UNIX for Advanced & Expert Users

Hi,

does anyone know if a "standard" service exists to synchronize boot on a multi-server system ?

I'm managing a system with four servers: a database server, an application server and two web servers. What I need is that when the system turns on (for example for a cold start due to electrical issues), the application server waits for the database server to be up and running and the web servers wait for the application server to be up and running.

Right now I'm using a timed "turn-on" with a switched PDU but since boot time is not consistent over time this method is not optimal.

Any hint ?

Thank you (and happy new year) !!

You could perhaps start all 4 servers at the same time but not start the applications automatically or perhaps only the database server. The script that starts the database could have a ssh command to the application server to start the application start script. This script in turn could start the two webserver applications.

Alternatively you could start all servers at the same time and put a wait loop at the beginning of the application start script on the application server that keeps checking if the database is availably. Likewise the webserver start scripts could check if the application is available.

The second option is probably the most trouble-free.

^^^
Thanks, the second option was what I was thinking too.

But do you know if a service to do this kind of things already exist or should I use custom shell scripts to wait and check for the services on the other servers ?
I always prefer to use "standards" methods if they are available.

Thank you !!

I would take each service out of the rc and then have a cron ping each server, if each ping returns a result then have it run another perl script ssh and exec each init script. Make sense?

Alternatively you could use telnet or e.g. nmap to test whether a service is listening at the intended port of the remote server, so that the script knows that the application is running on that server and can then break the wait loop (perhaps after an additional timeout if need be).