PHP-rdkafka is a thin librdkafka binding providing a working PHP 5 / PHP 7 Kafka client.
It supports the high level and low level consumers, producer, and metadata APIs.
The API ressembles as much as possible to librdkafka's, and is fully documented here.
The source of the documentation can be found here
Installing librdkafka dependency (Ubuntu and Debian)
If you install by running:
apt-get -y install librdkafka-devlibrdkafka version 0.x.x will be installed.
You can check currently installed version of librdkafka-dev by running:
apt-cache policy librdkafka-devhttps://arnaud-lb.github.io/php-rdkafka/phpdoc/rdkafka.setup.html
https://arnaud-lb.github.io/php-rdkafka/phpdoc/rdkafka.examples.html
Configuration parameters used below can found in Librdkafka Configuration reference
For producing, we first need to create a producer, and to add brokers (Kafka servers) to it:
<?php
$conf = new RdKafka\Conf();
$conf->set('log_level', LOG_DEBUG);
$conf->set('debug', 'all');
$rk = new RdKafka\Producer($conf);
$rk->addBrokers("10.0.0.1:9092,10.0.0.2:9092");
Next, we create a topic instance from the producer:
<?php
$topic = $rk->newTopic("test");From there, we can produce as much messages as we want, using the produce method:
<?php
$topic->produce(RD_KAFKA_PARTITION_UA, 0, "Message payload");The first argument is the partition. RD_KAFKA_PARTITION_UA stands for
unassigned, and lets librdkafka choose the partition.
The second argument are message flags and should be either 0
or RD_KAFKA_MSG_F_BLOCK to block produce on full queue.
The message payload can be anything.
This should be done prior to destroying a producer instance
to make sure all queued and in-flight produce requests are completed
before terminating. Use a reasonable value for $timeout_ms.
$rk->flush($timeout_ms);In case you don't care about sending messages that haven't been sent yet,
you can use purge() before calling flush():
// Forget messages that are not fully sent yet
$rk->purge(RD_KAFKA_PURGE_F_QUEUE);
$rk->flush($timeout_ms);The RdKafka\KafkaConsumer class supports automatic partition assignment/revocation. See the example here.
We first need to create a low level consumer, and to add brokers (Kafka servers) to it:
<?php
$conf = new Conf();
$conf->set('log_level', LOG_LEVEL);
$conf->set('debug', 'all');
$rk = new RdKafka\Consumer($conf);
$rk->addBrokers("10.0.0.1,10.0.0.2");Next, create a topic instance by calling the newTopic() method, and start
consuming on partition 0:
<?php
$topic = $rk->newTopic("test");
// The first argument is the partition to consume from.
// The second argument is the offset at which to start consumption. Valid values
// are: RD_KAFKA_OFFSET_BEGINNING, RD_KAFKA_OFFSET_END, RD_KAFKA_OFFSET_STORED.
$topic->consumeStart(0, RD_KAFKA_OFFSET_BEGINNING);Next, retrieve the consumed messages:
<?php
while (true) {
// The first argument is the partition (again).
// The second argument is the timeout.
$msg = $topic->consume(0, 1000);
if (null === $msg) {
continue;
} elseif ($msg->err) {
echo $msg->errstr(), "\n";
break;
} else {
echo $msg->payload, "\n";
}
}Consuming from multiple topics and/or partitions can be done by telling librdkafka to forward all messages from these topics/partitions to an internal queue, and then consuming from this queue:
Creating the queue:
<?php
$queue = $rk->newQueue();Adding topic partitions to the queue:
<?php
$topic1 = $rk->newTopic("topic1");
$topic1->consumeQueueStart(0, RD_KAFKA_OFFSET_BEGINNING, $queue);
$topic1->consumeQueueStart(1, RD_KAFKA_OFFSET_BEGINNING, $queue);
$topic2 = $rk->newTopic("topic2");
$topic2->consumeQueueStart(0, RD_KAFKA_OFFSET_BEGINNING, $queue);Next, retrieve the consumed messages from the queue:
<?php
while (true) {
// The only argument is the timeout.
