Shrinking a Pnfs Database

All Pnfs Databases are Gdbm Files compatible with gdbm version 1.7.3 and higher. The specification of gdbm claims that a database never shrinks automatically when deleting records, BUT unused space is reused before the database is extended.
So, under normal conditions it should not become necessary to shrink a pnfs database because filesystems tend to increase over time.
But if you really need the space you may write your own small tool which uses the gdbm_reorganize function. Make sure the dbserver for the reorganized database is not running. Or you may use some pnfs tools to shrink the database.

Using PNFS tools to shrink a database

The available tools to modify the database were not intended to be used by a user directy, with the result, that they are neither convenient nor for easy to use.

Extend your PATH to have the the necessary tools available. You need $pnfs/tools and $pnfs/dbfs. $pnfs is the pnfs variable in /usr/etc/pnfsSetup.
For the following procedure you need the database name as well as the database full path. Use mdb show to find whatever you need.

Shut down the database you want to modify.

          mdb disable <databaseName>
          mdb update

Get the database id of the desired database.

        md2tool <databaseFullPath> getdbid

Create a new empty database which will be the destination of the shrinked database.
```
        md2tool none create <databaseFullPath>.small <databaseID>
    
```
Take the <databaseID> from the getdbid command.
Copy the large database to the newly created one.
```
        md3shrink <databaseFullPath> <databaseFullPath>.small
    
```
The copy process may take awhile because it goes throu some unnecessary pnfs layers. For each internal record one line of output is written to stdout specifying the pnfsID of the record. So stdout should be written to /dev/null to speedup the copy process. If errors are reported to stderr, the database is somehow corrupted and it is not recommended to proceed further.

Backup or remove the large database and rename the small one.

        mv  <databaseFullPath>  <databaseFullPath>.backup
        mv  <databaseFullPath>.small  <databaseFullPath>

Enable the database again.

        mdb enable <databaseName>
        mdb update