背景

由于公司想要基于flink的日志做实时预警功能，故需要实时接入，并刷入es进行分析。

注意点

日志接入必须异步，不能影响服务性能

kafka集群宕机，依旧能够提交flink任务且运行任务

kafka集群挂起恢复，可以依旧续写实时运行日志

自定义Appender

在类上加上@Plugin注解，标记为自定义appender

@Plugin(name = "KafkaAppender", category = "Core", elementType = "appender", printObject = true)
public final class KafkaAppender extends AbstractAppender {}

在类加上@PluginFactory和@PluginAttribute来配合log4j.properties来传递参数

@PluginFactory
    public static KafkaAppender createAppender(
            /** 发送到的Topic */
            @PluginAttribute("topic") String topic,
            /** Kafka地址 */
            @PluginAttribute("kafkaBroker") String kafkaBroker,
            /** 设置的数据格式Layout */
            @PluginElement("Layout") Layout<? extends Serializable> layout,
            @PluginAttribute("name") String name,
            @PluginAttribute("append") boolean append,
            /** 日志等级 */
            @PluginAttribute("level") String level,
            /** 设置打印包含的包名，前缀匹配，逗号分隔多个 */
            @PluginAttribute("includes") String includes,
            /** 设置打印不包含的包名，前缀匹配,同时存在会被排除，逗号分隔多个 */
            @PluginAttribute("excludes") String excludes) {
        return new KafkaAppender(name, topic, kafkaBroker, null, layout, append, level, includes, excludes);
    }

在append中对每一条日志进行处理

    @Override
    public void append(LogEvent event) {
        if (event.getLevel().isMoreSpecificThan(this.level)) {
            if (filterPackageName(event)) {
                return;
            }
            try {
                if (producer != null) {
                    CompletableFuture.runAsync(() -> {
                        producer.send(new ProducerRecord<String, String>(topic, getLayout().toSerializable(event).toString()));
                    }, executorService);
                }
            } catch (Exception e) {
                LOGGER.error("Unable to write to kafka for appender [{}].", this.getName(), e);
                LOGGER.error("Unable to write to kafka in appender: " + e.getMessage(), e);
            } finally {
            }
        }
    }

源码地址https://gitee.com/czshh0628/realtime-log-appender

log4j配置文件

日志接入kafka

在flink的conf目录的log4j.properties里添加如下配置

# 自定义的Kafka配置
rootLogger.appenderRef.kafka.ref=KafkaAppender
appender.kafka.type=KafkaAppender
appender.kafka.name=KafkaAppender
# 日志发送到的Topic
appender.kafka.topic=cdc
# Kafka Broker
appender.kafka.kafkaBroker=xxx:9092,xxx:9092
# kerberos认证
http://appender.kafka.keyTab=xxx
http://appender.kafka.principal=xxx
# 发送到Kafka日志等级
appender.kafka.level=info
# 过滤指定包名的文件
appender.kafka.includes=com.*,org.apache.hadoop.yarn.client.*,org.*
## kafka的输出的日志pattern
appender.kafka.layout.type=PatternLayout
appender.kafka.layout.pattern={"logFile":"${sys:log.file}","taskId":"${sys:taskId}","taskVersion":"${sys:taskVersion}","logTime":"%d{yyyy-MM-dd HH:mm:ss,SSS}","logMsg":"%-5p %-60c %x - %m","logThrow":"%throwable"}

日志接入elasticsearch

# 自定义的es的配置
rootLogger.appenderRef.es.ref=EsAppender
appender.es.type=EsAppender
appender.es.name=EsAppender
appender.es.hostname=bigdata
appender.es.port=9200
appender.es.index=flink_logs
appender.es.fetchSize=100
appender.es.fetchTime=5000
appender.es.level=info
appender.es.includes=com.*,org.apache.hadoop.yarn.client.*,org.*
appender.es.layout.type=PatternLayout
appender.es.layout.pattern={"logFile":"${sys:log.file}","taskId":"${sys:taskId}","taskVersion":"${sys:taskVersion}","logTime":"%d{yyyy-MM-dd'T'HH:mm:ss,SSS'Z'}","logMsg":"%-5p %-60c %x - %m","logThrow":"%throwable"}

启动脚本

bin/flink run -m yarn-cluster -p 1 -yjm 1024 -ytm 1024 -ys 1 -yD env.java.opts="-DtaskId=1000 -DtaskVersion=1.0" -c czs.study.flinkcdc.mysql.MySqlCdcStream test.jar

实现效果