Mailman 3 March 2009 - Twisted

[Twisted-Python] how to create state machines?
by Doug Farrell March 27, 2009

March 27, 2009

Hi all, I'm still getting my 'sea legs' with Twisted, so this may be a novice question, but I'm wondering what would be a good way to create a state machine with Twisted? For instance I've written quite a bit of code in different languages that creates co-operative, context switching schemes using state machines. Here would be a primative example in psuedocode: Switch(state): Case INIT: Initialize() state = CMD1 Case CMD1: CommandOne() state = WAITCMD1 Case WAITCMD1: If CommandOneDone() state = DONE Else state = WAITCMD1 Case DONE: state = EXIT This state machine initializes a system, issues an asynchronous command that takes time to complete (say a motor move command) and then waits for that command to be done before exiting. In the context of a framework that is calling this state machine, the WAITCMD1 is executed repeatedly (polling) while the asynchronous command completes. A system can be constructed with lots of little state machines like this and be made to look like it is doing many things at once even though the execution of the program is single threaded. I understand (pretty much) the Twisted framework is like this and implmenting event handlers like connectionMade(), etc., is a state machine, but I'm wondering how to implement what I've outlined above in one of Twisted's state event handlers, or inside something like callLater(). For example, let's say I want to use Twisted to run a long running daemon process that has an XMLRPC interface. That XMLRPC interface is an interface to a state machine inside Twisted that allows the caller to change state, or start new state machines, or get the status of a running state machine. In that case I'm thinking I want a state machine the runs continuously in the Twisted loop, structured like the example above; co-operatively yielding back to Twisted, but running non-stop. Something like callLater(1, stateMachine), but non-stop, without even the 1 second call loop. Is this something I can do with callLater()? Would using a deferred be a better option? I see deferred's callBack() system as a connecting states together, but I'm not sure how a WAITONCMD1 kind of thing would work, where the deferred callBack(d, waitTillDone) would have to keep coming back to itself till some condition is met? Thanks in advance for your help, Doug

2 1

[Twisted-Python] Re: Twisted-Python] how to create state machines?
by Andrew Francis March 27, 2009

March 27, 2009

Hi Doug: >Message: 1 >Date: Thu, 26 Mar 2009 14:26:17 -0400 >From: "Doug Farrell" <dfarrell(a)mypublisher.com> >Subject: [Twisted-Python] how to create state machines? >To: <twisted-python(a)twistedmatrix.com> >Message-ID: <318F79422ADC5041A93343721F84474D01A8BB43(a)exchange.mypublisher.local> >Content-Type: text/plain; charset="us-ascii" >This state machine initializes a system, issues an asynchronous command >that takes time to complete (say a motor move command) and then waits >for that command to be done before exiting. In the context of a >framework that is calling this state machine, the WAITCMD1 is executed >repeatedly (polling) while the asynchronous command completes. A system >can be constructed with lots of little state machines like this and be >made to look like it is doing many things at once even though the >execution of the program is single threaded. >I understand (pretty much) the Twisted framework is like this and >implmenting event handlers like connectionMade(), etc., is a state >machine, but I'm wondering how to implement what I've outlined above in >one of Twisted's state event handlers, or inside something like >callLater(). For example, let's say I want to use Twisted to run a long >running daemon process that has an XMLRPC interface. That XMLRPC >interface is an interface to a state machine inside Twisted that allows >the caller to change state, or start new state machines, or get the >status of a running state machine. In that case I'm thinking I want a >state machine the runs continuously in the Twisted loop, structured like >the example above; co-operatively yielding back to Twisted, but running >non-stop. Something like callLater(1, stateMachine), but non-stop, >without even the 1 second call loop. If I understand your correctly, I don't think you need to implement a state machine to simulate concurrency with Twisted - Twisted does a lot of that for you. You can think of a Twisted application as a state machine - the callback being the state and the completion of the operation and the calling of the callback is the transition. These callbacks at runtime act like a thread of execution. def Initialize(...): # do something deferred = someFunctionThatReturnsADeferred() deferred.addCallback(State2) def State2(...): # do something deferred = someFunctionThatReturnsADeferred() deferred.addCallback(State3) def State3(someData): # do something if someData == 'State4': deferred = someFunctionThatReturnsADeferred() func = State4 elif someData == 'State5': deferred = someOtherFunctionThatReturnsADeferred() func = State5 ... deferred.addCallback(func) if __name__ = "__main__": initialize() reactor.run() As for the Twisted loop. Well, you don't really see the Twisted loop since that is hidden in the reactor. Also you should distinguish between writing a new protocol and using an existing one. In the case of XMLRPC, creating the server isn't the problem. http://twistedmatrix.com/projects/web/documentation/howto/xmlrpc.html Once a XMLRPC server is created, Twisted will take responsibility for creating new instances (or threads if you want to see it that way). If you still need a state machine, then the only hiccup I can see is sharing state machine (if you really need one) between XMLRPC method invocations. Cheers, Andrew

