Migrate VMs on outage without reboot

[Pourya Mehdinejad]

we have a 3 nodes cluster which are correctly working with high availability,
Last night, one of our nodes had an outage, so the HA-Manager migrated all VMs to another node as expected.
But I noticed that all of our VMs get rebooted after migration, this caused some issue with VMs, since some of them are hosting databases, this unexpected reboot may result in data corruption, etc ...

Is there anyway so we can ensure that even on a node outage, all VMs get live migrated to another node with reboot ?
This is really critical for us.

 

[wolfgang]

Hi,

what you mean called Fault Tolerance and is not supported yet.
The main problem with Fault Tolerance you have to mirror the memory always,
so you would need min 40GBit dedicated redundant network for this service.

 

[hk@]

Well, besides databases should support proper locking and transaction-processing and therefore be reboot-safe in terms of data corruption if a node breaks, starting the VM on another node should leave no data corruption if things are done the right way.

Having said this, I know non-transaction-databases and -apps are famous, there you should consider doing a fail-safe database-server-setup not relying on any operation-system failsafe, eg mysql offers clustering and and postgresql does so too and so on.

 

[Pourya Mehdinejad]

Hi,

what you mean called Fault Tolerance and is not supported yet.
The main problem with Fault Tolerance you have to mirror the memory always,
so you would need min 40GBit dedicated redundant network for this service.

We can provide 10Gb SFPs to each of our servers dedicated to this service, but can we have it on Proxmox, maybe as a paid service or it's not supported at all ?

 

[wolfgang]

This feature in not stable yet.
And as hk@ wrote it makes more sense to achieve this at DB level,
because in this way you have only keep the data in sync and not the hole os.

 

[Pourya Mehdinejad]

This feature in not stable yet.
And as hk@ wrote it makes more sense to achieve this at DB level,
because in this way you have only keep the data in sync and not the hole os.

I understand that, But it would be best if none of our users experience downtime or interruption on their Machines.
we are strongly looking for this feature..
Is there any other solution ?

 

[LnxBil]

Is there any other solution ?

Use a proper database cluster - as it was already said before.

Oracle Real Application Cluster (RAC) is very nice for that. You need at least two nodes and your client application will not notice any downtime with the use of Transparent Failover (TAF). If a node crashes, the connection will automatically switched to another node and an ongoing transaction is rolled back and redone.