[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]

RE: [Linux-cluster] Tiebreaker IP Address



As you said, Fencing is a nice way of saying "make sure the non-responsive
node can not write anything to our disks, by whatever means necessary".
This usually involves the equivalent of pulling the power plug out of the
non-responsive node.  Why be so harsh?  Why not do a normal shutdown?

A "normal" shutdown will always flush buffers to disk. The most important thing is the integrity of our data. If the cluster has determined the node is not functioning properly, we don't want to give it the opportunity to write bad/corrupted data to our disk. By "pulling the plug" it will not be able to do so.

So does that means that even in any case of cluster failure (suppose a
network fail), the node will shutdown abnormally only, or it will be a clean
shutdown. And once a node is shutdown due to a failure, will the node
automatically come up or does it need to be manually brought up.

As mentioned before, fencing is "pulling the plug" .. if fencing is set up correctly, the node will reboot and rejoin the cluster.

Barry
-----Original Message-----
From: linux-cluster-bounces redhat com
[mailto:linux-cluster-bounces redhat com] On Behalf Of Barry Brimer
Sent: Sunday, January 20, 2008 7:31 PM
To: linux clustering
Subject: Re: [Linux-cluster] Tiebreaker IP Address

Can any one explain me what exactly is the tiebreaker IP and how does it
function? What is the use if we set the tiebreaker IP as the Default
Gateway
address?

In clustering, it is important that the cluster nodes are able to
communicate with one another.  It is also important that the cluster nodes
agree on the status of the cluster.  To acheive this, various methods are
used to communicate between cluster nodes to inform the other nodes that
this node is active and participating in the cluster.  Quorum is usually
defined as "greater than one half".  In a cluster larger than 2 nodes,
the cluster nodes can determine that if they stop receiving cluster
communications (usually referred to as heartbeat) from a particular node,
they assume that the non-responsive node is not functioning correctly, and
one of the remaining nodes in the cluster will fence the non-responsive
node.  Fencing is a nice way of saying "make sure the non-responsive node
can not write anything to our disks, by whatever means necessary".  This
usually involves the equivalent of pulling the power plug out of the
non-responsive node.  Why be so harsh?  Why not do a normal shutdown?  If
the non-responsive node has data in buffers that has not been written to
disk, and the other cluster nodes feel that this node is having a problem,
they want to ensure that the non-responsive node can not write its buffers
out to disk, in order to make sure that the non-responsive node has no
chance of corrupting the data used by the cluster.  This is all fine,
because if you have greater than 2 nodes, you should be able to get
agreement by a majority on whether a node is functioning, and therefore
whether the cluster is allowed to operate.  In a two-node cluster, we need
to have some other way to determine which cluster member is healthy, and
which one isn't.  If a cluster node were functioning correctly, it would
be able to reach its default gateway.  Therefore the tiebreaker IP address
is the default gateway because both machines should be able to reach it if
they were functioning properly.  Therefore if one node is able to reach
the tiebreaker IP address, and one isn't, it is assumed that the properly
running node is the one that can reach the default gateway, and that
allows the tie to be broken and allows that node to fence the other node.

Barry

--
Linux-cluster mailing list
Linux-cluster redhat com
https://www.redhat.com/mailman/listinfo/linux-cluster


<<<<   Disclaimer Message  >>>>
"This email and any files transmitted with it are confidential and intended solely for the use of the individual or entity to whom they are addressed. If you are not the named addressee, please notify the sender immediately after deleting this e-mail from your system and do not disseminate, distribute or copy this e-mail. The sender does not accept liability for any errors or omissions in the contents of this message, which arise as a result of erroneous e-mail transmission."
[Mohsin Haider Darwish LLC & Group Companies, PO.Box 880, Ruwi-112, Oman]

--
Linux-cluster mailing list
Linux-cluster redhat com
https://www.redhat.com/mailman/listinfo/linux-cluster

!DSPAM:4796b976281132343459193!




[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]