Whoops. Forgot to send this to the list.
---------- Forwarded message ----------
From: Eric A <[email protected]>
Date: Mon, May 12, 2008 at 12:50 PM
Subject: Re: barriers off by default?
To: Eric Sandeen <[email protected]>
On Mon, May 12, 2008 at 11:53 AM, Eric Sandeen <[email protected]> wrote:
> > From reading the documentation, I was under the impression that write
> > barriers don't always do what they're supposed to do.
>
> Which documentation?
>
Documentation/block/barrier.txt
There's a section that starts:
"* Error handling. Currently, block layer will report error to upper
layer if any of requests in an ordered sequence fails. Unfortunately,
this doesn't seem to be enough. "
Cheers,
Eric
Eric A wrote:
> Documentation/block/barrier.txt
>
> There's a section that starts:
>
> "* Error handling. Currently, block layer will report error to upper
> layer if any of requests in an ordered sequence fails. Unfortunately,
> this doesn't seem to be enough. "
And ends:
"As the probability of this happening is very low and the drive should
be faulty, implementing the fix is probably an overkill. But, still,
it's there."
I'm not sure it's an argument for disabling barriers in general. :)
-Eric
On Mon, May 12, 2008 at 1:00 PM, Eric Sandeen <[email protected]> wrote:
> Eric A wrote:
> > Documentation/block/barrier.txt
> >
> > There's a section that starts:
> >
> > "* Error handling. Currently, block layer will report error to upper
> > layer if any of requests in an ordered sequence fails. Unfortunately,
> > this doesn't seem to be enough. "
>
>
>
> And ends:
>
> "As the probability of this happening is very low and the drive should
> be faulty, implementing the fix is probably an overkill. But, still,
> it's there."
>
> I'm not sure it's an argument for disabling barriers in general. :)
I agree; it's just something that's been bugging me since I started
searching for a solution for zfs-fuse. Maybe performance is the
reason? Flushing the write cache is kind of slow.
Cheers,
Eric
Eric A wrote:
> I agree; it's just something that's been bugging me since I started
> searching for a solution for zfs-fuse. Maybe performance is the
> reason? Flushing the write cache is kind of slow.
So is journal=ordered, so is journaling at all for that matter.
But if 32MB of your just-flushed-from-log metadata evaporates on power
loss, it was all a bit of a waste IMHO...
-Eric