Andrew Beekhof wrote:
On Jan 31, 2008, at 7:53 PM, Steve Wray wrote:
Andrew Beekhof wrote:
On Jan 31, 2008, at 1:53 AM, Steve Wray wrote:
Ok I now have cib.xml working but the behavior of the cluster is
still strange.
I took the code from the pingd documentation and inserted it into
the cib.xml as follows:
[snip]
The documentation is not clear on this, but is that the correct
place to insert the code fragment?
Yes.
Cool, I had wondered if I was supposed to use some tool to insert
this; seems a bit odd to manually edit a file that appears to be
automatically updated.
The observed behavior is now that if the passive node loses network
connectivity
you didn't simulate this by unplugging eth0 did you?
The two nodes on this testbed are virtual machines running under Xen.
I can't unplug eth0. I took the network interface down.
Which amounts to the same thing and isn't a good idea.
What you've just done is create a split-brain scenario (try google) and
without stonith enabled is a really good way to corrupt your data.
pingd can only help you when you loose external connectivity - not when
you loose connectivity to the other cluster node(s)
Its a simple rule.
If the node loses connectivity to *everything* then
if it was already passive it should stay passive.
if it was active it should assume that the other node is still good
and go passive.
Its not rocket surgery.
The node is supposed to be providing a network service; if it can't
connect to the network *at* *all* theres no point in it becoming the
active node, now is there?
And for what its worth, I believe that its ipfail which has a problem
with this not pingd.
In fact, the pingd documentation appears to provide for exactly the
scenario I've outlined... its just that, following the instructions to
the best of my ability, I've been unable to get it to work.
From the pingd documentation:
<quote>
It is sometimes desirable to shut a particular service down if ping
connectivity is lost. This rule will prohibit the service from running
anywhere that there is no ping connectivity to the outside world, and
all nodes with some connectivity are treated as the same, regardless of
how many ping nodes are accessible.
</quote>
which is what I want to achieve and which should protect from
'split-brain' in event of total loss of network.
_______________________________________________
Linux-HA mailing list
[email protected]
http://lists.linux-ha.org/mailman/listinfo/linux-ha
See also: http://linux-ha.org/ReportingProblems
