[Linux-HA] colocating resources on failed restart :CRM-Stable-4a0d4e40eeb0

Andrew Beekhof beekhof at gmail.com
Mon Nov 6 07:48:46 MST 2006 

On 11/6/06, Andrew Beekhof <beekhof at gmail.com> wrote:
> On 11/5/06, Alex and Gill Strachan <asgks at yahoo.com> wrote:
> > Hopefully an easier question to understand.  I was confused after reading my last post !
>
> I see you've made some progress, but this is probably worth answering anyway.
>
> What has happened is that you said "try to put aims where the ip is".
> This means that the ip will be placed first, and then later aims will
> put in the same place.
>
> However when it comes time to put aims in the same place, we find that
> it cant run there and so we put it somewhere else (which we can do
> because you used < INFINITY)
>
> Reversing the direction of the colocation constraints may have helped
> you here (because we'd have decided where to put aims first).

Its also worth pointing out that one needs to think carefully about
the collocation constraints.

If we have an IP, FileSystem and a WWW server, then this is probably wrong:

   collocate(FS with IP)
   collocate(WWW with FS)

because a failing filesystem (that could potentially be mounted
elsewhere) wont be able to take the IP with it (since its already
allocated somewhere).

Instead this is better:
   collocate(IP with FS)
   collocate(WWW with FS)

The filesystem is the weak link here, so tell us to find a good place
for it to run, and only then place the rest accordingly.

Remember, one still needs to add rsc_ordering constraints :-)


This is new behavior that will be in 2.0.8 and a consequence of the
weak-collocation work.  I'm still not 100% sure its in its final form.

