Resolved Issues
The following MapR issues, which were reported by customers, are resolved in Version 5.1.0.
Installation and Configuration
14230: The difficulty of verifying that files were installed by a MapR package has
been resolved. On the Ubuntu platform, you can check the md5sums against the installed files
by running src/support/tools/verifySum.sh
. On the RHEL platform, you can
use sha256sum
against the installed files by running
verifySum.sh
.
19532: Web browsers no longer fail to connect to the MCS due to a Weak Ephemeral Diffie-Hellman Key.
19742: Restarting warden no longer resets permissions on the root directory to
777
.
Failed to move central config file, error Invalid cross-device link.
19790: The warden service starts successfully when you reboot the operating system
with the abrt-ccpp
component enabled.
19808: When you install MapR, the TLSv1 protocol is now automatically disabled for services that run a web server.
20543: When you install MapR on a CentOS node, the nfs-utils
package will be installed with the mapr-nfs
package if it is already not
installed on the system.
20695: When you install MapR on a CentOS node, nss version 3.19 is installed as a dependency to prevent SSL connection errors.
20916: Fixed the issue causing a map task to report a successful completion instead of a failure when the combiner thread fails with an out-of-memory error.
21134: When you install MapR on a node, the tcp_syn_retries
parameter is now automatically set to 4
.
21147: hdfs-default.xml
and hdfs-site.xml
are now
included in the hadoop-hdfs-*.jar
to resolve the issue where Oozie is
unable to run hadoop commands when centralized logging is enabled.
21230: You can now enable insecure protocols, such as TLSv1 and SSLv3, that MapR disables by default.
21523: For systems with SSDs, MapR now automatically configures the Linux IO
scheduler for the drives in a storage pool based on the disk type. That is, on SSD-based
systems, MapR will set the scheduler type to NOOP
.
MapR-DB
11646: Improved auto-compaction to compact segments that have completely expired and to mark partitions in need of TTL compaction.
19388: Added C-API functionality that allows the scanner to return results for a particular timestamp or between two timestamps.
19430, 21638: MapR-DB applications failed with a segmentation fault because of memory corruptions in the C API.
19501: Truncating a table with the HBase shell truncate command no longer leads to a Stale File Handle error when the table is subsequently accessed.
MapR-FS
11828: Fixed the issue that caused write operations on a volume to hang.
14447: Fixed the issue that caused NFS client IOs to break during a VIP failover.
14482: MapR-FS native C APIs now support impersonation.
16679: Fixed the issue that caused CLDB to consider itself unstable and go into restart mode as a result of unknown containers in memory.
17298: Fixed the issue that caused a memory leak after closing a file using the MapR-FS C API.
17642: Fixed the issue that caused space reservation for container resynchronization to be out of sync with the amount of space needed.
17762: Fixed the issue where the desksetup script does not populate
/lib/udev/rules.d/99-mapr-disk.rules
and the disk setup fails due to
wrong device permissions.
18193: Fixed the issue that caused read pages to be reused after they were freed.
18981: Application calls to hdfsGetHosts()
no longer fail with a
“could not get fidmap” error message.
19027: ShimLoader now uses a NULL value when the Thread Context Class Loader (TCCL) is not set.
19231: Fixed the issue that caused the client to crash while handling a mix of compressed and uncompressed data in the same write request.
19568: Fixed the issue that caused mirroring to fail with the following error:
MapInCidsToMirrorCids failed for volume
19628: The mapr user now has the permission necessary to stop/start the fileserver in order to add disks from the MCS or maprcli.
20377: Fixed the issue that caused a buffer overflow even when a limit was set on Bytebuffer.
20648: When you put a node into maintenance mode, the
-timeoutminutes
setting no longer times out before the set timeout
duration.
20671: Fixed the issue that caused mirror intra-volume resynchronization to use slow resync slots (rather than fast resync slots) when there is no data change.
20816: A JBoss application was failing with the following error:
java.lang.UnsatisfiedLinkError:
com.mapr.security.JNISecurity.SetClusterOption
.
20843: Fixed the issue related to MFS crashing, which caused multiple MFS servers to go down and volume data to be unavailable until the MFS nodes were restarted.
21041: Increasing the RpcTimeout value in the core.xml
file will no
longer disable hardmount, which allows the FileClient to try all CLDBs.
21171: Fixed the issue related to MFS restarting many times with an "mfs is potentially deadlocked" error.
21224: The snapshot delete operation will now be throttled to prevent it from competing with other operations.
21259: Fixed the issue that caused MFS to crash on multiple nodes while restoring a volume dump.
21335: When a mirror volume was created on a separate cluster from the source volume, a dump of the mirror was created, and the dump was copied to the source cluster, an attempted restore of the dump would hang. This sequence of operations is successful in Version 5.1.
21372: The issue causing degradation in performance during a scheduled snapshot delete operation has been fixed.
21407: A restore of a dump from a standard volume no longer hangs. MFS authentication issues that caused this problem have been resolved.
21696: The TOO_MANY_CONTAINERS
alarm will now be raised when there
are containers above the threshold limit. With this fix:
- The threshold limit for generating the alarm is now on a per-node basis, but must be configured cluster-wide.
- The threshold limit for RW containers can be configured and is, by default, 50K. The threshold limit for RW and snapshot containers combined is 10 times the threshold limit for RW containers (for example, 10x50K=500K).
- You can no longer set and retrieve maxContainers using the
maprcli node modify
command.
21900: In the event of a node restart during a snapshot schedule, MFS will now keep accurate count of stale containers to ensure snapshot schedules run smoothly.
