Mailing List Archive: Linux-HA: Users

(Bug?) regarding resource_stickiness, master_slave and master-colocated groups

Index | Next | Previous | View Threaded

dk at in-telegence

Mar 13, 2008, 8:05 AM

Post #1 of 6 (677 views)
Permalink

(Bug?) regarding resource_stickiness, master_slave and master-colocated groups

Hi

during the writeup of ScoreCalculation on the wiki, I noticed something
strange. It'd be nice to know whether this is on purpose or a bug.

Test setup is:

2 nodes, 1 drbd device, a group of 3 resource which are to run on top of
the drbd master.

resource_stickiness is set to 100

If I use a colocation constraint with a score of infinity, the master
receives a stickiness bonus of 600.

If I change the colocation score to a numeric value (I tested 1000 and
5000), the bonus is reduced to 400.

I could explain the 600 as 2 * num_resources * stickiness, but I cannot
see where those 400 come from.

Is this a bug or (why?) is this intended?

Regards
Dominik
_______________________________________________
Linux-HA mailing list
Linux-HA [at] lists
http://lists.linux-ha.org/mailman/listinfo/linux-ha
See also: http://linux-ha.org/ReportingProblems

achapela.rexistros at gmail

Mar 13, 2008, 8:29 AM

Post #2 of 6 (601 views)
Permalink

Re: (Bug?) regarding resource_stickiness, master_slave and master-colocated groups [In reply to]

Dominik Klein escribió:
> Hi
>
> during the writeup of ScoreCalculation on the wiki, I noticed
> something strange. It'd be nice to know whether this is on purpose or
> a bug.
>
> Test setup is:
>
> 2 nodes, 1 drbd device, a group of 3 resource which are to run on top
> of the drbd master.
>
> resource_stickiness is set to 100
>
> If I use a colocation constraint with a score of infinity, the master
> receives a stickiness bonus of 600.
>
> If I change the colocation score to a numeric value (I tested 1000 and
> 5000), the bonus is reduced to 400.
>
> I could explain the 600 as 2 * num_resources * stickiness, but I
> cannot see where those 400 come from.
>
> Is this a bug or (why?) is this intended?
I think there is a BUG related with master_slave resources. I have
opened this bug:
http://developerbugs.linux-foundation.org/show_bug.cgi?id=1852 and today
I have no response... :(

Is this which are you talking about ?
>
> Regards
> Dominik
> _______________________________________________
> Linux-HA mailing list
> Linux-HA [at] lists
> http://lists.linux-ha.org/mailman/listinfo/linux-ha
> See also: http://linux-ha.org/ReportingProblems
>

_______________________________________________
Linux-HA mailing list
Linux-HA [at] lists
http://lists.linux-ha.org/mailman/listinfo/linux-ha
See also: http://linux-ha.org/ReportingProblems

dejanmm at fastmail

Mar 13, 2008, 10:24 AM

Post #3 of 6 (599 views)
Permalink

Re: (Bug?) regarding resource_stickiness, master_slave and master-colocated groups [In reply to]

Hi,

On Thu, Mar 13, 2008 at 04:29:24PM +0100, Adrian Chapela wrote:
> Dominik Klein escribi?:
>> Hi
>>
>> during the writeup of ScoreCalculation on the wiki, I noticed something
>> strange. It'd be nice to know whether this is on purpose or a bug.
>>
>> Test setup is:
>>
>> 2 nodes, 1 drbd device, a group of 3 resource which are to run on top of
>> the drbd master.
>>
>> resource_stickiness is set to 100
>>
>> If I use a colocation constraint with a score of infinity, the master
>> receives a stickiness bonus of 600.
>>
>> If I change the colocation score to a numeric value (I tested 1000 and
>> 5000), the bonus is reduced to 400.
>>
>> I could explain the 600 as 2 * num_resources * stickiness, but I cannot
>> see where those 400 come from.
>>
>> Is this a bug or (why?) is this intended?
> I think there is a BUG related with master_slave resources. I have opened
> this bug: http://developerbugs.linux-foundation.org/show_bug.cgi?id=1852
> and today I have no response... :(
>
> Is this which are you talking about ?

I think that it's unrelated.

There has been no answer yet to your bug report, because our CRM
developer's not around right now, but he should soon be back.

