在 PHP 内部，系统维护着连接状态，其状态有三种可能的情况：
当 PHP 脚本正常地运行 NORMAL 状态时，连接为有效。当远程客户端中断连接时，ABORTED 状态的标记将会被打开。远程客户端连接的中断通常是由用户点击 STOP 按钮导致的。当连接时间超过 PHP 的时限（请参阅 set_time_limit() 函数）时，TIMEOUT 状态的标记将被打开。
可以决定脚本是否需要在客户端中断连接时退出。有时候让脚本完整地运行会带来很多方便，即使没有远程浏览器接受脚本的输出。默认的情况是当远程客户端连接中断时脚本将会退出。该处理过程可由 php.ini 的 ignore_user_abort 或由 httpd.conf 设置中对应的"php_value ignore_user_abort"以及 ignore_user_abort() 函数来控制。如果没有告诉 PHP 忽略用户的中断，脚本将会被中断，除非通过 register_shutdown_function() 设置了关闭触发函数。通过该关闭触发函数，当远程用户点击 STOP 按钮后，脚本再次尝试输出数据时，PHP 将会检测到连接已被中断，并调用关闭触发函数。
脚本也有可能被内置的脚本计时器中断。默认的超时限制为 30 秒。这个值可以通过设置 php.ini 的 max_execution_time 或 httpd.conf 设置中对应的"php_value max_execution_time"参数或者 set_time_limit() 函数来更改。当计数器超时的时候，脚本将会类似于以上连接中断的情况退出，先前被注册过的关闭触发函数也将在这时被执行。在该关闭触发函数中，可以通过调用 connection_status() 函数来检查超时是否导致关闭触发函数被调用。如果超时导致了关闭触发函数的调用，该函数将返回 2。
需要注意的一点是 ABORTED 和 TIMEOUT 状态可以同时有效。这在告诉 PHP 忽略用户的退出操作时是可能的。PHP 将仍然注意用户已经中断了连接但脚本仍然在运行的情况。如果到了运行的时间限制，脚本将被退出，设置过的关闭触发函数也将被执行。在这时会发现函数 connection_status() 返回 3。
The CONNECTION_XXX constants that are not listed here for some reason are:
0 = CONNECTION_NORMAL
1 = CONNECTION_ABORTED
2 = CONNECTION_TIMEOUT
3 = CONNECTION_ABORTED & CONNECTION_TIMEOUT
Number 3 is effectively tested like this:
if (CONNECTION_ABORTED & CONNECTION_TIMEOUT)
echo 'Connection both aborted and timed out';
01
02
03
04
05
06
07
08
09
10
11
12
13
14
15
16
17
18
/*
* Anti-Pattern
*/
# Connect
mysql_connect('localhost', 'username', 'password') or die('Could not connect: ' . mysql_error());
# Choose a database
mysql_select_db('someDatabase') or die('Could not select database');
# Perform database query
$query = "SELECT * from someTable";
$result = mysql_query($query) or die('Query failed: ' . mysql_error());
# Filter through rows and echo desired information
while ($row = mysql_fetch_object($result)) {
echo $row->name;
}
I had a lot of problems getting a redirect to work, after which my script was intended to keep working in the background. The redirect to another page of my site simply would only work once the original page had finished processing.
I finally found out what was wrong:
The session only gets closed by PHP at the very end of the script, and since access to the session data is locked to prevent more than one page writing to it simultaneously, the new page cannot load until the original processing has finished.
Solution:
Close the session manually when redirecting using session_write_close():
<?php
ignore_user_abort(true);
set_time_limit(0);
$strURL = "PUT YOUR REDIRCT HERE";
header("Location: $strURL", true);
header("Connection: close", true);
header("Content-Encoding: none\r\n");
header("Content-Length: 0", true);
flush();
ob_flush();
session_write_close();
// Continue processing...
sleep(100);
exit;
?>
But careful:
Make sure that your script doesn't write to the session after session_write_close(), i.e. in your background processing code. That won't work. Also avoid reading, remember, the next script may already have modified the data.
So try to read out the data you need prior to redirecting.
I was quite stuck when trying to make my script redirect the client to another URL and then continue processing. The reason was php-fpm. All possible buffer flushes did not work, unless I called fastcgi_finish_request();
For example:
<?php
// redirecting...
ignore_user_abort(true);
header("Location: ".$redirectUrl, true);
header("Connection: close", true);
header("Content-Length: 0", true);
ob_end_flush();
flush();
fastcgi_finish_request(); // important when using php-fpm!
sleep (5); // User won't feel this sleep because he'll already be away
// do some work after user has been redirected
?>
This simple function outputs a string and closes the connection. It considers compression using "ob_gzhandler"
It took me a little while to put this all together, mostly because setting the encoding to none, as some people noted here, didn't work.
