Changed Features and Behaviors in Cloudera Manager 4
The following sections describe what’s changed in each Cloudera Manager release.
What's Changed in Cloudera Manager 4.7.0
- The default values of a number of YARN properties have changed to more sensible values.
Customers with the old default values will have their configuration changed to the new
default values automatically upon upgrading Cloudera Manager to version 4.7 or later.
- mapreduce.map.memory.mb: 512MiB -> 1024MiB
- mapreduce.reduce.memory.mb: 512MiB -> 1024MiB
- mapreduce.map.java.opts.max.heap: 394MiB -> 787MiB
- mapreduce.reduce.java.opts.max.heap: 394MiB -> 787MiB
- mapreduce.reduce.shuffle.parallelcopies: 5 -> 10
- mapreduce.task.io.sort.factor: 10 -> 64
- mapreduce.task.io.sort.mb: 100 -> 256
- yarn.log-aggregation-enable: false -> true
- yarn.app.mapreduce.am.resource.mb: 1536 -> 1024
Note that it is possible that if more than 1024 was actually needed for yarn.app.mapreduce.am.resource.mb, then changing it from 1536 to 1024 could cause a problem. Even less likely, increasing some of the other parameters could create memory / CPU problems and degrade performance in some scenarios. We expect that these new defaults are better in most cases.
- The default value of dfs.namenode.checkpoint.txns has increased from 40,000 to 1 million. This improves the efficiency of HDFS checkpointing.
What's Changed in Cloudera Manager 4.6.0
- The All Services and individual services pages combine status and health information into a single status value.
- The Administration pages are now accessed from the new Administration menu on the main navigation bar. The gear icon no longer appears. In addition, the tabs on the Administration page have been removed; you can access the individual pages from the Administration menu.
- A new Diagnose menu provides access to the commands for searching for logs and events.
- The cluster-level Rolling Restart (available with the Enterprise version of Cloudera Manager) has been modified so that it performs restarts host-by-host, rather than by roles across all DataNodes. This is to avoid the possibility that all TaskTrackers or RegionServers might be blacklisted during the restart.
- The Cloudera Manager documentation set has been re-organized to
make it easier to find content. In particular, the Cloudera Manager User Guide has been
split into several separate books:
- Managing Clusters with Cloudera Manager, which covers configuring services and roles, adding security, configuring HA, and upgrading CDH among other topics.
- Cloudera Manager Monitoring and Diagnostics Guide, covering service and activity monitoring, events and alerts, reports, logging, and charting.
- Cloudera Manager Administration Guide, covering working with the server and agents, user authentication, and upgrading Cloudera Manager, among other topics.
- To avoid a conflict with the HBase Rest Web UI port in CDH4.3, the Cloudera Manager Host Monitor (HMON) default port has been changed from 8085 to 8090.
- In the tsquery language, the like operator has been renamed to rlike to reflect that it supports regular expression matching.
- The Remove From Cluster command has been enhanced so you can now remove a host from a cluster in a single operation, leaving it available to be added to another cluster managed by the same Cloudera Manager server.