HACMP, DMS, timeout..

[ogniemi]

Hello,

On test cluster, to increase the time before the cluster goes into reconfig too long I've change the "clstgmgr" sysbsystem start switch on both cluster nodes as follows:

chssys -s clstrmgr -a "-u 600000"

(AIX 5.2 ML02, HACMP 4.5.0.10)


The problem is that having set this value (10 minutes) when I start HACMP it halts a node. (the status is like after "halt -q&quot

Did anyone met such problem?


The "smitty" window freezes with status:

COMMAND STATUS

Command: running stdout: yes stderr: no

Before command completion, additional instructions may appear below.

Dec 1 2003 15:28:50Starting execution of /usr/sbin/cluster/etc/rc.cluster
with parameters : -boot -N -b -i
0513-029 The portmap Subsystem is already active.
Multiple instances are not supported.
0513-029 The inetd Subsystem is already active.
Multiple instances are not supported.
Loaded kernel extension kmid = 35243456
dms init
Dec 1 2003 15:29:00Checking for srcmstr active...Dec 1 2003 15:29:00complete.
6102 - 0:00 syslogd
Setting routerevalidate to 1
Dec 1 2003 15:29:00
/usr/sbin/cluster/utilities/clstart : called with flags -sm

0513-059 The clstrmgr Subsystem has been started. Subsystem PID is 15390.


Broadcast message from root@clnode1 (tty) at 15:29:03 ...

Starting Cluster Manager (clstrmgr) subsystem on clnode1


0513-059 The snmpd Subsystem has been started. Subsystem PID is 16550.

 

[gileb]

hi,
Do you have a dump ? how do you know it is a DMS ?
As far as i know, getting in reconfiguration too long is not a problem, if cluster is only long to start, it gets correct at the end and returns to normal...

 

[ogniemi]

The change was the only I did on both cluster nodes. Before, when the timeout was set to 360secs cluster was up and running.

After the crash, I restarted the LPAR and removed "-u 600000" switch from clstrmgr subsystem and cluster works again - starts without any problem.

So I had repeated the operation with "chssys -s clstrmgr -a "-u 600000"" on both nodes and crash happened again. It means that claster cannot start with such value of timeout, or maybe it is not supported with version 4.5.

BTW:
I cannot found it documented on any IBM's site, docu - no HACMP redbook for version 4.5 mentions about changing the time before the cluster goes into reconfig too long )


regards,m.

 

[gileb]

i am not sure this parameter is taken in 4.5...
You should instead modify the "smitty cm_time_before_warning"

hope that helps