kamaelia europython tutorial

http://www.kamaelia.org/PragmaticConcurrency sparks.m@gmail.com

Kamaelia: Pragmatic Concurrency

A Tutorial

Michael SparksEuropython '09, Birmingham UK

About me• - Been using python for several years• - Always been interested in concurrency

• - Kamaelia aims to embody safe practices.

• - Work at BBC R&D's Northern Lab, based in Manchester, moving to Media City:UK in 2011

• - Kamaelia is born out of a variety of R&D projects, and shaped by needs, not aesthetics or purity.

• Kamaelia is not an active R&D project. It is used in R&D projects, and hence actively maintained.

About me•

Disclaimer: Like you, I'm doing this on my time, not the BBC's. This doesn't represent BBC opinion on anything.

About Kamaelia• - Also born from a desire to make concurrency

in programs easier to work with• - Because it's nice in the shell – I want software

concurrency that easy :-)• - Expressiveness is favoured over performance,

but not to preclude optimisation• - Means we're cautious about adding syntactic

sugar.•

About Kamaelia• - Adapts Unix Philosophy to make a program

concurrent internally, but with the purpose of simplifying maintenance.

• Unix Philosophy:• Write programs that do one thing and do it well.

Write programs to work together.Write programs to handle text streams, because that is a universal interface.

• --Doug McIlroy•

About Kamaelia• - Adapts Unix Philosophy to make a program

concurrent internally, but with the purpose of simplifying maintenance.

• Kamaelia Philosophy:• Write components that do one thing and do it well.

Write components to work together. Write components to handle python object streams, because that is a universal interface.

• --With apologies to Doug McIlroy•

What we're covering• - An overview of Kamaelia

• & it's view on concurrency• - How to build a mini Kamaelia

• – to get under the hood.• - Building components & systems• - Examining larger systems, and debugging.• - Building a large(ish) system

Time Estimates• First part:• - An overview of Kamaelia (lightning talkstyle : 5-10 mins)• - How to build a mini Kamaelia (exercise 40-60 mins)• - Starting building components & systems (remainder)

After break:

• - More advanced stuff (demo/etc 20 mins)• - Larger systems and debugging. (20 mins)• - Building a large(ish) system (exercise 40 mins)

Format• I generally welcome questions at any point• That said...

• 'Except in the overview• - the rest of this morning should answer them!!

• Copious notes provided - covers more than today

• Materials available from URL below• Is a mixture of “explain then do”.

Perhaps hold questions for during “do” :-) ?

Caveat•

• * First time I've given this tutorial !• * If we run short of time on any section, we'll

skip ahead.•

• * BUT, have copious notes that cover everything.

Questions?

• ... before we dive in?•

Part 1•

• First part:• - An overview of Kamaelia (lightning talk style : 5-10

mins)• - How to build a mini Kamaelia (exercise 40-60 mins)• - Starting building components & systems (remainder)

Kamaelia, a 20:20 overview

20:20 PresentationA presentation style based on the “Pecha Kucha”style 20 slides, 20 seconds each(Similar to a lightning talk)

Hardware finally going massively concurrent ...

.... PS3, high end servers, trickling down to desktops, laptops)

“And one language to inthe darkness bind them”... can just we REALLY abandon 50 years of code for Erlang, Haskelland occam?

Opportunity!

Problems

“many hands make lightwork” but Viewed as Hard... do we just have crap tools?

Missing Something?Fundamental Control Structures... in imperative languages number greater than 3! (despite what you get taught...!)

Control Structure Traditional Abstraction Biggest Pain Points

Sequence Function Global VarSelection Function Global VarIteration Function Global VarParallel Thread Shared Data

Usually Skipped Lost or duplicate update are most common bugs

Regarding using concurrency, what sort of applications are we

talking about here?

Desktop

Network

Novice

3rd Party

trainee

Desktop

Network

Novice

3rd Party

P2P Whiteboard

ER DB Modeller Kids

Programming(logo) Simple

Speak 'n Write

UGCBackend

Transcoder

Think backendneeded foryoutube/flickrtype systems

Compose

Shot ChangeDetection

MobileReframing

Macro“record

everything”

Podcasts

Email &SpamSMTP

GreylistingPop3Proxy ClientSide

Spam Tools

IRCWeb

Serving

trainee

AWS(Amazon)

SednaXMLDBXMPP

pubsubQt

Gtkmicroblogging

MiniAxonScriptReaderMediaPreviewon MobileReliableMulticast

P2P RadioTorrent3D SystemsRealtime MusicPaint AppP2P Web ServerSecure “phone”Social Net Vis...

Core Approach: Concurrent things with comms points Generally send messages Keep data private, don't share

control

outbox

signal

control

outbox

signal

But I must share data?Use Software Transactional Memoryie version control for variables.

1. Check out the collection of values you wish to work on2. Change them3. Check them back in 4. If conflict/clash, go back to 1

Perspectives in APIs! (1/2)

control

outbox

signal

1st Person - I change my state

2nd Person – YOUwant to me to dosomething(you sendme a message)

3rd Person –Bla shoulddo something(I send a message)

If you have concurrency it becomes natural to think in terms of 1st 2nd and 3rd person. This affects an API's structure, and can be vital for understanding it!

