# Copyright (c) Twisted Matrix Laboratories.
# See LICENSE for details.
"""
Tests for L{twisted.runner.procmon}.
"""
import pickle
from twisted.internet.error import ProcessDone, ProcessExitedAlready, ProcessTerminated
from twisted.internet.task import Clock
from twisted.internet.testing import MemoryReactor
from twisted.logger import globalLogPublisher
from twisted.python.failure import Failure
from twisted.runner.procmon import LoggingProtocol, ProcessMonitor
from twisted.trial import unittest
class DummyProcess:
"""
An incomplete and fake L{IProcessTransport} implementation for testing how
L{ProcessMonitor} behaves when its monitored processes exit.
@ivar _terminationDelay: the delay in seconds after which the DummyProcess
will appear to exit when it receives a TERM signal
"""
pid = 1
proto = None
_terminationDelay = 1
def __init__(
self,
reactor,
executable,
args,
environment,
path,
proto,
uid=None,
gid=None,
usePTY=0,
childFDs=None,
):
self.proto = proto
self._reactor = reactor
self._executable = executable
self._args = args
self._environment = environment
self._path = path
self._uid = uid
self._gid = gid
self._usePTY = usePTY
self._childFDs = childFDs
def signalProcess(self, signalID):
"""
A partial implementation of signalProcess which can only handle TERM and
KILL signals.
- When a TERM signal is given, the dummy process will appear to exit
after L{DummyProcess._terminationDelay} seconds with exit code 0
- When a KILL signal is given, the dummy process will appear to exit
immediately with exit code 1.
@param signalID: The signal name or number to be issued to the process.
@type signalID: C{str}
"""
params = {"TERM": (self._terminationDelay, 0), "KILL": (0, 1)}
if self.pid is None:
raise ProcessExitedAlready()
if signalID in params:
delay, status = params[signalID]
self._signalHandler = self._reactor.callLater(
delay, self.processEnded, status
)
def processEnded(self, status):
"""
Deliver the process ended event to C{self.proto}.
"""
self.pid = None
statusMap = {
0: ProcessDone,
1: ProcessTerminated,
}
self.proto.processEnded(Failure(statusMap[status](status)))
class DummyProcessReactor(MemoryReactor, Clock):
"""
@ivar spawnedProcesses: a list that keeps track of the fake process
instances built by C{spawnProcess}.
@type spawnedProcesses: C{list}
"""
def __init__(self):
MemoryReactor.__init__(self)
Clock.__init__(self)
self.spawnedProcesses = []
def spawnProcess(
self,
processProtocol,
executable,
args=(),
env={},
path=None,
uid=None,
gid=None,
usePTY=0,
childFDs=None,
):
"""
Fake L{reactor.spawnProcess}, that logs all the process
arguments and returns a L{DummyProcess}.
"""
proc = DummyProcess(
self,
executable,
args,
env,
path,
processProtocol,
uid,
gid,
usePTY,
childFDs,
)
processProtocol.makeConnection(proc)
self.spawnedProcesses.append(proc)
return proc
class ProcmonTests(unittest.TestCase):
"""
Tests for L{ProcessMonitor}.
"""
def setUp(self):
"""
Create an L{ProcessMonitor} wrapped around a fake reactor.
"""
self.reactor = DummyProcessReactor()
self.pm = ProcessMonitor(reactor=self.reactor)
self.pm.minRestartDelay = 2
self.pm.maxRestartDelay = 10
self.pm.threshold = 10
def test_reprLooksGood(self):
"""
Repr includes all details
"""
self.pm.addProcess("foo", ["arg1", "arg2"], uid=1, gid=2, env={})
representation = repr(self.pm)
self.assertIn("foo", representation)
self.assertIn("1", representation)
self.assertIn("2", representation)
def test_simpleReprLooksGood(self):
"""
Repr does not include unneeded details.
Values of attributes that just mean "inherit from launching
process" do not appear in the repr of a process.
"""
self.pm.addProcess("foo", ["arg1", "arg2"], env={})
representation = repr(self.pm)
self.assertNotIn("(", representation)
self.assertNotIn(")", representation)
def test_getStateIncludesProcesses(self):
"""
The list of monitored processes must be included in the pickle state.
"""
self.pm.addProcess("foo", ["arg1", "arg2"], uid=1, gid=2, env={})
self.assertEqual(
self.pm.__getstate__()["processes"], {"foo": (["arg1", "arg2"], 1, 2, {})}
)
def test_getStateExcludesReactor(self):
"""
The private L{ProcessMonitor._reactor} instance variable should not be
included in the pickle state.
