v4.6.1.7 Release notes
Software version
Release Date: February 17, 2021
See v4.6.1.7 for download information.
Software upgrade support
In Unravel v4.6.1.7, the following path is supported for the RPM upgrade:
4.5.4.0 to 4.6.1.7
Sensor upgrade
A sensor upgrade is mandatory.
Certified platforms
Review your platform's compatibility matrix before you upgrade or install.
Updates to Unravel's configuration properties
Refer to v4.6.x - Updates to Unravel properties.
Installation and Upgrade instructions
Refer to Upgrading Unravel server for instructions to upgrade Unravel.
New features
MapR
For Unravel 4.6.1.x releases, MapR 6.0.1 is tested and is compatible from Unravel version 4.6.1.7 onwards.
Reports
Cluster Impala Workload report is added, which presents the workload details in a heatmap. This report is supported only on CDH platforms.
Improvements and enhancements
Tez tab for Hive is displayed for MapR after you set the following property:
com.unravel.cluster.type=mapr
Unsupported
Hive versions lesser than 1.10 are no longer supported.
Impala is not supported for MapR.
Kerberos is not supported for MapR.
Migration Planning and Capacity Forecasting are not supported for MapR.
Bug fixes
Applications
Some of the Hive apps do not show data/DAG. (ASP-954)
UI
Cluster Impala Workload page is displayed on MapR clusters where Impala is not supported. (UIX-3658)
CDP
In CDP 7.1.5, the rowSizePattern does not match for most of the operators in an Impala query. (ASP-951)
EMR: Hive metrics may not be published in the RUNNING state. (HIVE-135)
INSERT statements are not included in the App count on HDI. (DATAPAGE-256)
The accessed partition does not support Hive on Tez etc. It only supports Hive on MR. (DATAPAGE-250)
Datapage does not load the total table and partition size KPI's (DATAPAGE-379)
Resource metrics will not be collected for JDK versions that are equal to or greater than 9 without using the extra JVM switches as mentioned below:
Workaround: All metrics can be collected if the following options are used for sensor JVM
--add-exports java.base/jdk.internal.perf=ALL-UNNAMED --add-opens jdk.management/com.sun.management.internal=ALL-UNNAMED
For example:
JDK9_MODULE="--add-exports java.base/jdk.internal.perf=ALL-UNNAMED --add-opens jdk.management/com.sun.management.internal=ALL-UNNAMED" ENABLED_SENSOR_FOR_DRIVER="spark.driver.extraJavaOptions=$JDK9_MODULE -Dcom.unraveldata.client.rest.request.timeout.ms=500 -Dcom.unraveldata.client.rest.conn.timeout.ms=500 -Dcom.unraveldata.client.rest.queue=2000 -javaagent:unravel-agent-pack-bin.zip/btrace-agent.jar=libs=spark-2.3,config=driver,stdout=true,debug=true" ENABLED_SENSOR_FOR_EXECUTOR="spark.executor.extraJavaOptions=$JDK9_MODULE -javaagent:unravel-agent-pack-bin.zip/btrace-agent.jar=libs=spark-2.3,config=executor"
These JVM options are valid for JDK9 - JDK15. Without these JVM options, the following metrics are not collected:
ProcessCpuLoad, ProcessCpuTime, SystemCpuLoad
PerfCounters metrics
gc load metric sensor for MR application will not load on EMR.
Spark applications that run in the local mode do not use YARN resources, so these applications will be shown using 0 resources in Unravel. (PLATFORM-2809)
Intermittent issue in fetching the Btrace data for MR applications. (PLATFORM-2807)
Killed MR apps do not display RM diagnostics. (PLATFORM-2815)
Certain users and queues are missing in the Cluster Compare report. (REPORT-1342)
In the Cluster Compare report, the same value trends are highlighted instead of trends with different values. This is fixed in the New UX. (REPORT-1479)
There is a lag seen while streaming Spark SQL applications. (PLATFORM-2764)
For PySpark applications, the processCPUTime and the processCPULoad are not captured properly. (USPARK-626)
The number of TEZ applications in an Oozie workflow is sometimes incorrect. (PLATFORM-2403)
For streaming applications, the global search with application ID does not work. (UIX-3312)