Date: Fri, 21 May 1999 23:45:08 -0400
Message-Id: <199905220345.XAA01544@Mail.MeepZor.Com>
From: Rodent of Unusual Size <Ken.Coar@Golux.Com>
To: Apache HTTP developers <new-httpd@Apache.Org>
Subject: [STATUS] (apache-apr) Fri May 21 23:45:08 EDT 1999
Sender: new-httpd-owner@apache.org
Precedence: bulk
Reply-To: new-httpd@apache.org

Apache Portable Runtime STATUS:
Last modified at [$Date: 1999/04/29 19:34:19 $]

Release:

    0.1  : In pre-alpha development

Plan:

    Everyone discuss and wrangle out the issues on new-httpd,
    and commit changes as appropriate.

Showstoppers:
Committed Code Changes:

    . docs/ directory created and initial function pre-spec added
    . include/ directory created and start of APRStatus macros added
    . pthreads/ directory tree created and populated with latest
      work on Apache 1.3.5-dev port to pthreads inserted

Available Patches:
In progress:
    Logging statements
        We have some comments along the lines of "log this here".
        Let's actually make those logs at some point. 

    Modules
        Ensure thread safety.  Does not include mod_proxy, yet.
	Modules that don't work:
            mod_digest.c     --Anybody have a client that can transmit Digest
			       Auth?

	Questionable modules: (Not tested.)         
	    mod_log_agent.c   -- Are we keeping these in 2.0 after Ken's
            mod_log_referer.c -- mod_log_config changes.  If yes, I'll test
			      -- them later.
    
    Warning scrubbing
        Get rid of all reasonable warnings with the strictest compiler flags,
        to help weed out bugs

    Everything

Needs patch:

    When the server runs at a high load, and the load later drops off,
    the server tries to kill off some children. The problem is that the
    children selected may have threads blocked on the accept lock, so
    they will never exit. This wasn't a problem with 1.3, because each
    process got kicked out of the lock with a signal. This could be solved
    (Thanks, Dean) with a server-wide pipe that replaces the SIGWINCH signal.
    This would wake up every thread when a single child death signal is sent
    over the pipe.

    Related to the above: On Red Hat 5.2, when we SIGTERM the server and
    the server is in this state, the children waiting on fcntl will not
    die, even though nobody else is holding the lock, verified with lsof.
    The children eventually will be SIGKILLed by the parent. This only
    happens in USE_MULTI_ACCEPT mode, and the problem goes away when
    USE_INTRAPROCESS_SERIALIZED_ACCEPT is undefined.  Dean has a couple of
    suggestions for tracing down this problem in
    <Pine.LNX.3.96dg4.990420221323.30703c-100000@twinlark.arctic.org>

Open issues:

  * Whether every APR routine should return a status code
  * What the status codes should look like
  * Bringing in the libap stuff

  * Optimize Optimize Optimize

  FEATURE SET FOR APR

Closed issues:
    Accept Thread Serialization
        Implemented as a blocking queue.  1 accept thread passes connection
        to queue (size of queue is config directive, default # of threads).
        All other threads block on queue.
    Process/Thread hybridization
        put monitor process back into place.  Was removed for initial port, 
	it's time we make sure the server is running, and if not, start 
	it again.
    Bringing this code up-to-date with the changes in the 1.3 branch.
        It's now up-to-date as of the morning of Feb 6th, 1999.
    Scoreboard
	Design the scoreboard appropriately for a hybrid process web
	server. This includes adding a tid field to the scoreboard.
    Signal Handling
        Investigate signal handling in Apache-pthreads.  SIGUSR1 is not a
	usable signal on Linux 2.0, because pthreads uses it to communicate
	between threads.  SIGWINCH was used for graceful restarts.
    Process/thread management
        Put (back) in the logic to manage the number of processes.
	Managing threads per process other than with a configuration
	constant is in doubt.
    Accept Serialization across Processes
	We now use one mutex per socket.
    Timeouts.  (only soft timeouts)
	Keepalive and Initail Conection timeouts are done.  Are there others?
	CGI?
    Modules that work:
        mod_env.c mod_negotiation.c mod_autoindex.c mod_dir.c mod_cgi.c 
        mod_actions.c mod_speling.c mod_userdir.c mod_alias.c 
        mod_expires.c mod_headers.c mod_so.c mod_setenvif.c 
        mod_auth_anon.c mod_access.c mod_include.c mod_imap.c 
        mod_log_config.c mod_asis.c mod_rewrite.c mod_mime.c 
        mod_usertrack.c mod_unique_id.c mod_cern_meta.c
        mod_auth.c (I'm assuming mod_auth* also work)
        mod_mime_magic.c mod_status.c mod_mmap_static.c
    Mod_proxy