This is one we've found that makes sense

control

outbox

signal

private real methods

Messagesfrom publicinboxes

Messages sentto public outboxes

Perspectives in APIs! (2/2)

Also, thinkabout stdin

Also, thinkabout stdout

If you have concurrency it becomes natural to think in terms of 1st 2nd and 3rd person. This affects an API's structure, and can be vital for understanding it!

This is one we've found that makes sense

control

private real methods

Messagesfrom publicinboxes

Actor SystemsDistinction can be unclear,potential source of ambiguity*

No outbox conceptPossible issues withrate limiting*Hardcodes recipientin the sender

*system dependent issue

Advantages of outboxes

control

outbox

signal

No hardcoding of recipientallows: - Late Binding - Dynamic rewiring

Concurrency Patterns as Reusable Code ... a concurrency DSL

A Core Concurrency DSLPipeline(A,B,C)Graphline(A=A,B=B, C=C, linkages = {})Tpipe(cond, C)Seq(A,B,C), PAR(), ALT()Backplane(“name”), PublishTo(“name”), SubscribeTo(“name”)Carousel(...)PureTransformer(...)StatefulTransformer(...)PureServer(...)MessageDemuxer(...)Source(*messages)NullSink

Some of these are work in progress – they've been identified as useful, but not implemented as chassis, yet

Pipeline ExamplePipeline( MyGamesEventsComponent(up="p", down="l", left="a", right="s"), BasicSprite("cat.png", name = "cat", border=40),).activate() MyGames

EventsComponent

BasicSprite

Graphline ExampleGraphline( NEXT = Button(...), PREVIOUS = Button(...), FIRST = Button(...), LAST = Button(...), CHOOSER = Chooser(...), IMAGE = Image(...), ...).run()

FIRST(button)

Chooser

LAST(button)

PREVIOUS(button)

NEXT(button)

Server Example

MainServer Core

Remote User

Protocol Handler Factory

Socket handler

Protocol handler

datato

datafromuser

Created at runtime to handle the connection

Server Example

MainServer Core

Protocol Handler Factory

You therefore need to provide this bit.

Server Examplefrom Kamaelia.Chassis.ConnectedServer import ServerCorefrom Kamaelia.Util.PureTransformer import PureTransformer

def greeter(*argv, **argd): return PureTransformer(lambda x: "hello" +x)

class GreeterServer(ServerCore): protocol=greeter port=1601

GreeterServer().run()

Backplane Example# Streaming Server for raw DVB of Radio 1 Backplane(“Radio”).activate()

Pipeline( DVB_Multiplex(850.16, [6210], feparams), # RADIO ONE PublishTo("RADIO"),).activate()

def radio(*argv,**argd): return SubscribeTo(“RADIO”)

ServerCore(protocol=radio, port=1600).run()

So that's the 5 minute version

Short Q&A before we move on?

Part 1•

Mini Axon• Kamaelia is divided into two halves• * One part handles all the concurrency

stuff, providing you a component model• * The other is a large collection of

components and some apps using them.

Mini Axon

• Axon is the part that handles concurrency and provides the component model, and is the key to understanding why & how Kamaelia works.

Mini Axon

• ... is a collection of exercises where you build just such a beast.

Mini Axon•

• Generators* Python's smallest unit of concurrency

• Microprocesses* Generators with context

• Scheduler* Something to run lots of microprocesses

• Components* A microprocess with boxes (treated as in & outboxes)

• Postman* Something to do deliveries

Mini Axon•

Generators• * Python's smallest unit of concurrency • * Single function you call, get a

generator back• * Can then call it's .next() method to

* Get a new value from it * Give it CPU time

Fibonacci Generator• def fib(a,b):

while 1: yield a a, b = b, b + a

• Demo

Fibonacci Generator• >>> def fib(a,b):

... while 1:

... yield a

... a, b = b, b + a

...>>> g = fib(1,1)>>> g<generator object at 0xb7bf9c6c>>>> [ g.next() for _ in range(10) ][1, 1, 2, 3, 5, 8, 13, 21, 34, 55]

Lots of Generators• Using the same fib generator, make 10 of them:

>>> GS = [ fib(x,x) for x in range(10) ]

• And “run” them:>>> [ G.next() for G in GS ][0, 1, 2, 3, 4, 5, 6, 7, 8, 9]>>> [ G.next() for G in GS ][0, 1, 2, 3, 4, 5, 6, 7, 8, 9]>>> [ G.next() for G in GS ][0, 2, 4, 6, 8, 10, 12, 14, 16, 18]>>> [ G.next() for G in GS ][0, 3, 6, 9, 12, 15, 18, 21, 24, 27]>>> [ G.next() for G in GS ][0, 5, 10, 15, 20, 25, 30, 35, 40, 45]

Generators as co-routines• def fib(a,b):

while 1: yield 1 # Just to say “keep running me” print a a, b = b, b + a

• Demo

Generators as co-routines• def printer(tag):

while 1: yield 1 # Makes it a generator print tag

• Demo

Mini Axon•

Microprocesses• * Generators with context•

microprocess exercise• Write a class microprocess with methods:

