… | |
… | |
62 | |
62 | |
63 | // unloop was called, so exit |
63 | // unloop was called, so exit |
64 | return 0; |
64 | return 0; |
65 | } |
65 | } |
66 | |
66 | |
67 | =head1 DESCRIPTION |
67 | =head1 ABOUT THIS DOCUMENT |
|
|
68 | |
|
|
69 | This document documents the libev software package. |
68 | |
70 | |
69 | The newest version of this document is also available as an html-formatted |
71 | The newest version of this document is also available as an html-formatted |
70 | web page you might find easier to navigate when reading it for the first |
72 | web page you might find easier to navigate when reading it for the first |
71 | time: L<http://pod.tst.eu/http://cvs.schmorp.de/libev/ev.pod>. |
73 | time: L<http://pod.tst.eu/http://cvs.schmorp.de/libev/ev.pod>. |
|
|
74 | |
|
|
75 | While this document tries to be as complete as possible in documenting |
|
|
76 | libev, its usage and the rationale behind its design, it is not a tutorial |
|
|
77 | on event-based programming, nor will it introduce event-based programming |
|
|
78 | with libev. |
|
|
79 | |
|
|
80 | Familarity with event based programming techniques in general is assumed |
|
|
81 | throughout this document. |
|
|
82 | |
|
|
83 | =head1 ABOUT LIBEV |
72 | |
84 | |
73 | Libev is an event loop: you register interest in certain events (such as a |
85 | Libev is an event loop: you register interest in certain events (such as a |
74 | file descriptor being readable or a timeout occurring), and it will manage |
86 | file descriptor being readable or a timeout occurring), and it will manage |
75 | these event sources and provide your program with events. |
87 | these event sources and provide your program with events. |
76 | |
88 | |
… | |
… | |
110 | name C<loop> (which is always of type C<ev_loop *>) will not have |
122 | name C<loop> (which is always of type C<ev_loop *>) will not have |
111 | this argument. |
123 | this argument. |
112 | |
124 | |
113 | =head2 TIME REPRESENTATION |
125 | =head2 TIME REPRESENTATION |
114 | |
126 | |
115 | Libev represents time as a single floating point number, representing the |
127 | Libev represents time as a single floating point number, representing |
116 | (fractional) number of seconds since the (POSIX) epoch (somewhere near |
128 | the (fractional) number of seconds since the (POSIX) epoch (somewhere |
117 | the beginning of 1970, details are complicated, don't ask). This type is |
129 | near the beginning of 1970, details are complicated, don't ask). This |
118 | called C<ev_tstamp>, which is what you should use too. It usually aliases |
130 | type is called C<ev_tstamp>, which is what you should use too. It usually |
119 | to the C<double> type in C, and when you need to do any calculations on |
131 | aliases to the C<double> type in C. When you need to do any calculations |
120 | it, you should treat it as some floating point value. Unlike the name |
132 | on it, you should treat it as some floating point value. Unlike the name |
121 | component C<stamp> might indicate, it is also used for time differences |
133 | component C<stamp> might indicate, it is also used for time differences |
122 | throughout libev. |
134 | throughout libev. |
123 | |
135 | |
124 | =head1 ERROR HANDLING |
136 | =head1 ERROR HANDLING |
125 | |
137 | |
… | |
… | |
609 | |
621 | |
610 | This value can sometimes be useful as a generation counter of sorts (it |
622 | This value can sometimes be useful as a generation counter of sorts (it |
611 | "ticks" the number of loop iterations), as it roughly corresponds with |
623 | "ticks" the number of loop iterations), as it roughly corresponds with |
612 | C<ev_prepare> and C<ev_check> calls. |
624 | C<ev_prepare> and C<ev_check> calls. |
613 | |
625 | |
|
|
626 | =item unsigned int ev_loop_depth (loop) |
|
|
627 | |
|
|
628 | Returns the number of times C<ev_loop> was entered minus the number of |
|
|
629 | times C<ev_loop> was exited, in other words, the recursion depth. |
|
|
630 | |
|
|
631 | Outside C<ev_loop>, this number is zero. In a callback, this number is |
|
|
632 | C<1>, unless C<ev_loop> was invoked recursively (or from another thread), |
|
|
633 | in which case it is higher. |
|
|
634 | |
|
|
635 | Leaving C<ev_loop> abnormally (setjmp/longjmp, cancelling the thread |
|
|
636 | etc.), doesn't count as exit. |
|
|
637 | |
614 | =item unsigned int ev_backend (loop) |
638 | =item unsigned int ev_backend (loop) |
615 | |
639 | |
616 | Returns one of the C<EVBACKEND_*> flags indicating the event backend in |
640 | Returns one of the C<EVBACKEND_*> flags indicating the event backend in |
617 | use. |
641 | use. |
618 | |
642 | |
… | |
… | |
632 | |
656 | |
633 | This function is rarely useful, but when some event callback runs for a |
657 | This function is rarely useful, but when some event callback runs for a |
634 | very long time without entering the event loop, updating libev's idea of |
658 | very long time without entering the event loop, updating libev's idea of |
635 | the current time is a good idea. |
659 | the current time is a good idea. |
636 | |
660 | |
637 | See also "The special problem of time updates" in the C<ev_timer> section. |
661 | See also L<The special problem of time updates> in the C<ev_timer> section. |
638 | |
662 | |
639 | =item ev_suspend (loop) |
663 | =item ev_suspend (loop) |
640 | |
664 | |
641 | =item ev_resume (loop) |
665 | =item ev_resume (loop) |
642 | |
666 | |
… | |
… | |
799 | |
823 | |
800 | By setting a higher I<io collect interval> you allow libev to spend more |
824 | By setting a higher I<io collect interval> you allow libev to spend more |
801 | time collecting I/O events, so you can handle more events per iteration, |
825 | time collecting I/O events, so you can handle more events per iteration, |
802 | at the cost of increasing latency. Timeouts (both C<ev_periodic> and |
