mirror of
https://github.com/wassname/catalyst.git
synced 2026-07-11 15:31:20 +08:00
Merge branch 'new_world_order' of github.com:quantopian/zipline into new_world_order
This commit is contained in:
@@ -0,0 +1,106 @@
|
||||
import zmq
|
||||
import pytz
|
||||
from datetime import datetime, timedelta
|
||||
|
||||
from unittest2 import TestCase
|
||||
from collections import defaultdict
|
||||
|
||||
from zipline.finance.trading import SIMULATION_STYLE
|
||||
from zipline.core.devsimulator import AddressAllocator
|
||||
from zipline.lines import SimulatedTrading
|
||||
|
||||
from zipline.utils.test_utils import (
|
||||
drain_zipline,
|
||||
check,
|
||||
setup_logger,
|
||||
teardown_logger,
|
||||
launch_component,
|
||||
create_monitor,
|
||||
launch_monitor
|
||||
)
|
||||
|
||||
|
||||
from zipline.core import Component
|
||||
from zipline.core.component import ComponentSocketArgs
|
||||
from zipline.protocol import (
|
||||
DATASOURCE_FRAME,
|
||||
DATASOURCE_UNFRAME
|
||||
)
|
||||
|
||||
from zipline.gens.tradegens import SpecificEquityTrades
|
||||
from zipline.gens.utils import hash_args
|
||||
from zipline.gens.zmqgen import gen_from_poller
|
||||
|
||||
import logbook
|
||||
log = logbook.Logger('ComponentTestCase')
|
||||
|
||||
allocator = AddressAllocator(1000)
|
||||
|
||||
|
||||
class ComponentTestCase(TestCase):
|
||||
|
||||
leased_sockets = defaultdict(list)
|
||||
|
||||
def setUp(self):
|
||||
self.zipline_test_config = {
|
||||
'allocator' : allocator,
|
||||
'sid' : 133,
|
||||
'devel' : False,
|
||||
'results_socket' : allocator.lease(1)[0],
|
||||
'simulation_style' : SIMULATION_STYLE.FIXED_SLIPPAGE
|
||||
}
|
||||
self.ctx = zmq.Context()
|
||||
setup_logger(self)
|
||||
|
||||
def tearDown(self):
|
||||
self.ctx.term()
|
||||
teardown_logger(self)
|
||||
|
||||
def test_specific_equity_source(self):
|
||||
filter = [1,2,3,4]
|
||||
#Set up source a. One minute between events.
|
||||
args_a = tuple()
|
||||
kwargs_a = {
|
||||
'sids' : [1,2],
|
||||
'start' : datetime(2012,6,6,0,tzinfo=pytz.utc),
|
||||
'delta' : timedelta(minutes = 1),
|
||||
'filter' : filter,
|
||||
'count' : 100
|
||||
}
|
||||
|
||||
c_id = SpecificEquityTrades.__name__ + hash_args(args_a, kwargs_a)
|
||||
mon = create_monitor(allocator)
|
||||
|
||||
out_socket_args = ComponentSocketArgs(
|
||||
style=zmq.PUSH,
|
||||
uri=allocator.lease(1)[0],
|
||||
bind=True
|
||||
)
|
||||
|
||||
c = Component(
|
||||
SpecificEquityTrades,
|
||||
args_a,
|
||||
kwargs_a,
|
||||
c_id,
|
||||
out_socket_args,
|
||||
DATASOURCE_FRAME,
|
||||
mon
|
||||
)
|
||||
|
||||
mon.manage(set([c.get_id]))
|
||||
mon_proc = launch_monitor(mon)
|
||||
|
||||
# launch in a process
|
||||
proc = launch_component(c)
|
||||
|
||||
pull_socket = self.ctx.socket(zmq.PULL)
|
||||
pull_socket.connect(out_socket_args.uri)
|
||||
poller = zmq.Poller()
|
||||
poller.register(pull_socket, zmq.POLLIN)
|
||||
unframe = DATASOURCE_UNFRAME
|
||||
for msg in gen_from_poller(poller, pull_socket, unframe):
|
||||
# assert things about the messages.
|
||||
log.info(msg)
|
||||
|
||||
pull_socket.close()
|
||||
log.info("DONE!")
|
||||
@@ -1,7 +1,7 @@
|
||||
from zipline.utils.test_utils import setup_logger, teardown_logger
|
||||
from unittest2 import TestCase, skip
|
||||
|
||||
from zipline.core.monitor import Controller
|
||||
from zipline.core.monitor import Monitor
|
||||
|
||||
class TestMonitor(TestCase):
|
||||
def setUp(self):
|
||||
@@ -15,12 +15,12 @@ class TestMonitor(TestCase):
|
||||
pub_socket = 'tcp://127.0.0.1:5000'
|
||||
route_socket = 'tcp://127.0.0.1:5001'
|
||||
|
||||
con = Controller(pub_socket, route_socket)
|
||||
con.manage([])
|
||||
mon = Monitor(pub_socket, route_socket)
|
||||
mon.manage([])
|
||||
|
||||
def test_init_topology(self):
|
||||
pub_socket = 'tcp://127.0.0.1:5000'
|
||||
route_socket = 'tcp://127.0.0.1:5001'
|
||||
|
||||
con = Controller(pub_socket, route_socket, )
|
||||
con.manage([ 'a', 'b', 'c', 'd' ])
|
||||
mon = Monitor(pub_socket, route_socket, )
|
||||
mon.manage([ 'a', 'b', 'c', 'd' ])