NFS
17115: When the debug log level was set in the nfsserver.conf
file,
this value did not take effect. In Version 5.1, the log level is read from the configuration
file.
18153: During a manual failover operation, NFS response time increased when the OS was rebooted.
19939: Base and POSIX client licenses can now be applied even if a partner package is installed before initializing the cluster.
20013: The nfsserver.log
file was not being re-created after being
removed at the end of its retention time. This file is now rolled over periodically or when
its size exceeds the configured value of 1GB.
20322: Fixed the issue related to the GID list that caused MFS to return an error. The NFS server was not adding the GID sent by the NFS client to its list of GIDs.
20514: Reduced the latency that NFS clients experience in the event of an uncontrolled node failure.
CLDB
19592: When a volume name is provided as input, MapR now validates the input, rejecting invalid input such as a script, which prevents an alert from appearing on the browser.
19743: The maprcli alarm
commands no longer return unsupported
system alarms, and the CLDB no longer raises the unsupported alarms.
POSIX Client
19978: Fixed the issue related to latency that POSIX clients experience in the event of an uncontrolled node failure.
21337: When the mapr-loopbacknfs
service was restarted, its status
was reported as FAILED
. This misleading status message is no longer
printed.
20558: When you remove nodes with the fileserver and nfsserver roles using the MapR
Control System or the maprcli node remove –hostids <hostnames> -service
<nfsserver>
command, the nodes no longer appear in the NFS nodes tab of the
MCS or when you run the maprcli node list –nfsnodes 1
command.
21277: Fixed the issue related to the shared memory segment that prevented the
loopbacknfs
service from starting after installation.
YARN
18672: YARN applications no longer fail when label-based scheduling and time-based resource reservations are enabled at the same time.
20026: Resolved the issue related to
UserGroupInformation.initialize()
attempting to resolve KDC realm even in
unsecured mode. This resulted in applications that are using Hadoop to appear to hang if the
network card was disconnected.
20287: Fixed the issue related to YARN Resource Manager crashing when FileSystemStateStore fails to delete the files.
20419: YARN-3493 had to be backported to Hadoop 2.7.0 because of an exception in the Resource Manager.
20898: A YARN application timed out because some of the reducers became stuck in LOCALIZED state.
21006: When the number of vcores
in a queue was specified with a
decimal value, the FairScheduler read the value after the decimal point as the
vcores
value.
21338: YARN job client no longer fails when it tries to connect to a port that is
not within the range defined by the
yarn.app.mapreduce.am.job.client.port-range
property.
21146: An attempt to copy files including symbolic links failed even when the
distcp -i
option (ignore failures) was used.
22261: After a ResourceManager restart, the Scheduler tab in the RM UI now shows correct information about the job that is running (below the Cluster Metrics and User Metrics tabs).
22242: The Resource Manager no longer fails at startup with a null pointer exception.
Warden
19926: Warden no longer silently updates configure.sh
to set
isDB
to true
in cases where
configure.sh
was run with the -noDB
option.
MapReduce
14265: When the TaskTracker starts up, it sets the reserved physical memory based on the initial calculation from warden. However, when an adjustment was made to the slots (for example, based on CPU, memory, or MFS disks), the reserved physical memory was not adjusted downwards. In version 5.1, the TaskTracker physical memory limit is set after slots calculation.
20117: The JobTracker UI no longer fails with a NullPointerException whenever a job with invalid configuration is submitted to the JobTracker.
20498: Fixed the JobTracker performance problem caused by a cached configuration object. Now, the object is created statically and used for all task completion events.
20725: When an application such as Oozie spawns a job within a job, job failure longer occurs due to static order initialization of the Configuration object.
21107: Fixed the issue related to the JobClient failing with the error "java.io.IOException: Connection timed out" on the client node.
21226: Multiple users can access the JobTracker view in the MCS if you disable the
new mapred.jobtracker.ui.showcounters
JobTracker option to prevent the
CPU/memory counters from displaying.
Scheduling
20644, 20854: Fixed the issue that caused mirrors to fail with multiple schedules
stopped. Diagnostics showed, with the maprcli schedule list
command, a
0
value in the inuse
column even when there were
multiple associated volumes.
21713: Fair Scheduler queue names that contain spaces no longer cause the Resource Manager to shut down. In Version 5.1. leading and/or trailing spaces in queue names are trimmed internally and associated applications proceed. A queue name that contains spaces only (no other characters) causes applications to fail with an exception.
MCS
16095: The MapR UI Jetty web server has been updated and now provides an option to disable the SSLv3 protocol.
<URL> returns invalid code: 404
<DATE> <TIME>,088 ERROR com.mapr.baseutils.URLProbingUtility
Security
14197: The mapr superuser, designated administrative users, and users with cluster ACL permissions set to at least login can nowaccess the CLDB view in the MCS when security is enabled.
19823: MapR now supports selective auditing of certain filesystem and table operations.
20061: You do not have to list the cluster name as the first entry in the
mapr-clusters.conf
file to log into a secure cluster with Kerberos
enabled.
21326: Access of a protected file over NFS was denied but not audited in the file server. This operation is now audited.
21327: Fixed the issue that caused connections to hosts on secure clusters to fail.
java.lang.UnsatisfiedLinkError: com.mapr.security.JNISecurity.SetClusterOption
22012: After applying a 4.0.2 patch, only the mapr
user could run
the hadoop fs -ls /
command.
22332: Version 3.2.1 of the Apache commons-collections library contained a remote execution vulnerability. To fix this problem,
MapR Version 5.1 packages use Version 3.2.2 of the library
(commons-collections-3.2.2.jar
).