subversion-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Daniel Shahaf <danie...@elego.de>
Subject Re: svnsync and locks
Date Thu, 05 Apr 2012 18:19:00 GMT
C. Michael Pilato wrote on Thu, Apr 05, 2012 at 12:19:17 -0400:
> On 04/05/2012 11:26 AM, Julian Foad wrote:
> > How would an admin arrange for svnsync to synchronize locks
> > (reserved-checkout locks, that is)?
> > 
> > I was talking to Philip and he mentioned that he'd been thinking about
> > this.  It seems to us that the only way available currently is for
> > post-[un]lock on the master to rsync the whole 'locks' directory to the
> > slave.  (That's for FSFS; no idea if there's an equivalent for BDB.)
> > That doesn't seem satisfactory, for several reasons.  One issue is it
> > isn't guaranteed to happen in the right order relative to commits.
> > 
> > In terms of *preventing* a user committing to a locked file without
> > holding the lock, you don't need the locks to be present on the slave, of
> > course, because it's the master not the slave that will process a commit.
> > But if we don't sync locks onto the slave, then users checking out and
> > updating from the slave will not see the correct set of locks, which is
> > unhelpful.
> > 
> > Could we teach svnsync to sync locks?
> > 
> > If we did have a way to sync locks, there would then be locks on the
> > slave, and how would "svnsync sync" then make commits?  I can't think how
> > it could know what lock tokens it should provide with the commit; the
> > master kept no record of them, nor even of the fact that such locks
> > existed on the master at that earlier point in time.  I suppose svnsync
> > would have to make its commit to the slave in a way that bypasses all
> > lock checking.  Or maybe there are ways we could make it supply the right
> > list of lock tokens, but I can't think of a way.  Bypassing all locks
> > should be fine in this scenario.
> 
> See http://subversion.tigris.org/issues/show_bug.cgi?id=3457 for earlier
> manifestations of these thoughts and links to possibly to some more
> complications.

Also http://subversion.tigris.org/issues/show_bug.cgi?id=3750

Daniel
(who currently wonders whether replacing the FSFS locks/ by an SQLite db
will make some of the problems easier)

Mime
View raw message