Ambari Management Pack Development

Table of Contents

• Overview
• Ambari Zookeeper Config Lifecycle
• Adding a new property
• How to identify errors in MPack changes
• Testing changes without cycling Vagrant build
• Configuration involving dependency services
• Kerberos
• Best practices
• Upgrading MPack Services

Overview

Typically, Ambari Management Pack development will be done in the Vagrant environments. These instructions are specific to Vagrant, but can be adapted for other environemnts (e.g. make sure to be on the correct nodes for server vs agent files)

There are two MPacks:

• Metron - contains artifacts for deploying the Metron service
• Elasticsearch - contains artifacts for installing Elasticsearch and Kibana services

There is an mpack.json file for each which describes what services the mpack will contain, versions, etc.

Alongside this are two directories, addon-services and common-services. Below the Metron MPack is described, but this also applies similarly to the Elasticsearch MPack.

The layout of /common-services/METRON/CURRENT is

• /configuration
  • This contains a set of *-env.xml files, relevent to particular components or the service as a whole. These are where properties are defined.
• /package
  • /files
    • Contains files that get used as provided, in particular Elasticsearch templates.
  • /scripts
    • A set of Python scripts that interface with Ambari to manage setup and install
    • /params
      • A set of Python scripts for managing parameters from the *-env.xml files
    • /templates
      • A set of Jinja template files which can be populated with properties
• /quicklinks
  • Contains quicklinks.json to define the Ambari quicklinks that should be used in the UI
• themes
  • Manages the Ambari UI themes blocks for organizing the configuration by relevant area.
• kerberos.json
  • Defines the keytabs and other Kerberos configuration to be used when Kerberizing a cluster
• metainfo.xml
  • Defines the METRON service, along with required packages, services, etc. If you need to have the MPack install a new package (e.g. RPM, DEB), add it here.
• service_advisor.py
  • Handles component layout and validation, along with handling some configurations for other services or that needs configs from other services.

The layout of /addon-services/METRON/CURRENT is

• /repos
  • Contains repoinfo.xml that defines repositories to install packages from
• metainfo.xml
  • Limited info version of /common-services/METRON/CURRENT/metainfo.xml
• role_command_order.json
  • Defines the order of service startup and other actions relative to each other.

Ambari Zookeeper Config Lifecycle

The source of truth for Metron configuration resides in Zookeeper as a series of JSON documents. There are actually multiple locations that you can populate the Zookeeper configs from:

• $METRON_HOME/config/zookeeper
• Stellar REPL
• Management UI
• Ambari

Any change that works with Metron configuration stored in Zookeeper should always pull the latest config from Zookeeper first before making any changes and publishing them back up to avoid potentially overwriting the latest configs with stale data. The lifecycle for configuration management with Ambari is as follows:

• First start/install for parsers, enrichment, indexing, or the profiler
  • Perform a PUSH from local disk to Zookeeper
• All component start/restart operations for parsers, enrichment, indexing, or the profiler
  • Perform a JSON patch (RFC 6902) on the configs in Zookeeper and push the patched, pretty formatted config up to Zookeeper. Patching is currently applicable to to global.json only.
  • Pull all configs to the local file system.

The patching mechanism we introduced to Ambari for managing the global config is by default additive in nature, though it will also overwrite any existing values with the latest values from Ambari. This means that you can add any properties to the global config outside of Ambari you like, provided you are not modifying values explicitly managed by Ambari. This is important because you will not get any errors when modifying the configuration outside of Ambari. Instead, if you modified es.ip and changed global.json outsided Ambari, you would not see this change in Ambari. Meanwhile, the indexing topology would be using the new value stored in Zookeeper. Remember, Zookeeper is the source of truth. If you were to restart a Metron topology component via Ambari, that es.ip property would now be set back to the value stored in Ambari. So in summary, if it’s managed by Ambari, only change that value via Ambari. If it’s not, then feel free to use whatever mechanism you like, just be sure to pull the latest config from Zookeeper before making any changes. We make no effort to handle race conditions. It’s last commit wins. Below are the global config properties managed by Ambari:

• es.clustername
• es.ip
• es.date.format
• parser.error.topic
• update.hbase.table
• update.hbase.cf
• profiler.client.period.duration
• profiler.client.period.duration.units

Adding a new property

1. Add the property to the appropriate *-env.xml file found in METRON.CURRENT/configuration.

   <property>
       <name>new_property</name>
       <description>New Property description</description>
       <value>Default Value</value>
       <display-name>New Property Pretty Name</display-name>
   </property>
   

The appropriate *-env.xml file should be selected based on which component depends on the property. This allows Ambari to accurately restart only the affected components when the property is changed. If a property is in metron-env.xml, Ambari will prompt you to restart all Metron components.

