Mailing List Archive: Linux-HA: Pacemaker

Unnecessary shuffling of master/slave resources and maybe a split brain

Index | Next | Previous | View Threaded

achapela.rexistros at gmail

Nov 19, 2008, 10:59 AM

Post #1 of 13 (1513 views)
Permalink

Unnecessary shuffling of master/slave resources and maybe a split brain

Hello,

after many tests I found some problems in my config. For that I have
attached my hb_reports for two cases.

The first situation is a problem with connectivity on node2, the master
node. This node has unplugged the ethernet cable, and for a minutes it
has no connectivity at all.
Node1 knew problem and it took the control very well but my problem is
the node2 is considering itself as a master node, this could be
considered as a split brain ? When technicians plugged again the wire
two nodes had comunication again. Node2 after a "negotiation" became the
master node again.

The second situation is similar. Someone unplug the power cable of
node2. Node1 took control again and it became the master node. Then
someone plug the power cable again and node2 started up. When Heartbeat
is started, after a while starting and stoping resources, node2 became
the master node again. But in this situation node1 remained in a death
situation.

Could you have a look ? I think this bug I found few moths ago could be
the reason: http://developerbugs.linux-foundation.org/show_bug.cgi?id=1852

Thank you!

Attachments:

no_connectivity.tar.gz (182 KB)

stickiness.tar.gz (199 KB)

dk at in-telegence

Nov 20, 2008, 12:18 AM

Post #2 of 13 (1455 views)
Permalink

Re: Unnecessary shuffling of master/slave resources and maybe a split brain [In reply to]

Only looked at scenario 1.

Adrian Chapela wrote:
> Hello,
>
> after many tests I found some problems in my config. For that I have
> attached my hb_reports for two cases.
>
> The first situation is a problem with connectivity on node2, the master
> node. This node has unplugged the

"the"? From node1's ha.cf in the no-connectivity report, you have only
one communication channel which is a configuration error as is and is
not supported in any case as far as i know.

> ethernet cable, and for a minutes it
> has no connectivity at all.

This is exactly the place where some sort of stonith should happen.

> Node1 knew problem and it took the control very well but my problem is
> the node2 is considering itself as a master node, this could be
> considered as a split brain ?

sure

> When technicians plugged again the wire
> two nodes had comunication again. Node2 after a "negotiation" became the
> master node again.

You now have more master instances than configured allowed. So
everything regarding this drbd resource is stopped and started to sort
things out.

Regards
Dominik

_______________________________________________
Pacemaker mailing list
Pacemaker [at] clusterlabs
http://list.clusterlabs.org/mailman/listinfo/pacemaker

achapela.rexistros at gmail

Nov 20, 2008, 12:44 AM

Post #3 of 13 (1457 views)
Permalink

Re: Unnecessary shuffling of master/slave resources and maybe a split brain [In reply to]

Dominik Klein escribió:
> Only looked at scenario 1.
>
> Adrian Chapela wrote:
>> Hello,
>>
>> after many tests I found some problems in my config. For that I have
>> attached my hb_reports for two cases.
>>
>> The first situation is a problem with connectivity on node2, the
>> master node. This node has unplugged the
>
> "the"? From node1's ha.cf in the no-connectivity report, you have only
> one communication channel which is a configuration error as is and is
> not supported in any case as far as i know.

What are you saying me ? I should have another communication channel as
a COM port... I could test this with a null modem cable.
>
>> ethernet cable, and for a minutes it has no connectivity at all.
>
> This is exactly the place where some sort of stonith should happen.
>
>> Node1 knew problem and it took the control very well but my problem
>> is the node2 is considering itself as a master node, this could be
>> considered as a split brain ?
>
> sure
>
>> When technicians plugged again the wire two nodes had comunication
>> again. Node2 after a "negotiation" became the master node again.
>
> You now have more master instances than configured allowed. So
> everything regarding this drbd resource is stopped and started to sort
> things out.
>
> Regards
> Dominik
>
> _______________________________________________
> Pacemaker mailing list
> Pacemaker [at] clusterlabs
> http://list.clusterlabs.org/mailman/listinfo/pacemaker
>

_______________________________________________
Pacemaker mailing list
Pacemaker [at] clusterlabs
http://list.clusterlabs.org/mailman/listinfo/pacemaker

dk at in-telegence

Nov 20, 2008, 1:02 AM

Post #4 of 13 (1460 views)
Permalink

Re: Unnecessary shuffling of master/slave resources and maybe a split brain [In reply to]

> What are you saying me ? I should have another communication channel as
> a COM port... I could test this with a null modem cable.

