You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@subversion.apache.org by kf...@collab.net on 2001/08/29 23:56:53 UTC
DB_INCOMPLETE solved, small new problem appearing though
I'm happy to report that Sander Striker's patch did the trick.
Though my `mass-commit' script is still running (just got past commit
#150), it's gotten no errors so far. While it's been doing commits
over ra_dav, I've been running `svnlook' repeatedly on the repository,
*and* doing simultaneous checkouts from the same repository. And
occasionally cleaning out old log files using `db_archive'. :-)
The mass-commit script never even pauses for breath. No errors from
the script, no errors in /var/log/httpd/error_log, and `svnlook'
reports a consistent repository every time. As best I can determine
from watching all of /proc/*/fd/, no one is holding open file
descriptors longer than necessary. Apache process sizes all look
reasonable too.
Oh wait, now it's on commit #284, and *still* no errors.
I have noticed that the update driver can get confused during those
simultaneous checkouts (i.e., a checkout going on during mass-commit)
and error out in get_editor.c:847 ("object of same name already
exists"). This will need looking into. Have done some detective work
already, and have a hypothesis about the cause -- Greg Stein or Ben,
if you take a look at issue #468, I think there's a chance you'll know
right away what's going on.
Ahh, mass-commit just finished commit #1000 with no errors.
So the summary of this mail is: mostly win with a little bit lose. :-)
-K
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: DB_INCOMPLETE solved, small new problem appearing though
Posted by kf...@collab.net.
Wonderful developments!
(Please note, however, that what you're browsing is not yet the true
repository, it's just a test repos that will get blown away and
re-imported when we really hit M3.)
-K
Joe Orton <jo...@manyfish.co.uk> writes:
> On Thu, Aug 30, 2001 at 02:58:37AM +0200, Branko �ibej wrote:
> > Bravo! My hat's off.
> >
> > I'm jumping up and down with glee right now, because I can actually
> > browse the repository on svn.collab.net with an ordinary web browser.
> > And it's fast, doesn't feel like ther's a 100meg database sitting behind it.
> >
> >
> > (Well, almost: for directories, I get "No default response to GET for a
> > collection.". Fix that, and we've got browsing on HEAD for free. (Hmm.
> > Looking at one of the gifs in notes/guis/pics, I get a textual
> > representation. Guess what "websvn" will do with the mime-type props? :-))
>
> You can point your favourite DAV client at it too :)
>
> $ cadaver svn.collab.net/repos/svn/
>
> (though cadaver is a bit unhappy about the lack of properties exposed)
>
> joe
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
> For additional commands, e-mail: dev-help@subversion.tigris.org
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: DB_INCOMPLETE solved, small new problem appearing though
Posted by Greg Stein <gs...@lyra.org>.
On Thu, Aug 30, 2001 at 07:35:28AM +0100, Joe Orton wrote:
> On Thu, Aug 30, 2001 at 02:58:37AM +0200, Branko �ibej wrote:
> > Bravo! My hat's off.
> >
> > I'm jumping up and down with glee right now, because I can actually
> > browse the repository on svn.collab.net with an ordinary web browser.
> > And it's fast, doesn't feel like ther's a 100meg database sitting behind it.
"Fast" is a good thing, but I don't think we're going to set any requests-
per-second record :-) It is quite cool that it is fast even *before* we
have spent any time worrying about it.
(and I have an item or two on the table to speed it up even more)
[ some future day, we may want to analyze the performance, but it should be
quite fine for a while; the above-mentioned item is adjusting the mod_dav
API to ask for buckets from the backend (much more efficient than read
loops); it would also be good for a request to do an "internal" redirect
to the version resource; that resource can be cached, unlike the "public"
URL; internal Apache caches can then satisfy future requests without
hitting the Berkeley DB; damn I'm happy to be hosted in Apache... ]
> > (Well, almost: for directories, I get "No default response to GET for a
> > collection.". Fix that, and we've got browsing on HEAD for free. (Hmm.
That one is a bit tough. mod_dav shuts down collection requests even before
the repository gets a chance. Need to think more on that one.
> > Looking at one of the gifs in notes/guis/pics, I get a textual
> > representation. Guess what "websvn" will do with the mime-type props? :-))
Should be a reasonably simple fix to repos.c::dav_svn_set_headers(), so that
browsing a jpg will return a real image :-)
> You can point your favourite DAV client at it too :)
>
> $ cadaver svn.collab.net/repos/svn/
>
> (though cadaver is a bit unhappy about the lack of properties exposed)
What properties, specifically? IIRC, there are only a few you look for. It
shouldn't be hard at all to update the support for just those.
