[I'll Cc the lvm-devel list on this]
Neil,
what you are stating here is a bit distressing,
since i am using LVM over raid5.
I'd like to help in solving this kind of problems...
On Mon, Feb 26, 2001 at 01:04:23PM +1100, Neil Brown wrote:
> On Thursday February 22, [EMAIL PROTECTED] wrote:
>
> [Sorry about the delay in replying, but as this mail was not addressed
> to me ("neilb" did not appear in headers) it didn't get a very high
> priority in my mail reading. Please, when sending email to me,
> address it to me so my filtering can work. Thanks]
>
> >
> > hello,
> > looking at your patches i was trying to understand patch-J-r5align
> > does it mean that that particular request is not handled by the
> > stripe cache?
>
> Yes. A request that is not properly aligned cannot be handled by the
> stripe cache. I'm not certain exactly what will happen, but I suspect
> that it would mostly work but would occasionally cause data corruption
> if a differently aligned request for a nearby stripe were written at
> the same time. Occasional corruption is not something that I want!
>
> >
> > besides printing the warning (wouldn't it be better to print it only once
> > or once in a while, instead of at every request)
>
> Arguable it would be better to rate-limit the printing of the message,
> but really, I don't want anyone to use raid5 in a situaiton where
> requests might not be aligned, and I want them to get a very clear
> message this it doesn't work.
>
>
> >
> > anyway once you properly configure your LVM system, misaligned requests happen only
> > during vg operations, which are rare.
i actually goofed on this, i was thinking something else...
what lvm does during vg operations is reading data in non 4k sized blocks
(same happens when using dd on the md device)
Feb 24 21:56:30 Moskowskaya kernel: raid5: switching cache buffer size, 4096 --> 1024
Feb 24 21:56:30 Moskowskaya kernel: raid5: switching cache buffer size, 1024 --> 4096
Feb 24 21:56:30 Moskowskaya kernel: raid5: switching cache buffer size, 0 --> 4096
> But even once is too often if it could corrupt your data.
>
> I don't know much about the details of LVM, but if you can create
> partitions that are not on a 4K boundry in the real device, then you
> could certainly create a situation with lots of misaligned requests.
i don't think this is the case since LVM divides the physical disk
in chunks called PE sized "from 8KB to 512MB in powers of 2 and unit KB"
> I hope to remove this restriction eventually, but until then, I want
> people to be able to know when potentially bad things are happening.
>
> NeilBrown
>
> -
> To unsubscribe from this list: send the line "unsubscribe linux-raid" in
> the body of a message to [EMAIL PROTECTED]
--
Luca Berra -- [EMAIL PROTECTED]
Communication Media & Services S.r.l.
/"\
\ / ASCII RIBBON CAMPAIGN
X AGAINST HTML MAIL
/ \
-
To unsubscribe from this list: send the line "unsubscribe linux-raid" in
the body of a message to [EMAIL PROTECTED]