You need at least 2 connections between the nodes. It never was different.

It once was true that a null modem oder serial cable as the second
connection was enough - from my _personal_ experience, having 2 ethernet
connections is better (because faster and able to cope with the rather
high traffic of v2 (pacemaker)).

Regards
Dominik

_______________________________________________
Pacemaker mailing list
Pacemaker [at] clusterlabs
http://list.clusterlabs.org/mailman/listinfo/pacemaker

achapela.rexistros at gmail

Nov 20, 2008, 1:21 AM

Post #5 of 13 (1457 views)
Permalink

Re: Unnecessary shuffling of master/slave resources and maybe a split brain [In reply to]

achapela.rexistros at gmail

Nov 20, 2008, 3:03 AM

Post #6 of 13 (1452 views)
Permalink

Re: Unnecessary shuffling of master/slave resources and maybe a split brain [In reply to]

Dominik Klein escribió:
>> What are you saying me ? I should have another communication channel
>> as a COM port... I could test this with a null modem cable.
>
> You need at least 2 connections between the nodes. It never was
> different.
>
> It once was true that a null modem oder serial cable as the second
> connection was enough - from my _personal_ experience, having 2
> ethernet connections is better (because faster and able to cope with
> the rather high traffic of v2 (pacemaker)).
I tested two options. I think the best option is a second ethernet card.
I have configured this option but I saw that my cluster didn't do a
failover.
The two connections are in different networks (One in 192.168.18.X and
other in 172.16.1.0) but I don't know why pingd isn't going down all of
resources.

I have this connectivity constraint:
<rsc_location id="mail-connectivity" rsc="Mail-drbd">
<rule id="mail-pingd-prefer-rule" score="-INFINITY"
role="Master">
<expression id="mail-pingd-prefer"
attribute="pingd" operation="lt" value="1000"/>
</rule>
</rsc_location>

<rsc_location id="samba-connectivity" rsc="Samba-drbd">
<rule id="samba-pingd-exclude-rule" score="-INFINITY" >
<expression id="samba-pingd-exclude"
attribute="pingd" operation="lt" value="1000"/>
</rule>
</rsc_location>

And this colocation:
<rsc_colocation id="mail_drbrd_rule" rsc="Mail" with-rsc="Mail-drbd"
with-rsc-role="Master" score="INFINITY"/>
<rsc_colocation id="samba_drbrd_rule" rsc="Samba" with-rsc="Samba-drbd"
with-rsc-role="Master" score="INFINITY"/>

The node lost connectivity with 192.168.18.0 (I defined all host in
this network for pingd) and pingd value should be less than 1000.
>
> Regards
> Dominik
>
> _______________________________________________
> Pacemaker mailing list
> Pacemaker [at] clusterlabs
> http://list.clusterlabs.org/mailman/listinfo/pacemaker
>

_______________________________________________
Pacemaker mailing list
Pacemaker [at] clusterlabs
http://list.clusterlabs.org/mailman/listinfo/pacemaker

dk at in-telegence

Nov 20, 2008, 4:52 AM

Post #7 of 13 (1453 views)
Permalink

Re: Unnecessary shuffling of master/slave resources and maybe a split brain [In reply to]

