… | |
… | |
380 | |
380 | |
381 | There is a slight catch to child watchers, however: you usually start |
381 | There is a slight catch to child watchers, however: you usually start |
382 | them *after* the child process was created, and this means the process |
382 | them *after* the child process was created, and this means the process |
383 | could have exited already (and no SIGCHLD will be sent anymore). |
383 | could have exited already (and no SIGCHLD will be sent anymore). |
384 | |
384 | |
385 | Not all event models handle this correctly (POE doesn't), but even for |
385 | Not all event models handle this correctly (neither POE nor IO::Async |
|
|
386 | do, see their AnyEvent::Impl manpages for details), but even for event |
386 | event models that *do* handle this correctly, they usually need to be |
387 | models that *do* handle this correctly, they usually need to be loaded |
387 | loaded before the process exits (i.e. before you fork in the first |
388 | before the process exits (i.e. before you fork in the first place). |
388 | place). |
389 | AnyEvent's pure perl event loop handles all cases correctly regardless |
|
|
390 | of when you start the watcher. |
389 | |
391 | |
390 | This means you cannot create a child watcher as the very first thing in |
392 | This means you cannot create a child watcher as the very first thing in |
391 | an AnyEvent program, you *have* to create at least one watcher before |
393 | an AnyEvent program, you *have* to create at least one watcher before |
392 | you "fork" the child (alternatively, you can call "AnyEvent::detect"). |
394 | you "fork" the child (alternatively, you can call "AnyEvent::detect"). |
393 | |
395 | |
… | |
… | |
697 | AnyEvent::Impl::Tk based on Tk, very bad choice. |
699 | AnyEvent::Impl::Tk based on Tk, very bad choice. |
698 | AnyEvent::Impl::Qt based on Qt, cannot be autoprobed (see its docs). |
700 | AnyEvent::Impl::Qt based on Qt, cannot be autoprobed (see its docs). |
699 | AnyEvent::Impl::EventLib based on Event::Lib, leaks memory and worse. |
701 | AnyEvent::Impl::EventLib based on Event::Lib, leaks memory and worse. |
700 | AnyEvent::Impl::POE based on POE, not generic enough for full support. |
702 | AnyEvent::Impl::POE based on POE, not generic enough for full support. |
701 | |
703 | |
|
|
704 | # warning, support for IO::Async is only partial, as it is too broken |
|
|
705 | # and limited toe ven support the AnyEvent API. See AnyEvent::Impl::Async. |
|
|
706 | AnyEvent::Impl::IOAsync based on IO::Async, cannot be autoprobed (see its docs). |
|
|
707 | |
702 | There is no support for WxWidgets, as WxWidgets has no support for |
708 | There is no support for WxWidgets, as WxWidgets has no support for |
703 | watching file handles. However, you can use WxWidgets through the |
709 | watching file handles. However, you can use WxWidgets through the |
704 | POE Adaptor, as POE has a Wx backend that simply polls 20 times per |
710 | POE Adaptor, as POE has a Wx backend that simply polls 20 times per |
705 | second, which was considered to be too horrible to even consider for |
711 | second, which was considered to be too horrible to even consider for |
706 | AnyEvent. Likewise, other POE backends can be used by AnyEvent by |
712 | AnyEvent. Likewise, other POE backends can be used by AnyEvent by |
… | |
… | |
906 | "PERL_ANYEVENT_STRICT" |
912 | "PERL_ANYEVENT_STRICT" |
907 | AnyEvent does not do much argument checking by default, as thorough |
913 | AnyEvent does not do much argument checking by default, as thorough |
908 | argument checking is very costly. Setting this variable to a true |
914 | argument checking is very costly. Setting this variable to a true |