>
> >
> > Okay
> >
> > From this output
> > ============
> > Last updated: Sun Nov  5 18:27:58 2006
> > Current DC: sinfids3a2 (ec74bd17-2016-4d32-a694-0f6983121cd9)
> > 3 Nodes configured.
> > 9 Resources configured.
> > ============
> >
> > Node: sinfids3b1 (338afa76-8997-4d66-8381-fc36ec4b456b): online
> >         resource_sinfids3B_vip  (heartbeat::ocf:IPaddr)
> >
> > Node: sinfids3a2 (ec74bd17-2016-4d32-a694-0f6983121cd9): online
> >         resource_sinfids3A_aims (lsb:aims)           <---ALL BY ITSELF
> >
> > Node: sinfids3a1 (b757aece-0e47-41e5-92b7-6a80b4f3eea7): online
> >         resource_sinfids3A_drbd (heartbeat:drbddisk)
> >         resource_sinfids3A_fs   (heartbeat::ocf:Filesystem)
> >         resource_sinfids3A_smb  (lsb:smb)
> >         resource_sinfids3A_oralsnr      (heartbeat::ocf:oralsnr)
> >         resource_sinfids3A_vip  (heartbeat::ocf:IPaddr)
> >         resource_sinfids3_vip   (heartbeat::ocf:IPaddr)
> >         resource_sinfids3A_oracle       (heartbeat::ocf:oracle)
> >
> > How can I configure cib.xml so that if resource resource_sinfids3A_aims is relocated to another node the other 3A resources follow?
> >
> >
> >
> >
> > ----- Original Message ----
> > From: Alex and Gill Strachan <asgks at yahoo.com>
> > To: linux-ha <linux-ha at lists.linux-ha.org>
> > Sent: Sunday, 5 November, 2006 2:38:07 PM
> > Subject: [Linux-HA] colocating resources on failed restart :CRM-Stable-4a0d4e40eeb0
> >
> > I have a group of resources linked by the name 3A, these resources must always run together so I allocated large co-location scores.
> >
> > When the resource_sinfids3A_aims fails and it is moved to another node I need all of the 3A resources to move with it and to start before.
> >
> > e.g.
> > resource_sinfids3A_aims  fails on node 3a2
> > hb restarts and reduces node weight for that node..
> > resource_sinfids3A_aims  fails on node 3a2
> > hb is unable to restart on node 3a2 so decides to relocate to 3a1
> >
> > ...How do I inform hb to stop all the other 3A resources on 3a2 and move
> > ...everything to 3a1, also starting in a particular order.
> >
> > Why didn't the colocation scores help in keeping the 3A resources together?
> >
> >
> > I originally had colocation scores of INFINITY for the 3A group but this then prevents the ability to specifiy that resource smb can fail 3 times while resource aims can only fail once.
> >
> >
> > I originally had this working by using groups and on_fail="fence" but it doesn't offer enough flexibility.
> >
> > e.g.
> > I would like heartbeat to restart smb on failure 3 times before moving to another node; using resource_stickiness.  When using groups the restart of smb would trigger a stop of all higher resources, then start smb followed by start the higher resources.  This behaviour was not wanted.
> >
> >
> >
> >
> >
> >
> > ============
> > Last updated: Sun Nov  5 14:02:46 2006
> > Current DC: sinfids3a2 (ec74bd17-2016-4d32-a694-0f6983121cd9)
> > 3 Nodes configured.
> > 9 Resources configured.
> > ============
> >
> > Node: sinfids3b1 (338afa76-8997-4d66-8381-fc36ec4b456b): online
> >         resource_sinfids3B_vip  (heartbeat::ocf:IPaddr)
> > Node: sinfids3a2 (ec74bd17-2016-4d32-a694-0f6983121cd9): online
> >         resource_sinfids3A_drbd (heartbeat:drbddisk)
> >         resource_sinfids3A_fs   (heartbeat::ocf:Filesystem)
> >         resource_sinfids3A_smb  (lsb:smb)
> >         resource_sinfids3A_vip  (heartbeat::ocf:IPaddr)
> >         resource_sinfids3A_oralsnr      (heartbeat::ocf:oralsnr)
> >         resource_sinfids3_vip   (heartbeat::ocf:IPaddr)
> >         resource_sinfids3A_oracle       (heartbeat::ocf:oracle)
> >         resource_sinfids3A_aims (lsb:aims)
> > Node: sinfids3a1 (b757aece-0e47-41e5-92b7-6a80b4f3eea7): online
> >
> >
> >
> >        <rsc_order id="order_sinfids3_sinfids3A" from="resource_sinfids3_vip" type="after" to="resource_sinfids3A_vip"/>
> >        <rsc_order id="order_sinfids3_sinfids3B" from="resource_sinfids3_vip" type="after" to="resource_sinfids3B_vip"/>
> >        <rsc_order id="order_sinfids3A_drbd" from="resource_sinfids3A_drbd" type="after" to="resource_sinfids3A_vip"/>
> >        <rsc_order id="order_sinfids3A_fs" from="resource_sinfids3A_fs" type="after" to="resource_sinfids3A_drbd"/>
> >        <rsc_order id="order_sinfids3A_smb" from="resource_sinfids3A_smb" type="after" to="resource_sinfids3A_fs"/>
> >        <rsc_order id="order_sinfids3A_oracle" from="resource_sinfids3A_oracle" type="after" to="resource_sinfids3A_fs"/>
> >        <rsc_order id="order_sinfids3A_oralsnr" from="resource_sinfids3A_oralsnr" type="after" to="resource_sinfids3A_oracle"/>
> >        <rsc_order id="order_sinfids3A_aims" from="resource_sinfids3A_aims" type="after" to="resource_sinfids3A_oralsnr"/>
> >
> >        <rsc_colocation id="colocation_sinfids3_sinfids3A" from="resource_sinfids3_vip" to="resource_sinfids3A_vip" score="9000"/>
> >        <rsc_colocation id="colocation_sinfids3_sinfids3B" from="resource_sinfids3_vip" to="resource_sinfids3B_vip" score="3000"/>
> >
> >        <rsc_colocation id="colocation_sinfids3A_drbd" from="resource_sinfids3A_drbd" to="resource_sinfids3A_vip" score="100000"/>
> >        <rsc_colocation id="colocation_sinfids3A_fs" from="resource_sinfids3A_fs" to="resource_sinfids3A_drbd" score="100000"/>
> >        <rsc_colocation id="colocation_sinfids3A_smb" from="resource_sinfids3A_smb" to="resource_sinfids3A_fs" score="100000"/>
> >        <rsc_colocation id="colocation_sinfids3A_oracle" from="resource_sinfids3A_oracle" to="resource_sinfids3A_fs" score="100000"/>
> >        <rsc_colocation id="colocation_sinfids3A_oralsnr" from="resource_sinfids3A_oralsnr" to="resource_sinfids3A_oracle" score="100000"/>
> >        <rsc_colocation id="colocation_sinfids3A_aims" from="resource_sinfids3A_aims" to="resource_sinfids3A_oralsnr" score="100000"/>
> >
> >
> >        <primitive class="lsb" type="aims" id="resource_sinfids3A_aims" restart_type="restart">
> >          <operations>
> >            <op name="monitor" timeout="240s" id="aims_sinfids3A_aims_mon" interval="180s"/>
> >          </operations>
> >          <instance_attributes id="resource_sinfids3A_aims">
> >            <attributes>
> >              <nvpair id="resource_sinfids3A_aims-target_role" name="target_role" value="started"/>
> >            </attributes>
> >          </instance_attributes>
> >        </primitive>
> >
> >
> >
> >
> >
> >
> > Send instant messages to your online friends http://au.messenger.yahoo.com
> > _______________________________________________
> > Linux-HA mailing list
> > Linux-HA at lists.linux-ha.org
> > http://lists.linux-ha.org/mailman/listinfo/linux-ha
> > See also: http://linux-ha.org/ReportingProblems
> >
> >
> >
> >
> >
> >
> > Send instant messages to your online friends http://au.messenger.yahoo.com
> >
> > _______________________________________________
> > Linux-HA mailing list
> > Linux-HA at lists.linux-ha.org
> > http://lists.linux-ha.org/mailman/listinfo/linux-ha
> > See also: http://linux-ha.org/ReportingProblems
> >
> >
> >
>

More information about the Linux-HA mailing list