Source code for evennia.server.profiling.dummyrunner

"""
Dummy client runner

This module implements a stand-alone launcher for stress-testing
an Evennia game. It will launch any number of fake clients. These
clients will log into the server and start doing random operations.
Customizing and weighing these operations differently depends on
which type of game is tested. The module contains a testing module
for plain Evennia.

Please note that you shouldn't run this on a production server!
Launch the program without any arguments or options to see a
full step-by-step setup help.

Basically (for testing default Evennia):

 - Use an empty/testing database.
 - set PERMISSION_ACCOUNT_DEFAULT = "Builder"
 - start server, eventually with profiling active
 - launch this client runner

If you want to customize the runner's client actions
(because you changed the cmdset or needs to better
match your use cases or add more actions), you can
change which actions by adding a path to

   DUMMYRUNNER_ACTIONS_MODULE = <path.to.your.module>

in your settings. See utils.dummyrunner_actions.py
for instructions on how to define this module.

"""


import random
import sys
import time
from argparse import ArgumentParser

import django
from twisted.conch import telnet
from twisted.internet import protocol, reactor
from twisted.internet.task import LoopingCall

django.setup()
import evennia  # noqa

evennia._init()

from django.conf import settings  # noqa

from evennia.commands.cmdset import CmdSet  # noqa
from evennia.commands.command import Command  # noqa
from evennia.utils import mod_import, time_format  # noqa
from evennia.utils.ansi import strip_ansi  # noqa

# Load the dummyrunner settings module

DUMMYRUNNER_SETTINGS = mod_import(settings.DUMMYRUNNER_SETTINGS_MODULE)
if not DUMMYRUNNER_SETTINGS:
    raise IOError(
        "Error: Dummyrunner could not find settings file at %s"
        % settings.DUMMYRUNNER_SETTINGS_MODULE
    )
IDMAPPER_CACHE_MAXSIZE = settings.IDMAPPER_CACHE_MAXSIZE

DATESTRING = "%Y%m%d%H%M%S"
CLIENTS = []

# Settings

# number of clients to launch if no input is given on command line
NCLIENTS = 1
# time between each 'tick', in seconds, if not set on command
# line. All launched clients will be called upon to possibly do an
# action with this frequency.
TIMESTEP = DUMMYRUNNER_SETTINGS.TIMESTEP
# chance of a client performing an action, per timestep. This helps to
# spread out usage randomly, like it would be in reality.
CHANCE_OF_ACTION = DUMMYRUNNER_SETTINGS.CHANCE_OF_ACTION
# spread out the login action separately, having many accounts create accounts
# and connect simultaneously is generally unlikely.
CHANCE_OF_LOGIN = DUMMYRUNNER_SETTINGS.CHANCE_OF_LOGIN
# Port to use, if not specified on command line
TELNET_PORT = DUMMYRUNNER_SETTINGS.TELNET_PORT or settings.TELNET_PORTS[0]
#
NCONNECTED = 0  # client has received a connection
NLOGIN_SCREEN = 0  # client has seen the login screen (server responded)
NLOGGING_IN = 0  # client starting login procedure
NLOGGED_IN = 0  # client has authenticated and logged in

# time when all clients have logged_in
TIME_ALL_LOGIN = 0
# actions since all logged in
TOTAL_ACTIONS = 0
TOTAL_LAG_MEASURES = 0
# lag per 30s for all logged in
TOTAL_LAG = 0
TOTAL_LAG_IN = 0
TOTAL_LAG_OUT = 0


INFO_STARTING = """
    Dummyrunner starting using {nclients} dummy account(s). If you don't see
    any connection messages, make sure that the Evennia server is
    running.

    TELNET_PORT = {port}
    IDMAPPER_CACHE_MAXSIZE = {idmapper_cache_size} MB
    TIMESTEP = {timestep} (rate {rate}/s)
    CHANCE_OF_LOGIN = {chance_of_login}% per time step
    CHANCE_OF_ACTION = {chance_of_action}% per time step
    -> avg rate (per client, after login): {avg_rate} cmds/s
    -> total avg rate (after login): {avg_rate_total} cmds/s

    Use Ctrl-C (or Cmd-C) to stop/disconnect all clients.

    """

ERROR_NO_MIXIN = """
    Error: Evennia is not set up for dummyrunner. Before starting the
    server, make sure to include the following at *the end* of your
    settings file (remove when not using dummyrunner!):

        from evennia.server.profiling.settings_mixin import *

    This will change the settings in the following way:
        - change PERMISSION_ACCOUNT_DEFAULT to 'Developer' to allow clients
          to test all commands
        - change PASSWORD_HASHERS to use a faster (but less safe) algorithm
          when creating large numbers of accounts at the same time
        - set LOGIN_THROTTLE/CREATION_THROTTLE=None to disable it

    If you don't want to use the custom settings of the mixin for some
    reason, you can change their values manually after the import, or
    add DUMMYRUNNER_MIXIN=True to your settings file to avoid this
    error completely.

