The aim of this filter is to aggregate informations available among several events (typically log lines) belonging to a same task, and finally push aggregated information into final task event.
Ruby
Latest commit 203f369 Jan 8, 2017 @fbaligand fbaligand committed with fbaligand release v2.5.1

README.md

Logstash Filter Aggregate Documentation

Travis Build Status

The aim of this filter is to aggregate information available among several events (typically log lines) belonging to a same task, and finally push aggregated information into final task event.

You should be very careful to set logstash filter workers to 1 (-w 1 flag) for this filter to work correctly otherwise events may be processed out of sequence and unexpected results will occur.

Example #1

  • with these given logs :
     INFO - 12345 - TASK_START - start
     INFO - 12345 - SQL - sqlQuery1 - 12
     INFO - 12345 - SQL - sqlQuery2 - 34
     INFO - 12345 - TASK_END - end
  • you can aggregate "sql duration" for the whole task with this configuration :
     filter {
         grok {
             match => [ "message", "%{LOGLEVEL:loglevel} - %{NOTSPACE:taskid} - %{NOTSPACE:logger} - %{WORD:label}( - %{INT:duration:int})?" ]
         }

         if [logger] == "TASK_START" {
             aggregate {
                 task_id => "%{taskid}"
                 code => "map['sql_duration'] = 0"
                 map_action => "create"
             }
         }

         if [logger] == "SQL" {
             aggregate {
                 task_id => "%{taskid}"
                 code => "map['sql_duration'] += event.get('duration')"
                 map_action => "update"
             }
         }

         if [logger] == "TASK_END" {
             aggregate {
                 task_id => "%{taskid}"
                 code => "event.set('sql_duration', map['sql_duration'])"
                 map_action => "update"
                 end_of_task => true
                 timeout => 120
             }
         }
     }
  • the final event then looks like :
{
         "message" => "INFO - 12345 - TASK_END - end",
    "sql_duration" => 46
}

the field sql_duration is added and contains the sum of all sql queries durations.

Example #2 : no start event

  • If you have the same logs than example #1, but without a start log :
     INFO - 12345 - SQL - sqlQuery1 - 12
     INFO - 12345 - SQL - sqlQuery2 - 34
     INFO - 12345 - TASK_END - end
  • you can also aggregate "sql duration" with a slightly different configuration :
     filter {
         grok {
             match => [ "message", "%{LOGLEVEL:loglevel} - %{NOTSPACE:taskid} - %{NOTSPACE:logger} - %{WORD:label}( - %{INT:duration:int})?" ]
         }

         if [logger] == "SQL" {
             aggregate {
                 task_id => "%{taskid}"
                 code => "map['sql_duration'] ||= 0 ; map['sql_duration'] += event.get('duration')"
             }
         }

         if [logger] == "TASK_END" {
             aggregate {
                 task_id => "%{taskid}"
                 code => "event.set('sql_duration', map['sql_duration'])"
                 end_of_task => true
                 timeout => 120
             }
         }
     }
  • the final event is exactly the same than example #1
  • the key point is the "||=" ruby operator.
    it allows to initialize 'sql_duration' map entry to 0 only if this map entry is not already initialized

Example #3 : no end event

Third use case: You have no specific end event.

A typical case is aggregating or tracking user behaviour. We can track a user by its ID through the events, however once the user stops interacting, the events stop coming in. There is no specific event indicating the end of the user's interaction.

In this case, we can enable the option 'push_map_as_event_on_timeout' to enable pushing the aggregation map as a new event when a timeout occurs.
In addition, we can enable 'timeout_code' to execute code on the populated timeout event. We can also add 'timeout_task_id_field' so we can correlate the task_id, which in this case would be the user's ID.

  • Given these logs:
    INFO - 12345 - Clicked One
    INFO - 12345 - Clicked Two
    INFO - 12345 - Clicked Three
  • You can aggregate the amount of clicks the user did like this:
    filter {
        grok {
                 match => [ "message", "%{LOGLEVEL:loglevel} - %{NOTSPACE:user_id} - %{GREEDYDATA:msg_text}" ]
        }

        aggregate {
            task_id => "%{user_id}"
            code => "map['clicks'] ||= 0; map['clicks'] += 1;"
            push_map_as_event_on_timeout => true
            timeout_task_id_field => "user_id"
            timeout => 600 # 10 minutes timeout
            timeout_tags => ['_aggregatetimeout']
            timeout_code => "event.set('several_clicks', event.get('clicks') > 1)"
        }
    }
  • After ten minutes, this will yield an event like:
    {
        "user_id": "12345",
        "clicks": 3,
        "several_clicks": true,
        "tags": [
            "_aggregatetimeout"
        ]
    }

Example #4 : no end event and tasks come one after the other

Fourth use case : like example #3, you have no specific end event, but also, tasks come one after the other.
That is to say : tasks are not interlaced. All task1 events come, then all task2 events come, ...
In that case, you don't want to wait task timeout to flush aggregation map.

