Re: On register_shutdown_function: What might be the problem?

Thank you for replying. Sorry for being long again. I tried your suggestion of this: <?php ignore_user_abort(true); header("Location: redirect2.html"); echo "foo\n"; flush(); for($i=0;$i<10;$i++) { echo $i; sleep(1); } $fp = fopen("/tmp/foo.txt","a"); fputs($fp,$i); fclose($fp); ?> The browser did not get the redirect page until the new $i(10) has been appended to file foo.txt. I also tried ----------------------------------------------- ob_start(); header("Location: http://liang.ns2user.info/php/y.html"); echo "foo\n"; ob_end_flush(); ----------------------------------------------- And does not work. I think it might be the case of what you suggest:

If the client sticks around after seeing the redirect and doesn't redirect until after the server has closed the connection, then there is no way to force a close.

So may I draw the conclusions: 1. Server in php can not actively close the connection to make the client see the redirect page immediately after the header has been set? 2. The connection is not closed even when the main php end and the shutdown function is running? If above conclusions are not true, could be somewhere that might be wrong due to the apache-php configuration or what? If those are true, I think I have to re-design the project flow control and it will take sometime while I have already finished the code assuming no above problem. Also I think this might result an awkward design for the project. (I mean jsp will be easier to achieve a relatively more elegant design). I use several other program langurages for many years, but new to php, so I probably made some assumption base on my experiences of other langurages but not true. The project I am working now to add enhancements is written on php. I try to keep the main design of it. I did not expect it would be difficult to achieve this: 1. write an xml page 2. let http client get this xml page in time 3. continue do something time consuming now I have difficulty in 2 and 3. My client is normally perl LWP::UserAgent, and sometimes browser. I wonder how to let the client get the page in time before I can continue to finish the rest of the work. Otherwise the client will timeout and get no result. Please help! Thank you. Liang

I really didn't follow all that. But this stuff is not that complex. header() sets a header to be sent when output goes out. header() does not send an actual header at that point. If you don't send any output, then the headers won't go out until the script terminates. If you have any sort of output buffering enabled, then even if you think you are sending something, you may only be buffering it, and in that case again the headers won't go out until the request is done. Also, ignore_user_abort() controls whether or not your script will be terminated when we are able to detect that the user has aborted. A user abort is defined as nobody being around to read the data we are sending out. If you don't send anything, we can't detect if the browser has gone away. Generally browsers will redirect as soon as they see a Location: redirect header, but that could be client-speficic. If the client sticks around after seeing the redirect and doesn't redirect until after the server has closed the connection, then there is no way to force a close. If it does redirect and close its end of the connection, then if you set ignore_user_abort(false) your script will be terminated as soon as it tries to send something further and your shutdown function will be called at that point. -Rasmus Liang ZHONG wrote: Hi Rasmus, This may be a little bit long, sorry for taking your time. It still does not work as expected. I tried some experiment, and found that if I called some function or write some code line other then calling header(), the register_shutdown_function and other part of codes work as expected . For example: <?php set_time_limit(5); function f(){ set_time_limit(10); //doing something time consuming } some_function(); ?> The time limit of 5 will be the limit of the some_function() and the 10 will be the limit of function f() respectively. Code example: ----------------------------------------------------------------------------------------------------------------------- <?php set_time_limit(1); ignore_user_abort(true); function say_goodbye() {

       $st = connection_status();
       print "Status 1: ".$st."\n";
       set_time_limit(10);
       $st = connection_status();
       print "Status 2: ".$st."\n";

       $count=20000000;
       for($i=0; $i<$count; $i++){    }
           print "End!\n";
           exec("touch /home/.nappy/liang/liang.ns2user.info/php/bbb");

} register_shutdown_function("say_goodbye"); print "Sleeping...\n"; $count=10000000;