    Warning: Don't run dummyrunner on a production database! It will
    create a lot of spammy objects and accounts!
    """


ERROR_FEW_ACTIONS = """
    Dummyrunner settings error: The ACTIONS tuple is too short: it must
    contain at least login- and logout functions.
    """


HELPTEXT = """
DO NOT RUN THIS ON A PRODUCTION SERVER! USE A CLEAN/TESTING DATABASE!

This stand-alone program launches dummy telnet clients against a
running Evennia server. The idea is to mimic real accounts logging in
and repeatedly doing resource-heavy commands so as to stress test the
game. It uses the default command set to log in and issue commands, so
if that was customized, some of the functionality will not be tested
(it will not fail, the commands will just not be recognized).  The
running clients will create new objects and rooms all over the place
as part of their running, so using a clean/testing database is
strongly recommended.

Setup:
  1) setup a fresh/clean database (if using sqlite, just safe-copy
     away your real evennia.db3 file and create a new one with
     `evennia migrate`)
  2) in server/conf/settings.py, add

        PERMISSION_ACCOUNT_DEFAULT="Builder"

     This is so that the dummy accounts can test building operations.
     You can also customize the dummyrunner by modifying a setting
     file specified by DUMMYRUNNER_SETTINGS_MODULE

  3) Start Evennia like normal, optionally with profiling (--profile)
  4) Run this dummy runner via the evennia launcher:

        evennia --dummyrunner <nr_of_clients>

  5) Log on and determine if game remains responsive despite the
     heavier load. Note that if you activated profiling, there is a
     considerate additional overhead from the profiler too so you
     should usually not consider game responsivity when using the
     profiler at the same time.
  6) If you use profiling, let the game run long enough to gather
     data, then stop the server cleanly using evennia stop or @shutdown.
     @shutdown. The profile appears as
     server/logs/server.prof/portal.prof (see Python's manual on
     cProfiler).

Notes:

The dummyrunner tends to create a lot of accounts all at once, which is
a very heavy operation. This is not a realistic use-case - what you want
to test is performance during run. A large
number of clients here may lock up the client until all have been
created. It may be better to connect multiple dummyrunners instead of
starting one single one with a lot of accounts. Exactly what this number
is depends on your computer power. So start with 10-20 clients and increase
until you see the initial login slows things too much.

"""


