Converting existing module/objects to threads

[jdlists]

I have inheirted some existing code, that i will explain in a moment,
have needed to extend and ultimately should be able to run in threads.
I've done a bunch of work with python but very little with threads and
am looking for some pointers on how to implement, and if the lower
level modules/objects need to be rewritten to use threading.local for
all local variables.

I have a module that communicates with a hardware device, which reads
data off of sensors, that can only talk with one controller at a time.
The controller (my module) needs to (in its simplest form) init,
configure the device, request data, and write out xml, sleep, repeat.

The new request is that the device needs to be queried until a
condition is true, and then start requesting data. So an instance of a
controller needs to be deadicated to a hardware device forever, or
until the program ends....which ever comes first.

This currently works in a non-threaded version, but only for one device
at a time, there is a need to create a single windows(yeach) service
that talks to many of these devices at once. I don't need worker
threads that handle seperate portions of the entire job, i need a
single application to spawn multiple processes to run through the
entire communication from configure to report, sleep until the next
interval time and run again. The communication could last from 1
minute to 10 minutes before it ends.


Here is the code layout in pseudocode.

module.Object - controller.Main - handles all socket communications

class subcontroller(controller.Main):
def __init__(self,id,configurationFile):
controller.Main.__init__(self)
// instantiate variables and local objects that handle
configuration, logic and data output

def configure(self,configurationFile):
//read configurationFile and configure device

def process(self):
while 1:
//based on configuration file, query the device until condition
is true and then write xml, sleep until time to repeat and run again.

within controller there are 5 objects and subcontroller is a sinlge
object that loads other objects from the inherited controller.System

I'm trying to figure out how difficult it is going to be to convert
this to a threaded application. The original controller.Main is built
to talk to devices in series, never in parallel. so no objects are
considered to be thread safe, but no instance of any of the objects
should need to share resources with any other instance of teh same
object. they would all have unique configuration files and talk to
devices on unique ip/ports.

on a unix system, forking,while potentially not optimal, would be a
fine solution, unfortunantely this needs to run on windows.

I know i have left out many details, but hopefully this is enough to at
least enable some kind soles to lend an opinnion or two.

many thanks
jd

 

[Gabriel Genellina]

This currently works in a non-threaded version, but only for one device
at a time, there is a need to create a single windows(yeach) service
that talks to many of these devices at once. I don't need worker
threads that handle seperate portions of the entire job, i need a
single application to spawn multiple processes to run through the
entire communication from configure to report, sleep until the next
interval time and run again. The communication could last from 1
minute to 10 minutes before it ends.

Consider using the asyncore module instead of threads.


--
Gabriel Genellina
Softlab SRL





__________________________________________________
Preguntá. Respondé. Descubrí.
Todo lo que querías saber, y lo que ni imaginabas,
está en Yahoo! Respuestas (Beta).
¡Probalo ya!
http://www.yahoo.com.ar/respuestas

 

[Ben Finney]

So an instance of a controller needs to be deadicated to a hardware
device forever, or until the program ends....which ever comes first.

I hope you've got tests in place for both of those conditions

 

[martdi]

I have inheirted some existing code, that i will explain in a moment,
have needed to extend and ultimately should be able to run in threads.
I've done a bunch of work with python but very little with threads and
am looking for some pointers on how to implement, and if the lower
level modules/objects need to be rewritten to use threading.local for
all local variables.

I have a module that communicates with a hardware device, which reads
data off of sensors, that can only talk with one controller at a time.
The controller (my module) needs to (in its simplest form) init,
configure the device, request data, and write out xml, sleep, repeat.

The new request is that the device needs to be queried until a
condition is true, and then start requesting data. So an instance of a
controller needs to be deadicated to a hardware device forever, or
until the program ends....which ever comes first.

This currently works in a non-threaded version, but only for one device
at a time, there is a need to create a single windows(yeach) service
that talks to many of these devices at once. I don't need worker
threads that handle seperate portions of the entire job, i need a
single application to spawn multiple processes to run through the
entire communication from configure to report, sleep until the next
interval time and run again. The communication could last from 1
minute to 10 minutes before it ends.


