Viewing Container Logs on the Web UI

Scenario

When yarn.log-aggregation-enable of YARN is set to true, the container log aggregation function is enabled. Log aggregation indicates that after applications are run on YARN, NodeManager aggregates all container logs of the node to HDFS and deletes local logs. For details, see Configuring Container Log Aggregation.

However, all logs will be aggregated to an HDFS directory and can only be viewed by accessing an HDFS file. Open-source Spark and YARN do not support the function of viewing aggregated logs on the web UI.

Spark supports this function. As shown in Figure 1, the AggregatedLogs tab is added to the HistoryServer page. You can click logs to view aggregated logs.

Figure 1 Aggregated log page

Configuration Description

To display logs on the web UI, aggregated logs need to be parsed and presented. Spark parses aggregation logs using JobHistoryServer of Hadoop. Therefore, you can use the spark.jobhistory.address parameter to specify the URL of the JobHistoryServer page to parse and present the logs.

Navigation path for setting parameters:

1. Install the Spark client.

   For details, see Installing a Client.

2. Log in to the Spark client node as the client installation user.

   Set the following parameter in the spark-defaults.conf configuration file of the Spark client, or set the parameter using --conf when submitting the application.

   

   • This function depends on JobHistoryServer of Hadoop. Therefore, ensure that JobHistoryServer is running properly before using the log aggregation function.
   • If the parameter value is empty, the AggregatedLogs tab page still exists, but you cannot view logs by clicking logs.
   • The aggregated container logs can be viewed only when the application is running and event log files of the application exist on HDFS.
   • You can click the log link on the Executors page to view the logs of a running task. After the task completes, the logs are aggregated to HDFS, and the log link on the Executors page becomes invalid. In this case, you can click logs on the AggregatedLogs page to view the aggregated logs.

   Table 1 Parameter description

   Parameter	Description
   spark.jobhistory.address	URL of the JobHistoryServer page. The format is http(s)://ip:port/jobhistory. For example, https://10.92.115.1:26014/jobhistory. The default value is empty, indicating that container aggregation logs cannot be viewed on the web UI. Restart the service for the configuration to take effect.