Lustre devel - Feb 2007 - How to identify Lustre tunables in /proc?

I''m working on a script to take a snapshot of Lustre tunable values  
in /proc.

I''d rather not hard code a list of files that will have to be updated  
every time a tunable is added or removed.

One strategy that has been suggested to me is to look for files that  
contain "min" or "max" as part of the name. I think there
some that
this strategy will miss, though, like "lru_size" in the namespaces
area.

The approach I have tentatively settled on is to look for files with  
at least one write bit turned on.

Is this a reasonable approach? Will it find all the tunable files?  
Does it select files that are not tunable? Can anyone suggest a  
better way of doing this?

Thanks,
--
Tom Barron
tbarron@ornl.gov
865.576.1620
Oak Ridge National Laboratory

Tom Barron wrote:
> I''m working on a script to take a snapshot of Lustre tunable
values in
> /proc.
>
> I''d rather not hard code a list of files that will have to be
updated
> every time a tunable is added or removed.
>
> One strategy that has been suggested to me is to look for files that 
> contain "min" or "max" as part of the name. I think
there some that
> this strategy will miss, though, like "lru_size" in the
namespaces area.
>
> The approach I have tentatively settled on is to look for files with 
> at least one write bit turned on.
>
> Is this a reasonable approach? Will it find all the tunable files? 
> Does it select files that are not tunable? Can anyone suggest a better 
> way of doing this?
Using the permissions bits is a reasonable way to go.  Tunables would 
have root rw permission.  There are some write-only files that perform 
specific actions; these are not tunables and should probably be avoided 
by your script.

cfs21:~# ll /proc/fs/lustre/mds/lustre-MDT0000/
total 0
-rw-r--r--  1 root root 0 Feb  1 16:04 atime_diff
-r--r--r--  1 root root 0 Feb  1 16:04 blocksize
--w-------  1 root root 0 Feb  1 16:04 evict_client
dr-xr-xr-x  2 root root 0 Feb  1 16:04 exports/
-r--r--r--  1 root root 0 Feb  1 16:04 filesfree
-r--r--r--  1 root root 0 Feb  1 16:04 filestotal
-r--r--r--  1 root root 0 Feb  1 16:04 fstype
-rw-r--r--  1 root root 0 Feb  1 16:04 group_acquire_expire
-rw-r--r--  1 root root 0 Feb  1 16:04 group_expire_interval
--w-------  1 root root 0 Feb  1 16:04 group_flush
--w-------  1 root root 0 Feb  1 16:04 group_info
-rw-r--r--  1 root root 0 Feb  1 16:04 group_upcall

On Feb 01, 2007  17:01 -0500, Tom Barron wrote:
> I''m working on a script to take a snapshot of Lustre tunable
values
> in /proc.
Is this for data gathering purposes (e.g. record parameters for a test),
or is it to restore these values at a later time?  If it is the latter,
then in 1.6 this can be done with "tunefs.lustre" in a persistent
manner.

Cheers, Andreas
--
Andreas Dilger
Principal Software Engineer
Cluster File Systems, Inc.