• __init__(self)* Takes no arguments

* Uses super to call superclass __init__ method• main(self)

* No arguments * Should yield 1

microprocess answer•

• class microprocess(object): def __init__(self): super(microprocess, self).__init__() def main(self): yield 1

Generally, we'll skip answers in this presentation, they're all in the web version of this tutorial here:

• http://www.kamaelia.org/MiniAxon/

• (and in the notes :-)

microprocess usage• class printer(microprocess):

def __init__(self, tag): super(printer, self).__init__() self.tag = tag def main(self): while 1: yield 1 print self.tag

• “Printer” isn't particularly interesting, but allows things like components, but let's see how this can be used.

microprocess usage 2• >>> X = printer("Something")

>>> G = X.main()>>> X,G(<__main__.printer object at 0xb7bfd2ac>, <generator object at 0xb7bfd4cc>)>>> G.next()1>>> G.next()Something1>>> X.tag = "Something else">>> G.next()Something else1

Mini Axon•

Scheduler•

* Something to run lots of microprocesses•

scheduler exercise• Write a class scheduler with 3 methods:

• __init__(self)* Uses super to call superclass __init__ method

* subclasses microprocess * Creates 2 queues – active & newqueue

• main(self)* Is a generator

* Runs the microprocesses activated• activateMicroprocess(self, someprocess): * Calls somprocess.main() * Adds generator to newqueue

scheduler exercise• Scheduler logic

• main(self)* loops 100 times, yields 1 at start of loop

* loop through generators in self.active * call their .next() method * If result is not -1 and no StopIteration, append to newqueue * at end of loop, newqueue becomes active * newqueue reset to empty list

scheduler usage• Using same printer class..

>>> X = printer("Hello World")>>> Y = printer("Game Over") # :-)

• >>> myscheduler = scheduler()>>> myscheduler.activateMicroprocess(X)>>> myscheduler.activateMicroprocess(Y)

• >>> for _ in myscheduler.main():... pass

• <prints Hello world/Game over repeatedly>

Mini Axon•

Component•

• * microprocess with a standard interface.• boxes as inboxes/outboxes

component exercise• Write a class component subclass of microprocess with 4 methods:

• __init__(self)

send(self, value, outboxname)

recv(self, inboxname)

dataReady(self, inboxname)

• Behaviour coming up!

component exercise 1/4• __init__(self) logic:

* Add an attribute self.boxes , default value:

{ “inbox”: [], “outbox”: [] }• * Ensure you call the superclass

__init__ method appropriately

component exercise 2/4• send(self,value, outboxname) logic:

Finds the list named outboxname in self.boxes, and appends value to it.

• Before: X.send(“hello”, “outbox”) { “inbox”: [], “outbox”: [] }

• After:

{ “inbox”: [], “outbox”: [“hello”] }•

component exercise 3/4• recv(self, inboxname) logic:

Finds the list named inboxname in self.boxes, and pops the first value

• Given:

{ “inbox”: [“hello”, “world”], “outbox”: [] }• self.recv(“inbox”) returns, “hello” leaving...

{ “inbox”: [“world”], “outbox”: [] }•

component exercise 4/4•

dataReady(self, inboxname) logic: Finds the list named inboxname in self.boxes: returns the length of the list.

• Given:

{ “inbox”: [“hello”, “world”], “outbox”: [] }• dataReady(“inbox”) --> 2

(allows if self.dataReady(“inbox”) )

component usage•

Until we add a means for data to get from outboxes to inboxes, using components is no more interesting than microprocesses

Mini Axon•

Postman• * Something to do deliveries•

• Note: This is more conceptual in real Axon,

but was real in v. old Axon. ie real Axon is more efficient!

postman exercise• Write a class postman subclass of microprocess

with 2 methods:• __init__(self, source, sourcebox, sink, sinkbox)

main(self)

• Behaviour coming up!

postman exercise 1/4•

__init__(self, source, sourcebox,

sink, sinkbox) logic: * Copy all the arguments as local attributes

• * Ensure you call the superclass __init__ method

appropriately!

postman exercise 2/4•

main(self) logic:

• In a loop:• yield a non -1 value (eg 1)

Check if source's sourcebox has dataReady

• If it has, use recv to collect it from there, and sink's send method to deliver it.

Mini Axon•

• Using it!

Producer/Consumer 1/3•

• class Producer(component): def __init__(self, message): super(Producer, self).__init__() self.message = message

• def main(self): while 1: yield 1 self.send(self.message, "outbox")

• class Consumer(component): def main(self): count = 0 while 1: yield 1 count += 1 if self.dataReady("inbox"): data = self.recv("inbox") print data, count

• p = Producer("Hello World")c = Consumer()postie = postman(p, "outbox", c, "inbox")

• myscheduler = scheduler()myscheduler.activateMicroprocess(p)myscheduler.activateMicroprocess(c)myscheduler.activateMicroprocess(postie)

• for _ in myscheduler.main(): pass

Producer/Consumer Output•

• Hello World 2Hello World 3Hello World 4

• ...

