Continuing on with building an Oracle RAC cluster on NFS using Amazon EC2, following this build guide.
The setup will be two Oracle RAC nodes and one NFS node. This means that the NFS node is a Single Point Of Failure (SPOF) however this is not meant to be a production system rather a system to provide a environment to perform testing and learning.
I looked at various build guides for enabling High Availability NFS, essentially multiple nodes providing NFS using DRBD, however this requires a block device again. There is a new product released called ElasticDrive which provides a NBD driver for S3. I will be going back and using this product to see if I can get both Openfiler and OCFS2 working with it later.
A bit later…
I got to the point of installing the clusterware and it yet another EC2 related snag!
The lack of a 2nd ethernet device and the inability to setup a Virtual IP address (vip). Whilst I was able follow this post to fake another ethernet device by copying /etc/sysconfig/network-scripts/if-cfg-eth0 to if-cfg-eth0:1, the inability to setup up an virtual IP is going to kill this.
If you have been following this long list of posts on trying to get Oracle RAC working on EC2 this is pretty frustrating to say the least.
I thought about this for a while and did some reading of the EC2 forums. The next attempt I am going to run Qemu inside the Amazon VM (Virtual Machine). So a VM within a VM.
Qemu can convert VMware images to Qemu images so I can go back to using the good Oracle RAC on VMware install guides published by Howard Rogers at the Dizwell site.
Once I have the Oracle RAC install built under VMware I can convert that to run in EC2.
The bonus if this works is
- I can run multiple mini VMs inside one EC2 (Xen) VM.
- When Amazon get direct support for VMware happening I can reuse those VMware images directly.
- If I want to setup virtualization on another hosting services I am already halfway there.
- I can try pointing these Qemu Oracle RAC VMs at external shared disk or NFS and possibly run up a 4 node RAC cluster on one or two EC2 VMs
So that is where I am at. The work done so far with NFS is not lost as I mentioned. I have demostrated that you can provide more storage as required any application which requires more than the normal 150G /mnt mount point provided by EC2.