   for($i=0; $i<$count; $i++){    }

print "Done!\n"; ?> ------------------------------------------------------------------------------------------------------------------------ -bash-2.05b$ curl -N liang.ns2user.info/php/v.php Sleeping... <br /> <b>Fatal error</b>: Maximum execution time of 1 second exceeded in <b>/home/.nappy/liang/liang.ns2user.info/php/v.php</b> on line <b>30</b><br /> Status 1: 2 Status 2: 2 End! ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ if I change the time limit from 10 to 5 in function f() -bash-2.05b$ curl -N liang.ns2user.info/php/v.php Sleeping... <br /> <b>Fatal error</b>: Maximum execution time of 1 second exceeded in <b>/home/.nappy/liang/liang.ns2user.info/php/v.php</b> on line <b>30</b><br /> Status 1: 2 Status 2: 2 <br /> <b>Fatal error</b>: Maximum execution time of 5 seconds exceeded in <b>/home/.nappy/liang/liang.ns2user.info/php/v.php</b> on line <b>14</b><br /> ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Change both the time limit to 10 will result: -bash-2.05b$ curl -N liang.ns2user.info/php/v.php Sleeping... Done! Status 1: 0 Status 2: 0 End! ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ But if the some_function is header() then the rule above does not work. It seems the function f()'s time limit also rule the header(). Only after the registered shutdown function finishes runing normally or by hit the expire time limit, will the header() return page to browser/http user agent. Example code with suggestion from Rasmus as: ------------------------------------------------------------------------------------------------------------------ <?php set_time_limit(5); function f(){ set_time_limit(100); $count=500000000; for($i=0; $i<$count; $i++){ //sit here and loop for a bit so we can have time to hit Stop... echo "a \n"; flush(); } echo "end"; exec("touch /tmp/aaa"); } register_shutdown_function('f'); ignore_user_abort(true); header("Content-type: text/plain"); header("Location: y.html"); echo "foo\n"; flush(); for($i=0;$i<10;$i++) { echo $i; sleep(1); } $fp = fopen("/tmp/foo.txt","a"); fputs($fp,$i); fclose($fp); ?> ------------------------------------------------------------------------------------------------------------------ After the file /tmp/foo.txt has been created, before the file /tmp/aaa being created, the y.html will not get to the browser or perl program using LWP::UserAgent. I think it is no way to close the connection actively by the php program to deliver the page sooner to end the http request, even I add lines and make code like this does not work: ---------------------------------------------------------------- header("Content-type: text/plain"); header("Cache-Control: no-store, no-cache"); header("Location: y.html"); header("Connection: close"); ---------------------------------------------------------------- My project act as a broker for user agent to a library database server called z39.50. Upon the request from user agent, my program need to connect a z39.50 server, getting data back 1 by 1, transforming to sepcial xml format and sending back to the request party. When the data repository is huge (sometimes up to million records), I have to get partial data transform and send back to user agent (normally a piece of perl code, called harvester) with a resumption token. The harvester will in a while loop send out another http request with the last resumption token and fetch data of next part, until finish all data fetching. The time to connect the database and do the query is a constant overhead. So it is not a good design that program need to connect the database and query upon each request with or with out resumption token. So my design is to connect to database and query only when the first initial request comes, and reponse back with partical data using header() and continue getting back data from z39 server. Upon next request, the program will only need to make sure the data needed has already stored in the harddrive and transform them and send them back. Since the user harvester agent normally has a 180 second timeout, it is necessary to respond within that period of time. I really need your suggestion. Thank you very much again. Liang

Try somthing lik

e this:

<?php ignore_user_abort(true); header("Location: redirect2.html"); echo "foo\n"; flush(); for($i=0;$i<10;$i++) { echo $i; sleep(1); } $fp = fopen("/tmp/foo.txt","a"); fputs($fp,$i); fclose($fp); ?> Liang ZHONG wrote:

Sorry, does not seem to work here. The code below takes minutes to

show

up in browser. Any more suggestion?

---------------------------------------------------------------------------------------

<?php set_time_limit(5); function f(){ set_time_limit(100); $count=500000000; for($i=0; $i<$count; $i++){ } echo "end"; exec("touch /home/.nappy/liang/liang.ns2user.info/php/aaa"); } register_shutdown_function('f'); ignore_user_abort(true); header("Content-type: text/plain"); header("Location: y.html"); $count=50000; for($i=0; $i<$count; $i++){ echo " \n"; } flush(); ?>

---------------------------------------------------------------------------------------

Liang

If you don't flush some output after setting the header() then the headers won't go out until the end of the request. So do something

like:

ignore_user_abort(true); header("Location: http://whatever"); echo "foo\n"; flush(); Then whatever comes after this should run and the browser is long

gone.

-Rasmus Liang ZHONG wrote:

I think I did not express myself clearly. What I want is to be able to redirect user an existing page (let them get it immediately), and to close the connection actively, NOT

passively

by user abort, at last, to run the function in background. But the redirecting using function header() with location has a problem that header() always does return the page to user after the entire program, including registered showdown function finish running,

which is

against the will. I put a time consuming task into a function that registered to be a shutdown function and hoping it runs after the user has got the redirected page and the connection has been closed.

But

experiements (using browsers, curl command line tool as well as perl::LWP code) show that the user got the redirected page only after the shutdown function finished, which is against the description

of

register_shutdown_function at php website. It seems only header() function use to redirect page has this problem (not executed until register_shutdown_function finished) while

other

functions like print()/echo(), exec() have not. The code looks like:

---------------------------------------------------------------------------------------------

<?php set_time_limit(1); function f(){ set_time_limit(20); $count=50000000; for($i=0; $i<$count; $i++){ } echo "end"; exec("touch /home/.nappy/liang/liang.ns2user.info/php/aaa"); } register_shutdown_function('f'); header("Content-type: text/plain"); header("Location: y.html"); ?>

---------------------------------------------------------------------------------------------

http client who sends the request to the php program will only get the page back as response after function f finsihes (file aaa

created).

Changing the $count will make a lot difference. My BIGGEST question is: How to make user get the redirect page immediately after the

header() is

called, and not until function f() ends, while making sure that

the

function f() will finally fully (and might slowly) execute? Thank you very much for kindly replying. With high respect, Liang

Liang ZHONG wrote:

My Question is: What is the correct way to keep the function running after I

redirect an

existing page to http client (which I want the client get

immediately)

and then immediately close the connection?

ignore_user_abort(true); -Rasmus -- PHP General Mailing List (http://www.php.net/) To unsubscribe, visit: http://www.php.net/unsub.php

-- PHP General Mailing List (http://www.php.net/) To unsubscribe, visit: http://www.php.net/unsub.php