|
||||
|
||||
+2
-2
@@ -6,14 +6,14 @@ Zipline
|
||||
# it is a place to expose the public interfaces.
|
||||
|
||||
import protocol # namespace
|
||||
from core.monitor import Controller
|
||||
from core.monitor import Monitor
|
||||
from lines import SimulatedTrading
|
||||
from core.host import ComponentHost
|
||||
from utils.protocol_utils import ndict
|
||||
|
||||
__all__ = [
|
||||
SimulatedTrading,
|
||||
Controller,
|
||||
Monitor,
|
||||
ComponentHost,
|
||||
protocol,
|
||||
ndict
|
||||
|
||||
@@ -1,9 +1,9 @@
|
||||
from host import ComponentHost
|
||||
from component import Component
|
||||
from monitor import Controller
|
||||
from monitor import Monitor
|
||||
|
||||
__all__ = [
|
||||
Component,
|
||||
Controller,
|
||||
Monitor,
|
||||
ComponentHost
|
||||
]
|
||||
|
||||
+267
-449
@@ -11,10 +11,13 @@ import logbook
|
||||
import traceback
|
||||
import humanhash
|
||||
from setproctitle import setproctitle
|
||||
from collections import namedtuple
|
||||
|
||||
# pyzmq
|
||||
import zmq
|
||||
|
||||
from zipline.gens.zmqgen import gen_from_poller
|
||||
|
||||
from zipline.core.monitor import PARAMETERS
|
||||
|
||||
from zipline.protocol import (
|
||||
@@ -26,146 +29,85 @@ from zipline.protocol import (
|
||||
EXCEPTION_FRAME
|
||||
)
|
||||
|
||||
log = logbook.Logger('Component')
|
||||
|
||||
from zipline.exceptions import ComponentNoInit
|
||||
log = logbook.Logger('Component')
|
||||
|
||||
class KillSignal(Exception):
|
||||
def __init__(self):
|
||||
pass
|
||||
|
||||
ComponentSocketArgs = namedtuple('ComponentSocket',['uri','style','bind'])
|
||||
|
||||
class Component(object):
|
||||
|
||||
"""
|
||||
Base class for components. Defines the the base messaging
|
||||
interface for components.
|
||||
|
||||
:param addresses: a dict of name_string -> zmq port address strings.
|
||||
Must have the following entries
|
||||
|
||||
:param data_address: socket address used for data sources to stream
|
||||
their records. Will be used in PUSH/PULL sockets
|
||||
between data sources and a Feed. Bind will always
|
||||
be on the PULL side (we always have N producers and
|
||||
1 consumer)
|
||||
|
||||
:param feed_address: socket address used to publish consolidated feed
|
||||
from serialization of data sources
|
||||
will be used in PUB/SUB sockets between Feed and
|
||||
Transforms. Bind is always on the PUB side.
|
||||
|
||||
:param merge_address: socket address used to publish transformed
|
||||
values. will be used in PUSH/PULL from many
|
||||
transforms to one Merge Bind will always be on
|
||||
the PULL side (we always have N producers and
|
||||
1 consumer)
|
||||
|
||||
:param results_address: socket address used to publish merged data
|
||||
source feed and transforms to clients will be
|
||||
used in PUB/SUB from one Merge to one or many
|
||||
clients. Bind is always on the PUB side.
|
||||
|
||||
bind/connect methods will return the correct socket type for each
|
||||
address.
|
||||
|
||||
"""
|
||||
|
||||
# ------------
|
||||
# Construction
|
||||
# ------------
|
||||
|
||||
abstract = True
|
||||
#__metaclass__ = WorkflowMeta
|
||||
def __init__(self,
|
||||
gen_func,
|
||||
gen_args,
|
||||
gen_kwargs,
|
||||
component_id,
|
||||
out_socket_args,
|
||||
frame,
|
||||
monitor,
|
||||
in_socket_args=None,
|
||||
unframe=None
|
||||
):
|
||||
|
||||
def __init__(self, *args, **kwargs):
|
||||
self.zmq = None
|
||||
self.context = None
|
||||
self.addresses = None
|
||||
self.waiting = None
|
||||
assert component_id, \
|
||||
"Every component needs a unique and invariant identifier"
|
||||
assert isinstance(component_id, basestring), \
|
||||
"Components must have string IDs"
|
||||
assert isinstance(out_socket_args, ComponentSocketArgs), \
|
||||
"out_socket_args args must be ComponentSocketArgs"
|
||||
|
||||
if in_socket_args:
|
||||
assert isinstance(in_socket_args, ComponentSocketArgs), \
|
||||
"in_socket_args args must be ComponentSocketArgs"
|
||||
|
||||
# -----------------
|
||||
# Generator
|
||||
# -----------------
|
||||
self.component_id = component_id
|
||||
self.gen_args = gen_args
|
||||
self.gen_kwargs = gen_kwargs
|
||||
self.gen_func = gen_func
|
||||
self.generator = None
|
||||
self.frame = frame
|
||||
|
||||
# lock for waiting on monitor "GO"
|
||||
self.waiting = None
|
||||
|
||||
# -----------------
|
||||
# ZMQ properties
|
||||
# -----------------
|
||||
self.in_socket_args = in_socket_args
|
||||
self.out_socket_args = out_socket_args
|
||||
self.zmq = None
|
||||
self.context = None
|
||||
self.out_socket = None
|
||||
self.in_socket = None
|
||||
self.monitor = monitor
|
||||
self.unframe = unframe
|
||||
|
||||
self.out_socket = None
|
||||
self.killed = False
|
||||
self.controller = None
|
||||
# timeout on heartbeat is very short to avoid burning
|
||||
# cycles on heartbeating. unit is milliconds
|
||||
self.heartbeat_timeout = 0
|
||||
# TODO: state_flag is deprecated, remove
|
||||
# TODO: error_state is deprecated, remove
|
||||
self.state_flag = COMPONENT_STATE.OK
|
||||
self.error_state = COMPONENT_FAILURE.NOFAILURE
|
||||
self.on_done = None
|
||||
self.state_flag = COMPONENT_STATE.OK
|
||||
|
||||
self._exception = None
|
||||
self.fail_time = None
|
||||
self.start_tic = None
|
||||
self.stop_tic = None
|
||||
self.note = None
|
||||
self.confirmed = False
|
||||
self.devel = False
|
||||
self.socks = None
|
||||
self.last_ping = None
|
||||
# track time of last ping we received from monitor
|
||||
self.last_ping = None
|
||||
|
||||
# Humanhashes make this way easier to debug because they stick
|
||||
# in your mind unlike a 32 byte string of random hex.
|
||||
self.guid = uuid.uuid4()
|
||||
self.huid = humanhash.humanize(self.guid.hex)
|
||||
|
||||
# This is where component specific constructors should be
|
||||
# defined. Arguments passed to init are threaded through.
|
||||
self.init(*args, **kwargs)
|
||||
|
||||
def init(self):
|
||||
"""
|
||||
Subclasses should override this to extend the setup for the
|
||||
class. Shouldn't have side effects.
