[Pacemaker] pacemaker service start failed.

Takatoshi MATSUO matsuo.tak at gmail.com
Fri Nov 9 04:24:25 UTC 2012


2012/11/9 Andrew Beekhof <andrew at beekhof.net>:
> On Fri, Nov 9, 2012 at 12:25 PM, Takatoshi MATSUO <matsuo.tak at gmail.com> wrote:
>> Hi Andrew
>>
>> Pgsql RA uses crm_attribute to  change slave's master-socre from master like
>>  # crm_attribute -l reboot  -N host2 -n "master-pgsql:1" -v "1000"
>> Because the Slave of PostgreSQL can't get replication status.
>>
>> In addition the RA uses "uname -n" to compare hostname to
>> $OCF_RESKEY_CRM_meta_notify_****_uname.
>> Mysql RA does the same way.
>>
>> But we can't use "crm_node --name" in Pacemaker 1.0.x.
>> How do we keep compatibility?
>> HOSTNAME=`crm_node --name 2>/dev/null || uname -n` ?  :(
>
> I'd do two things...
>
> First, add support for crm_node --name to the next 1.0 release (just
> returning uname(2))
> Second, add this to ocf-binaries...
>
> : ${UNAME_N:=uname -n}
>
> and replace all references to uname -n with $UNAME_N [1]
> That makes it easy for anyone that needs "name != uname" (or for
> distros that want to support it by default) to change.
>
> Make sense?

Yes.
Thank you.

Of course "crm_node --name" returns different value
after restarting Pacemaker 2.0  with different corosync.conf ?

I am concerned that we need to change value of
pgsql's node_list  parameter which lists node names
participated in replication.

>
> [1] Looks like this is the full list:
> # grep -l "uname -n" *
> ./Filesystem
> ./WinPopup
> ./EvmsSCC
> ./db2
> ./mysql
> ./AudibleAlarm
> ./pgsql
> ./MailTo

Regards,
Takatoshi MATUSO




More information about the Pacemaker mailing list