One area of cluster confusion involves the cron (8) daemon. The crontab files for each member (found in /var/spool/cron/crontabs) are member-specific – /var/spool/cron is a CDSL.
# fln /var/spool/cron /var/spool/cron -> ../cluster/members/{memb}/spool/cron
Not only are the crontab files member-specific, but the cron daemon is also not cluster-aware.
# clu_ps | grep -E "^member|^=|^-|cron" | grep -v grep ================================================================================ member1 (molari) ps -ef -------------------------------------------------------------------------------- root 525300 524289 0.0 13:47:26 ?? 0:00.02 /usr/sbin/cron ================================================================================ member2 (sheridan) ps -ef -------------------------------------------------------------------------------- root 1049553 1048577 0.0 13:47:25 ?? 0:00.01 /usr/sbin/cron ================================================================================
This means that on each member, cron acts independently from all the other members, and that a cron job to backup a data directory would run on the same file systems at the same times on all members in the cluster. Obviously not a good arrangement!
This problem is solved in one of the Best Practices documents that can be found on HP's Best Practices website:
http://www.tru64unix.compaq.com/docs/best_practices/clus_bps.html
This Best Practices document shows you how to set up a CAA service to handle cron jobs that need to be addressed in a cluster-aware fashion.
For more on setting up CAA, refer to Chapter 23 and Chapter 24.