[Linux-cluster] gfs_tools reclaim

[Robert Peterson]

Ananda Bhattacharya wrote:
>
> Hi I was wondering what the differences between gfs_tool reclaim  and 
> gfs_tool shrink?
>
> And secondly is there a need for these commands to be run?
>
> -Anand
>
Hi Anand,

This is how I think it works (and as always, I could be wrong):

gfs_tool reclaim takes GFS metadata on disk that is no longer being 
used, and
   converts it back to free space.  That's normally not an issue because 
those unused
   metadata blocks will usually be reused anyway.  For example: create 
files A then
   file B.  When you delete file A, you have an unused metadata block 
for A's inode,
   which has been deleted.  If you allocate more space to file B, it 
normally won't
   reuse A's deleted inode because it's "freed metadata" rather than 
"free space."
   However, if your file system is out of "free space", I think a 
reclaim automatically
   happens and it will reuse that space anyway.  Now let's say that you 
create file C.
   GFS can now reuse the freed metadata (inode) formally used for A, to 
keep the
   new metadata (inode) for C.  So in either case, the deleted metadata 
blocks will be
   reused when GFS sees fit, and therefore, you shouldn't need to run 
this command
   unless you have special needs.

   Incidentally, GFS2 no longer has the distinction between "free space" 
and "free
   metadata space", so free space is free space regardless of whether it 
was previously
   used for metadata or user data.

gfs_tool shrink throws out unused gfs locks from memory (cache).  
Ordinarily, gfs
   manages its own glocks according to the tuning parameters.  So 
there's really no
   need to do this unless you have a special problem regarding the 
caching of gfs locks.

Regards,

Bob Peterson
Red Hat Cluster Suite