• Hello World 96Hello World 97Hello World 98

• Not 100, because of yields before start of scheduler loop. (scheduler terminates early)

Mini Axon•

• Using it for more useful stuff

File Reader•

• class FileReader(component):

• def __init__(self, filename): super(FileReader, self).__init__() self.file = open(filename, "rb",0)

• def main(self): yield 1 for line in self.file.xreadlines(): self.send(line, "outbox") yield 1

Multicast sender 1/2• class Multicast_sender(component):

def __init__(self, laddr, lport, daddr, dport): super(Multicast_sender, self).__init__() self.laddr = laddr self.lport = lport self.daddr = daddr self.dport = dport

... continued

Multicast sender 2/2• class Multicast_sender(component):

... continued

def main(self): sock = socket.socket(socket.AF_INET, socket.SOCK_DGRAM, socket.IPPROTO_UDP) sock.bind((self.laddr,self.lport)) sock.setsockopt(socket.IPPROTO_IP, socket.IP_MULTICAST_TTL, 10) while 1: if self.dataReady("inbox"): data = self.recv("inbox") l = sock.sendto(data, (self.daddr,self.dport) ) yield 1

FileReader -> MulticastSender•

• reader = FileReader("Ulysses")sender = Multicast_sender("0.0.0.0", 0, "224.168.2.9", 1600)postie = Postman(reader, "outbox", sender, "inbox")

• myscheduler = scheduler()myscheduler.activateMicroprocess(reader)myscheduler.activateMicroprocess(sender)myscheduler.activateMicroprocess(postie)

• for _ in myscheduler.main(): pass

Mini Axon --> Axon•

• The rest is:Syntactic Sugar

Ways of using it Optimisations (eg direct delivery – no postman) + a couple of other ideas we'll come to (STM, threadedcomponents, services, processcomponents)

Questions?•

• ... before we look how to build some real components, and how to use them?

Part 1•

Building a Video Recorder•

• Before the break, we'll build a simple video recorder.• Approach:• 1 Build a webcam• 2 Clean up the code for “normal” reuse• 3 Componentise in least effort manner• 4 Separate input/transform/display parts• 5 Hook webcam up to a dirac encoder and filewriter•

Step 1: Build a simple Webcam• Pygame 1.9.1 alpha has basic Linux webcam support

which works nicely, so we're using that.• First some initialisation

• import pygameimport pygame.camera

• pygame.init()pygame.camera.init()

Step 1: Build a simple Webcam• The some definitions:

• displaysize = (800, 600)capturesize = (640, 480)imagesize = (352, 288)imageorigin = (0,0)device = “/dev/video0”

Step 1: Build a simple Webcam• Initialise the display, allocate a camera, and activate it

• display = pygame.display.set_mode(displaysize)camera = pygame.camera.Camera(device, capturesize)camera.start()

Step 1: Build a simple Webcam• Loop round capturing images and display them

• while 1: snapshot = camera.get_image() snapshot = pygame.transform.scale(snapshot, imagesize) display.blit(snapshot, imageorigin) pygame.display.flip()

• Demo!

Step 2: Clean up for reuse• Given a collection of config options, wrapping this in a

class makes sense.• As before these parts are unchanged:

• import pygameimport pygame.camera

Step 2: Clean up for reuse• We then define a class, and put the config options as

class attributes:• class VideoCapturePlayer(object):• displaysize = (800, 600) capturesize = (640, 480) imagesize = (352, 288) imageorigin = (0,0) device = “/dev/video0”

Step 2: Clean up for reuse• Put our initialisation in the __init__. Allow the user to

override our defaults

def __init__(self, **argd): self.__dict__.update(**argd) super(VideoCapturePlayer,self).__init__() self.display = pygame.display.set_mode(self.displaysize) self.camera = pygame.camera.Camera(self.device, self.capturesize) self.camera.start()

Step 2: Clean up for reuse• Wrap up body of loop in a method

def get_and_flip(self): snapshot = self.camera.get_image() snapshot = pygame.transform.scale(snapshot, self.imagesize) self.display.blit(snapshot, self.imageorigin) pygame.display.flip()

Step 2: Clean up for reuse• Provide a hook to start it going, wrapping the main loop.

def main(self): while 1: self.get_and_flip()

• Then run it!

VideoCapturePlayer().main()

• Demo!

Step 3: Componentise•

• In this case, least effort approach is to use a threaded component. Would could make it a generator component later if needed.

• Changes:• * imports

* class's baseclass / extraction of display_flip into a method. (to simplify later integration with existing components)* how we run it.

• Imports change to

import pygameimport pygame.camerafrom Axon.ThreadedComponent import threadedcomponent

• Baseclass changes & initialiser changes:

class VideoCapturePlayer(threadedcomponent):... def __init__(self, **argd): # no longer update __dict__ here super(VideoCapturePlayer,self).__init__(**argd)...

• Extract display flipping out to a separate method:

def pygame_display_flip(self): pygame.display.flip()

def get_and_flip(self): snapshot = self.camera.get_image() snapshot = pygame.transform.scale(snapshot, self.imagesize) self.display.blit(snapshot, self.imageorigin) self.pygame_display_flip()

• Change to how we run:

• We had:• VideoCapturePlayer().main()

• We now:• VideoCapturePlayer().run()

• Demo!

