These documents are Copyright (c) 2009-2010 by Nick Mathewson, and are made available under the Creative Commons Attribution-Noncommercial-Share Alike license, version 3.0. Future versions may be made available under a less restrictive license.
Additionally, the source code examples in these documents are also licensed under the so-called "3-Clause" or "Modified" BSD license. See the license_bsd file distributed with these documents for the full terms.
For the latest version of this document, see http://www.wangafu.net/~nickm/libevent-book/TOC.html
To get the source for the latest version of this document, install git and run "git clone git://github.com/nmathewson/libevent-book.git"
Working with an event loop
Running the loop
Once you have an event_base with some events registered (see the next section about how to create and register events), you will want Libevent to wait for events and alert you about them.
#define EVLOOP_ONCE 0x01 #define EVLOOP_NONBLOCK 0x02 int event_base_loop(struct event_base *base, int flags);
By default, the event_base_loop() function runs an event_base until there are no more events registered in it. To run the loop, it repeatedly checks whether any of the registered events has triggered (for example, if a read event’s file descriptor is ready to read, or if a timeout event’s timeout is ready to expire). Once this happens, it marks all triggered events as "active", and starts to run them.
You can change the behavior of event_base_loop() by setting one or more flags in its flags argument. If EVLOOP_ONCE is set, then the loop will stop immediately after running any active events. If EVLOOP_NONBLOCK is set, then the loop will not wait for events to trigger: it will only check whether any events are ready to trigger immediately.
When it is done, event_base_loop() returns 0 if it exited normally, and -1 if it exited because of some unhandled error in the backend.
To aid in understanding, here’s an approximate summary of the event_base_loop algorithm:
while (any events are registered with the loop) {
if (EVLOOP_NONBLOCK was set, or any events are already active)
If any registered events have triggered, mark them active.
else
Wait until at least one event has triggered, and mark it active.
for (p = 0; p < n_priorities; ++p {
if (any event with priority of p is active) {
Run all active events with priority of p.
break; /* Do not run any events of a less important priority */
}
}
if (EVLOOP_ONCE was set or EVLOOP_NONBLOCK was set)
break;
}
As a convenience, you can also call:
int event_base_dispatch(struct event_base *base);
The event_base_dispatch() call is the same as event_base_loop(), with no flags set. Thus, it keeps running until there are no more registered events or until event_base_loopbreak() or event_base_loopexit() is called.
These functions are defined in <event2/event.h>. They have existed since Libevent 1.0.
Stopping the loop
If you want an active event loop to stop running before all events are removed from it, you have two slightly different functions you can call.
int event_base_loopexit(struct event_base *base,
const struct timeval *tv);
int event_base_loopbreak(struct event_base *base);
The event_base_loopexit() function tells an event_base to stop looping after a given time has elapsed. If the tv argument is NULL, the event_base stops looping without a delay. If the event_base is currently running callbacks for any active events, it will continue running them, and not exit until they have all been run.
The event_base_loopbreak() function tells the event_base to exit its loop immediately. It differs from event_base_loopexit(base, NULL) in that if the event_base is currently running callbacks for any active events, it will exit immediately after finishing the one it’s currently processing.
Both of these methods return 0 on success and -1 on failure.
#include <event2/event.h>
/* Here's a callback function that calls loopbreak */
void cb(int sock, short what, void *arg)
{
struct event_base *base = arg;
event_base_loopbreak(base);
}
void main_loop(struct event_base *base, evutil_socket_t watchdog_fd)
{
struct event *watchdog_event;
/* Construct a new event to trigger whenever there are any bytes to
read from a watchdog socket. When that happens, we'll call the
cb function, which will make the loop exit immediately without
running any other active events at all.
*/
watchdog_event = event_new(base, watchdog_fd, EV_READ, cb, base);
event_add(watchdog_event, NULL);
event_base_dispatch(base);
}
#include <event2/event.h>
void run_base_with_ticks(struct event_base *base)
{
struct timeval ten_sec;
ten_sec.tv_sec = 10;
ten_sec.tv_usec = 0;
/* Now we run the event_base for a series of 10-second intervals, printing
"Tick" after each. For a much better way to implement a 10-second
timer, see the section below about persistent timer events. */
while (1) {
/* This schedules an exit ten seconds from now. */
event_base_loopexit(base, &ten_sec);
event_base_dispatch(base);
puts("Tick");
}
}
Sometimes you may want to tell whether your call to event_base_dispatch() or event_base_loop() exited normally, or because of a call to event_base_loopexit() or event_base_break(). You can use these functions to tell whether loopexit or break was called:
int event_base_got_exit(struct event_base *base); int event_base_got_break(struct event_base *base);
These two functions will return true if the loop was stopped with event_base_loopexit() or event_base_break() respectively, and false otherwise. Their values will be reset the next time you start the event loop.
These functions are declared in <event2/event.h>. The event_break_loopexit() function was first implemented in Libevent 1.0c; event_break_loopbreak() was first implemented in Libevent 1.4.3.
Checking the internal time cache
Sometimes you want to get an approximate view of the current time inside an event callback, and you want to get it without calling gettimeofday() yourself (presumably because your OS implements gettimeofday() as a syscall, and you’re trying to avoid syscall overhead).
From within a callback, you can ask Libevent for its view of the current time when it began executing this round of callbacks:
int event_base_gettimeofday_cached(struct event_base *base,
struct timeval *tv_out);
The event_base_gettimeofday_cached() function sets the value of its tv_out argument to the cached time if the event_base is currently executing callbacks. Otherwise, it calls evutil_gettimeofday() for the actual current time. It returns 0 on success, and negative on failure.
Note that since the timeval is cached when Libevent starts running callbacks, it will be at least a little inaccurate. If your callbacks take a long time to run, it may be very inaccurate.
This function was new in Libevent 2.0.4-alpha.
Dumping the event_base status
void event_base_dump_events(struct event_base *base, FILE *f);
For help debugging your program (or debugging Libevent!) you might sometimes want a complete list of all events added in the event_base and their status. Calling event_base_dump_events() writes this list to the stdio file provided.
The list is meant to be human-readable; its format will change in future versions of Libevent.
This function was introduced in Libevent 2.0.1-alpha.
Obsolete event loop functions
As discussed above, older versions of Libevent APIs had a global notion of a "current" event_base.
Some of the event loop functions in this section had variants that operated on the current base. These functions behaved as the current functions, except that they took no base argument.
| Current function | Obsolete current-base version |
|---|---|
event_base_dispatch() |
event_dispatch() |
event_base_loop() |
event_loop() |
event_base_loopexit() |
event_loopexit() |
event_base_loopbreak() |
event_loopbreak() |
|
Note
|
Because event_base did not support locking before Libevent 2.0, these functions weren’t completely threadsafe: it was not permissible to call the _loopbreak() or _loopexit() functions from a thread other than the one executing the event loop. |