<?php
function outputStringAndCloseConnection2($stringToOutput)
{
set_time_limit(0);
ignore_user_abort(true);
// buffer all upcoming output - make sure we care about compression:
if(!ob_start("ob_gzhandler"))
ob_start();
echo $stringToOutput;
// get the size of the output
$size = ob_get_length();
// send headers to tell the browser to close the connection
header("Content-Length: $size");
header('Connection: close');
// flush all output
ob_end_flush();
ob_flush();
flush();
// close current session
if (session_id()) session_write_close();
}
?>
PHP changes directory on connection abort so code like this will not do what you want:
<?php
function abort()
{
if(connection_aborted())
unlink('file.ini');
}
register_shutdown_function('abort');
?>
actually it will delete file in apaches's root dir so if you want to unlink file in your script's dir on abort or write to it you have to store directory
<?php
function abort()
{
global $dsd;
if(connection_aborted())
unlink($dsd.'/file.ini');
}
register_shutdown_function('abort');
$dsd=getcwd();
?>
hey, thanks to arr1, and it is very useful for me, when I need to return to the user fast and then do something else.
When using the codes, it nearly drive me mad and I found another thing that may affect the codes:
Content-Encoding: gzip
This is because the zlib is on and the content will be compressed. But this will not output the buffer until all output is over.
So, it may need to send the header to prevent this problem.
now, the code becomes:
<?php
ob_end_clean();
header("Connection: close\r\n");
header("Content-Encoding: none\r\n");
ignore_user_abort(true); // optional
ob_start();
echo ('Text user will see');
$size = ob_get_length();
header("Content-Length: $size");
ob_end_flush(); // Strange behaviour, will not work
flush(); // Unless both are called !
ob_end_clean();
//do processing here
sleep(5);
echo('Text user will never see');
//do some processing
?>
connection_status() return ABORTED state ONLY if the client disconnects gracefully (with STOP button). In this case the browser send the RST TCP packet that notify PHP the connection is closed.
But.... If the connection is stopped by networs troubles (wifi link down by exemple) the script doesn't know that the client is disconnected :(
I've tried to use fopen("php://output") with stream_select() on writting to detect write locks (due to full buffer) but php give me this error : "cannot represent a stream of type Output as a select()able descriptor"
So I don't know how to detect correctly network trouble connection...
in regards of posting from:
arr1 at hotmail dot co dot uk
if you use/write sessions you need to do this before:
(otherwise it does not work)
session_write_close();
and if wanted:
ignore_user_abort(TRUE);
instead of ignore_user_abort();
Closing the users browser connection whilst keeping your php script running has been an issue since 4.1, when the behaviour of register_shutdown_function() was modified so that it would not automatically close the users connection.
sts at mail dot xubion dot hu
Posted the original solution:
<?php
header("Connection: close");
ob_start();
phpinfo();
$size=ob_get_length();
header("Content-Length: $size");
ob_end_flush();
flush();
sleep(13);
error_log("do something in the background");
?>
Which works fine until you substitute phpinfo() for
echo ('text I want user to see'); in which case the headers are never sent!
The solution is to explicitly turn off output buffering and clear the buffer prior to sending your header information.
example:
<?php
ob_end_clean();
header("Connection: close");
ignore_user_abort(); // optional
ob_start();
echo ('Text the user will see');
$size = ob_get_length();
header("Content-Length: $size");
ob_end_flush(); // Strange behaviour, will not work
flush(); // Unless both are called !
// Do processing here
sleep(30);
echo('Text user will never see');
?>
Just spent 3 hours trying to figure this one out, hope it helps someone :)
Tested in:
IE 7.5730.11
Mozilla Firefox 1.81
The point mentioned in the last comment isn't always the case.
If a user's connection is lost half way through an order processing script is confirming a user's credit card/adding them to a DB, etc (due to their ISP going down, network trouble... whatever) and your script tries to send back output (such as, "pre-processing order" or any other type of confirmation), then your script will abort -- and this could cause problems for your process.
I have an order script that adds data to a InnoDB database (through MySQL) and only commits the transactions upon successful completion. Without ignore_user_abort(), I have had times when a user's connection dropped during the processing phase... and their card was charged, but they weren't added to my local DB.
So, it's always safe to ignore any aborts if you are processing sensitive transactions that should go ahead, whether your user is "watching" on the other end or not.
These functions are very useful for example if you need to control when a visitor in your website place an order and you need to check if he/she didn't clicked the submit button twice or cancelled the submit just after have clicked the submit button.
If your visitor click the stop button just after have submitted it, your script may stop in the middle of the process of registering the products and do not finish the list, generating inconsistency in your database.
With the ignore_user_abort() function you can make your script finish everything fine and after you can check with register_shutdown_function() and connection_aborted() if the visitor cancelled the submission or lost his/her connection. If he/she did, you can set the order as not confirmed and when the visitor came back, you can present the old order again.
To prevent a double click of the submit button, you can disable it with javascript or in your script you can set a flag for that order, which will be recorded into the database. Before accept a new submission, the script will check if the same order was not placed before and reject it. This will work fine, as the script have finished the job before.
Note that if you use ob_start("callback_function") in the begin of your script, you can specify a callback function that will act like the shutdown function when our script ends and also will let you to work on the generated page before send it to the visitor.