909 | value will cause AnyEvent to load "AnyEvent::Strict" and then to |
915 | value will cause AnyEvent to load "AnyEvent::Strict" and then to |
910 | thoroughly check the arguments passed to most method calls. If it |
916 | thoroughly check the arguments passed to most method calls. If it |
911 | finds any problems it will croak. |
917 | finds any problems, it will croak. |
912 | |
918 | |
913 | In other words, enables "strict" mode. |
919 | In other words, enables "strict" mode. |
914 | |
920 | |
915 | Unlike "use strict", it is definitely recommended ot keep it off in |
921 | Unlike "use strict", it is definitely recommended to keep it off in |
916 | production. Keeping "PERL_ANYEVENT_STRICT=1" in your environment |
922 | production. Keeping "PERL_ANYEVENT_STRICT=1" in your environment |
917 | while developing programs can be very useful, however. |
923 | while developing programs can be very useful, however. |
918 | |
924 | |
919 | "PERL_ANYEVENT_MODEL" |
925 | "PERL_ANYEVENT_MODEL" |
920 | This can be used to specify the event model to be used by AnyEvent, |
926 | This can be used to specify the event model to be used by AnyEvent, |
… | |
… | |
1208 | EV/Any 100000 224 2.88 0.34 0.27 EV + AnyEvent watchers |
1214 | EV/Any 100000 224 2.88 0.34 0.27 EV + AnyEvent watchers |
1209 | CoroEV/Any 100000 224 2.85 0.35 0.28 coroutines + Coro::Signal |
1215 | CoroEV/Any 100000 224 2.85 0.35 0.28 coroutines + Coro::Signal |
1210 | Perl/Any 100000 452 4.13 0.73 0.95 pure perl implementation |
1216 | Perl/Any 100000 452 4.13 0.73 0.95 pure perl implementation |
1211 | Event/Event 16000 517 32.20 31.80 0.81 Event native interface |
1217 | Event/Event 16000 517 32.20 31.80 0.81 Event native interface |
1212 | Event/Any 16000 590 35.85 31.55 1.06 Event + AnyEvent watchers |
1218 | Event/Any 16000 590 35.85 31.55 1.06 Event + AnyEvent watchers |
|
|
1219 | IOAsync/Any 16000 989 38.10 32.77 11.13 via IO::Async::Loop::IO_Poll |
|
|
1220 | IOAsync/Any 16000 990 37.59 29.50 10.61 via IO::Async::Loop::Epoll |
1213 | Glib/Any 16000 1357 102.33 12.31 51.00 quadratic behaviour |
1221 | Glib/Any 16000 1357 102.33 12.31 51.00 quadratic behaviour |
1214 | Tk/Any 2000 1860 27.20 66.31 14.00 SEGV with >> 2000 watchers |
1222 | Tk/Any 2000 1860 27.20 66.31 14.00 SEGV with >> 2000 watchers |
1215 | POE/Event 2000 6328 109.99 751.67 14.02 via POE::Loop::Event |
1223 | POE/Event 2000 6328 109.99 751.67 14.02 via POE::Loop::Event |
1216 | POE/Select 2000 6027 94.54 809.13 579.80 via POE::Loop::Select |
1224 | POE/Select 2000 6027 94.54 809.13 579.80 via POE::Loop::Select |
1217 | |
1225 | |
… | |
… | |
1246 | few of them active), of course, but this was not subject of this |
1254 | few of them active), of course, but this was not subject of this |
1247 | benchmark. |
1255 | benchmark. |
1248 | |
1256 | |
1249 | The "Event" module has a relatively high setup and callback invocation |
1257 | The "Event" module has a relatively high setup and callback invocation |
1250 | cost, but overall scores in on the third place. |
1258 | cost, but overall scores in on the third place. |
|
|
1259 | |
|
|
1260 | "IO::Async" performs admirably well, about on par with "Event", even |
|
|
1261 | when using its pure perl backend. |
1251 | |
1262 | |
1252 | "Glib"'s memory usage is quite a bit higher, but it features a faster |
1263 | "Glib"'s memory usage is quite a bit higher, but it features a faster |
1253 | callback invocation and overall ends up in the same class as "Event". |
1264 | callback invocation and overall ends up in the same class as "Event". |
1254 | However, Glib scales extremely badly, doubling the number of watchers |
1265 | However, Glib scales extremely badly, doubling the number of watchers |
