On Tue, Jun 17, 2008 at 10:27, Junko IKEDA <[EMAIL PROTECTED]> wrote: >> When the lrmd process falls, lrmd reboots. >> But, the monitor stops after having rebooted. >> >> In this status, lrmd cannot detect the obstacle of the resource after it. >> >> Actually, there may be little possibility that lrmd reboots. >> But, I think that it is necessary when I think about the worst case. >> >> When it is necessary for lrmd to restart, I hope that a system reboots. >> Like ccm and crmd .... >> (Possibly a similar thing may apply to a stonithd process.) >> >> Will it be difficult to hope for this improvement in Heartbeat? >> >> Best Regards, >> Hideo Yamauchi. > > Hi, > > on Yamauchi-san's test environment, lrmd would restart if it's killed (# > kill -9 PID). > for mine, lrmd is sure to restart, but crmd catches some critical error and > end up to system reboot due to the suicide code. > see attached node-a/ha-log: around line 171 > > * kill -9 lrmd process > heartbeat[28075]: 2008/06/17_16:48:51 WARN: Managed > /usr/lib64/heartbeat/lrmd -r process 28086 killed by signal 9 [SIGKILL - > Kill, unblockable]. > > * crmd lose the connection with lrm > crmd[28089]: 2008/06/17_16:48:51 CRIT: lrm_dispatch: LRM Connection failed > > If you remember something, please let me know. > The version of Heartbeat, OS, glib are the same. > Can anything make an impact?
I can only think of two things - system load (or CPU power, both of which would affect the timing) and whether you both have stonith enabled. Actually a third thing is the use of "crm yes" or "crm respawn" would also affect this. Are any of these things different between the two clusters? > Which is the appropriate behavior for heartbeat when lrmd is killed; > restart lrmd or system reboot? Partly it depends on your preference. My personal belief is that when STONITH is enabled, rebooting the node when any of the v2 processes exit offers no additional protection to beyond that provided by STONITH and makes the recovery time longer. Others have been known to disagree :-) _______________________________________________ Linux-HA mailing list [email protected] http://lists.linux-ha.org/mailman/listinfo/linux-ha See also: http://linux-ha.org/ReportingProblems