"""
self.assertNotIn("_reactor", self.pm.__getstate__())
def test_addProcess(self):
"""
L{ProcessMonitor.addProcess} only starts the named program if
L{ProcessMonitor.startService} has been called.
"""
self.pm.addProcess("foo", ["arg1", "arg2"], uid=1, gid=2, env={})
self.assertEqual(self.pm.protocols, {})
self.assertEqual(self.pm.processes, {"foo": (["arg1", "arg2"], 1, 2, {})})
self.pm.startService()
self.reactor.advance(0)
self.assertEqual(list(self.pm.protocols.keys()), ["foo"])
def test_addProcessDuplicateKeyError(self):
"""
L{ProcessMonitor.addProcess} raises a C{KeyError} if a process with the
given name already exists.
"""
self.pm.addProcess("foo", ["arg1", "arg2"], uid=1, gid=2, env={})
self.assertRaises(
KeyError, self.pm.addProcess, "foo", ["arg1", "arg2"], uid=1, gid=2, env={}
)
def test_addProcessEnv(self):
"""
L{ProcessMonitor.addProcess} takes an C{env} parameter that is passed to
L{IReactorProcess.spawnProcess}.
"""
fakeEnv = {"KEY": "value"}
self.pm.startService()
self.pm.addProcess("foo", ["foo"], uid=1, gid=2, env=fakeEnv)
self.reactor.advance(0)
self.assertEqual(self.reactor.spawnedProcesses[0]._environment, fakeEnv)
def test_addProcessCwd(self):
"""
L{ProcessMonitor.addProcess} takes an C{cwd} parameter that is passed
to L{IReactorProcess.spawnProcess}.
"""
self.pm.startService()
self.pm.addProcess("foo", ["foo"], cwd="/mnt/lala")
self.reactor.advance(0)
self.assertEqual(self.reactor.spawnedProcesses[0]._path, "/mnt/lala")
def test_removeProcess(self):
"""
L{ProcessMonitor.removeProcess} removes the process from the public
processes list.
"""
self.pm.startService()
self.pm.addProcess("foo", ["foo"])
self.assertEqual(len(self.pm.processes), 1)
self.pm.removeProcess("foo")
self.assertEqual(len(self.pm.processes), 0)
def test_removeProcessUnknownKeyError(self):
"""
L{ProcessMonitor.removeProcess} raises a C{KeyError} if the given
process name isn't recognised.
"""
self.pm.startService()
self.assertRaises(KeyError, self.pm.removeProcess, "foo")
def test_startProcess(self):
"""
When a process has been started, an instance of L{LoggingProtocol} will
be added to the L{ProcessMonitor.protocols} dict and the start time of
the process will be recorded in the L{ProcessMonitor.timeStarted}
dictionary.
"""
self.pm.addProcess("foo", ["foo"])
self.pm.startProcess("foo")
self.assertIsInstance(self.pm.protocols["foo"], LoggingProtocol)
self.assertIn("foo", self.pm.timeStarted.keys())
def test_startProcessAlreadyStarted(self):
"""
L{ProcessMonitor.startProcess} silently returns if the named process is
already started.
"""
self.pm.addProcess("foo", ["foo"])
self.pm.startProcess("foo")
self.assertIsNone(self.pm.startProcess("foo"))
def test_startProcessUnknownKeyError(self):
"""
L{ProcessMonitor.startProcess} raises a C{KeyError} if the given
process name isn't recognised.
"""
self.assertRaises(KeyError, self.pm.startProcess, "foo")
def test_stopProcessNaturalTermination(self):
"""
L{ProcessMonitor.stopProcess} immediately sends a TERM signal to the
named process.
"""
self.pm.startService()
self.pm.addProcess("foo", ["foo"])
self.assertIn("foo", self.pm.protocols)
# Configure fake process to die 1 second after receiving term signal
timeToDie = self.pm.protocols["foo"].transport._terminationDelay = 1
# Advance the reactor to just before the short lived process threshold
# and leave enough time for the process to die
self.reactor.advance(self.pm.threshold)
# Then signal the process to stop
self.pm.stopProcess("foo")
# Advance the reactor just enough to give the process time to die and
# verify that the process restarts
self.reactor.advance(timeToDie)
# No further time is required to pass here but the reactor must
# iterate due to implementation details. See the comment in
# test_stopProcessForcedKill.
self.reactor.advance(0)
# We expect it to be restarted immediately
self.assertEqual(self.reactor.seconds(), self.pm.timeStarted["foo"])
def test_stopProcessForcedKill(self):
"""
L{ProcessMonitor.stopProcess} kills a process which fails to terminate
naturally within L{ProcessMonitor.killTime} seconds.