826 | at the cost of increasing latency. Timeouts (both C<ev_periodic> and |
803 | C<ev_timer>) will be not affected. Setting this to a non-null value will |
827 | C<ev_timer>) will be not affected. Setting this to a non-null value will |
804 | introduce an additional C<ev_sleep ()> call into most loop iterations. |
828 | introduce an additional C<ev_sleep ()> call into most loop iterations. The |
|
|
829 | sleep time ensures that libev will not poll for I/O events more often then |
|
|
830 | once per this interval, on average. |
805 | |
831 | |
806 | Likewise, by setting a higher I<timeout collect interval> you allow libev |
832 | Likewise, by setting a higher I<timeout collect interval> you allow libev |
807 | to spend more time collecting timeouts, at the expense of increased |
833 | to spend more time collecting timeouts, at the expense of increased |
808 | latency/jitter/inexactness (the watcher callback will be called |
834 | latency/jitter/inexactness (the watcher callback will be called |
809 | later). C<ev_io> watchers will not be affected. Setting this to a non-null |
835 | later). C<ev_io> watchers will not be affected. Setting this to a non-null |
… | |
… | |
811 | |
837 | |
812 | Many (busy) programs can usually benefit by setting the I/O collect |
838 | Many (busy) programs can usually benefit by setting the I/O collect |
813 | interval to a value near C<0.1> or so, which is often enough for |
839 | interval to a value near C<0.1> or so, which is often enough for |
814 | interactive servers (of course not for games), likewise for timeouts. It |
840 | interactive servers (of course not for games), likewise for timeouts. It |
815 | usually doesn't make much sense to set it to a lower value than C<0.01>, |
841 | usually doesn't make much sense to set it to a lower value than C<0.01>, |
816 | as this approaches the timing granularity of most systems. |
842 | as this approaches the timing granularity of most systems. Note that if |
|
|
843 | you do transactions with the outside world and you can't increase the |
|
|
844 | parallelity, then this setting will limit your transaction rate (if you |
|
|
845 | need to poll once per transaction and the I/O collect interval is 0.01, |
|
|
846 | then you can't do more than 100 transations per second). |
817 | |
847 | |
818 | Setting the I<timeout collect interval> can improve the opportunity for |
848 | Setting the I<timeout collect interval> can improve the opportunity for |
819 | saving power, as the program will "bundle" timer callback invocations that |
849 | saving power, as the program will "bundle" timer callback invocations that |
820 | are "near" in time together, by delaying some, thus reducing the number of |
850 | are "near" in time together, by delaying some, thus reducing the number of |
821 | times the process sleeps and wakes up again. Another useful technique to |
851 | times the process sleeps and wakes up again. Another useful technique to |
822 | reduce iterations/wake-ups is to use C<ev_periodic> watchers and make sure |
852 | reduce iterations/wake-ups is to use C<ev_periodic> watchers and make sure |
823 | they fire on, say, one-second boundaries only. |
853 | they fire on, say, one-second boundaries only. |
|
|
854 | |
|
|
855 | Example: we only need 0.1s timeout granularity, and we wish not to poll |
|
|
856 | more often than 100 times per second: |
|
|
857 | |
|
|
858 | ev_set_timeout_collect_interval (EV_DEFAULT_UC_ 0.1); |
|
|
859 | ev_set_io_collect_interval (EV_DEFAULT_UC_ 0.01); |
|
|
860 | |
|
|
861 | =item ev_invoke_pending (loop) |
|
|
862 | |
|
|
863 | This call will simply invoke all pending watchers while resetting their |
|
|
864 | pending state. Normally, C<ev_loop> does this automatically when required, |
|
|
865 | but when overriding the invoke callback this call comes handy. |
|
|
866 | |
|
|
867 | =item ev_set_invoke_pending_cb (loop, void (*invoke_pending_cb)(EV_P)) |
|
|
868 | |
|
|
869 | This overrides the invoke pending functionality of the loop: Instead of |
|
|
870 | invoking all pending watchers when there are any, C<ev_loop> will call |
|
|
871 | this callback instead. This is useful, for example, when you want to |
|
|
872 | invoke the actual watchers inside another context (another thread etc.). |
|
|
873 | |
|
|
874 | If you want to reset the callback, use C<ev_invoke_pending> as new |
|
|
875 | callback. |
|
|
876 | |
|
|
877 | =item ev_set_loop_release_cb (loop, void (*release)(EV_P), void (*acquire)(EV_P)) |
|
|
878 | |
|
|
879 | Sometimes you want to share the same loop between multiple threads. This |
|
|
880 | can be done relatively simply by putting mutex_lock/unlock calls around |
|
|
881 | each call to a libev function. |
|
|
882 | |
|
|
883 | However, C<ev_loop> can run an indefinite time, so it is not feasible to |
|
|
884 | wait for it to return. One way around this is to wake up the loop via |
|
|
885 | C<ev_unloop> and C<av_async_send>, another way is to set these I<release> |
|
|
886 | and I<acquire> callbacks on the loop. |
|
|
887 | |
|
|
888 | When set, then C<release> will be called just before the thread is |
|
|
889 | suspended waiting for new events, and C<acquire> is called just |
|
|
890 | afterwards. |
|
|
891 | |
|
|
892 | Ideally, C<release> will just call your mutex_unlock function, and |
|
|
893 | C<acquire> will just call the mutex_lock function again. |
|
|
894 | |
|
|
895 | =item ev_set_userdata (loop, void *data) |
|
|
896 | |
|
|
897 | =item ev_userdata (loop) |
|
|
898 | |
|
|
899 | Set and retrieve a single C<void *> associated with a loop. When |
|
|
900 | C<ev_set_userdata> has never been called, then C<ev_userdata> returns |
|
|
901 | C<0.> |
|
|
902 | |
|
|
903 | These two functions can be used to associate arbitrary data with a loop, |
|
|
904 | and are intended solely for the C<invoke_pending_cb>, C<release> and |
|
|
