Fail over cluster configuration on windows servers

[Derek Sylvester]

Hi,

Trying to get a fail over cluster configuration up and running on windows server 2012 R2 (not a mirrored cluster but a fail over cluster as described in https://www.rabbitmq.com/pacemaker.html). Has anyone done this? If so, do you have any detailed steps of how?

Thanks.

[Derek Sylvester]

Obviously I'm not asking how to use pacemaker etc on windows. Just trying to get the rabbitmq nodes configured correctly (i.e. to use the same shared storage location and the same node name).

[Derek Sylvester]

Or is there a windows version of the instruction anywhere?

On Monday, July 25, 2016 at 3:26:15 PM UTC+1, Derek Sylvester wrote:

[Michael Klishin]

Derek,

There is no Windows version for our Pacemaker guide that I know of. I personally don't have

any experience with Pacemaker on Windows to comment on how significant the differences

can be.

--
You received this message because you are subscribed to the Google Groups "rabbitmq-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to rabbitmq-user...@googlegroups.com.
To post to this group, send email to rabbitm...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

--

MK

Staff Software Engineer, Pivotal/RabbitMQ

[Derek Sylvester]

Hi Michael,

As I mentioned, it's not really pacemaker on windows I was looking for. I guess I made a hash of explaining the question. I'm trying to understand how, on windows servers, to achieve the same thing that is described in that article on Linux. Specifically the sections about configuring RabbitMQ to use a shared storage location (a simple network share) and how to set up the nodes such that they have are a hot standby for each other (I can configure MS failover cluster and network storage with no issues, I just can't get two nodes to be happy using the same storage location or configure them with the same node name etc).

Thanks.

[Michael Klishin]

Pacemaker does not have two live nodes with the same node name, that's not supported

by the Erlang runtime for fairly obvious reasons.

Shared storage is also something to approach very carefully because concurrent writes from

2 nodes will step over each others feet.

Most of what Pacemaker does is to ensure only one node uses the shared resource (e.g. IP address or

node name).

Search list archives for "hot stand-by", there are earlier discussions about how to maintain a stand-by with

exchange federation and TTL for messages.

[Joseph L. Casale]

On Tuesday, July 26, 2016 at 2:19:43 AM UTC-6, Derek Sylvester wrote:
> I just can't get two nodes to be happy using the same storage location or configure them with the same node name etc).

What does this provide over and above the existing facilities which are native to the broker?

Mirrored queues and consumers with either multiple host connection strategies or indirection

with a load balancer?

The two concepts are not interchangeable, when the application is itself not fault tolerant

you then need some level of platform availability with a virtual ip (nfs etc) however RabbitMQ

natively implements this and its known to work so you don't have to re-invent the wheel.

jlc

[Derek Sylvester]

Ok, so the example on your web site that I linked to suggests that, for high availability with a two node configuration, only a single node runs at a time, with the other being an identically configured standby but with the broker not running. Some higher level service determines when the current live node is down and starts up the other.

"With the high availability setup described in this guide, when a node fails, the durable queues and the persistent messages within them can be recovered by a different node...

The main trick to HA Rabbit is to ensure that when the passive node becomes the active node, it must have the same node-name as the failed node. It must also have read and write access to the files in the shared storage."

So, the two nodes use the same mnesia persistent storage, with only on node active at a time. And both nodes must have the same node name. The example goes on to describe how this is done on linux using pacemaker and drbd. What I'm trying to do is configure node1 to use a network share, and node2 to use the same network share. And for node2 to have the same node name as node1 so that when it comes only it can pickup where node1 left off and read the same mnesia database. This is what the suggested HA configuration is, is it not?

So the question is how to configure RabbitMQ on windows servers to achieve this. I can find no examples of how this has been done. Surely this is possible on windows, or am I completely missing something here?

Thanks.

[Derek Sylvester]

Hi Joseph, I'm not trying to reinvent the wheel here (at least I hope not). Is configure a similar windows server based setup as the example on the RabbitMQ web site for HA (not a cluster with mirrored queues)

Thanks.

[Joseph L. Casale]

On Tuesday, July 26, 2016 at 12:05:52 PM UTC-6, Derek Sylvester wrote:
> Hi Joseph, I'm not trying to reinvent the wheel here (at least I hope not). Is configure

> a similar windows server based setup as the example on the RabbitMQ web site for

> HA (not a cluster with mirrored queues)

I can not account for the article however, pacemaker, corosync and the likes are used

when the applications themselves can not account for high availability.

RabbitMQ can and client software exists for many platforms that can take advantage

of it.

Cut out all of those moving parts, even if you get it working, I assure you it will be far

more fragile than letting RabbitMQ handle HA and using good client software.

jlc