Doc #63842 [Nab]: Infinite loop in example code

Edit report at https://bugs.php.net/bug.php?id=63842&edit=1 ID: 63842 User updated by: dean at ovts dot com dot au Reported by: dean at ovts dot com dot au Summary: Infinite loop in example code Status: Not a bug Type: Documentation Problem Package: Documentation problem PHP Version: Irrelevant Block user comment: N Private report: N New Comment: Well, I've looked over the relevant libcurl docs, and taken a look at how others on the internet are using curl_multi, and based on that, here's what I think represents best-practice, as example code for including in the PHP documentation pages: $MAX_SIMULTANEOUS = 50; // Adjust to whatever number of maximum simultaneous requests you think is appropriate. // Fill array $reqList[] with whatever data you're using to generate your CURL requests, // one request per array element, with consecutive indices starting at 0. // note that requests at the END of the array will be started first. if ($i = count($reqList)) // check that there is something to do... { $handleMap = new SplObjectStorage; // use this to associate curl handles with the data that created them $curlMH = curl_multi_init(); $x = $i - $MAX_SIMULTANEOUS; if ($x<0) $x = 0; while ($i>$x) initRequest($reqList[--$i]); do { while (($mrc = curl_multi_exec($curlMH, $active)) === CURLM_CALL_MULTI_PERFORM); if ($mrc !== CURLM_OK) break; // Shouldn't normally ever happen; look at the list of CURLM_ errors to see when it might. while ($info = curl_multi_info_read($curlMH)) // a request has completed { $ch = $info['handle']; if (($errNo = $info['result']) === CURLE_OK) processResult($handleMap[$ch], curl_multi_getcontent($ch)); else ... // handle failed CURL request, (e.g. write to the error log or database, output a message, etc.) if ($i) initRequest($reqList[--$i]); // start new request if one is waiting curl_multi_remove_handle($curlMH, $ch); curl_close($ch); $handleMap->detach($ch); // clean up completed request } // wait for next CURL operation to complete, or sleep a short time if CURL is busy but unable to "block" using curl_multi_select if ($active && (curl_multi_select($curlMH) === -1)) usleep(50); } while ($active); if ($mrc !== CURLM_OK) ... // optinally handle CURLM_ errors (e.g. write to error log, output message, etc.) // clean up foreach ($handleMap as $ch) { curl_multi_remove_handle($curlMH, $ch); curl_close($ch); } curl_multi_close($curlMH); $reqList = $curlMH = $handleMap = null; } function initRequest($reqData) { global $curlMH, $handleMap; // process $reqData to get $url (the request URL), and any POST data, or other CURL options. // also generate $custReqData, which will be passed to processResult to identify the request // when it has completed (may be the same as $reqData). $ch = curl_init($url); curl_setopt($ch, ...); // set other CURL options curl_setopt($ch, ...); curl_multi_add_handle($curlMH, $ch); $handleMap[$ch] = $custReqData; } function processResult($custReqData, $response) { // process $response (the data returned from the CURL request) // using $custReqData to identify which request it belongs to. } Previous Comments: ------------------------------------------------------------------------ [2013-02-07 19:02:39] karl at silverglassworks dot com The current example is definitely an incorrect way of doing things, and was causing an infinite loop for me today. From the cURL documentation (http://curl.haxx.se/libcurl/c/curl_multi_fdset.html): ****************************************************************************** When libcurl returns -1 in max_fd, it is because libcurl currently does something that isn't possible for your application to monitor with a socket and unfortunately you can then not know exactly when the current action is completed using select(). When max_fd returns with -1, you need to wait a while and then proceed and call curl_multi_perform anyway. How long to wait? I would suggest 100 milliseconds at least, but you may want to test it out in your own particular conditions to find a suitable value. ****************************************************************************** So in the situation where libcurl can't use select(), it will always return -1, and thus the example will be an infinite loop, as I have been experiencing. I changed my code to the following, and it now works: while ($active && $mrc == CURLM_OK) { if (curl_multi_select($mh) === -1) { usleep(100); } do { $mrc = curl_multi_exec($mh, $active); } while ($mrc == CURLM_CALL_MULTI_PERFORM); } ------------------------------------------------------------------------ [2013-01-16 19:21:35] dean at ovts dot com dot au Googleguy, under bug #61141, several people report that they do indeed get an infinite loop because curl_multi_select returns -1 indefinitely... ------------------------------------------------------------------------ [2013-01-16 19:14:19] dean at ovts dot com dot au If the loop is entered and curl_multi_select then returns -1, the variables determining the loop condition remain unchanged, so curl_multi_select is simply called repeatedly until such time as it returns something other than -1. Googleguy, I don't quite understand what you were saying: Is it that eventually curl_multi_select is *GUARANTEED* to return something other than -1? *IF* that's true then there's no infinite loop. However, it's still very unclear exectly how/why this code is supposed to work. I'd like to see an explanation/clarification in the docs of the return values from curl_multi_exec and curl_multi_select, and of the $active flag. What exactly does a -1 returned from curl_multi_select imply? And (again) what would it mean if $active were true, but the return value from curl_multi_exec wasn't CURLM_OK? At the moment, without any such explanation in the docs, the example code is like an opaque magic formula for how curl_multi_exec should be used. It's difficult to trust it, when there are many "alternative versions" posted on the internet, and some on other php docs pages too! ------------------------------------------------------------------------ [2013-01-16 18:31:56] googleguy@php.net *Correction to my last comment "This is not a bug, because the condition of the loop is NOT dependent upon curl_multi_select." ------------------------------------------------------------------------ [2013-01-16 18:31:05] googleguy@php.net This is not a bug, because the condition of the loop is dependent upon curl_multi_select. If the curl_multi_select returns -1 the code within the loop body is simply not executed. This still does not create an infinite loop under normal conditions because eventually if all of the curl handles timeout they are freed from the select resource handle. ------------------------------------------------------------------------ The remainder of the comments for this report are too long. To view the rest of the comments, please view the bug report online at https://bugs.php.net/bug.php?id=63842 -- Edit this bug report at https://bugs.php.net/bug.php?id=63842&edit=1