905 | C<acquire> callbacks described above, but of course can be (ab-)used for |
|
|
906 | any other purpose as well. |
824 | |
907 | |
825 | =item ev_loop_verify (loop) |
908 | =item ev_loop_verify (loop) |
826 | |
909 | |
827 | This function only does something when C<EV_VERIFY> support has been |
910 | This function only does something when C<EV_VERIFY> support has been |
828 | compiled in, which is the default for non-minimal builds. It tries to go |
911 | compiled in, which is the default for non-minimal builds. It tries to go |
… | |
… | |
1083 | integer between C<EV_MAXPRI> (default: C<2>) and C<EV_MINPRI> |
1166 | integer between C<EV_MAXPRI> (default: C<2>) and C<EV_MINPRI> |
1084 | (default: C<-2>). Pending watchers with higher priority will be invoked |
1167 | (default: C<-2>). Pending watchers with higher priority will be invoked |
1085 | before watchers with lower priority, but priority will not keep watchers |
1168 | before watchers with lower priority, but priority will not keep watchers |
1086 | from being executed (except for C<ev_idle> watchers). |
1169 | from being executed (except for C<ev_idle> watchers). |
1087 | |
1170 | |
1088 | See L< |
|
|
1089 | |
|
|
1090 | This means that priorities are I<only> used for ordering callback |
|
|
1091 | invocation after new events have been received. This is useful, for |
|
|
1092 | example, to reduce latency after idling, or more often, to bind two |
|
|
1093 | watchers on the same event and make sure one is called first. |
|
|
1094 | |
|
|
1095 | If you need to suppress invocation when higher priority events are pending |
1171 | If you need to suppress invocation when higher priority events are pending |
1096 | you need to look at C<ev_idle> watchers, which provide this functionality. |
1172 | you need to look at C<ev_idle> watchers, which provide this functionality. |
1097 | |
1173 | |
1098 | You I<must not> change the priority of a watcher as long as it is active or |
1174 | You I<must not> change the priority of a watcher as long as it is active or |
1099 | pending. |
1175 | pending. |
1100 | |
|
|
1101 | The default priority used by watchers when no priority has been set is |
|
|
1102 | always C<0>, which is supposed to not be too high and not be too low :). |
|
|
1103 | |
1176 | |
1104 | Setting a priority outside the range of C<EV_MINPRI> to C<EV_MAXPRI> is |
1177 | Setting a priority outside the range of C<EV_MINPRI> to C<EV_MAXPRI> is |
1105 | fine, as long as you do not mind that the priority value you query might |
1178 | fine, as long as you do not mind that the priority value you query might |
1106 | or might not have been clamped to the valid range. |
1179 | or might not have been clamped to the valid range. |
|
|
1180 | |
|
|
1181 | The default priority used by watchers when no priority has been set is |
|
|
1182 | always C<0>, which is supposed to not be too high and not be too low :). |
|
|
1183 | |
|
|
1184 | See L<WATCHER PRIORITY MODELS>, below, for a more thorough treatment of |
|
|
1185 | priorities. |
1107 | |
1186 | |
1108 | =item ev_invoke (loop, ev_TYPE *watcher, int revents) |
1187 | =item ev_invoke (loop, ev_TYPE *watcher, int revents) |
1109 | |
1188 | |
1110 | Invoke the C<watcher> with the given C<loop> and C<revents>. Neither |
1189 | Invoke the C<watcher> with the given C<loop> and C<revents>. Neither |
1111 | C<loop> nor C<revents> need to be valid as long as the watcher callback |
1190 | C<loop> nor C<revents> need to be valid as long as the watcher callback |
… | |
… | |
1176 | #include <stddef.h> |
1255 | #include <stddef.h> |
1177 | |
1256 | |
1178 | static void |
1257 | static void |
1179 | t1_cb (EV_P_ ev_timer *w, int revents) |
1258 | t1_cb (EV_P_ ev_timer *w, int revents) |
1180 | { |
1259 | { |
1181 | struct my_biggy big = (struct my_biggy * |
1260 | struct my_biggy big = (struct my_biggy *) |
1182 | (((char *)w) - offsetof (struct my_biggy, t1)); |
1261 | (((char *)w) - offsetof (struct my_biggy, t1)); |
1183 | } |
1262 | } |
1184 | |
1263 | |
1185 | static void |
1264 | static void |
1186 | t2_cb (EV_P_ ev_timer *w, int revents) |
1265 | t2_cb (EV_P_ ev_timer *w, int revents) |
1187 | { |
1266 | { |
1188 | struct my_biggy big = (struct my_biggy * |
1267 | struct my_biggy big = (struct my_biggy *) |
1189 | (((char *)w) - offsetof (struct my_biggy, t2)); |
1268 | (((char *)w) - offsetof (struct my_biggy, t2)); |
1190 | } |
1269 | } |
|
|
1270 | |
|
|
1271 | =head2 WATCHER PRIORITY MODELS |
|
|
1272 | |
|
|
1273 | Many event loops support I<watcher priorities>, which are usually small |
|
|
1274 | integers that influence the ordering of event callback invocation |
|
|
1275 | between watchers in some way, all else being equal. |
|
|
1276 | |
|
|
1277 | In libev, Watcher priorities can be set using C<ev_set_priority>. See its |
|
|
1278 | description for the more technical details such as the actual priority |
|
|
1279 | range. |
|
|
1280 | |
|
|
1281 | There are two common ways how these these priorities are being interpreted |
|
|
1282 | by event loops: |
|
|
1283 | |
|
|
1284 | In the more common lock-out model, higher priorities "lock out" invocation |
|
|
1285 | of lower priority watchers, which means as long as higher priority |
|
|
1286 | watchers receive events, lower priority watchers are not being invoked. |
|
|
1287 | |
|
|
1288 | The less common only-for-ordering model uses priorities solely to order |
|
|
1289 | callback invocation within a single event loop iteration: Higher priority |
|
|
1290 | watchers are invoked before lower priority ones, but they all get invoked |
|
|
1291 | before polling for new events. |
|
|
1292 | |
|
|
1293 | Libev uses the second (only-for-ordering) model for all its watchers |
|
|
1294 | except for idle watchers (which use the lock-out model). |
|
|
1295 | |
|
|
1296 | The rationale behind this is that implementing the lock-out model for |
|
|