Step 4: Split into input & output•

• Now we can split the component in two:• * VideoCaptureSource• * Surface Displayer

• The Video capture source now needs to be self regulating,so it needs to sleep during the loop, relative to a target frame rate.

• VideoCaptureSource class preamble:• import timeclass VideoCaptureSource(threadedcomponent): capturesize = (352, 288) delay = 0 fps = -1 device = “/dev/video0”

• VideoCaptureSource initialiser• def __init__(self, **argd): super(VideoCaptureSource, self).__init__(**argd) self.camera = pygame.camera.Camera(self.device, self.capturesize) self.camera.start() if self.fps != -1: self.delay = 1.0/self.fps self.snapshot = None

• Main loop body now just captures images, capturing a reference

• def capture_one(self):• self.snapshot = None self.snapshot = self.camera.get_image()

• Main loop still runs it, and sends the image out an outbox, and then sleeps.

• def main(self):• while 1: self.capture_one() self.send(self.snapshot, “outbox”) time.sleep(self.delay)

• The surface displayer takes the other code chunks• from Axon.Component import component• class SurfaceDisplayer(component): displaysize = (800,600) imagesize = (352, 288) imageorigin = (0,0)

• Has it's own initialiser...• def __init__(self, **argd):

super(SurfaceDisplayer, self).__init__(**argd) self.display = pygame.display.set_mode(self.displaysize)

Retains the following method unchanged:• def pygame_display_flip(self):

pygame.display.slip()

Step 4: Split into input & output• Then mainbody waits for surfaces to display, sleeping

when there isn't any:• def main(self):

• while 1: while self.dataReady(“inbox”): snapshot = self.recv(“inbox”) snapshot = pygame.transform.scale(snapshot, self.imagesize) self.display.blit(snapshot, self.imageorigin)

• while not self.anyReady(): self.pause() yield 1

• yield 1

• We then join these back together in a pipeline:• from Kamaelia.Chassis.Pipeline import Pipeline

• Pipeline(

• VideoCaptureSource(),

• SurfaceDisplayer(),

• ).run()

• Demo!

Step 5: Recording• Working back...•

• We want to write to a file... (SimpleWriter)• We want to write dirac encoded video DiracEncoder• That expects YUV Frames (ToYUV420_planar)• That expects RGB data + metadata

PureTransformation of somr RGB data• Which needs a picture source

• Which is where we started.

Step 5: Recording• Our imports therefore need to add:•

• from Kamaelia.File.Writing import SimpleFileWriter• from Kamaelia.Codec.Dirac import DiracEncoder• from Kamaelia.Video.PixFormatConversion import ToYUV420_planar• from Kamaelia.Util.PureTransformer import PureTransformer•

• And we delete everything related to display

Step 5: Recording• We then need to join it all together:• Pipeline(

• VideoCaptureSource(), PureTransformer(lambda F : \ {"rgb" : pygame.image.tostring(F, "RGB"), "size" : (352, 288), "pixformat" : "RGB_interleaved", }), ToYUV420_planar(), DiracEncoder(preset="CIF", encParams={"num_L1":0}), SimpleFileWriter("X.drc"),

• ).run()

Step 5: Recording•

• This then records dirac encoded video, which we can now playback with a simple dirac player!

• Demo !•

Questions?• ... before we break?•

End of Part 1•

• Before the second half, download and install Kamaelia from the link below, if you haven't already.

• http://tinyurl.com/kot49x •

Kamaelia: Pragmatic Concurrency

A Tutorial

Michael SparksEuropython '09, Birmingham UK

Part 2•

• Second part:• - More advanced stuff (demo/etc 20 mins)• - Larger systems, embedding and debugging. (20 mins)• - Building a large(ish) system (exercise 40 mins)

Questions?• ... before we carry on?•

Part 2•

• Second part:• - More advanced stuff (demo/etc 20 mins)• - Larger systems and debugging. (20 mins)• - Building a large(ish) system (exercise 40 mins)

More advanced Stuff• Have seen how to build a basic component and a basic

system - covers a wide set of problems. Now we broaden the scope.

• If we have time we'll come back to Kamaelia's STM model & concepts of services, and (experimental) multiple process support.

Well behaved Shutdown• Components should expect to receive these messages on their

“control” inbox, and behave as follows:

• Axon.Ipc.shutdownMicroprocess – if you receive this, you are expected to shutdown immediately, and to pass this message on.

• Axon.Ipc.producerFinished – if you receive this, someone sending you data has shutdown. You may want to shutdown depending on your application's logic. You may process all outstanding messages in this case. You may wish to pass this message on, or change it to shutdownMicroprocess if it was unexpected.

• Due to different component needs no syntactic sugar exists for this, but common cases are being discussed at present.

Components• Inheritable default values – we've already seen this!• class VideoCapturePlayer(threadedcomponent):

displaysize = (800, 600) capturesize = (640, 480) imagesize = (352, 288) imageorigin = (0,0) device = "/dev/video0"

• Also usable:• VideoCapturePlayer(device = "/dev/video0")

VideoCapturePlayer(device = "/dev/video1")

Components• Inheritable default values – other places:

class ServerCore(component): port=1601 protocol=None socketOptions=None TCPS=TCPServer ....