[docs]class CmdDummyRunnerEchoResponse(Command): """ Dummyrunner command measuring the round-about response time from sending to receiving a result. Usage: dummyrunner_echo_response <timestamp> Responds with dummyrunner_echo_response:<timestamp>,<current_time> The dummyrunner will send this and then compare the send time with the receive time on both ends. """ key = "dummyrunner_echo_response"
[docs] def func(self): # returns (dummy_client_timestamp,current_time) self.msg(f"dummyrunner_echo_response:{self.args},{time.time()}") if self.caller.account.is_superuser: print(f"cmddummyrunner lag in: {time.time() - float(self.args)}s")
[docs]class DummyRunnerCmdSet(CmdSet): """ Dummyrunner injected cmdset. """
[docs] def at_cmdset_creation(self): self.add(CmdDummyRunnerEchoResponse())
# ------------------------------------------------------------ # Helper functions # ------------------------------------------------------------ ICOUNT = 0
[docs]def idcounter(): """ Makes unique ids. Returns: str: A globally unique id. """ global ICOUNT ICOUNT += 1 return str("{:03d}".format(ICOUNT))
GCOUNT = 0
[docs]def gidcounter(): """ Makes globally unique ids. Returns: count (int); A globally unique counter. """ global GCOUNT GCOUNT += 1 return "%s_%s" % (time.strftime(DATESTRING), GCOUNT)
[docs]def makeiter(obj): """ Makes everything iterable. Args: obj (any): Object to turn iterable. Returns: iterable (iterable): An iterable object. """ return obj if hasattr(obj, "__iter__") else [obj]
# ------------------------------------------------------------ # Client classes # ------------------------------------------------------------
[docs]class DummyClient(telnet.StatefulTelnetProtocol): """ Handles connection to a running Evennia server, mimicking a real account by sending commands on a timer. """
[docs] def report(self, text, clientkey): pad = " " * (25 - len(text)) tim = round(time.time() - self.connection_timestamp) print( f"{text} {clientkey}{pad}\t" f"conn: {NCONNECTED} -> " f"welcome screen: {NLOGIN_SCREEN} -> " f"authing: {NLOGGING_IN} -> " f"loggedin/tot: {NLOGGED_IN}/{NCLIENTS} (after {tim}s)" )
[docs] def connectionMade(self): """ Called when connection is first established. """ global NCONNECTED # public properties self.cid = idcounter() self.key = f"Dummy-{self.cid}" self.gid = f"{time.strftime(DATESTRING)}_{self.cid}" self.istep = 0 self.exits = [] # exit names created self.objs = [] # obj names created self.connection_timestamp = time.time() self.connection_attempt = 0 self.action_started = 0 self._connected = False self._loggedin = False self._logging_out = False self._ready = False self._report = "" self._cmdlist = [] # already stepping in a cmd definition self._login = self.factory.actions[0] self._logout = self.factory.actions[1] self._actions = self.factory.actions[2:] reactor.addSystemEventTrigger("before", "shutdown", self.logout) NCONNECTED += 1 self.report("-> connected", self.key) reactor.callLater(30, self._retry_welcome_screen)
def _retry_welcome_screen(self): if not self._connected and not self._ready: # we have connected but not received anything for 30s. # (unclear why this would be - overload?) # try sending a look to get something to start with self.report("?? retrying welcome screen", self.key) self.sendLine(bytes("look", "utf-8")) # make sure to check again later reactor.callLater(30, self._retry_welcome_screen) def _print_statistics(self): global TIME_ALL_LOGIN, TOTAL_ACTIONS global TOTAL_LAG, TOTAL_LAG_MEASURES, TOTAL_LAG_IN, TOTAL_LAG_OUT tim = time.time() - TIME_ALL_LOGIN avgrate = round(TOTAL_ACTIONS / tim) lag = TOTAL_LAG / (TOTAL_LAG_MEASURES or 1) lag_in = TOTAL_LAG_IN / (TOTAL_LAG_MEASURES or 1) lag_out = TOTAL_LAG_OUT / (TOTAL_LAG_MEASURES or 1) TOTAL_ACTIONS = 0 TOTAL_LAG = 0 TOTAL_LAG_IN = 0 TOTAL_LAG_OUT = 0 TOTAL_LAG_MEASURES = 0 TIME_ALL_LOGIN = time.time() print( f".. running 30s average: ~{avgrate} actions/s " f"lag: {lag:.2}s (in: {lag_in:.2}s, out: {lag_out:.2}s)" ) reactor.callLater(30, self._print_statistics)
[docs] def dataReceived(self, data): """ Called when data comes in over the protocol. We wait to start stepping until the server actually responds Args: data (str): Incoming data. """ global NLOGIN_SCREEN, NLOGGED_IN, NLOGGING_IN, NCONNECTED global TOTAL_ACTIONS, TIME_ALL_LOGIN global TOTAL_LAG, TOTAL_LAG_MEASURES, TOTAL_LAG_IN, TOTAL_LAG_OUT if not data.startswith(b"\xff"): # regular text, not a telnet command if NCLIENTS == 1: print("dummy-client sees:", str(data, "utf-8")) if not self._connected: # waiting for connection # wait until we actually get text back (not just telnet # negotiation) # start client tick d = LoopingCall(self.step) df = max(abs(TIMESTEP * 0.001), min(TIMESTEP / 10, 0.5)) # dither next attempt with random time timestep = TIMESTEP + (-df + (random.random() * df)) d.start(timestep, now=True).addErrback(self.error) self.connection_attempt += 1 self._connected = True NLOGIN_SCREEN += 1 NCONNECTED -= 1 self.report("<- server sent login screen", self.key) elif self._loggedin: if not self._ready: # logged in, ready to run NLOGGED_IN += 1 NLOGGING_IN -= 1 self._ready = True self.report("== logged in", self.key) if NLOGGED_IN == NCLIENTS and not TIME_ALL_LOGIN: # all are logged in! We can start collecting statistics print(".. All clients connected and logged in!") TIME_ALL_LOGIN = time.time() reactor.callLater(30, self._print_statistics) elif TIME_ALL_LOGIN: TOTAL_ACTIONS += 1 try: data = strip_ansi(str(data, "utf-8").strip()) if data.startswith("dummyrunner_echo_response:"): # handle special lag-measuring command. This returns # dummyrunner_echo_response:<starttime>,<midpointtime> now = time.time() _, data = data.split(":", 1) start_time, mid_time = (float(part) for part in data.split(",", 1)) lag_in = mid_time - start_time lag_out = now - mid_time total_lag = now - start_time # full round-about time TOTAL_LAG += total_lag TOTAL_LAG_IN += lag_in TOTAL_LAG_OUT += lag_out TOTAL_LAG_MEASURES += 1 except Exception: pass
