Feature: task time limits (and other related stuff)

.github/workflows/macos_l11.yml

@@ -25,6 +25,9 @@ jobs:
       with:
         php-version: '8.2'
 
+    - name: "Homebrew: Install GNU Core Utilities"
+      run: yes | brew install coreutils
+
     - name: Validate composer.json and composer.lock
       run: composer validate --strict
 

README.md

@@ -13,13 +13,17 @@ Utilize Laravel Processes to run PHP code asynchronously.
 - Restrictions from `laravel/serializable-closure` apply (see [their README](https://github.com/laravel/serializable-closure))
 - Hands-off execution: no built-in result-checking, check the results yourself (e.g. via database, file cache, etc)
 
+This library internally uses an Artisan command to run the async code, which is similar to Laravel 11 [Concurrency](https://laravel.com/docs/11.x/concurrency).
+
 ## Why should I want this?
 This library is very helpful for these cases:
-- You want a minimal-setup async for easy vertical scaling
+- You want a cross-platform minimal-setup async for easy vertical scaling
 - You want to start quick-and-dirty async tasks right now (e.g. prefetching resources, pinging remote, etc.)
   - Best is if your task only has very few lines of code
 - Laravel 11 [Concurrency](https://laravel.com/docs/11.x/concurrency) is too limiting; e.g.:
   - You want to do something else while waiting for results
+  - You want to conveniently limit the max (real) execution time of the concurrent tasks
+- And perhaps more!
 
 Of course, if you are considering extreme scaling (e.g. Redis queues in Laravel, multi-worker clusters, etc.) or guaranteed task execution, then this library is obviously not for you.
 
@@ -32,6 +36,12 @@ composer require vectorial1024/laravel-process-async
 
 This library supports Unix and Windows; see the Testing section for more details.
 
+### Extra requirements for Unix
+If you are on Unix, check that you also have the following:
+- GNU Core Utilities (`coreutils`)
+  - MacOS do `brew install coreutils`!
+  - Other Unix distros should check if `coreutils` is preinstalled
+
 ## Change log
 Please see `CHANGELOG.md`.
 
@@ -59,6 +69,24 @@ $task->start();
 // the task is now run in another PHP process, and will not report back to this PHP process.
 ```
 
+### Task time limits
+You can set task time limits before you start them, but you cannot change them after the tasks are started. When the time limit is reached, the async task is killed.
+
+The default time limit is 30 real seconds. You can also choose to not set any time limit, in this case the (CLI) PHP `max_execution_time` directive will control the time limit.
+
+Note: `AsyncTaskInterface` contains an implementable method `handleTimeout` for you to define timeout-related cleanups (e.g. write to some log that the task has timed out). This method is still called when the PHP `max_execution_time` directive is triggered.
+
+```php
+// start with the default time limit...
+$task->start();
+
+// start task with a different time limit...
+$task->withTimeLimit(15)->start();
+
+// ...or not have any limits at all (beware of orphaned processes!)
+$task->withoutTimeLimit()->start();
+```
+
 ## Testing
 PHPUnit via Composer script:
 ```sh

src/AsyncTask.php

@@ -1,12 +1,16 @@
 <?php
 
+declare(strict_types=1);
+
 namespace Vectorial1024\LaravelProcessAsync;
 
 use Closure;
 use Illuminate\Process\InvokedProcess;
 use Illuminate\Support\Facades\Process;
 use Laravel\SerializableClosure\SerializableClosure;
+use LogicException;
 use loophp\phposinfo\OsInfo;
+use RuntimeException;
 
 /**
  * The common handler of an AsyncTask; this can be a closure (will be wrapped inside AsyncTask) or an interface instance.
@@ -25,6 +29,65 @@ class AsyncTask
      */
     private InvokedProcess|null $runnerProcess = null;
 
