Mercurial repository

Mercurial repository

[Daniel Berlin]

As I mentioned I would do on overseers a few months ago (at least, I'm
pretty sure i did, it's in my sent mail :P), some gcc developers
requested a mercurial mirror of the svn repository.I just got around
to it now.

I have thus installed mercurial, and it's web component (which is a
simple python script) on sourceware.

The mirror is publicly available, and I am watching our logs and load
from it to make sure it's not taking up a lot of cpu (so far, it is
taking up roughly nothing).

There are no RHEL packages for it, but I am looking into creating some
so that it is nicely managed.

This does not mean gcc is moving to mercurial, or anything of the sort.
This may happen at some point in the distant future, it may not.

--Dan

Re: Mercurial repository

[Jonathan Larmour]

Daniel Berlin wrote:
> 
> I have thus installed mercurial, and it's web component (which is a
> simple python script) on sourceware.
> 
> The mirror is publicly available, and I am watching our logs and load
> from it to make sure it's not taking up a lot of cpu (so far, it is
> taking up roughly nothing).

It was observed before the recent reboot that sourceware was exhibiting a 
high level of disk activity, contributing to a high overall load, resulting 
in things like dropped anonsvn/cvs connections etc.  The nature of disk 
activity is that it's difficult to point the finger at the precise cause, 
but the mercurial script that pulls from svn seemed to be cropping up a lot.

So if you're going to be monitoring mercurial, please also consider its 
effect on disk, not just CPU. This won't show up in 'top' etc.

For now though, the load seems lower, and I couldn't say for definite what 
the difference is, but I do note the mercurial scripts no longer seem to be 
running (since the reboot).

Have a look at http://sourceware.org/mrtg/summary.html and in particular 
note http://sourceware.org/mrtg/pageio.html and how it changes at start of 
week 20. I don't know whether, if or how that corresponds to when mercurial 
(or the mercurial pull from svn script) started operation.

Jifl
-- 
--["No sense being pessimistic, it wouldn't work anyway"]-- Opinions==mine

Re: Mercurial repository

[Daniel Berlin]

On 5/30/07, Jonathan Larmour <jifl@jifvik.org> wrote:
> Daniel Berlin wrote:
> >
> > I have thus installed mercurial, and it's web component (which is a
> > simple python script) on sourceware.
> >
> > The mirror is publicly available, and I am watching our logs and load
> > from it to make sure it's not taking up a lot of cpu (so far, it is
> > taking up roughly nothing).
>
> It was observed before the recent reboot that sourceware was exhibiting a
> high level of disk activity, contributing to a high overall load, resulting
> in things like dropped anonsvn/cvs connections etc.  The nature of disk
> activity is that it's difficult to point the finger at the precise cause,
> but the mercurial script that pulls from svn seemed to be cropping up a lot.
>
> So if you're going to be monitoring mercurial, please also consider its
> effect on disk, not just CPU. This won't show up in 'top' etc.
>
> For now though, the load seems lower, and I couldn't say for definite what
> the difference is, but I do note the mercurial scripts no longer seem to be
> running (since the reboot).
This was probably the  script that was doing the initial mirroring of
svn to mercurial.
It took *way* longer than expected, mostly due to svn.

Anyway, i'll keep an eye out.