[Pacemaker] Re: Understanding rules: location, colocation, order. Using with Master/Slave
Andrew Beekhof
beekhof at gmail.com
Tue Oct 28 15:27:55 EDT 2008
On Tue, Oct 28, 2008 at 18:22, Adrian Chapela
<achapela.rexistros at gmail.com> wrote:
> Serge Dubrouski escribió:
>>
>> On Tue, Oct 28, 2008 at 6:29 AM, Adrian Chapela
>> <achapela.rexistros at gmail.com> wrote:
>>
>>>
>>> Serge Dubrouski escribió:
>>>
>>>>
>>>> On Mon, Oct 27, 2008 at 12:43 PM, Adrian Chapela
>>>> <achapela.rexistros at gmail.com> wrote:
>>>>
>>>>
>>>>>
>>>>> Serge Dubrouski escribió:
>>>>>
>>>>>
>>>>>>
>>>>>> Something like this:
>>>>>>
>>>>>> <rsc_order id="drbd0_before_myGroup" first="ms-drbd0"
>>>>>> then="myGroup" then-action="start" first-action="promote"/>
>>>>>> <rsc_colocation id="myGroup_on_drbd0" rsc="myGroup"
>>>>>> with-rsc="ms-drbd0" with-rsc-role="Master" score="INFINITY"/>
>>>>>> <rsc_location id="primNode" rsc="myGroup">
>>>>>> <rule id="prefered_primNode" score="1000">
>>>>>> <expression attribute="#uname" id="expression.id2242728"
>>>>>> operation="eq" value="fc-node1"/>
>>>>>> </rule>
>>>>>> </rsc_location>
>>>>>>
>>>>>> See that cib.xml that I sent you a couple of days ago. First rule will
>>>>>> promote DRBD before starting a group, second will collocate master and
>>>>>> a group, third one will place group and master to the desired node.
>>>>>>
>>>>>>
>>>>>>
>>>>>
>>>>> Yes, I bases my rules on yours, but I can't update my config with your
>>>>> rules
>>>>> directly.
>>>>> <rsc_order id="drbd0_before_myGroup" first="ms-drbd0" then="mail_Group"
>>>>> then-action="start" first-action="promote"/>
>>>>>
>>>>> This could be like:
>>>>> <rsc_order id="drbd0_before_myGroup" from="mail_Group" action="start"
>>>>> to="ms-drbd0" to_action="promote"/>
>>>>>
>>>>> What is the version of your heartbeat ? My 2.99.1 heartbeat didn't
>>>>>
>>>>>
>>>>
>>>> It's processed by pacemaker, not heartbeat. That rule worked all right
>>>> under 0.6, 0.7, 1.0:
>>>>
>>>> Refresh in 3s...
>>>>
>>>> ============
>>>> Last updated: Mon Oct 27 14:52:47 2008
>>>> Current DC: fc-node2 (ad6f19b7-228a-48b7-bae0-f95a838bde2a)
>>>> 2 Nodes configured.
>>>> 3 Resources configured.
>>>> ============
>>>>
>>>> Node: fc-node1 (b88f98c6-50f2-463a-a6eb-51abbec645a9): online
>>>> Node: fc-node2 (ad6f19b7-228a-48b7-bae0-f95a838bde2a): online
>>>>
>>>> Full list of resources:
>>>>
>>>> Clone Set: DoFencing
>>>> child_DoFencing:0 (stonith:external/xen0): Started fc-node1
>>>> child_DoFencing:1 (stonith:external/xen0): Started fc-node2
>>>> Master/Slave Set: ms-drbd0
>>>> drbd0:0 (ocf::heartbeat:drbd): Master fc-node1
>>>> drbd0:1 (ocf::heartbeat:drbd): Started fc-node2
>>>> Resource Group: myGroup
>>>> myIP (ocf::heartbeat:IPaddr): Started fc-node1
>>>> fs0 (ocf::heartbeat:Filesystem): Started fc-node1
>>>> myPgsql (ocf::heartbeat:pgsql): Started fc-node1
>>>>
>>>> [root at fc-node1 crm]# rpm -qa | grep pacemaker
>>>> libpacemaker3-1.0.0-2.1
>>>> pacemaker-1.0.0-2.1
>>>> [root at fc-node1 crm]#
>>>>
>>>>
>>>> What error do you get?
