Async Process

This is considered an advanced topic.

Synchronous versus Asynchronous

Most program code operates synchronously. This means that each statement in your code gets processed and finishes before the next can begin. This makes for easy-to-understand code. It is also a requirement in many cases - a subsequent piece of code often depend on something calculated or defined in a previous statement.

Consider this piece of code:

print "before call ..."
long_running_function()
print "after call ..."

When run, this will print "before call ...", after which the long_running_function gets to work for however long time. Only once that is done, the system prints "after call ...". Easy and logical to follow. Most of Evennia work in this way and often it’s important that commands get executed in the same strict order they were coded.

Evennia, via Twisted, is a single-process multi-user server. In simple terms this means that it swiftly switches between dealing with player input so quickly that each player feels like they do things at the same time. This is a clever illusion however: If one user, say, runs a command containing that long_running_function, all other players are effectively forced to wait until it finishes.

Now, it should be said that on a modern computer system this is rarely an issue. Very few commands run so long that other users notice it. And as mentioned, most of the time you want to enforce all commands to occur in strict sequence.

When delays do become noticeable and you don’t care in which order the command actually completes, you can run it asynchronously. This makes use of the run_async() function in src/utils/utils.py:

run_async(function, *args, **kwargs)

Where function will be called asynchronously with *args and **kwargs. Example:

from evevnnia import utils
print "before call ..."
utils.run_async(long_running_function)
print "after call ..."

Now, when running this you will find that the program will not wait around for long_running_function to finish. In fact you will see "before call ..." and "after call ..." printed out right away. The long-running function will run in the background and you (and other users) can go on as normal.

Customizing asynchronous operation

A complication with using asynchronous calls is what to do with the result from that call. What if long_running_function returns a value that you need? It makes no real sense to put any lines of code after the call to try to deal with the result from long_running_function above - as we saw the "after call ..." got printed long before long_running_function was finished, making that line quite pointless for processing any data from the function. Instead one has to use callbacks.

utils.run_async takes reserved kwargs that won’t be passed into the long-running function:

  • at_return(r) (the callback) is called when the asynchronous function (long_running_function above) finishes successfully. The argument r will then be the return value of that function (or None).

    def at_return(r):
        print r
    
  • at_return_kwargs - an optional dictionary that will be fed as keyword arguments to the at_return callback.

  • at_err(e) (the errback) is called if the asynchronous function fails and raises an exception. This exception is passed to the errback wrapped in a Failure object e. If you do not supply an errback of your own, Evennia will automatically add one that silently writes errors to the evennia log. An example of an errback is found below:

def at_err(e):
    print "There was an error:", str(e)
  • at_err_kwargs - an optional dictionary that will be fed as keyword arguments to the at_err errback.

An example of making an asynchronous call from inside a Command definition:

from ev import utils
from game.gamesrc.commands.basecommand import Command

class CmdAsync(Command):

   key = "asynccommand"

   def func(self):

       def long_running_function():
           #[... lots of time-consuming code
           return final_value

       def at_return(r):
           self.caller.msg("The final value is %s" % r)

       def at_err(e):
           self.caller.msg("There was an error: %s" % e)

       # do the async call, setting all callbacks
       utils.run_async(long_running_function, at_return, at_err)

That’s it - from here on we can forget about long_running_function and go on with what else need to be done. Whenever it finishes, the at_return function will be called and the final value will pop up for us to see. If not we will see an error message.

Process Pool

Note: The Process pool is currently not available nor supported, so the following section should be ignored. An old and incompatible version of the procpool can be found in the evennia/procpool repository if you are interested.

The ProcPool is an Evennia subsystem that launches a pool of processes based on the ampoule package (included with Evennia). When active, run_async will use this pool to offload its commands. ProcPool is deactivated by default, it can be turned on with settings.PROCPOOL_ENABLED. It should be noted that the default SQLite3 database is not suitable for for multiprocess operation. So if you use ``ProcPool`` you should consider switching to another database such as MySQL or PostgreSQL.

The Process Pool makes several additional options available to run_async. The following keyword arguments make sense when ProcPool is active:

  • use_thread - this force-reverts back to thread operation (as above). It effectively deactivates all additional features ProcPool offers.
  • proc_timeout - this enforces a timeout for the running process in seconds; after this time the process will be killed.
  • at_return, at_err - these work the same as above.

In addition to feeding a single callable to run_async, the first argument may also be a source string. This is a piece of python source code that will be executed in a subprocess via ProcPool. Any extra keyword arguments to run_async that are not one of the reserved ones will be used to specify what will be available in the execution environment.

There is one special variable used in the remote execution: _return. This is a function, and all data fed to _return will be returned from the execution environment and appear as input to your at_return callback (if it is defined). You can call _return multiple times in your code - the return value will then be a list.

Example:

from src.utils.utils import run_async

source = """
from time import sleep
sleep(5) # sleep five secs
val = testvar + 5
_return(val)
_return(val + 5)
"""

# we assume myobj is a character retrieved earlier
# these callbacks will just print results/errors
def callback(ret):
    myobj.msg(ret)
def errback(err):
    myobj.msg(err)
testvar = 3

# run async
run_async(source, at_return=callback, at_err=errback, testvar=testvar)

# this will return '[8, 13]'

You can also test the async mechanism from in-game using the @py command:

@py from src.utils.utils import run_async;run_async("_return(1+2)",at_return=self.msg)

Note: The code execution runs without any security checks, so it should not be available to unprivileged users. Try contrib.evlang.evlang.limited_exec for running a more restricted version of Python for untrusted users. This will use run_async under the hood.

delay

The delay function is a much simpler sibling to run_async. It is in fact just a way to delay the execution of a command until a future time. This is equivalent to something like time.sleep() except delay is asynchronous while sleep would lock the entire server for the duration of the sleep.

def callback(obj):
   obj.msg("Returning!")
delay(10, caller, callback=callback)

This will delay the execution of the callback for 10 seconds. This function is explored much more in the Command Duration Tutorial.

Assorted notes

Note that the run_async will try to launch a separate thread behind the scenes. Some databases, notably our default database SQLite3, does not allow concurrent read/writes. So if you do a lot of database access (like saving to an Attribute) in your function, your code might actually run slower using this functionality if you are not careful. Extensive real-world testing is your friend here.

Overall, be careful with choosing when to use asynchronous calls. It is mainly useful for large administration operations that have no direct influence on the game world (imports and backup operations come to mind). Since there is no telling exactly when an asynchronous call actually ends, using them for in-game commands is to potentially invite confusion and inconsistencies (and very hard-to-reproduce bugs).

The very first synchronous example above is not really correct in the case of Twisted, which is inherently an asynchronous server. Notably you might find that you will not see the first before call ... text being printed out right away. Instead all texts could end up being delayed until after the long-running process finishes. So all commands will retain their relative order as expected, but they may appear with delays or in groups.

Further reading

Technically, run_async is just a very thin and simplified wrapper around a Twisted Deferred object; the wrapper sets up a separate thread and assigns a default errback also if none is supplied. If you know what you are doing there is nothing stopping you from bypassing the utility function, building a more sophisticated callback chain after your own liking.