david at allouche.net
Wed Jan 17 19:06:56 GMT 2007
Jelmer Vernooij wrote:
> On Wed, 2007-01-17 at 12:54 +0100, David Allouche wrote:
> > That's very interesting. A number of Launchpad imports fail in a
> > not-deterministic way with "failed REPORT". Currently I just classify
> > those failures as "unreliable server", but I was puzzled by a pattern
> > I noticed: the failures did not happen in a really random way.
> I've seen that one as well. It appears to especially occur often with
> sourceforge repositories. I suspect it has to do with timeouts, but I'm
> not sure (as sourceforge uses https only so I can't use wireshark).
One fix I plan to implement for cscvs is to get all svn communication go through a single session-like object which can retry failed requests.
That should fix those requests that fail transiently due to a load peak on the server. But it might also unintentionally work around this bug with svn.ra.get_dir :(
> > I would very much like if you could notify me when you encounter weird bugs like this
> Sure, I'll try to keep you informed. Can you please do the same if you
> encounter strange bugs?
> > (or like the "need to escape control chars in commit log for bzrlib" bug) as launchpad-cscvs and svn2bzr _will_ have the same problems.
> That's actually a bzr bug (the escaping should happen in bzr's revision serializer), which
> I hope to fix properly soon.
Ack that, but it appears to happen particularly with svn imports. So although it's not specifically a svn problem, it's still a svn-related problem.
BTW, my svn2bzr branch copied the fix from bzr-svn, before I made it use a customised MinimalTree (from cvsps-import) which also implemented this fix.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Size: 189 bytes
Desc: not available
Url : https://lists.ubuntu.com/archives/bazaar/attachments/20070117/9a8f24c8/attachment.pgp
More information about the bazaar