epicserver 0.1

Epic Server
===========
Epic Server is a runtime/harness for running code that offers multi-process and
multi-machine execution. It handles aspects such as auto-discovery and
messaging and abstracts these in an easy to use for to allow you to focus on
writing the code to run your services.

Note: This is alpha software. The below talks about features coming in the 0.1
release, the project is in a runnable state at the moment but does not include
the multi machine/process execution at this time, this is expected to be
finished shortly before official release

Example
--------
from epicserver.objects import Entity

class GameEntity(Entity): pass

class Player(GameEntity):
score = 0
async enter_level(self, barrels):
for barrel in barrels:
# shoot the barrel
killed = await barrel.on_hit()
if killed
self.score += 1

async get_score(self):
return self.score

class Barrel(GameEntity):
team = "Society for the prevention of cruelty to barrels'
async on_hit(self):
print("What, did you expect me to explode?")
return False

async def setup(pool)
pool.bind(GameEntity)
barrels = [pool.Barrel[i] for i in range(30)]
player = pool.Player['Bob: Slayer of armoured cylinders']

await player.enter_level(barrels)
score = await player.get_score()

print(f"New High scrore for player: {score}")


Features
---------
* Based on raw async/await
* Inter object based messaging that looks like normal async method calls
* Automatic 'fabric' creation, stiching togther multiple processes on the same
or seperate machines.
* Automatic discovery of machines
* Virtual actors/objects, no explicit instantiation required
* Isolated and pluggable systems allow swapping in diffrent implementations
(eg Serialisation, networking)
* Automatic object percistance to disk (aka 'saving')
* Garbage colleciton of objects
* Python >= 3.6 only support


Use Cases
----------
* Distributed Game Backend
* Agent Simulation
* Communciation {latform (instant messaging, twiiter like systems)
* Graph based processing

Epic Server is ideal for anything that is messaging heavy and designed around
objects that need to communicate

Epix Server is not a 'drop in' solution to python to gain concurrency.
Applications seeking to exploit the parallelism of Epic Server will need to be
written to use it. Code that uses fork or threads can be adapted to work (by
making each fork/thread a separate object) however Epic Server is focused on
high levels of parallelism (executing contexts is 10s to 100s of times the number
of CPU's) and short lived 'calculations'

Inspriaration
--------------
Epicserver draws fromconcepts from many diffrent languages and libraries, a
short list is included below

* Orleans from microsoft research (dotnet/C#). Used as the machinary behind
Halo 5 and other games
* Erlang
* Stackless Python
* AsyncIO
* SansIO
* Zope/ZODB

While the concepts of many of the above can be ported straight to python,
modification of the concepts to make them feel fammilar to python programmers
where undertaken in an attempt to make the resulting framework more pythonic
and fit better with existing python ecosystems.


Implimentation
--------------
Epic Server consists of an event loop running in the main thread that is
responsible for the execution of work, as well as several background IO threads
responsible for sending and reciving IO requests as well as handling IO (both
disk and network). Additional threads can be utilized for other blocking calls
that may arrise such as Database access.

All access to thes IO threads is dont via message passing and mediated by the
event loop

IO Loop
========
Epic Server is implemented as an IO Loop with a number of 'syscalls' that
correspond to high level concepts. Rather than focusing on how to perform the
IO, Epic Server presents primitives such as 'persist to database' or 'Call
remote object' leaving the implementation up to the server and allowing you to
avoid having IO routines embedded in your business logic that would need to be
mocked out for testing.

By moving the IO out of the program and replacing it with a token based syscall
system, one can simply and quickly test buisness logic without havign to mock
out te IO routines or simulate services. Testing becomes a simple process of
sending the code messages and checing the repsponses.

Moving the IO out of the code also allows high performance implemetnations such
as TIPC or ininiband to be plugged in without modifcation of code, this applies
to which database is in use or the searlisation library allowing futher updates
to features and performance without having to rework (or test) code.

Concurrency
============
A number of the implementations this project was inspired by have very explict
ideas in regards to the concurrency model and interleaving of requests, a short
summary is below:

* Orleans: Objects are single threaded, only one request may be performed on
an object at a time
* Erlang: Actors are lite-weight processes that communicate via mailboxes
* Stackless Python: Follows CPython/GIL Semantics
* AsyncIO: Cooperative multitasking, yields control flow at yield/await
statements, Guaranteed exclusive execution between these two points (Providing
threading is not used)

Epic Server is composed of multiple Python processes that may be running on the
same or different machines. Each Python process has one 'execution' thread that
corresponds to the main thread as well as a number of background threads for IO
(DB access, sending requests and receiving requests)..

From the point of view of user code, a method call has exclusive access tot he
CPU between 'yield points' (yield/await). At these points, the running context
may be changed and other code may be executed. This is identical to AsyncIOs
guarantees and a lot less restrictive (and dangerous) than other frameworks
such as Orleans while being more in line with what python programmers expect.

In contrast to the Orleans approach the above means that multiple instances of
the same method call can be executing and will be interleaved. Should method
level or object level locking be required to introduce Orleans like semantics
be required then a decorator will need to be used to ensure this.

The Use of IO threads should not interfere with the execution thread as these
are interacted with via message passing and mediated by the event loop.
Mutation of the messages after being 'handed off' however may cause
undesirable effects and as such all 'sent' messages should not be further
modified and references to them dropped.

As an object can only be running in a single process at a time, Parallelism can
be achieved by executing multiple different objects in multiple processes at
once. Spreading the work over multiple objects can be used to achieve speedups