Spark应用监控解决方案--使用Prometheus和Grafana监控Spark应用

Spark任务启动后，我们通常都是通过跳板机去Spark UI界面查看对应任务的信息，一旦任务多了之后，这将会是让人头疼的问题。如果能将所有任务信息集中起来监控，那将会是很完美的事情。

通过Spark官网指导文档，发现Spark只支持以下sink

Each instance can report to zero or more sinks. Sinks are contained in the org.apache.spark.metrics.sink package:

• ConsoleSink: Logs metrics information to the console.
• CSVSink: Exports metrics data to CSV files at regular intervals.
• JmxSink: Registers metrics for viewing in a JMX console.
• MetricsServlet: Adds a servlet within the existing Spark UI to serve metrics data as JSON data.
• GraphiteSink: Sends metrics to a Graphite node.
• Slf4jSink: Sends metrics to slf4j as log entries.
• StatsdSink: Sends metrics to a StatsD node.

没有比较常用的Influxdb和Prometheus ~~~

谷歌一把发现要支持influxdb需要使用第三方包，比较有参考意义的是这篇，Monitoring Spark Streaming with InfluxDB and Grafana ,在提交任务的时候增加file和配置文件，但成功永远不会这么轻松。。。

写入influxdb的数据都是以application_id命名的，类似这种application_1533838659288_1030_1_jvm_heap_usage，也就是说每个任务的指标都是在单独的表，最终我们展示在grafana不还得一个一个配置么？

显然这个不是我想要的结果，最终目的就是：一次配置后每提交一个任务自动会在监控上看到。

谷歌是治愈一切的良药，终究找到一个比较完美的解决方案，就是通过graphite_exporter中转数据后接入Prometheus，再通过grafana展示出来。

所以，目前已经实践可行的方案有两个

方案一：

监控数据直接写入influxdb，再通过grafana读取数据做展示，步骤如下：

1.在spark下 conf/metrics.properties 加入以下配置

master.source.jvm.class=org.apache.spark.metrics.source.JvmSource
worker.source.jvm.class=org.apache.spark.metrics.source.JvmSource
driver.source.jvm.class=org.apache.spark.metrics.source.JvmSourc
executor.source.jvm.class=org.apache.spark.metrics.source.JvmSource

*.sink.influx.class=org.apache.spark.metrics.sink.InfluxDbSink
*.sink.influx.protocol=http
*.sink.influx.host=xx.xx.xx.xx
*.sink.influx.port=8086
*.sink.influx.database=sparkonyarn
*.sink.influx.auth=admin:admin

2.在提交任务的时候增加以下配置，并确保以下jar存在

--files /spark/conf/metrics.properties \

--conf spark.metrics.conf=metrics.properties \
--jars /spark/jars/metrics-influxdb-1.1.8.jar,/spark/jars/spark-influx-sink-0.4.0.jar \
--conf spark.driver.extraClassPath=metrics-influxdb-1.1.8.jar:spark-influx-sink-0.4.0.jar \
--conf spark.executor.extraClassPath=metrics-influxdb-1.1.8.jar:spark-influx-sink-0.4.0.jar

缺点：application_id发生变化需要重新配置grafana

方案二（目前在用的）：

通过graphite_exporter将原生数据通过映射文件转化为有 label 维度的 Prometheus 数据

1.下载graphite_exporter，解压后执行以下命令，其中graphite_exporter_mapping需要我们自己创建，内容为数据映射文件

nohup ./graphite_exporter --graphite.mapping-config=graphite_exporter_mapping &

例如

mappings:
- match: '*.*.jvm.*.*'
  name: jvm_memory_usage
  labels:
    application: $1
    executor_id: $2
    mem_type: $3
    qty: $4

会将数据转化成 metric name 为 jvm_memory_usage，label 为 application，executor_id，mem_type，qty 的格式。

application_1533838659288_1030_1_jvm_heap_usage -> jvm_memory_usage{application="application_1533838659288_1030",executor_id="driver",mem_type="heap",qty="usage"}

2.配置 Prometheus 从 graphite_exporter 获取数据，重启prometheus

/path/to/prometheus/prometheus.yml

scrape_configs:
  - job_name: 'spark'
    static_configs:
    - targets: ['localhost:9108']

3.在spark下 conf/metrics.properties 加入以下配置

master.source.jvm.class=org.apache.spark.metrics.source.JvmSource
worker.source.jvm.class=org.apache.spark.metrics.source.JvmSource
driver.source.jvm.class=org.apache.spark.metrics.source.JvmSourc
executor.source.jvm.class=org.apache.spark.metrics.source.JvmSource

*.sink.graphite.class=org.apache.spark.metrics.sink.GraphiteSink
*.sink.graphite.protocol=tcp
*.sink.graphite.host=xx.xx.xx.xx
*.sink.graphite.port=9109
*.sink.graphite.period=5
*.sink.graphite.unit=seconds

4.提交spark任务的时候增加  --files /spark/conf/metrics.properties

5.最后在grafana创建prometheus数据源，创建需要的指标，最终效果如下，有新提交的任务不需要再配置监控，直接选择application_id就可以看对应的信息

需要用到的jar包

https://repo1.maven.org/maven2/com/izettle/metrics-influxdb/1.1.8/metrics-influxdb-1.1.8.jar

https://mvnrepository.com/artifact/com.palantir.spark.influx/spark-influx-sink

模板

mappings:
- match: '*.*.executor.filesystem.*.*'
  name: filesystem_usage
  labels:
    application: $1
    executor_id: $2
    fs_type: $3
    qty: $4

- match: '*.*.executor.threadpool.*'
  name: executor_tasks
  labels:
    application: $1
    executor_id: $2
    qty: $3

- match: '*.*.executor.jvmGCTime.count'
  name: jvm_gcTime_count
  labels:
    application: $1
    executor_id: $2

- match: '*.*.executor.*.*'
  name: executor_info
  labels:
    application: $1
    executor_id: $2
    type: $3
    qty: $4

- match: '*.*.jvm.*.*'
  name: jvm_memory_usage
  labels:
    application: $1
    executor_id: $2
    mem_type: $3
    qty: $4

- match: '*.*.jvm.pools.*.*'
  name: jvm_memory_pools
  labels:
    application: $1
    executor_id: $2
    mem_type: $3
    qty: $4

- match: '*.*.BlockManager.*.*'
  name: block_manager
  labels:
    application: $1
    executor_id: $2
    type: $3
    qty: $4

- match: '*.driver.DAGScheduler.*.*'
  name: DAG_scheduler
  labels:
    application: $1
    type: $2
    qty: $3

- match: '*.driver.*.*.*.*'
  name: task_info
  labels:
    application: $1
    task: $2
    type1: $3
    type2: $4
    qty: $5

graphite_exporter_mapping

参考资料

https://github.com/palantir/spark-influx-sink

https://spark.apache.org/docs/latest/monitoring.html

https://www.linkedin.com/pulse/monitoring-spark-streaming-influxdb-grafana-christian-g%C3%BCgi

https://github.com/prometheus/prometheus/wiki/Default-port-allocations

https://github.com/prometheus/graphite_exporter

https://prometheus.io/download/

https://rokroskar.github.io/monitoring-spark-on-hadoop-with-prometheus-and-grafana.html

https://blog.csdn.net/lsshlsw/article/details/82670508

https://www.jianshu.com/p/274380bb0974