Components• Inheritable default values:

class MailServer(ServerCore): port=25 protocol=SMTPProtocol

• Usage:• MailServer().run()

MailServer(port=8025).run()MailServer(socketOptions=...).run()

Components• Inheritable default values - nests

class MyReconfiguredMailServer(MailServer): port=25 protocol=SMTPProtocol socketOptions = socketOptions=(socket.SOL_SOCKET, socket.SO_REUSEADDR, 1) class TCPS(TCPServer): CSA=MyDebugWrapper(ConnectedSocketAdapter)

• Allows for deep reconfiguration of components, without rewriting.

Components• Inheritable default values – real usage – Kamaelia Grey

class GreylistServer(MoreComplexServer): logfile = config["greylist_log"] debuglogfile = config["greylist_debuglog"] socketOptions=(socket.SOL_SOCKET, socket.SO_REUSEADDR, 1) port = config["port"] class TCPS(TCPServer): CSA = NoActivityTimeout(ConnectedSocketAdapter, timeout=config["inactivity_timeout"], debug=False) class protocol(GreyListingPolicy): servername = config["servername"] serverid = config["serverid"] smtp_ip = config["smtp_ip"] smtp_port = config["smtp_port"] allowed_senders = config["allowed_senders"] allowed_sender_nets = config["allowed_sender_nets"] ...

Pipelines• Seen how Pipelines can join components together.• However also can do:

• Pipeline( Lsystem(), ConsoleEchoer(tag="\n", forwarder=True), Damage(),

circular = True,).run()

• To enable a feedback loop. (Demo)

Pipelines• Transcode Engine• Pipeline(

DirectoryWatcher(watch = conf["main_incoming_queue"]), ImageMover(destdir = conf["image_queue"]),).activate()

• Pipeline( DirectoryWatcher(watch = conf["image_queue"]), ImageTranscoder(destdir = conf["image_moderation_queue"]),).activate()

• Pipeline( DirectoryWatcher(watch = conf["main_incoming_queue"]), VideoMover(destdir = conf["video_queue"]),).activate()

• Pipeline( DirectoryWatcher(watch = conf["video_queue"]), VideoTranscoder(destdir = conf["video_moderation_queue"]),).run()

Pipelines• Reusable Dirac Video Source:• def DiracWebCam():

return Pipeline( VideoCaptureSource(), PureTransformer(lambda F : \ {"rgb" : pygame.image.tostring(F, "RGB"), "size" : (352, 288), "pixformat" : "RGB_interleaved", }), ToYUV420_planar(), DiracEncoder(preset="CIF", encParams={"num_L1":0}), )

Graphlines• Graphlines are like pipelines, but any shape. For

example, a simple presentation tool:• Graphline(

CHOOSER = Chooser(items = files), IMAGE = Image(size=(800,600), position=(8,48)), NEXT = Button(caption="Next", msg="NEXT", position=(72,8)), PREVIOUS = Button(caption="Previous", msg="PREV", position=(8,8)), FIRST = Button(caption="First", msg="FIRST",position=(256,8)), LAST = Button(caption="Last", msg="LAST",position=(320,8)), linkages = { ("NEXT","outbox") : ("CHOOSER","inbox"), ("PREVIOUS","outbox") : ("CHOOSER","inbox"), ("FIRST","outbox") : ("CHOOSER","inbox"), ("LAST","outbox") : ("CHOOSER","inbox"), ("CHOOSER","outbox") : ("IMAGE","inbox"), }).run()

Graphlines•

• * Swiss Army Knife of Kamaelia components• * Allows almost any topology• - BUT an outbox may only link to one inbox• - many outbox may link to one inbox• * Can contain any component – including pipelines and

graphlines•

Graphlines• SMS Outbound processor• Graphline(

SMS_SOURCE = Pipeline( DirectoryWatcher(watch = "outgoingsms"), FilesToProcessSource(), FileSlurper(tuplemode=True), ), CLEANER = Pipeline( PureTransformer(lambda (x,y): y ), Mover(), ), SENDER = Pipeline( PureTransformer(lambda (x,y): cjson.decode(x) ), SMSSender(), ), SPLIT = TwoWaySplitter(),

Graphlines• SMS Outbound processor (cont)• Linkages = {

("SMS_SOURCE","outbox") : ("SPLIT","inbox"), ("SPLIT", "outbox") : ("CLEANER", "inbox"), ("SPLIT", "outbox2") : ("SENDER", "inbox"), }).run()

• Linkages to or from “self” (or “”) link to the Graphline's boxes.• Graphline(

SPLIT = SomeSplitter() MERGE = SomeMerger() P1 = SomeTransformer() P2 = SomeTransformer() P3 = SomeTransformer() linkages { (“self”, “inbox”) : (“SPLIT”, “inbox”), (“MERGE”, “outbox”) : (“self”, “outbox”), ... }

