Monitoring the state of Spark applications

To evaluate the performance of Spark applications in a Yandex Data Processing cluster, you can check the following:

• Application list
• Application logs
• Application queue
• Application details
• Resources allocated to the application
• Persisted RDDs
• List of SQL queries and their execution plans

Checking the application listChecking the application list

1. Go to the folder page and select Yandex Data Processing.
2. Click the cluster name.
3. Under UI Proxy, select the YARN Resource Manager Web UI interface.

It shows information about all running and completed applications.

Checking application logsChecking application logs

1. Go to the folder page and select Yandex Data Processing.

2. Click the cluster name.

3. Under UI Proxy, select the YARN Resource Manager Web UI interface.

4. Find the application you need and click its ID in the ID column.

   This will open a window with information about the application's performance and a table with a list of application run attempts.

5. Click the link next to the attempt in question in the Logs column.

Checking the application queueChecking the application queue

1. Go to the folder page and select Yandex Data Processing.
2. Click the cluster name.
3. Under UI Proxy, select the YARN Resource Manager Web UI interface.
4. In the left-hand menu, go to Scheduler.

The Application Queues section shows the queue of applications and resources used by them.

Checking application detailsChecking application details

This will open the Spark History Server Web UI window with details of the selected application:

• Event Timeline: History of job runs with info about added and removed executors
• Active Jobs: List of jobs being run or waiting to be run.
• Completed Jobs: List of finished jobs.

For each job, the table specifies:

• Start time (Submitted)
• Duration
• Stages: Succeeded/Total
• Tasks: Succeeded/Total

Checking resources allocated to the applicationChecking resources allocated to the application

1. Go to the folder page and select Yandex Data Processing.
2. Click the cluster name.
3. Under UI Proxy, select the Spark History Server Web UI interface.
4. In the top menu, go to Executors.

The UI will display two tables:

• Summary: High-level information, such as the number and status of executors and resources used.
• Executors: Information about each executor.

The tables specify the following:

• Amount of resources available per resource executor.
• Number of running and completed tasks.
• Task duration (Task Time), including the time spent for garbage collection (GC Time).

Checking persisted RDDsChecking persisted RDDs

1. Go to the folder page and select Yandex Data Processing.
2. Click the cluster name.
3. Under UI Proxy, select the Spark History Server Web UI interface.
4. In the top menu, go to Storage.

The UI displays the list of cacheable tables (RDDs). For each RDD, it shows information about the used memory and disk space, as well as caching progress.

To view detailed statistics, click the RDD name.

Checking the list of SQL queries and their execution plansChecking the list of SQL queries and their execution plans

1. Go to the folder page and select Yandex Data Processing.
2. Click the cluster name.
3. Under UI Proxy, select the Spark History Server Web UI interface.
4. In the top menu, go to SQL.

The table lists executed SQL queries, including their start time and duration.

To see the query execution plan, click the query text in the Description column. The query execution plan is displayed as a flowchart. To view it as text, click Details at the bottom of the figure.

The query execution plan contains statistics for each operator with the number of completed tasks and their duration. If the query is still running, the current statistics will be shown.