1 0

[Twisted-Python] pb and deferToThread for geographic data services - Newbie woes
by Jeff Cavner March 26, 2009

March 26, 2009

I am trying to use Twisted to provide a high volume (with lots of simultaneous requests) set of services that process and deliver several different types of geographic data. I am brand new to Twisted. But a user would come into the system through a Turbogears controller, where I intend to write a Twisted perspective broker client to a pb server. The server will have different functions that ideally would operate in an asynchronous fashion. That is, the client would make a request for some processing on one portion of the problem, while returning control and then another remote function would be called to work on another part of the problem. My thinking was that I would have to use pb with deferToThread. But I am having a hard time finding any good examples that use both pb and deferToThread. In other words for a simple pb example like the following I am unsure of where to add the deferToThread bits. from twisted.spread import pb from twisted.internet import reactor from twisted.python import util factory = pb.PBClientFactory() reactor.connectTCP("localhost", 8789, factory) d = factory.getRootObject() d.addCallback(lambda object: object.callRemote("echo", "hello network")) d.addCallback(lambda echo: 'server echoed: '+echo) d.addErrback(lambda reason: 'error: '+str(reason.value)) d.addCallback(util.println) d.addCallback(lambda _: reactor.stop()) reactor.run() My attempt to modify the code above to take advantage of two remote function calls (one long, and one short) to emulate longer processing times by a function that should run while control is returned and the another function called is woefully synchronous. Here is that code: from twisted.spread import pb from twisted.internet import reactor from twisted.python import util from twisted.internet import threads def getWMS(): d.addCallback(lambda object: object.callRemote("long", "hello long")) d.addCallback(lambda long: 'server echoed: '+long) d.addErrback(lambda reason: 'error: '+str(reason.value)) def gotWMS(someobject): d.addCallback(util.println) def getShort(): e.addCallback(lambda object: object.callRemote("short", "hello short")) e.addCallback(lambda echo: 'server echoed: '+echo) e.addErrback(lambda reason: 'error: '+str(reason.value)) def gotShort(someobject): e.addCallback(util.println) factory = pb.PBClientFactory() reactor.connectTCP("localhost", 8789, factory) d = factory.getRootObject() mythread = threads.deferToThread(getWMS) mythread.addCallback(gotWMS) e = factory.getRootObject() mythread2 = threads.deferToThread(getShort) mythread2.addCallback(gotShort) reactor.run() The long call always returns first and therefore stalls the return from the short function. I was hoping, expecting to get the short call response back before the long. I am sure I am not thinking about any of this correctly. Any pointers????? Additionally does one have to get the rootObject for each function call as I have done here? Jeff Cavner jcavner(a)edac.unm.edu

1 0

[Twisted-Python] How to split a huge 'service' class.
by Peter Cai March 26, 2009

March 26, 2009

Hi all, I am writing a network application based on twisted and following the "finger service" tutorial in twisted document. I wrote multiple protocols classes and a single service class. This allows client to choose the most suitable protocol to access services. My problem is : the service class now become really huge, about 3000 thousands line of code. Then I moved these code that really do the job in service class to several classes, and the service would call these methods of these classes to finish calls from protocol classes. But still, this would cause another problem : now the service classes seems to be useless. It contains lots small methods that almost do nothing. That has a bad smell for me. So, if you are in my situation, what will you do？ -- look to the things around you,the immediate world around you, if you are alive,it will mean something to you ——Paul Strand

