zloop (3) - Linux Manuals

zloop: event-driven reactor


zloop - event-driven reactor


// Callback function for reactor socket activity
typedef int (zloop_reader_fn) (
    zloop_t *loop, zsock_t *reader, void *arg);

// Callback function for reactor events (low-level)
typedef int (zloop_fn) (
    zloop_t *loop, zmq_pollitem_t *item, void *arg);

// Callback for reactor timer events
typedef int (zloop_timer_fn) (
    zloop_t *loop, int timer_id, void *arg);

//  Create a new zloop reactor
CZMQ_EXPORT zloop_t *
    zloop_new ();

//  Destroy a reactor
    zloop_destroy (zloop_t **self_p);

//  Register socket reader with the reactor. When the reader has messages,
//  the reactor will call the handler, passing the arg. Returns 0 if OK, -1
//  if there was an error. If you register the same socket more than once,
//  each instance will invoke its corresponding handler.
    zloop_reader (zloop_t *self, zsock_t *sock, zloop_reader_fn handler, void *arg);

//  Cancel a socket reader from the reactor. If multiple readers exist for
//  same socket, cancels ALL of them.
    zloop_reader_end (zloop_t *self, zsock_t *sock);

//  Configure a registered reader to ignore errors. If you do not set this,
//  then readers that have errors are removed from the reactor silently.
    zloop_reader_set_tolerant (zloop_t *self, zsock_t *sock);

//  Register low-level libzmq pollitem with the reactor. When the pollitem
//  is ready, will call the handler, passing the arg. Returns 0 if OK, -1
//  if there was an error. If you register the pollitem more than once, each
//  instance will invoke its corresponding handler. A pollitem with
//  socket=NULL and fd=0 means 'poll on FD zero'.
    zloop_poller (zloop_t *self, zmq_pollitem_t *item, zloop_fn handler, void *arg);

//  Cancel a pollitem from the reactor, specified by socket or FD. If both
//  are specified, uses only socket. If multiple poll items exist for same
//  socket/FD, cancels ALL of them.
    zloop_poller_end (zloop_t *self, zmq_pollitem_t *item);

//  Configure a registered poller to ignore errors. If you do not set this,
//  then poller that have errors are removed from the reactor silently.
    zloop_poller_set_tolerant (zloop_t *self, zmq_pollitem_t *item);

//  Register a timer that expires after some delay and repeats some number of
//  times. At each expiry, will call the handler, passing the arg. To run a
//  timer forever, use 0 times. Returns a timer_id that is used to cancel the
//  timer in the future. Returns -1 if there was an error.
    zloop_timer (zloop_t *self, size_t delay, size_t times, zloop_timer_fn handler, void *arg);

//  Cancel a specific timer identified by a specific timer_id (as returned by
//  zloop_timer).
    zloop_timer_end (zloop_t *self, int timer_id);

//  Register a ticket timer. Ticket timers are very fast in the case where
//  you use a lot of timers (thousands), and frequently remove and add them.
//  The main use case is expiry timers for servers that handle many clients,
//  and which reset the expiry timer for each message received from a client.
//  Whereas normal timers perform poorly as the number of clients grows, the
//  cost of ticket timers is constant, no matter the number of clients. You
//  must set the ticket delay using zloop_set_ticket_delay before creating a
//  ticket. Returns a handle to the timer that you should use in
//  zloop_ticket_reset and zloop_ticket_delete.
    zloop_ticket (zloop_t *self, zloop_timer_fn handler, void *arg);

//  Reset a ticket timer, which moves it to the end of the ticket list and
//  resets its execution time. This is a very fast operation.
    zloop_ticket_reset (zloop_t *self, void *handle);

//  Delete a ticket timer. We do not actually delete the ticket here, as
//  other code may still refer to the ticket. We mark as deleted, and remove
//  later and safely.
    zloop_ticket_delete (zloop_t *self, void *handle);

//  Set the ticket delay, which applies to all tickets. If you lower the
//  delay and there are already tickets created, the results are undefined.
    zloop_set_ticket_delay (zloop_t *self, size_t ticket_delay);

