On Fri, Oct 09, 2009 at 09:58:14AM +0100, Daniel P. Berrange wrote:
> On Fri, Oct 09, 2009 at 10:19:08AM +0200, Dan Kenigsberg wrote:
> > On Wed, Oct 07, 2009 at 12:59:39PM +0100, Daniel P. Berrange wrote:
> > > On Tue, Oct 06, 2009 at 07:04:29PM +0200, Dan Kenigsberg wrote:
> > > > Would someone help me have a shrink-wrapped solution for obtaining libvirt
> > > > events in python?
> > >
> > > I decided to re-write the demo program so that is shows a serious
> > > production kwalitee event loop implementation that can be used in
> > > real world applications. I think you'll find this much nicer :-)
> >
> > It sure look nicer (though I still don't get the hows and whys).
> > However, it seems that you have an issue with python's time.time()
> > measured in seconds, not milliseconds.
>
> Here's a version with more docs & the time in milliseconds
>
> #!/usr/bin/python -u
> #
> #
> #
> #################################################################################
> # Start off by implementing a general purpose event loop for anyones use
> #################################################################################
>
> import sys
> import getopt
> import os
> import libvirt
> import select
> import errno
> import time
> import threading
>
> #
> # This general purpose event loop will support waiting for file handle
> # I/O and errors events, as well as scheduling repeatable timers with
> # a fixed interval.
> #
> # It is a pure python implementation based around the poll() API
> #
> class virEventLoopPure:
> # This class contains the data we need to track for a
> # single file handle
> class virEventLoopPureHandle:
> def __init__(self, handle, fd, events, cb, opaque):
> self.handle = handle
> self.fd = fd
> self.events = events
> self.cb = cb
> self.opaque = opaque
>
> def get_id(self):
> return self.handle
>
> def get_fd(self):
> return self.fd
>
> def get_events(self):
> return self.events
>
> def set_events(self, events):
> self.events = events
>
> def dispatch(self, events):
> self.cb(self.handle,
> self.fd,
> events,
> self.opaque[0],
> self.opaque[1])
>
> # This class contains the data we need to track for a
> # single periodic timer
> class virEventLoopPureTimer:
> def __init__(self, timer, interval, cb, opaque):
> self.timer = timer
> self.interval = interval
> self.cb = cb
> self.opaque = opaque
> self.lastfired = 0
>
> def get_id(self):
> return self.timer
>
> def get_interval(self):
> return self.interval
>
> def set_interval(self, interval):
> self.interval = interval
>
> def get_last_fired(self):
> return self.lastfired
>
> def set_last_fired(self, now):
> self.lastfired = now
>
> def dispatch(self):
> self.cb(self.timer,
> self.opaque[0],
> self.opaque[1])
>
>
> def __init__(self, debug=False):
> self.debugOn = debug
> self.poll = select.poll()
> self.pipetrick = os.pipe()
> self.nextHandleID = 1
> self.nextTimerID = 1
> self.handles = []
> self.timers = []
> self.quit = False
>
> # The event loop can be used from multiple threads at once.
> # Specifically while the main thread is sleeping in poll()
> # waiting for events to occur, another thread may come along
> # and add/update/remove a file handle, or timer. When this
> # happens we need to interrupt the poll() sleep in the other
> # thread, so that it'll see the file handle / timer changes.
> #
> # Using OS level signals for this is very unreliable and
> # hard to implement correctly. Thus we use the real classic
> # "self pipe" trick. A anonymous pipe, with one end registered
> # with the event loop for input events. When we need to force
> # the main thread out of a poll() sleep, we simple write a
> # single byte of data to the other end of the pipe.