1297 | watchers is not well supported by most kernel interfaces, and most event |
|
|
1298 | libraries will just poll for the same events again and again as long as |
|
|
1299 | their callbacks have not been executed, which is very inefficient in the |
|
|
1300 | common case of one high-priority watcher locking out a mass of lower |
|
|
1301 | priority ones. |
|
|
1302 | |
|
|
1303 | Static (ordering) priorities are most useful when you have two or more |
|
|
1304 | watchers handling the same resource: a typical usage example is having an |
|
|
1305 | C<ev_io> watcher to receive data, and an associated C<ev_timer> to handle |
|
|
1306 | timeouts. Under load, data might be received while the program handles |
|
|
1307 | other jobs, but since timers normally get invoked first, the timeout |
|
|
1308 | handler will be executed before checking for data. In that case, giving |
|
|
1309 | the timer a lower priority than the I/O watcher ensures that I/O will be |
|
|
1310 | handled first even under adverse conditions (which is usually, but not |
|
|
1311 | always, what you want). |
|
|
1312 | |
|
|
1313 | Since idle watchers use the "lock-out" model, meaning that idle watchers |
|
|
1314 | will only be executed when no same or higher priority watchers have |
|
|
1315 | received events, they can be used to implement the "lock-out" model when |
|
|
1316 | required. |
|
|
1317 | |
|
|
1318 | For example, to emulate how many other event libraries handle priorities, |
|
|
1319 | you can associate an C<ev_idle> watcher to each such watcher, and in |
|
|
1320 | the normal watcher callback, you just start the idle watcher. The real |
|
|
1321 | processing is done in the idle watcher callback. This causes libev to |
|
|
1322 | continously poll and process kernel event data for the watcher, but when |
|
|
1323 | the lock-out case is known to be rare (which in turn is rare :), this is |
|
|
1324 | workable. |
|
|
1325 | |
|
|
1326 | Usually, however, the lock-out model implemented that way will perform |
|
|
1327 | miserably under the type of load it was designed to handle. In that case, |
|
|
1328 | it might be preferable to stop the real watcher before starting the |
|
|
1329 | idle watcher, so the kernel will not have to process the event in case |
|
|
1330 | the actual processing will be delayed for considerable time. |
|
|
1331 | |
|
|
1332 | Here is an example of an I/O watcher that should run at a strictly lower |
|
|
1333 | priority than the default, and which should only process data when no |
|
|
1334 | other events are pending: |
|
|
1335 | |
|
|
1336 | ev_idle idle; // actual processing watcher |
|
|
1337 | ev_io io; // actual event watcher |
|
|
1338 | |
|
|
1339 | static void |
|
|
1340 | io_cb (EV_P_ ev_io *w, int revents) |
|
|
1341 | { |
|
|
1342 | // stop the I/O watcher, we received the event, but |
|
|
1343 | // are not yet ready to handle it. |
|
|
1344 | ev_io_stop (EV_A_ w); |
|
|
1345 | |
|
|
1346 | // start the idle watcher to ahndle the actual event. |
|
|
1347 | // it will not be executed as long as other watchers |
|
|
1348 | // with the default priority are receiving events. |
|
|
1349 | ev_idle_start (EV_A_ &idle); |
|
|
1350 | } |
|
|
1351 | |
|
|
1352 | static void |
|
|
1353 | idle_cb (EV_P_ ev_idle *w, int revents) |
|
|
1354 | { |
|
|
1355 | // actual processing |
|
|
1356 | read (STDIN_FILENO, ...); |
|
|
1357 | |
|
|
1358 | // have to start the I/O watcher again, as |
|
|
1359 | // we have handled the event |
|
|
1360 | ev_io_start (EV_P_ &io); |
|
|
1361 | } |
|
|
1362 | |
|
|
1363 | // initialisation |
|
|
1364 | ev_idle_init (&idle, idle_cb); |
|
|
1365 | ev_io_init (&io, io_cb, STDIN_FILENO, EV_READ); |
|
|
1366 | ev_io_start (EV_DEFAULT_ &io); |
|
|
1367 | |
|
|
1368 | In the "real" world, it might also be beneficial to start a timer, so that |
|
|
1369 | low-priority connections can not be locked out forever under load. This |
|
|
1370 | enables your program to keep a lower latency for important connections |
|
|
1371 | during short periods of high load, while not completely locking out less |
|
|
1372 | important ones. |
1191 | |
1373 | |
1192 | |
1374 | |
1193 | =head1 WATCHER TYPES |
1375 | =head1 WATCHER TYPES |
1194 | |
1376 | |
1195 | This section describes each watcher in detail, but will not repeat |
1377 | This section describes each watcher in detail, but will not repeat |
… | |
… | |
1221 | descriptors to non-blocking mode is also usually a good idea (but not |
1403 | descriptors to non-blocking mode is also usually a good idea (but not |
1222 | required if you know what you are doing). |
1404 | required if you know what you are doing). |
1223 | |
1405 | |
1224 | If you cannot use non-blocking mode, then force the use of a |
1406 | If you cannot use non-blocking mode, then force the use of a |
1225 | known-to-be-good backend (at the time of this writing, this includes only |
1407 | known-to-be-good backend (at the time of this writing, this includes only |
1226 | C<EVBACKEND_SELECT> and C<EVBACKEND_POLL>). |
1408 | C<EVBACKEND_SELECT> and C<EVBACKEND_POLL>). The same applies to file |
|
|
1409 | descriptors for which non-blocking operation makes no sense (such as |
|
|
1410 | files) - libev doesn't guarentee any specific behaviour in that case. |
1227 | |
1411 | |
1228 | Another thing you have to watch out for is that it is quite easy to |
1412 | Another thing you have to watch out for is that it is quite easy to |
1229 | receive "spurious" readiness notifications, that is your callback might |
1413 | receive "spurious" readiness notifications, that is your callback might |
1230 | be called with C<EV_READ> but a subsequent C<read>(2) will actually block |
1414 | be called with C<EV_READ> but a subsequent C<read>(2) will actually block |
1231 | because there is no data. Not only are some backends known to create a |
