access denied errors (bug 1576) still a problem?

[Baz <br...@gmail.com>]

Hi there,
with the current stable releases of svn, tortoisesvn, intellij we're
still getting the 'access denied' bug or variations of it once or
twice a week on windows XP.
http://subversion.tigris.org/issues/show_bug.cgi?id=1576
(that bug says closed, but the thread that goes along with it explains
that the patch just reduces the likelihood of being bitten)

Our incidence of this seems a bit higher than I'd expect in a stable
release, but I've noticed that we seem particularly prone to the issue
in one project, where one of the versioned directories has 500+ files
in it (oh... don't ask).

I just wanted to check if anyone knew if there is something
directory-size dependent in this bug? e.g. .svn/entries is 157k,
there's a lot of .svn-base files that would need touched in an update,
etc; I guess this might increase the window of vulnerability. Is it
worth me trying some experiments (with ludicrously large dirs) to see
if I can reproduce this more often? There seemed to be some trouble
trying to trigger it before.

Or am I on a wild goose chase for something thats resolved next release...

It would be really nice if there was a way to at least automate the
solution to this. Right now we move the broken folder out of the tree,
delete its .svn directories, update the tree to recover the missing
folder, copy the code back, and manually correct any failed 'adds'. A
better way would be to use a variation on 'svn update' that recovered
the .svn dirs without touching the working copy. 'svn repair' perhaps?

-Baz

PS we've already taken the step of excluding the folders involved from
virus scans and google desktop indexing. Its likely we have more
tag-along processes though.

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

Re: access denied errors (bug 1576) still a problem?

[Baz <br...@gmail.com>]

On 5/2/06, D.J. Heap <dj...@gmail.com> wrote:
> Are you using the native Win32 client or cygwin?  AFAIK, the cygwin
> client doesn't include the workaround code.  Which clients actually
> get the error and are they built using the native libraries or cygwin
> libraries or something else?  Could there really be other processes
> holding onto the files (editors, etc.)?

We're using tortoise and intellij mainly - tortoise is native and
intellij uses javasvn (which I believe is independent of the svn
libs). I've also wondered about this being a source of error -
tortoise looking at a window in the background while intellij updates
it, for example. I can test this too.

I do use the cygwin client as well, but only from time to time. Theres
no correlation with the problem.

> The workaround code basically waits and retries for around 10 seconds
> per file -- that's quite a long time to wait for tag-alongs to finish
> whatever they were doing with each file.

Ah - that doesn't sound like what we're seeing, unless other work is
done in parallel with the wait. The commit/update where the bug occurs
is taking a normal amount of time, but this would be longer than 10s
anyway due to the size of the tree. I don't recall seeing 10s pauses
though.

> Also, you shouldn't have a broken tree afterwards.  You should be able
> to retry the operation.  If you are getting broken working copies,
> then please mail a good reproduction recipe to the dev list -- that's
> a bug.

Ok, I'll see if I can nail it down. We are getting (very) broken
working copies, but I presumed that since its in the faq its a known
issue. I'll set up a test repository and stress it a bit.

Cheers
Baz

PS sorry for the dupe. Must-hit-reply-to-all....doh.

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

Re: access denied errors (bug 1576) still a problem?

["D.J. Heap" <dj...@gmail.com>]

On 5/2/06, Baz <br...@gmail.com> wrote:
> Hi there,
> with the current stable releases of svn, tortoisesvn, intellij we're
> still getting the 'access denied' bug or variations of it once or
> twice a week on windows XP.
> http://subversion.tigris.org/issues/show_bug.cgi?id=1576
> (that bug says closed, but the thread that goes along with it explains
> that the patch just reduces the likelihood of being bitten)
>
> Our incidence of this seems a bit higher than I'd expect in a stable
> release, but I've noticed that we seem particularly prone to the issue
> in one project, where one of the versioned directories has 500+ files
> in it (oh... don't ask).
[snip]

Are you using the native Win32 client or cygwin?  AFAIK, the cygwin
client doesn't include the workaround code.  Which clients actually
get the error and are they built using the native libraries or cygwin
libraries or something else?  Could there really be other processes
holding onto the files (editors, etc.)?

The workaround code basically waits and retries for around 10 seconds
per file -- that's quite a long time to wait for tag-alongs to finish
whatever they were doing with each file.

Also, you shouldn't have a broken tree afterwards.  You should be able
to retry the operation.  If you are getting broken working copies,
then please mail a good reproduction recipe to the dev list -- that's
a bug.

DJ

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