Hobbit following cluster nodes (or can I include a file in localclient.cfg)
Hi,
I have a passive/active cluster on 2 Linux servers :
- on machine MA :
- we use to run the applications A01 and A02
- with the IP addresses IP0A and logical IP0A:1
- we want to monitor file systems FS01, FS02, applications A01 and A02.
* on machine MB :
- we use to run the applications B01 and B02
- with the IP addresses IP0B and logical IP0B:1
- we want to monitor file systems FS03, FS04, applications B01 and
B02.
My first solution was to install one hobbit client on each machine MA and MB. Problems occurs when the cluster switch from a node to the second node :
- Applications A01 and A02 could be started on machine MB who keep running applications B01 and B02. Machine MB will have 4 file systems now : FS01, FS02, FS03, FS04 to monitor and 3 IP addresses : IP0B, IP0B:1 and IP0A:1
- Also, it possible to have the opposite : all applications (and file systems and IPs) could be on machine MA.
- Finally, applications can be switch from MA to MB : FS03, FS04, B01 and B02 and IP0B:1 could be run on MA and machine MB could run FS01, FS02, A01 and A02 and IP0A:1
As you can guess, it's a total mess to monitor. Everything is in red when machine MA acts as MB and so on. We had several ideas :
- Install a hobbit Client linked to the application cluster group of [A01,A02] and [B01,B02], it will follow the applications who needs to be monitored. -> problem : when everything is running on only one machine, the second one switch to PURPLE (none hobbit clients report datas)
- Install a hobbit Client linked to the application cluster group of [A01,A02] and [B01,B02] and also install another hobbit client on each machines -> problem : with 2 machines, one could be running 3 hobbit client are the same time, image with 4 node cluster, one machine could run 5 hobbit clients each on them checking RAM, CPU, disk space... Not a solution neither.
- Install a Hobbit client on each machine, switch the hobbit client configuration to the server (not LOCAL mode) -> problem : I still have install 2 clients each following the application cluster group. The hobbit client use the first IP of the machine IP0A on machine MA (eth0). Can I force to use the virtual IP (IP0A:1 = eth0:1) ? 4)..
Any ideas is welcome! Thanks by advance!
On Tuesday 02 February 2010, L.M.J wrote:
Hi,
I have a passive/active cluster on 2 Linux servers : For AIX, I created a script that runs on each cluster node. Each node has a normal hobbit client running. I also have an entry in bb-hosts for each clustered application.
I have a config file where I link the ip address to the Xymon name I used in bb- hosts (or you can use an other method in stead of the ip address to figure out wich applications are running a cluster node).
Each cluster nodes runs a external script that uses the previous config file to figure out wich applications are running. For each application, it generates a file with the same syntax as the normal hobbit client does and it sends the content with the correct hobbit client name to the xymon server. You can see the syntax if you click on 'Client data available' on a disk check.
So basically, I have a hobbit client running per application on the cluster that behaves exactly like a normal client. I even configured external tests per application in the config file.
Hope this helps.
Stef
On Tue, 2 Feb 2010 10:05:32 +0100, Stef Coene <stef.coene at docum.org> wrote:
On Tuesday 02 February 2010, L.M.J wrote:
Hi,
I have a passive/active cluster on 2 Linux servers : For AIX, I created a script that runs on each cluster node. Each node has a normal hobbit client running. I also have an entry in bb-hosts for each clustered application.
I have a config file where I link the ip address to the Xymon name I used in bb-hosts (or you can use an other method in stead of the ip address to figure out wich applications are running a cluster node).
Each cluster nodes runs a external script that uses the previous config file to figure out wich applications are running. For each application, it generates a file with the same syntax as the normal hobbit client does and it sends the content with the correct hobbit client name to the xymon server. You can see the syntax if you click on 'Client data available' on a disk check.
So basically, I have a hobbit client running per application on the cluster that behaves exactly like a normal client. I even configured external tests per application in the config file.
Thanks Stef for the info but it's not that clear for. Is it possible to share with some sample config and script file please? In fact, you create a script in "client/ext/" that's generate the right "client/etc/localclient.cfg" ?
participants (2)
-
linuxmasterjedi@free.fr
-
stef.coene@docum.org