Re: Race condition in storeClose()

On Wed, Nov 22, 2000, Brian Degenhardt wrote:

>
> I'm getting lots of these:
>
> 2000/11/22 13:24:02| WARNING: failed to unpack meta data
> 2000/11/22 13:24:11| WARNING: failed to unpack meta data
> 2000/11/22 13:24:13| WARNING: failed to unpack meta data
>
> Isn't that the sign of a partially written object?

Hrm. You shouldn't get those with a partially written object - you will
only get those if you fail to write the metadata out correctly.

Ok, its good to know that cancelling the IO is the right thing to do.
I guess I'd better clue myself up on the aufs code and abort things
in the right order ..

For those out there using diskd, diskd will require the outstanding IOs
to be placed on a queue/list so they can be tagged as "aborted".

And yes Andres, we have to allow the reads to be aborted whilst allowing
the writes to complete.

adrian

-- 
Adrian Chadd			"God: Damn! I left pot everywhere!
<adrian@creative.net.au>	  Now I'll have to create Republicans!"
				    - Bill Hicks