svnsync.txt   [plain text]


                           ==== What Is It? ====

svnsync is a tool for creating and maintaining read-only mirrors of
subversion repositories.  It works by replaying commits that occurred
in one repository and committing it into another.

                           ==== Basic Setup ====

First, you need to create your destination repository:

$ svnadmin create dest

Because svnsync uses revprops to keep track of bookkeeping information
(and because it copies revprops from the source to the destination)
it needs to be able to change revprops on your destination repository.
To do this you'll need to set up a pre-revprop-change hook script that
lets the user you'll run svnsync as make arbitrary propchanges.

$ cat <<'EOF' > dest/hooks/pre-revprop-change
#!/bin/sh
USER="$3"

if [ "$USER" = "svnsync" ]; then exit 0; fi

echo "Only the svnsync user can change revprops" >&2
exit 1
EOF
$ chmod +x dest/hooks/pre-revprop-change

$ svnsync init --username svnsync file://`pwd`/dest \
                                  http://svn.example.org/source/repos
Copied properties for revision 0
$

Note that the arguments to 'svnsync init' are two arbitrary repository
URLs.  The first is the destination, which must be empty, and the second
is the source.

Now you can just run the 'svnsync sync' command to synchronize pending
revisions.  This will copy any revisions that exist in the source repos
but don't exist in the destination repos.

$ svnsync sync file://`pwd`/dest
Committed revision 1.
Copied properties for revision 1.
Committed revision 2.
Copied properties for revision 2.
Committed revision 3.
Copied properties for revision 3.
...

                              ==== Locks ====

If you kill a sync while it's occurring there's a chance that it might
leave the repository "locked".  svnsync ensures that only one svnsync
process is copying data into a given destination repository at a time
by creating a svn:sync-lock revprop on revision zero of the destination
repository.  If that property is there, but you're sure no svnsync is
actually running, you can unlock the repository by deleting that revprop.

$ svn pdel --revprop -r 0 svn:sync-lock file://`pwd`/dest

                               ==== FAQ ====

Q: So what can I do with this thing anyway?

A: Well, anything that's read-only.  As long as you don't commit changes
   to the destination repository you're all set.  This means destination
   repositories are good for providing offsite mirrors, read-only mirrors,
   etc.

Q: What if I want to check out from a mirror, but commit to the master?

A: That's possible, but requires some gymnastics.  You see, each repository
   has its own UUID, which is stored in the working copy, so if you check
   out from the mirror, and then do a 'svn switch --relocate' to point to
   the master it'll error out.  To make this work you need to make sure that
   the mirrors have the same UUID as the master.  You can read a repository
   UUID with 'svnlook uuid' or 'svn info', and change it with
   'svnadmin setuuid'.

   Once both the mirror and master repositories have the same UUID you can
   safely check out from the mirror and commit to the master, all you have
   to do is do a 'svn switch --relocate' to point your working copy to the
   master before a commit.

   Note that you should NEVER commit changes to a mirror other than
   via svnsync, so to avoid accidentally doing so you may want to add
   a start-commit hook script that disallows commits from users other
   than the one you run svnsync as, and a pre-lock hook script that
   disallows all filesystem lock requests (svnsync will never create
   these locks, but its attempt to commit can be blocked by them).

Q: What version of Subversion is required to use svnsync?

A: The source repository must be running Subversion 1.4 or newer, since
   svnsync uses a new RA layer command that was added in 1.4.  On the other
   hand, the destination repository can be any version of Subversion, since
   all svnsync is doing is committing changes using the regular RA APIs.

Q: Do I need to run svnsync on the same machine as one of the
   repositories?

A: While you do need direct access to the destination repository to
   set up a pre-revprop-change hook script, after that point svnsync
   communicates with both repositories through the same "repository
   access" layer that svn uses to connect to remote repositories. So
   svnsync does not have to be run on the same machine as either
   repository; it can communicate with both repositories over any of
   Subversion's RA protocols (svn://, svn+ssh://, http://, https://,
   or file:///).  In fact, you don't need any special permissions on
   the source repository at all.

Q: How does svnsync deal with parts of the master repository that I'm not
   authorized to read?

A: svnsync will simply not copy parts of the repository that you
   cannot read; files copied from "private" parts of the repository
   into "public" parts will look as if they have been added from
   scratch.  If a revision only modifies files that you cannot read,
   it will appear to be empty.  (Just like with "svn log", log
   messages from revisions you cannot read part of will be empty.)

Q: Can I mirror a subdirectory of a master repository?

A: As of Subversion 1.5, it is possible to limit svnsync to a subdirectory
   of the master repository.
   This is most useful when the master repository is organized in projects, 
   and you want to sync only one project.
   Example showing svnsync of project1 in the master repository:
   /project1
      /branches
      /tags
      /trunk
   /project2
      /branches
      /tags
      /trunk

   The following commands will sync all changes in /project1 to the target 
   repository:
$ svnsync init file://`pwd`/dest http://svn.example.org/source/repos/project1
$ svnsync sync file://`pwd`/dest
   
   Note: this syntax only allows you to limit the scope of svnsync to
   /project1. It does not: 
   - allow you to sync two or more projects from the master repository. 
   - recognize renames of project1. Example, if the original name of project1
   was secretproject, only the changes starting from the revision in which the
   rename to project1 was committed will be synced.

   If you need any of these abilities right now, you may want to look into SVK
   (http://svk.bestpractical.com/).

Q: What happens when I change a revprop on the master server?

A: That depends, did you change it on a revision that had already been
   mirrored or one that's still waiting to be mirrored.  If the revision
   hasn't been mirrored yet, the new revprop will just get copied across
   normally when the next sync happens.  If not, then you've got a small
   problem.  You see, since revprops aren't versioned, there's no way to
   detect (via the Subversion RA APIs anyway) that it's been changed, so
   the next time you run a sync svnsync has no way to tell that it has
   changed.  There is a way for you to build your own solution though,
   just use the 'svnsync copy-revprops' command.  The usual technique is
   either to put an explicit call to it in your master repository's
   post-revprop-change script, or to have the post-revprop-change script
   record the fact that a change occurred, and then later on have some
   job look through the list of changes and copy them over for you.

Q: How can I relocate the source repository for svnsync?

A: A simple way to relocate a svnsync-ed repository is to change the
   revision property (of the mirror) that stores the source repository's
   URL.  Just use this command:

$ svn propset svn:sync-from-url --revprop -r 0 NEW_SOURCE_URL MIRROR_URL

  NOTE: Don't use `svn propedit`, because editors may append an EOL
  character to NEW_SOURCE_URL that will lead svnsync complaining
  "svnsync: Malformed URL for repository".