1415 | because there is no data. Not only are some backends known to create a |
… | |
… | |
1352 | year, it will still time out after (roughly) one hour. "Roughly" because |
1536 | year, it will still time out after (roughly) one hour. "Roughly" because |
1353 | detecting time jumps is hard, and some inaccuracies are unavoidable (the |
1537 | detecting time jumps is hard, and some inaccuracies are unavoidable (the |
1354 | monotonic clock option helps a lot here). |
1538 | monotonic clock option helps a lot here). |
1355 | |
1539 | |
1356 | The callback is guaranteed to be invoked only I<after> its timeout has |
1540 | The callback is guaranteed to be invoked only I<after> its timeout has |
1357 | passed. If multiple timers become ready during the same loop iteration |
1541 | passed (not I<at>, so on systems with very low-resolution clocks this |
1358 | then the ones with earlier time-out values are invoked before ones with |
1542 | might introduce a small delay). If multiple timers become ready during the |
1359 | later time-out values (but this is no longer true when a callback calls |
1543 | same loop iteration then the ones with earlier time-out values are invoked |
1360 | C<ev_loop> recursively). |
1544 | before ones of the same priority with later time-out values (but this is |
|
|
1545 | no longer true when a callback calls C<ev_loop> recursively). |
1361 | |
1546 | |
1362 | =head3 Be smart about timeouts |
1547 | =head3 Be smart about timeouts |
1363 | |
1548 | |
1364 | Many real-world problems involve some kind of timeout, usually for error |
1549 | Many real-world problems involve some kind of timeout, usually for error |
1365 | recovery. A typical example is an HTTP request - if the other side hangs, |
1550 | recovery. A typical example is an HTTP request - if the other side hangs, |
… | |
… | |
1409 | C<after> argument to C<ev_timer_set>, and only ever use the C<repeat> |
1594 | C<after> argument to C<ev_timer_set>, and only ever use the C<repeat> |
1410 | member and C<ev_timer_again>. |
1595 | member and C<ev_timer_again>. |
1411 | |
1596 | |
1412 | At start: |
1597 | At start: |
1413 | |
1598 | |
1414 | ev_timer_init (timer, callback); |
1599 | ev_init (timer, callback); |
1415 | timer->repeat = 60.; |
1600 | timer->repeat = 60.; |
1416 | ev_timer_again (loop, timer); |
1601 | ev_timer_again (loop, timer); |
1417 | |
1602 | |
1418 | Each time there is some activity: |
1603 | Each time there is some activity: |
1419 | |
1604 | |
… | |
… | |
1481 | |
1666 | |
1482 | To start the timer, simply initialise the watcher and set C<last_activity> |
1667 | To start the timer, simply initialise the watcher and set C<last_activity> |
1483 | to the current time (meaning we just have some activity :), then call the |
1668 | to the current time (meaning we just have some activity :), then call the |
1484 | callback, which will "do the right thing" and start the timer: |
1669 | callback, which will "do the right thing" and start the timer: |
1485 | |
1670 | |
1486 | ev_timer_init (timer, callback); |
1671 | ev_init (timer, callback); |
1487 | last_activity = ev_now (loop); |
1672 | last_activity = ev_now (loop); |
1488 | callback (loop, timer, EV_TIMEOUT); |
1673 | callback (loop, timer, EV_TIMEOUT); |
1489 | |
1674 | |
1490 | And when there is some activity, simply store the current time in |
1675 | And when there is some activity, simply store the current time in |
1491 | C<last_activity>, no libev calls at all: |
1676 | C<last_activity>, no libev calls at all: |
… | |
… | |
1888 | some child status changes (most typically when a child of yours dies or |
2073 | some child status changes (most typically when a child of yours dies or |
1889 | exits). It is permissible to install a child watcher I<after> the child |
2074 | exits). It is permissible to install a child watcher I<after> the child |
1890 | has been forked (which implies it might have already exited), as long |
2075 | has been forked (which implies it might have already exited), as long |
1891 | as the event loop isn't entered (or is continued from a watcher), i.e., |
2076 | as the event loop isn't entered (or is continued from a watcher), i.e., |
1892 | forking and then immediately registering a watcher for the child is fine, |
2077 | forking and then immediately registering a watcher for the child is fine, |
1893 | but forking and registering a watcher a few event loop iterations later is |
2078 | but forking and registering a watcher a few event loop iterations later or |
1894 | not. |
2079 | in the next callback invocation is not. |
1895 | |
2080 | |
1896 | Only the default event loop is capable of handling signals, and therefore |
2081 | Only the default event loop is capable of handling signals, and therefore |
1897 | you can only register child watchers in the default event loop. |
2082 | you can only register child watchers in the default event loop. |
|
|
2083 | |
|
|
2084 | Due to some design glitches inside libev, child watchers will always be |
|
|
2085 | handled at maximum priority (their priority is set to C<EV_MAXPRI> by |
|
|
2086 | libev) |
1898 | |
2087 | |
1899 | =head3 Process Interaction |
2088 | =head3 Process Interaction |
1900 | |
2089 | |
1901 | Libev grabs C<SIGCHLD> as soon as the default event loop is |
2090 | Libev grabs C<SIGCHLD> as soon as the default event loop is |
1902 | initialised. This is necessary to guarantee proper behaviour even if |
2091 | initialised. This is necessary to guarantee proper behaviour even if |
… | |
… | |
2254 | // no longer anything immediate to do. |
2443 | // no longer anything immediate to do. |
2255 | } |
2444 | } |
2256 | |
2445 | |
2257 | ev_idle *idle_watcher = malloc (sizeof (ev_idle)); |
2446 | ev_idle *idle_watcher = malloc (sizeof (ev_idle)); |
2258 | ev_idle_init (idle_watcher, idle_cb); |
2447 | ev_idle_init (idle_watcher, idle_cb); |
2259 | ev_idle_start (loop, idle_cb); |
