On Mon, Apr 18, 2011 at 7:41 AM, Daniel Shahaf <d...@daniel.shahaf.name> wrote: > On Mon, 18 Apr 2011 07:08 -0400, "Mark Phippard" <markp...@gmail.com> wrote: >> I know why it fails, but I would not expect it to fail as a user, even >> with a proxy. I did not look at Arwin's test but it does not require >> a WC to show the failure. This also fails: >> >> $ svn mkdir url://branches/branch1 >> $ svn mkdir url://branches/branch2 > > Because in the second mkdir, the slave's idea of HEAD is behind master's and > user's idea of HEAD.
As I said, I know why it fails. That does not mean we should not identify these problems and look for ways to solve them. If proxy was a first class feature it would know you were running a command where the all of the HTTP requests should be proxied and this would not fail. > I agree it would be nice if this worked, but given that we have to remain > sane to people who open an RA session to the slave before it has > synced up I'm afraid it might be tricky to address. (That corresponds to the > case of concurrent commits --- two people attempting Mark's two > mkdirs concurrently.) > > Separately: we might want to teach the wc that one repos_url (out of several > with the same repos_uuid) is preferred in given circumstances... > then we could have '^mirror/' and '^master/' syntaxes. </random-thoughts> To reiterate, there is no WC involved here. So a change like that is not needed. This problem exists with commands that are run on URL's alone. As you know most of our SVN commands involved many HTTP requests and the problem here is that some of them are handled by the proxy and others are proxied to the master. A solution would likely require the proxy to have more awareness of the SVN command being executed so that it knew to proxy all of the requests back to the master rather than handle any of them off the local replica. -- Thanks Mark Phippard http://markphip.blogspot.com/
