Enrichment

Introduction

The enrichment topology is a topology dedicated to taking the data from the parsing topologies that have been normalized into the Metron data format (e.g. a JSON Map structure with original_message and timestamp) and

  • Enriching messages with external data from data stores (e.g. hbase) by adding new fields based on existing fields in the messages.
  • Marking messages as threats based on data in external data stores
  • Marking threat alerts with a numeric triage level based on a set of Stellar rules.

Enrichment Architecture

Architecture

Unified Enrichment Topology

There is an experimental unified enrichment topology which is shipped. Currently the architecture, as described above, has a split/join in order to perform enrichments in parallel. This poses some issues in terms of ease of tuning and reasoning about performance.

In order to deal with these issues, there is an alternative enrichment topology which uses data parallelism as opposed to the split/join task parallelism. This architecture uses a worker pool to fully enrich any message within a worker. This results in

  • Fewer bolts in the topology
  • Each bolt fully operates on a message.
  • Fewer network hops

Unified Architecture

This architecture is fully backwards compatible; the only difference is how the enrichment will operate on each message (in one bolt where the split/join is done in a threadpool as opposed to split across multiple bolts).

Using It

In order to use this, you will need to

  • Edit $METRON_HOME/bin/start_enrichment_topology.sh and adjust it to use remote-unified.yaml instead of remote.yaml
  • Restart the enrichment topology.

Configuring It

There are two parameters which you might want to tune in this topology. Both of them are topology configuration adjustable in the flux file $METRON_HOME/config/flux/enrichment/remote-unified.yaml:

  • metron.threadpool.size : The size of the threadpool. This can take a number or a multiple of the number of cores (e.g. 5C to 5 times the number of cores). The default is 2C.
  • metron.threadpool.type : The type of threadpool. (note: descriptions taken from here).
    • FIXED is a fixed threadpool of size n. n threads will process tasks at the time, when the pool is saturated, new tasks will get added to a queue without a limit on size. Good for CPU intensive tasks. This is the default.
    • WORK_STEALING is a work stealing threadpool. This will create and shut down threads dynamically to accommodate the required parallelism level. It also tries to reduce the contention on the task queue, so can be really good in heavily loaded environments. Also good when your tasks create more tasks for the executor, like recursive tasks.

In order to configure the parallelism for the enrichment bolt and threat intel bolt, the configurations will be taken from the respective join bolt parallelism. When proper ambari support for this is added, we will add its own property.

Enrichment Configuration

The configuration for the enrichment topology, the topology primarily responsible for enrichment and threat intelligence enrichment, is defined by JSON documents stored in zookeeper.

There are two types of configurations at the moment, global and sensor specific.

Global Configuration

There are a few enrichments which have independent configurations, such as from the global config. You can also configure the enrichment topology’s writer batching settings.

Also, see the “Global Configuration” section for more discussion of the global config.

GeoIP

Metron supports enrichment of IP information using GeoLite2. The location of the file is managed in the global config.

geo.hdfs.file

The location on HDFS of the GeoLite2 database file to use for GeoIP lookups. This file will be localized on the storm supervisors running the topology and used from there. This is lazy, so if this property changes in a running topology, the file will be localized from HDFS upon first time the file is used via the geo enrichment.

Writer Batching

enrichment.writer.batchSize

The size of the batch that is written to Kafka at once. Defaults to 15 (size of 1 disables batching).

enrichment.writer.batchTimeout

The timeout after which a batch will be flushed even if batchSize has not been met. Optional. If unspecified, or set to 0, it defaults to a system-determined duration which is a fraction of the Storm parameter topology.message.timeout.secs. Ignored if batchSize is 1, since this disables batching.

Sensor Enrichment Configuration

The sensor specific configuration is intended to configure the individual enrichments and threat intelligence enrichments for a given sensor type (e.g. snort).