$msg = $queue->consume(1000);
if (null === $msg) {
continue;
} elseif ($msg->err) {
echo $msg->errstr(), "\n";
break;
} else {
echo $msg->payload, "\n";
}
}librdkafka per default stores offsets on the broker.
If you're using local file for offset storage, then by default the file is created in the current directory, with a
name based on the topic and the partition. The directory can be changed by setting the offset.store.path
configuration property.
Other interesting properties are: auto.commit.interval.ms, auto.commit.enable, group.id, max.poll.interval.ms.
auto.commit.interval.ms and auto.commit.enable work in tandem: unless you specify otherwise, consumers WILL
commit automatically in the background (at least high-level ones). If you need control and want to commit manually,
then you want to set auto.commit.enable to 'false'.
group.id is responsible for setting your consumer group ID and it should be unique (and should
not change). Kafka uses it to recognize applications and store offsets for them.
<?php
$topicConf = new RdKafka\TopicConf();
$topicConf->set("auto.commit.interval.ms", 1e3);
$topic = $rk->newTopic("test", $topicConf);
$topic->consumeStart(0, RD_KAFKA_OFFSET_STORED);Librdkafka Configuration reference
librdkafka will buffer up to 1GB of messages for each consumed partition by default. You can lower memory usage by reducing the value of the queued.max.messages.kbytes parameter on your consumers.
Each consumer and procuder instance will fetch topics metadata at an interval defined by the topic.metadata.refresh.interval.ms parameter. Depending on your librdkafka version, the parameter defaults to 10 seconds, or 600 seconds.
librdkafka fetches the metadata for all topics of the cluster by default. Setting topic.metadata.refresh.sparse to the string "true" makes sure that librdkafka fetches only the topics he uses.
Setting topic.metadata.refresh.sparse to "true", and topic.metadata.refresh.interval.ms to 600 seconds (plus some jitter) can reduce the bandwidth a lot, depending on the number of consumers and topics.
This setting allows librdkafka threads to terminate as soon as librdkafka is done with them. This effectively allows your PHP processes / requests to terminate quickly.
When enabling this, you have to mask the signal like this:
<?php
// once
pcntl_sigprocmask(SIG_BLOCK, array(SIGIO));
// any time
$conf->set('internal.termination.signal', SIGIO);Maximum time a broker socket operation may block. A lower value improves responsiveness at the expense of slightly higher CPU usage.
Reducing the value of this setting improves shutdown speed. The value defines the maximum time librdkafka will block in one iteration of a read loop. This also defines how often the main librdkafka thread will check for termination.
This defines the maximum and default time librdkafka will wait before sending a batch of messages. Reducing this setting to e.g. 1ms ensures that messages are sent ASAP, instead of being batched.
This has been seen to reduce the shutdown time of the rdkafka instance, and of the PHP process / request.
Here is a configuration optimized for low latency. This allows a PHP process / request to send messages ASAP and to terminate quickly.
<?php
$conf = new \RdKafka\Conf();
$conf->set('socket.timeout.ms', 50); // or socket.blocking.max.ms, depending on librdkafka version
if (function_exists('pcntl_sigprocmask')) {
pcntl_sigprocmask(SIG_BLOCK, array(SIGIO));
$conf->set('internal.termination.signal', SIGIO);
} else {
$conf->set('queue.buffering.max.ms', 1);
}
$producer = new \RdKafka\Producer($conf);
$consumer = new \RdKafka\Consumer($conf);Polling after producing can also be important to reduce termination times:
$producer->produce(...);
while ($producer->getOutQLen() > 0) {
$producer->poll(1);
}
https://arnaud-lb.github.io/php-rdkafka/phpdoc/book.rdkafka.html
The source of the documentation can be found here
If the documentation is not enough, feel free to ask a questions on the php-rdkafka channels on Gitter or Google Groups.
Because your IDE is not able to auto discover php-rdkadka api you can consider usage of external package providing a set of stubs for php-rdkafka classes, functions and constants: kwn/php-rdkafka-stubs
If you would like to contribute, thank you :)
Before you start, please take a look at the CONTRIBUTING document to see how to get your changes merged in.
Documentation copied from librdkafka.
Authors: see contributors.
php-rdkafka is released under the MIT license.