php_http_parserPHP的HTTP解析擴展
php_http_parser 是基于node.js http-parser的PHP擴展，可用于實現(xiàn)純異步PHP程序
libcurl提供了異步調(diào)用方式，有兩種風格：
-
ONE MULTI HANDLE MANY EASY HANDLES：加入多個easy handle后執(zhí)行curl_multi_perform方法。該方法在php curl擴展中有對應實現(xiàn)。但最后一步curl_multi_perform是阻塞的。
-
MULTI_SOCKET，這個是真正的非阻塞方法，但需要自行實現(xiàn)event loop，且封裝較為困難，目前在php中沒有對應實現(xiàn)。經(jīng)過調(diào)研，curl_multi_socket_action跟php內(nèi)核結(jié)合困難度很高。
除此之外，基本上沒有真正的實現(xiàn)異步http請求的php擴展。目前僅有部分純php實現(xiàn)的版本，比如tsf中的http client實現(xiàn)。 使用純php實現(xiàn)的問題主要受限于http解析的性能。因此考慮將這一模塊用擴展的方式來實現(xiàn)。node.js http-parser就是一個很好的c語言的http解析庫。 php_http_parser就是對其做的一個封裝，在php中暴露出相應的接口。
為了實現(xiàn)真正的非阻塞請求，仍然需要自己實現(xiàn)event loop。目前推薦結(jié)合swoole使用，以獲得更好的性能。
使用方式
$buffs = array("HTTP/1.1 301 Moved Permanently\r\n"
,"Location: http://www.google.com/\r\n"
,"Content-Type: text/html; charset=UTF-8\r\n"
,"Date: Sun, 26 Apr 2009 11:11:49 GMT\r\n"
,"Expires: Tue, 26 May 2009 11:11:49 GMT\r\n"
,"Cache-Control: public, max-age=2592000\r\n"
,"Server: gws\r\n"
,"Content-Length: 193\r\n"
,"\r\n"
,"<HTML><HEAD><meta http-equiv=\"content-type\" content=\"text/html;charset=utf-8\">\n"
,"<TITLE>301 Moved</TITLE></HEAD><BODY>\n"
,"<H1>301 Moved</H1>\n"
,"The document has moved\n"
,"<A HREF=\"http://www.google.com/\">here</A>.\r\n"
,"<A HREF=\"http://www.google.com/\">here</A>.\r\n"
,"<A HREF=\"http://www.google.com/\">here</A>.\r\n"
,"<A HREF=\"http://www.google.com/\">here</A>.\r\n"
,"<A HREF=\"http://www.google.com/\">here</A>.\r\n"
,"</BODY></HTML>\r\n");
$hp = new HttpParser();
foreach($buffs as $buff){
$ret = $hp->execute($buff);
if($ret !== false){
echo $ret;
break;
}
}
雖然http請求可能分包發(fā)送，HttpParser會將所有包合并在一起后，出發(fā)body事件，然后調(diào)用相應的回調(diào)方法。 諸如header回調(diào)，目前暫未實現(xiàn)。另外，此處需要自行實現(xiàn)timeout邏輯。
示例代碼是結(jié)合swoole_client與swPromise框架實現(xiàn)的一個異步http client。 籍此可以實現(xiàn)真正的非阻塞的PHP程序。
class HttpClientFuture implements FutureIntf {
protected $url = null;
protected $post = null;
protected $proxy = false;
public function __construct($url, $post = array(), $proxy = array()) {
$this->url = $url;
$this->post = $post;
if($proxy){
$this->proxy = $proxy;
}
}
public function run(Promise &$promise) {
$cli = new \swoole_client ( SWOOLE_TCP, SWOOLE_SOCK_ASYNC );
$urlInfo = parse_url ( $this->url );
if(!isset($urlInfo ['port']))$urlInfo ['port'] = 80;
$httpParser = new \HttpParser();
$cli->on ( "connect", function ($cli)use($urlInfo){
$host = $urlInfo['host'];
if($urlInfo['port'])$host .= ':'.$urlInfo['port'];
$req = array();
$req[] = "GET {$this->url} HTTP/1.1\r\n";
$req[] = "User-Agent: PHP swAsync\r\n";
$req[] = "Host:{$host}\r\n";
$req[] = "Connection:close\r\n";
$req[] = "\r\n";
$req = implode('', $req);
$cli->send ( $req );
} );
$cli->on ( "receive", function ($cli, $data = "") use(&$httpParser, &$promise) {
$ret = $httpParser->execute($data);
if($ret !== false){
$cli->close();
$promise->accept(['http_data'=>$ret]);
}
} );
$cli->on ( "error", function ($cli) use(&$promise) {
$promise->reject ();
} );
$cli->on ( "close", function ($cli) {
} );
if($this->proxy){
$cli->connect ( $this->proxy['host'], $this->proxy ['port'], 1 );
}else{
$cli->connect ( $urlInfo ['host'], $urlInfo ['port'], 1 );
}
}
}
性能
[web@gz-web01 php_http_parser]$ time /data/server/php/bin/php http_parser.php 2000000 real 0m11.489s user 0m11.435s sys 0m0.017s
1個worker進程
./http_load -fetches 20000 -parallel 100 9502.listasync 20000 fetches, 100 max parallel, 2.02e+06 bytes, in 5.94536 seconds 101 mean bytes/connection 3363.97 fetches/sec, 339761 bytes/sec msecs/connect: 0.0473873 mean, 1.155 max, 0.019 min msecs/first-response: 29.6366 mean, 51.736 max, 15.22 min HTTP response codes: code 200 -- 20000 -bash: history: write error: Success
2個worker進程
./http_load -fetches 20000 -parallel 100 9502.listasync 20000 fetches, 100 max parallel, 2.02e+06 bytes, in 3.17119 seconds 101 mean bytes/connection 6306.77 fetches/sec, 636984 bytes/sec msecs/connect: 0.0643583 mean, 1.211 max, 0.023 min msecs/first-response: 15.7489 mean, 32.425 max, 3.242 min HTTP response codes: code 200 -- 20000 -bash: history: write error: Success
4個woker進程
./http_load -fetches 20000 -parallel 100 9502.listasync 20000 fetches, 100 max parallel, 2.02e+06 bytes, in 1.57194 seconds 101 mean bytes/connection 12723.2 fetches/sec, 1.28504e+06 bytes/sec msecs/connect: 0.0815263 mean, 1.349 max, 0.02 min msecs/first-response: 7.65904 mean, 22.568 max, 1.221 min HTTP response codes: code 200 -- 20000 -bash: history: write error: Success
8個woker進程
./http_load -fetches 20000 -parallel 100 9502.listasync 20000 fetches, 100 max parallel, 2.02e+06 bytes, in 1.02967 seconds 101 mean bytes/connection 19423.8 fetches/sec, 1.9618e+06 bytes/sec msecs/connect: 0.147502 mean, 1.575 max, 0.014 min msecs/first-response: 3.17218 mean, 22.566 max, 0.339 min HTTP response codes: code 200 -- 20000 -bash: history: write error: Success
評論
圖片
表情