On Aug 25, 2008, at 05:17, Tobi Delbruck wrote:
> When we try to commit small changes to chip design schematic EDIF
> files
> which are large (>20MB) text files we often get the bug that
> the commit hangs after uploading.
>
> Observing the server shows that diff is spinning with 100% CPU for
> >30 minutes.
> Occasionally we also get server crashes due (in one case) to a SCSI
> timeout.
> But no other VM's on the same hardware have this SCSI timeout, so
> it is probably not hardware.
> The server is an OpenBSD release 4.3. VM. The diff is not consuming
> too much memory,
> just is spinning.
>
> Committing from WinXP using TortoiseSVN 1.5. Server is running svn
> 1.4.4 apache module
> and we are using filesystem, not Berkeley DB.
>
> If we kill the commit job on the client and try to commit again, it
> fails with out of date warning. We we try
> to update the client, tortoisesvn complains about corrupted local
> copy. But the commit apparently succeeds
> because if we delete the local copy and update, we get the new
> version that we tried to commit.
You say you observe a "diff" process eating all CPU on the server? I
don't think Subversion spawns a separate diff process, either on the
client or server. Perhaps you have a post-commit hook installed on
the server which e.g. sends out a commit email, and perhaps that
script is running "diff" for you? You may need to configure that
script to skip the diff part for files past a certain size.
---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe_at_subversion.tigris.org
For additional commands, e-mail: users-help_at_subversion.tigris.org
Received on 2008-08-26 02:19:21 CEST