Driven Plugin Guide: Troubleshooting the Driven Agent

version 2.2.6

Driven Plugin Guide: Troubleshooting the Driven Agent

1. Prerequisites: 1.1. Check your team’s demands and expectations

1.2. Check your system privileges and network access
2. Installing the Driven Plugin: 2.1. Downloading

2.2. Using the Driven Plugin
3. Configuring: 3.1. Driven Server Hosts

3.2. API Key

3.3. Retry Rules

3.4. Disk Buffer

3.5. Default Application Tags

3.6. Application Heartbeat Frequency

3.7. Slice Data Suppression

3.8. Slice Data on Completion

3.9. Java Command Suppression

3.10. Failed Flows

3.11. Pending Flows

3.12. Control Plugin Logging

3.13. Disable Plugin

3.14. Plugin Property Reference
4. Advanced Installing and Configuring: 4.1. Scripted Installation

4.2. Integrating the Driven Plugin for Amazon Elastic MapReduce
5. Troubleshooting the Driven Agent: 5.1. Applications cannot send data to the Driven Server

5.2. Driven Agent is making the application seem slower

Troubleshooting the Driven Agent

This Troubleshooting section is a list of tips for solving common issues that users encounter while running applications using the Driven Agent.

If you still have problems after reading this page, feel free to post your queries to Driven Forums or email us at support@concurrentinc.com.

If your issue is not covered here, see the following chapter:

Troubleshooting the Driven Agent

Note	If you are having issues with any application other than Cascading, there may be framework specific configurations that need to be modified.

Applications cannot send data to the Driven Server

This happens when your client application cannot communicate to the Driven Server.

Verify that your Driven application is up by logging in to the web browser interface.
Ensure that the Driven server host location is reachable from your Hadoop cluster. The configured location is stored in the $HADOOP_CONF/cascading-service.properties file.

Driven Agent is making the application seem slower

The Driven Agent and Plugin execute within the application JVM, they do not execute on the cluster or send data from within the cluster.

In some cases, the client side application may take longer to exit than usual creating the perception the client application is running slower with Driven enabled.

In fact, the cluster side computation is un-affected, but before the client side of the application can exit, the Driven Agent and Plugin must complete transmitting data about the completed application.

If the Hadoop JobTracker is under load or there are connectivity issues between the Hadoop JobTracker or the Driven Server, the Agent and Plugin will delay exiting the client JVM.

If the Driven Agent and Plugin are monitoring an application on a very large cluster with many thousands of tasks per job, retrieving data from the Hadoop JobTracker and transmitting this information to the Driven Server can create delays.

You can suppress transmission of the task (Driven uses the term slice) data by setting the following property in the cascading-service.properties file:

driven.protocol.slice.suppress=true

Note

The process of the Driven Agent and Plugin collecting and sending the telemetry data is decoupled from the actual execution of your application on the cluster. While your client application is gated on completion of both the application processing and the Driven Plugin transmission, no SLAs are compromised as a result of the additional latency that may be introduced due to data transmission.