Re: [RFC] ext3 multi-mount protection

On Jul 14, 2006  13:06 +0100, Stephen C. Tweedie wrote:
> I'm not sure that the filesystem is the right place to do this.  It will
> probably work OK for lightly loaded systems and to protect against some
> common administration accidents, but for real protection in a clustered
> environment we need to care about a lot more failure scenarios.  One
> pretty common failure mode is temporary timeouts on a node, either a
> device driver timing out and suspending IO for some seconds, or
> something like VM pressure causing bits of the kernel to effectively
> lock up for a while.  A 10-second kick-the-sequence-number loop isn't
> going to protect against those.

What would you propose instead?  The problem is that all of the existing
HA tools have failures that have bitten us here, so we just can't trust
them.

> If we do decide we want this, then I don't think that the kernel is the
> right place for it.  That's for two reasons --- first, it's never going
> to get all the hard cases (timeouts etc) right anyway, so reducing the
> point of having it in the kernel;

Though not eliminating it entirely, because it does avoid some latency
and swapping issues, and guarantees that the daemon doesn't get killed
by OOM, or fails to start or is otherwise misconfigured.

> secondly, we need it for e2fsck and tune2fs just as much as for mount.

By all means yes, e2fsck and tune2fs (to a lesser degree, since it is
generally safe to run on a mounted fs) need to be able to understand
this INCOMPAT feature, and as a result will also understand its
semantics.  This was one of the reasons for adding an INCOMPAT flag,
and the other was so that unaware kernels won't be able to mount such
filesystems and trample on them.

Cheers, Andreas
--
Andreas Dilger
Principal Software Engineer
Cluster File Systems, Inc.



-------------------------------------------------------------------------
Using Tomcat but need to do more? Need to support web services, security?
Get stuff done quickly with pre-integrated technology to make your job easier
Download IBM WebSphere Application Server v.1.0.1 based on Apache Geronimo
http://sel.as-us.falkag.net/sel?cmd=lnk&kid=120709&bid=263057&dat=121642