Newsgroups: php.internals
Path: news.php.net
Xref: news.php.net php.internals:47014
Mailing-List: contact internals-help@lists.php.net; run by ezmlm
Message-ID: <F2.C8.00376.AC9D18B4@pb1.pair.com>
To: internals@lists.php.net
Followup-To: php.internals
Lines: 164
Date: Sun, 21 Feb 2010 19:11:46 -0600
MIME-Version: 1.0
Content-Type: multipart/mixed; boundary="nextPart4156758.TtaCxCBeP4"
Subject: multi-jobs run-tests.php
From: geissert@php.net (Raphael Geissert)

--nextPart4156758.TtaCxCBeP4
Content-Type: text/plain; charset="ISO-8859-1"
Content-Transfer-Encoding: 7Bit

Hi,

I've spent the last hours working on making run-tests.php be able to run 
tests in parallel. The main reason being the time it takes to run the whole 
testsuite even on multicore CPUs.

Attached is the first set of changes needed. It uses pcntl's fork and 
sysvmsg to send some of the results back to the parent process.

Main functionalities still work, but the summary doesn't include failed 
tests.

I plan to move away from sysvmsg because with the current implementation 
killing the parent process can leave message queues open (nothing that can't 
be fixed by adding a proper exit handler) and sysvmsg is not enabled by 
default. Probably to something php://memory-based.

To support all the features under multi-jobs mode I think the proper fix is 
to cleanup the code so that run_test() stores the state information via an 
interface. This would make the code cleaner and would let multi-jobs mode 
just change the state-storing interface.

How should I handle this? just by committing the changes? first sending them 
here (to internals)? by contacting somebody else?

Cheers,
-- 
Raphael Geissert
--nextPart4156758.TtaCxCBeP4
Content-Type: text/x-patch; name="php_parallel_run-tests.patch"
Content-Transfer-Encoding: 8Bit
Content-Disposition: attachment; filename="php_parallel_run-tests.patch"

diff --git a/run-tests.php b/run-tests.php
index 81f218b..9516490 100755
--- a/run-tests.php
+++ b/run-tests.php
@@ -451,6 +451,12 @@ if ($compression) {
 	$output_file = 'compress.zlib://' . $output_file . '.gz';
 }
 
+$max_jobs = false;
+
+if (extension_loaded('pcntl') && extension_loaded('sysvmsg')) {
+    $max_jobs = 1;
+}
+
 $just_save_results = false;
 $leak_check = false;
 $html_output = false;
@@ -551,6 +557,14 @@ if (isset($argc) && $argc > 1) {
 					$ini_overwrites[] = $argv[++$i];
 					break;
 				//case 'h'
+				case 'j':
+					if ($max_jobs !== false) {
+						$max_jobs = (int)$argv[++$i];
+						if ($max_jobs < 1) {
+							$max_jobs = 1;
+						}
+					}
+					break;
 				case '--keep-all':
 					foreach($cfgfiles as $file) {
 						$cfg['keep'][$file] = true;
@@ -665,6 +679,9 @@ Options:
     -d foo=bar  Pass -d option to the php binary (Define INI entry foo
                 with value 'bar').
 
+    -j <n>      Maximum number of jobs to run in parallel (defaults to 1,
+                requires pcntl and sysvmsg).
+
     -m          Test for memory leaks with Valgrind.
 
     -p <php>    Specify PHP executable to run.
@@ -1082,9 +1099,30 @@ function system_with_timeout($commandline, $env = null, $stdin = null)
 	return $data;
 }
 
+function reap_jobs($queue, &$results)
+{
+	// process messages aggresively to avoid
+	// saturating the queue, stalling childs
+	// waiting for space in the queue to exit
+	while (msg_receive($queue, 0, $null, 4096, $msg, true, MSG_IPC_NOWAIT))
+		$results[$msg['index']] = $msg['result'];
+	return pcntl_wait($null);
+}
+
 function run_all_tests($test_files, $env, $redir_tested = null)
 {
 	global $test_results, $failed_tests_file, $php, $test_cnt, $test_idx;
+	global $max_jobs;
+
+	$run_once = 0;
+
+	if ($max_jobs !== false) {
+		$jobs_avail = $max_jobs;
+		do {
+			$queue_key = mt_rand();
+		} while (function_exists('msg_queue_exists') && msg_queue_exists($queue_key));
+		$queue = msg_get_queue($queue_key);
+	}
 
 	foreach($test_files as $name) {
 
@@ -1100,15 +1138,56 @@ function run_all_tests($test_files, $env, $redir_tested = null)
 			$index = $name;
 		}
 		$test_idx++;
+
+		if ($max_jobs && ($max_jobs > 1 || $jobs_avail == 0)) {
+			if (!$jobs_avail) {
+				if (reap_jobs($queue, $test_results) > 0)
+				    $jobs_avail++;
+			}
+			$pid = pcntl_fork();
+			if ($pid == -1) {
+			    // Do nothing special, test will be run by
+			    // the parent process.
+			    // Decrement $jobs_avail and $max_jobs so that if
+			    // future calls to fork fail we will end up
+			    // disabling multi-jobs support without
+			    // leaving childs behind.
+			    $jobs_avail--;
+			    $max_jobs--;
+			} else if ($pid) {
+			    // parent process
+			    $jobs_avail--;
+			    continue;
+			} else {
+			    // child process:
+			    $max_jobs = false;
+			    $run_once = 1;
+			}
+		}
+
 		$result = run_test($php, $name, $env);
 
 		if (!is_array($name) && $result != 'REDIR') {
-			$test_results[$index] = $result;
+			// send the results back to the parent process
+			// if there's an active queue
+			if (isset($queue))
+				msg_send($queue, 1, array('index' => $index, 'result' => $result));
+			else
+				$test_results[$index] = $result;
 			if ($failed_tests_file && ($result == 'XFAILED' || $result == 'FAILED' || $result == 'WARNED' || $result == 'LEAKED')) {
 				fwrite($failed_tests_file, "$index\n");
 			}
 		}
+		if ($run_once) exit;
 	}
+
+	if ($max_jobs !== false) {
+		// check for messages one last time before removing the queue
+		// reap jobs at the same time
+		while (reap_jobs($queue, $test_results) > 0);
+		msg_remove_queue($queue);
+	}
+
 }
 
 //


--nextPart4156758.TtaCxCBeP4--