Adrian Chapela wrote:
> Dominik Klein escribió:
>>> What are you saying me ? I should have another communication channel
>>> as a COM port... I could test this with a null modem cable.
>>
>> You need at least 2 connections between the nodes. It never was
>> different.
>>
>> It once was true that a null modem oder serial cable as the second
>> connection was enough - from my _personal_ experience, having 2
>> ethernet connections is better (because faster and able to cope with
>> the rather high traffic of v2 (pacemaker)).
> I tested two options. I think the best option is a second ethernet card.
> I have configured this option but I saw that my cluster didn't do a
> failover.
> The two connections are in different networks (One in 192.168.18.X and
> other in 172.16.1.0) but I don't know why pingd isn't going down all of
> resources.
>
> I have this connectivity constraint:
> <rsc_location id="mail-connectivity" rsc="Mail-drbd">
> <rule id="mail-pingd-prefer-rule" score="-INFINITY"
> role="Master">
> <expression id="mail-pingd-prefer"
> attribute="pingd" operation="lt" value="1000"/>
> </rule>
> </rsc_location>
>
>
>
> <rsc_location id="samba-connectivity" rsc="Samba-drbd">
> <rule id="samba-pingd-exclude-rule" score="-INFINITY" >
> <expression id="samba-pingd-exclude"
> attribute="pingd" operation="lt" value="1000"/>
> </rule>
> </rsc_location>
>
> And this colocation:
> <rsc_colocation id="mail_drbrd_rule" rsc="Mail" with-rsc="Mail-drbd"
> with-rsc-role="Master" score="INFINITY"/>
> <rsc_colocation id="samba_drbrd_rule" rsc="Samba" with-rsc="Samba-drbd"
> with-rsc-role="Master" score="INFINITY"/>
>
> The node lost connectivity with 192.168.18.0 (I defined all host in
> this network for pingd) and pingd value should be less than 1000.

Should be or is?

Could you post your xml including the status section in this situation?

Regards
Dominik

_______________________________________________
Pacemaker mailing list
Pacemaker [at] clusterlabs
http://list.clusterlabs.org/mailman/listinfo/pacemaker

achapela.rexistros at gmail

Nov 20, 2008, 8:24 AM

Post #8 of 13 (1461 views)
Permalink

Re: Unnecessary shuffling of master/slave resources and maybe a split brain [In reply to]

Dominik Klein escribió:
> Adrian Chapela wrote:
>> Dominik Klein escribió:
>>>> What are you saying me ? I should have another communication
>>>> channel as a COM port... I could test this with a null modem cable.
>>>
>>> You need at least 2 connections between the nodes. It never was
>>> different.
>>>
>>> It once was true that a null modem oder serial cable as the second
>>> connection was enough - from my _personal_ experience, having 2
>>> ethernet connections is better (because faster and able to cope with
>>> the rather high traffic of v2 (pacemaker)).
>> I tested two options. I think the best option is a second ethernet
>> card. I have configured this option but I saw that my cluster didn't
>> do a failover.
>> The two connections are in different networks (One in 192.168.18.X
>> and other in 172.16.1.0) but I don't know why pingd isn't going down
>> all of resources.
>>
>> I have this connectivity constraint:
>> <rsc_location id="mail-connectivity" rsc="Mail-drbd">
>> <rule id="mail-pingd-prefer-rule" score="-INFINITY"
>> role="Master">
>> <expression id="mail-pingd-prefer"
>> attribute="pingd" operation="lt" value="1000"/>
>> </rule>
>> </rsc_location>
>>
>>
>>
>> <rsc_location id="samba-connectivity" rsc="Samba-drbd">
>> <rule id="samba-pingd-exclude-rule" score="-INFINITY" >
>> <expression id="samba-pingd-exclude"
>> attribute="pingd" operation="lt" value="1000"/>
>> </rule>
>> </rsc_location>
>>
>> And this colocation:
>> <rsc_colocation id="mail_drbrd_rule" rsc="Mail" with-rsc="Mail-drbd"
>> with-rsc-role="Master" score="INFINITY"/>
>> <rsc_colocation id="samba_drbrd_rule" rsc="Samba"
>> with-rsc="Samba-drbd" with-rsc-role="Master" score="INFINITY"/>
>>
>> The node lost connectivity with 192.168.18.0 (I defined all host in
>> this network for pingd) and pingd value should be less than 1000.
>
> Should be or is?
It Should be changing value but it isn't changing it.
(To know I am using: watch -n1 "cibadmin -Ql | grep status-.*-pingd" )

Pingd is writing this to log;
pingd[3181]: 2008/11/20_17:17:42 info: ping_read: Retrying...
>
> Could you post your xml including the status section in this situation?
Yes, first I will compile the new version of Pacemaker because pingd was
broken in other versions and this could be the same.
>
> Regards
> Dominik
>
> _______________________________________________
> Pacemaker mailing list
> Pacemaker [at] clusterlabs
> http://list.clusterlabs.org/mailman/listinfo/pacemaker
>

