What Can I Do If an Error Occurs when I Access the Application Page Because the Application Cached by HistoryServer Is Recycled?
Question
An error occurs when I access a Spark application page on the HistoryServer page.
Check the HistoryServer logs. The "FileNotFound" exception is found. The related logs are as follows:
2016-11-22 23:58:03,694 | WARN | [qtp55429210-232] | /history/application_1479662594976_0001/stages/stage/ | org.sparkproject.jetty.servlet.ServletHandler.doHandle(ServletHandler.java:628) java.io.FileNotFoundException: ${BIGDATA_HOME}/tmp/spark/jobHistoryTemp/blockmgr-5f1f6aca-2303-4290-9845-88fa94d78480/09/temp_shuffle_11f82aaf-e226-46dc-b1f0-002751557694 (No such file or directory)
Answer
If a Spark application with a large number of tasks is run on the HistoryServer page, the memory overflows to disk and files with the temp_shuffle prefix are generated.
By default, HistoryServer caches 50 Spark applications (determined by the spark.history.retainedApplications configuration item). When the number of Spark applications in the memory exceeds 50, HistoryServer reclaims the first cached Spark application and clears the corresponding temp_shuffle file.
When a user is viewing Spark applications to be recycled, the temp_shuffle file may not be found. As a result, the current page cannot be accessed.
If the preceding problem occurs, use either of the following methods to solve the problem:
- Access the HistoryServer page of the Spark application again. The correct page information is displayed.
- If more than 50 Spark applications need to be accessed at the same time, increase the value of spark.history.retainedApplications.
Log in to FusionInsight Manager and choose Cluster > Services > Spark. Click Configurations then All Configurations. In the navigation tree on the left, click JobHistory and select Page. Then, configure the parameter as follows:
Table 1 Parameter description Parameter
Description
Default Value
spark.history.retainedApplications
Number of Spark applications cached by HistoryServer. When the number of applications to be cached exceeds the value of this parameter, HistoryServer reclaims the first cached Spark application.
50
Feedback
Was this page helpful?
Provide feedbackThank you very much for your feedback. We will continue working to improve the documentation.See the reply and handling status in My Cloud VOC.
For any further questions, feel free to contact us through the chatbot.
Chatbot