Hi, I want to try the HA cluser in proxmox, I read post and "how to" wiki, I need some confirmation to start: 1. Basic system could be: 3 server and external iSCSI NAS where put VD (visible to all 3 server)? 2. Is there a problem in hardware are promiscuous if migrate VM from one server to another? 3. With this installation is mandatory to have fences device? Thanks for help
Needs for HA system
- Thread starter nicothebrush
- Start date
Hi,
They all should be in on one location, to prevent network latency.
If you use for Vm vcpu type kvm64 it will work.
with HA you should always use fencing.
This is the only way you can be sure the an node is down and you do not start a resource twice.
Yes this is possible.
They all should be in on one location, to prevent network latency.
If you use for Vm vcpu type kvm64 it will work.
with HA you should always use fencing.
This is the only way you can be sure the an node is down and you do not start a resource twice.
Thanks for reply Wolfgang, I comment "inline"
... or in a "HA-like" installation, may I use only 3 server and if one go in fail I manually start VM on the other? (I thought that with 3 server, 2 server can "decide" if the remain one is in fail mode)
>>> OK I'll use it and all server stay in one network LAN!
>>> I read about fencing device but for a test (minimal) installation what do you advise to buy?
... or in a "HA-like" installation, may I use only 3 server and if one go in fail I manually start VM on the other? (I thought that with 3 server, 2 server can "decide" if the remain one is in fail mode)
The tricky part is between the fail and failover. when a node is down, the other nodes dont know if it just hung, or something else happend. if they decide to failover, and the "down" nodes was just hung up for a bit, you'll suddenly have to virtual machines (two instances of the "same" one) writing to the same storage. the only safe way, is to put that node in a known state.
fencing is not as hard as it seems, but you do need to test it very carefully because a missed setting could turn a single node failure into a cluster wide failure. if you have a managed switch you can use that, but its really better to get power fencing. network managed power switches dont cost as much as disaster recovery.
fencing is not as hard as it seems, but you do need to test it very carefully because a missed setting could turn a single node failure into a cluster wide failure. if you have a managed switch you can use that, but its really better to get power fencing. network managed power switches dont cost as much as disaster recovery.
Thanks pixel,
so I my installation, if I understand your tips, I'd try to attach a fence-UPS to every server? do you now some brand?
so I my installation, if I understand your tips, I'd try to attach a fence-UPS to every server? do you now some brand?
for testing only you can use the pve-fencing agent this fence only the vm.
this is only to get an idea about it.
to use this you must install fence-agents-pve
read here
https://pve.proxmox.com/wiki/Fencing
this is only to get an idea about it.
to use this you must install fence-agents-pve
read here
https://pve.proxmox.com/wiki/Fencing
Last edited by a moderator:
APC are known to work well, either independent power switch or integrated into a UPS.
Thanks tom,
do you have a model in particular? I try to google APC, UPS, fence ... but ... no particular model found...
(and often I see fenche for garden )
do you have a model in particular? I try to google APC, UPS, fence ... but ... no particular model found...
(and often I see fenche for garden )at work we use apc7901. each has 8 sockets and one network port. our servers have dual power supplies so have 2, each on different power sources.
Power fencing is great. If your nodes have an ipmi card of some sort, e.g. iDRAC or iLO you can also use that
Theres also network fencing if you have a decent managed switch, but thats really more of a while you wait for the power fencing hardware.
exposing the management interface (ipmi, idrac) etc is a big risk. your giving console access to a production network. id only do that if theres no other option. ive also had an idrac interface hang and had to have someone manuall power cycle that node. this happend when we were waiting for the fencing hardware.
exposing the management interface (ipmi, idrac) etc is a big risk. your giving console access to a production network. id only do that if theres no other option. ive also had an idrac interface hang and had to have someone manuall power cycle that node. this happend when we were waiting for the fencing hardware.