|
||||
"""
|
||||
raise ComponentNoInit(self.__class__)
|
||||
self.guid = uuid.uuid4()
|
||||
self.huid = humanhash.humanize(self.guid.hex)
|
||||
|
||||
|
||||
# ------------
|
||||
# Core Methods
|
||||
# ------------
|
||||
|
||||
def open(self):
|
||||
"""
|
||||
Open the connections needed to start doing work.
|
||||
"""
|
||||
raise NotImplementedError
|
||||
|
||||
def ready(self):
|
||||
"""
|
||||
Return ``True`` if and only if the component has finished
|
||||
execution.
|
||||
"""
|
||||
return self.state_flag in [COMPONENT_STATE.DONE, \
|
||||
COMPONENT_STATE.EXCEPTION]
|
||||
|
||||
def successful(self):
|
||||
"""
|
||||
Return ``True`` if and only if the component has finished
|
||||
execution successfully, that is, without raising an error.
|
||||
"""
|
||||
return self.state_flag == COMPONENT_STATE.DONE and not \
|
||||
self.exception
|
||||
|
||||
@property
|
||||
def exception(self):
|
||||
"""
|
||||
Holds the exception that the component failed on, or ``None`` if
|
||||
the component has not failed.
|
||||
"""
|
||||
return self._exception
|
||||
|
||||
def do_work(self):
|
||||
raise NotImplementedError
|
||||
|
||||
def init_zmq(self):
|
||||
self.zmq = zmq
|
||||
self.context = self.zmq.Context()
|
||||
self.zmq_poller = self.zmq.Poller
|
||||
# The the process title so you can watch it in top
|
||||
setproctitle(self.__class__.__name__)
|
||||
return
|
||||
|
||||
def _run(self):
|
||||
"""
|
||||
@@ -174,33 +116,32 @@ class Component(object):
|
||||
|
||||
The core logic of the all components is run here.
|
||||
"""
|
||||
# The process title so you can watch it in top, ps.
|
||||
setproctitle(self.gen_func.__name__)
|
||||
|
||||
log.info("Start %r" % self)
|
||||
log.info("Pid %s" % os.getpid())
|
||||
log.info("Group %s" % os.getpgrp())
|
||||
|
||||
self.start_tic = time.time()
|
||||
|
||||
self.done = False # TODO: use state flag
|
||||
self.sockets = []
|
||||
|
||||
self.init_zmq()
|
||||
|
||||
self.setup_poller()
|
||||
|
||||
self.setup_control()
|
||||
self.open()
|
||||
|
||||
self.signal_ready()
|
||||
self.lock_ready()
|
||||
self.wait_ready()
|
||||
|
||||
# -----------------------
|
||||
# YOU SHALL NOT PASS!!!!!
|
||||
# -----------------------
|
||||
# ... until the controller signals GO
|
||||
# ... until the monitor signals GO
|
||||
|
||||
self.loop()
|
||||
for event in self.generator:
|
||||
self.heartbeat()
|
||||
msg = self.frame(event)
|
||||
self.out_socket.send(msg)
|
||||
|
||||
self.stop_tic = time.time()
|
||||
self.signal_done()
|
||||
|
||||
def run(self, catch_exceptions=True):
|
||||
"""
|
||||
@@ -214,119 +155,10 @@ class Component(object):
|
||||
else:
|
||||
# if we get a kill signal, forcibly close all the
|
||||
# sockets.
|
||||
# exc_info = sys.exc_info()
|
||||
# self.relay_exception(exc_info[0], exc_info[1], exc_info[2])
|
||||
self.teardown_sockets()
|
||||
|
||||
finally:
|
||||
self.shutdown()
|
||||
log.info("Exiting %r" % self)
|
||||
|
||||
def working(self):
|
||||
"""
|
||||
Controls when the work loop will start and end
|
||||
|
||||
If we encounter an exception or signal done exit.
|
||||
|
||||
Overload for higher order behavior.
|
||||
"""
|
||||
return (not self.done)
|
||||
|
||||
def loop(self, lockstep=True):
|
||||
"""
|
||||
Loop to do work while we still have work to do.
|
||||
"""
|
||||
while self.working():
|
||||
self.heartbeat()
|
||||
self.do_work()
|
||||
|
||||
def runtime(self):
|
||||
if self.ready() and self.start_tic and self.stop_tic:
|
||||
return self.stop_tic - self.start_tic
|
||||
|
||||
def heartbeat(self, timeout=0):
|
||||
# wait for synchronization reply from the host
|
||||
self.socks = dict(self.poll.poll(timeout))
|
||||
|
||||
# ----------------
|
||||
# Control Dispatch
|
||||
# ----------------
|
||||
assert self.control_in, 'Component does not have a control_in socket'
|
||||
|
||||
# If we're in devel mode drop out because the controller
|
||||
# isn't guaranteed to be around anymore
|
||||
if self.devel:
|
||||
log.warn("Skipping heartbeat because of devel flag")
|
||||
return
|
||||
|
||||
if self.socks.get(self.control_in) == zmq.POLLIN:
|
||||
msg = self.control_in.recv()
|
||||
event, payload = CONTROL_UNFRAME(msg)
|
||||
|
||||
# ===========
|
||||
# Heartbeat
|
||||
# ===========
|
||||
|
||||
# The controller will send out a single number packed in
|
||||
# a CONTROL_FRAME with ``heartbeat`` event every
|
||||
# (n)-seconds. The component then has n seconds to
|
||||
# respond to it. If not then it will be considered as
|
||||
# malfunctioning or maybe CPU bound.
|
||||
|
||||
if event == CONTROL_PROTOCOL.HEARTBEAT:
|
||||
# Heart outgoing
|
||||
heartbeat_frame = CONTROL_FRAME(
|
||||
CONTROL_PROTOCOL.OK,
|
||||
payload
|
||||
)
|
||||
|
||||
self.last_ping = float(payload)
|
||||
# Echo back the heartbeat identifier to tell the
|
||||
# controller that this component is still alive and
|
||||
# doing work
|
||||
self.control_out.send(heartbeat_frame)
|
||||
|
||||
|
||||
# =========
|
||||
# Soft Kill
|
||||
# =========
|
||||
|
||||
# Try and clean up properly and send out any reports or
|
||||
# data that are done during a clean shutdown. Inform the
|
||||
# controller that we're done.
|
||||
elif event == CONTROL_PROTOCOL.SHUTDOWN:
|
||||
self.signal_done()
