From: Iain Barker (ibarker@aastra.com)
Date: Fri Jun 13 2003 - 15:48:44 EDT
Responses as follows:
Bryan Lavelle: A customer reported there was no problem during mismatched
cluster upgrade, but not confident that is the correct procedure.
Johan Brusche: V1.6 and V5.x Trucluster do NOT tolerate each other on the
memory channel. At least one of them will panic during the am_i_alone scan
at boot time.
Bluejay Adametz: Confirms the panic scenario above, and suggests
disconnecting one hub from each node, such that each node thinks it is alone
in the cluster for the duration of the upgrade.
Most likely I will proceed as suggested by Bluejay. I think there could also
be issues due to the shared SCSI bus and differences due to the requirements
for clusterwide filesystem. In summary, trucluster upgrade is not going to
be trivial. I have to do this for 140 nodes so I guess some careful
preparation will be required in the lab.
-----Original Message-----
From: Iain Barker [mailto:ibarker@aastra.com]
Sent: Saturday, 24 May, 2003 13:04
To: tru64-unix-managers@ornl.gov
Subject: TruCluster memory channel during Tru64 upgrade from 4.0f to 5.1b
Hi managers.
I have a question regarding TruCluster upgrade and software compatibiltiy
across memory channel
The setup is a two node Tru64 4.0f cluster running TruCluster 1.6, with dual
memory channel hubs and two memory channel adapters for each node as the
cluster interconnect. We also have two shared SCSI busses between the nodes.
I want update the cluster to Tru64 5.1b shortly, but have to keep at least
one of the systems active so it has to be some type of rolling in-service
upgrade.
As there is no direct Tru64 upgrade path between those two releases I was
intending to remove one system from the cluster, install Tru64 5.1b (either
a new install, or an update install through 5.0a and 5.1a as intermediates)
and then boot the node to create a new, separate cluster and not attempt to
connect to the old one. Then once that first upgraded node is running and
stable with our application, take down the second node and add it to the new
cluster to form the completed 5.1b two node cluster.
It doesn't matter to us that TruCluster will not be running between nodes
during the upgrade, our application can handle running on a single system
disconnected from the cluster with just Ethernet to transfer application
data from the old node to the new node, and I can offline the shared SCSI
bus on both nodes for the duration of the upgrade as we have additional
disks on SCSI busses local to each node.
But I wonder if having two different versions of TruCluster running on
opposite sides of the memory channel would cause problems, even though they
would not be attempting to communicate and form a cluster. For example, if
the 4.0f system is running the application and the first node is booted to
5.1b, will the mchan probe during 5.1b boot cause any problem to the 4.0f
node? The last thing I want is a kernel panic on the 4.0f node, causing both
nodes to be out of service.
If anyone has attempted such an upgrade I would be grateful for any comments
on your experiences, and also if there is an HP best practice for this type
of rolling upgrade.
thanks.
Iain
This archive was generated by hypermail 2.1.7 : Sat Apr 12 2008 - 10:49:23 EDT