1255 | increases the processing time by more than a factor of four, making it |
1266 | increases the processing time by more than a factor of four, making it |
… | |
… | |
1326 | single "request", that is, reading the token from the pipe and |
1337 | single "request", that is, reading the token from the pipe and |
1327 | forwarding it to another server. This includes deleting the old timeout |
1338 | forwarding it to another server. This includes deleting the old timeout |
1328 | and creating a new one that moves the timeout into the future. |
1339 | and creating a new one that moves the timeout into the future. |
1329 | |
1340 | |
1330 | Results |
1341 | Results |
1331 | name sockets create request |
1342 | name sockets create request |
1332 | EV 20000 69.01 11.16 |
1343 | EV 20000 69.01 11.16 |
1333 | Perl 20000 73.32 35.87 |
1344 | Perl 20000 73.32 35.87 |
|
|
1345 | IOAsync 20000 157.00 98.14 epoll |
|
|
1346 | IOAsync 20000 159.31 616.06 poll |
1334 | Event 20000 212.62 257.32 |
1347 | Event 20000 212.62 257.32 |
1335 | Glib 20000 651.16 1896.30 |
1348 | Glib 20000 651.16 1896.30 |
1336 | POE 20000 349.67 12317.24 uses POE::Loop::Event |
1349 | POE 20000 349.67 12317.24 uses POE::Loop::Event |
1337 | |
1350 | |
1338 | Discussion |
1351 | Discussion |
1339 | This benchmark *does* measure scalability and overall performance of the |
1352 | This benchmark *does* measure scalability and overall performance of the |
1340 | particular event loop. |
1353 | particular event loop. |
1341 | |
1354 | |
1342 | EV is again fastest. Since it is using epoll on my system, the setup |
1355 | EV is again fastest. Since it is using epoll on my system, the setup |
1343 | time is relatively high, though. |
1356 | time is relatively high, though. |
1344 | |
1357 | |
1345 | Perl surprisingly comes second. It is much faster than the C-based event |
1358 | Perl surprisingly comes second. It is much faster than the C-based event |
1346 | loops Event and Glib. |
1359 | loops Event and Glib. |
|
|
1360 | |
|
|
1361 | IO::Async performs very well when using its epoll backend, and still |
|
|
1362 | quite good compared to Glib when using its pure perl backend. |
1347 | |
1363 | |
1348 | Event suffers from high setup time as well (look at its code and you |
1364 | Event suffers from high setup time as well (look at its code and you |
1349 | will understand why). Callback invocation also has a high overhead |
1365 | will understand why). Callback invocation also has a high overhead |
1350 | compared to the "$_->() for .."-style loop that the Perl event loop |
1366 | compared to the "$_->() for .."-style loop that the Perl event loop |
1351 | uses. Event uses select or poll in basically all documented |
1367 | uses. Event uses select or poll in basically all documented |
… | |
… | |
1407 | THE IO::Lambda BENCHMARK |
1423 | THE IO::Lambda BENCHMARK |
1408 | Recently I was told about the benchmark in the IO::Lambda manpage, which |
1424 | Recently I was told about the benchmark in the IO::Lambda manpage, which |
1409 | could be misinterpreted to make AnyEvent look bad. In fact, the |
1425 | could be misinterpreted to make AnyEvent look bad. In fact, the |
1410 | benchmark simply compares IO::Lambda with POE, and IO::Lambda looks |
1426 | benchmark simply compares IO::Lambda with POE, and IO::Lambda looks |
1411 | better (which shouldn't come as a surprise to anybody). As such, the |
1427 | better (which shouldn't come as a surprise to anybody). As such, the |
1412 | benchmark is fine, and shows that the AnyEvent backend from IO::Lambda |
1428 | benchmark is fine, and mostly shows that the AnyEvent backend from |
1413 | isn't very optimal. But how would AnyEvent compare when used without the |