_______________________________________________
Pacemaker mailing list
Pacemaker [at] clusterlabs
http://list.clusterlabs.org/mailman/listinfo/pacemaker

achapela.rexistros at gmail

Nov 20, 2008, 10:45 AM

Post #9 of 13 (1450 views)
Permalink

Re: Unnecessary shuffling of master/slave resources and maybe a split brain [In reply to]

Dominik Klein escribió:
> Adrian Chapela wrote:
>> Dominik Klein escribió:
>>>> What are you saying me ? I should have another communication
>>>> channel as a COM port... I could test this with a null modem cable.
>>>
>>> You need at least 2 connections between the nodes. It never was
>>> different.
>>>
>>> It once was true that a null modem oder serial cable as the second
>>> connection was enough - from my _personal_ experience, having 2
>>> ethernet connections is better (because faster and able to cope with
>>> the rather high traffic of v2 (pacemaker)).
>> I tested two options. I think the best option is a second ethernet
>> card. I have configured this option but I saw that my cluster didn't
>> do a failover.
>> The two connections are in different networks (One in 192.168.18.X
>> and other in 172.16.1.0) but I don't know why pingd isn't going down
>> all of resources.
>>
>> I have this connectivity constraint:
>> <rsc_location id="mail-connectivity" rsc="Mail-drbd">
>> <rule id="mail-pingd-prefer-rule" score="-INFINITY"
>> role="Master">
>> <expression id="mail-pingd-prefer"
>> attribute="pingd" operation="lt" value="1000"/>
>> </rule>
>> </rsc_location>
>>
>>
>>
>> <rsc_location id="samba-connectivity" rsc="Samba-drbd">
>> <rule id="samba-pingd-exclude-rule" score="-INFINITY" >
>> <expression id="samba-pingd-exclude"
>> attribute="pingd" operation="lt" value="1000"/>
>> </rule>
>> </rsc_location>
>>
>> And this colocation:
>> <rsc_colocation id="mail_drbrd_rule" rsc="Mail" with-rsc="Mail-drbd"
>> with-rsc-role="Master" score="INFINITY"/>
>> <rsc_colocation id="samba_drbrd_rule" rsc="Samba"
>> with-rsc="Samba-drbd" with-rsc-role="Master" score="INFINITY"/>
>>
>> The node lost connectivity with 192.168.18.0 (I defined all host in
>> this network for pingd) and pingd value should be less than 1000.
>
> Should be or is?
>
> Could you post your xml including the status section in this situation?
Here is my cib.xml
>
> Regards
> Dominik
>
> _______________________________________________
> Pacemaker mailing list
> Pacemaker [at] clusterlabs
> http://list.clusterlabs.org/mailman/listinfo/pacemaker
>

Attachments:

node2_cib.xml (58.5 KB)

achapela.rexistros at gmail

Nov 20, 2008, 10:49 AM

Post #10 of 13 (1454 views)
Permalink

Re: Unnecessary shuffling of master/slave resources and maybe a split brain [In reply to]

Dominik Klein escribió:
> Adrian Chapela wrote:
>> Dominik Klein escribió:
>>>> What are you saying me ? I should have another communication
>>>> channel as a COM port... I could test this with a null modem cable.
>>>
>>> You need at least 2 connections between the nodes. It never was
>>> different.
>>>
>>> It once was true that a null modem oder serial cable as the second
>>> connection was enough - from my _personal_ experience, having 2
>>> ethernet connections is better (because faster and able to cope with
>>> the rather high traffic of v2 (pacemaker)).
>> I tested two options. I think the best option is a second ethernet
>> card. I have configured this option but I saw that my cluster didn't
>> do a failover.
>> The two connections are in different networks (One in 192.168.18.X
>> and other in 172.16.1.0) but I don't know why pingd isn't going down
>> all of resources.
>>
>> I have this connectivity constraint:
>> <rsc_location id="mail-connectivity" rsc="Mail-drbd">
>> <rule id="mail-pingd-prefer-rule" score="-INFINITY"
>> role="Master">
>> <expression id="mail-pingd-prefer"
>> attribute="pingd" operation="lt" value="1000"/>
>> </rule>
>> </rsc_location>
>>
>>
>>
>> <rsc_location id="samba-connectivity" rsc="Samba-drbd">
>> <rule id="samba-pingd-exclude-rule" score="-INFINITY" >
>> <expression id="samba-pingd-exclude"
>> attribute="pingd" operation="lt" value="1000"/>
>> </rule>
>> </rsc_location>
>>
>> And this colocation:
>> <rsc_colocation id="mail_drbrd_rule" rsc="Mail" with-rsc="Mail-drbd"
>> with-rsc-role="Master" score="INFINITY"/>
>> <rsc_colocation id="samba_drbrd_rule" rsc="Samba"
>> with-rsc="Samba-drbd" with-rsc-role="Master" score="INFINITY"/>
>>
>> The node lost connectivity with 192.168.18.0 (I defined all host in
>> this network for pingd) and pingd value should be less than 1000.
>
> Should be or is?
It could be a problem with pingd again, I couldn't see pingd updating
the value.
I have attached hb_report.
>
> Could you post your xml including the status section in this situation?

