Apache Ambari Upgrade
Also available as:
PDF

Perform Rolling Upgrade

  1. Ambari checks that your cluster meets prerequisites. A dialog displays the results:

    1. If any required prerequisites are not met, the result displays an error.

      You cannot proceed with the upgrade until you make the appropriate corrections and return to Perform Upgrade again.

    2. If any optional prerequisites are not met, the result displays a warning.

      You may proceed with the upgrade.

    3. Ambari displays a list of configuration changes that occur during the upgrade.

  2. When the prerequisite checks complete, the upgrade starts. The time required to perform the upgrade depends on many factors. As part of the upgrade process, each component in the cluster restarts in a serial fashion. The stop/start times contribute to the total upgrade time.

  3. The upgrade process includes the following stages. Some stages require that you complete an action during normal operation.

    If any stage fails, the upgrade stops and prompts you for action.

    Stage

    Description

    Action Required

    Prepare Upgrade

    You should stop all YARN queues, all long-running applications on Slider, and deactivate & kill all running Storm topologies.

    Perform the actions to prepare for the upgrade.

    Prepare BackupsThis step prompts you to confirm that you have taken proper backups before proceeding.You must acknowledge the prompt for database backups.

    ZooKeeper

    All ZooKeeper servers are upgraded and restarted.

    None

    Ranger

    Ranger Admin and UserSync servers are upgraded and restarted.

    None. If Ranger Admin does not function after the upgrade completes, run the upgrade scripts manually, then Retry Upgrading Ranger.

    Core Masters

    This stage upgrades the master components of core services. This includes JournalNodes & NameNodes (HDFS), HistoryServer (MapReduce2), ResourceManager & ATS (YARN) and HBase Masters (HBase).

    None

    All Service Checks

    All Service Checks are performed against the cluster.

    Any service check that fail prompt you to Ignore and Continue, Downgrade or Retry. If you selected the Skip all Service Check failures option, you will only be prompted when all Service Checks complete.

    Core Slaves

    This stage upgrades the slave components of core services. This includes DataNodes (HDFS), RegionServers (HBase) and NodeManagers (YARN). This is done in two batches: 20% of the slaves first, then the remaining slaves.

    After the first 20% batch completes, you are prompted to verify the cluster is operating correctly.

    All Service Checks

    All Service Checks are performed against the cluster.

    Any service check that fail prompt you to Ignore and Continue, Downgrade or Retry. If you selected the Skip all Service Check failures option, you will only be prompted when all Service Checks complete.

    Spark

    The Spark Job History Server and clients are upgraded.

    None

    Oozie

    The Oozie Server and clients are upgraded.

    None

    Falcon

    The Falcon Server and clients are upgraded.

    None

    Client Components

    All remaining clients are upgraded.

    None

    All Service Checks

    All Service Checks are performed against the cluster.

    Any service check that fails prompts you to Ignore and Continue, Downgrade or Retry. If you selected the Skip all Service Check failures option, you will only be prompted when all Service Checks complete.

    Kafka

    The Kafka Brokers are upgraded.

    None

    Knox

    The Knox Gateways are upgraded.

    None

    Storm

    Storm does not support rolling upgrade from a previous HDP version to HDP-2.5.

    The rolling upgrade process prompts you to stop Storm topologies, perform the upgrade, and redeploy your topologies.

    Slider

    The Slider clients are upgraded.

    None

    Flume

    The Flume agents are upgraded.

    None

    Finalize Upgrade Pre-Check

    Checks if any hosts were not upgraded, either because the host was in Maintenance Mode, or one or more components on the host failed to upgrade (and were skipped).

    Click the list that displays # hosts for details on the hosts (and their components) that are not upgraded. You can Pause Upgrade, delete the hosts and return to finalize.

    Finalize

    The component upgrades are complete. You are prompted to Finalize. Finalizing completes the upgrade process and saves the cluster state.

    Prompted to Finalize, Finalize Later or Downgrade.

  4. When the rolling upgrade stages complete, may choose to Finalize the upgrade, to Finalize Later or to Downgrade. Finalizing later gives you a chance to perform more validation on the cluster. Downgrade moves the cluster version back to the previous version (basically: reverses the upgrade process stages). Once finalized, you cannot downgrade back to the previous version.

    [Note]Note

    If you choose to finalize later, both versions will be listed on the Stack and Versions tab with the starting version displaying as Current. It is not until you finalize that Ambari makes the target version the current version. Also, until you finalize, you will not be able to perform operational changes to the cluster (such as move components, change configurations, etc).

  5. Click Finalize to complete the rolling upgrade process.

More Information

Rolling Upgrade Prerequisites

Retry Ranger Upgrade