[docs] def connectionLost(self, reason): """ Called when loosing the connection. Args: reason (str): Reason for loosing connection. """ if not self._logging_out: self.report("XX lost connection", self.key)
[docs] def error(self, err): """ Error callback. Args: err (Failure): Error instance. """ print(err)
[docs] def counter(self): """ Produces a unique id, also between clients. Returns: counter (int): A unique counter. """ return gidcounter()
[docs] def logout(self): """ Causes the client to log out of the server. Triggered by ctrl-c signal. """ self._logging_out = True cmd = self._logout(self)[0] self.report(f"-> logout/disconnect ({self.istep} actions)", self.key) self.sendLine(bytes(cmd, "utf-8"))
[docs] def step(self): """ Perform a step. This is called repeatedly by the runner and causes the client to issue commands to the server. This holds all "intelligence" of the dummy client. """ global NLOGGING_IN, NLOGIN_SCREEN rand = random.random() if not self._cmdlist: # no commands ready. Load some. if not self._loggedin: if rand < CHANCE_OF_LOGIN or NLOGGING_IN < 10: # lower rate of logins, but not below 1 / s # get the login commands self._cmdlist = list(makeiter(self._login(self))) NLOGGING_IN += 1 # this is for book-keeping NLOGIN_SCREEN -= 1 self.report("-> create/login", self.key) self._loggedin = True else: # no login yet, so cmdlist not yet set return else: # we always pick a cumulatively random function crand = random.random() cfunc = [func for (cprob, func) in self._actions if cprob >= crand][0] self._cmdlist = list(makeiter(cfunc(self))) # at this point we always have a list of commands if rand < CHANCE_OF_ACTION: # send to the game cmd = str(self._cmdlist.pop(0)) if cmd.startswith("dummyrunner_echo_response"): # we need to set the timer element as close to # the send as possible cmd = cmd.format(timestamp=time.time()) self.sendLine(bytes(cmd, "utf-8")) self.action_started = time.time() self.istep += 1 if NCLIENTS == 1: print(f"dummy-client sent: {cmd}")
[docs]class DummyFactory(protocol.ReconnectingClientFactory): protocol = DummyClient initialDelay = 1 maxDelay = 1 noisy = False
[docs] def __init__(self, actions): "Setup the factory base (shared by all clients)" self.actions = actions
# ------------------------------------------------------------ # Access method: # Starts clients and connects them to a running server. # ------------------------------------------------------------
[docs]def start_all_dummy_clients(nclients): """ Initialize all clients, connect them and start to step them Args: nclients (int): Number of dummy clients to connect. """ global NCLIENTS NCLIENTS = int(nclients) actions = DUMMYRUNNER_SETTINGS.ACTIONS if len(actions) < 2: print(ERROR_FEW_ACTIONS) return # make sure the probabilities add up to 1 pratio = 1.0 / sum(tup[0] for tup in actions[2:]) flogin, flogout, probs, cfuncs = ( actions[0], actions[1], [tup[0] * pratio for tup in actions[2:]], [tup[1] for tup in actions[2:]], ) # create cumulative probabilies for the random actions cprobs = [sum(v for i, v in enumerate(probs) if i <= k) for k in range(len(probs))] # rebuild a new, optimized action structure actions = (flogin, flogout) + tuple(zip(cprobs, cfuncs)) # setting up all clients (they are automatically started) factory = DummyFactory(actions) for i in range(NCLIENTS): reactor.connectTCP("127.0.0.1", TELNET_PORT, factory) # start reactor reactor.run()
# ------------------------------------------------------------ # Command line interface # ------------------------------------------------------------ if __name__ == "__main__": try: settings.DUMMYRUNNER_MIXIN except AttributeError: print(ERROR_NO_MIXIN) sys.exit() # parsing command line with default vals parser = ArgumentParser(description=HELPTEXT) parser.add_argument( "-N", nargs=1, default=1, dest="nclients", help="Number of clients to start" ) args = parser.parse_args() nclients = int(args.nclients[0]) print( INFO_STARTING.format( nclients=nclients, port=TELNET_PORT, idmapper_cache_size=IDMAPPER_CACHE_MAXSIZE, timestep=TIMESTEP, rate=1 / TIMESTEP, chance_of_login=CHANCE_OF_LOGIN * 100, chance_of_action=CHANCE_OF_ACTION * 100, avg_rate=(1 / TIMESTEP) * CHANCE_OF_ACTION, avg_rate_total=(1 / TIMESTEP) * CHANCE_OF_ACTION * nclients, ) ) # run the dummyrunner TIME_START = t0 = time.time() start_all_dummy_clients(nclients=nclients) ttot = time.time() - t0 # output runtime print("... dummy client runner stopped after %s." % time_format(ttot, style=3))