Dear All,
I am newbie to Linux Clustering, i have 2 standalone CentOS servers, i want to setup a cluster on those servers, to synchronize between each other, and to make a one as standby to the other, if a one fails the other will switchover.
I will mostly use Apache, Mysql, and PHP.
I have read the "Cluster Administration" document, i found that there are multiple methods to setup the cluster, actually i want to ask expert people in the clustering, which method is the most proper one, and should i have use a fence device, which one i will preferably use.
Thanks for assistance.
_________________________________________________________________ Hotmail: Powerful Free email with security by Microsoft. https://signup.live.com/signup.aspx?id=60969
On 10-07-05 04:36 PM, Torintino T wrote:
How you implement the application fail-over will depend on your approach for each of you high-availability applications. Have you thought about how you will sync data (DRBD, GFS2, NFS, etc)?
As for the cluster itself, that is, the stack underneath the applications, it is not too hard to setup. Install the clustering group and then configure the '/etc/ais/openais.conf' file and the '/etc/cluster.conf' file. Of course, how to do that is a bit bigger question.
As for fencing; Yes, yes and definitely Yes. If your servers have IPMI (ie; Dell's DRAC, HP's iLO, etc) then use IPMI for fencing. Otherwise you will need an external fence device like an addressable PDU.
A word of note; Clustering on CentOS is fairly old now. Old enough, that I've personally moved off of it Fedora 13 for non-production use in order to prepare for the "real soon now" RHEL/CentOS 6 release. If you're not imminently deploying your cluster, you may want to explore this.
Hi Digimer
Thanks for your helpful reply.
Actually i didn't determine yet by which method i will sync the data until i will overlooking into searching.
Regarding fencing, i will use 2 standalone Dell R210 servers.
So which Linux distribution, you recommend me to use because i am about to implement it in the next few days.
and Which the method i will follow to create the cluster, referencing to the "Cluster Administration" document. as i see : Luci, Conga, Cluster Manager, etc..
Your help is appreciated
Thanks
_________________________________________________________________ Hotmail: Free, trusted and rich email service. https://signup.live.com/signup.aspx?id=60969
Digimer,
to add something i don't want to use RHEL because of the license fees.
I just would like from you please to give me the roadmap to follow it.
Thanks alot
From: torintino2@live.com To: linux@alteeve.com; centos@centos.org Subject: RE: [CentOS] CentOS Cluster Date: Tue, 6 Jul 2010 00:18:25 +0300
Hi Digimer
Thanks for your helpful reply.
Actually i didn't determine yet by which method i will sync the data until i will overlooking into searching.
Regarding fencing, i will use 2 standalone Dell R210 servers.
So which Linux distribution, you recommend me to use because i am about to implement it in the next few days.
and Which the method i will follow to create the cluster, referencing to the "Cluster Administration" document. as i see : Luci, Conga, Cluster Manager, etc..
Your help is appreciated
Thanks
Hotmail: Free, trusted and rich email service. Get it now. _________________________________________________________________ Your E-mail and More On-the-Go. Get Windows Live Hotmail Free. https://signup.live.com/signup.aspx?id=60969
On 10-07-05 06:06 PM, Torintino T wrote:
The best thing I can do at this time is point you to my *incomplete* How-To I was working on for CentOS. It's not perfect, it's not even vetted. In fact, I've set it aside and have started re-working it for Fedora 13 at this time, with the goal of eventually re-writing it again for CentOS 6.
With that convoluted warning and caveat, take a look at this:
http://wiki.alteeve.com/index.php/2-Node_CentOS5_Cluster
The work-in-progress Fedora 13 variant is here:
http://wiki.alteeve.com/index.php/Two_Node_Fedora_13_Cluster
The CentOS original document will not be completed any further than it is. I've decided to move away from the old cluster stack. The Fedora variant is very actively being developed (I put a couple hours into each night) as is not really safe to follow.
This makes it hard for me to recommend an operating system to base your cluster on. Obviously, you want the stability of a server-class operating system like CentOS. However, you'll be working hard to get an old-fashioned cluster off the ground. As such, I will leave that decision up to you.
As for a roadmap - No one can give you that. No two clusters are the same. They are designed to solve a particular problem. What I can give is advise;
Answer these questions:
1) What problems are you trying to solve with a cluster.
