Couple of questions:

1. Are you going to add ability to 'reload' configuration instead of doing a
restart of pound? Closing all client connections for a configuration change is
not a good solution especially when you have several thousand current
connections open.

2. Are you going to add ability to keep configuration for different sites in
different
files and 'include' those configuration files into main config file something
like Apache does?

Thank you.





 
---------------------------------
 Get your own web address.
 Have a HUGE year through Yahoo! Small Business.

The include facility is relatively trivial build outside of pound with a
startup script (and Robert has declined to implement that feature in the
past).

The reload capability is important, but Robert has also been less than
enthusiastic about that.  I would recommend that you push for that feature
over the other.


On 2/20/07 4:38 AM, "Dennis" <ccweb03(at)yahoo.com> wrote:
[...]

On 2/19/07, Ted Dunning <tdunning(at)veoh.com> wrote:[...]

Yeah, nice.. but not so important.
[...]

This is shocking. We're using a reverse proxy, mostly for
high-availability reasons (I think most people feel the same), and
restarting it is quite disruptive. I'd be interested in hearing why
the generally-accepted behavior of SIGHUP isn't important to pound. :)

 -Charlie

>> The reload capability is important, but Robert has also been less
than[...][...]

In my personal experience, arguments for features are much stronger when 
  they include patches.  Now I can't speak for anyone to say that your 
patch would be accepted, but I am guessing it would help a lot towards 
making it reality.  Why not take a stab at it?

Regards,[...]

Also,  if you decide to do this, remember that Robert really, really likes
to be able to keep the pound in root jail.

That might mean that the right answer is to use the control socket to send
the new config to the process.

Another interesting option is to invoke a new pound and have a signal to the
old pound that says to stop accepting new connections.  The problem with
that approach is that session information might be hard to transfer (but
writing to a file in the root jail just before you stop listening is a
viable option.  A second signal can tell the new instance to read the
session file and start accepting connections.  There would be a fraction of
a millisecond of down-time with this approach, but it would be fantastically
easier than changing all of the in-memory structures on the fly.


On 2/20/07 5:14 PM, "Dave Steinberg" <dave(at)redterror.net> wrote:

>>> The reload capability is important, but Robert has also been less
than
>>> enthusiastic about that.  I would recommend that you push for that
>>> feature
>>> over the other.[...][...]

Also, feel free to check the archives, this has been discussed to
*death* at least 5 or 6 times.

I'm encouraged that with poundctl, perhaps as it matures this type of
hot-reload functionality might be available out of band.  (For instance,
if you can serialize your sessions and re-inject them post-reload, the
problem is solved)

Joseph Gooch
Sapphire Suite Product Manager
K12 Systems, Inc.
[...]
with a[...]
the[...]
than[...]
pound(at)apsis.ch.[...]