2448 | ev_idle_start (loop, idle_watcher); |
2260 | |
2449 | |
2261 | |
2450 | |
2262 | =head2 C<ev_prepare> and C<ev_check> - customise your event loop! |
2451 | =head2 C<ev_prepare> and C<ev_check> - customise your event loop! |
2263 | |
2452 | |
2264 | Prepare and check watchers are usually (but not always) used in pairs: |
2453 | Prepare and check watchers are usually (but not always) used in pairs: |
… | |
… | |
2357 | struct pollfd fds [nfd]; |
2546 | struct pollfd fds [nfd]; |
2358 | // actual code will need to loop here and realloc etc. |
2547 | // actual code will need to loop here and realloc etc. |
2359 | adns_beforepoll (ads, fds, &nfd, &timeout, timeval_from (ev_time ())); |
2548 | adns_beforepoll (ads, fds, &nfd, &timeout, timeval_from (ev_time ())); |
2360 | |
2549 | |
2361 | /* the callback is illegal, but won't be called as we stop during check */ |
2550 | /* the callback is illegal, but won't be called as we stop during check */ |
2362 | ev_timer_init (&tw, 0, timeout * 1e-3); |
2551 | ev_timer_init (&tw, 0, timeout * 1e-3, 0.); |
2363 | ev_timer_start (loop, &tw); |
2552 | ev_timer_start (loop, &tw); |
2364 | |
2553 | |
2365 | // create one ev_io per pollfd |
2554 | // create one ev_io per pollfd |
2366 | for (int i = 0; i < nfd; ++i) |
2555 | for (int i = 0; i < nfd; ++i) |
2367 | { |
2556 | { |
… | |
… | |
2597 | event loop blocks next and before C<ev_check> watchers are being called, |
2786 | event loop blocks next and before C<ev_check> watchers are being called, |
2598 | and only in the child after the fork. If whoever good citizen calling |
2787 | and only in the child after the fork. If whoever good citizen calling |
2599 | C<ev_default_fork> cheats and calls it in the wrong process, the fork |
2788 | C<ev_default_fork> cheats and calls it in the wrong process, the fork |
2600 | handlers will be invoked, too, of course. |
2789 | handlers will be invoked, too, of course. |
2601 | |
2790 | |
|
|
2791 | =head3 The special problem of life after fork - how is it possible? |
|
|
2792 | |
|
|
2793 | Most uses of C<fork()> consist of forking, then some simple calls to ste |
|
|
2794 | up/change the process environment, followed by a call to C<exec()>. This |
|
|
2795 | sequence should be handled by libev without any problems. |
|
|
2796 | |
|
|
2797 | This changes when the application actually wants to do event handling |
|
|
2798 | in the child, or both parent in child, in effect "continuing" after the |
|
|
2799 | fork. |
|
|
2800 | |
|
|
2801 | The default mode of operation (for libev, with application help to detect |
|
|
2802 | forks) is to duplicate all the state in the child, as would be expected |
|
|
2803 | when I<either> the parent I<or> the child process continues. |
|
|
2804 | |
|
|
2805 | When both processes want to continue using libev, then this is usually the |
|
|
2806 | wrong result. In that case, usually one process (typically the parent) is |
|
|
2807 | supposed to continue with all watchers in place as before, while the other |
|
|
2808 | process typically wants to start fresh, i.e. without any active watchers. |
|
|
2809 | |
|
|
2810 | The cleanest and most efficient way to achieve that with libev is to |
|
|
2811 | simply create a new event loop, which of course will be "empty", and |
|
|
2812 | use that for new watchers. This has the advantage of not touching more |
|
|
2813 | memory than necessary, and thus avoiding the copy-on-write, and the |
|
|
2814 | disadvantage of having to use multiple event loops (which do not support |
|
|
2815 | signal watchers). |
|
|
2816 | |
|
|
2817 | When this is not possible, or you want to use the default loop for |
|
|
2818 | other reasons, then in the process that wants to start "fresh", call |
|
|
2819 | C<ev_default_destroy ()> followed by C<ev_default_loop (...)>. Destroying |
|
|
2820 | the default loop will "orphan" (not stop) all registered watchers, so you |
|
|
2821 | have to be careful not to execute code that modifies those watchers. Note |
|
|
2822 | also that in that case, you have to re-register any signal watchers. |
|
|
2823 | |
2602 | =head3 Watcher-Specific Functions and Data Members |
2824 | =head3 Watcher-Specific Functions and Data Members |
2603 | |
2825 | |
2604 | =over 4 |
2826 | =over 4 |
2605 | |
2827 | |
2606 | =item ev_fork_init (ev_signal *, callback) |
2828 | =item ev_fork_init (ev_signal *, callback) |
… | |
… | |
3496 | defined to be C<0>, then they are not. |
3718 | defined to be C<0>, then they are not. |
3497 | |
3719 | |
3498 | =item EV_MINIMAL |
3720 | =item EV_MINIMAL |
3499 | |
3721 | |
3500 | If you need to shave off some kilobytes of code at the expense of some |
3722 | If you need to shave off some kilobytes of code at the expense of some |
3501 | speed, define this symbol to C<1>. Currently this is used to override some |
3723 | speed (but with the full API), define this symbol to C<1>. Currently this |
3502 | inlining decisions, saves roughly 30% code size on amd64. It also selects a |
3724 | is used to override some inlining decisions, saves roughly 30% code size |
3503 | much smaller 2-heap for timer management over the default 4-heap. |
3725 | on amd64. It also selects a much smaller 2-heap for timer management over |
|
|
3726 | the default 4-heap. |
|
|
3727 | |
|
|
3728 | You can save even more by disabling watcher types you do not need |
|
|
3729 | and setting C<EV_MAXPRI> == C<EV_MINPRI>. Also, disabling C<assert> |
|
|
3730 | (C<-DNDEBUG>) will usually reduce code size a lot. |
|
|
3731 | |
|
|
3732 | Defining C<EV_MINIMAL> to C<2> will additionally reduce the core API to |
|
|
3733 | provide a bare-bones event library. See C<ev.h> for details on what parts |
|
|
3734 | of the API are still available, and do not complain if this subset changes |
|
|
3735 | over time. |
3504 | |
3736 | |
3505 | =item EV_PID_HASHSIZE |
3737 | =item EV_PID_HASHSIZE |