Just like the global config, the format is a JSON stored in zookeeper. The configuration is a complex JSON object with the following top level fields:

  • enrichment : A complex JSON object representing the configuration of the enrichments
  • threatIntel : A complex JSON object representing the configuration of the threat intelligence enrichments

The enrichment Configuration

Field Description Example
fieldToTypeMap In the case of a simple HBase enrichment (i.e. a key/value lookup), the mapping between fields and the enrichment types associated with those fields must be known. This enrichment type is used as part of the HBase key. Note: applies to hbaseEnrichment only. "fieldToTypeMap" : { "ip_src_addr" : [ "asset_enrichment" ] }
fieldMap The map of enrichment bolts names to configuration handlers which know how to split the message up. The simplest of which is just a list of fields. More complex examples would be the stellar enrichment which provides stellar statements. Each field listed in the array arg is sent to the enrichment referenced in the key. Cardinality of fields to enrichments is many-to-many. "fieldMap": {"hbaseEnrichment": ["ip_src_addr","ip_dst_addr"]}
config The general configuration for the enrichment "config": {"typeToColumnFamily": { "asset_enrichment" : "cf" } }

The config map is intended to house enrichment specific configuration. For instance, for the hbaseEnrichment, the mappings between the enrichment types to the column families is specified.

The fieldMapcontents are of interest because they contain the routing and configuration information for the enrichments.
When we say ‘routing’, we mean how the messages get split up and sent to the enrichment adapter bolts.
The simplest, by far, is just providing a simple list as in

    "fieldMap": {
      "geo": [
        "ip_src_addr",
        "ip_dst_addr"
      ],
      "host": [
        "ip_src_addr",
        "ip_dst_addr"
      ],
      "hbaseEnrichment": [
        "ip_src_addr",
        "ip_dst_addr"
      ]
      }

Based on this sample config, both ip_src_addr and ip_dst_addr will go to the geo, host, and hbaseEnrichment adapter bolts.

Stellar Enrichment Configuration

For the geo, host and hbaseEnrichment, this is sufficient. However, more complex enrichments may contain their own configuration. Currently, the stellar enrichment is more adaptable and thus requires a more nuanced configuration.

At its most basic, we want to take a message and apply a couple of enrichments, such as converting the hostname field to lowercase. We do this by specifying the transformation inside of the config for the stellar fieldMap. There are two syntaxes that are supported, specifying the transformations as a map with the key as the field and the value the stellar expression:

    "fieldMap": {
       ...
      "stellar" : {
        "config" : {
          "hostname" : "TO_LOWER(hostname)"
        }
      }
    }

Another approach is to make the transformations as a list with the same var := expr syntax as is used in the Stellar REPL, such as:

    "fieldMap": {
       ...
      "stellar" : {
        "config" : [
          "hostname := TO_LOWER(hostname)"
        ]
      }
    }

Sometimes arbitrary stellar enrichments may take enough time that you would prefer to split some of them into groups and execute the groups of stellar enrichments in parallel. Take, for instance, if you wanted to do an HBase enrichment and a profiler call which were independent of one another. This usecase is supported by splitting the enrichments up as groups.

Consider the following example:

    "fieldMap": {
       ...
      "stellar" : {
        "config" : {
          "malicious_domain_enrichment" : {
            "is_bad_domain" : "ENRICHMENT_EXISTS('malicious_domains', ip_dst_addr, 'enrichments', 'cf')"
          },
          "login_profile" : [
            "profile_window := PROFILE_WINDOW('from 6 months ago')", 
            "global_login_profile := PROFILE_GET('distinct_login_attempts', 'global', profile_window)",
            "stats := STATS_MERGE(global_login_profile)",
            "auth_attempts_median := STATS_PERCENTILE(stats, 0.5)", 
            "auth_attempts_sd := STATS_SD(stats)",
            "profile_window := null", 
            "global_login_profile := null", 
            "stats := null"
          ]
        }
      }
    }

Here we want to perform two enrichments that hit HBase and we would rather not run in sequence. These enrichments are entirely independent of one another (i.e. neither relies on the output of the other). In this case, we’ve created a group called malicious_domain_enrichment to inquire about whether the destination address exists in the HBase enrichment table in the malicious_domains enrichment type. This is a simple enrichment, so we can express the enrichment group as a map with the new field is_bad_domain being a key and the stellar expression associated with that operation being the associated value.

