Question about nfs cluster

[Jan VdV]

Hi,

I'm a student and as a graduating project I chose to set up a high availability system with load balancing.
Two load balancers (one as a backup...) provide connections to two application servers running a webserver, sql and perhaps some other services. These application servers get their content from two data servers running a nfs.

I followed the howto on setting up a highly available and loadbalanced apache server and the one about a highly available nfs server (both written by Falko Timme). All went well except when rebooting after all was set up, I lost all functionality on the data servers. The screen output was completely messed up, and the system didn't respond anymore until I disabled heartbeat.

My question is: when BOTH data servers go down, is the nfs lost and if so, why? Is there a solution?

I've set up my computers in a network lab of my university where also other student are working. There's a good chance at least one of those students will power off my setup and I don't want all my work disappear at the flip of a switch...

[falko]

Does nfs-server get started when the system is booted? Do you see it in the output of ps aux?
Any errors in the log files?

[Jan VdV]

First of all, because of the total loss of the data servers(well, I'm new to all of this, perhaps it could be saved ;-)), I ended up re-installing them. I did not save the error logs so I cannot provide you with that information.

At the moment the nfs cluster is working like a dream, but I never rebooted both servers at the same time. I only tested if the nfs still worked when one server was down.
The nfs-server is started at boottime and no errors are generated.

Since I don't know what the outcome will be, I'm not keen on shutting both of them down.

But reading between the lines: your answer suggests that the setup should work when both servers go down and are rebooted at the same time?

[martinfst]

In general, a cluster configuration should be started one-by-one if both servers went down. It's a bad idea to make both servers fight who's 1st to take precedence.
I am not familiar with this howto, but as a seasoned sysadmin on linux clusters, thought I'd give you this advise.

[Jan VdV]

@martinfst:

I appreciate your input. Being a student, any information is welcome, especially from experts like you and Mr. Timme.
I'm going to set a small time delay at boottime on one server to avoid that problem. Hopefully it helps.

[falko]

In general, a cluster configuration should be started one-by-one if both servers went down.
This is true. If it is possible, don't shut down both servers at the same time.