2) What research have you done that concludes that a cluster will introduce fewer problems than it solves.
3) Are you prepared to learn and then maintain the cluster?
4) Are you willing and able to take the time to throughly test and debug your cluster before going live? Do not expect to get your cluster live in anything less than a month's time.
Clustering is not something you can briefly jump into and then move on from.
On 7/6/10, Digimer linux@alteeve.com wrote:
Hi, pardon me for jumping into this thread midway. I've been looking at the cluster thing, being still lost as to what/how exactly so would like to know what are the reasons you choose to roll your cluster using this approach instead of using something like Eucalyptus?
to add something i don't want to use RHEL because of the license fees.
I just would like from you please to give me the roadmap to follow it.
Hence the reason you are in a CentOS list? We also have RHEL's cluster software which is obviously robust, to say the least. Their mailing list is active and very competent.
Forget about rolling your own setup with newer versions of software given the experience you have, it's hard enough. Use the tools provided by upstream and you'll have a higher success rate. I seriously doubt you have requirements that exceed the capabilities of the upstream packages.
I have a setup exactly like what you describe and use DRBD instead of shared storage given the active/passive nature, it works quite well.
Roll your sleeves up and bang one out in the lab, then ask questions...
You should give a thorough read of the rhel cluster wiki, for eg. http://sources.redhat.com/cluster/wiki/FAQ/CMAN http://sources.redhat.com/cluster/wiki/FAQ/GeneralQuestions http://sources.redhat.com/cgi-bin/cvsweb.cgi/cluster/doc/usage.txt?cvsroot=c...
I would also suggest not using a high level tool to do your configuring, you'll miss so many nuances in your learning curve, so forget Luci/Ricci and system-config-cluster.
You'll want "cman.x86_64 rgmanager.x86_64 openais.x86_64" for this minimal setup as you can skip the clustered file system bits and just use DRBD with ext.
You'll need to pay special attention to fencing, the cluster requires this so it can operate with safe assumptions. Your drbd conf needs to deal with and react to split brain syndrome.
Anyway, that's lots for now...
jlc
2010/7/5 Torintino T torintino2@live.com:
There are several options depending on your requirements and your resources.
For synchronizing the web root, you can opt for something as simple as rsync or checkout from a central repository. For mysql, there are some creative methods including fully clustered mysql servers ($$), dump/restore from primary to secondary.
If you opt for shared storage there are a few other options but it requires a little more complex setup.
Easiest situation is a SAN volume that you can swing from node1 to node2. This can be backed by an actual SAN, by a Linux host running iSCSI services, DRBD, GNBD, etc.. Shared storage requires fencing, either through the "standard" mechanisms or via logic on the application side.
If you just have two nodes and no central storage, IMHO the easiest setup would be a DRBD volume. Use luci to build the cluster with a virtual IP that swings between the two nodes. This is rather simple:
Install drbd, luci, ricci.
Configure the drbd volumes between the two nodes. Ample documentation is available and the process is fairly trivial. If at all possible, build a second network for this traffic.
Configure clustering, using luci, on the first node. At minimum you'd setup a parent service with some child services of the web root and mysql database mount, virtual IP, and the actual httpd/mysql services. For people new to CentOS clustering, this can be a little confusing as the cluster "service" you setup initially is not just a network service such as apache, dns, etc., but an application. The application then has dependent "services" which can include a filesystem, ip address, daemons, etc..
Note that "exclusive" doesn't mean a service runs on just one node, but that *only* that service runs on a node. If you set a service as exclusive, unless you have a separate node for each exclusive service, you may run intro frustration when you try to failover nodes.
Configure your fencing. At simplest case you can configure a policy that will, in essence say "Die!" to the other node. If the other node doesn't die, the active node can kill the other node in various and sundry ways including pulling the power, shutting down the node via the virtual machine host, etc.. You can also be a bit more polite and install GFS and let the GFS service handle who gets the resources. The tradeoff is in complexity.
Once it's configured on the first node, import the luci configuration onto the second node.
-
Digimer -
Emmanuel Noobadmin -
John R Pierce -
Joseph L. Casale -
Kwan Lowe -
Torintino T