In contrast, the stellar enrichment group login_profile is interacting with the profiler, has multiple temporary expressions (i.e. profile_window, global_login_profile, and stats) that are useful only within the context of this group of stellar expressions. In this case, we would need to ensure that we use the list construct when specifying the group and remember to set the temporary variables to null so they are not passed along.

In general, things to note from this section are as follows:

  • The stellar enrichments for the stellar enrichment adapter are specified in the config for the stellar enrichment adapter in the fieldMap
  • Groups of independent (i.e. no expression in any group depend on the output of an expression from an other group) may be executed in parallel
  • If you have the need to use temporary variables, you may use the list construct. Ensure that you assign the variables to null before the end of the group.
  • Ensure that you do not assign a field to a stellar expression which returns an object which JSON cannot represent.
  • Fields assigned to Maps as part of stellar enrichments have the maps unfolded, similar to the HBase Enrichment
    • For example the stellar enrichment for field foo which assigns a map such as foo := { 'grok' : 1, 'bar' : 'baz'} would yield the following fields:
      • foo.grok == 1
      • foo.bar == 'baz'

The threatIntel Configuration

Field Description Example
fieldToTypeMap In the case of a simple HBase threat intel enrichment (i.e. a key/value lookup), the mapping between fields and the enrichment types associated with those fields must be known. This enrichment type is used as part of the HBase key. Note: applies to hbaseThreatIntel only. "fieldToTypeMap" : { "ip_src_addr" : [ "malicious_ips" ] }
fieldMap The map of threat intel enrichment bolts names to fields in the JSON messages. Each field is sent to the threat intel enrichment bolt referenced in the key. Each field listed in the array arg is sent to the enrichment referenced in the key. Cardinality of fields to enrichments is many-to-many. "fieldMap": {"hbaseThreatIntel": ["ip_src_addr","ip_dst_addr"]}
triageConfig The configuration of the threat triage scorer. In the situation where a threat is detected, a score is assigned to the message and embedded in the indexed message. "riskLevelRules" : { "IN_SUBNET(ip_dst_addr, '192.168.0.0/24')" : 10 }
config The general configuration for the Threat Intel "config": {"typeToColumnFamily": { "malicious_ips","cf" } }

The config map is intended to house threat intel specific configuration. For instance, for the hbaseThreatIntel threat intel adapter, the mappings between the enrichment types to the column families is specified. The fieldMap configuration is similar to the enrichment configuration in that the adapters available are the same.

The triageConfig field is also a complex field and it bears some description:

Field Description Example
riskLevelRules This is a list of rules (represented as Stellar expressions) associated with scores with optional names and comments see below
aggregator An aggregation function that takes all non-zero scores representing the matching queries from riskLevelRules and aggregates them into a single score. "MAX"

A risk level rule is of the following format:

  • name : The name of the threat triage rule
  • comment : A comment describing the rule
  • rule : The rule, represented as a Stellar statement
  • score : Associated threat triage score for the rule
  • reason : Reason the rule tripped. Can be represented as a Stellar statement

An example of a rule is as follows:

    "riskLevelRules" : [ 
        { 
          "name" : "is internal"
        , "comment" : "determines if the destination is internal."
        , "rule" : "IN_SUBNET(ip_dst_addr, '192.168.0.0/24')"
        , "score" : 10
        , "reason" : "FORMAT('%s is internal', ip_dst_addr)"
        }
                       ]

The supported aggregation functions are:

  • MAX : The max of all of the associated values for matching queries
  • MIN : The min of all of the associated values for matching queries
  • MEAN : The mean of all of the associated values for matching queries
  • SUM : The sum of all the associated values for matching queries
  • POSITIVE_MEAN : The mean of the positive associated values for the matching queries.