• Allows graphlines to nest, and higher level abstractions

Graphlines

• Whiteboard using nesting graphlines:• Pipeline(

chunks_to_lines(), lines_to_tokenlists(), Graphline( ROUTER = Router( ((lambda T : T[0]=="SOUND"), "audio"), ((lambda T : T[0]!="SOUND"), "whiteboard"), ), WHITEBOARD = FilteringPubsubBackplane(whiteboardBackplane), AUDIO = Pipeline( SimpleDetupler(1), # remove 'SOUND' tag SpeexDecode(3), FilteringPubsubBackplane(audioBackplane, dontRemoveTag=True), RawAudioMixer(), SpeexEncode(3), Entuple(prefix=["SOUND"],postfix=[]), ),

Graphlines

• Whiteboard using nesting graphlines:• Linkages = {

# incoming messages go to a router ("", "inbox") : ("ROUTER", "inbox"),

• # distribute messages to appropriate destinations ("ROUTER", "audio") : ("AUDIO", "inbox"), ("ROUTER", "whiteboard") : ("WHITEBOARD", "inbox"),

• # aggregate all output ("AUDIO", "outbox") : ("", "outbox"), ("WHITEBOARD", "outbox") : ("", "outbox"),

• # shutdown routing ("", "control") : ("ROUTER", "control"), ("ROUTER", "signal") : ("AUDIO", "control"), ("AUDIO", "signal") : ("WHITEBOARD", "control"), ("WHITEBOARD", "signal") : ("", "signal") }, ), tokenlists_to_lines(),)

Graphlines

Backplanes•

• * For where you want 1 to many or many to many• * Declare a backplane• * Components can publish data to it• * Components may subscribe to it.• * Subscribers get a copy of all data sent•

Backplanes• Can be useful for updating a central state.•

• For example players locations can be posted here:• Backplane("PLAYERS").activate()

Backplanes• Players Post their updates like this:• Pipeline(

• MyGamesEventsComponent(up="p", down="l", left="a", right="s"),

• BasicSprite("cat.png", name = "cat", border=40),

• PureTransformer(lambda x: ("Cat ", x)),

• PublishTo("PLAYERS"),

• ).activate()

Backplanes• The system can analyse their updates like this:• Pipeline(

• SubscribeTo("PLAYERS"),

• PlayerAnalyser(),

• Distancer(),

• ConsoleEchoer(),

• ).activate()

Backplanes & Servers• When combined with a server, you instantly get a pub/

sub capable server, or splitter, or merger.• Backplane(“SPLIT”).activate()

Pipeline( DiracWebCam(), PublishTo(“SPLIT”),).activate()

def VideoProtocol(**argd): return SubscribeTo(“SPLIT”)

• ServerCore(protocol=VideoProtocol, port=1700).run()

Backplanes & Servers• When combined with a server, you instantly get a pub/

sub capable server, or splitter, or merger.• Backplane(“CHAT”).activate()

def ChatProtocol(**argd): return Pipelines( SubscribeTo(“CHAT”) NullComponent(), PublishTo(“CHAT”) )

• ServerCore(protocol=ChatProtocol, port=1700).run()

PAR & Seq• PAR & Seq are wrapper components with concepts

nabbed from Occam. (hence their names)• * Seq runs each component it's given, one after

another, wiring up inboxes/outboxes such that each component handles it. Useful in staged protocols (later)

• * PAR runs all the components concurrently. Their output is merged. Shutdown messages sent to PAR are forwarded to all subcomponents – making it useful for managing system shutdown.

Using PAR• PAR can be used to simplify some systems:• Pipeline(

PAR( Button(caption="Next", msg="NEXT", position=(72,8)), Button(caption="Previous", msg="PREV", position=(8,8)), Button(caption="First", msg="FIRST" ,position=(256,8)), Button(caption="Last", msg="LAST", position=(320,8)), ),

• Chooser(items = files), Image(size=(800,600), position=(8,48)),).run()

• This is the same slideshow as the previous Graphline...

Using PAR• ... to run & shutdown multiple subsystems• Pipeline(

timedShutdown(TTL=15), PAR( Pipeline( ReadFileAdaptor(file, readmode="bitrate", bitrate = 300000*8/5), DiracDecoder(), MessageRateLimit(framerate), VideoOverlay(position=(260,48), size=(200,300)), ), Pipeline( ReadFileAdaptor(file, readmode="bitrate", bitrate = 2280960*8), DiracDecoder(), ToRGB_interleaved(), VideoSurface(size=(200, 300), position=(600,48)), ),

Using PAR• ... to run & shutdown multiple subsystems• Pipeline(

PAR(Button(caption="Next", msg="NEXT", position=(72,8)), Button(caption="Previous", msg="PREV", position=(8,8)), Button(caption="First", msg="FIRST" ,position=(256,8)), Button(caption="Last", msg="LAST", position=(320,8)), ), Chooser(items = files), Image(size=(200,300), position=(8,48), maxpect=(200,300)), ), Pipeline( Textbox(size=(200,300), position=(8,360)), TextDisplayer(size=(200,300), position=(228,360)), ), Ticker(size=(200,300), position=(450,360)), ),).run()

Using Seq• An example from the mobile reframer:• Seq( "Decoding & separating frames...",

Graphline( MAXF = DetermineMaxFrameNumber(edlfile), DO = Carousel( ... ), STOP = TriggeredOneShot(""),... ), "Processing edits...", Graphline( REFRAMING = ReframeVideo(edlfile... SOUND = PassThroughAudio(edlfile... ENCODING = ReEncode(outFileName...... ), "Cleaning up...", StopSelector(), ).run()

Using Seq• An example of passing state in a protocol handler:• def CompositeBulletinBoardProtocol(**argd):

ConnectionInfo = {} ConnectionInfo.update(argd) return Seq( Authenticator(State = ConnectionInfo), UserRetriever(State = ConnectionInfo), MessageBoardUI(State = ConnectionInfo), StateSaverLogout(State = ConnectionInfo), )

ServerCore(protocol=CompositeBulletinBoardProtocol, ...)