Here is the code layout in pseudocode.

module.Object - controller.Main - handles all socket communications

class subcontroller(controller.Main):
def __init__(self,id,configurationFile):
controller.Main.__init__(self)
// instantiate variables and local objects that handle
configuration, logic and data output

def configure(self,configurationFile):
//read configurationFile and configure device

def process(self):
while 1:
//based on configuration file, query the device until condition
is true and then write xml, sleep until time to repeat and run again.

within controller there are 5 objects and subcontroller is a sinlge
object that loads other objects from the inherited controller.System

I'm trying to figure out how difficult it is going to be to convert
this to a threaded application. The original controller.Main is built
to talk to devices in series, never in parallel. so no objects are
considered to be thread safe, but no instance of any of the objects
should need to share resources with any other instance of teh same
object. they would all have unique configuration files and talk to
devices on unique ip/ports.

on a unix system, forking,while potentially not optimal, would be a
fine solution, unfortunantely this needs to run on windows.

I know i have left out many details, but hopefully this is enough to at
least enable some kind soles to lend an opinnion or two.

many thanks
jd

Taking a look at asyncore could be worthwhile, but if you want to
implement it with threads, you may be able to do it this way:

In your main file, from where you start the program, let's call it
main:

main(self)
Load Required configuration
spawn threads (1 for each controller)
define a queue object from module queue.queue used for communication
with threads
enter an infinite loop that checks for the conditions
once conditions are met, notify the proper thread


class ControllerThread(threading.Thread):
def __init__(self):
define a queue here, to process messages from the main
call threading.Thread.__init__(self)

define method to load config for thread objects
(you might want to pass an argument to init to load your configs
from a file)
(you might also want to pass the queue of the main program to the
thread to send it messages)
define methods to post messages to the queue like read, send to the
machine, stop, ...

define the run method that is what will be called when you start
your thread.
this method should enter an infinite loop that will check if
something has to be done (check in the queue).





hope this might help you
good luck

 

[jdlists]

Consider using the asyncore module instead of threads.

I think that is a good point and I am considering using
asyncore/asynchat... i'm a little confused as to how i can make this
model work. There is no server communication without connection from
the client (me), which happens on intervals, not when data is available
on a socket or when the socket is available to be written, which is
always. Basically i need to determine how to trigger the asynchat
process based on time. in another application that i write, i'm the
server and the chat process happens every time the client wakes
up...easy and perfect for asyncore

That is a solution i'd like to persue, but am having a hard time
getting my head around that as well.

 

[jdlists]

Taking a look at asyncore could be worthwhile, but if you want to
implement it with threads, you may be able to do it this way:

In your main file, from where you start the program, let's call it
main:

main(self)
Load Required configuration
spawn threads (1 for each controller)
define a queue object from module queue.queue used for communication
with threads
enter an infinite loop that checks for the conditions
once conditions are met, notify the proper thread


class ControllerThread(threading.Thread):
def __init__(self):
define a queue here, to process messages from the main
call threading.Thread.__init__(self)

define method to load config for thread objects
(you might want to pass an argument to init to load your configs
from a file)
(you might also want to pass the queue of the main program to the
thread to send it messages)
define methods to post messages to the queue like read, send to the
machine, stop, ...

define the run method that is what will be called when you start
your thread.
this method should enter an infinite loop that will check if
something has to be done (check in the queue).





hope this might help you
good luck

thanks for the comments. the ControllerThread already extends a class.
does this cause problems with classes that must extent
threading.Thread. Normally it should not matter, but with threads i'm
unsure. Should i just instantiate the object that i'm normally
extending in the ControllerThread.__init__, and call it from the
self.classthatusedtoextend.method(), or does it not matter.

again, thanks.

 

[martdi]

I am not sure if I understand you question well, but:

in the __init__ of the thread subclass, you can instantiate an object
of the class that makes the work

or

ControllerThread could extend both classes and i don't think there
would be a problem.