"""
self.pm.startService()
self.pm.addProcess("foo", ["foo"])
self.assertIn("foo", self.pm.protocols)
self.reactor.advance(self.pm.threshold)
proc = self.pm.protocols["foo"].transport
# Arrange for the fake process to live longer than the killTime
proc._terminationDelay = self.pm.killTime + 1
self.pm.stopProcess("foo")
# If process doesn't die before the killTime, procmon should
# terminate it
self.reactor.advance(self.pm.killTime - 1)
self.assertEqual(0.0, self.pm.timeStarted["foo"])
self.reactor.advance(1)
# We expect it to be immediately restarted. While no actual time
# should need to pass for this to happen, the reactor will need to
# iterate a couple times because the implementation uses `callLater`
# (twice!) to schedule the restart and no delayed call can run sooner
# than the reactor iteration after it is scheduled.
self.reactor.pump([0, 0])
self.assertEqual(self.reactor.seconds(), self.pm.timeStarted["foo"])
def test_stopProcessUnknownKeyError(self):
"""
L{ProcessMonitor.stopProcess} raises a C{KeyError} if the given process
name isn't recognised.
"""
self.assertRaises(KeyError, self.pm.stopProcess, "foo")
def test_stopProcessAlreadyStopped(self):
"""
L{ProcessMonitor.stopProcess} silently returns if the named process
is already stopped. eg Process has crashed and a restart has been
rescheduled, but in the meantime, the service is stopped.
"""
self.pm.addProcess("foo", ["foo"])
self.assertIsNone(self.pm.stopProcess("foo"))
def test_outputReceivedCompleteLine(self):
"""
Getting a complete output line on stdout generates a log message.
"""
events = []
self.addCleanup(globalLogPublisher.removeObserver, events.append)
globalLogPublisher.addObserver(events.append)
self.pm.addProcess("foo", ["foo"])
# Schedule the process to start
self.pm.startService()
# Advance the reactor to start the process
self.reactor.advance(0)
self.assertIn("foo", self.pm.protocols)
# Long time passes
self.reactor.advance(self.pm.threshold)
# Process greets
self.pm.protocols["foo"].outReceived(b"hello world!\n")
self.assertEquals(len(events), 1)
namespace = events[0]["log_namespace"]
stream = events[0]["stream"]
tag = events[0]["tag"]
line = events[0]["line"]
self.assertEquals(namespace, "twisted.runner.procmon.ProcessMonitor")
self.assertEquals(stream, "stdout")
self.assertEquals(tag, "foo")
self.assertEquals(line, "hello world!")
def test_ouputReceivedCompleteErrLine(self):
"""
Getting a complete output line on stderr generates a log message.
"""
events = []
self.addCleanup(globalLogPublisher.removeObserver, events.append)
globalLogPublisher.addObserver(events.append)
self.pm.addProcess("foo", ["foo"])
# Schedule the process to start
self.pm.startService()
# Advance the reactor to start the process
self.reactor.advance(0)
self.assertIn("foo", self.pm.protocols)
# Long time passes
self.reactor.advance(self.pm.threshold)
# Process greets
self.pm.protocols["foo"].errReceived(b"hello world!\n")
self.assertEquals(len(events), 1)
namespace = events[0]["log_namespace"]
stream = events[0]["stream"]
tag = events[0]["tag"]
line = events[0]["line"]
self.assertEquals(namespace, "twisted.runner.procmon.ProcessMonitor")
self.assertEquals(stream, "stderr")
self.assertEquals(tag, "foo")
self.assertEquals(line, "hello world!")
def test_outputReceivedCompleteLineInvalidUTF8(self):
"""
Getting invalid UTF-8 results in the repr of the raw message
"""
events = []
self.addCleanup(globalLogPublisher.removeObserver, events.append)
globalLogPublisher.addObserver(events.append)
self.pm.addProcess("foo", ["foo"])
# Schedule the process to start
self.pm.startService()
# Advance the reactor to start the process
self.reactor.advance(0)
self.assertIn("foo", self.pm.protocols)
# Long time passes
self.reactor.advance(self.pm.threshold)
# Process greets
self.pm.protocols["foo"].outReceived(b"\xffhello world!\n")
self.assertEquals(len(events), 1)
message = events[0]
namespace = message["log_namespace"]
stream = message["stream"]
tag = message["tag"]
output = message["line"]
self.assertEquals(namespace, "twisted.runner.procmon.ProcessMonitor")
self.assertEquals(stream, "stdout")
self.assertEquals(tag, "foo")
self.assertEquals(output, repr(b"\xffhello world!"))
