Last month I had to free a filesystem in HP-UX, took around 12 hours to
release the freespace, we tried to speed up by umounting the filesystem,
reboot the server but still.
On 4/21/06, Kevin Closson <kevinc_at_polyserve.com> wrote:
>
> >>>My understanding was that he tested with fuser or lsof and
> >>>found no other users using that file? If that is so, then OS
> >>>may be little slow in recognizing that the situation has changed.
>
> Impossible. There is no "sweep up" demon for this (as it were) in
> any Unix derivation. There is an OS thread in the bowels of unlink(1)
> waiting on a kernel semaphore for the lock on the inode held
> (incremented)
> by the processes that have file descriptors (or mmap(1)s too of course)
> on the file. They close explicitly (e.g., close(1)) or implicitly
> (e.g., exit(1) exec(1)) and the ref count decrements and when
> the count is right the original poor lingering thread in
> unlink(1) gets his deed done.
>
> I remember one really cherry scenario where (a lot like this
> installments on this thread) a customer with a US $1,000,000
> SMP was looking for the sqlplus executable. He blamed the
> daylights out of the OS because he didn't understand the semantics
> of UFS file manipulation. Lo and behold, we started digging in and
> found that his Oracle 7 Oracle home had 1 file in it and that
> was ${ORACLE_HOME}/bin/oracle. Can anyone guess how that scenario
> came to be?
>
> rm -fr *
>
>
>
>
>
> >>>--
> >>>Mladen Gogala
> >>>http://www.mgogala.com
> >>>
> >>>--
> >>>http://www.freelists.org/webpage/oracle-l
> >>>
> >>>
> >>>
> --
> http://www.freelists.org/webpage/oracle-l
>
>
>
--
http://www.freelists.org/webpage/oracle-l
Received on Fri Apr 21 2006 - 02:10:55 CDT
