA. L'HA è per i casi in cui un nodo non funziona correttamente e si guasta, quindi questo verrà rilevato dagli altri nodi nel cluster e le VM verranno spostate su nodi funzionanti.

Thank you very much, everything was very clear and how I had sensed it. I will have to monitor the situation in another way and trigger the migration with the APIs in the event of a disaster