make cloud fully HA on Compute nodes

asked 2016-09-24 02:42:20 -0500

SinaMaleki gravatar image

Hi all .

We have choose Mirantis Fuel as our cloud installer platform and I put 3 nodes as controller and i can see that mirantis fuel make controller Fully HA . So in this case the controller nodes are fully HA .

at first step i need to have Compute nodes HA too . So what should I do and is there any documention ?

Thanks .

edit retag flag offensive close merge delete

4 answers

Sort by ยป oldest newest most voted

answered 2017-02-06 09:42:46 -0500

sampath-priyankara gravatar image

Hi SinaMaleki, I think, compute node HA is a common requirement for ops who run legacy pet workloads. You can use openstack Masakari for compute node HA.

edit flag offensive delete link more

answered 2018-07-27 01:11:13 -0500

Madhu gravatar image

To make compute node HA first step should configure shared storage Iscsi LUN /NFS and mount to all compute node, while configure compute node nova user id and group id should be same for all compute node to maintain same id for nova follow the same steps for all compute node .

edit flag offensive delete link more

answered 2017-02-10 14:48:39 -0500

mting gravatar image

I tried use gluster for replcation the instance directory, it works well with 8 compute.

first setup gluster on all compute, then mount the instance directory on gluster fs

one thing need to be careful is that the nova userid/groupid should be same on all server

edit flag offensive delete link more

answered 2016-10-28 14:56:16 -0500

michael-w-turvey gravatar image

Hi SinaMaleki,

You're right that by using three controller nodes, you are able to configure the control plane to be HA. This is because the various services that constitute the control plane are written in such a way that they run on all three nodes, and if any of those three nodes has a failure, the other two nodes are able to continue to communicate with each other and process requests. Specifically, these software services were written to handle a failure of one control node and continue functioning normally.

When you are asking for an HA compute node, what I believe you are asking for a virtual machine to continue running even in the event of a failure of an underlying physical server. (If that's not accurate, then please correct me). Implementing something like this in a general sense on modern hardware isn't practical. You would effectively have to checkpoint VM CPU instruction cycles across multiple physical systems. While I can't say for certain that such a distributed VM doesn't exist, I imagine it would run exceptionally slowly.

Instead, it is best to take the approach that the OpenStack control plane software does, and develop or configure your applications so they are distributed across multiple virtual machines in a way that makes them tolerant to at least a failure of a single instance.

edit flag offensive delete link more

Get to know Ask OpenStack

Resources for moderators

Question Tools

1 follower


Asked: 2016-09-24 02:42:20 -0500

Seen: 411 times

Last updated: Jul 27 '18