  • A typical case is aggregating results from jdbc input plugin.
  • Given that you have this SQL query : SELECT country_name, town_name FROM town ORDER BY country_name
  • Using jdbc input plugin, you get these 3 events from :
  { "country_name": "France", "town_name": "Paris" }
  { "country_name": "France", "town_name": "Marseille" }
  { "country_name": "USA", "town_name": "New-York" }
  • And you would like these 2 result events to push them into elasticsearch :
  { "country_name": "France", "town_name": [ "Paris", "Marseille" ] }
  { "country_name": "USA", "town_name": [ "New-York" ] }
  • You can do that using push_previous_map_as_event aggregate plugin option :
     filter {
         aggregate {
             task_id => "%{country_name}"
             code => "
                map['town_name'] ||= []
                event.to_hash.each do |key,value|
                    map[key] = value unless map.has_key?(key)
                    map[key] << value if map[key].is_a?(Array) and !value.is_a?(Array)
                end
             "
             push_previous_map_as_event => true
             timeout => 5
             timeout_tags => ['aggregated']
         }

         if "aggregated" not in [tags] {
            drop {}
         }
     }
  • The key point is that each time aggregate plugin detects a new country_name, it pushes previous aggregate map as a new logstash event (with 'aggregated' tag), and then creates a new empty map for the next country
  • When 5s timeout comes, the last aggregate map is pushed as a new event
  • Finally, initial events (which are not aggregated) are dropped because useless

How it works

  • the filter needs a "task_id" to correlate events (log lines) of a same task
  • at the task beggining, filter creates a map, attached to task_id
  • for each event, you can execute code using 'event' and 'map' (for instance, copy an event field to map)
  • in the final event, you can execute a last code (for instance, add map data to final event)
  • after the final event, the map attached to task is deleted
  • in one filter configuration, it is recommanded to define a timeout option to protect the filter against unterminated tasks. It tells the filter to delete expired maps
  • if no timeout is defined, by default, all maps older than 1800 seconds are automatically deleted
  • all timeout options have to be defined in only one aggregate filter per task_id pattern.
    Timeout options are : timeout, timeout_code, push_map_as_event_on_timeout, push_previous_map_as_event, timeout_task_id_field, timeout_tags
  • if code execution raises an exception, the error is logged and event is tagged '_aggregateexception'

Use Cases

  • extract some cool metrics from task logs and push them into task final log event (like in example #1 and #2)
  • extract error information in any task log line, and push it in final task event (to get a final event with all error information if any)
  • extract all back-end calls as a list, and push this list in final task event (to get a task profile)
  • extract all http headers logged in several lines to push this list in final task event (complete http request info)
  • for every back-end call, collect call details available on several lines, analyse it and finally tag final back-end call log line (error, timeout, business-warning, ...)
  • Finally, task id can be any correlation id matching your need : it can be a session id, a file path, ...

Aggregate Plugin Options

  • task_id :
    The expression defining task ID to correlate logs.
    This value must uniquely identify the task in the system.
    This option is required.
    Example value : "%{application}%{my_task_id}"

  • code:
    The code to execute to update map, using current event.
    Or on the contrary, the code to execute to update event, using current map.
    You will have a 'map' variable and an 'event' variable available (that is the event itself).
    This option is required.
    Example value : "map['sql_duration'] += event.get('duration')"

  • map_action:
    Tell the filter what to do with aggregate map (default : "create_or_update").
    create: create the map, and execute the code only if map wasn't created before
    update: doesn't create the map, and execute the code only if map was created before
    create_or_update: create the map if it wasn't created before, execute the code in all cases
    Default value: create_or_update

  • end_of_task:
    Tell the filter that task is ended, and therefore, to delete aggregate map after code execution.
    Default value: false

  • aggregate_maps_path:
    The path to file where aggregate maps are stored when logstash stops and are loaded from when logstash starts.
    If not defined, aggregate maps will not be stored at logstash stop and will be lost.
    Must be defined in only one aggregate filter (as aggregate maps are global).
    Example value : "/path/to/.aggregate_maps"

  • timeout:
    The amount of seconds after a task "end event" can be considered lost.
    When timeout occurs for a task, The task "map" is evicted.
    Timeout can be defined for each "task_id" pattern.
    If no timeout is defined, default timeout will be applied : 1800 seconds.

  • timeout_code
    The code to execute to complete timeout generated event, when 'push_map_as_event_on_timeout' or 'push_previous_map_as_event' is set to true.
    The code block will have access to the newly generated timeout event that is pre-populated with the aggregation map.
    If 'timeout_task_id_field' is set, the event is also populated with the task_id value
    Example value: "event.set('state', 'timeout')"

  • push_map_as_event_on_timeout
    When this option is enabled, each time a task timeout is detected, it pushes task aggregation map as a new logstash event.
    This enables to detect and process task timeouts in logstash, but also to manage tasks that have no explicit end event.
    Default value: false

  • push_previous_map_as_event:
    When this option is enabled, each time aggregate plugin detects a new task id, it pushes previous aggregate map as a new logstash event, and then creates a new empty map for the next task.
    WARNING: this option works fine only if tasks come one after the other. It means : all task1 events, then all task2 events, etc...
    Default value: false

  • timeout_task_id_field
    This option indicates the timeout generated event's field for the "task_id" value.
    The task id will then be set into the timeout event. This can help correlate which tasks have been timed out.
    This field has no default value and will not be set on the event if not configured.
    Example:
    If the task_id is "12345" and this field is set to "my_id", the generated timeout event will contain 'my_id' key with '12345' value.

  • timeout_tags
    Defines tags to add when a timeout event is generated and yield.
    Default value: []

Changelog

Read CHANGELOG.md.

Need Help?

Need help? Try #logstash on freenode IRC or the https://discuss.elastic.co/c/logstash discussion forum.

Want to contribute?

Read BUILD.md.