Example Configuration

An example configuration for the YAF sensor is as follows:

{
  "enrichment": {
    "fieldMap": {
      "geo": [
        "ip_src_addr",
        "ip_dst_addr"
      ],
      "host": [
        "ip_src_addr",
        "ip_dst_addr"
      ],
      "hbaseEnrichment": [
        "ip_src_addr",
        "ip_dst_addr"
      ]
    }
  ,"fieldToTypeMap": {
      "ip_src_addr": [
        "playful_classification"
      ],
      "ip_dst_addr": [
        "playful_classification"
      ]
    }
  },
  "threatIntel": {
    "fieldMap": {
      "hbaseThreatIntel": [
        "ip_src_addr",
        "ip_dst_addr"
      ]
    },
    "fieldToTypeMap": {
      "ip_src_addr": [
        "malicious_ip"
      ],
      "ip_dst_addr": [
        "malicious_ip"
      ]
    },
    "triageConfig" : {
      "riskLevelRules" : [ 
        {
          "rule" : "ip_src_addr == '10.0.2.3' or ip_dst_addr == '10.0.2.3'",
          "score" : 10
        }
      ],
      "aggregator" : "MAX"
    }
  }
}

ThreatIntel alert levels are emitted as a new field “threat.triage.level.” So for the example above, an incoming message that trips the ip_src_addr rule will have a new field threat.triage.level=10.

Example Enrichment via Stellar

Let’s walk through doing a simple enrichment using Stellar on your cluster using the Squid topology.

Install Prerequisites

Now let’s install some prerequisites:

  • Squid client via yum install squid
  • ES Head plugin via /usr/share/elasticsearch/bin/plugin install mobz/elasticsearch-head

Start Squid via service squid start

Adjust Enrichment Configurations for Squid to Call Stellar

Let’s adjust the configurations for the Squid topology to annotate the messages using some Stellar functions.

  • Edit the squid enrichment configuration at $METRON_HOME/config/zookeeper/enrichments/squid.json (this file will not exist, so create a new one) to add some new fields based on stellar queries:
{
  "enrichment" : {
    "fieldMap": {
      "stellar" : {
        "config" : {
          "numeric" : {
                      "foo": "1 + 1"
                      }
          ,"ALL_CAPS" : "TO_UPPER(source.type)"
        }
      }
     }
  },
  "threatIntel" : {
    "fieldMap":{
     "stellar" : {
        "config" : {
          "bar" : "TO_UPPER(source.type)"
        }
      } 
    },
    "triageConfig" : {
    }
  }
}

We have added the following fields as part of the enrichment phase of the enrichment topology:

  • foo == 2
  • ALL_CAPS == SQUID

We have added the following as part of the threat intel:

  • bar == SQUID

Please note that foo and ALL_CAPS will be applied in separate workers due to them being in separate groups.

  • Upload new configs via $METRON_HOME/bin/zk_load_configs.sh --mode PUSH -i $METRON_HOME/config/zookeeper -z node1:2181
  • Make the Squid topic in kafka via /usr/hdp/current/kafka-broker/bin/kafka-topics.sh --zookeeper node1:2181 --create --topic squid --partitions 1 --replication-factor 1

Start Topologies and Send Data

Now we need to start the topologies and send some data:

  • Start the squid topology via $METRON_HOME/bin/start_parser_topology.sh -k node1:6667 -z node1:2181 -s squid
  • Generate some data via the squid client:
    • squidclient http://yahoo.com
    • squidclient http://cnn.com
  • Send the data to kafka via cat /var/log/squid/access.log | /usr/hdp/current/kafka-broker/bin/kafka-console-producer.sh --broker-list node1:6667 --topic squid
  • Browse the data in elasticsearch via the ES Head plugin @ http://node1:9200/_plugin/head/ and verify that in the squid index you have two documents
  • Ensure that the documents have new fields foo, bar and ALL_CAPS with values as described above.

Note that we could have used any Stellar statements here, including calling out to HBase via ENRICHMENT_GET and ENRICHMENT_EXISTS or even calling a machine learning model via Model as a Service.