def test_outputReceivedPartialLine(self):
"""
Getting partial line results in no events until process end
"""
events = []
self.addCleanup(globalLogPublisher.removeObserver, events.append)
globalLogPublisher.addObserver(events.append)
self.pm.addProcess("foo", ["foo"])
# Schedule the process to start
self.pm.startService()
# Advance the reactor to start the process
self.reactor.advance(0)
self.assertIn("foo", self.pm.protocols)
# Long time passes
self.reactor.advance(self.pm.threshold)
# Process greets
self.pm.protocols["foo"].outReceived(b"hello world!")
self.assertEquals(len(events), 0)
self.pm.protocols["foo"].processEnded(Failure(ProcessDone(0)))
self.assertEquals(len(events), 1)
namespace = events[0]["log_namespace"]
stream = events[0]["stream"]
tag = events[0]["tag"]
line = events[0]["line"]
self.assertEquals(namespace, "twisted.runner.procmon.ProcessMonitor")
self.assertEquals(stream, "stdout")
self.assertEquals(tag, "foo")
self.assertEquals(line, "hello world!")
def test_connectionLostLongLivedProcess(self):
"""
L{ProcessMonitor.connectionLost} should immediately restart a process
if it has been running longer than L{ProcessMonitor.threshold} seconds.
"""
self.pm.addProcess("foo", ["foo"])
# Schedule the process to start
self.pm.startService()
# advance the reactor to start the process
self.reactor.advance(0)
self.assertIn("foo", self.pm.protocols)
# Long time passes
self.reactor.advance(self.pm.threshold)
# Process dies after threshold
self.pm.protocols["foo"].processEnded(Failure(ProcessDone(0)))
self.assertNotIn("foo", self.pm.protocols)
# Process should be restarted immediately
self.reactor.advance(0)
self.assertIn("foo", self.pm.protocols)
def test_connectionLostMurderCancel(self):
"""
L{ProcessMonitor.connectionLost} cancels a scheduled process killer and
deletes the DelayedCall from the L{ProcessMonitor.murder} list.
"""
self.pm.addProcess("foo", ["foo"])
# Schedule the process to start
self.pm.startService()
# Advance 1s to start the process then ask ProcMon to stop it
self.reactor.advance(1)
self.pm.stopProcess("foo")
# A process killer has been scheduled, delayedCall is active
self.assertIn("foo", self.pm.murder)
delayedCall = self.pm.murder["foo"]
self.assertTrue(delayedCall.active())
# Advance to the point at which the dummy process exits
self.reactor.advance(self.pm.protocols["foo"].transport._terminationDelay)
# Now the delayedCall has been cancelled and deleted
self.assertFalse(delayedCall.active())
self.assertNotIn("foo", self.pm.murder)
def test_connectionLostProtocolDeletion(self):
"""
L{ProcessMonitor.connectionLost} removes the corresponding
ProcessProtocol instance from the L{ProcessMonitor.protocols} list.
"""
self.pm.startService()
self.pm.addProcess("foo", ["foo"])
self.assertIn("foo", self.pm.protocols)
self.pm.protocols["foo"].transport.signalProcess("KILL")
self.reactor.advance(self.pm.protocols["foo"].transport._terminationDelay)
self.assertNotIn("foo", self.pm.protocols)
def test_connectionLostMinMaxRestartDelay(self):
"""
L{ProcessMonitor.connectionLost} will wait at least minRestartDelay s
and at most maxRestartDelay s
"""
self.pm.minRestartDelay = 2
self.pm.maxRestartDelay = 3
self.pm.startService()
self.pm.addProcess("foo", ["foo"])
self.assertEqual(self.pm.delay["foo"], self.pm.minRestartDelay)
self.reactor.advance(self.pm.threshold - 1)
self.pm.protocols["foo"].processEnded(Failure(ProcessDone(0)))
self.assertEqual(self.pm.delay["foo"], self.pm.maxRestartDelay)
def test_connectionLostBackoffDelayDoubles(self):
"""
L{ProcessMonitor.connectionLost} doubles the restart delay each time
the process dies too quickly.
"""
self.pm.startService()
self.pm.addProcess("foo", ["foo"])
self.reactor.advance(self.pm.threshold - 1) # 9s
self.assertIn("foo", self.pm.protocols)
self.assertEqual(self.pm.delay["foo"], self.pm.minRestartDelay)
# process dies within the threshold and should not restart immediately
self.pm.protocols["foo"].processEnded(Failure(ProcessDone(0)))
self.assertEqual(self.pm.delay["foo"], self.pm.minRestartDelay * 2)
def test_startService(self):
"""
L{ProcessMonitor.startService} starts all monitored processes.