1. Add the property to the metron_theme.json file found in METRON.CURRENT/themes if the property was added to a component-specific *-env.xml file (metron-parsers-env.xml for example) and not metron-env.xml. This is necessary for the property to be displayed in the correct tab of the Metron Configs section in the Ambari UI. Using other properties as a guide, add the property to /configuration/placement/configs for proper placement and also to /configuration/widgets for a specific widget type.

2. Reference the property in METRON.CURRENT/package/scriptes/params/params_linux.py, unless it will be used in Ambari’s status command. It will be stored in a variable. The name doesn’t have to match, but it’s preferred that it does. Make sure to use replace metron-env the correct *-env file, as noted above.

   new_property = config['configurations']['metron-env']['new_property']
   

   If this property will be used in the status command, instead make this change in METRON.CURRENT/package/scriptes/params/status_params.py. Afterwards, in params_linux.py, reference the new property:

   new_property = status_params.new_property
   

   This behavior is because Ambari doesn’t send all parameters to the status, so it needs to be explicitly provided. Also note that status_params.py parameters are not automatically pulled into the params_linux.py namespace, so we explicitly choose the variables to include. See https://docs.python.org/2/howto/doanddont.html#at-module-level for more info.

3. Ambari master services can then import the params:

   from params import params
   env.set_params(params)
   

4. The *_commands.py files receive the params as an input from the master services. Once this is done, they can be accessed via the variable we set above:

   self.__params.new_property
   

<property>
    <name>metron_apps_hdfs_dir</name>
    <value>/apps/metron</value>
    <description>Metron apps HDFS dir</description>
    <display-name>Metron apps HDFS dir</display-name>
</property>

metron_apps_hdfs_dir = config['configurations']['metron-env']['metron_apps_hdfs_dir']

metron_apps_hdfs_dir = status_params.metron_apps_hdfs_dir

from resource_management.libraries.functions import format
# ...
hdfs_grok_patterns_dir = format("{metron_apps_hdfs_dir}/patterns")

metron_apps_indexed_hdfs_dir = format(format(config['configurations']['metron-indexing-env']['metron_apps_indexed_hdfs_dir']))

<property>
    <name>metron_apps_hdfs_dir</name>
    <value>/apps/metron</value>
    <description>Metron apps HDFS dir</description>
    <display-name>Metron apps HDFS dir</display-name>
</property>

<property>
    <name>metron_apps_indexed_hdfs_dir</name>
    <value>{{metron_apps_hdfs_dir}}/indexing/indexed</value>
    <description>Indexing bolts will write to this HDFS directory</description>
    <display-name>Metron apps indexed HDFS dir</display-name>
</property>

def __init__(self, params):
    if params is None:
        raise ValueError("params argument is required for initialization")
    self.__params = params
    # Other initialization

def stop(self, env, upgrade_type=None):
    from params import params
    env.set_params(params)
    commands = ParserCommands(params)
    commands.stop_parser_topologies()

def init_parsers(self):
    Logger.info(
        "Copying grok patterns from local directory '{0}' to HDFS '{1}'".format(self.__params.local_grok_patterns_dir,
                                                                                self.__params.hdfs_grok_patterns_dir))

    self.__params.HdfsResource(self.__params.hdfs_grok_patterns_dir,
                               type="directory",
                               action="create_on_execute",
                               owner=self.__params.metron_user,
                               mode=0755,
                               source=self.__params.local_grok_patterns_dir)

    Logger.info("Done initializing parser configuration")

<property>
    <name>metron_log_dir</name>
    <value>/var/log/metron</value>
    <description>Log directory for metron</description>
    <display-name>Metron log dir</display-name>
</property>

metron_log_dir = config['configurations']['metron-env']['metron_log_dir']

METRON_LOG_DIR="{{metron_log_dir}}"

from resource_management.core.resources.system import File
from resource_management.core.source import Template

def configure(self, env, upgrade_type=None, config_dir=None):
    # Do stuff

    env.set_params(params)
    File(format("/etc/sysconfig/metron"),
         content=Template("metron.j2")
         )
    # Do stuff

...
METRON_LOG_DIR="/var/log/metron"
...

{
  "configuration": {
    "layouts": [...],
    "placement": {
      "configs": [
        {
          "config": "metron-enrichment-env/enrichment_workers",
          "subsection-name": "subsection-enrichment-storm"
        },
        ...
      ]
    },
    "widgets": [
      {
        "config": "metron-enrichment-env/enrichment_workers",
        "widget": {
          "type": "text-field"
        }
      }
    ]
  }
}

How to identify errors in MPack changes

Typically, any errors are thrown at one of two times:

1. Attempting to install Metron as a whole. These are typically service level definition errors, not property errors. Logs will often be found in /var/log/ambari-server/ambari-server.log. Often the log will indicate it was unable to load a particular service or subservice with some more detail.
2. Running the actual functionality. These typically tend to cause errors in the UI at runtime. Logs are usually found in /var/log/ambari-agent/ambari-agent.log.

