We combine these. We run as a service, but have a daily cron, with random
time spread among our hosts, to stop/start the service and clean up stale
.pid files. This is more of a hold over from our early days more than
anything, but it works, doesn't cause issues, and keeps the runs spread
out.
O
My take on it is to run it from our nagios server. What better way to
monitor the puppet runs than by executing that run as part of the check?
retry intervals also help push changes out much quicker if they could take
multiple runs etc.
We also run a single daily cron job.
Chris
On 8 October 201
