[nylug-talk] resuming interrupted copy, kinda - or rsync?

[Ron Guerin]

Peter C. Norton wrote:
> On Fri, Oct 26, 2007 at 02:47:08PM -0400, jh wrote:

>> Don't journaled file systems prevent that from happening? Aren't all 
>> changes atomic?
> 
> Yes and no. An individual write may be complete, but that write is for
> a block or a group of blocks, and associated metadata changes (with
> journalled metadata, which ext3 has, but some filesystems haven't had
> in the past). However, that doesn't mean that the file was completely
> written at the time of the crash, just that the filesystems state
> wasn't corrupted because of the interruption.
> 
> Eg. if you hit ^c instead of having a power outage, it doesn't
> guarantee that the file being copied was fully copied, just that the
> last atomic operation (write, link, unlink, sync) on that file
> completed.
> 
> Also, the filesystem may not save you from underlying raid,
> controller, or memory issues especially during a cascading power
> outage.

I know of at least one high-profile "oops" and several smaller ones
which traced back to two issues, the first being that a lot of hardware
RAID doesn't work as claimed, and second that commodity hardware often
just isn't going to provide the support that your journaling file system
and other stuff really needs to have your back.

Since I started writing this, in a superb bit of irony, a client had a
RAID "oops".  Combine that with an "easy to use" Microsoft OS, and
you've got yourself one big bucket of misery.

- Ron