Wow, I had no idea this was supported.  But it looks like we do, given the instructions later in that TechNote:

(quoting begins)
NetBackup Master Server or Media Server (disk backups only) can be installed in a local zone (whole root).

Hi CRZ,

Finally I had the time to test. Here's what I've done:

1. Created a full root zone (used create -b in zonecfg)
2. Manually installed with pkgadd the packages VRTSicsco VRTSpbx inside the zone
3. from the global zone I have installed VRTSat and VRTSaz (must be installed from there as the pkginfo has SUNW_PKG_ALLZONES=true)
4. then I've executed the install script from the CD

Everything was installed properly and I've choosen the local zone name as the master server and the EMM database name. When it asked me about the media server I have answered no, not configure any at the time.
After installation finished successfully I managed to connect to the master server using jnbSA.

I will now configure a media server and check backup and restore. I expect everything to work perfectly.

Thanks for the help!

Cheers,
Miro.

Hi yktglr,

Sorry for late answer I've just got way too busy with other things.
Unfortunately you can't control the robot from the zone. Physical devices shouldn't be managed by a non-global zone. You can mount a raw disk, but I'm afraid you wont be able to do SCSI reservations on any tape drive (to lock the drive). So configuring disk stage also doesn't make sense.
You should use a separate Media Server.

Cheers,
Miro.

Hi Crz,

I'm not sure if you're still reading this post, but I would have an interesting question.
Ok we have the Master running in a zone. Is it supported to configure this zone to be a Sun Cluster "Failover local zone"?
The Failover zone switches from one node to another. So it takes all the disks and IPs with itself. So theoretically this should work without running the cluster_config script. As the hostname and the master server name will always be the same, as well the IP address.

Thanks,
Miro.

Hi Miro,

I honestly have no clue what the answer to that question is, but it's awfully interesting.  :)

In theory, and assuming I'm correctly interpreting what you've said, NetBackup wouldn't see your master as part of a cluster - just as a "normal" server - so I would speculate that you're right about the failover behaviour...NetBackup wouldn't know any different, so why should it act any different?

I must stress "in theory" here, though.  I'm awful with hypotheticals and can't give you any definitive answer one way or the other.

I would tend to think this DOES fall under a supported configuration, however - if we support a master in the local zone, this configuration would seem to fall under that.

(DISCLAIMER: I may be totally wrong.  Please do not take my word for it.  If you really need to know, please open a case with support and get it in writing from an authoritative source.)

If anybody's tried this, please speak up and let us know how it's working (or not working)!

Hi Crz,

As far as I know the failover zone is a cluster resource on it's own. I might be able to test this setup later this month, but I'm not sure of that yet.
Anyway if somebody feels interested in this, then the sun cluster data service docs for HA containers (zones) can be found here.
The thing is that this data service agent has two options of which you can choose to monitor and manage applications inside the zone. One option is to use scripts like init scripts and the other is to use solaris Service Management Facility. Either way we have the chance to check for ports or any command outputs like bpdbjobs which can idicate us if netbackup is up and running so we can make decisions whether to fail over or not. Also we can provide start and stop scripts eg to run bp.kill_all if we want to stop netbackup, before we would failover the zone.
In this setup I think NBU (as you mentioned) doesn't see the difference between a failover zone or a single local zone as everything moves together.

Hope I can try it out soon. If I have any news I'll update you.
Thanks again Crz, I appreciate your input.

Cheers,
Miro.