Exemplo n.º 1
0
    def start(self):
        """
        Start fork loop.
        """

        # Load configuration here to make sure that everything for it is loaded
        self.cfg = getConfiguration()

        # Must import here because we don't have DB on bootup yet
        from Globals import DB
        # TODO: Fetch adapter with interface
        self.storage_index = FileStorageIndex(DB.storage)

        # SIGCHLD tells us that child process has really died and we can spawn
        # new child
        registerHandler(signal.SIGCHLD, self._waitChildToDieAndScheduleNew)

        # With SIGUSR1 child can tell that it dies by request, not by exception
        # etc.
        registerHandler(signal.SIGUSR1, self._childIsGoingToDie)

        self.loop()
Exemplo n.º 2
0
class ForkLoop(object):

    def __init__(self):

        self.fork = True # Create child on start
        self.active = False
        self.pause = False
        self.killed_child = True
        self.forking = False
        self.exit = False

        self.parent_pid = os.getpid()
        self.child_pid = None

        # Timers
        self.boot_started = None
        self.child_started = None

        self.cfg = None
        self.storage_index = None

    def isChild(self):
        return self.child_pid == 0

    def startBootTimer(self):
        if not self.boot_started:
            self.boot_started = time.time()

    def startChildBooTimer(self):
        self.child_started = time.time()

    def isChildAlive(self):

        if self.isChild():
            return True

        return (self.child_pid is not None
            and os.path.exists("/proc/%i" % self.child_pid))

    def _scheduleFork(self, signum=None, frame=None):
        self.fork = True

    def _childIsGoingToDie(self, signum=None, frame=None):
        self.killed_child = True

    def start(self):
        """
        Start fork loop.
        """

        # Load configuration here to make sure that everything for it is loaded
        self.cfg = getConfiguration()

        # Must import here because we don't have DB on bootup yet
        from Globals import DB
        # TODO: Fetch adapter with interface
        self.storage_index = FileStorageIndex(DB.storage)


        # SIGCHLD tells us that child process has really died and we can spawn
        # new child
        registerHandler(signal.SIGCHLD, self._waitChildToDieAndScheduleNew)

        # With SIGUSR1 child can tell that it dies by request, not by exception
        # etc.
        registerHandler(signal.SIGUSR1, self._childIsGoingToDie)

        self.loop()




    def loop(self):
        """
        Magic happens here
        """

        registerHandler(signal.SIGINT, self._parentExitHandler)
        registerHandler(signal.SIGTERM, self._parentExitHandler)

        self.active = True

        logger.info("Fork loop starting on parent. PID %i" % os.getpid())
        while True:
            self.forking = False

            if self.exit:
                return

            if self.fork:
                self.fork = False

                if self.pause:
                    # Pause mode. No forks now.
                    continue

                if not self.killed_child:
                    errline()
                    errline("Child died on bootup. Pausing fork loop for now. ")
                    errline("Fix possible errors and save edits and we'll try booting again.")
                    errline("Waiting...")

                    # Child died because of unknown reason. Mark it as killed
                    # and go into pause mode.
                    self.killed_child = True
                    self.pause = True
                    continue

                if self.isChildAlive():
                    # Child is still alive for some reason. Lets wait few
                    # rounds for it to die.
                    continue

                self.forking = True
                self.startChildBooTimer()
                self.child_pid = os.fork()
                if self.child_pid == 0:
                    break
                self.killed_child = False

            time.sleep(1)

        logger.setChildLogger()
        logger.info("Forked new child. Installing reloadable products...")

        self._prepareNewChild()

        self.forking = False

        logger.info("Booted up new new child in %s seconds. PID %i" % (
            time.time() - self.child_started,  os.getpid()))

        notify(NewChildIsReady(self))

    def _prepareNewChild(self):
        """
        Prepare newly forked child. Make sure that it can properly read DB
        and install deferred products.
        """

        # Register exit listener. We cannot immediately spawn new child when we
        # get a modified event. Must wait that child has closed database etc.
        atexit.register(self._childExitHandler)

        # Make sure that PID files and locks stay here, because dying child
        # will clear them.
        self.makeLockFile()
        self.makePidFile()

        self.storage_index.restore()

        notify(NewChildForked(self))

        autoinclude.include_deferred()
        fiveconfigure.install_deferred()


    def spawnNewChild(self):
        """
        STEP 1 (parent): New child spawning starts by killing the current
        child.
        """


        if not self.active:
            raise CannotSpawnNewChild("Loop not started yet")

        if self.forking:
            raise CannotSpawnNewChild("Serious forking action is already going on. Cannot fork now.")

        if self.child_pid is None:
            raise CannotSpawnNewChild("No killing yet. Not started child yet")


        self.pause = False

        if not self.killed_child or self.isChild():
            self._killChild()
        else:
            # Ok, we already have sent the SIGINT the child, but asking for new child
            logger.info("Not sending SIGINT because we already killed the child. Just scheduling new fork.")
            self._scheduleFork()

        self.killed_child = True


    def _killChild(self):
        if self.isChild():
            # Signal parent that this is requested kill, not an error situation
            os.kill(self.parent_pid, signal.SIGUSR1)
            # Kill itself
            os.kill(os.getpid(), signal.SIGINT)
        else:
            os.kill(self.child_pid, signal.SIGINT)

    def _parentExitHandler(self, signum=None, frame=None):
        if self.isChild():
            return

        self.exit = True

        if self.isChildAlive():
            logger.info("Parent dying. Killing child first.")
            self._killChild()

    def _childExitHandler(self):
        """
        STEP 2 (child): Child is about to die. Fix DB.
        """

        self.storage_index.save()


    def _waitChildToDieAndScheduleNew(self, signal=None, frame=None):
        """
        STEP 3 (parent): Child told us via SIGCHLD that we can spawn new child
        """

        # Acknowledge dead child
        os.wait()

        # Schedule new
        self._scheduleFork()


    # Modified from Zope2/Startup/__init__.py
    def makePidFile(self):

        if not self.cfg.zserver_read_only_mode:
            # write the pid into the pidfile if possible
            try:
                if os.path.exists(self.cfg.pid_filename):
                    os.unlink(self.cfg.pid_filename)
                f = open(self.cfg.pid_filename, 'w')
                f.write(str(self.parent_pid))
                f.close()
            except IOError:
                pass


    # Modified from Zope2/Startup/__init__.py
    def makeLockFile(self):
        if not self.cfg.zserver_read_only_mode:
            from Zope2.Startup.misc.lock_file import lock_file
            lock_filename = self.cfg.lock_filename
            try:
                if os.path.exists(lock_filename):
                    os.unlink(lock_filename)
                lockfile = open(lock_filename, 'w')
                lock_file(lockfile)
                lockfile.write(str(self.parent_pid))
                lockfile.flush()
            except IOError:
                pass