cancel
Showing results for 
Search instead for 
Did you mean: 

VxFS checkpoints in Solaris non-global zones?

AAlmroth
Level 6
Partner Accredited

We are looking about for ways to use VxFS checkpoints for a file system mounted in a non-global zone, but as the VRTSvxfs package is hollow, no vxfs related commands exist in the zone.
Preferrably we would like to have NBU to control it, but any scripted way will be considered as well. Also NBU would have to run within the zone to integrate with the application (pre/post backup scripts).

I've been searching on the topic, but haven't really found anything of use. My immediate conclusion was that it isn't supported... :(
From what I understand, a possible, but not really good solution, would be to mount the file systems in the global zone and then share it to the zone. But as we run VCS on top, we would like to control this with the DiskGroup and Mount resources to present the volumes within the zones only.

Any ideas?

/A

1 ACCEPTED SOLUTION

Accepted Solutions

AAlmroth
Level 6
Partner Accredited
An update: The VxFS filesystems are present in the global zone but not visible with the df command, but they are present when running mount command. This enables us to run fsckptadm in the global zone for the file systems "mounted" and used within the zones.

The downside is that, as we run the applications on a 3 node VCS cluster, we don't know for sure on which physical node the application is running at the moment without some scripting. We are looking at using a NBU policy which starts the job on all three nodes, and a pre-snapshot script, checking with VCS whether the application service group is on that node. If it is not, the script does nothing (and the backup will be of zero bytes). If the application is present, an ssh script session in to the zone, run the quisce, and then checkpoint/backup continues in the global zone. Another downside is that the backup will be cataloged as run on the physical node, and not the virtual name used by the application, but we think with documention, the ops staff can work around this upon potential restores.

So: problem solved...

It would certainly be a very nice feature in NBU if it could be made zone aware and integration with the global zone when using VxFS checkpoints/snapshots... Sort of an off-host backup but local on the physical box... :)

/A

View solution in original post

5 REPLIES 5

Gaurav_S
Moderator
Moderator
   VIP    Certified

I believe you are right... you will need to take the checkpoints from the global zone.... I don't think fsckptadm is zone aware....

If you are running VCS, you can configure VCS in such a way to get DG/Mount & Zone resource on the top in the service group to failover ...

Gaurav

AAlmroth
Level 6
Partner Accredited
An update: The VxFS filesystems are present in the global zone but not visible with the df command, but they are present when running mount command. This enables us to run fsckptadm in the global zone for the file systems "mounted" and used within the zones.

The downside is that, as we run the applications on a 3 node VCS cluster, we don't know for sure on which physical node the application is running at the moment without some scripting. We are looking at using a NBU policy which starts the job on all three nodes, and a pre-snapshot script, checking with VCS whether the application service group is on that node. If it is not, the script does nothing (and the backup will be of zero bytes). If the application is present, an ssh script session in to the zone, run the quisce, and then checkpoint/backup continues in the global zone. Another downside is that the backup will be cataloged as run on the physical node, and not the virtual name used by the application, but we think with documention, the ops staff can work around this upon potential restores.

So: problem solved...

It would certainly be a very nice feature in NBU if it could be made zone aware and integration with the global zone when using VxFS checkpoints/snapshots... Sort of an off-host backup but local on the physical box... :)

/A

Gaurav_S
Moderator
Moderator
   VIP    Certified

Hello,

good to hear that.... probably you can post this as an Idea in the NBU ideas on Connect forum...

Some of developer may respond to you on this :)

I didn't get one point though, you say that VxFS filesystems are present in global zone & they are not visible by df command while they are visible with mount command ... ! I wouldn't imagine that, FS should be visible by df command as well...

Gaurav

AAlmroth
Level 6
Partner Accredited
The difference between df and mount caught me too, but as we could see the mounted file system in the global zone with the mount command, we tried fsckptadm, and it worked. I read about this deep down somewhere in the manuals, but can't find tthe reference now.

The VCS Mount agent is zone aware and it doesn't run in the Container, but in the global zone from what I read in the bundled agent guide. This could explain why the mount point is only visible in the zone with the df command.

I guess if we had used loopback devices instead, the mount points would be visible in the global zone as well as in the non-global zones. This is true for the root file system for each zone, as those are mounted in the global zone.

Will start working on a RFE for NBU now... :)

/A

g_lee
Level 6
Re: "discrepancy" between the df and mount output - it sounds like the filesystem was delegated to the non-global zone (as opposed to just added as a loopback mount) so the ngz has exclusive access to that mount - this is why the filesystem is not seen in df output, but is seen in mount (ie: this is the expected behaviour).

Doc reference here - VxFS 5.0MP3 (Solaris) Administrator's Guide:
http://sfdoccentral.symantec.com/sf/5.0MP3/solaris/html/vxfs_admin/apds03.htm