Troubleshooting a CDH Upgrade Using Cloudera Manager

HDFS DataNodes fail to start

After upgrading, HDFS DataNodes fail to start with exception:

Exception in secureMainjava.lang.RuntimeException: Cannot start datanode because the configured max locked memory size (dfs.datanode.max.locked.memory) of 4294967296 bytes is more than the datanode's available RLIMIT_MEMLOCK ulimit of 65536 bytes.

Possible Reasons

HDFS caching, which is enabled by default in CDH 5 and higher, requires new memlock functionality from Cloudera Manager Agents.

Possible Solutions

Do the following:

  1. Stop all CDH and managed services.
  2. On all hosts with Cloudera Manager Agents, hard-restart the Agents. Before performing this step, ensure you understand the semantics of the hard_restart command by reading Hard Stopping and Restarting Agents.
    • RHEL-compatible 7 and higher:
      $ sudo service cloudera-scm-agent next_stop_hard
      $ sudo service cloudera-scm-agent restart
    • All other Linux distributions:
      sudo service cloudera-scm-agent hard_restart
  3. Start all services.

Upgrading to CDH 5.8.0 or CDH 5.8.1 When Using the Flume Kafka Client

Due to the change of offset storage from ZooKeeper to Kafka in the CDH 5.8 Flume Kafka client, data might not be consumed by the Flume agents, or might be duplicated (if during an upgrade to CDH 5.8.0 or CDH 5.8.1. To prevent this, perform the steps described below before you upgrade your system.

The upgrade process is based on this example configuration:
tier1.sources  = source1
tier1.channels = channel1
tier1.sinks = sink1

tier1.sources.source1.type = org.apache.flume.source.kafka.KafkaSource
tier1.sources.source1.zookeeperConnect = zkhost:2181
tier1.sources.source1.topic = flumetopic
tier1.sources.source1.groupId = flume
tier1.sources.source1.channels = channel1
tier1.sources.source1.interceptors = i1 i2
tier1.sources.source1.interceptors.i1.type = timestamp
tier1.sources.source1.interceptors.i2.type = host = 100

tier1.channels.channel1.type =
tier1.channels.channel1.groupId = flumechannel
tier1.channels.channel1.capacity = 10000
tier1.channels.channel1.transactionCapacity = 1000

tier1.sinks.sink1.type = hdfs
tier1.sinks.sink1.hdfs.path = /tmp/kafka/%{topic}/
tier1.sinks.sink1.hdfs.filePrefix = %{host}-
tier1.sinks.sink1.hdfs.rollInterval = 60
tier1.sinks.sink1.hdfs.rollSize = 0
tier1.sinks.sink1.hdfs.rollCount = 0
tier1.sinks.sink1.hdfs.fileType = DataStream = channel1
tier1.sinks.sink1.hdfs.kerberosKeytab = $KERBEROS_KEYTAB
tier1.sinks.sink1.hdfs.kerberosPrincipal = $KERBEROS_PRINCIPAL

Perform the following steps to upgrade CDH.

  1. If you are using a version lower than CDH 5.7, first upgrade to CDH 5.7. If for some reason you cannot upgrade to CDH 5.7, contact your Cloudera Sales Engineer for assistance, or file a support case with specific versions from which and to which you are upgrading.
  2. Add the following sections to the source and channel for the Flume configuration:
    # Added for source upgrade compatability
    tier1.sources.source1.kafka.bootstrap.servers = broker1:9092,broker2:9092 = kafka
    tier1.sources.source1.kafka.dual.commit.enabled = true = flume
    tier1.sources.source1.kafka.topics = flumetopic
    # Added for channel upgrade compatability
    tier1.channels.channel1.kafka.topic = flumechannel1
    tier1.channels.channel1.kafka.bootstrap.servers = broker1:9092,broker2:9092 = flumechannel = kafka
    tier1.channels.channel1.kafka.dual.commit.enabled = true
  3. Restart (or rolling restart) the Flume agents. This switches to Kafka, but keeps both the Kafka and ZooKeeper offsets updated because the dual.commit.enabled property is set to true. Confirm that Kafka messages are flowing through the Flume servers. Updating the offsets only occurs when new messages are consumed, so there must be at least one Kafka message consumed by the Flume agent, or one event passed through the Flume channel. Use the following commands to verify that Flume is properly updating the offsets in Kafka (the egrep command is used to match the correct topic names: in this example, flumetopic and flumechannel) :
    echo "exclude.internal.topics=false" > /tmp/consumer.config 
    kafka-console-consumer --consumer.config /tmp/consumer.config 
    --formatter "kafka.coordinator.GroupMetadataManager\$OffsetsMessageFormatter" 
    --zookeeper zkhost:2181 --topic __consumer_offsets  |egrep -e "flumetopic|flumechannel1"

    Output should be similar to the following and show that the Flume source and/or channel topics offsets are being incremented:

    [flume,flumetopic,0]::[OffsetMetadata[70,cf9e5630-214e-4689-9869-5e077c936ffb],CommitTime 1469827951129,ExpirationTime 1469914351129]
    [flumechannel,flumechannel1,0]::[OffsetMetadata[61,875e7a82-1c22-43be-acaa-eb4d63e7f71e],CommitTime 1469827951128,ExpirationTime 1469914351128]
    [flumechannel,flumechannel1,0]::[OffsetMetadata[62,66bda888-0a70-4a02-a286-7e2e7d14050d],CommitTime 1469827951131,ExpirationTime 1469914351131]
  4. Perform the upgrade to CDH 5.8.0 or CDH 5.8.1. The Flume agents are restarted during the process. Flume continues to consume the source topic where it left off, and the sinks continue draining from the Kafka channels where they left off. Post upgrade, remove the following deprecated properties from flume.conf because they are no longer used in CDH 5.8.0 or higher:
    tier1.sources.source1.zookeeperConnect = zkhost:2181
    tier1.sources.source1.topic = flumetopic
    tier1.sources.source1.groupId = flume
    tier1.channels.channel1.groupId = flumechannel