[Dirvish] using --inplace ?

[Paul Slootman]

On Fri 01 Sep 2006, Dave Howorth wrote:
> Paul Slootman wrote:
> >I was chekcing what rsync does in a typical dirvish run. Basically lots
> >of stat()s of the file in the destination directory, stat()s of the file
> >in the link-dest directory, then (when the file doesn't yet exist) an
> >open() for writing in the destination directory with a temporary
> >filename, and then a rename() at the end.
> 
> What does it do in the case where the file does exist but has changed?

The way dirvish works, the destination directory is always empty; the
file may exist in the --link-dest directory, but then rsync will also
write first to a temp filename and then rename it without the --inplace
option.

> >I considered that the use of a temp name and then a rename doesn't
> >really make much sense in dirvish usage. Since rsync 2.6.3 (30 Sep 2004)
> >it knows about --inplace.
> 
> But according to the man page, I think we can only use it from 2.6.4 
> onwards:
> "Prior to rsync 2.6.4 --inplace was also incompatible with 
> --compare-dest and --link-dest."

Indeed, it complains about protocol 29 being needed, as I discovered
this morning :-)  A good way of detecting systems with outdated rsync
versions, BTW; those systems that failed to backup last night now have
rsync 2.6.8 installed.

> >This will save 2 system calls per new or
> >changed file. Doesn't sound like much, but it's a lot of metadata
> >updates that will impact journalling filesystems.
> 
> I'm all for saving system calls (and context switches :) so this sounds 
> like a good idea.
> 
> >Can anyone think of a reaon why using it would be a bad idea, besides
> >the fact that an older rsync version may be in place (pun intended),
> >in which case it should be upgraded anyway?
> >
> >I've actually modified dirvish already on one of the systems I have
> >dirvish running on to use --inplace, I'll check tomorrow how it went.
> 
> It always comes down to how to know something didn't break, doesn't it?

Indeed; however, luckily dirvish is fairly anal about checking results,
so I'm fairly confident that nothing will break without being noticed.

I'll continue testing.


Paul Slootman