Behavior of failed duplication job

[NB-OPS]

Hi,

NetBackup 7.5 on Solaris.

We back up from data source to VTL and then it gets duplicated to tapes using SLP.

Scenario:
Prior to a duplication job starting, current tape status are:
Tape 1: has images from the previously successful duplications but it is not full
Tape 2: has no image; in scratch pool
Tape 3: has no image; in scratch pool

A duplication job based on SLP is launched and the tapes are used in the order of Tape 1, 2, and 3.

Duplication from VTL to physical tape library uses Tape 1 until it's full. Then it uses Tape 2 and it writes until full as well. Then it uses Tape 3 but in the middile of writing to Tape 3, the duplication job failed for whatever reason (I wonder: Is out-of-tape considered to be a failure?)

At this moment, what happens to Tape 1, 2, and 3 from the "images on media" and catalog's point of view?

My thought is:
For both images on media and in the catalog:
  Tape 1: has images on itself from the previously successful backups AND data from this dup job up until it got full; corresponding entries in the catalog are not removed.
  Tape 2: has data from this dup job and remain as 'full'; corresponding entries in the catalog are not removed.
  Tape 3: has no good data and goes back to the scratch pool; corresponding entries in the catalog are removed.

NetBackup support said because the duplication job has failed, even though Tape 1 and Tape 2 contain data from this dup job, all data on all tapes for this failed dup job are removed from the catalog AND both Tape 2 and Tape 3 go back to the scratch pool while Tape 1 is not be marked full.

If that's the case, before the duplication job fails and I took out Tape1 and 2 because they were full at that moment, Tape 1 and 2 are actually useless even though the contain images on the physical tapes. So catalog and physical media have mismatched information?

What is really happening? Could someone shed some lights on this?

Thanks!

[RiaanBadenhorst · Accepted solution]

Any images that got duplicated are complete, and they'll remain on those tapes. It does it one by one. You can verify this by looking at the images listed in the job details and checking the catalog.

If its was one giant backup job that spanned 3 tapes, then yes, it would release those media.

[NB-OPS]

Riaan,

Thanks for your input. I guess I did not make myself clear. The way I understand it, a backup job corresponds to a backup image. When an image is duplicated and requires multiple media, each media stores a fragment of the image. Based on what you said, if the duplication fails, all fragments relevent to this duplication job on all media used are cleared and the catalog is updated accordingly?

If this is the case and if I take out the full media (Tape 1 & 2 above) while the duplication job is writing to Tape 3, the catalog will be updated and Tape 2 will be moved to the Scratch pool while Tape 1 is marked as Active but not Full (ie. the state before this duplication job was started)?

Thanks,

[sdo · Accepted solution]

Yes and yes. But media are only made free/scratch after media cleanup has occurred, which usually takes place once every 12 hours.

[RiaanBadenhorst]

Correct, and as SDO stated.

[sdo]

You can see when the major cleanup events occur using:

# bperror -hoursago 9999 -U | grep -i "clean"