Thanks,

Dejan

>> Regards
>> Dominik
>> _______________________________________________
>> Linux-HA mailing list
>> Linux-HA [at] lists
>> http://lists.linux-ha.org/mailman/listinfo/linux-ha
>> See also: http://linux-ha.org/ReportingProblems
>>
>
> _______________________________________________
> Linux-HA mailing list
> Linux-HA [at] lists
> http://lists.linux-ha.org/mailman/listinfo/linux-ha
> See also: http://linux-ha.org/ReportingProblems
_______________________________________________
Linux-HA mailing list
Linux-HA [at] lists
http://lists.linux-ha.org/mailman/listinfo/linux-ha
See also: http://linux-ha.org/ReportingProblems

dk at in-telegence

Mar 14, 2008, 12:54 AM

Post #4 of 6 (600 views)
Permalink

Re: (Bug?) regarding resource_stickiness, master_slave and master-colocated groups [In reply to]

Adrian Chapela wrote:
> Dominik Klein escribió:
>> Hi
>>
>> during the writeup of ScoreCalculation on the wiki, I noticed
>> something strange. It'd be nice to know whether this is on purpose or
>> a bug.
>>
>> Test setup is:
>>
>> 2 nodes, 1 drbd device, a group of 3 resource which are to run on top
>> of the drbd master.
>>
>> resource_stickiness is set to 100
>>
>> If I use a colocation constraint with a score of infinity, the master
>> receives a stickiness bonus of 600.
>>
>> If I change the colocation score to a numeric value (I tested 1000 and
>> 5000), the bonus is reduced to 400.
>>
>> I could explain the 600 as 2 * num_resources * stickiness, but I
>> cannot see where those 400 come from.
>>
>> Is this a bug or (why?) is this intended?
> I think there is a BUG related with master_slave resources. I have
> opened this bug:
> http://developerbugs.linux-foundation.org/show_bug.cgi?id=1852 and today
> I have no response... :(
>
> Is this which are you talking about ?

No, but I experienced that as well. I don't know why it happens, but I
think you can get around it.

Please try this:


<rsc_location id="rscloc-ms-drbd1:0" rsc="drbd1:0">
<rule id="rscloc-ms-drbd1:0-rule1" score="500">
<expression id="rscloc-ms-drbd1:0-rule1-expr"
attribute="#uname" operation="eq" value="node1"/>
</rule>
<rule id="rscloc-ms-drbd1:0-rule2" score="-500">
<expression id="rscloc-ms-drbd1:0-rule2-expr"
attribute="#uname" operation="eq" value="node2"/>
</rule>
</rsc_location>

<rsc_location id="rscloc-ms-drbd1:1" rsc="drbd1:1">
<rule id="rscloc-ms-drbd1:1-rule1" score="500">
<expression id="rscloc-ms-drbd1:1-rule1-expr"
attribute="#uname" operation="eq" value="node2"/>
</rule>
<rule id="rscloc-ms-drbd1:1-rule2" score="-500">
<expression id="rscloc-ms-drbd1:1-rule2-expr"
attribute="#uname" operation="eq" value="node1"/>
</rule>
</rsc_location>

Solved that problem for me.

Regards
Dominik
_______________________________________________
Linux-HA mailing list
Linux-HA [at] lists
http://lists.linux-ha.org/mailman/listinfo/linux-ha
See also: http://linux-ha.org/ReportingProblems

dk at in-telegence

Mar 14, 2008, 7:54 AM

Post #5 of 6 (609 views)
Permalink

Re: (Bug?) regarding resource_stickiness, master_slave and master-colocated groups [In reply to]

> Solved that problem for me.

At least with a colocated resource I have to add.

Regards
Dominik
_______________________________________________
Linux-HA mailing list
Linux-HA [at] lists
http://lists.linux-ha.org/mailman/listinfo/linux-ha
See also: http://linux-ha.org/ReportingProblems

dk at in-telegence

Mar 14, 2008, 8:00 AM

Post #6 of 6 (609 views)
Permalink

Re: (Bug?) regarding resource_stickiness, master_slave and master-colocated groups [In reply to]

Dominik Klein wrote:
>> Solved that problem for me.
>
> At least with a colocated resource I have to add.

Urghs. Friday afternoon ...

I just wanted to verify that and it turns out my method does not restart
the whole thing on a slave failure. Thats true.

But it does still restart the whole thing if I shutdown the slave node
and let it rejoin the cluster. In fact, what I see is that for a short
time, BOTH nodes become "standby". Right after that, both nodes are
shown as "online" and then the master_slave resource including all clone
instances and colocated resources are restarted.

Regards
Dominik
_______________________________________________
Linux-HA mailing list
Linux-HA [at] lists
http://lists.linux-ha.org/mailman/listinfo/linux-ha
See also: http://linux-ha.org/ReportingProblems

Index | Next | Previous | View Threaded

Interested in having your list archived? Contact Gossamer Threads