|
||||
|
||||
# =========
|
||||
# Hard Kill
|
||||
# =========
|
||||
|
||||
# Just exit.
|
||||
elif event == CONTROL_PROTOCOL.KILL:
|
||||
self.kill()
|
||||
|
||||
# In case we didn't receive a ping, send a pre-emptive
|
||||
# pong to the monitor.
|
||||
elif hasattr(self, 'control_out') and \
|
||||
self.last_ping and \
|
||||
time.time() - self.last_ping > 1:
|
||||
# send a ping ahead of schedule
|
||||
pre_pong = time.time()
|
||||
heartbeat_frame = CONTROL_FRAME(
|
||||
CONTROL_PROTOCOL.OK,
|
||||
str(pre_pong)
|
||||
)
|
||||
|
||||
# Echo back the heartbeat identifier to tell the
|
||||
# controller that this component is still alive and
|
||||
# doing work
|
||||
self.control_out.send(heartbeat_frame, self.zmq.NOBLOCK)
|
||||
self.last_ping = pre_pong
|
||||
elif self.last_ping and \
|
||||
time.time() - self.last_ping > PARAMETERS.MAX_COMPONENT_WAIT:
|
||||
# monitor is gone without sending the shutdown
|
||||
# signal, do a hard exit.
|
||||
self.kill()
|
||||
|
||||
# ----------------------------
|
||||
# Cleanup & Modes of Failure
|
||||
@@ -342,16 +174,6 @@ class Component(object):
|
||||
for sock in self.sockets:
|
||||
sock.close()
|
||||
|
||||
def shutdown(self):
|
||||
"""
|
||||
Clean shutdown.
|
||||
|
||||
Tear down after normal operation.
|
||||
"""
|
||||
if self.on_done:
|
||||
log.warn("{id} calling done.".format(id=self.get_id))
|
||||
self.on_done()
|
||||
|
||||
def kill(self):
|
||||
"""
|
||||
Unclean shutdown.
|
||||
@@ -359,9 +181,64 @@ class Component(object):
|
||||
Tear down ( fast ) as a mode of failure in the simulation or on
|
||||
service halt.
|
||||
"""
|
||||
# sys.exit(1)
|
||||
raise KillSignal()
|
||||
|
||||
def signal_exception(self, exc=None, scope=None):
|
||||
"""
|
||||
All exceptions inside any component should boil back to
|
||||
this handler.
|
||||
|
||||
Will inform the system that the component has failed and how it
|
||||
has failed.
|
||||
"""
|
||||
self.state_flag = COMPONENT_STATE.EXCEPTION
|
||||
exc_type, exc_value, exc_traceback = sys.exc_info()
|
||||
|
||||
# if a downstream component fails, this component may try
|
||||
# sending when there are zero connections to the socket,
|
||||
# which will raise ZMQError(EAGAIN). So, it doesn't make
|
||||
# sense to relay this exception to Monitor and the rest
|
||||
# of the zipline.
|
||||
if isinstance(exc, zmq.ZMQError) and exc.errno == zmq.EAGAIN:
|
||||
log.warn("{id} raised a ZMQError(EAGAIN) not relaying"\
|
||||
.format(id=self.get_id))
|
||||
return
|
||||
|
||||
# sys.stdout.write(trace)
|
||||
log.exception("Unexpected error in run for {id}.".format(id=self.get_id))
|
||||
|
||||
try:
|
||||
log.info('{id} sending exception to monitor'\
|
||||
.format(id=self.get_id))
|
||||
msg = EXCEPTION_FRAME(
|
||||
exc_traceback,
|
||||
exc_type.__name__,
|
||||
exc_value.message
|
||||
)
|
||||
|
||||
exception_frame = CONTROL_FRAME(
|
||||
CONTROL_PROTOCOL.EXCEPTION,
|
||||
msg
|
||||
)
|
||||
self.control_out.send(exception_frame, self.zmq.NOBLOCK)
|
||||
# The monitor should relay the exception back
|
||||
# to all zipline components. Wait here until the
|
||||
# notice arrives, and we can assume other zipline
|
||||
# components have broken out of their message
|
||||
# loops.
|
||||
for i in xrange(PARAMETERS.MAX_COMPONENT_WAIT):
|
||||
self.heartbeat(timeout=1000)
|
||||
log.warn("{id} never heard back from monitor."\
|
||||
.format(id=self.get_id))
|
||||
|
||||
except KillSignal:
|
||||
log.info("{id} received confirmation from monitor"\
|
||||
.format(id=self.get_id))
|
||||
except:
|
||||
log.exception("Exception waiting for monitor reply")
|
||||
|
||||
|
||||
|
||||
# ----------------------
|
||||
# Internal Maintenance
|
||||
# ----------------------
|
||||
@@ -405,7 +282,7 @@ class Component(object):
|
||||
# Go
|
||||
# ====
|
||||
|
||||
# A distributed lock from the controller to ensure
|
||||
# A distributed lock from the monitor to ensure
|
||||
# synchronized start.
|
||||
|
||||
if event == CONTROL_PROTOCOL.HEARTBEAT:
|
||||
@@ -417,7 +294,7 @@ class Component(object):
|
||||
log.info('Prestart Heartbeat ' + self.get_id)
|
||||
|
||||
elif event == CONTROL_PROTOCOL.GO:
|
||||
# Side effectful call from the controller to unlock
|
||||
# Side effectful call from the monitor to unlock
|
||||
# and begin doing work only when the entire topology
|
||||
# of the system beings to come online
|
||||
log.info('Unlocking ' + self.__class__.__name__)
|
||||
@@ -429,7 +306,7 @@ class Component(object):
|
||||
|
||||
# Try and clean up properly and send out any reports or
|
||||
# data that are done during a clean shutdown. Inform the
|
||||
# controller that we're done.
|
||||
# monitor that we're done.
|
||||
elif event == CONTROL_PROTOCOL.SHUTDOWN:
|
||||
self.signal_done()
|
||||
break
|
||||
@@ -449,97 +326,90 @@ class Component(object):
|
||||
self.kill()
|
||||
break
|
||||
|
||||
def heartbeat(self, timeout=0):
|
||||
# wait for synchronization reply from the host
|
||||
socks = dict(self.poll.poll(timeout))
|
||||
|
||||
# ----------------
|
||||
# Control Dispatch
|
||||
# ----------------
|
||||
assert self.control_in, 'Component does not have a control_in socket'
|
||||
|
||||
if socks.get(self.control_in) == zmq.POLLIN:
|
||||
msg = self.control_in.recv()
|
||||
event, payload = CONTROL_UNFRAME(msg)