1 0

[Twisted-Python] strange server crash
by Alec Matusis March 25, 2009

March 25, 2009

I have been operating a busy twisted TCP server for quite some time (1 year+). Yesterday, it strangely crashed under peak load for that day (it has been under more load in the past). # grep segfault /var/log/messages Mar 20 19:12:15 serv2 kernel: [17687209.144548] twistd[10701]: segfault at 8 rip 41eccf rsp 7fff33675270 error 6 I cannot understand the reason of the crash, except that its some very rare race condition. While the code is quite cumbersome, here are some general details: 8 min before the crash, the server started giving repeated error on clients disconnect, that I have never seen before in the last year. It first produced this twice: 2009/03/20 19:04 -0700 [ClientProtocol,318930,70.132.227.150] Unhandled Error Traceback (most recent call last): File "/nail/encap/Twisted-2.5.0/lib/python2.4/site-packages/twisted/application/a pp.py", line 113, in runReactorWithLogging reactor.run() File "/nail/encap/Twisted-2.5.0/lib/python2.4/site-packages/twisted/internet/posi xbase.py", line 220, in run self.mainLoop() File "/nail/encap/Twisted-2.5.0/lib/python2.4/site-packages/twisted/internet/posi xbase.py", line 231, in mainLoop self.doIteration(t) File "/nail/encap/Twisted-2.5.0/lib/python2.4/site-packages/twisted/internet/epol lreactor.py", line 181, in doPoll log.callWithLogger(selectable, _drdw, selectable, fd, event) --- <exception caught here> --- File "/nail/encap/Twisted-2.5.0/lib/python2.4/site-packages/twisted/python/log.py ", line 48, in callWithLogger return callWithContext({"system": lp}, func, *args, **kw) File "/nail/encap/Twisted-2.5.0/lib/python2.4/site-packages/twisted/python/log.py ", line 33, in callWithContext return context.call({ILogContext: newCtx}, func, *args, **kw) File "/nail/encap/Twisted-2.5.0/lib/python2.4/site-packages/twisted/python/contex t.py", line 59, in callWithContext return self.currentContext().callWithContext(ctx, func, *args, **kw) File "/nail/encap/Twisted-2.5.0/lib/python2.4/site-packages/twisted/python/contex t.py", line 37, in callWithContext return func(*args,**kw) File "/nail/encap/Twisted-2.5.0/lib/python2.4/site-packages/twisted/internet/epol lreactor.py", line 212, in _doReadOrWrite self._disconnectSelectable(selectable, why, inRead) File "/nail/encap/Twisted-2.5.0/lib/python2.4/site-packages/twisted/internet/posi xbase.py", line 252, in _disconnectSelectable selectable.readConnectionLost(f) File "/nail/encap/Twisted-2.5.0/lib/python2.4/site-packages/twisted/internet/tcp. py", line 405, in readConnectionLost self.connectionLost(reason) File "/nail/encap/Twisted-2.5.0/lib/python2.4/site-packages/twisted/internet/tcp. py", line 416, in connectionLost protocol.connectionLost(reason) File "/nail/live/im/myserv/protocols.py", line 170, in connectionLost self.session.logout() File "/nail/live/im/myserv/service.py", line 1199, in logout if self.nick in session.sellers: exceptions.AttributeError: Deferred instance has no attribute 'sellers' and then a different error in the same line with the same call stack, that was repeated about 1000 times for different logging out clients: 2009/03/20 19:04 -0700 [ClientProtocol,326158,99.245.246.187] Unhandled Error Traceback (most recent call last): File "/nail/encap/Twisted-2.5.0/lib/python2.4/site-packages/twisted/application/a pp.py", line 113, in runReactorWithLogging reactor.run() File "/nail/encap/Twisted-2.5.0/lib/python2.4/site-packages/twisted/internet/posi xbase.py", line 220, in run self.mainLoop() File "/nail/encap/Twisted-2.5.0/lib/python2.4/site-packages/twisted/internet/posi xbase.py", line 231, in mainLoop self.doIteration(t) File "/nail/encap/Twisted-2.5.0/lib/python2.4/site-packages/twisted/internet/epol lreactor.py", line 181, in doPoll log.callWithLogger(selectable, _drdw, selectable, fd, event) --- <exception caught here> --- File "/nail/encap/Twisted-2.5.0/lib/python2.4/site-packages/twisted/python/log.py ", line 48, in callWithLogger return callWithContext({"system": lp}, func, *args, **kw) File "/nail/encap/Twisted-2.5.0/lib/python2.4/site-packages/twisted/python/log.py ", line 33, in callWithContext return context.call({ILogContext: newCtx}, func, *args, **kw) File "/nail/encap/Twisted-2.5.0/lib/python2.4/site-packages/twisted/python/contex t.py", line 59, in callWithContext return self.currentContext().callWithContext(ctx, func, *args, **kw) File "/nail/encap/Twisted-2.5.0/lib/python2.4/site-packages/twisted/python/contex t.py", line 37, in callWithContext return func(*args,**kw) File "/nail/encap/Twisted-2.5.0/lib/python2.4/site-packages/twisted/internet/epol lreactor.py", line 212, in _doReadOrWrite self._disconnectSelectable(selectable, why, inRead) File "/nail/encap/Twisted-2.5.0/lib/python2.4/site-packages/twisted/internet/posi xbase.py", line 252, in _disconnectSelectable selectable.readConnectionLost(f) File "/nail/encap/Twisted-2.5.0/lib/python2.4/site-packages/twisted/internet/tcp. py", line 405, in readConnectionLost self.connectionLost(reason) File "/nail/encap/Twisted-2.5.0/lib/python2.4/site-packages/twisted/internet/tcp. py", line 416, in connectionLost protocol.connectionLost(reason) File "/nail/live/im/myserv/protocols.py", line 170, in connectionLost self.session.logout() File "/nail/live/im/myserv/service.py", line 1199, in logout if self.nick in session.sellers: exceptions.AttributeError: WLResultSet instance has no attribute 'sellers' This last error repeated about 1000 times for different clients, after which the segfault occurred. The problem is, that session is an instance of a Session class that is defined in the code, it is NEITHER a Deferred NOR an instance of WLResultSet class- these are completely different classes, and theres no possible way in the code for variable session to become an instance of Deferred. The code is approximately like this: def logout(self): self.on_logout(self.do_logout) for session in self.service.client_sessions: ß self.service.client_sessions is a dictionary of custom Session class instances if self.nick in session.sellers: ß error here, it claims that session is a Deferred or sometimes an instance of a completely unrelated class #Some statements The function self.on_logout sometimes (rarely) dispatches a deferred (it has maybeDeferred in it), such that its argument self.on_logout is called when that deferred returns (otherwise its called synchronously). My question is, is it possible to muck up deferred in such a way, that after dispatching a deferred chain inside self.on_logout(self.do_logout) , the remaining statements would be executed in some unexpected local scope, such that I would see these weird errors where my for loop iterator would suddenly go over Deferred instances, instead of the dictionary of Session instances that I define in the code ?? Also, why did these errors lead to segfault of the whole twisted process after 8 minutes?