3506 | |
3738 | |
3507 | C<ev_child> watchers use a small hash table to distribute workload by |
3739 | C<ev_child> watchers use a small hash table to distribute workload by |
3508 | pid. The default size is C<16> (or C<1> with C<EV_MINIMAL>), usually more |
3740 | pid. The default size is C<16> (or C<1> with C<EV_MINIMAL>), usually more |
… | |
… | |
3694 | default loop and triggering an C<ev_async> watcher from the default loop |
3926 | default loop and triggering an C<ev_async> watcher from the default loop |
3695 | watcher callback into the event loop interested in the signal. |
3927 | watcher callback into the event loop interested in the signal. |
3696 | |
3928 | |
3697 | =back |
3929 | =back |
3698 | |
3930 | |
|
|
3931 | =head4 THREAD LOCKING EXAMPLE |
|
|
3932 | |
3699 | =head3 COROUTINES |
3933 | =head3 COROUTINES |
3700 | |
3934 | |
3701 | Libev is very accommodating to coroutines ("cooperative threads"): |
3935 | Libev is very accommodating to coroutines ("cooperative threads"): |
3702 | libev fully supports nesting calls to its functions from different |
3936 | libev fully supports nesting calls to its functions from different |
3703 | coroutines (e.g. you can call C<ev_loop> on the same loop from two |
3937 | coroutines (e.g. you can call C<ev_loop> on the same loop from two |
… | |
… | |
3788 | way (note also that glib is the slowest event library known to man). |
4022 | way (note also that glib is the slowest event library known to man). |
3789 | |
4023 | |
3790 | There is no supported compilation method available on windows except |
4024 | There is no supported compilation method available on windows except |
3791 | embedding it into other applications. |
4025 | embedding it into other applications. |
3792 | |
4026 | |
|
|
4027 | Sensible signal handling is officially unsupported by Microsoft - libev |
|
|
4028 | tries its best, but under most conditions, signals will simply not work. |
|
|
4029 | |
3793 | Not a libev limitation but worth mentioning: windows apparently doesn't |
4030 | Not a libev limitation but worth mentioning: windows apparently doesn't |
3794 | accept large writes: instead of resulting in a partial write, windows will |
4031 | accept large writes: instead of resulting in a partial write, windows will |
3795 | either accept everything or return C<ENOBUFS> if the buffer is too large, |
4032 | either accept everything or return C<ENOBUFS> if the buffer is too large, |
3796 | so make sure you only write small amounts into your sockets (less than a |
4033 | so make sure you only write small amounts into your sockets (less than a |
3797 | megabyte seems safe, but this apparently depends on the amount of memory |
4034 | megabyte seems safe, but this apparently depends on the amount of memory |
… | |
… | |
3801 | the abysmal performance of winsockets, using a large number of sockets |
4038 | the abysmal performance of winsockets, using a large number of sockets |
3802 | is not recommended (and not reasonable). If your program needs to use |
4039 | is not recommended (and not reasonable). If your program needs to use |
3803 | more than a hundred or so sockets, then likely it needs to use a totally |
4040 | more than a hundred or so sockets, then likely it needs to use a totally |
3804 | different implementation for windows, as libev offers the POSIX readiness |
4041 | different implementation for windows, as libev offers the POSIX readiness |
3805 | notification model, which cannot be implemented efficiently on windows |
4042 | notification model, which cannot be implemented efficiently on windows |
3806 | (Microsoft monopoly games). |
4043 | (due to Microsoft monopoly games). |
3807 | |
4044 | |
3808 | A typical way to use libev under windows is to embed it (see the embedding |
4045 | A typical way to use libev under windows is to embed it (see the embedding |
3809 | section for details) and use the following F<evwrap.h> header file instead |
4046 | section for details) and use the following F<evwrap.h> header file instead |
3810 | of F<ev.h>: |
4047 | of F<ev.h>: |
3811 | |
4048 | |
… | |
… | |
3847 | |
4084 | |
3848 | Early versions of winsocket's select only supported waiting for a maximum |
4085 | Early versions of winsocket's select only supported waiting for a maximum |
3849 | of C<64> handles (probably owning to the fact that all windows kernels |
4086 | of C<64> handles (probably owning to the fact that all windows kernels |
3850 | can only wait for C<64> things at the same time internally; Microsoft |
4087 | can only wait for C<64> things at the same time internally; Microsoft |
3851 | recommends spawning a chain of threads and wait for 63 handles and the |
4088 | recommends spawning a chain of threads and wait for 63 handles and the |
3852 | previous thread in each. Great). |
4089 | previous thread in each. Sounds great!). |
3853 | |
4090 | |
3854 | Newer versions support more handles, but you need to define C<FD_SETSIZE> |
4091 | Newer versions support more handles, but you need to define C<FD_SETSIZE> |
3855 | to some high number (e.g. C<2048>) before compiling the winsocket select |
4092 | to some high number (e.g. C<2048>) before compiling the winsocket select |
3856 | call (which might be in libev or elsewhere, for example, perl does its own |
4093 | call (which might be in libev or elsewhere, for example, perl and many |
3857 | select emulation on windows). |
4094 | other interpreters do their own select emulation on windows). |
3858 | |
4095 | |
3859 | Another limit is the number of file descriptors in the Microsoft runtime |
4096 | Another limit is the number of file descriptors in the Microsoft runtime |
3860 | libraries, which by default is C<64> (there must be a hidden I<64> fetish |
4097 | libraries, which by default is C<64> (there must be a hidden I<64> |
3861 | or something like this inside Microsoft). You can increase this by calling |