1429 | IO::Lambda isn't very optimal. But how would AnyEvent compare when used |
1414 | extra baggage? To explore this, I wrote the equivalent benchmark for |
1430 | without the extra baggage? To explore this, I wrote the equivalent |
1415 | AnyEvent. |
1431 | benchmark for AnyEvent. |
1416 | |
1432 | |
1417 | The benchmark itself creates an echo-server, and then, for 500 times, |
1433 | The benchmark itself creates an echo-server, and then, for 500 times, |
1418 | connects to the echo server, sends a line, waits for the reply, and then |
1434 | connects to the echo server, sends a line, waits for the reply, and then |
1419 | creates the next connection. This is a rather bad benchmark, as it |
1435 | creates the next connection. This is a rather bad benchmark, as it |
1420 | doesn't test the efficiency of the framework, but it is a benchmark |
1436 | doesn't test the efficiency of the framework or much non-blocking I/O, |
1421 | nevertheless. |
1437 | but it is a benchmark nevertheless. |
1422 | |
1438 | |
1423 | name runtime |
1439 | name runtime |
1424 | Lambda/select 0.330 sec |
1440 | Lambda/select 0.330 sec |
1425 | + optimized 0.122 sec |
1441 | + optimized 0.122 sec |
1426 | Lambda/AnyEvent 0.327 sec |
1442 | Lambda/AnyEvent 0.327 sec |
… | |
… | |
1432 | |
1448 | |
1433 | AnyEvent/select/nb 0.085 sec |
1449 | AnyEvent/select/nb 0.085 sec |
1434 | AnyEvent/EV/nb 0.068 sec |
1450 | AnyEvent/EV/nb 0.068 sec |
1435 | +state machine 0.134 sec |
1451 | +state machine 0.134 sec |
1436 | |
1452 | |
1437 | The benchmark is also a bit unfair (my fault) - the IO::Lambda |
1453 | The benchmark is also a bit unfair (my fault): the IO::Lambda/POE |
1438 | benchmarks actually make blocking connects and use 100% blocking I/O, |
1454 | benchmarks actually make blocking connects and use 100% blocking I/O, |
1439 | defeating the purpose of an event-based solution. All of the newly |
1455 | defeating the purpose of an event-based solution. All of the newly |
1440 | written AnyEvent benchmarks use 100% non-blocking connects (using |
1456 | written AnyEvent benchmarks use 100% non-blocking connects (using |
1441 | AnyEvent::Socket::tcp_connect and the asynchronous pure perl DNS |
1457 | AnyEvent::Socket::tcp_connect and the asynchronous pure perl DNS |
1442 | resolver), so AnyEvent is at a disadvantage here as non-blocking |
1458 | resolver), so AnyEvent is at a disadvantage here, as non-blocking |
1443 | connects generally require a lot more bookkeeping and event handling |
1459 | connects generally require a lot more bookkeeping and event handling |
1444 | than blocking connects (which involve a single syscall only). |
1460 | than blocking connects (which involve a single syscall only). |
1445 | |
1461 | |
1446 | The last AnyEvent benchmark additionally uses AnyEvent::Handle, which |
1462 | The last AnyEvent benchmark additionally uses AnyEvent::Handle, which |
1447 | offers similar expressive power as POE and IO::Lambda (using |
1463 | offers similar expressive power as POE and IO::Lambda, using |
1448 | conventional Perl syntax), which means both the echo server and the |
1464 | conventional Perl syntax. This means that both the echo server and the |
1449 | client are 100% non-blocking w.r.t. I/O, further placing it at a |
1465 | client are 100% non-blocking, further placing it at a disadvantage. |
1450 | disadvantage. |
|
|
1451 | |
1466 | |
1452 | As you can see, AnyEvent + EV even beats the hand-optimised "raw sockets |
1467 | As you can see, the AnyEvent + EV combination even beats the |
1453 | benchmark", while AnyEvent + its pure perl backend easily beats |
1468 | hand-optimised "raw sockets benchmark", while AnyEvent + its pure perl |