|
||||
|
||||
# ===========
|
||||
# Heartbeat
|
||||
# ===========
|
||||
|
||||
# The monitor will send out a single number packed in
|
||||
# a CONTROL_FRAME with ``heartbeat`` event every
|
||||
# (n)-seconds. The component then has n seconds to
|
||||
# respond to it. If not then it will be considered as
|
||||
# malfunctioning or maybe CPU bound.
|
||||
|
||||
if event == CONTROL_PROTOCOL.HEARTBEAT:
|
||||
# Heart outgoing
|
||||
heartbeat_frame = CONTROL_FRAME(
|
||||
CONTROL_PROTOCOL.OK,
|
||||
payload
|
||||
)
|
||||
|
||||
self.last_ping = float(payload)
|
||||
# Echo back the heartbeat identifier to tell the
|
||||
# monitor that this component is still alive and
|
||||
# doing work
|
||||
self.control_out.send(heartbeat_frame)
|
||||
|
||||
|
||||
# =========
|
||||
# Soft Kill
|
||||
# =========
|
||||
|
||||
# Try and clean up properly and send out any reports or
|
||||
# data that are done during a clean shutdown. Inform the
|
||||
# monitor that we're done.
|
||||
elif event == CONTROL_PROTOCOL.SHUTDOWN:
|
||||
self.signal_done()
|
||||
|
||||
# =========
|
||||
# Hard Kill
|
||||
# =========
|
||||
|
||||
# Just exit.
|
||||
elif event == CONTROL_PROTOCOL.KILL:
|
||||
self.kill()
|
||||
|
||||
# In case we didn't receive a ping, send a pre-emptive
|
||||
# pong to the monitor.
|
||||
elif self.last_ping and time.time() - self.last_ping > 1:
|
||||
# send a ping ahead of schedule
|
||||
pre_pong = time.time()
|
||||
heartbeat_frame = CONTROL_FRAME(
|
||||
CONTROL_PROTOCOL.OK,
|
||||
str(pre_pong)
|
||||
)
|
||||
|
||||
# Echo back the heartbeat identifier to tell the
|
||||
# monitor that this component is still alive and
|
||||
# doing work
|
||||
self.control_out.send(heartbeat_frame, self.zmq.NOBLOCK)
|
||||
self.last_ping = pre_pong
|
||||
elif self.last_ping and \
|
||||
time.time() - self.last_ping > PARAMETERS.MAX_COMPONENT_WAIT:
|
||||
# monitor is gone without sending the shutdown
|
||||
# signal, do a hard exit.
|
||||
self.kill()
|
||||
|
||||
|
||||
def signal_ready(self):
|
||||
log.info(self.__class__.__name__ + ' is ready')
|
||||
|
||||
if hasattr(self, 'control_out'):
|
||||
frame = CONTROL_FRAME(
|
||||
CONTROL_PROTOCOL.READY,
|
||||
''
|
||||
)
|
||||
self.control_out.send(frame)
|
||||
|
||||
def signal_cancel(self):
|
||||
self.done = True
|
||||
|
||||
# TODO: no hasattr hacks
|
||||
#if not self.controller:
|
||||
if hasattr(self, 'control_out'):
|
||||
frame = CONTROL_FRAME(
|
||||
CONTROL_PROTOCOL.SHUTDOWN,
|
||||
None
|
||||
)
|
||||
self.control_out.send(frame)
|
||||
|
||||
# then proceeds to do shutdown(), and teardown_sockets()
|
||||
# to complete the process
|
||||
|
||||
def signal_exception(self, exc=None, scope=None):
|
||||
"""
|
||||
All exceptions inside any component should boil back to
|
||||
this handler.
|
||||
|
||||
Will inform the system that the component has failed and how it
|
||||
has failed.
|
||||
"""
|
||||
|
||||
if scope == 'algo':
|
||||
self.error_state = COMPONENT_FAILURE.ALGOEXCEPT
|
||||
else:
|
||||
self.error_state = COMPONENT_FAILURE.HOSTEXCEPT
|
||||
|
||||
self.state_flag = COMPONENT_STATE.EXCEPTION
|
||||
# mark the time of failure so we can track the failure
|
||||
# progogation through the system.
|
||||
|
||||
self.stop_tic = time.time()
|
||||
|
||||
self._exception = exc
|
||||
exc_type, exc_value, exc_traceback = sys.exc_info()
|
||||
|
||||
# if a downstream component fails, this component may try
|
||||
# sending when there are zero connections to the socket,
|
||||
# which will raise ZMQError(EAGAIN). So, it doesn't make
|
||||
# sense to relay this exception to Monitor and the rest
|
||||
# of the zipline.
|
||||
if isinstance(exc, zmq.ZMQError) and exc.errno == zmq.EAGAIN:
|
||||
log.warn("{id} raised a ZMQError(EAGAIN) not relaying"\
|
||||
.format(id=self.get_id))
|
||||
return
|
||||
|
||||
# sys.stdout.write(trace)
|
||||
log.exception("Unexpected error in run for {id}.".format(id=self.get_id))
|
||||
|
||||
if hasattr(self, 'control_out') and self.control_out:
|
||||
try:
|
||||
log.info('{id} sending exception to controller'\
|
||||
.format(id=self.get_id))
|
||||
msg = EXCEPTION_FRAME(
|
||||
exc_traceback,
|
||||
exc_type.__name__,
|
||||
exc_value.message
|
||||
)
|
||||
|
||||
exception_frame = CONTROL_FRAME(
|
||||
CONTROL_PROTOCOL.EXCEPTION,
|
||||
msg
|
||||
)
|
||||
self.control_out.send(exception_frame, self.zmq.NOBLOCK)
