Known Issues at Release (Release 6.2.0)

You may encounter the following known issues after upgrading to Release 6.2. This list is current as of the release date.

Where available, the workaround for an issue is also documented in this topic. HPE regularly releases maintenance releases and patches to fix issues. We recommend checking the release notes for any subsequent maintenance releases to see if one or more of these issues are fixed.

File Migrate

CORE-645: The FileMigrate feature does not work with HPE Ezmeral Data Fabric versions 6.2, and 6.2.1. The FileMigrate feature needs to be upgraded to support JDK 11 and AWS S3 1.11.X versions.

Installation and Configuration Issues

You can see generic installation issues here: Installer Known Issues.

SPYG-1136

During a manual installation or upgrade, Collectd provided in core 6.1.0 won't start on RHEL / CentOS 8.2 because it expects the Python 2 libraries to be installed, and RHEL / CentOS 8.2 provides the Python 3 libraries instead. This issue does not affect installations or upgrades performed using the Installer.

Workaround: Before installing the monitoring components, check to see if Python 2 is installed. If the following error is generated, try installing Python 2 on RHEL / CentOS 8.2:

failed: libpython2.7.so.1.0: cannot open shared object file

IN-2637: After a manual installation, Oozie and Hive services can fail to connect to a MySQL or MariaDB database because the server time-zone value is unrecognized or represents more than one time zone. The issue affects your installation if you applied the mapr-patch released on or after February 21, 2021 (including the latest mapr-patch). This issue affects manual installations but is fixed in Installer 1.14.0.0.; Workaround: For manual installations, you must configure either the server or JDBC driver (using the serverTimezone configuration property) to use a more specific time-zone value if you want to utilize time-zone support. After running configure.sh but before starting the Oozie or Hive services, update the serverTimezone parameter in the hive-site.xml or oozie-site.xml. For more information, see MySQL Bug #95036.

IN-2935

On RHEL or CentOS 8.3, new installations using the Installer can fail with the following error message:

mount.nfs: access denied by server while mounting localhost:/mapr

This happens when the Installer cannot start the mapr-loopbacknfs service because the RHEL or CentOS NFS/NFS4 service is running.

Workaround: Edit the /etc/systemd/system/mapr-loopbacknfs.service file to add the following Conflicts directive to the nfs-mountd.service:

[Unit]
Description=MapR Technologies, Inc. loopbacknfs service
After=rc-local.service
After=network.target syslog.target
Conflicts=nfs-mountd.service

The Conflicts command stops nfs-mountd before installation so it cannot interfere with starting mapr-loopbacknfs. After editing the loopbacknfs.service file, perform a daemon reload using the following command, and then retry the installation:

systemctl daemon-reload

IN-2947: For Installer 1.16.0 on Ubuntu 18.04, the Extend Cluster operation fails for clusters larger than three nodes. The operation fails on nodes where ZooKeeper is not installed. The failure occurs because the Installer attempts to update the ZooKeeper service file on a node that has no roles file for ZooKeeper.; Workaround: Make sure ZooKeeper is running on every node in the cluster, and retry the Extend Cluster operation.

Monitoring Issues

ES-77, FLUD-55

During an upgrade from EEP 6.x to EEP 7.0.0 or EEP 7.0.1, some monitoring components do not get updated because of an error in the fourth digit of the package version. This issue can occur during manual upgrades or upgrades performed using the Installer. The affected components can include any or all of the following:

Elasticsearch
Fluentd
Grafana
Kibana

See Reinstalling Monitoring Components After an Upgrade.

Performance Issues

MFS-10838

In release 6.2.0, the default setting for Java garbage collection (GC) is ParallelGC. Garbage-collection settings can influence performance. If you notice a difference in the performance of Spark or MapReduce jobs as compared to release 6.1.0, try reverting to the G1GC setting, which is the Java 11 default. You can do this by overwriting the Java properties for the am, map, and reduce containers in the mapred-site.xml file:

<property>
  <name>yarn.app.mapreduce.am.command-opts</name>
  <value>-Xmx1024m --add-opens java.base/java.lang=ALL-UNNAMED</value>
</property>

<property>
  <name>mapreduce.map.java.opts</name>
  <value>-Xmx900m --add-opens java.base/java.lang=ALL-UNNAMED</value>
</property>

<property>
  <name>mapreduce.reduce.java.opts</name>
  <value>-Xmx2560m --add-opens java.base/java.lang=ALL-UNNAMED</value>
</property>

To change the garbage-collection (GC) default setting:

Comment out the following lines in hadoop_home/etc/hadoop/yarn-env.sh and mapred-env.sh:

YARN_OPTS="$YARN_OPTS -XX:+UseParallelGC"

HADOOP_OPTS="$HADOOP_OPTS -XX:+UseParallelGC"

Restart Hadoop services, as described in node services.

For more information about garbage collection, see Types of Java Garbage Collectors.

Timeline Server

See Hive-on-Tez User Interface Known Issues.

Permission Issues

MFS-6776: Teragen jobs fail to complete when run as the root user, because of a permission issue on the application directory. The directory should have permission-mode bits 750 and not 700 as it is at present.
Workaround: Set the permission-mode bits on this directory to 750.

Control System Issues

MON-7568: If you are using SPNEGO authentication, Control System (MCS) UI is irresponsive after the session timeout.; Workaround: Use a new browser tab after the session timeout.

MON-5653

The system displays an incorrect number of failed nodes for the NFSv4 service on the Services page when the failed node count is clicked.

Workaround: Use the maprcli node list command to get the node count.

MAPRDB-2309

Setting permissions on Column Families from the Control System causes issues with table fields.

Workaround: Set permissions using the maprcli table cf colperm set command directly.

CORE-538

ACL is not removed when using the Control System to delete an ACL from a volume with multiple ACLs.

Workaround: None

MON-5933

Table creation from the Control System fails when both public and a user are selected for permissions.

Workaround: None

MapR-DB Issues

MAPRDB-2529

If you run the following commands without a license, the commands hang:

maprcli stream replica add
maprcli stream edit (hangs when you run it with the -compact option)

Workaround: Add a license as described in Adding a License.

MAPRDB-2092

In the HPE Ezmeral Data Fabric Database, adding a table index or replicating a table fails if the cluster administrator (MAPR_USER) does not have write access to the parent volume of the table.

Workaround: For more information, see this support advisory. If a patch is not available or the available patch has not been applied, another workaround is to add the MAPR_USER to the writeACE for the table parent volume.