Re: TCPIP 5.4: obsolete information in HELP, what are new defaults (clusteralias)

From: Phillip Helbig---remove CLOTHES to reply (helbig_at_astro.multiCLOTHESvax.de)
Date: 03/24/05 

Date: Thu, 24 Mar 2005 07:08:37 +0000 (UTC)

In article <C%s0e.9634$C7.8696@news-server.bigpond.net.au>,
"NoName@Nospam.com" <NoEmail@ThisAddress.com> writes:

> As you hope, the cluster alias functionality will continue to work after
> upgrading to V5.4. No configuration changes are necessary. So, you can
> explore failSAFE IP at your leisure and knowing that the old cluster alias
> functionality continues to work as it did in V5.3 and prior versions.

OK.

> For similar functionality to SHOW INTER/CLUSTER, you will need to use
> "ifconfig -a". This will display all the configured addresses. If there are
> standby addresses, then it will tell you where the address is active.
> Unlike the IP Cluster Alias, failSAFE supports many instances of standby
> addresses across a cluster.

I guess I can use

$ pipe ifconfig -a | sea sys$pipe aaa.bbb.ccc.ddd

where aaa.bbb.ccc.ddd is the cluster-alias address, then check $STATUS
to see if the machine which executed the command has the cluster alias.

The reason I need to do this in a procedure is the following: I am
running the OSU HTTP Server on a cluster with several virtual hosts
(i.e. for virtual web servers which all have the same IP address, the
server returns different pages depending on the name of the host
specified in the HTTP request). Without this fancy feature, running it
on a cluster is easy---just run an instance on each node and it will
listen on both the specific address and the cluster-alias address.
Using the multiple virtual servers on one machine is easy as well. The
problem is, in order to use the virtual-server stuff, one must specify a
NUMERIC IP address in the configuration files. Obviously, I want to use
the same configuration files for all nodes. However, I can only specify
an IP address which really exists on that node. Specifying the specific
address won't work since that would mean different configuration files
on each cluster. Specifying the cluster-alias address will only work on
the node which actually currently has the cluster alias. So, what I do
is run the server only on the node which has the cluster alias. In
order to do this, I use SHOW INTERFACE/CLUSTER in a PIPE command with
SEARCH and check the status. Since I'm interested only in failover and
not in load balancing, this means that the other nodes would just have
an idle server anyway, until one of them got the cluster alias. In
order to have failover functionality, I have a batch job which
periodically checks if the server is still running and if it isn't, it
submits a job on each node to start it, with the result that due to the
test above it only starts on the node with the cluster alias.

If anyone knows a more elegant way to run the OSU server in a cluster
with several virtual hosts, let me know!

I realise that a test like the one I am using is not very elegant, only
marginally better than parsing the output of a command. It would be
nice if there were a lexical function to indicate whether a machine
currently has the cluster-alias address and/or what that address is.

With the ifconfig command, apparently I have to know the cluster-alias
address and hard-code it (at least at some level) in the command,
whereas with the old version I don't have to search for the address
itself, but just for IMPERSONATOR.

> Actually, the work to continue to add functionality to the TCPIP CLI is far
> from negligible. We now have *many* diverse BSD-based utilities for
> managing various parts of TCPIP and the layered services. It is very
> difficult to justify the cost of wrapping a VMS-style CLI around otherwise
> functional (albeit - perhaps more cryptic) utilities that have been more or
> less accepted as industry standard. Keeping in mind, extending the CLI
> takes someone's time away from developing additional features or addressing
> problem reports, etc. This is the usual issue of trade-offs here, and we
> hope we are keeping most of our customers happy most of the time.

Sic transit gloria mundi. Yes, I see your point. On the other hand, it
is nice to have ONE CLI with an integrated HELP, SET and SHOW commands
etc. Also, since three letters are needed to be unique, typos are
caught more easily than with a unix-style command.

Don't forget to remove SHOW INTERFACE/CLUSTER from HELP (or, perhaps
better, replace it with a note saying how to do something similar with
another command).

Relevant Pages

  • RE: [Maybe SPAM] Re: how to set time on the cluster nodes
    ... "scshutdown" command, rebooted in non-cluster mode & set the same time ... Subject: Re: how to set time on the cluster nodes ... Hi Sanjay, ... as an ntp server, you can use the ntpdate command... ...
    (SunManagers)
  • Re: Help with MSDTC settings
    ... On both nodes, stop the MSDTC service. ... To do this, type the following command at a command prompt, and then press ... After the COM Cluster Wizard has finished running, ...
    (microsoft.public.windows.server.clustering)
  • Re: Updating VGs that are part of a cluster
    ... Think the real issue was I accidentally put a -L in the "varyonvg" command ... Group Descriptor Areas of only the disks that are members of the same volume ... Then sync up cluster. ...
    (AIX-L)
  • Re: Under what conditions will evict hang?
    ... If you want remove a node from server cluster, ... Use /Stop command option, ... Windows 2000 node and try to evict a Windows Server 2003, ...
    (microsoft.public.windows.server.clustering)
  • Re: Finding the Active Cluster Node
    ... > One option would be to use the cluster.exe command line utility ... Say my cluster name is MyCluster and I want to know the node name ... > Resource Group Node Status ... > Microsoft highly recommends that users with Internet access update their ...
    (microsoft.public.sqlserver.clustering)