Jim Paris wrote:
FWIW a month or so ago I used mdadm + 2.6.4 and constructed a 5x250Gb RAID 5 array with one drive missing.And here's a trace showing problems even without a filesystem.
Writing data near the end is fatal. This does not happen if I write
to the first 1G of the array.
Sorry, that test was bogus, and I needed to learn how to use mdadm. I haven't actually managed to cause corruption on a raw device with no
filesystem. However, copying a single 200MB file onto Reiserfs will
cause corruption. It takes a lot more work (e.g. actually copying an
installed system onto it), but XFS shows eventual corruption as well,
so it's not specific to the filesystem type.
I see no problems if I start the array with a complete set of disks; the corruption only happens if it starts degraded (tested with both 1 and 2 disks missing, and with the missing disks being at both the beginning and the end). This happens on Linux 2.6.3 and 2.6.7, with mdadm 1.5.0 and 1.4.0, with and without CONFIG_LBD. RAID-5 works correctly in all tested configurations. I have tried varying the number of disks in the array.
When I added the missing drive and reconstruction had finished I had fs corruption.
I used the reiser tools to fix it but lost an awful lot of data.
I reported it in detail here [http://marc.theaimsgroup.com/?l=linux-raid&m=108687793611905&w=2] and got zero response <shrug>
Since then it's been fine.
I don't have much faith in it though ;)
David
PS around that time there was a patch [http://marc.theaimsgroup.com/?l=linux-raid&m=108635099921570&w=2]
for a bug in the RAID5 resync code.
it was only for the raid5.c
It doesn't look raid5 algorithm specific ... :)
-
To unsubscribe from this list: send the line "unsubscribe linux-raid" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