6 8

[Twisted-Python] So how does everyone use an ORM in twisted?
by Ryan Lepidi March 23, 2009

March 23, 2009

I have searched a bit, and it seems as though you can't use SQL Alchemy with twisted due to SA being thread unsafe. The only solution I found was sAsync, but this is old and unmaintained. What is everyone else doing for database access for a multiuser server? Is everyone really using raw sql queries or is there some easy solution that I'm missing? I'm not even stuck on SQL Alchemy; if there is some other ORM that works in this case, I'd be willing to learn it.

3 2

[Twisted-Python] Stackless Twisted shocking performance
by Simon Pickles March 23, 2009

March 23, 2009

Hi, Now I am first to admit my problem lies within my app design somewhere, but I thought I'd raise it here for advice. I have a server made from a central stackless python hub, with a boost python c++ interface, to which c++ clients connect. This hub is 'surrounded' by stackless python zones which are passed work by the hub. The zones send the updates to the hub which passes them onto the clients. The hub and zones communicate using twisted.perspectivebroker. This all seemed to work well, until I began to load test. It quickly became clear the MAJOR bottleneck was the transmission of data between the hub and zone - the twisted bit. As more updates are sent zone->hub, per second, the incoming client->hub->zone messages take SECONDS to get thru! This is with approx 50 messages per second between a single zone and the hub. Is this the sort of performance I can expect? If so, what I my alternatives? I am considering ditching the twisted part and using stacklessSockets instead. Thanks for any advice Simon

3 3

[Twisted-Python] Weekly Bug Summary
by exarkun＠twistedmatrix.com March 22, 2009

March 22, 2009

1 0

[Twisted-Python] Re: [Stackless] Stackless Twisted shocking
by Andrew Francis March 21, 2009

March 21, 2009

Hi Simon: --- On Sat, 3/21/09, stackless-request(a)stackless.com <stackless-request(a)stackless.com> wrote: > As more updates are sent zone->hub, per second, the > incoming client->hub->zone messages take SECONDS to get thru! > This is with approx 50 messages per second between a single zone and the >hub. > Is this the sort of performance I can expect? I am not an expert in performance tuning... and I have to do some performance testing soon. That said, I have problems envisioning the architecture. Does a request cause a C++ client to be spawned? Are the C++ clients logging into the server independent of requests? There are so many things missing. However the first thing I would ask is what sort of performance can I reasonably expect? What does the workflow look like? Where are the delays? Also what does stuff like CPU, memory utilization, paging look like as you increase the load? Are messages queuing up somewhere? > If so, what I my alternatives? I am considering ditching > the twisted part and using stacklessSockets instead. Simon, PB Twisted is much higher level than stacklessSockets. Cheers, Andrew

1 0

[Twisted-Python] LineReceiver MAX_LENGTH limitation
by Pet March 18, 2009

March 18, 2009

Hi, in LineReceiver there is MAX_LENGTH = 16384 limitation. Is it possible to set it bigger at start of an application? Regards, Pet

2 2