Unfortunately, because errors tend to occur at runtime, it’s often necessary to add things like logging statements, or even just throw errors to print out in the Ambari UI.

The primary solution to these is to look in the logs for exceptions, see what’s going wrong (Property doesn’t exist? Malformed file couldn’t be loaded?), and adjust appropriately.

Testing changes without cycling Vagrant build

There are techniques we can use to avoid spinning down and spinning back up Vagrant.

/var/lib/ambari-server/resources/mpacks/metron-ambari.mpack-0.4.0.0/common-services
/var/lib/ambari-server/resources/mpacks/elasticsearch-ambari.mpack-0.4.0.0/common-services
/var/lib/ambari-agent/cache/common-services

/var/lib/ambari-server/resources/mpacks/metron-ambari.mpack-0.4.0.0/common-services/METRON/0.4.0/package/scripts/enrichment_commands.py
/var/lib/ambari-agent/cache/common-services/METRON/0.4.0/package/scripts/enrichment_commands.py

[root@node1 ~]# find /var/lib/ -name enrichment_commands.py
/var/lib/ambari-server/resources/mpacks/metron-ambari.mpack-0.4.0.0/common-services/METRON/0.4.0/package/scripts/enrichment_commands.py
/var/lib/ambari-agent/cache/common-services/METRON/0.4.0/package/scripts/enrichment_commands.py

Configuration involving dependency services

Metron can define expectations on other services, e.g. Storm’s topology.classpath should be /etc/hbase/conf:/etc/hadoop/conf. This happens in METRON.CURRENT/service_advisor.py.

The value is defined in a map in getSTORMSiteDesiredValues(self, is_secured). This map is used by validateSTORMSiteConfigurations to ensure that the value used by the component is what we need it to be. Generally, only the map should need to be modified.

Properties from the other services can also be used and examined. We use this to build the appropriate URL for Storm, because it differs based on Kerberos security. For example, to retrieve from another service, access the configurations array, retrieve the appropriate config file, retrieve the properties, and finally the desired property.

stormUIServerPort = services["configurations"]["storm-site"]["properties"]["ui.port"]

The security of the cluster can be checked with

is_secured = self.isSecurityEnabled(services)

Note that the configuration expectations will be properly enforced if the cluster is upgraded (e.g. the Storm URL is properly updated) and it does not need to occur manually.

Kerberos

Any scripts interacting with underlying secured tech needs to have kinit run as needed. This includes anything interacting with HDFS, HBase, Kafka, Storm, etc. This kinit should be run in a conditional statement, to avoid breaking non-Kerberized clusters.

Ambari can run kinit, as a given user with the appropriate keytab like so:

if self.__params.security_enabled:
    metron_security.kinit(self.__params.kinit_path_local,
                          self.__params.metron_keytab_path,
                          self.__params.metron_principal_name,
                          execute_user=self.__params.metron_user)

The security_enabled param is already made available, along with appropriate keytabs for metron, hdfs, kafka, and hbase users.

Best practices

• Write scripts to be idempotent. The pattern currently used is to write a file out when a task is finished, e.g. setting up ACLs or tables. For example, when indexing is configured, a file is written out and checked based on a property.

  def set_configured(self):
      File(self.__params.indexing_configured_flag_file,
           content="",
           owner=self.__params.metron_user,
           mode=0755)
  

  This is checked in the indexing master

  if not commands.is_configured():
      commands.init_kafka_topics()
      commands.init_hdfs_dir()
      commands.set_configured()
  

• Ensure ACLs are properly managed. This includes Kafka and HBase. Often this involves a config file written out as above because this isn’t idempotent!

  • Make sure to kinit as the correct user for setting up ACLs in a secured cluster. This is usually kafka for Kafka and hbase for HBase.
  • See set_hbase_acls in METRON.CURRENT/package/scripts/enrichment_commands.py for an HBase example
  • See init_kafka_acls in METRON.CURRENT/package/scripts/enrichment_commands.py and METRON.CURRENT/package/scripts/metron_service.py for an Kafka example

Upgrading MPack Services

Apache Metron currently provides one service as part of its Metron MPack

• Metron

Apache Metron currently provides two services as part of its Elasticsearch MPack

• Elasticsearch
• Kibana

There is currently no mechanism provided for multi-version or backwards compatibility. If you upgrade a service, e.g. Elasticsearch 2.x to 5.x, that is the only version that will be supported by Ambari via MPack.

The main steps for upgrading a service are split into add-on and common services for each service within the MPack as follows:

• Update the common services
  • Change the service directory to use the new product version number
  • Update metainfo.xml
• Update the add-on services
  • Change the service directory to use the new product version number
  • Update repoinfo.xml
  • Update metainfo.xml
• Update mpack.json