> self.debug("Self pipe watch %d write %d" %(self.pipetrick[0], self.pipetrick[1]))
> self.poll.register(self.pipetrick[0], select.POLLIN)
>
> def debug(self, msg):
> if self.debugOn:
> print msg
>
>
> # Calculate when the next timeout is due to occurr, returning
> # the absolute timestamp for the next timeout, or 0 if there is
> # no timeout due
> def next_timeout(self):
> next = 0
> for t in self.timers:
> last = t.get_last_fired()
> interval = t.get_interval()
> if interval < 0:
> continue
> if next == 0 or (last + interval) < next:
> next = last + interval
>
> return next
>
> # Lookup a virEventLoopPureHandle object based on file descriptor
> def get_handle_by_fd(self, fd):
> for h in self.handles:
> if h.get_fd() == fd:
> return h
> return None
>
> # Lookup a virEventLoopPureHandle object based on its event loop ID
> def get_handle_by_id(self, handleID):
> for h in self.handles:
> if h.get_id() == handleID:
> return h
> return None
>
>
> # This is the heart of the event loop, performing one single
> # iteration. It asks when the next timeout is due, and then
> # calcuates the maximum amount of time it is able to sleep
> # for in poll() pending file handle events.
> #
> # It then goes into the poll() sleep.
> #
> # When poll() returns, there will zero or more file handle
> # events which need to be dispatched to registered callbacks
> # It may also be time to fire some periodic timers.
> #
> # Due to the coarse granularity of schedular timeslices, if
> # we ask for a sleep of 500ms in order to satisfy a timer, we
> # may return upto 1 schedular timeslice early. So even though
> # our sleep timeout was reached, the registered timer may not
> # technically be at its expiry point. This leads to us going
> # back around the loop with a crazy 5ms sleep. So when checking
> # if timeouts are due, we allow a margin of 20ms, to avoid
> # these pointless repeated tiny sleeps.
> def run_once(self):
> sleep = -1
> next = self.next_timeout()
> self.debug("Next timeout due at %d" % next)
> if next > 0:
> now = int(time.time() * 1000)
> if now >= next:
> sleep = 0
> else:
> sleep = next - now
>
> self.debug("Poll with a sleep of %d" % sleep)
> events = self.poll.poll(sleep)
Thanks! but there's still a units problem here (should be sleep/1000. )
| poll(...)
| poll([timeout=-1[, maxevents=-1]]) -> [(fd, events), (...)]
|
| Wait for events on the epoll file descriptor for a maximum time of timeout
| in seconds (as float). -1 makes poll wait indefinitely.
| Up to maxevents are returned to the caller.
>
> # Dispatch any file handle events that occurred
> for (fd, revents) in events:
> # See if the events was from the self-pipe
> # telling us to wakup. if so, then discard
> # the data just continue
> if fd == self.pipetrick[0]:
> data = os.read(fd, 1)
> continue
>
> h = self.get_handle_by_fd(fd)
> if h:
> self.debug("Dispatch fd %d handle %d events %d" % (fd, h.get_id(), revents))
> h.dispatch(self.events_from_poll(revents))
>
> now = int(time.time() * 1000)
> for t in self.timers:
> interval = t.get_interval()
> if interval < 0:
> continue
>
> want = t.get_last_fired() + interval
> # Deduct 20ms, since schedular timeslice
> # means we could be ever so slightly early
> if now >= (want-20):
> self.debug("Dispatch timer %d now %s want %s" % (t.get_id(), str(now), str(want)))
> t.set_last_fired(now)
> t.dispatch()
>
>
> # Actually the event loop forever
> def run_loop(self):
> self.quit = False
> while not self.quit:
> self.run_once()
>
> def interrupt(self):
> os.write(self.pipetrick[1], 'c')
>
>
> # Registers a new file handle 'fd', monitoring for 'events' (libvirt
> # event constants), firing the callback cb() when an event occurs.
