How to delete all data related to a specific filespace?

[tivolimistery]

Hi All,

we are going to destroy an HSM managed GPFS file system.
But, first, we would like delete all data (about 1PB) related to this GPFS file system and free up a lot of tapes (assigned to the primary storage pool and to the copy storage pool).

This is our first "massive deletion" and I would ask you some suggestion about the right way to do it.

In particular, I would know if I have to delete all files from file system and wait for reclamation effects or:

- simply remove related filespace from TSM server;
- delete related storage pool from TSM server;
- delete each single tape from storage pool.

Thank you in advance.

 

[marclant]

- simply remove related filespace from TSM server;
- delete related storage pool from TSM server;
- delete each single tape from storage pool.

You don't need to delete the volumes if you delete the filespace and vice-versa.

You can do it in one of 2 ways.
Method 1 is if you don't know which volumes are owned by that node
1 - delete the filespace(s)
2 - wait for reclamation to run (both onsite and offsite) or kick it off manually
3 - delete the stgpool(s) if empty
This method will have less commands because don't need to touch every volume


Method 2 if you know which volumes are vor that node
1 - delete each primary volume individually with the option discarddata=yes
2 - delete the stgpool(s)
This method will have a large number of commands because do need to touch every volume.

 

[tivolimistery]

Thank you very much for your support.
So (just for confirmation), if I understand we can do it in one of 2 ways you suggested or deleting all files from GPFS file system waiting for reclamation, right?

Thank you again.
Have a great day

 

[marclant]

deleting all files from GPFS file system waiting for reclamation, right?

That would work too, but may take longer. Deleting the filespaces is likely the quickest way.

 

[teeraw]

You don't need to delete the volumes if you delete the filespace and vice-versa.

You can do it in one of 2 ways.
Method 1 is if you don't know which volumes are owned by that node
1 - delete the filespace(s)
2 - wait for reclamation to run (both onsite and offsite) or kick it off manually
3 - delete the stgpool(s) if empty
This method will have less commands because don't need to touch every volume


Method 2 if you know which volumes are vor that node
1 - delete each primary volume individually with the option discarddata=yes
2 - delete the stgpool(s)
This method will have a large number of commands because do need to touch every volume.

Do any ways impacted to plenty of db log will fill-in during processing?
And, how about interval time to do this by 2 ways?

 

[marclant]

Do any ways impacted to plenty of db log will fill-in during processing?
And, how about interval time to do this by 2 ways?

It won't be hard on the active log because those will all be relatively short transactions. Depending on how many million files there are to delete, it could potentially have an impact if the archive log is small and the number of files to delete is extremely large.

And, how about interval time to do this by 2 ways?

It depends. Option 1 is much less work for you, less typing, just kick off the delete and grab a coffee. Option 2 is more work for you as you have to issue every delete command manually unless you are good as scripting. Option 2 could potentially be faster because you can delete multiple volumes in parallel, but not too many as to not overload the server. Can't tell you how many in parallel your server can handle, too many variables, start small and increase slowly if you feel it can handle more. Thins to monitor before adding more: CPU, memory, DB/log disks response time, and archive log free space.