Changes in VCS 5.0 Solaris

[Gene_Henriksen]

There are several features to look for (be sure to read the Release Notes).

1) no more .stale file when you have the configuration open. If you set the cluster attribute "BackupInterval", VCS will save the current config, when open and changes are made, to main.cf.autobackup. When you close the config, it will save the running config and remove the autobackup copies. If you do "hastop -all -force" when you have the config open, the cluster will come back up with the main.cf.

2) RemoteGroup agent that can interact with a VCS service group on a remote cluster.

3) support for VLAN interfaces, IO Fencing in DMP mode, 5 levels of Service Group dependencies plus more.

Be sure to RTFM before trying all of these out.

There are some new agents and some old ones have been dropped (Service Group Disk Heartbeats is one, Disk and Disk Reservation - SCSI 2 type- )

Also Virtual Fire Drill, no relation to the Global Cluster Fire Drill, but it allows you to check to see if certain features (such as mount points) exist on other servers to more quickly verify that you could failover.

Have fun.

[Hywel_Mallett]

SFWHA features the BackupInterval attribute. I'm not sure if I'll use this. I like the failsafe-ness ( I just made that word up) of the previous method.
The RemoteGroup agent is there - this is for multiple cluster environments, so not for me.
Gene, can you expand a bit on VLAN support? There doesn't appear to be any reference to this in the SFWHA documentation.
Service Group dependencies - didn't this used to be just two levels? We actually need one service group to be dependent on two other service groups. We can't do this with 4.3MP1.
Haven't SCSI 2 type disk and mount agents been deprecated (but still included) for a little while now?
Also I can't find any reference to the Virtual Fire Drill. What you say it is certainly sounds good.

The Simulator now lets you delete simulated clusters. I put in an enhancement request for that. Nice to know Symantec are listening. :)

[Gene_Henriksen]

The VLAN support in the UNIX version allows you to monitor the VLAN devices. Here is the release notes writeup:
The NIC, MultiNIC, MultiNICA, and MultiNICB agents now support VLAN
interfaces. The agents do not configure the NICs, but can monitor them.
See the OS vendor�s documentation on how to configure VLAN on your host, and
ensure that the switch or router connected to such an interface is compatible
with your configuration. Both server-side and switch-side VLAN configurations
are supported.

You aren't the only one who asked for cluster deletion in the simulator. We had to add information to the VCS course on how to get rid of one and it was too #$$% complicated.

The 5 levels of service group dependency (up from 3) is more for support of parallel applications such as Oracle RAC.

VFD seems to be UNIX only in this release. Perhaps the next release of Windows.... There are some things that seem harder (or useless) to implement in Windows. MS would not allow us to introduce our file system into the Windows world, no UNIX vendor has felt that way, and so some of our volume/file system integration doesn't exist.

The Disk/DiskRes in UNIX was just dropped. In the 5.0 Windows version we dropped the Mount and DiskRes also. In order to use VCS you must use our SFW disk groups.

The BackupInterval is actually a very good thing. We have had problems over the years with users not closing the configuration and then a crash leaves the cluster down. This way, you still come up with the original config, while all your changes have been saved for you.

[Hywel_Mallett]

The BackupInterval - wouldn't you say this was a workaround for people not closing the config? If an admin hasn't closed the config when they've finished making any changes, I'd rather they get a virtual slapped wrist!
If the config was open because you've been making changes, and the cluster is shut down for any reason, then the original config may no longer be valid.
Maybe I'm just being picky ;)

[Gene_Henriksen]

It is largely a matter of trying to keep the clusters up as much as possible. Yes, if you always had to close the config or VCS would use rslapd (the imaginary remote slap daemon to make sure the admin does things correctly), then we could have .stale to protect the cluster config.

Apparently a larger percentage of users prefer the clusters come back up after a reboot with an open config.

[Hywel_Mallett]

Well then I think a larger percentage of users should get their procedures in order ;)

[Gene_Henriksen]

To mis-quote a famous English poet:

Ours is not to question which
Ours is but to do and bitch

[Hywel_Mallett]

VCS guru, literary expert, is there anything Gene doesn't know about? :)

[antonella_arche]

Hi there,

about the changes in VCS 5.0 for Solaris I have a couple of questions on building custom C++ agents: the library /usr/lib/libvcsagfw.so.2 has changed dimension compared to 4.1 version. It has increase its size from 3120kb to 3264kb.
Can I compile my 4.1 compatible agent using the 4.1 library and run using the 5.0 library? in other terms is the library in VCS 5.0 backward compatible?
This would be useful to me as some of my customers maybe using VCS 4.1 and others VCS 5.0 and I would like to produce a single agent binary.

[Gene_Henriksen]

Appendix A of the Cluster Server Agent's Development Guide is all the help I can give you on this. I am not a C++ programmer.

[antonella_arche]

It's not there but it was worth to try since this info in nowhere to be found in documentation. Thanks anyway.