cancel
Showing results for 
Search instead for 
Did you mean: 

Configuring a standalone CP Server

szhargrave
Level 3

Hi

I'm trialling an installation of SF Cluster File System across 2 nodes.  All is working fine with no fencing configured, which I'm now trying to sort.  The nodes are VMware guests, and so I have to use CP Server based fencing since SCSI-3 PR isn't supported.

I've successfully installed a single-node VCS machine elsewhere on the LAN, and configured it as a CP Server.  From the storage cluster I can run cpsadm -s 10.34.8.95 -a ping_cps and ping successfully.

The problem comes when I try to configure sfcfs fencing with installsfcfs -fencing

I tell it to configure CP client fencing without SCSI-3 PR with a single CP server (for now whilst testing).  Everything goes fine until it tries to configure and I get: -

Adding the client cluster to the CP Server 10.34.8.95 ......................................................................................... Failed

Looking in the install logs, I see: -

0 14:53:07 cmd /usr/bin/ssh -x -o NumberOfPasswordPrompts=0 -o StrictHostKeyChecking=no 10.34.8.95 "LANG=C LC_ALL=C /opt/VRTScps/bin/cpsadm -s 10.34.8.95 -a add_clus -p 14250 -c fatwire_uat -u {6d95a2b8-1dd2-11b2-bc43-99380954473a} 2>&1
echo _CMD_EXIT_=\$?" 2>&1
0 14:53:08 cmd stdout=
LLT lltstat ERROR V-14-2-15000 open /dev/llt failed: No such file or directory
CPS ERROR V-97-1400-647 Unable to get nodeid value from lltstat -N output.
CPS ERROR V-97-1400-133 Client unable to communicate with CPS.  This might be due to mismatched protocol version

Basically it ssh's to the CP server and attempts to run a number of cpsadm commands, all of which fail because /dev/llt does not exist.

In the CP Server, llt is not running, since the cluster was configured as a single-node cluster (solely for CP Server) and therefore llt and gab are not running.

So, what do I need to do to get this to work?  It is my understanding that single-node VCS running CP Server is a perfectly valid configuration, and will therefore not be running LLT, so why is it trying to use this?

Thanks,

 

Simon

1 ACCEPTED SOLUTION

Accepted Solutions

AHerr
Level 5
Employee Accredited Certified

Hi Simon,

You have captured the installation of your environment very clearly.
There are a couple of requirements for CPS.

Frist, if you have trusted ssh setup, then you can use installvcs -fencing to connect from your cluster to a CPS.  A single CP Server is supported but only for lab environments.  Anything that relies on fencing should have at least 3 coordination points.  Without trusted ssh, you can setup a relationship with CPS by manually registering the cluster and nodes.

Second, there was a change in the previous release that required LLT/GAB to be installed and configured to start VCS, even on a single node.  We are working on removing this requirement in a future version of the product.

If you have additional questions, please let me know.

 

Also, as a side note:  If you are interested in learning about the new features coming out in the next release of Storage Foundation and Cluster Server, please apply to the beta group:

https://www-secure.symantec.com/connect/groups/storage-foundation-and-veritas-cluster-server-60-beta-program



Regards,
Anthony

View solution in original post

3 REPLIES 3

AHerr
Level 5
Employee Accredited Certified

Hi Simon,

You have captured the installation of your environment very clearly.
There are a couple of requirements for CPS.

Frist, if you have trusted ssh setup, then you can use installvcs -fencing to connect from your cluster to a CPS.  A single CP Server is supported but only for lab environments.  Anything that relies on fencing should have at least 3 coordination points.  Without trusted ssh, you can setup a relationship with CPS by manually registering the cluster and nodes.

Second, there was a change in the previous release that required LLT/GAB to be installed and configured to start VCS, even on a single node.  We are working on removing this requirement in a future version of the product.

If you have additional questions, please let me know.

 

Also, as a side note:  If you are interested in learning about the new features coming out in the next release of Storage Foundation and Cluster Server, please apply to the beta group:

https://www-secure.symantec.com/connect/groups/storage-foundation-and-veritas-cluster-server-60-beta-program



Regards,
Anthony

szhargrave
Level 3

Thanks Anthony

Absolutely I will be configuring 3 CPS before production, I'm using 1 just to get things working in our PoC.  We have shared keys configured so I'm using the install script with -fencing to configure.

So what's the recommendation for setting up a Standalone VCS server running CPS if llt/gab have to be running?  When I install VCS on the single machine, it says you can proceed without GAB and LLT which is what I was doing: -

If you plan to run VCS on a single node without any need for adding cluster node online, you have an option to proceed without starting GAB and LLT.
Starting GAB and LLT is recommended.

Do you want to start GAB and LLT? [y,n,q] (y) y

If I say no then I'm asked to configure heartbeats etc.  Presumably for this to work I need 2 NICs in the server (even though it won't really heartbeat?)  Can I use loopback?  Are there any implications at any stage that could put the standalone cluster at risk of self-fencing, for example if a NIC goes down will GAB panic the kernel?

I'm keen to avoid adding the risk to the other services running on a system that has had CPS bolted on due to self-fencing.

Is there a paper that documents how to install standalone VCS running CPS including how to configure LLT/GAB since it is required?

Thanks again, Simon

mikebounds
Level 6
Partner Accredited

I think the VCS config wizard allows you to configure one heartbeat, but if not you can amend manually.  For most operating systems you just need to create:

/etc/llthosts containing:

 0 nodename

/etc/llttab containing:

 

 set-node /etc/VRTSvcs/conf/sysname
set-cluster ClusterId
link link1 NicDevice - ether - -
 
 
 /etc/VRTSvcs/conf/sysname containing:
 nodename
 
/etc/gabtab containing:
 /sbin/gabconfig -c -n 1 
 
Linux is exception (which you maybe using if you using vmware) which in addition requires you to edit /etc/sysconfig/vcs and change line "ONENODE=yes to ONENODE=no
 
To start llt and gab, you can reboot, or start manually using:
lltconfig -c and /sbin/gabconfig -c -n 1 (or source /etc/gabtab)
 
Note when you start VCS, you do not need to use "hastart -onenode" as the onenode flag means don't use llt and gab, not you have onenode.
 
Mike