tencent cloud

Overview

When using Flink, you need to monitor its task running status to know whether the tasks run normally and troubleshoot faults. TMP integrates Pushgateway to allow Flink to write metrics and provides an out-of-the-box Grafana monitoring dashboard for it.

Prerequisites

1. The EMR product you purchased includes the Flink component, and a Flink task is running in your instance.
2. You have created a TKE cluster in the region and VPC of your TMP instance.

Directions

Integrating product

Getting Pushgateway access configuration

1. Log in to the EMR console, select the corresponding instance, and select Basic Info > Instance Info to get the Pushgateway address and token.
   
2. Get the APPID on the Account Info page.

Modifying Flink configuration

1. Log in to the EMR console, select the corresponding instance, and select Cluster Service.
2. Find the Flink configuration item and select Configuration Management in the Operation column on the right to enter the configuration management page.
3. On the right of the page, click Add Configuration Item and add the following configuration items one by one:

   Configuration Item	Default Value	Data Type	Description	Suggestion
   metrics.reporter.promgateway.class	None	String	Name of the Java class for exporting metrics to Pushgateway	-
   metrics.reporter.promgateway. jobName	None	String	Push task name	Specify an easily understandable string
   metrics.reporter.promgateway. randomJobNameSuffix	true	Boolean	Whether to add a random string after the task name	Set it to `true`. If no random string is added, metrics of different Flink tasks will overwrite each other
   metrics.reporter.promgateway. groupingKey	None	String	Global label added to each metric in the format of `k1=v1;k2=v2`	Add the EMR instance ID to distinguish between the data of different instances, such as `instance_id=emr-xxx`
   metrics.reporter.promgateway. interval	None	Time	Time interval for pushing metrics, such as 30s	We recommend you set the value to about 1 minute
   metrics.reporter.promgateway.host	None	String	Pushgateway service address	It is the service address of the TMP instance in the console
   metrics.reporter.promgateway.port	-1	Integer	Pushgateway service port	It is the port of the TMP instance in the console
   metrics.reporter.promgateway. needBasicAuth	false	Boolean	Whether the Pushgateway service requires authentication	Set it to `true`, as the Pushgateway of TMP requires authentication
   metrics.reporter.promgateway.user	None	String	Username for authentication	It is your `APPID`
   metrics.reporter.promgateway. password	None	String	Password for authentication	It is the access token of the TMP instance in the console
   metrics.reporter.promgateway. deleteOnShutdown	true	Boolean	Whether to delete the corresponding metrics on the Pushgateway after the Flink task is completed	Set it to `true`

   Below is a sample configuration:

   metrics.reporter.promgateway.class: org.apache.flink.metrics.prometheus.PrometheusPushGatewayReporter
   metrics.reporter.promgateway.jobName: climatePredict
   metrics.reporter.promgateway.randomJobNameSuffix:true
   metrics.reporter.promgateway.interval: 60 SECONDS
   metrics.reporter.promgateway.groupingKey:instance_id=emr-xxxx
   metrics.reporter.promgateway.host: 172.xx.xx.xx
   metrics.reporter.promgateway.port: 9090
   metrics.reporter.promgateway.needBasicAuth: true
   metrics.reporter.promgateway.user: appid
   metrics.reporter.promgateway.password: token
   

Installing Flink Pushgateway plugin

The Pushgateway plugin in the official package currently does not support configuring the authentication information, but TMP requires authentication before data can be written. Therefore, we recommend you use the JAR package we provide. We have also submitted a pull request for supporting authentication to the Flink team.

1. To prevent class conflicts, if you have already used the official Flink plugin, run the following command to delete it first:

   cd /usr/local/service/flink/lib
   rm flink-metrics-prometheus*jar
   

2. In the EMR console, select the corresponding instance, and select Cluster Resource > Resource Management > Master to view the master node.
3. Click the instance ID to go to the CVM console, log in to the CVM instance, and run the following command to install the plugin:

   cd /usr/local/service/flink/lib
   wget https://rig-1258344699.cos.ap-guangzhou.myqcloud.com/flink/flink-metrics-prometheus_2.11-auth.jar -O flink-metrics-prometheus_2.11-auth.jar
   

Verifying

1. Run the flink run command on the master node to submit a new task and view the task log:

   grep metrics /usr/local/service/flink/log/flink-hadoop-client-*.log
   

2. If the log contains the following content, the configuration is successfully loaded:
   

   Note：

   As tasks previously submitted in the cluster use the old configuration file, their metrics are not reported.

Viewing monitoring information

1. In Integration Center in the target TMP instance, find Flink monitoring, install the corresponding Grafana dashboard, and then you can enable the Flink monitoring dashboard.
2. Enter Grafana and click to expand the Flink monitoring panel.
   
3. Click Flink Job List to view the monitoring information.
   
4. Click a job name or job ID in the table to view the job monitoring details.
   
5. Click Flink Cluster in the top-right corner to view the Flink cluster monitoring information.
   
6. Click a task name in the table to view the task monitoring details.
   

Integrating with alert feature

1. Log in to the TMP console and select the target TMP instance to enter the management page.
2. Click Alerting Rule and add the corresponding alerting rules. For more information, please see Creating Alerting Rule.