The Paste HTTP Server Thread Pool
=================================

This document describes how the thread pool in ``paste.httpserver``
works, and how it can adapt to problems.

Note all of the configuration parameters listed here are prefixed with
``threadpool_`` when running through a Paste Deploy configuration.

Error Cases
-----------

When a WSGI application is called, it's possible that it will block
indefinitely.  There's two basic ways you can manage threads:

* Start a thread on every request, close it down when the thread stops

* Start a pool of threads, and reuse those threads for subsequent
  requests

In both cases things go wrong -- if you start a thread every request
you will have an explosion of threads, and with it memory and a loss
of performance.  This can culminate in really high loads, swapping,
and the whole site grinds to a halt.

If you are using a pool of threads, all the threads can simply be used
up.  New requests go into a queue to be processed, but since that
queue never moves forward everyone will just block.  The site
basically freezes, though memory usage doesn't generally get worse.

Paste Thread Pool
-----------------

The thread pool in Paste has some options to walk the razor's edge
between the two techniques, and to try to respond usefully in most
cases.

The pool tracks all workers threads.  Threads can be in a few states:

* Idle, waiting for a request ("idle")

* Working on a request

  * For a reasonable amount of time ("busy")

  * For an unreasonably long amount of time ("hung")

* Thread that should die

  * An exception has been injected that should kill the thread, but it
    hasn't happened yet ("dying")

  * An exception has been injected, but the thread has persisted for
    an unreasonable amount of time ("zombie")

When a request comes in, if there are no idle worker threads waiting
then the server looks at the workers; all workers are busy or hung.
If too many are hung, another thread is opened up.  The limit is if
there are less than ``spawn_if_under`` busy threads.  So if you have
10 workers, ``spawn_if_under`` is 5, and there are 6 hung threads and
4 busy threads, another thread will be opened (bringing the number of
busy threads back to 5).  Later those threads may be collected again
if some of the threads become un-hung.  A thread is hung if it has
been working for longer than ``hung_thread_limit`` (default 30
seconds).

Every so often, the server will check all the threads for error
conditions.  This happens every ``hung_check_period`` requests
(default 100).  At this time if there are more than enough threads
(because of ``spawn_if_under``) some threads may be collected.  If any
threads have been working for longer than ``kill_thread_limit``
(default 1800 seconds, i.e., 30 minutes) then the thread will be
killed.

To kill a thread the ``ctypes`` module must be installed.  This will
raise an exception (``SystemExit``) in the thread, which should cause
the thread to stop.  It can take quite a while for this to actually
take effect, sometimes on the order of several minutes.  This uses a
non-public API (hence the ``ctypes`` requirement), and so it might not
work in all cases.  I've tried it in pure Python code and with a hung
socket, and in both cases it worked.  As soon as the thread is killed
(before it is actually dead) another worker is added to the pool.

If the killed thread lives longer than ``dying_thread_limit`` (default
300 seconds, 5 minutes) then it is considered a zombie.

Zombie threads are not handled specially unless you set
``max_zombies_before_die``.  If you set this and there are more than
this many zombie threads, then the entire process will be killed.
This is useful if you are running the server under some process
monitor, such as ``start-stop-daemon``, ``daemontools``, ``runit``, or
with ``paster serve --monitor``.  To make the process die, it may run
``os._exit``, which is considered an impolite way to exit a process
(akin to ``kill -9``).  It *will* try to run the functions registered
with ``atexit`` (except for the thread cleanup functions, which are
the ones which will block so long as there are living threads).

Notification
------------

If you set ``error_email`` (including setting it globally in a Paste
Deploy ``[DEFAULT]`` section) then you will be notified of two error
conditions: when hung threads are killed, and when the process is
killed due to too many zombie threads.

Missed Cases
------------

If you have a worker pool size of 10, and 11 slow or hung requests
come in, the first 10 will get handed off but the server won't know
yet that they will hang.  The last request will stay stuck in a queue
until another request comes in.  When a later request comes later
(after ``hung_thread_limit`` seconds) the server will notice the
problem and add more threads, and the 11th request will come through.

If a trickle of bad requests keeps coming in, the number of hung
threads will keep increasing.  At 100 the ``hung_check_period`` may
not clean them up fast enough.

Killing threads is not something Python really supports.  Corruption
of the process, memory leaks, or who knows what might occur.  For the
most part the threads seem to be killed in a fairly simple manner --
an exception is raised, and ``finally`` blocks do get executed.  But
this hasn't been tried much in production, so there's not much
experience with it.

watch_threads
-------------

If you want to see what's going on in your process, you can install
the application ``egg:Paste#watch_threads`` (in the
``paste.debug.watchthreads`` module).  This lets you see requests and
how long they have been running.  In Python 2.5 you can see tracebacks
of the running requests; before that you can only see request data
(URLs, User-Agent, etc).  If you set ``allow_kill = true`` then you
can also kill threads from the application.  The thread pool is
intended to run reliably without intervention, but this can help debug
problems or give you some feeling of what causes problems in the site.

This does open up privacy problems, as it gives you access to all the
request data in the site, including cookies, IP addresses, etc.  It
shouldn't be left on in a public setting.

socket_timeout
--------------

The HTTP server (not the thread pool) also accepts an argument
``socket_timeout``.  It is turned off by default.  You might find it
helpful to turn it on.