推送度量指标


偶尔你需要监控不能被获取的实例。它们可能被防火墙保护,或者它们生命周期太短而不能通过拉模式获取数据。Prometheus的Pushgateway允许你将这些实例的时间序列数据推送到Prometheus的代理任务中。结合Prometheus简单的文本导出格式,这使得即使没有客户库,也能使用shell脚本获取数据。

Java批量任务例子

这个例子主要说明, 如何执行一个批处理任务,并且在没有执行成功时报警

如果使用Maven,添加下面的代码到pom.xml文件中:

  1. <dependency>
  2. <groupId>io.prometheus</groupId>
  3. <artifactId>simpleclient</artifactId>
  4. <version>0.0.10</version>
  5. </dependency>
  6. <dependency>
  7. <groupId>io.prometheus</groupId>
  8. <artifactId>simpleclient_pushgateway</artifactId>
  9. <version>0.0.10</version>
  10. </dependency>

执行批量作业的代码:

  1. import io.prometheus.client.CollectorRegistry;
  2. import io.prometheus.client.Gauge;
  3. import io.prometheus.client.exporter.PushGateway;
  4. void executeBatchJob() throws Exception {
  5. CollectorRegistry registry = new CollectorRegistry();
  6. Gauge duration = Gauge.build()
  7. .name("my_batch_job_duration_seconds")
  8. .help("Duration of my batch job in seconds.")
  9. .register(registry);
  10. Gauge.Timer durationTimer = duration.startTimer();
  11. try {
  12. // Your code here.
  13. // This is only added to the registry after success,
  14. // so that a previous success in the Pushgateway is not overwritten on failure.
  15. Gauge lastSuccess = Gauge.build()
  16. .name("my_batch_job_last_success_unixtime")
  17. .help("Last time my batch job succeeded, in unixtime.")
  18. .register(registry);
  19. lastSuccess.setToCurrentTime();
  20. } finally {
  21. durationTimer.setDuration();
  22. PushGateway pg = new PushGateway("127.0.0.1:9091");
  23. pg.pushAdd(registry, "my_batch_job");
  24. }
  25. }

警报一个Pushgateway,如果需要的话,修改host和port

如果任务最近没有运行,请创建一个警报到Alertmanager。将以下内容添加到Pushgateway的Prometheus服务的记录规则中:record rules ALERT MyBatchJobNotCompleted IF min(time() - my_batch_job_last_success_unixtime{job="my_batch_job"}) > 60 * 60 FOR 5m WITH { severity="page" } SUMMARY "MyBatchJob has not completed successfully in over an hour"