Uploaded image for project: 'Ambari'
  1. Ambari
  2. AMBARI-10931

Blueprints processor needs stronger validation of Blueprint JSON structure

    XMLWordPrintableJSON

Details

    • Bug
    • Status: Resolved
    • Major
    • Resolution: Fixed
    • 2.0.0
    • 2.1.0
    • ambari-server

    Description

      The Blueprints processor needs to have stronger validation during the process of handling a POST for a newly-submitted Blueprint from a REST client.
      The Blueprints structure requires that the "configurations" element be an array of Maps, with each Map representing a given configuration type ("core-site", "storm-site", etc).
      Typically, the configurations structure should look like:
      ""configurations": [
      {
      "mapred-env":

      { "properties": { "mapreduce_log_dir_prefix": "/grid/0/log/hadoop-mapreduce" }

      }
      },
      { "mapred-site":

      { "properties": { "install-test-mapred-site": "install-test-mapred-site-VALUE" }

      }
      },
      { "yarn-env":

      { "properties": { "yarn_log_dir_prefix": "/grid/0/log/hadoop" }

      }
      }
      ]"
      The snippet above is just an example to illustrate the expected format.
      Recent changes to the Blueprint's structure to handle "properties_attributes" (such as "final") have modified the structure of a Blueprint, so that configuration properties are now contained in a "properties" map, internal to each configuration type. Creating new Blueprints using this newer format (the older format is still acceptable to the processor) require that the configuration types are still enclosed in a map, even though an internal map is also present, which may be a source of confusion.
      An example of an incorrect format would be:
      ""configurations": [
      {
      "mapred-env":

      { "properties": { "mapreduce_log_dir_prefix": "/grid/0/log/hadoop-mapreduce" }

      },
      "mapred-site":

      { "properties": { "install-test-mapred-site": "install-test-mapred-site-VALUE" }

      },
      "yarn-env":

      { "properties": { "yarn_log_dir_prefix": "/grid/0/log/hadoop" }

      }
      ]
      "
      Please note that in the example above a single map is created, with all configuration elements being added to that entry's configuration type. This causes a corrupted Blueprint to be stored and used for deployments, since all configuration elements added in the original Blueprint will be registered under one of the config types used in the original Blueprint. This will cause many types of cluster startup failures, as the configuration elements will be added to incorrect locations.

      When the Blueprint processor receives a POST submission with a new Blueprint that includes this incorrect format for "configurations", the processor must reject this submission, and send back a relevant error code/message, so that the user can resolve this problem prior to attempting a cluster deployment.
      This issue will not affect the older format for Blueprints, and so any backwards-compatible Blueprints used should work fine. This issue will only occur when new Blueprints are submitted that use the newer syntax/format for defining configuration properties (with support for properties_attributes) as well.
      This JIRA is being created to track this issue, as it could potentially be a user experience issue going forward, and should probably be resolved in Ambari 2.1, if possible.

      Attachments

        1. AMBARI-10931_01.patch
          12 kB
          Emil Anca
        2. AMBARI-10931_02.patch
          17 kB
          Emil Anca

        Issue Links

          Activity

            People

              eanca Emil Anca
              eanca Emil Anca
              Votes:
              0 Vote for this issue
              Watchers:
              3 Start watching this issue

              Dates

                Created:
                Updated:
                Resolved: