6.2 KiB
Task
At this stage, Swoole
has no way to hook
all blocking functions, which means that some functions will still cause process blocking
, which will affect the scheduling of coroutines. At this time, we can simulate coroutines by using the Task
component. In order to achieve the purpose of calling blocking functions without blocking the process, in essence, it is still multi-process running blocking functions, so the performance will be obviously inferior to the native coroutine, depending on the number of Task Worker
.
Install
composer require hyperf/task
Configure
Because Task is not the default component, you need to add Task
related configuration to server.php
when using it.
<?php
declare(strict_types=1);
use Hyperf\Server\Event;
return [
// Other irrelevant configuration items are omitted here
'settings' => [
// Number of Task Workers, configure the appropriate number based on your server configuration
'task_worker_num' => 8,
// Because `Task` mainly deals with methods that cannot be coroutined, it is recommended to set `false` here to avoid data confusion under coroutines
'task_enable_coroutine' => false,
],
'callbacks' => [
// Task callbacks
Event::ON_TASK => [Hyperf\Framework\Bootstrap\TaskCallback::class, 'onTask'],
Event::ON_FINISH => [Hyperf\Framework\Bootstrap\FinishCallback::class, 'onFinish'],
],
];
use
The Task component provides two usage methods: active method delivery
and annotation delivery
.
Active method delivery
<?php
use Hyperf\Utils\Coroutine;
use Hyperf\Utils\ApplicationContext;
use Hyperf\Task\TaskExecutor;
use Hyperf\Task\Task;
class MethodTask
{
public function handle($cid)
{
return [
'worker.cid' => $cid,
// Returns -1 when task_enable_coroutine is false, otherwise returns the corresponding coroutine ID
'task.cid' => Coroutine::id(),
];
}
}
$container = ApplicationContext::getContainer();
$exec = $container->get(TaskExecutor::class);
$result = $exec->execute(new Task([MethodTask::class, 'handle'], [Coroutine::id()]));
Using annotations
It is not particularly intuitive to use active method delivery
. Here we implement the corresponding #[Task]
annotation and rewrite the method call through AOP
. When in the Worker
process, it is automatically delivered to the Task
process, and the coroutine waits for the data to return.
<?php
use Hyperf\Utils\Coroutine;
use Hyperf\Utils\ApplicationContext;
use Hyperf\Task\Annotation\Task;
class AnnotationTask
{
#[Task]
public function handle($cid)
{
return [
'worker.cid' => $cid,
// Returns -1 when task_enable_coroutine=false, otherwise returns the corresponding coroutine ID
'task.cid' => Coroutine::id(),
];
}
}
$container = ApplicationContext::getContainer();
$task = $container->get(AnnotationTask::class);
$result = $task->handle(Coroutine::id());
use Hyperf\Task\Annotation\Task;
is required when using the#[Task]
annotation
The annotation supports the following parameters
| Configuration | Type | Default | Remarks | | :------: | :---: | :----: | :-------------------------------------- ----------------------: | | timeout | int | 10 | Task execution timeout | | workerId | int | -1 | Specifies the process ID of the task to be delivered (-1 means random delivery to an idle process) |
Appendix
Swoole does not have a list of coroutine functions for the time being
- mysql, the bottom layer uses libmysqlclient, which is not recommended, it is recommended to use pdo_mysql/mysqli that has already implemented coroutines
- mongo, the bottom layer uses mongo-c-client
- pdo_pgsql
- pdo_ori
- pdo_odbc
- pdo_firebird
MongoDB
Because
MongoDB
has no way to behook
, we can call it throughTask
. The following is a brief introduction to how to callMongoDB
through annotations.
Below we implement two methods insert
and query
. It should be noted that the manager
method cannot use Task
,
Because Task
will be processed in the corresponding Task process
, and then return the data from the Task process
to the Worker process
.
Therefore, the input and output parameters of the Task method
should not carry any IO
, such as returning an instantiated Redis
and so on.
<?php
declare(strict_types=1);
namespace App\Task;
use Hyperf\Task\Annotation\Task;
use MongoDB\Driver\BulkWrite;
use MongoDB\Driver\Manager;
use MongoDB\Driver\Query;
use MongoDB\Driver\WriteConcern;
class MongoTask
{
public Manager $manager;
#[Task]
public function insert(string $namespace, array $document)
{
$writeConcern = new WriteConcern(WriteConcern::MAJORITY, 1000);
$bulk = new BulkWrite();
$bulk->insert($document);
$result = $this->manager()->executeBulkWrite($namespace, $bulk, $writeConcern);
return $result->getUpsertedCount();
}
#[Task]
public function query(string $namespace, array $filter = [], array $options = [])
{
$query = new Query($filter, $options);
$cursor = $this->manager()->executeQuery($namespace, $query);
return $cursor->toArray();
}
protected function manager()
{
if ($this->manager instanceof Manager) {
return $this->manager;
}
$uri = 'mongodb://127.0.0.1:27017';
return $this->manager = new Manager($uri, []);
}
}
Use as follows
<?php
use App\Task\MongoTask;
use Hyperf\Utils\ApplicationContext;
$client = ApplicationContext::getContainer()->get(MongoTask::class);
$client->insert('hyperf.test', ['id' => rand(0, 99999999)]);
$result = $client->query('hyperf.test', [], [
'sort' => ['id' => -1],
'limit' => 5,
]);
Other options
If the Task mechanism cannot meet the performance requirements, you can try another open source project under the Hyperf organization GoTask. GoTask starts the Go process as the Swoole main process sidecar through the Swoole process management function, and uses the process communication to deliver the task to the sidecar for processing and receive the return value. It can be understood as the Go version of Swoole TaskWorker.