• Worth noting the similarity between this and wsgi.(Except Seq can contain graphlines, pipelines, etc too)

Part 2•

Larger Systems & Debugging•

Large systems which are long running can develop bugs which are awkward to debug.

• Kamaelia has a collection of tools you can use which we'll walk through / demonstrate next.

• Practical example.

Key Tools:• * Instrumentation using backplanes & logging• * Axon Visualiser• * Embeddable python console•

• Practical examples.

Part 2•

Building a Bulletin Board•

We're going to build an “old school” bulletin board system.

• * Someone telnets in & authenticates• * Can get help, exit or read messages• * Reading messages, they can read, reply,

exit reading, or get help• * State is restored/saved at session start/end

Building a Bulletin Board•

For practicality, we won't implement message posting, nor message deletion, nor persistent user state. (Though these would be useful exercises)

• All on-disk objects encoded as json objects: * users * Messages – stored in “folders” (directories) with filenames == messageid

• * message fields: from, to, __body__, date, message, reply-to, subject

Building a Bulletin Board• Getting started:• * Start with basic server for request/response

Just echo initially - BB1.py

• * make restarting quicker for debugging - BB2.py* abstract out “getting a line of data” - BB3.py* Then use that abstraction - BB4.py* Simplify “control” box handling - BB5.py* Abstract out reusable bits from domain specific – BB6.py

• We'll go through the code for these in the actual code files rather than on these slides

• (rather unwieldy as slides)

Building a Bulletin Board• Intermediate Plan• This is the logic of the user protocol

• Seq(

• Authenticator( <some shared state> ),

• UserRetriever( <some shared state> ),

• MessageBoardUI( <some shared state> ),

• StateSaverLogout( <some shared state> ),

Building a Bulletin Board• Intermediate Plan• * Tidy up control handling a touch more & add netPrint method

to reusable bit - BB7.py

• * Change Authenticator component to support the “passed on state”, change protocol handler factory to create that Seq pipeline. - BB8.py

• * Change Authenticator to authenticate against a password file and set “remoteuser” in the shared state - BB9.py

• * Write stubs for UserRetriever & StateSaverLogout - BB10.py

Building a Bulletin Board• Final UI Plan• * Change usage of waitMsg to yield self.waitMsg() -- BB11.py

• * Copy Authenticator's patten to create initial menu for MessageBoardUI, including stub for messages menu - BB12.py

• * Use same pattern for messages menu, use waitMsg pattern for logic. - BB13.py

• * Implement Folders to hold messages. Numbers as filenames (msg 1, 2, 3, 4) Messages as json encoded objects - BB14.py

• * Clean up to add a line oriented buffer to handle partial lines. (necessary for cross platform/real world) - BB15.py

Building a Bulletin Board• Final UI Plan• * Change usage of waitMsg to yield self.waitMsg() -- BB11.py

• * Copy Authenticator's patten to create initial menu for MessageBoardUI, including stub for messages menu - BB12.py

• * Use same pattern for messages menu, use waitMsg pattern for logic. - BB13.py

• * Implement Folders to hold messages. Numbers as filenames (msg 1, 2, 3, 4) Messages as json encoded objects - BB14.py

• * Clean up to add a line oriented buffer to handle partial lines. (necessary for cross platform/real world) - BB15.py

Summary• We've covered• * the 30,000 view of kamaelia

• * Building your own core

• * Building components, including a video recording application

• * Tools for building systems

• * Tools for debugging systems

• * Built a large(ish) system (an authenticated staged protocol)

• * Illustrated the majority of this using real world systems

Final Questions?•

Thank you!¬

kamaelia europython tutorial

Technology

download - europython

building systems with kamaelia

europython 2014 devops risk mitigation

a presentation of the openstack project - europython

gamers do rest - europython 2014

pythonic mathematics kirby urner europython 2005 göteborg,...

the django web framework (europython 2006)

diving into flask - europython · well-suited for web apis...

europython 2012 (05.07.2012, florence, italy) andreas...

europython 2016 bquery lightning talk

europython 2008: dataportability and python

the butler and the snake (europython) · the butler and the...

making your first openstack contribution (europython)

europython 2015 testmon

reportlab paragraphs reloaded-europython 2008

europython - machine learning for dummies with python

timeshift everything, miss nothing - mashup your pvr with...

practical concurrent systems made simple using kamaelia

beginner’s guide to machine learning...

the butler and the snake (europython 2015)