+    /**
+     * The maximum real time (in seconds) this task is allowed to run.
+     * @var int|null
+     */
+    private int|null $timeLimit = 30;
+
+    /**
+     * The value of constant("LARAVEL_START") for future usage. Apparently, constants are not available during shutdown functions.
+     * @var float|null
+     */
+    private float|null $laravelStartVal = null;
+
+    /**
+     * On Unix only. Indicates the process ID that can be used to track the "time elapsed" stat, which resolves to the following:
+     * - if the task was started under the `timeout` command, then the PID of said `timeout` command
+     * - else (i.e., started without time limit), the self PID
+     * 
+     * If not yet initialized or on Windows, then will be 0, which indicates an invalid PID.
+     * @var int
+     */
+    private int $timerProcID = 0;
+
+    /**
+     * On Unix only. Indicates whether a SIGINT has been received.
+     * @var bool
+     */
+    private bool $hasSigInt = false;
+
+    /**
+     * The string constant name for constant('LARAVEL_START'). Mainly to keep the code clean.
+     * @var string
+     */
+    private const LARAVEL_START = "LARAVEL_START";
+
+    /**
+     * The bitmask that can filter for fatal runtime errors.
+     * 
+     * Fatal errors other than the specific "time limit exceeded" error must not trigger the timeout handlers.
+     */
+    private const FATAL_ERROR_BITMASK = E_ERROR | E_PARSE | E_CORE_ERROR | E_COMPILE_ERROR;
+
+    /**
+     * Indicates whether GNU coreutils is found in the system; in particular, we are looking for the timeout command inside coreutils.
+     * 
+     * If null, indicates we haven't checked this yet.
+     * 
+     * Always null in Windows since Windows-side does not require GNU coreutils.
+     * @var bool|null
+     */
+    private static bool|null $hasGnuCoreUtils = null;
+
+    /**
+     * The name of the found timeout command inside GNU coreutils.
+     * 
+     * It is known that older MacOS environments might have "gtimeout" instead of "timeout".
+     * @var string|null
+     */
+    private static string|null $timeoutCmdName = null;
+
     /**
      * Creates an AsyncTask instance.
      * @param Closure|AsyncTaskInterface $theTask The task to be executed in the background.
@@ -48,6 +111,35 @@ public function __construct(Closure|AsyncTaskInterface $theTask)
     public function run(): void
     {
         // todo startup configs
+        // write down the LARAVEL_START constant value for future usage
+        $this->laravelStartVal = defined(self::LARAVEL_START) ? constant("LARAVEL_START") : null;
+
+        // install a timeout detector
+        // this single function checks all kinds of timeouts
+        register_shutdown_function([$this, 'shutdownCheckTaskTimeout']);
+        if (OsInfo::isWindows()) {
+            // windows can just use PHP's time limit
+            set_time_limit($this->timeLimit);
+        } else {
+            // assume anything not Windows to be Unix
+            // we already set it to kill this task after the timeout, so we just need to install a listener to catch the signal and exit gracefully
+            pcntl_async_signals(true);
+            pcntl_signal(SIGINT, function () {
+                // sicne we are already running with nohup, we can use SIGINT to indicate that a timeout has occurred.
+                // exit asap so that our error checking inside shutdown functions can take place outside of the usual max_execution_time limit
+                $this->hasSigInt = true;
+                exit();
+            });
+
+            // and we also need to see the command name of our parent, to correctly track time
+            $this->timerProcID = getmypid();
+            $parentPid = posix_getppid();
+            $parentCmd = exec("ps -p $parentPid -o comm=");
+            if ($parentCmd === "timeout" || $parentCmd === "gtimeout") {
+                // we should use the parent instead to time this task
+                $this->timerProcID = $parentPid;
+            }
+        }
 
         // then, execute the task itself
         if ($this->theTask instanceof SerializableClosure) {
@@ -77,12 +169,32 @@ public function start(): void
         if (OsInfo::isWindows()) {
             // basically, in windows, it is too tedioous to check whether we are in cmd or ps,
             // but we require cmd (ps won't work here), so might as well force cmd like this
+            // windows has real max time limit
             $this->runnerProcess = Process::quietly()->start("cmd /c start /b $baseCommand");
             return;
         }
         // assume anything not windows to be unix
         // unix use nohup
-        $this->runnerProcess = Process::quietly()->start("nohup $baseCommand");
+        // check time limit settings
+        $timeoutClause = "";
+        if ($this->timeLimit > 0) {
+            // do we really have timeout here?
+            if (static::$hasGnuCoreUtils === null) {
+                // haven't checked before; check
+                $tmpOut = exec("command -v timeout || command -v gtimeout");
+                $cmdName = !empty($tmpOut) ? $tmpOut : null;
+                unset($tmpOut);
+                static::$hasGnuCoreUtils = $cmdName !== null;
+                static::$timeoutCmdName = $cmdName;
+            }
+            if (static::$hasGnuCoreUtils === false) {
+                // can't do anything without GNU coreutils!
+                throw new RuntimeException("AsyncTask time limit requires GNU coreutils, but GNU coreutils was not installed");
+            }
+            // 2 is INT signal
+            $timeoutClause = static::$timeoutCmdName . " -s 2 {$this->timeLimit}";
+        }
+        $this->runnerProcess = Process::quietly()->start("nohup $timeoutClause $baseCommand >/dev/null 2>&1");
     }
 
