Cette page n'est pas encore disponible en français, sa traduction est en cours.
Si vous avez des questions ou des retours sur notre projet de traduction actuel, n'hésitez pas à nous contacter.

Overview

Because your log volume grows as your organization scales, the cost of ingesting and indexing in your downstream services (for example, log management solutions, SIEMs, and so forth) also rises. This guide walks you through using Observability Pipelines’ transforms to cut down on log volume and trim down the size of your logs to control your costs before data leaves your infrastructure or network.

Prerequisites

Use transforms to manage log volumes

In Observability Pipelines, a transform performs an action that modifies events, where events are logs flowing through the pipeline.

Deduplicate events

Use the dedupe transform to remove copies of data passing through your pipeline by adding the following component in your configuration.

transforms:
  my_transform_id:
    type: dedupe
    inputs:
      - my-source-or-transform-id
    cache: null
    fields: null
[transforms.my_transform_id]
type = "dedupe"
inputs = [ "my-source-or-transform-id" ]
{
  "transforms": {
    "my_transform_id": {
      "type": "dedupe",
      "inputs": [
        "my-source-or-transform-id"
      ],
      "cache": null,
      "fields": null
    }
  }
}

The Observability Pipelines Worker assigns every event a unique identifier to track deduplicated events. The cache option enables you to cache recent events to be used to check for duplicated data in the future, and defaults to 5,000 events. The fields option lists which fields are used to determine if an event is a duplicate.

Filter events

Use the filter transform when you want only certain logs that meet a specific criteria to pass through a component of your pipeline. For example, those conditions could be where the logs contain:

  • A specific tag, such as env.
  • A specific field value, such as the status field must be 400.

In those cases, insert a component that contains a filter transform to filter logs that uses the Datadog Processing Language (DPL) / Vector Remap Language (VRL) or Datadog Log Search syntax to set the conditions. Logs that don’t match the conditions are dropped.

The example below uses the filter transform and DPL/VRL to send only logs with a status of 500.

transforms:
  my_transform_id:
    type: filter
    inputs:
      - my-source-or-transform-id
    condition: 
      type: "vrl"
      source: ".status == 500"
[transforms.my_transform_id]
type = "filter"
inputs = [ "my-source-or-transform-id" ]

  [transforms.my_transform_id.condition]
  type = "vrl"
  source = ".status == 500"
{
  "transforms": {
    "my_transform_id": {
      "type": "filter",
      "inputs": [
        "my-source-or-transform-id"
      ],
      "condition": {
        "type": "vrl",
        "source": ".status == 500"
        }
    }
  }
}

Sample logs

When analyzing data that comes in large volumes or contains a lot of noise, such as CDN logs, sending all the logs to a destination is unnecessary. Instead, use the sample transform to send only the logs necessary to perform analysis that is statistically significant.

The exclude field excludes events from being sampled, and also supports DPL/VRL or Datadog Log Search syntax. The example below shows a configuration that samples every 10 events, which is set by the rate.

transforms:
  my_transform_id:
    type: sample
    inputs:
      - my-source-or-transform-id
    exclude: 
       type: "datadog_search"
       source: "*stack"
    rate: 10
[transforms.my_transform_id]
type = "sample"
inputs = [ "my-source-or-transform-id" ]
rate = 10

  [transforms.my_transform_id.exclude]
  type = "datadog_search"
  source = "*stack"
{
  "transforms": {
    "my_transform_id": {
      "type": "sample",
      "inputs": [
        "my-source-or-transform-id"
      ],
      "exclude": {
        "type": "datadog_search",
        "source": "*stack"
      },
      "rate": 10
    }
  }
}

Convert logs to metrics

In scenarios where you want to understand behavior over time, metrics around an event’s data point(s) would be more useful than a series of logs. As logs flow through your pipeline, use the log to metric transform to cut down on log volume by generating metrics based on specific tags.

You can generate four types of metrics:

  • Counter, which is useful to count the number of instances of logs with a specific tag. A count can be incremented, or reset to zero.
  • Distribution, which represents the distribution of sampled values. This is useful for generating summaries and histograms.
  • Gauge, which represents a single numerical value that can arbitrarily go up or down. This is useful to track values that fluctuate frequently.
  • Set, which consolidates unique values in an array. This can be useful to collect unique IP addresses, for example.

The example below illustrates a configuration for generating a counter metric , where metrics defines the key/value pairs to be added to the event.

transforms:
  my_transform_id:
    type: log_to_metric
    inputs:
      - my-source-or-transform-id
    metrics:
      - type: counter
        field: status
        name: response_total
        namespace: service
        tags:
          status: "{{status}}"
          host: "{{host}}"
[transforms.my_transform_id]
type = "log_to_metric"
inputs = [ "my-source-or-transform-id" ]

  [[transforms.my_transform_id.metrics]]
  type = "counter"
  field = "status"
  name = "response_total"
  namespace = "service"

    [transforms.my_transform_id.metrics.tags]
    status = "{{status}}"
    host = "{{host}}"
{
  "transforms": {
    "my_transform_id": {
      "type": "log_to_metric",
      "inputs": [
        "my-source-or-transform-id"
      ],
      "metrics": [
        {
          "type": "counter",
          "field": "status",
          "name": "response_total",
          "namespace": "service",
          "tags": {
            "status": "{{status}}",
            "host": "{{host}}"
          }
        }
      ]
    }
  }
}

If the following log is passed through the configuration above:

{
  "log": {
    "host": "10.22.11.222",
    "message": "Sent 200 in 54.2ms",
    "status": 200
  }
}

The following metric is generated:

{"metric":{"counter":{"value":1},"kind":"incremental","name":"response_total","namespace":"service","tags":{"host":"10.22.11.222","status":"200"}}}]

Collapse multiple events into a single log

In some cases, multiple logs can be consolidated into a single log. Thus, another way to cut down on log volume is to merge multiple logs into a single log. Use the reduce transform to reduce multiple logs into one.