"""
self.pm.addProcess("foo", ["foo"])
# Schedule the process to start
self.pm.startService()
# advance the reactor to start the process
self.reactor.advance(0)
self.assertIn("foo", self.pm.protocols)
def test_stopService(self):
"""
L{ProcessMonitor.stopService} should stop all monitored processes.
"""
self.pm.addProcess("foo", ["foo"])
self.pm.addProcess("bar", ["bar"])
# Schedule the process to start
self.pm.startService()
# advance the reactor to start the processes
self.reactor.advance(self.pm.threshold)
self.assertIn("foo", self.pm.protocols)
self.assertIn("bar", self.pm.protocols)
self.reactor.advance(1)
self.pm.stopService()
# Advance to beyond the killTime - all monitored processes
# should have exited
self.reactor.advance(self.pm.killTime + 1)
# The processes shouldn't be restarted
self.assertEqual({}, self.pm.protocols)
def test_restartAllRestartsOneProcess(self):
"""
L{ProcessMonitor.restartAll} succeeds when there is one process.
"""
self.pm.addProcess("foo", ["foo"])
self.pm.startService()
self.reactor.advance(1)
self.pm.restartAll()
# Just enough time for the process to die,
# not enough time to start a new one.
self.reactor.advance(1)
processes = list(self.reactor.spawnedProcesses)
myProcess = processes.pop()
self.assertEquals(processes, [])
self.assertIsNone(myProcess.pid)
def test_stopServiceCancelRestarts(self):
"""
L{ProcessMonitor.stopService} should cancel any scheduled process
restarts.
"""
self.pm.addProcess("foo", ["foo"])
# Schedule the process to start
self.pm.startService()
# advance the reactor to start the processes
self.reactor.advance(self.pm.threshold)
self.assertIn("foo", self.pm.protocols)
self.reactor.advance(1)
# Kill the process early
self.pm.protocols["foo"].processEnded(Failure(ProcessDone(0)))
self.assertTrue(self.pm.restart["foo"].active())
self.pm.stopService()
# Scheduled restart should have been cancelled
self.assertFalse(self.pm.restart["foo"].active())
def test_stopServiceCleanupScheduledRestarts(self):
"""
L{ProcessMonitor.stopService} should cancel all scheduled process
restarts.
"""
self.pm.threshold = 5
self.pm.minRestartDelay = 5
# Start service and add a process (started immediately)
self.pm.startService()
self.pm.addProcess("foo", ["foo"])
# Stop the process after 1s
self.reactor.advance(1)
self.pm.stopProcess("foo")
# Wait 1s for it to exit it will be scheduled to restart 5s later
self.reactor.advance(1)
# Meanwhile stop the service
self.pm.stopService()
# Advance to beyond the process restart time
self.reactor.advance(6)
# The process shouldn't have restarted because stopService has cancelled
# all pending process restarts.
self.assertEqual(self.pm.protocols, {})
class DeprecationTests(unittest.SynchronousTestCase):
"""
Tests that check functionality that should be deprecated is deprecated.
"""
def setUp(self):
"""
Create reactor and process monitor.
"""
self.reactor = DummyProcessReactor()
self.pm = ProcessMonitor(reactor=self.reactor)
def test_toTuple(self):
"""
_Process.toTuple is deprecated.
When getting the deprecated processes property, the actual
data (kept in the class _Process) is converted to a tuple --
which produces a DeprecationWarning per process so converted.
"""
self.pm.addProcess("foo", ["foo"])
myprocesses = self.pm.processes
self.assertEquals(len(myprocesses), 1)
warnings = self.flushWarnings()
foundToTuple = False
for warning in warnings:
self.assertIs(warning["category"], DeprecationWarning)
if "toTuple" in warning["message"]:
foundToTuple = True
self.assertTrue(foundToTuple, f"no tuple deprecation found:{repr(warnings)}")
def test_processes(self):
"""
Accessing L{ProcessMonitor.processes} results in deprecation warning
Even when there are no processes, and thus no process is converted
to a tuple, accessing the L{ProcessMonitor.processes} property
should generate its own DeprecationWarning.
"""
myProcesses = self.pm.processes
self.assertEquals(myProcesses, {})
warnings = self.flushWarnings()
first = warnings.pop(0)
self.assertIs(first["category"], DeprecationWarning)
self.assertEquals(warnings, [])
def test_getstate(self):
"""
Pickling an L{ProcessMonitor} results in deprecation warnings
"""
pickle.dumps(self.pm)
warnings = self.flushWarnings()
for warning in warnings:
self.assertIs(warning["category"], DeprecationWarning)
|