     /**
@@ -126,4 +238,125 @@ public static function fromBase64Serial(string $serial): ?static
             return null;
         }
     }
+
+    /**
+     * Returns the maximum real time this task is allowed to run. This also includes time spent on sleeping and waiting!
+     * 
+     * Null indicates unlimited time.
+     * @return int|null The time limit in seconds.
+     */
+    public function getTimeLimit(): int|null
+    {
+        return $this->timeLimit;
+    }
+
+    /**
+     * Sets the maximum real time this task is allowed to run. Chainable.
+     * 
+     * When the task reaches the time limit, the timeout handler (if exists) will be called.
+     * @param int $seconds The time limit in seconds.
+     * @return AsyncTask $this for chaining.
+     */
+    public function withTimeLimit(int $seconds): static
+    {
+        if ($seconds == 0) {
+            throw new LogicException("AsyncTask time limit must be positive (hint: use withoutTimeLimit() for no time limits)");
+        }
+        if ($seconds < 0) {
+            throw new LogicException("AsyncTask time limit must be positive");
+        }
+        $this->timeLimit = $seconds;
+        return $this;
+    }
+
+    /**
+     * Sets this task to run with no time limit (PHP INI `max_execution_time` may apply). Chainable.
+     * @return AsyncTask $this for chaining.
+     */
+    public function withoutTimeLimit(): static
+    {
+        $this->timeLimit = null;
+        return $this;
+    }
+
+    /**
+     * A shutdown function.
+     * 
+     * Upon shutdown, checks whether this is due to the task timing out, and if so, triggers the timeout handler.
+     * @return void
+     */
+    protected function shutdownCheckTaskTimeout(): void
+    {
+        if (!$this->hasTimedOut()) {
+            // shutdown due to other reasons; skip
+            return;
+        }
+
+        // timeout!
+        // trigger the timeout handler
+        if ($this->theTask instanceof AsyncTaskInterface) {
+            $this->theTask->handleTimeout();
+        }
+    }
+
+    /**
+     * During shutdown, checks whether this shutdown satisfies the "task timed out shutdown" condition.
+     * @return bool True if this task is timed out according to our specifications.
+     */
+    private function hasTimedOut(): bool
+    {
+        // we perform a series of checks to see if this task has timed out
+
+        // dedicated SIGINT indicates a timeout
+        if ($this->hasSigInt) {
+            return true;
+        }
+
+        // runtime timeout triggers a PHP fatal error
+        // this can happen on Windows by our specification, or on Unix when the actual CLI PHP time limit is smaller than the time limit of this task
+        $lastError = error_get_last();
+        if ($lastError !== null && ($lastError['type'] & self::FATAL_ERROR_BITMASK)) {
+            // has fatal error; is it our timeout error?
+            return str_contains($lastError['message'], "Maximum execution time");
+        }
+        unset($lastError);
+
+        // the remaining checks use the time-limit variable, so if it is unset, then there is nothing to check
+        if ($this->timeLimit <= 0) {
+            return false;
+        }
+
+        // not a runtime timeout; one of the following:
+        // it ended within the time limit; or
+        // on Unix, it ran out of time so it is getting a SIGTERM from us; or
+        // it somehow ran out of time, and is being manually detected and killed
+        if ($this->laravelStartVal !== null) {
+            // this is very important; in some test cases, this is being run directly by PHPUnit, and so LARAVEL_START will be null
+            // in this case, we have no idea when this task has started running, so we cannot deduce any timeout statuses
+
+            // check LARAVEL_START with microtime
+            $timeElapsed = microtime(true) - $this->laravelStartVal;
+            if ($timeElapsed >= $this->timeLimit) {
+                // yes
+                return true;
+            }
+
+            // if we are on Unix, and when we have set a task time limit, then the LARAVEL_START value is inaccurate
+            // because there will always be a small but significant delay between `timeout` start time and PHP start time.
+            // in this case, we will look at the pre-determined timer PID to ask about the actual elapsed time through the kernel's proc data
+            // this method should be slower than the microtime method
+            if (OsInfo::isUnix()) {
+                // get time elapsed in seconds
+                $tempOut = exec("ps -p {$this->timerProcID} -o etimes=");
+                // this must exist (we are still running!), otherwise it indicates the kernel is broken and we can go grab a chicken dinner instead
+                $timeElapsed = (int) $tempOut;
+                unset($tempOut);
+                // it seems like etimes can get random off-by-1 inaccuracies (e.g. timeout supposed to be 7, but etimes sees 6.99999... and prints "6")
+                return $timeElapsed >= $this->timeLimit;
+            }
+        }
+
+        // didn't see anything; assume is no
+        return false;
+    }
 }

src/AsyncTaskInterface.php

@@ -1,5 +1,7 @@
 <?php
 
+declare(strict_types=1);
+
 namespace Vectorial1024\LaravelProcessAsync;
 
 /**
@@ -16,4 +18,12 @@ interface AsyncTaskInterface
      * @return void
      */
     public function execute(): void;
+
+    /**
+     * Cleans up the task when the task runner has run out of time specified by its time limit.
+     * 
+     * Note: there is no need to call exit() again in this function.
+     * @return void
+     */
+    public function handleTimeout(): void;
 }