>>>>
>>>>
>>>
>>> I can update the configuration now. I have Heartbeat 2.99.1 + Pacemaker
>>> 1.0.
>>> Bur now, I can't put Master on the node. Have you check fc-node2 as a
>>> master
>>> node?
>>>
>>> How are you using pingd ? As a clone instance or configured in ha.cf ?
>>>
>>
>> pingd is broken in 1.0 :-( Andrew fixed it in the latest dev release.
>>
>
> OK, but Adrew said me this morning that bug was fixed in last stable code. I
> have downloaded Pacemaker-1-0-79d2ba7e502f but pingd seems broken as well.
I'm pretty sure it works - but its a good idea to log a bug (and add a
hb_report tarball) anyway
> If I delete the pingd rule Group is starting but if I don't delete the rule
> the Group is not starting.
>
> Another thing is the master selection, it is wrong in my case, using your
> config Sergei. Could you try changing the master node from fc-node1 and
> fc-node2 ?
>
> Thank you!
>
>>>>
>>>>>
>>>>> understand the rule.
>>>>>
>>>>>
>>>>>>
>>>>>> On Mon, Oct 27, 2008 at 12:24 PM, Adrian Chapela
>>>>>> <achapela.rexistros at gmail.com> wrote:
>>>>>>
>>>>>>
>>>>>>
>>>>>>>
>>>>>>> Serge Dubrouski escribió:
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>>>
>>>>>>>> On Mon, Oct 27, 2008 at 12:07 PM, Adrian Chapela
>>>>>>>> <achapela.rexistros at gmail.com> wrote:
>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>>>>
>>>>>>>>>> Hello!
>>>>>>>>>>
>>>>>>>>>> I am working on a cluster with a two Master/Slave instances.
>>>>>>>>>>
>>>>>>>>>> I have: 2 drbd Master/Slave instance, 1 pingd clone instance, 1
>>>>>>>>>> group
>>>>>>>>>> with
>>>>>>>>>> Filesystem resource.
>>>>>>>>>>
>>>>>>>>>> ms-drbd0 is the first Master/Slave
>>>>>>>>>> ms-drbd1 is the second Master/Slave
>>>>>>>>>> mail_Group is the first group, it depends on ms-drbd0
>>>>>>>>>> samba_Group is the second group, it depends on ms-drbd1
>>>>>>>>>>
>>>>>>>>>> I have the next rules:
>>>>>>>>>>
>>>>>>>>>> <rsc_order id="mail-drbd0_before_fs0" from="Montaxe_mail"
>>>>>>>>>> action="start"
>>>>>>>>>> to="ms-drbd0" to_action="promote"/>
>>>>>>>>>> <rsc_order id="samba-drbd1_before_fs0" from="Montaxe_samba"
>>>>>>>>>> action="start"
>>>>>>>>>> to="ms-drbd1" to_action="promote"/>
>>>>>>>>>> (starts Montaxe_mail when ms-drbd0 has been promoted, start
>>>>>>>>>> Montaxe_samba
>>>>>>>>>> when ms-drbd1 has been promoted. These rules are ok, I think)
>>>>>>>>>>
>>>>>>>>>> <rsc_colocation id="mail_Group_on_ms-drbd0" to="ms-drbd0"
>>>>>>>>>> to_role="master"
>>>>>>>>>> from="mail_Group" score="INFINITY"/>
>>>>>>>>>> <rsc_colocation id="samba_Group_on_ms-drbd1" to="ms-drbd1"
>>>>>>>>>> to_role="master" from="samba_Group" score="INFINITY"/>
>>>>>>>>>> (Run mail_Group only on the master node, run samba_Group on the
>>>>>>>>>> master
>>>>>>>>>> node)
>>>>>>>>>>
>>>>>>>>>> <rsc_location id="mail:drbd" rsc="ms-drbd0">
>>>>>>>>>> <rule id="rule:ms-drbd0" role="master" score="100">
>>>>>>>>>> <expression attribute="#uname" operation="eq"
>>>>>>>>>> value="debianquagga2"/>
>>>>>>>>>> </rule>
>>>>>>>>>> <rule id="mail_Group:pingd:rule" score="-INFINITY"
>>>>>>>>>> boolean_op="or">
>>>>>>>>>> <expression id="mail_Group:pingd:expr:undefined"
>>>>>>>>>> attribute="pingd"
>>>>>>>>>> operation="not_defined"/>
>>>>>>>>>> <expression id="mail_Group:pingd:expr:zero" attribute="pingd"
>>>>>>>>>> operation="lte" value="0"/>
>>>>>>>>>> </rule>
>>>>>>>>>> </rsc_location>
>>>>>>>>>> <rsc_location id="samba:drbd" rsc="ms-drbd1">
>>>>>>>>>> <rule id="rule:ms-drbd1" role="master" score="100">
