seldon core 通过 kafka 实现数据回流
一、官方方案
https://docs.seldon.io/projects/seldon-core/en/latest/analytics/logging.html?highlight=kafka
https://docs.seldon.io/projects/seldon-core/en/latest/streaming/kafka.html?highlight=kafka
示例:
apiVersion: machinelearning.seldon.io/v1
kind: SeldonDeployment
metadata:
name: cifar10
namespace: seldon
spec:
name: resnet32
predictors:
- graph:
implementation: TRITON_SERVER
logger:
mode: all
modelUri: gs://seldon-models/triton/tf_cifar10
name: cifar10
name: default
svcOrchSpec:
env:
- name: LOGGER_KAFKA_BROKER
value: seldon-kafka-plain-0.kafka:9092
- name: LOGGER_KAFKA_TOPIC
value: seldon
replicas: 1
protocol: v2
The two required environment variables are:
- LOGGER_KAFKA_BROKER : The Kafka Broker service endpoint.
- LOGGER_KAFKA_TOPIC : The kafka Topic to log the requests.
实践:
seldon-log-demo.yaml
[root@k8s-master seldon]# cat seldon-log-demo.yaml
apiVersion: machinelearning.seldon.io/v1
kind: SeldonDeployment
metadata:
name: iris-modelkafka-log
namespace: seldon
spec:
name: iris
predictors:
- graph:
implementation: SKLEARN_SERVER
logger:
mode: all
modelUri: gs://seldon-models/v1.12.0-dev/sklearn/iris
name: classifier
svcOrchSpec:
env:
- name: LOGGER_KAFKA_BROKER
value: 192.168.1.191:9092
- name: LOGGER_KAFKA_TOPIC
value: test-kafka
name: default
replicas: 1
重要的两个参数:
LOGGER_KAFKA_BROKER:192.168.1.191:9092
# topic 需要提前创建
LOGGER_KAFKA_TOPIC: test-kafka
部署模型:
[root@k8s-master seldon]# kubectl apply -f seldon-log-demo.yaml
[root@k8s-master seldon]# kubectl get pods -n seldon
NAME READY STATUS RESTARTS AGE
iris-modelkafka-log-default-0-classifier-7c985485b5-5gs8f 3/3 Running 0 7h37m
查看部署的svc:
[root@k8s-master ~]# kubectl get svc -n seldon
NAME TYPE CLUSTER-IP EXTERNAL-IP PORT(S) AGE
iris-modelkafka-log-default ClusterIP 10.233.50.157 <none> 8000/TCP,5001/TCP 7h29m
iris-modelkafka-log-default-classifier ClusterIP 10.233.63.196 <none> 9000/TCP,9500/TCP 7h29m
在k8s集群测试服务:
curl -X POST http://10.233.50.157:8000/api/v1.0/predictions \
-H 'Content-Type: application/json' \
-d '{ "data": { "ndarray": [[1,2,3,4]] } }'
测试打印:
[root@k8s-master ~]# curl -X POST http://10.233.50.157:8000/api/v1.0/predictions \
> -H 'Content-Type: application/json' \
> -d '{ "data": { "ndarray": [[1,2,3,4]] } }'
{"data":{"names":["t:0","t:1","t:2"],"ndarray":[[0.0006985194531162835,0.00366803903943666,0.995633441507447]]},"meta":{"requestPath":{"classifier":"seldonio/sklearnserver:1.14.1"}}}
[root@k8s-master ~]# curl -X POST http://10.233.50.157:8000/api/v1.0/predictions -H 'Content-Type: application/json' -d '{ "data": { "ndarray": [[1,2,3,4]] } }'
{"data":{"names":["t:0","t:1","t:2"],"ndarray":[[0.0006985194531162835,0.00366803903943666,0.995633441507447]]},"meta":{"requestPath":{"classifier":"seldonio/sklearnserver:1.14.1"}}}
监听kafka消息,获取回流日志:
offset = 7, value = { "data": { "ndarray": [[1,2,3,4]] } }
offset = 8, value = {"data":{"names":["t:0","t:1","t:2"],"ndarray":[[0.0006985194531162835,0.00366803903943666,0.995633441507447]]},"meta":{"requestPath":{"classifier":"seldonio/sklearnserver:1.14.1"}}}
offset = 9, value = { "data": { "ndarray": [[1,2,3,4]] } }
offset = 10, value = {"data":{"names":["t:0","t:1","t:2"],"ndarray":[[0.0006985194531162835,0.00366803903943666,0.995633441507447]]},"meta":{"requestPath":{"classifier":"seldonio/sklearnserver:1.14.1"}}}
二、kafka生产者和消费者
KafkaProducerDemo.java
import org.apache.kafka.clients.producer.KafkaProducer;
import org.apache.kafka.clients.producer.ProducerConfig;