[ needless to say, since the SVN client hasn't needed them, I haven't spent
effort to round out the dav support; but we certainly can do that... ]
Cheers,
-g
--
Greg Stein, http://www.lyra.org/
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: DB_INCOMPLETE solved, small new problem appearing though
Posted by Joe Orton <jo...@manyfish.co.uk>.
On Thu, Aug 30, 2001 at 02:58:37AM +0200, Branko �ibej wrote:
> Bravo! My hat's off.
>
> I'm jumping up and down with glee right now, because I can actually
> browse the repository on svn.collab.net with an ordinary web browser.
> And it's fast, doesn't feel like ther's a 100meg database sitting behind it.
>
>
> (Well, almost: for directories, I get "No default response to GET for a
> collection.". Fix that, and we've got browsing on HEAD for free. (Hmm.
> Looking at one of the gifs in notes/guis/pics, I get a textual
> representation. Guess what "websvn" will do with the mime-type props? :-))
You can point your favourite DAV client at it too :)
$ cadaver svn.collab.net/repos/svn/
(though cadaver is a bit unhappy about the lack of properties exposed)
joe
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: DB_INCOMPLETE solved, small new problem appearing though
Posted by Branko Čibej <br...@xbc.nu>.
Bravo! My hat's off.
I'm jumping up and down with glee right now, because I can actually
browse the repository on svn.collab.net with an ordinary web browser.
And it's fast, doesn't feel like ther's a 100meg database sitting behind it.
(Well, almost: for directories, I get "No default response to GET for a
collection.". Fix that, and we've got browsing on HEAD for free. (Hmm.
Looking at one of the gifs in notes/guis/pics, I get a textual
representation. Guess what "websvn" will do with the mime-type props? :-))
--
Brane �ibej <br...@xbc.nu> http://www.xbc.nu/brane/
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: DB_INCOMPLETE solved, small new problem appearing though
Posted by kf...@collab.net.
"Neal Reizer" <nr...@collab.net> writes:
> Great news! Congratulations!
Thanks, Neal!
Actually though (everyone), I am a little worried about the possible
effects of this problem:
> > I have noticed that the update driver can get confused during those
> > simultaneous checkouts (i.e., a checkout going on during mass-commit)
> > and error out in get_editor.c:847 ("object of same name already
> > exists"). This will need looking into. Have done some detective work
> > already, and have a hypothesis about the cause -- Greg Stein or Ben,
> > if you take a look at issue #468, I think there's a chance you'll know
> > right away what's going on.
Seems to me that it could result in a lying update, that is, your
checkout or update could finish and report that your working copy, or
some portion of it, is at revision X, and yet some objects claiming to
be at X would really have contents as if for Y (where Y > X).
Need to think about this more. For now, I've reclassifed issue #468
as an M3 bug, but would love to be talked out of that, if we can
establish that there's no danger of dishonest updates. :-) Bugs that
cause us inconvenience are okay in M3, but data corruption is another
matter.
-K
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
RE: DB_INCOMPLETE solved, small new problem appearing though
Posted by Neal Reizer <nr...@collab.net>.
Great news! Congratulations!
> -----Original Message-----
> From: kfogel@collab.net [mailto:kfogel@collab.net]
> Sent: Wednesday, August 29, 2001 4:57 PM
> To: dev@subversion.tigris.org
> Subject: DB_INCOMPLETE solved, small new problem appearing though
>
>
> I'm happy to report that Sander Striker's patch did the trick.
>
> Though my `mass-commit' script is still running (just got past commit
> #150), it's gotten no errors so far. While it's been doing commits
> over ra_dav, I've been running `svnlook' repeatedly on the repository,
> *and* doing simultaneous checkouts from the same repository. And
> occasionally cleaning out old log files using `db_archive'. :-)
>
> The mass-commit script never even pauses for breath. No errors from
> the script, no errors in /var/log/httpd/error_log, and `svnlook'
> reports a consistent repository every time. As best I can determine
> from watching all of /proc/*/fd/, no one is holding open file
> descriptors longer than necessary. Apache process sizes all look
> reasonable too.
>
> Oh wait, now it's on commit #284, and *still* no errors.
>
> I have noticed that the update driver can get confused during those
> simultaneous checkouts (i.e., a checkout going on during mass-commit)
> and error out in get_editor.c:847 ("object of same name already
> exists"). This will need looking into. Have done some detective work
> already, and have a hypothesis about the cause -- Greg Stein or Ben,
> if you take a look at issue #468, I think there's a chance you'll know
> right away what's going on.
>
> Ahh, mass-commit just finished commit #1000 with no errors.
>
> So the summary of this mail is: mostly win with a little bit lose. :-)
>
> -K
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
> For additional commands, e-mail: dev-help@subversion.tigris.org
>
>
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org