4098 | fetish or something like this inside Microsoft). You can increase this |
3862 | C<_setmaxstdio>, which can increase this limit to C<2048> (another |
4099 | by calling C<_setmaxstdio>, which can increase this limit to C<2048> |
3863 | arbitrary limit), but is broken in many versions of the Microsoft runtime |
4100 | (another arbitrary limit), but is broken in many versions of the Microsoft |
3864 | libraries. |
|
|
3865 | |
|
|
3866 | This might get you to about C<512> or C<2048> sockets (depending on |
4101 | runtime libraries. This might get you to about C<512> or C<2048> sockets |
3867 | windows version and/or the phase of the moon). To get more, you need to |
4102 | (depending on windows version and/or the phase of the moon). To get more, |
3868 | wrap all I/O functions and provide your own fd management, but the cost of |
4103 | you need to wrap all I/O functions and provide your own fd management, but |
3869 | calling select (O(n²)) will likely make this unworkable. |
4104 | the cost of calling select (O(n²)) will likely make this unworkable. |
3870 | |
4105 | |
3871 | =back |
4106 | =back |
3872 | |
4107 | |
3873 | =head2 PORTABILITY REQUIREMENTS |
4108 | =head2 PORTABILITY REQUIREMENTS |
3874 | |
4109 | |
… | |
… | |
3917 | =item C<double> must hold a time value in seconds with enough accuracy |
4152 | =item C<double> must hold a time value in seconds with enough accuracy |
3918 | |
4153 | |
3919 | The type C<double> is used to represent timestamps. It is required to |
4154 | The type C<double> is used to represent timestamps. It is required to |
3920 | have at least 51 bits of mantissa (and 9 bits of exponent), which is good |
4155 | have at least 51 bits of mantissa (and 9 bits of exponent), which is good |
3921 | enough for at least into the year 4000. This requirement is fulfilled by |
4156 | enough for at least into the year 4000. This requirement is fulfilled by |
3922 | implementations implementing IEEE 754 (basically all existing ones). |
4157 | implementations implementing IEEE 754, which is basically all existing |
|
|
4158 | ones. With IEEE 754 doubles, you get microsecond accuracy until at least |
|
|
4159 | 2200. |
3923 | |
4160 | |
3924 | =back |
4161 | =back |
3925 | |
4162 | |
3926 | If you know of other additional requirements drop me a note. |
4163 | If you know of other additional requirements drop me a note. |
3927 | |
4164 | |
… | |
… | |
3995 | involves iterating over all running async watchers or all signal numbers. |
4232 | involves iterating over all running async watchers or all signal numbers. |
3996 | |
4233 | |
3997 | =back |
4234 | =back |
3998 | |
4235 | |
3999 | |
4236 | |
|
|
4237 | =head1 GLOSSARY |
|
|
4238 | |
|
|
4239 | =over 4 |
|
|
4240 | |
|
|
4241 | =item active |
|
|
4242 | |
|
|
4243 | A watcher is active as long as it has been started (has been attached to |
|
|
4244 | an event loop) but not yet stopped (disassociated from the event loop). |
|
|
4245 | |
|
|
4246 | =item application |
|
|
4247 | |
|
|
4248 | In this document, an application is whatever is using libev. |
|
|
4249 | |
|
|
4250 | =item callback |
|
|
4251 | |
|
|
4252 | The address of a function that is called when some event has been |
|
|
4253 | detected. Callbacks are being passed the event loop, the watcher that |
|
|
4254 | received the event, and the actual event bitset. |
|
|
4255 | |
|
|
4256 | =item callback invocation |
|
|
4257 | |
|
|
4258 | The act of calling the callback associated with a watcher. |
|
|
4259 | |
|
|
4260 | =item event |
|
|
4261 | |
|
|
4262 | A change of state of some external event, such as data now being available |
|
|
4263 | for reading on a file descriptor, time having passed or simply not having |
|
|
4264 | any other events happening anymore. |
|
|
4265 | |
|
|
4266 | In libev, events are represented as single bits (such as C<EV_READ> or |
|
|
4267 | C<EV_TIMEOUT>). |
|
|
4268 | |
|
|
4269 | =item event library |
|
|
4270 | |
|
|
4271 | A software package implementing an event model and loop. |
|
|
4272 | |
|
|
4273 | =item event loop |
|
|
4274 | |
|
|
4275 | An entity that handles and processes external events and converts them |
|
|
4276 | into callback invocations. |
|
|
4277 | |
|
|
4278 | =item event model |
|
|
4279 | |
|
|
4280 | The model used to describe how an event loop handles and processes |
|
|
4281 | watchers and events. |
|
|
4282 | |
|
|
4283 | =item pending |
|
|
4284 | |
|
|
4285 | A watcher is pending as soon as the corresponding event has been detected, |
|
|
4286 | and stops being pending as soon as the watcher will be invoked or its |
|
|
4287 | pending status is explicitly cleared by the application. |
|
|
4288 | |
|
|
4289 | A watcher can be pending, but not active. Stopping a watcher also clears |
|
|
4290 | its pending status. |
|
|
4291 | |
|
|
4292 | =item real time |
|
|
4293 | |
|
|
4294 | The physical time that is observed. It is apparently strictly monotonic :) |
|
|
4295 | |
|
|
4296 | =item wall-clock time |
|
|
4297 | |
|
|
4298 | The time and date as shown on clocks. Unlike real time, it can actually |
|
|
4299 | be wrong and jump forwards and backwards, e.g. when the you adjust your |
|
|
4300 | clock. |
|
|
4301 | |
|
|
4302 | =item watcher |
|
|
4303 | |
|
|
4304 | A data structure that describes interest in certain events. Watchers need |
|
|
4305 | to be started (attached to an event loop) before they can receive events. |
|
|
4306 | |
|
|
4307 | =item watcher invocation |
|
|
4308 | |
|
|
4309 | The act of calling the callback associated with a watcher. |
|
|
4310 | |
|
|
4311 | =back |
|
|
4312 | |
4000 | =head1 AUTHOR |
4313 | =head1 AUTHOR |
4001 | |
4314 | |
4002 | Marc Lehmann <libev@schmorp.de>, with repeated corrections by Mikael Magnusson. |
4315 | Marc Lehmann <libev@schmorp.de>, with repeated corrections by Mikael Magnusson. |
4003 | |
4316 | |