import org.apache.kafka.clients.producer.ProducerRecord;
import org.apache.kafka.common.serialization.StringSerializer;
import java.util.Properties;
/**
* Kafka生产者Demo
*/
public class KafkaProducerDemo {
public static void main(String[] args) {
Properties props = new Properties();
// 服务器ip:端口号,集群用逗号分隔
props.put(ProducerConfig.BOOTSTRAP_SERVERS_CONFIG, "192.168.1.191:9092");
// key序列化指定类
props.put(ProducerConfig.KEY_SERIALIZER_CLASS_CONFIG, StringSerializer.class.getName());
// value序列化指定类
props.put(ProducerConfig.VALUE_SERIALIZER_CLASS_CONFIG, StringSerializer.class.getName());
// 生产者对象
KafkaProducer<String, String> producer = new KafkaProducer<>(props);
// 向test_topic发送hello, kafka
producer.send(new ProducerRecord<String, String>("test-kafka", "哇哈哈0000哈999"));
producer.close();
}
}
消费者:
KafkaConsumerDemo.java
import org.apache.kafka.clients.consumer.ConsumerConfig;
import org.apache.kafka.clients.consumer.ConsumerRecord;
import org.apache.kafka.clients.consumer.ConsumerRecords;
import org.apache.kafka.clients.consumer.KafkaConsumer;
import org.apache.kafka.clients.producer.KafkaProducer;
import org.apache.kafka.clients.producer.ProducerConfig;
import org.apache.kafka.clients.producer.ProducerRecord;
import org.apache.kafka.common.serialization.StringDeserializer;
import org.apache.kafka.common.serialization.StringSerializer;
import java.util.Arrays;
import java.util.Properties;
/**
* Kafka消费者Demo
*/
public class KafkaConsumerDemo {
public static void main(String[] args) {
Properties props = new Properties();
// 服务器ip:端口号,集群用逗号分隔
props.put(ConsumerConfig.BOOTSTRAP_SERVERS_CONFIG, "192.168.1.191:9092");
// 消费者指定组,名称可以随意,注意相同消费组中的消费者只能对同一个分区消费一次
props.put(ConsumerConfig.GROUP_ID_CONFIG, "test");
// 是否启用自动提交,默认true
props.put(ConsumerConfig.ENABLE_AUTO_COMMIT_CONFIG, true);
// 自动提交间隔时间1s
props.put(ConsumerConfig.AUTO_COMMIT_INTERVAL_MS_CONFIG, 1000);
// key反序列化指定类
props.put(ConsumerConfig.KEY_DESERIALIZER_CLASS_CONFIG, StringDeserializer.class.getName());
// value反序列化指定类,注意生产者与消费者要保持一致,否则解析出问题
props.put(ConsumerConfig.VALUE_DESERIALIZER_CLASS_CONFIG, StringDeserializer.class.getName());
// 消费者对象
KafkaConsumer<String, String> kafkaConsumer = new KafkaConsumer<>(props);
kafkaConsumer.subscribe(Arrays.asList("test-kafka"));
while (true) {
ConsumerRecords<String, String> records = kafkaConsumer.poll(100);
for (ConsumerRecord<String, String> record : records) {
System.out.printf("offset = %d, value = %s", record.offset(), record.value());
System.out.println();
}
}
}
}
pom.xml引入的依赖:
<?xml version="1.0" encoding="UTF-8"?>
<project xmlns="http://maven.apache.org/POM/4.0.0"
xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"
xsi:schemaLocation="http://maven.apache.org/POM/4.0.0 http://maven.apache.org/xsd/maven-4.0.0.xsd">
<modelVersion>4.0.0</modelVersion>
<groupId>org.example</groupId>
<artifactId>kafka-test</artifactId>
<version>1.0-SNAPSHOT</version>
<properties>
<maven.compiler.source>8</maven.compiler.source>
<maven.compiler.target>8</maven.compiler.target>
</properties>
<dependencies>
<dependency>
<groupId>org.apache.kafka</groupId>
<artifactId>kafka-clients</artifactId>
<version>3.2.0</version>
</dependency>
</dependencies>
</project>
实践
istio
我们选择在 seldon 命名空间下部署模型,所以这里先给 seldon 打个 Istio 自动注入的标签。这使得在 seldon 命名空间下创建的 k8s object 都会被 Istio 自动注入,纳入到 Istio 的管理。这里是为了使用 Istio 的路由管理功能:
kubectl label namespace seldon istio-injection=enabled
为者常成,行者常至
自由转载-非商用-非衍生-保持署名(创意共享3.0许可证)