ALM-18020 Yarn Task Execution Timeout

Updated on 2022-12-14 GMT+08:00

View PDF

Description

The system checks MapReduce and Spark tasks (except for permanent JDBC tasks) submitted to Yarn every 15 minutes. This alarm is generated when the task execution time exceeds the timeout duration specified by the user. However, the task can be properly executed. The client timeout parameter of MapReduce is mapreduce.application.timeout.alarm and that of Spark is spark.application.timeout.alarm. The unit is ms.

This alarm is cleared when the task is finished or terminated.

Attribute

Alarm ID	Alarm Severity	Auto Clear
18020	Minor	Yes

Parameters

Name	Meaning
Source	Specifies the cluster for which the alarm is generated.
ServiceName	Specifies the service for which the alarm is generated.
RoleName	Specifies the role for which the alarm is generated.
ApplicationName	Specifies the object (application ID) for which the alarm is generated.
Trigger Condition	Specifies the threshold for triggering the alarm.

Impact on the System

The alarm persists after task execution times out. However, the task can still be properly executed, so this alarm does not exert any impact on the system.

Possible Causes

The specified timeout duration is shorter than the required execution time.
The queue resources for task running are insufficient.
Task data skew occurs. As a result, some tasks process a large amount of data and take a long time to execute.

Procedure

Check whether the timeout interval is correctly set.

On FusionInsight Manager, choose O&M. In the navigation pane on the left, choose Alarm > Alarms. The Alarms page is displayed.
Select the alarm whose ID is 18020. In the alarm details, view Location to obtain the timeout task name and timeout duration.
Based on the task name and timeout interval, choose Cluster > Name of the desired cluster > Services > Yarn > ResourceManager (Active) to log in to the native Yarn page. Then find the task on the native page, check its StartTime and calculate the task execution time based on the current system time. Check whether the task execution time exceeds the timeout duration.
- If yes, go to 5.
- If no, go to 10.
Evaluate the expected task execution time based on the service and compare it with the task timeout interval. If the timeout interval is too short, set the timeout interval (mapreduce.application.timeout.alarm or spark.application.timeout.alarm) of the client to the task expected execution time. Run the task again and check whether the alarm is cleared.
- If yes, no further action is required.
- If no, go to 5.

Check whether the queue resources are sufficient.

Find the task on the native page and view the queue name of the task. Click Scheduler on the left of the native page. On the Applications Queues page, find the corresponding queue name and expand the queue details, as shown in the following figure.
Check whether the value of Used Resources in the queue details is approximately equal to the value of Max Resources, which indicates that the resources in the queue submitted by the task have been used up. If the queue resources are insufficient, choose Tenant Resources > Dynamic Resource Plan > Resource Distribution Policy on FusionInsight Manager and increase the value of Max Resources for the queue. Run the task again and check whether the alarm is cleared.
- If yes, no further action is required.
- If no, go to 7.

Check whethercoccurs.

On the native Yarn page, click task ID (for example, application_1565337919723_0002) > Tracking URL:ApplicationMaster > job_1565337919723_0002. The following page is displayed.
Choose Job > Map tasks or Job > Reduce tasks on the left and check whether the execution time of each Map or Reduce task differs greatly. If yes, task data skew occurs. In this case, you need to balance the task data.
Rectify the fault based on the preceding causes and perform the tasks again. Then, check whether the alarm persists.
- If yes, go to 10.
- If no, no further action is required.

Collect the fault information.

On FusionInsight Manager, choose O&M. In the navigation pane on the left, choose Log > Download.
Expand the Service drop-down list, and select Yarn for the target cluster.
Click in the upper right corner, and set Start Date and End Date for log collection to 10 minutes ahead of and after the alarm generation time, respectively. Then, click Download.
Contact O&M personnel and provide the collected logs.