>
> Regards
> Dominik
>
> _______________________________________________
> Pacemaker mailing list
> Pacemaker [at] clusterlabs
> http://list.clusterlabs.org/mailman/listinfo/pacemaker
>

Attachments:

report.tar.bz2 (30.1 KB)

beekhof at gmail

Dec 2, 2008, 6:11 AM

Post #11 of 13 (1383 views)
Permalink

Re: Unnecessary shuffling of master/slave resources and maybe a split brain [In reply to]

You created bug 2003 for this right? Or is this another case?

On Thu, Nov 20, 2008 at 19:49, Adrian Chapela
<achapela.rexistros [at] gmail> wrote:
> Dominik Klein escribió:
>>
>> Adrian Chapela wrote:
>>>
>>> Dominik Klein escribió:
>>>>>
>>>>> What are you saying me ? I should have another communication channel as
>>>>> a COM port... I could test this with a null modem cable.
>>>>
>>>> You need at least 2 connections between the nodes. It never was
>>>> different.
>>>>
>>>> It once was true that a null modem oder serial cable as the second
>>>> connection was enough - from my _personal_ experience, having 2 ethernet
>>>> connections is better (because faster and able to cope with the rather high
>>>> traffic of v2 (pacemaker)).
>>>
>>> I tested two options. I think the best option is a second ethernet card.
>>> I have configured this option but I saw that my cluster didn't do a
>>> failover.
>>> The two connections are in different networks (One in 192.168.18.X and
>>> other in 172.16.1.0) but I don't know why pingd isn't going down all of
>>> resources.
>>>
>>> I have this connectivity constraint:
>>> <rsc_location id="mail-connectivity" rsc="Mail-drbd">
>>> <rule id="mail-pingd-prefer-rule" score="-INFINITY"
>>> role="Master">
>>> <expression id="mail-pingd-prefer"
>>> attribute="pingd" operation="lt" value="1000"/>
>>> </rule>
>>> </rsc_location>
>>>
>>>
>>>
>>> <rsc_location id="samba-connectivity" rsc="Samba-drbd">
>>> <rule id="samba-pingd-exclude-rule" score="-INFINITY" >
>>> <expression id="samba-pingd-exclude"
>>> attribute="pingd" operation="lt" value="1000"/>
>>> </rule>
>>> </rsc_location>
>>>
>>> And this colocation:
>>> <rsc_colocation id="mail_drbrd_rule" rsc="Mail" with-rsc="Mail-drbd"
>>> with-rsc-role="Master" score="INFINITY"/>
>>> <rsc_colocation id="samba_drbrd_rule" rsc="Samba" with-rsc="Samba-drbd"
>>> with-rsc-role="Master" score="INFINITY"/>
>>>
>>> The node lost connectivity with 192.168.18.0 (I defined all host in this
>>> network for pingd) and pingd value should be less than 1000.
>>
>> Should be or is?
>
> It could be a problem with pingd again, I couldn't see pingd updating the
> value.
> I have attached hb_report.
>>
>> Could you post your xml including the status section in this situation?
>
>>
>> Regards
>> Dominik
>>
>> _______________________________________________
>> Pacemaker mailing list
>> Pacemaker [at] clusterlabs
>> http://list.clusterlabs.org/mailman/listinfo/pacemaker
>>
>
>
> _______________________________________________
> Pacemaker mailing list
> Pacemaker [at] clusterlabs
> http://list.clusterlabs.org/mailman/listinfo/pacemaker
>
>

_______________________________________________
Pacemaker mailing list
Pacemaker [at] clusterlabs
http://list.clusterlabs.org/mailman/listinfo/pacemaker

achapela.rexistros at gmail

Dec 2, 2008, 6:59 AM

Post #12 of 13 (1385 views)
Permalink

Re: Unnecessary shuffling of master/slave resources and maybe a split brain [In reply to]

Andrew Beekhof escribió:
> You created bug 2003 for this right? Or is this another case?
>

Yes, I created this bug. I will test this afternoon if it is possible.

