[Pacemaker] Best way to specify colocation and ordering

[Chris Picton]

>> On Fri, Aug 13, 2010 at 01:44:28PM +0000, Chris Picton wrote:
>> I have a drbd backed mysql server which has the following resources:
>> 
>> drbd0 -> lvm_data -> mount_data
>> drbd1 -> lvm_logs -> mount_logs
>> mysqld
>> floatingip
>> 
>> I would like the drbd based filesystems to start up in parallel.  Once
>> they have started, start mysql and the ip address.  Obviously the
>> reverse should happen on standby or shutdown.
>> 
>> all_drbd inf: ms_drbd1:Master ms_drbd0:Master
> 
> You want both on one node?

Yes - I need both masters on the same node - they are different 
filesystems on different physical disks, one for the database binlogs, 
one for the database data
 

>> if I do a crm resource
>> migrate MySQL, the active node shuts down the MySQL resource, but never
>> releases the drbd masters or dependant resources
> 
> The only thing that happened is mysql stop? Did you check logs for
> errors?

Yes MySQL stopped, and the dependant MailAlert and FloatingIP.  The logs 
indicate there was no attempt to stop the other resources at all - 

> 
>>  Master/Slave Set: ms_drbd0
>>      Masters: [ chris-test-02.ecntelecoms.za.net ] Slaves: [
>>      chris-test-01.ecntelecoms.za.net ]
>>  Master/Slave Set: ms_drbd1
>>      Masters: [ chris-test-02.ecntelecoms.za.net ] Slaves: [
>>      chris-test-01.ecntelecoms.za.net ]
>>  Resource Group: group_mysql
>>      MySQL      (ocf::ecn:MySQL.ocf):   Stopped MailAlert 
>>      (ocf::heartbeat:ECNAlert):	Stopped
>>  FloatingIP     (ocf::heartbeat:IPaddr2):       Stopped Resource Group:
>>  group_data
>>      lvm_data   (ocf::heartbeat:LVM):   Started chris-
>> test-02.ecntelecoms.za.net
>>      mount_data (ocf::heartbeat:Filesystem):    Started chris-
>> test-02.ecntelecoms.za.net
>>  Resource Group: group_logs
>>      lvm_logs   (ocf::heartbeat:LVM):   Started chris-
>> test-02.ecntelecoms.za.net
>>      mount_logs (ocf::heartbeat:Filesystem):    Started chris-
>> test-02.ecntelecoms.za.net
>>  Clone Set: STONITH-clone
>>      Started: [ chris-test-02.ecntelecoms.za.net chris-
>> test-01.ecntelecoms.za.net ]
>> 
>> 
>> Is there a better/more concise/more correct way of specifying the
>> colocations and orderings?
> 
> The constraints look OK.
> 
> Something like this should also work (example for the logs db):
> 
> order order_logs inf: ms_drbd1:promote group_logs group_mysql FloatingIP
> collocation col_logs inf: ms_drbd1:Master group_logs group_mysql
> FloatingIP

I have tried that, and resource start up in the correct order and 
together on the same node.

However, if I standby the 'master' node, I see the following in the logs 
a few seconds apart.  

info: rsc:FloatingIP:150: stop
info: rsc:drbd0:0:151: demote
info: rsc:drbd1:0:152: demote
info: rsc:MailAlert:153: stop
info: rsc:MySQL:154: stop

Ie, it is tring to demote the drbd resources before attempting to stop 
mysql, which can;t be done, as the drbd devices are still mounted.

Should I need to also specify the demote ordering, as well as the Slave 
colocations (to say MySQL cannot run where drbd is Slave?)

I had added:

colocation col_data inf: ms_drbd0:Master group_data group_mysql FloatingIP

colocation col_logs inf: ms_drbd1:Master group_logs group_mysql FloatingIP

order order_data inf: ms_drbd0:promote group_data group_mysql FloatingIP

order order_logs inf: ms_drbd1:promote group_logs group_mysql FloatingIP