> # Returns a unique integer identier for this handle, that should be
> # used to later update/remove it
> def add_handle(self, fd, events, cb, opaque):
> handleID = self.nextHandleID + 1
> self.nextHandleID = self.nextHandleID + 1
>
> h = self.virEventLoopPureHandle(handleID, fd, events, cb, opaque)
> self.handles.append(h)
>
> self.poll.register(fd, self.events_to_poll(events))
> self.interrupt()
>
> self.debug("Add handle %d fd %d events %d" % (handleID, fd, events))
>
> return handleID
>
> # Registers a new timer with periodic expiry at 'interval' ms,
> # firing cb() each time the timer expires. If 'interval' is -1,
> # then the timer is registered, but not enabled
> # Returns a unique integer identier for this handle, that should be
> # used to later update/remove it
> def add_timer(self, interval, cb, opaque):
> timerID = self.nextTimerID + 1
> self.nextTimerID = self.nextTimerID + 1
>
> h = self.virEventLoopPureTimer(timerID, interval, cb, opaque)
> self.timers.append(h)
> self.interrupt()
>
> self.debug("Add timer %d interval %d" % (timerID, interval))
>
> return timerID
>
> # Change the set of events to be monitored on the file handle
> def update_handle(self, handleID, events):
> h = self.get_handle_by_id(handleID)
> if h:
> h.set_events(events)
> self.poll.unregister(h.get_fd())
> self.poll.register(h.get_fd(), self.events_to_poll(events))
> self.interrupt()
>
> self.debug("Update handle %d fd %d events %d" % (handleID, h.get_fd(), events))
>
> # Change the periodic frequency of the timer
> def update_timer(self, timerID, interval):
> for h in self.timers:
> if h.get_id() == timerID:
> h.set_interval(interval);
> self.interrupt()
>
> self.debug("Update timer %d interval %d" % (timerID, interval))
> break
>
> # Stop monitoring for events on the file handle
> def remove_handle(self, handleID):
> handles = []
> for h in self.handles:
> if h.get_id() == handleID:
> self.poll.unregister(h.get_fd())
> self.debug("Remove handle %d fd %d" % (handleID, h.get_fd()))
> else:
> handles.append(h)
> self.handles = handles
> self.interrupt()
>
> # Stop firing the periodic timer
> def remove_timer(self, timerID):
> timers = []
> for h in self.timers:
> if h.get_id() != timerID:
> timers.append(h)
> self.debug("Remove timer %d" % timerID)
> self.timers = timers
> self.interrupt()
>
> # Convert from libvirt event constants, to poll() events constants
> def events_to_poll(self, events):
> ret = 0
> if events & libvirt.VIR_EVENT_HANDLE_READABLE:
> ret |= select.POLLIN
> if events & libvirt.VIR_EVENT_HANDLE_WRITABLE:
> ret |= select.POLLOUT
> if events & libvirt.VIR_EVENT_HANDLE_ERROR:
> ret |= select.POLLERR;
> if events & libvirt.VIR_EVENT_HANDLE_HANGUP:
> ret |= select.POLLHUP;
> return ret
>
> # Convert from poll() event constants, to libvirt events constants
> def events_from_poll(self, events):
> ret = 0;
> if events & select.POLLIN:
> ret |= libvirt.VIR_EVENT_HANDLE_READABLE;
> if events & select.POLLOUT:
> ret |= libvirt.VIR_EVENT_HANDLE_WRITABLE;
> if events & select.POLLNVAL:
> ret |= libvirt.VIR_EVENT_HANDLE_ERROR;
> if events & select.POLLERR:
> ret |= libvirt.VIR_EVENT_HANDLE_ERROR;
> if events & select.POLLHUP:
> ret |= libvirt.VIR_EVENT_HANDLE_HANGUP;
> return ret;
>
>
> ###########################################################################
> # Now glue an instance of the general event loop into libvirt's event loop
> ###########################################################################
>
> # This single global instance of the event loop wil be used for
> # monitoring libvirt events
> eventLoop = virEventLoopPure(debug=False)
>
> # This keeps track of what thread is running the event loop,
> # (if it is run in a background thread)
> eventLoopThread = None
>
>
> # These next set of 6 methods are the glue between the official
> # libvirt events API, and our particular impl of the event loop
> #
> # There is no reason why the 'virEventLoopPure' has to be used.
> # An application could easily may these 6 glue methods hook into
> # another event loop such as GLib's, or something like the python
> # Twisted event framework.