|
||||
# The controller should relay the exception back
|
||||
# to all zipline components. Wait here until the
|
||||
# notice arrives, and we can assume other zipline
|
||||
# components have broken out of their message
|
||||
# loops.
|
||||
for i in xrange(PARAMETERS.MAX_COMPONENT_WAIT):
|
||||
self.heartbeat(timeout=1000)
|
||||
log.warn("{id} never heard back from monitor."\
|
||||
.format(id=self.get_id))
|
||||
except KillSignal:
|
||||
log.info("{id} received confirmation from controller"\
|
||||
.format(id=self.get_id))
|
||||
except:
|
||||
log.exception("Exception waiting for controller reply")
|
||||
frame = CONTROL_FRAME(
|
||||
CONTROL_PROTOCOL.READY,
|
||||
''
|
||||
)
|
||||
self.control_out.send(frame)
|
||||
|
||||
def signal_done(self):
|
||||
"""
|
||||
@@ -550,20 +420,18 @@ class Component(object):
|
||||
# notify internal work loop that we're done
|
||||
self.done = True # TODO: use state flag
|
||||
|
||||
if hasattr(self, 'out_socket') and self.out_socket:
|
||||
msg = zmq.Message(str(CONTROL_PROTOCOL.DONE))
|
||||
self.out_socket.send(msg)
|
||||
msg = zmq.Message(str(CONTROL_PROTOCOL.DONE))
|
||||
self.out_socket.send(msg)
|
||||
|
||||
|
||||
if hasattr(self, 'control_out'):
|
||||
# notify controller we're done
|
||||
done_frame = CONTROL_FRAME(
|
||||
CONTROL_PROTOCOL.DONE,
|
||||
''
|
||||
)
|
||||
# notify monitor we're done
|
||||
done_frame = CONTROL_FRAME(
|
||||
CONTROL_PROTOCOL.DONE,
|
||||
''
|
||||
)
|
||||
|
||||
self.control_out.send(done_frame)
|
||||
log.info("[%s] sent control done" % self.get_id)
|
||||
self.control_out.send(done_frame)
|
||||
log.info("[%s] sent control done" % self.get_id)
|
||||
|
||||
# there is a narrow race condition where we finish just
|
||||
# after the Monitor accepts our prior heartbeat, but just
|
||||
@@ -571,43 +439,64 @@ class Component(object):
|
||||
# last heartbeat, and wait an unusually long time.
|
||||
self.heartbeat(timeout=5000)
|
||||
|
||||
|
||||
|
||||
|
||||
# -----------
|
||||
# Messaging
|
||||
# -----------
|
||||
|
||||
def setup_poller(self):
|
||||
def open(self):
|
||||
"""
|
||||
Setup the poller used for multiplexing the incoming data
|
||||
handling sockets.
|
||||
Open the connections needed to start doing work.
|
||||
Perform any setup that must be done within process.
|
||||
"""
|
||||
self.poll = self.zmq_poller()
|
||||
|
||||
def bind_data(self):
|
||||
return self.bind_pull_socket(self.addresses['data_address'])
|
||||
self.zmq = zmq
|
||||
self.context = self.zmq.Context()
|
||||
self.poll = self.zmq.Poller()
|
||||
|
||||
def connect_data(self):
|
||||
return self.connect_push_socket(self.addresses['data_address'])
|
||||
self.setup_control()
|
||||
|
||||
def bind_feed(self):
|
||||
return self.bind_pub_socket(self.addresses['feed_address'])
|
||||
if self.in_socket_args:
|
||||
self.in_socket = self.open_socket(self.in_socket_args)
|
||||
poller_gen = gen_from_poller(
|
||||
self.poller,
|
||||
self.in_socket,
|
||||
self.unframe
|
||||
)
|
||||
self.generator = self.gen_func(
|
||||
poller_gen,
|
||||
*self.gen_args,
|
||||
**self.gen_kwargs
|
||||
)
|
||||
else:
|
||||
self.generator = self.gen_func(*self.gen_args, **self.gen_kwargs)
|
||||
|
||||
def connect_feed(self):
|
||||
return self.connect_sub_socket(self.addresses['feed_address'])
|
||||
self.out_socket = self.open_socket(self.out_socket_args)
|
||||
|
||||
def bind_merge(self):
|
||||
return self.bind_pull_socket(self.addresses['merge_address'])
|
||||
def open_socket(self, sock_args):
|
||||
if sock_args.bind:
|
||||
return self.bind_socket(sock_args)
|
||||
else:
|
||||
return self.connect_socket(sock_args)
|
||||
|
||||
def connect_merge(self):
|
||||
return self.connect_push_socket(self.addresses['merge_address'])
|
||||
def bind_socket(self, sock_args):
|
||||
if sock_args.style == zmq.PULL:
|
||||
return self.bind_pull_socket(sock_args.uri)
|
||||
if sock_args.style == zmq.PUSH:
|
||||
return self.bind_push_socket(sock_args.uri)
|
||||
if sock_args.style == zmq.PUB:
|
||||
return self.bind_pub_socket(sock_args.uri)
|
||||
|
||||
def bind_result(self):
|
||||
return self.bind_push_socket(self.addresses['results_address'])
|
||||
raise Exception("Invalid socket arguments")
|
||||
|
||||
def connect_result(self):
|
||||
return self.connect_pull_socket(self.addresses['results_address'])
|
||||
def connect_socket(self, sock_args):
|
||||
if sock_args.style == zmq.PULL:
|
||||
return self.connect_pull_socket(sock_args.uri)
|
||||
if sock_args.style == zmq.PUSH:
|
||||
return self.connect_push_socket(sock_args.uri)
|
||||
if sock_args.style == zmq.SUB:
|
||||
return self.connect_sub_socket(sock_args.uri)
|
||||
|
||||
raise Exception("Invalid socket arguments")
|
||||
|
||||
def bind_push_socket(self, addr):
|
||||
push_socket = self.context.socket(self.zmq.PUSH)
|
||||
@@ -638,7 +527,6 @@ class Component(object):
|
||||
def connect_push_socket(self, addr):
|
||||
push_socket = self.context.socket(self.zmq.PUSH)
|
||||
push_socket.connect(addr)
|
||||
#push_socket.setsockopt(self.zmq.LINGER,0)
|
||||
self.sockets.append(push_socket)
|
||||
self.out_socket = push_socket
|
||||
|
||||
@@ -647,7 +535,6 @@ class Component(object):
|
||||
def bind_pub_socket(self, addr):
|
||||
pub_socket = self.context.socket(self.zmq.PUB)
|
||||
pub_socket.bind(addr)
|
||||
#pub_socket.setsockopt(self.zmq.LINGER, 0)
|
||||
self.out_socket = pub_socket
|
||||
|
||||
return pub_socket
|
||||
@@ -668,92 +555,29 @@ class Component(object):
|
||||
of the simulation and to forcefully tear down the simulation in
|
||||
case of a failure.
