【Flink系列二】构建实时计算平台——特别篇,用InfluxDb收集Flink Metrics

Influxdb 快速入门

原文地址:https://www.cnblogs.com/slankka/p/13865338.html

从Docker启动 Influxdb

docker pull influxdb:LATEST

docker run -d --name influxdb -p 8086:8086 \
      -v /opt/work/influxdb:/var/lib/influxdb \
      influxdb

进入Influxdb的Client

# docker exec -it influxdb influx
Connected to http://localhost:8086 version 1.8.3
InfluxDB shell version: 1.8.3

>create database flink # 创建Flink数据库

>use flink #为FLink创建RETENTION_POLICY(1)

>CREATE RETENTION POLICY one_hour ON flink DURATION 1h REPLICATION 1 #为FLink创建RETENTION_POLICY(2)

配置Flink

Flink ifluxdb reporter

修改FLINK_CONF

metrics.reporter.influxdb.class: org.apache.flink.metrics.influxdb.InfluxdbReporter
metrics.reporter.influxdb.host: <IP>
metrics.reporter.influxdb.port: 8086
metrics.reporter.influxdb.db: flink
metrics.reporter.influxdb.username: 
metrics.reporter.influxdb.password: 
metrics.reporter.influxdb.retentionPolicy: one_hour

以该配置启动的Flink作业,Flink会自动将指标写入Influxdb

【可选】安装Chronograf可视化界面

docker pull chronograf:LATEST
docker run --name chronograf -d -p 8888:8888 -v /opt/work/chronograf:/var/lib/chronograf chronograf

执行SQL可以查到CheckpointExternalPath

SELECT * FROM "flink"."one_hour"."jobmanager_job_lastCheckpointExternalPath"

效果图

Chronograf

优化

监控指标,一般使用Prometheus来做,而根据我的需求和实践来看,Influxdb仅用来接收lastCheckpointExternalPath这个指标。
经过大约半年多的观察,Influxdb 1.8,100个作业的情况下, 内存占用峰值会超过20GB,这个时候容器会自动重启,客户端无法上报。

因此需要对influxdb进行优化。这里记录一种最简单的优化,那就是直接减少指标数量:

package org.apache.flink.metrics.influxdb;

abstract class AbstractReporter<MetricInfo> implements MetricReporter {
	protected final Logger log = LoggerFactory.getLogger(getClass());

	protected final Map<Gauge<?>, MetricInfo> gauges = new HashMap<>();
	protected final Map<Counter, MetricInfo> counters = new HashMap<>();
	protected final Map<Histogram, MetricInfo> histograms = new HashMap<>();
	protected final Map<Meter, MetricInfo> meters = new HashMap<>();
	protected final MetricInfoProvider<MetricInfo> metricInfoProvider;

	protected AbstractReporter(MetricInfoProvider<MetricInfo> metricInfoProvider) {
		this.metricInfoProvider = metricInfoProvider;
	}

	@Override
	public void notifyOfAddedMetric(Metric metric, String metricName, MetricGroup group) {
		if (!metricName.equals("lastCheckpointExternalPath")) {
			return;
		}

经过验证,Flink 使用此Reporter,仅上报这一个指标。

posted @ 2020-10-23 17:38  一杯半盏  阅读(2200)  评论(0编辑  收藏  举报