Is the warning in http://subversion.tigris.org/faq.html#bdb-recovery still relevant?

[Eric Hanchrow <of...@blarg.net>]

Today on #svn sussman suggested that, since "svnadmin recover" gets an
exclusive lock on the repository before starting work, it is quite
safe to run at any time.  That suggests to me that the above-mentioned
warning is no longer needed.  Is this really the case?  And if so, is
the following paragraph ("Make absolutely sure you disable all access
to the repository before doing this ...") itself still relevant?
-- 
Using Unix is the computing equivalent of listening only to music
by David Cassidy.
        -- Rob Pike

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org

Re: Is the warning in http://subversion.tigris.org/faq.html#bdb-recovery still relevant?

[Philip Martin <ph...@codematters.co.uk>]

Eric Hanchrow <of...@blarg.net> writes:

> Today on #svn sussman suggested that, since "svnadmin recover" gets an
> exclusive lock on the repository before starting work, it is quite
> safe to run at any time.  That suggests to me that the above-mentioned
> warning is no longer needed.  Is this really the case?  And if so, is
> the following paragraph ("Make absolutely sure you disable all access
> to the repository before doing this ...") itself still relevant?

I believe it's safe to run "svnadmin recover" on a live repository
provided all other access also goes through libsvn_repos.  If there is
a lot of repository activity you may need to try repeatedly, or use
the --wait option, to get the exclusive lock.  Using the BDB tools
while running "svnadmin recover" or running db_recover while allowing
normal libsvn_repos access is likely to cause problems.  I guess we
could make the warning db_recover specific.

On a related note, I think the repository lock suffers from the
"per-process" problem that affected the libsvn_fs_fs lock, the lock
doesn't provide exclusive access within a multi-threaded process on
Unix.  That's not a problem the way we currently use it, since
svnadmin is single-threaded, but a multi-threaded program might have
problems.  I suppose we should either add a mutex or document the
restriction.

-- 
Philip Martin

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org