Slow hgweb.cgi

Dominik Psenner dpsenner at gmail.com
Wed Apr 13 01:44:09 CDT 2011


*argh* I could bang my head on the table. I just did:

1 cloned the complete hgweb.cgi to another url (hgserver/hg2)
2 set up a repository folder that is not a samba share (mkdir)
3 copied the biggest repository to that folder
4 ran a hg pull from my pc

Guess what? It went through within a jiffy. So the problem is indeed
smb/cifs. Yet I do not fully understand what's going on because it doesn't
seem to be a timeout issue where hgweb.cgi is just waiting for the smb
share.

I used to set it up with autofs:

-- quote auto.cifs --
$ cat /etc/auto.cifs
HGRoot -fstype=cifs,rw,credentials=/etc/samba/cred-file,uid=www-data,gid=hg
://"dc1/Disk_X/HGRoot/"&
-- eof quote--

The script goes into a heavy operation that consumes one core completely for
a rather long time where it has no obvious reason to do that. *hm* It feels
like there's something in hg or hgweb that causes problems, but I can't
really confirm it. It just feels like hgweb is somehow doing:

If samba
   Heavy loop
Continue

:-)

-- quote top --
3871 www-data  20   0  8188 4324 2160 D   27  0.4   0:00.47 python
/var/hg/hgweb.cgi
-- eof quote --

The faster version where the repository is not on the samba share doesn't
even pop up in top, so I'm unable to quote it.

Have you any ideas or suggestions I could try? Would an update to 1.8.2
help? 

> -----Original Message-----
> From: Dominik Psenner [mailto:dpsenner at gmail.com]
> Sent: Wednesday, April 13, 2011 8:05 AM
> To: 'Adrian Buehlmann'
> Cc: mercurial-devel at selenic.com
> Subject: RE: Slow hgweb.cgi
> 
> Good morning everyone.
> 
> Unfortunatly http://support.microsoft.com/kb/832161 is not our problem. We
> have WebClient disabled on that Win2003 Server. Our sysadmin further
> explained me that they have a huge ssd raid backend powered by fibre optic
> cables. So throughput should not be a problem at all.
> 
> Now I'm going to try moving the repository from samba into the virtual
> machine to see if the slow performance is still there without samba. At
> least we would be able to either blame or praise smb/cifs.
> 
> > -----Original Message-----
> > From: Dominik Psenner [mailto:dpsenner at gmail.com]
> > Sent: Tuesday, April 12, 2011 6:49 PM
> > To: Adrian Buehlmann
> > Cc: mercurial-devel at selenic.com
> > Subject: Re: Slow hgweb.cgi
> >
> > Great, that just looks promising. I look forward to test that tomorrow.
> > I'm going to clone the repositories somewhere else so that it is no
> > longer and see if the problem's still there without having the samba
> > storage involved. So we are going to know more tomorrow for sure!
> > --
> > Dominik Psenner
> > ## OpenPGP Key Signature #################################
> > # Key ID: B469318C                                       #
> > # Fingerprint: 558641995F7EC2D251354C3A49C7E3D1B469318C  #
> > ##########################################################
> >
> > On 04/12/2011 06:32 PM, Adrian Buehlmann wrote:
> > > On 2011-04-12 18:21, Dominik Psenner wrote:
> > >> Thanks for the response!
> > >>
> > >> Huh, dunno for sure .. gotta ask the server admin tomorrow. :-)
> > >
> > > I was pondering about http://support.microsoft.com/kb/832161
> > >
> > > If you can, ask the admin to stop the "WebClient service" on the
> server
> > > and see if there's a difference.
> > >
> > > We've been hit on TortoiseHg by that. The effect was dead slow access
> on
> > > Windows shares, which was triggered by the overlay handler going to
> high
> > > up in the directory in its search for .hg folders. Which triggered a
> > > massive speed penalty when hitting \\server or \\server\share (can't
> > > remember the details).
> 




More information about the Mercurial-devel mailing list