Multiple system disk cluster

[Wim Van den Wyngaert]

I have a 7.3 cluster with a common system disk called salpv1/2.

I have a 8.2 station with a local system disk called wsys07.

I want to add the 8.2 node to the cluster with cluster_config_lan (yes I should be on 7.3-2 but I wanted to try it anyway).

I had to change the decnet area of the 8.2 node and rebooted it afterwards. The node is reachable in lat, decnet and ip.

But cluster_config_lan doesn't accept the config. See enclosure. Show cluster doesn't show the new node in the cluster.

1) Why ? Is this the problem for which the whole cluster must be rebooted ? How to verify that ?
2) Why is it asking for the MOP service name when it is not a satelite ?

Wim

[Ian Miller.]

I would add the various parameters in into the V8.2 node by hand, edit SYLOGICALS on that system disk to mount the cluster common disk and reboot it.

[Karl Rohwedder]

You said, you rebootet the node WSYS07. Did it join the cluster? There should have been messages during the bootstrap.

I normally just set the SYSGEN parameters on the new node and copy over the CLUSTER_AUTHORIZE.DAT (in case of nodes with local system disks).

regards Kalle

[Wim Van den Wyngaert]

Ian,

I want a shared nothing but disks system.

Kalle,

The boot was before the cluster_config.

Wim

[Volker Halle]

Wim,

you don't have to do ANYTHING on node SALPV1/2 to add another node to the cluster, which boots from it's own system disk (except maybe think about and change EXPECTED_VOTES, but in a config like this, I would probably run WSYS07 with VOTES=0).

The work for allowing WSYS07 to join an existing cluster needs to be done on WSYS07 itself. You would need to run CLUSTER_CONFIG_LAN.COM to make this node a cluster member. For that you would need to know the cluster group number and password. As Kalle said, just copy CLUSTER_AUHTORIZE.DAT, if you don't know the password anymore.

Volker.

[Karl Rohwedder]

I think, you had to specifiy Satellite=y to
make cluster_config_lan use LAN as interconnect.

So it thinks, you want to add a 2nd bootnode, which had no clusterinterconnect (CI/DSSI/SHARED-SCSI) -> abort.

regards Kalle

[Ian Miller.]

you can copy CLUSTER_AUTHORIZE.DAT and
add to MODPARAMS.DAT
NISCS_LOAD_PEA0=1
VOTES=0
EXPECTED_VOTES=x
VAXCLUSTER=2
INTERCONNECT="NI"
MSCP_LOAD=1
MSCP_SERVE_ALL=1
etc
you should share the uaf, rightslist, etc so should mount the cluster common disk in SYLOGICALS

[Wim Van den Wyngaert]

It seems to work. I copied the clus auth file and I did :

NISCS_LOAD_PEA0=1
VOTES=0
EXPECTED_VOTES=x
VAXCLUSTER=2

Not (obsolete and only when station is serving disks) :

INTERCONNECT="NI"
MSCP_LOAD=1
MSCP_SERVE_ALL=1

And not :

Something with cluster_config_lan

Still have to solve some license issues ...

But I didn't find this in any documentation on hp.com. Any links on how to setup clusters of this kind and on when/how to use cluster_config ?

Wim

[Karl Rohwedder]

Licenses:

Make sure, you have a common LMF database (LMF$LICENSE is the logical which points to the file, should by on a common cluster disk).

And that all NOSHARE licenses are tight to a specific node /INCLUDE=...

regards Kalle

[Jan van den Ende]

Wim,


I want a shared nothing but disks system.


If I take that litteral, it is even simpler still!

If a system with VAXCLUSTER=2 boots, and it detects another cluster with the same cluster number and password, the system simply joins the cluster.

Initially, it is even "shared nothing, not even disks"

You CAN, if you so wish, keep the LMF database, SYSUAF, RIGHTSLIST, etc, all specific, that is, on the different system disks.
Now, IF you go mounting the disks shared, _YOU_ are responsible for the security issues that can arise if your SYSUAF and/or RIGHTSLIST databases contain conflicting info.
eg: USERA has UIC [x,y] in SYSUAF1, but that UIC is USERB in SYSUAF2.

Now, the files of USERA as seen from the system with SYSUAF1 _BELONG TO_ USERB as seen from a node with SYSUAF2.

But apart from that, you created a perfectly valid cluster.

-- and is you do not use VMS MOUNT, but NFS MOUNT for your disks, you have mimicked the basic concept of Unix clustering.


Proost.

Have one on me.

jpe

[Robert_Boyd]

There is a fair amount of information about how to do this kind of cluster setup in the documentation:

OpenVMS Cluster Systems: http://h71000.www7.hp.com/DOC/731FINAL/4477/4477PRO.HTML

Especially read Chapter 5 on creating a shared environment and Appendix B on building common files.

and

Guidelines for OpenVMS Cluster Configurations: http://h71000.www7.hp.com/DOC/82FINAL/6318/6318PRO.HTML

See Chapter 11 in particular on strategies for managing various kinds of cluster configurations. Section 11.3 discusses the mechanisms for creating a common environment.

Robert

[comarow]

If it shares an existing system disk, than the current cluster group number and password is correct.

Did you say it is a satelite?

What is the cluster typology? Are some nodes on the network and some not?
If it is a san or scsi disk, but not on a CI or memory channel ALL nodes must have niscs_load_pea0 set to 1, vaxcluster to 2.

I didn't see the error statement. What was it?

[Wim Van den Wyngaert]

Robert,

I find all theoretical issues but not on how to implement the cluster (or did I view the pages too fast ?). E.g. where is written when/how to use cluster_config and when to do setup yourself ?

Comarow,

Ethernet interconnect. Message in my original posting (enclosed).

Wim

[Volker Halle]

Wim,

for a configuration like this, just run @CLUSTER_CONFIG on the node to be added to the existing cluster with it's own system disk and use item 1:

1. ADD WSYS07 to existing cluster, or form a new cluster.

(this option will only be shown, if the local node is not yet running cluster software, e.g. F$GETSYI("CLUSTER_MEMBER") = FALSE).

As an example, when I installed V8.2 on our rx2600, I answered the following question:

Will this system be a member of an OpenVMS Cluster? (Yes/No) YES

during installation and after the first reboot, CLUSTER_CONFIG.COM was automatically run and allowed me to enter the necessary parameters for this node to join the existing cluster (SCS via LAN).

It looks like this variant of a cluster config does not seem to be explicitly covered in the manuals, but it's really only: make this standalone node a cluster member.

Volker.

[comarow]

It's simple.

Run cluster config.
Tell it is a cluster.
Don't say you have local page or swap files.
Do that later. Otherwise autogen will run.
Make sure you have added
vaxcluster 2
niscs_load_pea0 1
to modparams.dat
give it 0 votes until you are ready to go productions.
Set expected votes to 1, you must see the other node.

If you don't know cluster group number and pass word put anything in and copy cluster_authorization.dat over.

Make sure the original node has nisc_load_pea0 =1
vaxcluster 2
and has at least a vote.
It should start as a stand alone cluster.
If it's not do that first.

You can not have any duplicate volume labels or devices. That will stop booting.