Re: node and port alloclass, cannot add a node to the cluster

Anton Shterenlikht wrote:
For instance, I've a 2node Alpha-I64 cluster even though I never
run CLUSTER_CONFIG.COM or CLUSTER_CONFIG_LAN.COM. Somehow the cluster
was formed,


For a node to join an existing cluster, it needs to have the right cluster_authorize.dat file in its sys$system: as well as having its one proper SYSGEN parameters to enable the clustering code.

If the node is a satellite, it needs to be defined in the LANCP so its MOP requests can be answered, and it needs to have its own system root [SYSx.]. It is possible to set those up manually, but not recommended.

I now understand that actually one can only add a node of the same
architecture as that of a node on which CLUSTER_CONFIG.COM is run.

If you add a standalone node (with its own system disk), you can run CLUSTER_CONFIG on that node. You'll be prompted for the cluster information (group and password) and that procedure will create a local CLUSTER_AUTHORIZE.DAT and set the proper SYSGEN parameters.

When you add a satellite node, you run cluster_config on the boot node to define the satellite's parameters ( ethernet address, root name, node name etc). It then creates the [.SYSx...] structure along with the alias to VMS$COMMON.DIR and populates it with a basic SYSGEN/MODPARAMS data.


Logical names are then defined by the system manager based on how he wants his cluster to operate. (shared queues, shared SYSUAF etc etc).


>However, on my alpha and I64 they are all
defined in SYLOGICALS.TEMPLATE. Is there not a contradiction?

SYLOGICALS.TEMPLATE does not get executed. You are not supposed to modify the .TEMPLATE file. You are supposed to use it as "inspiration" to populate SYLOGICALS.COM which is the one that gets executed early in the boot stages.


I understood from Bob's reply that if I put the core system files
like e.g. SYSUAF.COM in SYS$COMMON:[SYSEXE] on one node, then I only need
to define the clusterwide logical names on the other node. Is that
correct? Does it matter which node?

It depends on your environment. If SYSUAF.DAT resides on a disk that is accessible only from NODE-A, consider the implications of NODE-B booting first and NODE-A remaining offline. If NODE-B defines SYSUAF to point to a non-existing device (since NODE-A hasn't booted), then anyone can access NODE-B without a password.

So only the node(s) who have direct access to the files (and who can serve it to nodes not having direct access) should define such logicals, and when other nodes join the cluster, the cluster-wide table gets copied and they automatically get the definitions for SYSUAF etc.

When a node boots and the central SYSUAF.DAT is not available, it should default to defining no logical name and having a minimally populated SYSUAF.DAT file in its SYS$SYSTEM to provide access to at least the system manager until the main node boots and provides access to te central SYSUAF as well as defining the clusterwide logical, after which, those other system automatically start to access the central one instead of the minimalist local one.

Why do the definitions in SYLOGICALS.TEMPLATE have /SYSTEM qualifiers:
Wouldn't /CLUSTER_SYSTEM be more appropriate?

Clusterwide logicals are a relatively recent addition of VMS (7.2 if I remember correctly). There is not yet a simple /CLUSTER_SYSTEM qualifier. Also, note that it is possible to create cluster-wide loggical name tables that are not "SYSTEM" (aka: think about group logical name tables that propagate across nodes.).
.

Relevant Pages

  • Re: cron facility wanted for VMS
    ... fail to run - such as disk failures. ... Cluster node serving the disk was not up (usually in cluster ... cluster for which your queue is setup to fail over with AUTOSTART. ... defined as a "Cluster Logical" Cluster Logicals are also quite useful ...
    (comp.os.vms)
  • Re: cron facility wanted for VMS
    ... fail to run - such as disk failures. ... There are three aspects of the OpenVMS Batch Facility which I rarely ... Cluster node serving the disk was not up (usually in cluster ... defined as a "Cluster Logical" Cluster Logicals are also quite useful ...
    (comp.os.vms)
  • Re: -COSI-I-NOTSYSCONCEAL, error
    ... It has been my experience that if you dare mention Rdb and "Cluster" in the ... cluster-wide logicals names are relatively new and IIRC ...
    (comp.databases.rdb)
  • Re: Backup/Copy tree with alias file
    ... >>in logicals defined as search lists, ... This dates back to V3/V4 days and the initial cluster code that would ...
    (comp.os.vms)
  • Re: $SHOW DEVICE/FILES enhancement request
    ... You can have a cluster where each machine has it's own ... >> installed the latest patches on the node with it's own system drive ... > We had lots of system disks. ... > SYSUAF, one RIGHTSLIST, ..., no matter how many system disks. ...
    (comp.os.vms)