|
||||
"""
|
||||
|
||||
# Allow for the possibility of not having a controller,
|
||||
# possibly the zipline devsimulator may not want this.
|
||||
if not self.controller:
|
||||
return
|
||||
|
||||
self.control_out = self.controller.message_sender(
|
||||
self.control_out = self.monitor.message_sender(
|
||||
identity = self.get_id,
|
||||
context = self.context,
|
||||
)
|
||||
|
||||
self.control_in = self.controller.message_listener(
|
||||
self.control_in = self.monitor.message_listener(
|
||||
context = self.context
|
||||
)
|
||||
|
||||
self.poll.register(self.control_in, self.zmq.POLLIN)
|
||||
self.sockets.extend([self.control_in, self.control_out])
|
||||
|
||||
# -----------
|
||||
# FSM Actions
|
||||
# -----------
|
||||
|
||||
#@property
|
||||
#def state(self):
|
||||
#if not hasattr(self, '_state'):
|
||||
#self._state = self.initial_state
|
||||
#else:
|
||||
#return self._state
|
||||
|
||||
#@state.setter
|
||||
#def state(self, new):
|
||||
#if not hasattr(self, '_state'):
|
||||
#self._state = self.initial_state
|
||||
|
||||
#old = self._state
|
||||
|
||||
#if (old, new) in self.workflow:
|
||||
#self._state = new
|
||||
#else:
|
||||
#raise RuntimeError("Invalid State Transition : %s -> %s" %(old, new))
|
||||
|
||||
# ---------------------
|
||||
# Description and Debug
|
||||
# ---------------------
|
||||
|
||||
def extern_logger(self):
|
||||
"""
|
||||
Pipe logs out to a provided logging interface.
|
||||
"""
|
||||
pass
|
||||
|
||||
def setup_extern_logger(self):
|
||||
"""
|
||||
Pipe logs out to a provided logging interface.
|
||||
"""
|
||||
pass
|
||||
|
||||
@property
|
||||
def get_id(self):
|
||||
"""
|
||||
The descriptive name of the component.
|
||||
The time invariant name for this component.
|
||||
Must be unique within this zipline.
|
||||
"""
|
||||
# Prevents the bug that Thomas ran into
|
||||
raise NotImplementedError
|
||||
|
||||
@property
|
||||
def get_type(self):
|
||||
"""
|
||||
The data flow type of the component.
|
||||
|
||||
- ``SOURCE``
|
||||
- ``CONDUIT``
|
||||
- ``SINK``
|
||||
|
||||
"""
|
||||
raise NotImplementedError
|
||||
|
||||
@property
|
||||
def get_pure(self):
|
||||
"""
|
||||
Describes whehter this component purely functional, i.e. for a
|
||||
given set of inputs is it guaranteed to always give the same
|
||||
output . Components that are side-effectful are, generally, not
|
||||
pure.
|
||||
"""
|
||||
return False
|
||||
return self.component_id
|
||||
|
||||
def debug(self):
|
||||
"""
|
||||
@@ -766,18 +590,12 @@ class Component(object):
|
||||
'pid' : os.getpid() ,
|
||||
'memaddress' : hex(id(self)) ,
|
||||
'ready' : self.successful() ,
|
||||
'succesfull' : self.ready() ,
|
||||
'successful' : self.ready() ,
|
||||
}
|
||||
|
||||
def __len__(self):
|
||||
"""
|
||||
Some components overload this for debug purposes
|
||||
"""
|
||||
raise NotImplementedError
|
||||
|
||||
def __repr__(self):
|
||||
"""
|
||||
Return a usefull string representation of the component to
|
||||
Return a useful string representation of the component to
|
||||
indicate its type, unique identifier, and computational context
|
||||
identifier name.
|
||||
"""
|
||||
|
||||
@@ -103,7 +103,7 @@ class ComponentHost(object):
|
||||
|
||||
|
||||
log.info('== Roll Call ==')
|
||||
log.info('Controller')
|
||||
log.info('Monitor')
|
||||
|
||||
self.launch_controller()
|
||||
|
||||
|
||||
@@ -38,7 +38,7 @@ class UnknownChatter(Exception):
|
||||
return """Component calling itself "%s" talking on unexpected channel""" % self.named
|
||||
|
||||
|
||||
log = logbook.Logger('Controller')
|
||||
log = logbook.Logger('Monitor')
|
||||
|
||||
# The scalars determining the timing of the monitor behavior for
|
||||
# the system.
|
||||
@@ -56,7 +56,7 @@ PARAMETERS = ndict(dict(
|
||||
SYSTEM_TIMEOUT = 50,
|
||||
))
|
||||
|
||||
class Controller(object):
|
||||
class Monitor(object):
|
||||
"""
|
||||
A N to M messaging system for inter component communication.
|
||||
|
||||
@@ -103,8 +103,6 @@ class Controller(object):
|
||||
self.route_socket = route_socket
|
||||
self.exception_socket = exception_socket
|
||||
|
||||
self.error_replay = OrderedDict()
|
||||
|
||||
self.missed_beats = Counter()
|
||||
|
||||
self.send_sighup = send_sighup
|
||||
@@ -499,7 +497,6 @@ class Controller(object):
|
||||
# Error Handling
|
||||
# --------------
|
||||
def exception(self, component, exception_data):
|
||||
self.error_replay[(component, time.time())] = exception_data
|
||||
log.error('Component in exception state: %s. Shutting down system and sending exception data to listeners.'\
|
||||
% component)
|
||||
# Send the exception message out to listeners.
|
||||
@@ -616,10 +613,6 @@ class Controller(object):
|
||||
self.associated.append(s)
|
||||
return s
|
||||
|
||||
def do_error_replay(self):
|
||||
for (component, time), error in self.error_replay.iteritems():
|
||||
log.info('Component Log for -- %s --:\n%s' % (component, error))
|
||||
|
||||
def kill(self):
|
||||
"""Aggressively exit the whole zipline.
