svnsync discrepancies again

[Ulrich Spörlein]

On Thu, 2020-07-09 at 10:20:53 -0400, Ryan Steinmetz wrote:
> 
> On (07/09/20 11:26), Ulrich Spörlein wrote:
> >Hi folks
> >
> >comparing my own output with what cgit produces has surfaced another diff
> >in the ports SVN metadata, specifically a 2s delta on r530303
> >
> >What I have in my SVN email (!) is this:
> >Author: sunpoet
> >Date: Thu Apr  2 12:03:10 2020
> >New Revision: 530303
> >URL: https://svnweb.freebsd.org/changeset/ports/530303
> >
> >Clicking that link, I get to this metadata instead:
> >Author: sunpoet
> >Date: Thu Apr 2 12:03:12 2020 UTC (3 months ago)
> >
> >My local svnsync copy has:
> >% svnlook info -r 530303 $PWD
> >sunpoet
> >2020-04-02 14:03:10 +0200 (Thu, 02 Apr 2020)
> >
> >The copy on cgit has:
> >% svn log -vc530303 file:///$PWD
> >------------------------------------------------------------------------
> >r530303 | sunpoet | 2020-04-02 12:03:12 +0000 (Thu, 02 Apr 2020) | 4 lines
> >
> >From cgit (within the cluster):
> >% svn log -c530303 https://svn.freebsd.org/ports
> >------------------------------------------------------------------------
> >r530303 | sunpoet | 2020-04-02 12:03:12 +0000 (Thu, 02 Apr 2020) | 4 lines
> >
> >From a random machine (outside the cluster):
> >% svn log -c530303 https://svn.freebsd.org/ports
> >------------------------------------------------------------------------
> >r530303 | sunpoet | 2020-04-02 14:03:10 +0200 (Thu, 02 Apr 2020) | 4 lines
> >
> >Why do we have garbage on svnweb.freebsd.org and why does a machine in the
> >cluster get pointed to that host?
> 
> Not sure I'd call this "garbage" nor would I immediately assume that 
> we've decided to "point cluster machines at garbage" and "give the rest 
> of the world the good stuff."
> 
> That being said, this is likely related to how svnsync works.  It's a 
> two step operation where commits are replayed first, followed by the 
> metadata.  Since this isn't atomic, there's a brief window that can 
> sometimes surface between these two operations.
> 
> svnweb has its own local copy of the repos that it looks at.  So do the 
> multiple public, geo-load balanced mirrors (svn.freebsd.org).

Yes, sadly I'm intimately familiar with the shortcomings of svnsync in
this regard. What I'm wondering is:
- who will fix this and
- where is the actual canonical source of our SVN and
- how can I get access to it from both a machine inside the cluster as
  well as outside?

That would be helpful to know.

Cheers
Uli