//  Set hard limit on number of timers allowed. Setting more than a small
//  number of timers (10-100) can have a dramatic impact on the performance
//  of the reactor. For high-volume cases, use ticket timers. If the hard
//  limit is reached, the reactor stops creating new timers and logs an
//  error.
    zloop_set_max_timers (zloop_t *self, size_t max_timers);

//  Set verbose tracing of reactor on/off
    zloop_set_verbose (zloop_t *self, bool verbose);

//  Start the reactor. Takes control of the thread and returns when the 0MQ
//  context is terminated or the process is interrupted, or any event handler
//  returns -1. Event handlers may register new sockets and timers, and
//  cancel sockets. Returns 0 if interrupted, -1 if cancelled by a handler.
    zloop_start (zloop_t *self);

//  Ignore zsys_interrupted flag in this loop. By default, a zloop_start will
//  exit as soon as it detects zsys_interrupted is set to something other than
//  zero. Calling zloop_ignore_interrupts will supress this behavior.
    zloop_ignore_interrupts (zloop_t *self);

//  Self test of this class
    zloop_test (bool verbose);


The zloop class provides an event-driven reactor pattern. The reactor handles zmq_pollitem_t items (pollers or writers, sockets or fds), and once-off or repeated timers. Its resolution is 1 msec. It uses a tickless timer to reduce CPU interrupts in inactive processes.

Please add @discuss section in ../src/zloop.c.


From zloop_test method.

//  Create two PAIR sockets and connect over inproc
zsock_t *output = zsock_new (ZMQ_PAIR);
assert (output);
zsock_bind (output, "inproc://zloop.test");

zsock_t *input = zsock_new (ZMQ_PAIR);
assert (input);
zsock_connect (input, "inproc://zloop.test");

zloop_t *loop = zloop_new ();
assert (loop);
zloop_set_verbose (loop, verbose);

//  Create a timer that will be cancelled
int timer_id = zloop_timer (loop, 1000, 1, s_timer_event, NULL);
zloop_timer (loop, 5, 1, s_cancel_timer_event, &timer_id);

//  After 20 msecs, send a ping message to output3
zloop_timer (loop, 20, 1, s_timer_event, output);

//  Set up some tickets that will never expire
zloop_set_ticket_delay (loop, 10000);
void *ticket1 = zloop_ticket (loop, s_timer_event, NULL);
void *ticket2 = zloop_ticket (loop, s_timer_event, NULL);
void *ticket3 = zloop_ticket (loop, s_timer_event, NULL);

//  When we get the ping message, end the reactor
rc = zloop_reader (loop, input, s_socket_event, NULL);
assert (rc == 0);
zloop_reader_set_tolerant (loop, input);
zloop_start (loop);

zloop_ticket_delete (loop, ticket1);
zloop_ticket_delete (loop, ticket2);
zloop_ticket_delete (loop, ticket3);

//  Check whether loop properly ignores zsys_interrupted flag
//  when asked to
zloop_destroy (&loop);
loop = zloop_new ();

bool timer_event_called = false;
zloop_timer (loop, 1, 1, s_timer_event3, &timer_event_called);

zsys_interrupted = 1;
zloop_start (loop);
//  zloop returns immediately without giving any handler a chance to run
assert (!timer_event_called);

zloop_ignore_interrupts (loop);
zloop_start (loop);
//  zloop runs the handler which will terminate the loop
assert (timer_event_called);
zsys_interrupted = 0;

//  cleanup
zloop_destroy (&loop);
assert (loop == NULL);

zsock_destroy (&input);
zsock_destroy (&output);


The czmq manual was written by the authors in the AUTHORS file.


Main web site: m[blue]m[]

Report bugs to the email <m[blue]zeromq-dev [at] lists.zeromq.orgm[][1]>


Copyright (c) 1991-2012 iMatix Corporation -- http://www.imatix.com Copyright other contributors as noted in the AUTHORS file. This file is part of CZMQ, the high-level C binding for 0MQ: http://czmq.zeromq.org This Source Code Form is subject to the terms of the Mozilla Public License, v. 2.0. If a copy of the MPL was not distributed with this file, You can obtain one at http://mozilla.org/MPL/2.0/. LICENSE included with the czmq distribution.


zeromq-dev [at] lists.zeromq.org
mailto:zeromq-dev [at] lists.zeromq.org