On Sun, Aug 11, 2013 at 12:42 AM, <dlellis_at_rockwellcollins.com> wrote:
> Hi Johan,
>
> I think you are spot on on both issues - thank you.
You're welcome :-).
> The performance issue relates to SVN externals stem, in my case, from new
> connections for each http request for externals versus keeping an open one
> for non-externals. Each connection results in a second or so delay.
> Multiply this by hundreds and updates that would take seconds, now takes ten
> minutes. There's been quite a bit of discussion about pooling connections I
> believe, but I don't know the current state of this effort or if its in any
> plan.
Hm, there were indeed some discussions (and patch attempts) recently,
but I'm not sure of the current state either. There is an open issue
about this in the issue tracker:
http://subversion.tigris.org/issues/show_bug.cgi?id=3763 (checkout
should re-use ra session for externals if possible)
It's marked as "started", but I'm not sure if that's correct status at
this point.
I'm cc'ing Ben Reser (who has the above issue assigned to him) and
Ivan Zhakov (who recenty wrote a patch for this). Perhaps they can
shed some light on the current state of "RA session caching".
As far as I remember from discussions during the Berlin Hackathon,
this was deemed too risky for 1.8, but perhaps within scope for 1.9.
> We're attempting to work around this by updating the subversion properties
> with svn update . --depth empty --ignore-externals and iterating over the
> externals and comparing to our local revision. If we see a difference, we
> update the file. This is the behavior I'd like to see externals take in the
> end (I'm sure there are good reasons for the current methodology). This
> issue and another issue that can corrupt the working copy are creating some
> barriers for us.
>
> Externals are becoming more and more important as we look to finer grained
> product line sharing of code.
>
> All that said, is there a work around to convincing SVN to recreate the file
> without a full fledge update?
No, I'm afraid I know of no workaround.
Perhaps some further variation of what you're already trying (but I'm
sure you did already many trial and error attempts).
--
Johan
Received on 2013-08-11 12:25:26 CEST