Hi there,
This does not fail on my quad CPU test box with CVS code. I seem
to remember a problem with append writes from a long time ago, but
not the specific fix [ -- there was a specific fix, I just can't
remember what it was ;-) ]. You might be able to dredge the CVS
logs if you need to stay at that version, otherwise I'd recommend
upgrading to more recent kernel code (2.4.17 is already dated).
FWIW, I tried these different test sizes:
-rw------- 1 root root 2.9M Sep 4 16:33 /mnt/xfs1/testfile.1176
-rw------- 1 root root 20M Sep 4 16:35 /mnt/xfs1/testfile.1183
-rw------- 1 root root 74M Sep 4 16:50 /mnt/xfs1/testfile.1233
cheers.
--
Nathan
On Wed, Sep 04, 2002 at 03:21:33PM +0900, hatakeyama wrote:
> Hi,
>
> I found an issue on dual cpu machine.
>
> Description:
> the xfs couldn't follow append-write semantics correctly.
> After the file was opened with O_APPEND flag, then seek the
> file position of EOF before each writing. After several seconds,
> we found that the data was not written one by one, that means
> some of them have been overwritten.
>
> Configuration:
> VERSION:2.4.17(xfs-0225-all.patch)
> Dual Intel pentium III 900
> 2G RAM
>
> --------
> You can reproduce the issue with the attached files as below:
>
> chdir to a xfs, run
>
> # test
>
> kill this process, the files size exceed 40 MB.
>
> and verify as below:
>
> # test2 testfile*
> --------
>
> Regards,
> hatake
|