Results 1 to 1 of 1
Okay, here is the set up. I have two Xen boxes, we'll call testxen1 and testxen2. These boxes each host failover postgres db boxes. There are currently two strings, we ...
- 05-26-2010 #1Just Joined!
- Join Date
- Nov 2006
- Posts
- 5
Creating Xen failover for hosted clusters
Okay, here is the set up. I have two Xen boxes, we'll call testxen1 and testxen2. These boxes each host failover postgres db boxes. There are currently two strings, we will call testdb1 and testdb2. Another pair are qadb1 and qadb2. Now testdb1 and qadb2 are hosted on testxen1. testdb2 and qadb1 are on testxen2. So if testxen1 goes down the boxes on testxen2 go live. Each db string has their own database. Now here is the rub, the powers that be decided iscsi was far superior to NFS and have replaced my NetApps with Equillogics. So now, I have the added complexity of having to use gfs.
So basically testdb1 and testdb2 need to share an iscsi volume. qadb1 and qadb2 need to share another iscsi volume. These are in failover, so if one detects a problem with the other then it takes over the running of the db.
Now despite working with linux since 99, I am officially a newb when it comes to iscsi/gfs. So I have been googling and finding how to get this all up. At first cman wouldn't load the module, but I found that though the kernel gets updated, they don't keep the modules in synch, so you have to downrev the kernel to get it to load the modules needed. It looks like DLM is the way to go, but trying to muddle through all of this. I did manage to get cman to start on testdb1 and testdb2, but did not include fencing at this time. I am hoping that my understanding of fencing is that if a problem is detected it will tell the server to reboot or something. One suggestion was to use fence_xvm, but that requires that you have the cluster on the dom0 or main host. Since the dom0 is two different hosts, not sure how that would work.
All these years we have been running a simple failover using a patched version of piranha from Sebastian Bonnet, that allows for scripting in failover. The standard only allowed scripting in load balancing mode. So my understanding of Red Hat cluster manager is weak. I can sort through that, but it seems that what I am trying to do is a bit complicated. I have not found any references to this type of configuration and how to set it up.
So basically, two xen boxes in failover. They host two different cluster domains, each hosting a db box on a different host machine. Each db string sharing an iscsi volume using gfs. (not set this up yet as this is my first foray into gfs also and trying to sort out all the pieces.)
Frankly, I hope that the speed of this setup blows nfs away, because it just seems like an awful convoluted way to set up file sharing. NFS has worked great for us all these years and is quick to set up. Set up share permissions on the host, create a mount point on the client and voila. Done. Move on. I have been spinning wheels on this project, after having to learn Xen, fairly easy and less than a day without distractions. Bogged down with cluster, and only just starting on the gfs.
So that's my story and sticking to it. I will continue to try to sort this out and get it up and running, but felt I could use another mind or two. Mine is about fried.


Reply With Quote