Server Load Balancing: Support for Fail-Over
Currently you can define a pool of servers > 2. And there is no way to define a failover cluster.
To do this, I would suggest allowing just a single server in the server list. Then creating a backup server list, in which you could put the backup machine.
For an example, see the load balancer option in pfSense. The fields would be like this:
primary server(s) list:
* servers in this list will receive a portion of incoming traffic.
backup server(s) list:
* servers in this will will only receive traffic if all the primary servers are inaccessible.
In this manner I can setup failover cluster by putting primary and secondary into their respective lists.
So we understand this request, currently the pool has fail over built in (dead nodes are removed and added in real time to the balancing) but you wish to have resources there (or a cluster of resources) doing no work, but becoming responsible for tasks/work when a failure happens in the primary node(s) only?
Udo Hoffner commented
I understand it as classical
active / passive mode
I suggest an acitive / maintanance (do not cut active connections, no new ones) / passive (to becomme active if n - Nodes failed or are in maintance-mode
Gert, your suggestion does indeed work. Thank you. I still think this feature would be nice to have as it would allow additional configurations besides the one I need it for. For example, the backup server in the pool could serve a static "sorry we're down" page...
Angelo, correct. The primary pool works as the server pool works today. A second backup pool does NOTHING until the primary pool is devoid of functioning servers. Then the backup pool becomes active.
For my use case, I will simply put one server each into primary/secondary pools.
Gert, not yet, I will try it out and post back here.
Have you tried the workaround i explained? It should deliver what you expect.
Our app cannot handle more than one active server at a time without modification. We just host the app for a customer and have no way to make said modifications, of course they have no desire to make changes to support our HA infrastructure.
Today we use a dedicated load balancer for this app (all load balancers I have ever used support fail over mode), we would like to remove this load balancer, and replace it with the capabilities of the Astaro.
you can define a fail-over cluster by using 'Availability Group' in combination with a standard 'DNAT'rule.
You are able to add multiple hosts into an availability group in a prioritized order. The group will always use the active host with the highest priority.
If you now use this availability group as the 'translate to' target host object in your HTTP DNAT rule like:
Internet -> HTTP -> External Interface translate to Availability Group -> HTTP
this way you have your fail-over mechanism with two servers.
I also understand your request, but what is the scenario where it makes sense to have fail-over clusters like (2x active, 2x fail-over) why not have all 4x servers active and evenly distribute load.
Is there a reason for that?