Skip to content
This repository has been archived by the owner on May 24, 2018. It is now read-only.

Latest commit

 

History

History
 
 

metron-elasticsearch

Folders and files

NameName
Last commit message
Last commit date

parent directory

..
src
src
 
 
README.md
README.md
 
 
pom.xml
pom.xml
 
 

README.md

Elasticsearch in Metron

Table of Contents

Introduction

Elasticsearch can be used as the real-time portion of the datastore resulting from metron-indexing.

Properties

es.clustername

The name of the elasticsearch Cluster. See here

es.ip

Specifies the nodes in the elasticsearch cluster to use for writing. The format is one of the following:

  • A hostname or IP address with a port (e.g. hostname1:1234), in which case es.port is ignored.
  • A hostname or IP address without a port (e.g. hostname1), in which case es.port is used.
  • A string containing a CSV of hostnames without ports (e.g. hostname1,hostname2,hostname3) without spaces between. es.port is assumed to be the port for each host.
  • A string containing a CSV of hostnames with ports (e.g. hostname1:1234,hostname2:1234,hostname3:1234) without spaces between. es.port is ignored.
  • A list of hostnames with ports (e.g. [ "hostname1:1234", "hostname2:1234"]). Note, es.port is NOT used in this construction.

es.port

The port for the elasticsearch hosts. This will be used in accordance with the discussion of es.ip.

es.date.format

The date format to use when constructing the indices. For every message, the date format will be applied to the current time and that will become the last part of the index name where the message is written to.

For instance, an es.date.format of yyyy.MM.dd.HH would have the consequence that the indices would roll hourly, whereas an es.date.format of yyyy.MM.dd would have the consequence that the indices would roll daily.

Upgrading to 5.6.2

Users should be prepared to re-index when migrating from Elasticsearch 2.3.3 to 5.6.2. There are a number of template changes, most notably around string type handling, that may cause issues when upgrading.

https://www.elastic.co/guide/en/elasticsearch/reference/5.6/setup-upgrade.html

Be aware that if you add a new string value and want to be able to filter and search on this value from the Alerts UI, you must add a mapping for that type to the appropriate Elasticsearch template. Below is more detail on how to choose the appropriate mapping type for your string value.

Type Mappings

Type mappings have changed quite a bit from ES 2.x -> 5.x. Here is a brief rundown of the biggest changes. More detailed references from Elasticsearch are provided in the Type Mapping References section below.

  • string fields replaced by text/keyword type

  • strings have new default mappings as follows

    {
  "type": "text",
  "fields": {
    "keyword": {
      "type": "keyword",
      "ignore_above": 256
    }
  }
}

  • There is no longer a _timestamp field that you can set "enabled" on. This field now causes an exception on templates. Replace with an application-created timestamp of "date" type.

The semantics for string types have changed. In 2.x, you have the concept of index settings as either "analyzed" or "not_analyzed" which basically means "full text" and "keyword", respectively. Analyzed text basically means the indexer will split the text using a text analyzer thus allowing you to search on substrings within the original text. "New York" is split and indexed as two buckets, "New" and "York", so you can search or query for aggregate counts for those terms independently and will match against the individual terms "New" or "York." "Keyword" means that the original text will not be split/analyzed during indexing and instead treated as a whole unit, i.e. "New" or "York" will not match in searches against the document containing "New York", but searching on "New York" as the full city name will. In 5.x language instead of using the "index" setting, you now set the "type" to either "text" for full text, or "keyword" for keywords.

Below is a table depicting the changes to how String types are now handled.

sort, aggregate, or access values ES 2.x ES 5.x Example
no 
"my_property" : {
  "type": "string",
  "index": "analyzed"
}

"my_property" : {
  "type": "text"
}
Additional defaults: "index": "true", "fielddata": "false" 		"New York" handled via in-mem search as "New" and "York" buckets. No aggregation or sort.
yes 
"my_property": {
  "type": "string",
  "index": "analyzed"
}

"my_property": {
  "type": "text",
  "fielddata": "true"
}
"New York" handled via in-mem search as "New" and "York" buckets. Can aggregate and sort.
yes 
"my_property": {
  "type": "string",
  "index": "not_analyzed"
}

"my_property" : {
  "type": "keyword"
}
"New York" searchable as single value. Can aggregate and sort. A search for "New" or "York" will not match against the whole value.
yes 
"my_property": {
  "type": "string",
  "index": "analyzed"
}

"my_property": {
  "type": "text",
  "fields": {
    "keyword": {
      "type": "keyword",
      "ignore_above": 256
    }
  }
}
"New York" searchable as single value or as text document, can aggregate and sort on the sub term "keyword."