>
> def virEventAddHandleImpl(fd, events, cb, opaque):
> global eventLoop
> return eventLoop.add_handle(fd, events, cb, opaque)
>
> def virEventUpdateHandleImpl(handleID, events):
> global eventLoop
> return eventLoop.update_handle(handleID, events)
>
> def virEventRemoveHandleImpl(handleID):
> global eventLoop
> return eventLoop.remove_handle(handleID)
>
> def virEventAddTimerImpl(interval, cb, opaque):
> global eventLoop
> return eventLoop.add_timer(interval, cb, opaque)
>
> def virEventUpdateTimerImpl(timerID, interval):
> global eventLoop
> return eventLoop.update_timer(timerID, interval)
>
> def virEventRemoveTimerImpl(timerID):
> global eventLoop
> return eventLoop.remove_timer(timerID)
>
> # This tells libvirt what event loop implementation it
> # should use
> def virEventLoopPureRegister():
> libvirt.virEventRegisterImpl(virEventAddHandleImpl,
> virEventUpdateHandleImpl,
> virEventRemoveHandleImpl,
> virEventAddTimerImpl,
> virEventUpdateTimerImpl,
> virEventRemoveTimerImpl)
>
> # Directly run the event loop in the current thread
> def virEventLoopPureRun():
> global eventLoop
> eventLoop.run_loop()
>
> # Spawn a background thread to run the event loop
> def virEventLoopPureStart():
> global eventLoopThread
> virEventLoopPureRegister()
> eventLoopThread = threading.Thread(target=virEventLoopPureRun, name="libvirtEventLoop")
> eventLoopThread.setDaemon(True)
> eventLoopThread.start()
>
>
> ##########################################################################
> # Everything that now follows is a simple demo of domain lifecycle events
> ##########################################################################
> def eventToString(event):
> eventStrings = ( "Added",
> "Removed",
> "Started",
> "Suspended",
> "Resumed",
> "Stopped",
> "Saved",
> "Restored" );
> return eventStrings[event];
>
> def myDomainEventCallback1 (conn, dom, event, detail, opaque):
> print "myDomainEventCallback1 EVENT: Domain %s(%s) %s %d" % (dom.name(), dom.ID(), eventToString(event), detail)
>
> def myDomainEventCallback2 (conn, dom, event, detail, opaque):
> print "myDomainEventCallback2 EVENT: Domain %s(%s) %s %d" % (dom.name(), dom.ID(), eventToString(event), detail)
>
> def usage():
> print "usage: "+os.path.basename(sys.argv[0])+" [uri]"
> print " uri will default to qemu:///system"
>
> def main():
> try:
> opts, args = getopt.getopt(sys.argv[1:], "h", ["help"] )
> except getopt.GetoptError, err:
> # print help information and exit:
> print str(err) # will print something like "option -a not recognized"
> usage()
> sys.exit(2)
> for o, a in opts:
> if o in ("-h", "--help"):
> usage()
> sys.exit()
>
> if len(sys.argv) > 1:
> uri = sys.argv[1]
> else:
> uri = "qemu:///system"
>
> print "Using uri:" + uri
>
> # Run a background thread with the event loop
> virEventLoopPureStart()
>
> vc = libvirt.open(uri)
>
> # Close connection on exit (to test cleanup paths)
> old_exitfunc = getattr(sys, 'exitfunc', None)
> def exit():
> print "Closing " + str(vc)
> vc.close()
> if (old_exitfunc): old_exitfunc()
> sys.exitfunc = exit
>
> #Add 2 callbacks to prove this works with more than just one
> vc.domainEventRegister(myDomainEventCallback1,None)
> vc.domainEventRegister(myDomainEventCallback2,None)
>
> # The rest of your app would go here normally, but for sake
> # of demo we'll just go to sleep. The other option is to
> # run the event loop in your main thread if your app is
> # totally event based.
> while 1:
> time.sleep(1)
>
>
> if __name__ == "__main__":
> main()
>
> --
> |: Red Hat, Engineering, London -o- http://people.redhat.com/berrange/ :|
> |: http://libvirt.org -o- http://virt-manager.org -o- http://ovirt.org :|
> |: http://autobuild.org -o- http://search.cpan.org/~danberr/ :|
> |: GnuPG: 7D3B9505 -o- F3C9 553F A1DA 4AC2 5648 23C1 B3DF F742 7D3B 9505 :|
--
Libvir-list mailing list
Libvir-list@xxxxxxxxxx
https://www.redhat.com/mailman/listinfo/libvir-list