Thank you!
> On Thu, Nov 20, 2008 at 19:49, Adrian Chapela
> <achapela.rexistros [at] gmail> wrote:
>
>> Dominik Klein escribió:
>>
>>> Adrian Chapela wrote:
>>>
>>>> Dominik Klein escribió:
>>>>
>>>>>> What are you saying me ? I should have another communication channel as
>>>>>> a COM port... I could test this with a null modem cable.
>>>>>>
>>>>> You need at least 2 connections between the nodes. It never was
>>>>> different.
>>>>>
>>>>> It once was true that a null modem oder serial cable as the second
>>>>> connection was enough - from my _personal_ experience, having 2 ethernet
>>>>> connections is better (because faster and able to cope with the rather high
>>>>> traffic of v2 (pacemaker)).
>>>>>
>>>> I tested two options. I think the best option is a second ethernet card.
>>>> I have configured this option but I saw that my cluster didn't do a
>>>> failover.
>>>> The two connections are in different networks (One in 192.168.18.X and
>>>> other in 172.16.1.0) but I don't know why pingd isn't going down all of
>>>> resources.
>>>>
>>>> I have this connectivity constraint:
>>>> <rsc_location id="mail-connectivity" rsc="Mail-drbd">
>>>> <rule id="mail-pingd-prefer-rule" score="-INFINITY"
>>>> role="Master">
>>>> <expression id="mail-pingd-prefer"
>>>> attribute="pingd" operation="lt" value="1000"/>
>>>> </rule>
>>>> </rsc_location>
>>>>
>>>>
>>>>
>>>> <rsc_location id="samba-connectivity" rsc="Samba-drbd">
>>>> <rule id="samba-pingd-exclude-rule" score="-INFINITY" >
>>>> <expression id="samba-pingd-exclude"
>>>> attribute="pingd" operation="lt" value="1000"/>
>>>> </rule>
>>>> </rsc_location>
>>>>
>>>> And this colocation:
>>>> <rsc_colocation id="mail_drbrd_rule" rsc="Mail" with-rsc="Mail-drbd"
>>>> with-rsc-role="Master" score="INFINITY"/>
>>>> <rsc_colocation id="samba_drbrd_rule" rsc="Samba" with-rsc="Samba-drbd"
>>>> with-rsc-role="Master" score="INFINITY"/>
>>>>
>>>> The node lost connectivity with 192.168.18.0 (I defined all host in this
>>>> network for pingd) and pingd value should be less than 1000.
>>>>
>>> Should be or is?
>>>
>> It could be a problem with pingd again, I couldn't see pingd updating the
>> value.
>> I have attached hb_report.
>>
>>> Could you post your xml including the status section in this situation?
>>>
>>> Regards
>>> Dominik
>>>
>>> _______________________________________________
>>> Pacemaker mailing list
>>> Pacemaker [at] clusterlabs
>>> http://list.clusterlabs.org/mailman/listinfo/pacemaker
>>>
>>>
>> _______________________________________________
>> Pacemaker mailing list
>> Pacemaker [at] clusterlabs
>> http://list.clusterlabs.org/mailman/listinfo/pacemaker
>>
>>
>>
>
> _______________________________________________
> Pacemaker mailing list
> Pacemaker [at] clusterlabs
> http://list.clusterlabs.org/mailman/listinfo/pacemaker
>
>

_______________________________________________
Pacemaker mailing list
Pacemaker [at] clusterlabs
http://list.clusterlabs.org/mailman/listinfo/pacemaker

achapela.rexistros at gmail

Dec 3, 2008, 9:51 AM

Post #13 of 13 (1378 views)
Permalink

Re: Unnecessary shuffling of master/slave resources and maybe a split brain [In reply to]

Adrian Chapela escribió:
> Andrew Beekhof escribió:
>> You created bug 2003 for this right? Or is this another case?
>>
>
> Yes, I created this bug. I will test this afternoon if it is possible.

