Spread out sitescript pulls to avoid hgweb bug

This is a workaround for a race condition in hgweb: http://bz.selenic.com/show_bug.cgi?id=3953

I looked a bit into fixing this, but it's definitely non-trivial. So I came up with a crude workaround instead. Note that sitescripts is not the only thing we pull, so we're probably going to see this again as we add more servers. Might make sense to have a script that handles updating and put the logic there.

[Wladimir Palant, 2013-09-03 13:34:00]

Judging by the lack of further error mails - LGTM :)

However, I strongly disagree with this as a long-term solution (I don't expect
Selenic to resolve it in the next few days). I would rather attempt
sequentializing config accesses with a local patch instead.

[Wladimir Palant, 2013-09-03 13:46:56]

I actually patched /usr/lib64/python2.7/site-packages/mercurial/ui.py now. At
the top I added:

    import thread
    config_lock = thread.allocate_lock()

And in setconfig() I added:

    with config_lock:

[Felix Dahlke, 2013-09-03 13:59:14]

On 2013/09/03 13:46:56, Wladimir Palant wrote:
> I actually patched /usr/lib64/python2.7/site-packages/mercurial/ui.py now. At
> the top I added:
> 
>     import thread
>     config_lock = thread.allocate_lock()
> 
> And in setconfig() I added:
> 
>     with config_lock:

We discussed it on IRC, but for the record: Patching Mercurial locally is
probably the best approach, yes. I only considered properly fixing it and
getting it merged upstream, that'd be more involved. This should suffice to
avoid this particular race condition.

[Wladimir Palant, 2013-10-01 09:45:08]

I've updated Mercurial which replaced my fix of course. So I've added the
following code to our hgwebdir.fcgi (not part of the Mercurial distribution,
based on an example of how Mercurial should be run):

# Serialize ui.setconfig() accesses to avoid hitting
http://bz.selenic.com/show_bug.cgi?id=3953
from mercurial.ui import ui
import thread

config_lock = thread.allocate_lock()
orig_setconfig = ui.setconfig

def new_setconfig(*args, **kwargs):
  with config_lock:
    orig_setconfig(*args, **kwargs)

ui.setconfig = new_setconfig

This should survive future Mercurial updates.

[Felix Dahlke, 2013-10-01 14:07:33]

On 2013/10/01 09:45:08, Wladimir Palant wrote:
> I've updated Mercurial which replaced my fix of course. So I've added the
> following code to our hgwebdir.fcgi (not part of the Mercurial distribution,
> based on an example of how Mercurial should be run):
> 
> # Serialize ui.setconfig() accesses to avoid hitting
> http://bz.selenic.com/show_bug.cgi?id=3953
> from mercurial.ui import ui
> import thread
> 
> config_lock = thread.allocate_lock()
> orig_setconfig = ui.setconfig
> 
> def new_setconfig(*args, **kwargs):
>   with config_lock:
>     orig_setconfig(*args, **kwargs)
> 
> ui.setconfig = new_setconfig
> 
> This should survive future Mercurial updates.

LGTM