If you want to set default string behavior for all strings for a given index and type, you can do so with a mapping similar to the following (replace ${your_type_here} accordingly):

# curl -XPUT 'http://${ES_HOST}:${ES_PORT}/_template/default_string_template' -d '
{
  "template": "*",
  "mappings" : {
    "${your_type_here}": {
      "dynamic_templates": [
        {
          "strings": {
            "match_mapping_type": "string",
            "mapping": {
              "type": "text"
            }
          }
        }
      ]
    }
  }
}
'

By specifying the "template" property with value "*" the template will apply to all indexes that have documents indexed of the specified type (${your_type_here}). This results in the following template.

# curl -XGET 'http://${ES_HOST}:${ES_PORT}/_template/default_string_template?pretty'
{
  "default_string_template" : {
    "order" : 0,
    "template" : "*",
    "settings" : { },
    "mappings" : {
      "${your_type_here}" : {
        "dynamic_templates" : [
          {
            "strings" : {
              "match_mapping_type" : "string",
              "mapping" : {
                "type" : "text"
              }
            }
          }
        ]
      }
    },
    "aliases" : { }
  }
}

Notes on other settings for types in ES

  • doc_values
    • on-disk data structure
    • provides access for sorting, aggregation, and field values
    • stores same values as _source, but in column-oriented fashion better for sorting and aggregating
    • not supported on text fields
    • enabled by default
  • fielddata
    • in-memory data structure
    • provides access for sorting, aggregation, and field values
    • primarily for text fields
    • disabled by default because the heap space required can be large
Type Mapping References

Using Metron with Elasticsearch 5.6.2

There is a requirement that all sensors templates have a nested alert field defined. This field is a dummy field. See Ignoring Unmapped Fields for more information

Without this field, an error will be thrown during ALL searches (including from UIs, resulting in no alerts being found for any sensor). This error will be found in the REST service's logs.

Exception seen:

QueryParsingException[[nested] failed to find nested object under path [alert]];

There are two steps to resolve this issue. First is to update the Elasticsearch template for each sensor, so any new indices have the field. This requires retrieving the template, removing an extraneous JSON field so we can put it back later, and adding our new field.

Make sure to set the ELASTICSEARCH variable appropriately. $SENSOR can contain wildcards, so if rollover has occurred, it's not necessary to do each index individually. The example here appends index* to get all indexes for the provided sensor.

export ELASTICSEARCH="node1"
export SENSOR="bro"
curl -XGET "http://${ELASTICSEARCH}:9200/_template/${SENSOR}_index*?pretty=true" -o "${SENSOR}.template"
sed -i '' '2d;$d' ./${SENSOR}.template
sed -i '' '/"properties" : {/ a\
"alert": { "type": "nested"},' ${SENSOR}.template

To manually verify this, you can optionally pretty print it again with:

python -m json.tool bro.template

We'll want to put the template back into Elasticsearch:

curl -XPUT "http://${ELASTICSEARCH}:9200/_template/${SENSOR}_index" -d @${SENSOR}.template

To update existing indexes, update Elasticsearch mappings with the new field for each sensor.

curl -XPUT "http://${ELASTICSEARCH}:9200/${SENSOR}_index*/_mapping/${SENSOR}_doc" -d '
{
  "properties" : {
    "alert" : {
      "type" : "nested"
    }
  }
}
'
rm ${SENSOR}.template

Installing Elasticsearch Templates

The stock set of Elasticsearch templates for bro, snort, yaf, error index and meta index are installed automatically during the first time install and startup of Metron Indexing service.

It is possible that Elasticsearch service is not available when the Metron Indexing Service startup, in that case the Elasticsearch template will not be installed.

For such a scenario, an Admin can have the template installed in two ways:

Method 1 - Manually from the Ambari UI by following the flow: Ambari UI -> Services -> Metron -> Service Actions -> Elasticsearch Template Install

Method 2 - Stop the Metron Indexing service, and start it again from Ambari UI. Note that the Metron Indexing service tracks if it has successfully installed the Elasticsearch templates, and will attempt to do so each time it is Started until successful.

Note: If you have made any customization to your index templates, then installing Elasticsearch templates afresh will lead to overwriting your existing changes. Please exercise caution.