I tested today and I reopened the bug because the bug is still present
for the case of unplugging an ethernet cable.

>
> Thank you!
>> On Thu, Nov 20, 2008 at 19:49, Adrian Chapela
>> <achapela.rexistros [at] gmail> wrote:
>>
>>> Dominik Klein escribió:
>>>
>>>> Adrian Chapela wrote:
>>>>
>>>>> Dominik Klein escribió:
>>>>>
>>>>>>> What are you saying me ? I should have another communication
>>>>>>> channel as
>>>>>>> a COM port... I could test this with a null modem cable.
>>>>>>>
>>>>>> You need at least 2 connections between the nodes. It never was
>>>>>> different.
>>>>>>
>>>>>> It once was true that a null modem oder serial cable as the second
>>>>>> connection was enough - from my _personal_ experience, having 2
>>>>>> ethernet
>>>>>> connections is better (because faster and able to cope with the
>>>>>> rather high
>>>>>> traffic of v2 (pacemaker)).
>>>>>>
>>>>> I tested two options. I think the best option is a second ethernet
>>>>> card.
>>>>> I have configured this option but I saw that my cluster didn't do a
>>>>> failover.
>>>>> The two connections are in different networks (One in 192.168.18.X
>>>>> and
>>>>> other in 172.16.1.0) but I don't know why pingd isn't going down
>>>>> all of
>>>>> resources.
>>>>>
>>>>> I have this connectivity constraint:
>>>>> <rsc_location id="mail-connectivity" rsc="Mail-drbd">
>>>>> <rule id="mail-pingd-prefer-rule" score="-INFINITY"
>>>>> role="Master">
>>>>> <expression id="mail-pingd-prefer"
>>>>> attribute="pingd" operation="lt" value="1000"/>
>>>>> </rule>
>>>>> </rsc_location>
>>>>>
>>>>>
>>>>>
>>>>> <rsc_location id="samba-connectivity" rsc="Samba-drbd">
>>>>> <rule id="samba-pingd-exclude-rule" score="-INFINITY" >
>>>>> <expression id="samba-pingd-exclude"
>>>>> attribute="pingd" operation="lt" value="1000"/>
>>>>> </rule>
>>>>> </rsc_location>
>>>>>
>>>>> And this colocation:
>>>>> <rsc_colocation id="mail_drbrd_rule" rsc="Mail" with-rsc="Mail-drbd"
>>>>> with-rsc-role="Master" score="INFINITY"/>
>>>>> <rsc_colocation id="samba_drbrd_rule" rsc="Samba"
>>>>> with-rsc="Samba-drbd"
>>>>> with-rsc-role="Master" score="INFINITY"/>
>>>>>
>>>>> The node lost connectivity with 192.168.18.0 (I defined all host
>>>>> in this
>>>>> network for pingd) and pingd value should be less than 1000.
>>>>>
>>>> Should be or is?
>>>>
>>> It could be a problem with pingd again, I couldn't see pingd
>>> updating the
>>> value.
>>> I have attached hb_report.
>>>
>>>> Could you post your xml including the status section in this
>>>> situation?
>>>> Regards
>>>> Dominik
>>>>
>>>> _______________________________________________
>>>> Pacemaker mailing list
>>>> Pacemaker [at] clusterlabs
>>>> http://list.clusterlabs.org/mailman/listinfo/pacemaker
>>>>
>>>>
>>> _______________________________________________
>>> Pacemaker mailing list
>>> Pacemaker [at] clusterlabs
>>> http://list.clusterlabs.org/mailman/listinfo/pacemaker
>>>
>>>
>>>
>>
>> _______________________________________________
>> Pacemaker mailing list
>> Pacemaker [at] clusterlabs
>> http://list.clusterlabs.org/mailman/listinfo/pacemaker
>>
>>
>
>

_______________________________________________
Pacemaker mailing list
Pacemaker [at] clusterlabs
http://list.clusterlabs.org/mailman/listinfo/pacemaker

Index | Next | Previous | View Threaded

Interested in having your list archived? Contact Gossamer Threads