Commit memory usage

["D.J. Heap" <dj...@shadyvale.net>]

When many targets are specified, svn_client_commit chews through lots of 
memory (~500 targets went through ~300MB of memory).  The attached patch 
introduces a subpool into the offending loop and seems to pass cursory 
testing.

Would someone mind reviewing it for correctness?  Feel free to 
tweak/test/commit if it looks desirable -- I'm going to be slow to 
respond for a couple of days.

I'm also investigating some heavy memory use on the server (svnserve) 
and client (during update) when hundreds of files are deleted in one 
commit, but I haven't repro'd it on HEAD yet -- hopefully by this 
weekend I will have details if it repros on HEAD.

DJ

Re: Commit memory usage

["D.J. Heap" <dj...@shadyvale.net>]

Ben Collins-Sussman wrote:
> On Tue, 2003-12-09 at 20:49, D.J. Heap wrote:
> 
> 
>>svn up -- the client memory grows to huge amounts
> 
> 
> This looks like an important scalability bug.  Can you file it, so we
> don't forget about it?  Maybe someone can fix it this week.
> 
> 

Issue 1635 filed.



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

Re: Commit memory usage

[Ben Collins-Sussman <su...@collab.net>]

On Tue, 2003-12-09 at 20:49, D.J. Heap wrote:

> svn up -- the client memory grows to huge amounts

This looks like an important scalability bug.  Can you file it, so we
don't forget about it?  Maybe someone can fix it this week.




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

Re: Commit memory usage

["D.J. Heap" <dj...@shadyvale.net>]

D.J. Heap wrote:
[snip]
 > I'm also investigating some heavy memory use on the server (svnserve)
> and client (during update) when hundreds of files are deleted in one 
> commit, but I haven't repro'd it on HEAD yet -- hopefully by this 
> weekend I will have details if it repros on HEAD.
> 
> DJ

I've just confirmed that these still reproduce on HEAD, but I won't have 
time to dig into them further until this weekend.  Recipe below for 
anyone interested.

Recipe running HEAD (r7968) svnserve and client on Windows:

create empty repo
check it out

copy lots of files into the root wc folder (no subdirs)
    (I used the files -- no subdirs -- from Apache's docs/manual
     directory and apr-iconv's ccs directory)

svn add *
svn commit
svn co to a secondary wc for later use
svn del *
svn commit -- without the above patch, both the client and the
               server memory grow to huge amounts; with the
               patch the client stays steady but the server still
               grows

cd to secondary wc
svn up -- the client memory grows to huge amounts


DJ


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