httpd-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Michael Handler <handler-list-new-ht...@grendel.net>
Subject Re: daemontools/foreground support in 1.3.*
Date Thu, 28 Feb 2002 09:00:34 GMT
Sorry for the lack of quick response, I had to go home for a family
funeral.

Justin Erenkrantz <jerenkrantz@ebuilt.com> writes:

> -0.  I personally believe that this shouldn't be backported.  If
> you want this, you should use 2.0.  Others will disagree vehemently
> though, and you may indeed garner enough +1s to get it in.
> 
> IMHO, 1.3 should be in bug-fix mode only.  No new features - as
> witnessed by our addition of AcceptMutex which screwed up 1.3.23
> on Solaris.  If you need this functionality, why not use 2.0 and
> help us get 2.0 to be GA?

I completely understand the desire to not to introduce substantial
changes into 1.3.* at this point, as well as encouraging people to
test the stability and correctness of 2.0. However:

* 2.0 is not ready for production deployment, and arguably won't
  be for some time. Thus, we need to keep running 1.3 in production
  until that time, and will need minor support to keep it running
  optimally in our environments. I'm perfectly happy to deploy 2.0
  in testing environments, and the daemontools support will make
  that abundantly easier, but I still need 1.3.* right now, and
  that means patching every release each time to run under daemontools.

  I first wrote this patch for 1.3.14 when I was doing the first
  software deploy at my current employer (August 2000). I thought
  about contributing it back to Apache then, but decided against
  it, since I was (mistakenly) informed that 2.0 already had this
  functionality, and I thought that 2.0 would be production quality
  "soon".
  
* This is a very minor change, on the order of 20 lines of code. I've
  personally been running this patch in production since 1.3.14,
  and I've found patches on the Internet that date back to 1.3.11
  and earlier, so there is clearly historical demand for this
  functionality, as well as reason to believe that the current patch
  is functional and correct.

  Traffic on the daemontools mailing list and the number of people
  who have downloaded the patches from my web server since I have
  announced them here and elsewhere shows that there is clearly
  substantial interest in daemontools and running all available
  server processes under supervise and svscan, including Apache's
  httpd. Yes, nothing stops me from distributing this functionality
  as a patch in the contrib tarball for each release, or from my
  web page, but that also means that binary packages or OS distributions
  most likely won't have this trivial but extremely useful bit of
  functionality included, and that means each binary maintainer
  either has to be lobbied to include this, or each sysadmin has
  to compile their own binaries, which strikes me as clear inefficiency
  that could easily be avoided.

This is a tiny change, and the functionality already exists in 2.0.
Lots of sysadmins are going to be stuck with 1.3.* for a while from
now, and this is a clear benefit to those of us who utilize
daemontools, without causing any harm to anyone.

I've had to write patches for several server applications to get
them running cleanly under daemontools, as have many other sysadmins,
but thanks to the work of Jos Backus and others, they've been
integrated back into the mainline of the distribution (rsync and
SAMBA come to mind as two recent examples). The 1.3.* Apache tree
is the final outlier in this regard in my common set of UNIX server
applications, and it would save many people work in the future to
have this integrated into the next 1.3.* release.

Am I reaching anyone here, or should I just be quiet?

-- 
handler@grendel.net (michael handler)                           washington, dc

Mime
View raw message