The example below uses a reduce transform configuration to merge multiple Ruby log exceptions events.

transforms:
  my_transform_id:
    type: reduce
    inputs:
      - my-source-or-transform-id
    group_by:
      - host
      - pid
      - tid
    merge_strategies:
      message: concat_newline
    starts_when: match(string!(.message), r'^[^\\s]')
[transforms.my_transform_id]
type = "reduce"
inputs = [ "my-source-or-transform-id" ]
group_by = [ "host", "pid", "tid" ]
starts_when = "match(string!(.message), r'^[^\\s]')"

[transforms.my_transform_id.merge_strategies]
  message = "concat_newline"
{
  "transforms": {
    "my_transform_id": {
      "type": "reduce",
      "inputs": [
        "my-source-or-transform-id"
      ],
      "group_by": [
        "host",
        "pid",
        "tid"
      ],
      "merge_strategies": {
        "message": "concat_newline"
      },
      "starts_when": "match(string!(.message), r'^[^\\s]')"
    }
  }
}

In the reduce transform, group_by is an ordered list of fields used to group events. In this example, the events are grouped by host, pid, and tid fields.

merge_strategies is a map of field names to custom merge strategies. There are different merge strategies, including array, where each value is appended to an array, and sum, which adds all numeric values. In this example, concat_newline is used, where each string value is concatenated, then delimited by a newline.

starts_when is a condition used to distinguish the first event of a transaction. If this condition resolves to true for an event, the previous transaction is flushed without this event, and a new transaction is started. In this example, events with .message that do not match the ^[^\\s] regular expression condition are reduced into a single event.

If the following Ruby exception logs are passed through the configuration above:

[{"log":{
    "host":"host-1.hostname.com",
    "message":"foobar.rb:6:in `/': divided by 0(ZeroDivisionError)",
    "pid":1234,
    "tid":5678,
    "timestamp":"2020-10-07T12:33:21.223543Z"}
},
{
"log":{
    "host":"host-1.hostname.com",
    "message":"from foobar.rb:6:in `bar'",
    "pid":1234,
    "tid":5678,
    "timestamp":"2020-10-07T12:33:21.223543Z"}
},
{
"log":{
    "host":"host-1.hostname.com",
    "message":"from foobar.rb:2:in `foo'",
    "pid":1234,
    "tid":5678,
    "timestamp":"2020-10-07T12:33:21.223543Z"}
},
{
"log":{
    "host":"host-1.hostname.com",
    "message":"from foobar.rb:9:in `\u003cmain\u003e'",
    "pid":1234,"tid":5678,
    "timestamp":"2020-10-07T12:33:21.223543Z"}
},
{
"log":{
    "host":"host-1.hostname.com",
    "message":"Hello world, I am a new log",
    "pid":1234,
    "tid":5678,
    "timestamp":"2020-10-07T12:33:22.123528Z"
}}]

The following logs are generated:

[{
"log": {
    "host":"host-1.hostname.com",
    "message":"foobar.rb:6:in `/': divided by 0 (ZeroDivisionError)\n
               from foobar.rb:6:in `bar'\n
               from foobar.rb:2:in `foo'\n
               from foobar.rb:9:in `\u003cmain\u003e'",
    "pid":1234,
    "tid":5678,
    "timestamp":"2020-10-07T12:33:21.223543Z"}
},
{
"log":{
    "host":"host-1.hostname.com",
    "message":"Hello world, I am a new log",
    "pid":1234,
    "tid":5678,
    "timestamp":"2020-10-07T12:33:22.123528Z"
}}]

Use transforms to manage log size

Removing unnecessary fields to trim down your logs

Logs can contain fields that are unnecessary. When processing terabytes of data a day, dropping fields that are superfluous can significantly reduce the total number of logs your destination ingests and indexes.

To remove unnecessary fields, use the DPL/VRL to remap your log data. The following example removes unnecessary tags using del.

transforms:
  my_transform_id:
    type: remap
    inputs:
      - my-source-or-transform-id
    source: |-
      del(.unecessary_env_field)
      del(.unecessary_service_field)
      del(.unecessary_tag_field)      
[transforms.my_transform_id]
type = "remap"
inputs = [ "my-source-or-transform-id" ]
source = """
del(.unecessary_env_field)
del(.unecessary_service_field)
del(.unecessary_tag_field)"""
{
  "transforms": {
    "my_transform_id": {
      "type": "remap",
      "inputs": [
        "my-source-or-transform-id"
      ],
      "source": "del(.unecessary_env_field)\ndel(.unecessary_service_field)\ndel(.unecessary_tag_field)"
    }
  }
}

Further Reading