Problems in multithreading usually happen when many threads try to
access the same ressource at the same time or that one thread waits for
an other thread to finish, and that other thread waits for the first
one to finish. The Queue module is threadsafe and it uses locking to
prevent multiple threads to access it at the same time. Using Queue is
a lot easier than having to manage locks by yourself.

As long as you do not share data between your classes like static
members, database connections, or I/0 on the same ressource, you
probably won't encounter problems.

 

[John Henry]

Making your code run in thread mode isn't the hard part. Just add
this:

import threading

class subcontrollerThread(threading.Thread, subcontroller):
def __init__(self,id,configurationFile):
threading.Thread.__init__(self)
subcontroller.__init__(self,id,configurationFile)
def run(self):
self.process()

threads=[]
# Say we have 5 of the subprocesses
for iThread in range(5):
th=subcontrollerThread(iThread,configurationFile)
threads.append(th)
th.start()

....main thread do whatever...


However, you have to make sure the code inside subcontroller is thread
safe. That's a topic in itself.

 

[Gabriel Genellina]

I think that is a good point and I am considering using
asyncore/asynchat... i'm a little confused as to how i can make this
model work. There is no server communication without connection from
the client (me), which happens on intervals, not when data is available
on a socket or when the socket is available to be written, which is
always. Basically i need to determine how to trigger the asynchat
process based on time. in another application that i write, i'm the
server and the chat process happens every time the client wakes
up...easy and perfect for asyncore

That is a solution i'd like to persue, but am having a hard time
getting my head around that as well.

You have to write your own dispatcher (inheriting from async_chat) as
any other protocol. You can call asyncore.loop whith count=1 (or 10,
but not None, so it returns after a few iterations) inside your *own*
loop. Inside your loop, when time comes, call
your_dispatcher.push(data) so the channel gets data to be sent.
Override collect_incoming_data() to get the response.
You can keep your pending requests in a priority queue (sorted by
time) and check the current time against the top element's time.

Maybe you could successfully implement your application using threads
- if none uses global variables, and no thread waits for another, it
may be safe. But I prefer to avoid threads whenever possible, at
least because debugging the application becomes harder.


--
Gabriel Genellina
Softlab SRL





__________________________________________________
Preguntá. Respondé. Descubrí.
Todo lo que querías saber, y lo que ni imaginabas,
está en Yahoo! Respuestas (Beta).
¡Probalo ya!
http://www.yahoo.com.ar/respuestas

 

[Hendrik van Rooyen]

I have inheirted some existing code, that i will explain in a moment,
have needed to extend and ultimately should be able to run in threads.
I've done a bunch of work with python but very little with threads and
am looking for some pointers on how to implement, and if the lower
level modules/objects need to be rewritten to use threading.local for
all local variables.

I have a module that communicates with a hardware device, which reads
data off of sensors, that can only talk with one controller at a time.
The controller (my module) needs to (in its simplest form) init,
configure the device, request data, and write out xml, sleep, repeat.

The new request is that the device needs to be queried until a
condition is true, and then start requesting data. So an instance of a
controller needs to be deadicated to a hardware device forever, or
until the program ends....which ever comes first.

This currently works in a non-threaded version, but only for one device
at a time, there is a need to create a single windows(yeach) service
that talks to many of these devices at once. I don't need worker
threads that handle seperate portions of the entire job, i need a
single application to spawn multiple processes to run through the
entire communication from configure to report, sleep until the next
interval time and run again. The communication could last from 1
minute to 10 minutes before it ends.

8<------------------------------------------------------------------------------

not sure if I understand this correctly - but I would in this position spawn new
threads, and use a global list of queues to interface between the new threads
and the old comms module, still talking to one device at a time, but now time
sliced. - in the comms module:

for q in list_of_queues:
see if anything to ask:
continue if not
ask it and put answer on reply queue

but then I am a Philistine coder, interested only in getting the job done...

- Hendrik