|
||||
"""
|
||||
|
||||
@@ -40,13 +40,13 @@ class ProcessSimulator(ComponentHost):
|
||||
# invoked by the host's open()
|
||||
|
||||
def launch_controller(self):
|
||||
proc = multiprocessing.Process(target=self.controller.run)
|
||||
proc = multiprocessing.Process(target=self.monitor.run)
|
||||
proc.start()
|
||||
self.con = proc
|
||||
|
||||
# Process specific
|
||||
self.controller_process = proc
|
||||
self.mapping[proc.pid] = 'Controller'
|
||||
self.monitor_process = proc
|
||||
self.mapping[proc.pid] = 'Monitor'
|
||||
|
||||
def launch_component(self, component):
|
||||
proc = multiprocessing.Process(target=component.run)
|
||||
@@ -81,7 +81,7 @@ class ProcessSimulator(ComponentHost):
|
||||
process.join(timeout=1)
|
||||
process.terminate()
|
||||
|
||||
self.controller.shutdown(soft=True)
|
||||
self.monitor.shutdown(soft=True)
|
||||
self.running = False
|
||||
|
||||
self.con.terminate()
|
||||
|
||||
@@ -19,8 +19,8 @@ def mock_raw_event(sid, dt):
|
||||
|
||||
def mock_done(id):
|
||||
return ndict({
|
||||
'dt' : "DONE",
|
||||
"source_id" : id,
|
||||
'dt' : "DONE",
|
||||
"source_id" : id,
|
||||
'tnfm_id' : id,
|
||||
'tnfm_value': None,
|
||||
'type' : 0
|
||||
@@ -43,7 +43,7 @@ def roundrobin(sources, namestrings):
|
||||
"""
|
||||
assert len(sources) == len(namestrings)
|
||||
mapping = OrderedDict(zip(namestrings, sources))
|
||||
|
||||
|
||||
# While our generators have not been exhausted, pull elements
|
||||
while mapping.keys() != []:
|
||||
for namestring, source in mapping.iteritems():
|
||||
|
||||
@@ -0,0 +1,27 @@
|
||||
import zmq
|
||||
import zipline.protocol as zp
|
||||
|
||||
def gen_from_pull_socket(socket_uri, context, unframe):
|
||||
"""
|
||||
A generator that takes a socket_uri, and yields
|
||||
messages from the poller until it gets a zp.CONTROL_PROTOCOL.DONE.
|
||||
"""
|
||||
pull_socket = context.socket(zmq.PULL)
|
||||
pull_socket.connect(socket_uri)
|
||||
poller = zmq.Poller()
|
||||
poller.register(pull_socket, zmq.POLLIN)
|
||||
|
||||
return gen_from_poller(poller, pull_socket, unframe)
|
||||
|
||||
def gen_from_poller(poller, in_socket, unframe):
|
||||
|
||||
while True:
|
||||
socks = dict(poller.poll(1000))
|
||||
|
||||
if socks.get(in_socket) == zmq.POLLIN:
|
||||
message = in_socket.recv()
|
||||
if message == str(zp.CONTROL_PROTOCOL.DONE):
|
||||
break
|
||||
else:
|
||||
event = unframe(message)
|
||||
yield event
|
||||
+7
-7
@@ -70,7 +70,7 @@ from zipline.transforms import BaseTransform
|
||||
from zipline.test_algorithms import TestAlgorithm
|
||||
from zipline.components import TradeSimulationClient
|
||||
from zipline.core.process import ProcessSimulator
|
||||
from zipline.core.monitor import Controller
|
||||
from zipline.core.monitor import Monitor
|
||||
from zipline.finance.trading import SIMULATION_STYLE
|
||||
|
||||
log = logbook.Logger('Lines')
|
||||
@@ -131,7 +131,7 @@ class SimulatedTrading(object):
|
||||
'results_address' : sockets[4],
|
||||
}
|
||||
|
||||
self.con = Controller(
|
||||
self.monitor = Monitor(
|
||||
# pub socket
|
||||
sockets[5],
|
||||
# route socket
|
||||
@@ -163,7 +163,7 @@ class SimulatedTrading(object):
|
||||
#setup transforms
|
||||
self.transforms = {}
|
||||
|
||||
self.sim.register_controller( self.con )
|
||||
self.sim.register_monitor( self.monitor )
|
||||
|
||||
|
||||
@staticmethod
|
||||
@@ -348,15 +348,15 @@ class SimulatedTrading(object):
|
||||
|
||||
return base | transforms | sources
|
||||
|
||||
def setup_controller(self):
|
||||
def setup_monitor(self):
|
||||
"""
|
||||
Prepare the controller to manage the topology specified
|
||||
Prepare the monitor to manage the topology specified
|
||||
by this line.
|
||||
"""
|
||||
self.con.manage(self.topology)
|
||||
self.monitor.manage(self.topology)
|
||||
|
||||
def simulate(self, blocking=True):
|
||||
self.setup_controller()
|
||||
self.setup_monitor()
|
||||
|
||||
self.started = True
|
||||
self.sim_context = self.sim.simulate()
|
||||
|
||||
@@ -1,3 +1,4 @@
|
||||
import multiprocessing
|
||||
import zmq
|
||||
import time
|
||||
import zipline.protocol as zp
|
||||
@@ -6,6 +7,7 @@ import blist
|
||||
from zipline.utils.date_utils import EPOCH
|
||||
from itertools import izip
|
||||
from logbook import FileHandler
|
||||
from zipline.core.monitor import Monitor
|
||||
|
||||
def setup_logger(test, path='/var/log/zipline/zipline.log'):
|
||||
test.log_handler = FileHandler(path)
|
||||
@@ -140,3 +142,33 @@ def assert_single_position(test, zipline):
|
||||
sid,
|
||||
"Portfolio should have one position in " + str(sid)
|
||||
)
|
||||
|
||||
|
||||
def launch_component(component):
|
||||
proc = multiprocessing.Process(target=component.run)
|
||||
proc.start()
|
||||
return proc
|
||||
|
||||
def launch_monitor(monitor):
|
||||
proc = multiprocessing.Process(target=monitor.run)
|
||||
proc.start()
|
||||
return proc
|
||||
|
||||
|
||||
def create_monitor(allocator):
|
||||
sockets = allocator.lease(3)
|
||||
mon = Monitor(
|
||||
# pub socket
|
||||
sockets[0],
|
||||
# route socket
|
||||
sockets[1],
|
||||
# exception socket to match tradesimclient's result
|
||||
# socket, because we want to relay exceptions to the
|
||||
# same listener
|
||||
sockets[2],
|
||||
# this controller is expected to run in a test, so no
|
||||
# need to signal the parent process on success or error.
|
||||
send_sighup=False
|
||||
)
|
||||
|
||||
return mon
|
||||
|
||||
Reference in New Issue
Block a user