1454 | IO::Lambda and POE. |
1469 | backend easily beats IO::Lambda and POE. |
1455 | |
1470 | |
1456 | And even the 100% non-blocking version written using the high-level (and |
1471 | And even the 100% non-blocking version written using the high-level (and |
1457 | slow :) AnyEvent::Handle abstraction beats both POE and IO::Lambda, even |
1472 | slow :) AnyEvent::Handle abstraction beats both POE and IO::Lambda by a |
1458 | thought it does all of DNS, tcp-connect and socket I/O in a non-blocking |
1473 | large margin, even though it does all of DNS, tcp-connect and socket I/O |
1459 | way. |
1474 | in a non-blocking way. |
1460 | |
1475 | |
1461 | The two AnyEvent benchmarks can be found as eg/ae0.pl and eg/ae2.pl in |
1476 | The two AnyEvent benchmarks programs can be found as eg/ae0.pl and |
1462 | the AnyEvent distribution, the remaining benchmarks are part of the |
1477 | eg/ae2.pl in the AnyEvent distribution, the remaining benchmarks are |
1463 | IO::lambda distribution and were used without any changes. |
1478 | part of the IO::lambda distribution and were used without any changes. |
1464 | |
1479 | |
1465 | SIGNALS |
1480 | SIGNALS |
1466 | AnyEvent currently installs handlers for these signals: |
1481 | AnyEvent currently installs handlers for these signals: |
1467 | |
1482 | |
1468 | SIGCHLD |
1483 | SIGCHLD |
1469 | A handler for "SIGCHLD" is installed by AnyEvent's child watcher |
1484 | A handler for "SIGCHLD" is installed by AnyEvent's child watcher |
1470 | emulation for event loops that do not support them natively. Also, |
1485 | emulation for event loops that do not support them natively. Also, |
1471 | some event loops install a similar handler. |
1486 | some event loops install a similar handler. |
|
|
1487 | |
|
|
1488 | If, when AnyEvent is loaded, SIGCHLD is set to IGNORE, then AnyEvent |
|
|
1489 | will reset it to default, to avoid losing child exit statuses. |
1472 | |
1490 | |
1473 | SIGPIPE |
1491 | SIGPIPE |
1474 | A no-op handler is installed for "SIGPIPE" when $SIG{PIPE} is |
1492 | A no-op handler is installed for "SIGPIPE" when $SIG{PIPE} is |
1475 | "undef" when AnyEvent gets loaded. |
1493 | "undef" when AnyEvent gets loaded. |
1476 | |
1494 | |
… | |
… | |
1512 | Similar considerations apply to $ENV{PERL_ANYEVENT_VERBOSE}, as that can |
1530 | Similar considerations apply to $ENV{PERL_ANYEVENT_VERBOSE}, as that can |
1513 | be used to probe what backend is used and gain other information (which |
1531 | be used to probe what backend is used and gain other information (which |
1514 | is probably even less useful to an attacker than PERL_ANYEVENT_MODEL), |
1532 | is probably even less useful to an attacker than PERL_ANYEVENT_MODEL), |
1515 | and $ENV{PERL_ANYEVENT_STRICT}. |
1533 | and $ENV{PERL_ANYEVENT_STRICT}. |
1516 | |
1534 | |
|
|
1535 | Note that AnyEvent will remove *all* environment variables starting with |
|
|
1536 | "PERL_ANYEVENT_" from %ENV when it is loaded while taint mode is |
|
|
1537 | enabled. |
|
|
1538 | |
1517 | BUGS |
1539 | BUGS |
1518 | Perl 5.8 has numerous memleaks that sometimes hit this module and are |
1540 | Perl 5.8 has numerous memleaks that sometimes hit this module and are |
1519 | hard to work around. If you suffer from memleaks, first upgrade to Perl |
1541 | hard to work around. If you suffer from memleaks, first upgrade to Perl |
1520 | 5.10 and check wether the leaks still show up. (Perl 5.10.0 has other |
1542 | 5.10 and check wether the leaks still show up. (Perl 5.10.0 has other |
1521 | annoying memleaks, such as leaking on "map" and "grep" but it is usually |
1543 | annoying memleaks, such as leaking on "map" and "grep" but it is usually |