>>>>>>>>>> <expression attribute="#uname" operation="eq"
>>>>>>>>>> value="debianquagga2"/>
>>>>>>>>>> </rule>
>>>>>>>>>> <rule id="samba_Group:pingd:rule" score="-INFINITY"
>>>>>>>>>> boolean_op="or">
>>>>>>>>>> <expression id="samba_Group:pingd:expr:undefined"
>>>>>>>>>> attribute="pingd"
>>>>>>>>>> operation="not_defined"/>
>>>>>>>>>> <expression id="samba_Group:pingd:expr:zero" attribute="pingd"
>>>>>>>>>> operation="lte" value="0"/>
>>>>>>>>>> </rule>
>>>>>>>>>> </rsc_location>
>>>>>>>>>> (Select debianquagga2 as Master and if the node lost its
>>>>>>>>>> connection
>>>>>>>>>> take
>>>>>>>>>> the score -INFINITY to do failover, it applies to ms-drbd0 and
>>>>>>>>>> ms-drbd1)
>>>>>>>>>>
>>>>>>>>>> With this rules all is working very well but the node selected as
>>>>>>>>>> master
>>>>>>>>>> isn't "debianquagga2", Why could be the reason ?
>>>>>>>>>>
>>>>>>>>>> I using Heartbeat 2.1.4
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>
>>>>>>>>> I have attached the cib xml file. If I delete two groups, Master is
>>>>>>>>> debianQuagga2, If not, Master is debianQuagga1.
>>>>>>>>>
>>>>>>>>>
>>>>>>>>>
>>>>>>>>>
>>>>>>>>
>>>>>>>> That probably has something to do with how scores are counted for
>>>>>>>> groups. In your rsc_location rule for masters you have a really low
>>>>>>>> score for assigning master role to debianquagga2. It's possible that
>>>>>>>> groups outscore them with default values. I'm not sure in that,
>>>>>>>> that's
>>>>>>>> just mu guess. You probably can check this with show score scripts.
>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>
>>>>>>> I will check that, thank you!
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>>>
>>>>>>>> I'd try to assign rsc_location rules to groups, not to master role.
>>>>>>>> Your collocation rule will control that groups are on the same nodes
>>>>>>>> with the masters. Or you can try to increase your scores from 100 to
>>>>>>>> something higher.
>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>
>>>>>>> Ok, but I need that node be a master before than start group because
>>>>>>> it
>>>>>>> depends on the master / slave resource, is that possible changing
>>>>>>> collocation ? Could you open my eyes with a simple example ?
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>>>>
>>>>>>>>> _______________________________________________
>>>>>>>>> Pacemaker mailing list
>>>>>>>>> Pacemaker at clusterlabs.org
>>>>>>>>> http://list.clusterlabs.org/mailman/listinfo/pacemaker
>>>>>>>>>
>>>>>>>>>
>>>>>>>>>
>>>>>>>>>
>>>>>>>>>
>>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>
>>>>>>> _______________________________________________
>>>>>>> Pacemaker mailing list
>>>>>>> Pacemaker at clusterlabs.org
>>>>>>> http://list.clusterlabs.org/mailman/listinfo/pacemaker
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>
>>>>>>
>>>>>
>>>>> _______________________________________________
>>>>> Pacemaker mailing list
>>>>> Pacemaker at clusterlabs.org
>>>>> http://list.clusterlabs.org/mailman/listinfo/pacemaker
>>>>>
>>>>>
>>>>>
>>>>
>>>>
>>>>
>>>
>>> _______________________________________________
>>> Pacemaker mailing list
>>> Pacemaker at clusterlabs.org
>>> http://list.clusterlabs.org/mailman/listinfo/pacemaker
>>>
>>>
>>
>>
>>
>>
>
>
> _______________________________________________
> Pacemaker mailing list
> Pacemaker at clusterlabs.org
> http://list.clusterlabs.org/mailman/listinfo/pacemaker
>
More information about the Pacemaker
mailing list