Re: ufsstat - testers / feedback wanted!

Poul-Henning Kamp wrote:
> In message <434EB85A.9030308_at_centtech.com>, Eric Anderson writes:
> 
>>Poul-Henning Kamp wrote:
> 
> 
>>>I would far to prefer to see a generic method that would work
>>>on all filesystems types, rather than a ufs specific tool.
>>>
>>>If you added counting code to the VOP_FOO() expansions, you should
>>>be able to gather statistics for all filesystem types.
>>>
>>>The collection could be controlled by a mount option if the impact
>>>is too big.
>>
>>That's an idea I thought of as well, but wanted to start small, since I 
>>really am new to C programming.  Now that I've done this part, I'll look 
>>at the various vop_* pieces and see if it's something in my grasp.
> 
> 
> I'd prefer for us to not commit a ufs-only solution, that would be
> enough to hold off a filesystem independent solution.

True, I see your point.

>>From what it sounds like, you'd also like to see per-mount point stats, 
>>but from the vfs layer, right?   If that's true, then do you have any 
>>suggestions on how to store the statistics for each mounted fs?
> 
> 
> In struct mount ?

Ok - so I'd place my struct vfsstats in the mount struct, so each fs 
records it's own stats if the mount option is enabled.  (reiterating to 
make sure I'm understanding correctly)

Also - is it best to increment the counters at the beginning of the 
operation, end, or something else?  For ufsstat, I placed it at the end, 
but for some operations, if they fail, the do not get recorded.  This 
may be what we want - only record completed ops, or maybe we want to 
record all requested operations, completed or not.  What are thoughts on 
this?

Thanks..
Eric


-- 
------------------------------------------------------------------------
Eric Anderson        Sr. Systems Administrator        Centaur Technology
Anything that works is better than anything that doesn't.
------------------------------------------------------------------------