Log Architecture & Structural Overview

The AWS CloudWatch Embedded Metric Format (EMF) Log Parser is an essential telemetry stream within the AWS Cloud Infrastructure ecosystem. Parse AWS CloudWatch Embedded Metric Format (EMF) structured JSON payloads to extract metrics and dimensions.

This schema defines a structure of 9 extracted attributes, including 2 numeric metrics and 7 string dimensions. In production observability architectures, these tokens provide high-cardinality indexing keys for telemetry pipelines before shipping to storage backends such as ClickHouse, Elasticsearch, Amazon S3, or Datadog.

Raw Telemetry Ingestion Profile

A typical raw event line for aws-cloudwatch-json averages 227 bytes across 9 tokens. Modern collectors such as Fluent Bit and Vector require zero-backtracking regular expressions to avoid CPU spikes during traffic surges.

Extracted Field Schema & Data Types

The transpiled Grok pattern extracts the following schema fields from each raw event line. Data collectors cast these values according to the typed mappings below.

Field Name	Inferred Type	Description & Collector Semantics
timestamp	integer	Unix epoch millisecond timestamp.
namespace	string	CloudWatch custom metric namespace.
dimension	string	Primary dimension name.
metric_name	string	Published CloudWatch metric name.
unit	string	Unit of measurement (Milliseconds, Count, Bytes).
dim_key	string	Dimension attribute key name.
dim_val	string	Dimension attribute value.
metric_key	string	Target metric key name.
metric_value	float	Numeric metric measurement value.

Common Regex Traps & Production Edge Cases

Engineers frequently encounter ingestion failures or pipeline drops due to subtle variations in real-world event logs. Watch out for these verified pitfalls:

1EMF logs must be valid JSON; unescaped quotation marks in nested payloads will break regex extraction.

2CloudWatch natively extracts EMF, but external collectors (Fluent Bit) require JSON parsers to prevent redundant metric creation.

3Timestamp is recorded in milliseconds, unlike standard Unix epoch seconds.

Production Collector Setup & Configurations

Pre-configured parser definitions ready to be dropped into your infrastructure repository.

Fluent Bit (parsers.conf)

Format: regex

# ==============================================================================
# Fluent Bit Parser Configuration (parsers.conf)
# ==============================================================================
[PARSER]
    Name        logpipeline_parser
    Format      regex
    Regex       ^\{"_aws":\{"Timestamp":(?<timestamp>(?:[+-]?(?:[0-9]+))),"CloudWatchMetrics":\[\{"Namespace":"(?<namespace>\b\w+\b)","Dimensions":\[\["(?<dimension>\b\w+\b)"\]\],"Metrics":\[\{"Name":"(?<metric_name>\b\w+\b)","Unit":"(?<unit>\b\w+\b)"\}\]\}\]\},"(?<dim_key>\b\w+\b)":"(?<dim_val>\S+)","(?<metric_key>\b\w+\b)":(?<metric_value>(?:(?:(?:[+-]?(?:[0-9]+(?:\.[0-9]+)?|\.[0-9]+)))))\}$
    Time_Key    timestamp
    Time_Format %Y-%m-%dT%H:%M:%S%z
    Types       timestamp:integer metric_value:float

# ==============================================================================
# Fluent Bit Pipeline Filter (fluent-bit.conf)
# ==============================================================================
[FILTER]
    Name         parser
    Match        *
    Key_Name     log
    Parser       logpipeline_parser
    Reserve_Data On

Vector.dev (Remap VRL)

parse_regex!

# ==============================================================================
# Vector.dev Remap Language (VRL) Transform
# Use inside a 'remap' transform in vector.yaml
# ==============================================================================
.parsed, err = parse_regex(.message, r'^\{"_aws":\{"Timestamp":(?<timestamp>(?:[+-]?(?:[0-9]+))),"CloudWatchMetrics":\[\{"Namespace":"(?<namespace>\b\w+\b)","Dimensions":\[\["(?<dimension>\b\w+\b)"\]\],"Metrics":\[\{"Name":"(?<metric_name>\b\w+\b)","Unit":"(?<unit>\b\w+\b)"\}\]\}\]\},"(?<dim_key>\b\w+\b)":"(?<dim_val>\S+)","(?<metric_key>\b\w+\b)":(?<metric_value>(?:(?:(?:[+-]?(?:[0-9]+(?:\.[0-9]+)?|\.[0-9]+)))))\}$')

if err == null {
    . = merge(., .parsed)
    del(.parsed)

    # Type coercions
    .timestamp = to_int!(.timestamp)
    .metric_value = to_float!(.metric_value)

} else {
    log("LogPipeline parsing warning: " + err, level: "warn")
}

# ==============================================================================
# vector.yaml Pipeline Component
# ==============================================================================
transforms:
  parse_logs:
    type: remap
    inputs: ["source_logs"]
    source: |
      .parsed, err = parse_regex(.message, r'^\{"_aws":\{"Timestamp":(?<timestamp>(?:[+-]?(?:[0-9]+))),"CloudWatchMetrics":\[\{"Namespace":"(?<namespace>\b\w+\b)","Dimensions":\[\["(?<dimension>\b\w+\b)"\]\],"Metrics":\[\{"Name":"(?<metric_name>\b\w+\b)","Unit":"(?<unit>\b\w+\b)"\}\]\}\]\},"(?<dim_key>\b\w+\b)":"(?<dim_val>\S+)","(?<metric_key>\b\w+\b)":(?<metric_value>(?:(?:(?:[+-]?(?:[0-9]+(?:\.[0-9]+)?|\.[0-9]+)))))\}$')
      if err == null {
        . = merge(., .parsed)
        del(.parsed)
      }

Datadog Log Pipeline Grok Parser

match_rules

# ==============================================================================
# Datadog Log Processing Pipeline Grok Parser
# Navigate to: Logs -> Configuration -> Pipelines -> Add Processor -> Grok Parser
# ==============================================================================

# Match Rule:
rule \{"_aws":\{"Timestamp":%{INT:timestamp},"CloudWatchMetrics":\[\{"Namespace":"%{WORD:namespace}","Dimensions":\[\["%{WORD:dimension}"\]\],"Metrics":\[\{"Name":"%{WORD:metric_name}","Unit":"%{WORD:unit}"\}\]\}\]\},"%{WORD:dim_key}":"%{NOTSPACE:dim_val}","%{WORD:metric_key}":%{NUMBER:metric_value}\}

# Complete Datadog Pipeline Processor JSON:
{
  "type": "grok-parser",
  "name": "LogPipeline Grok Parser",
  "is_enabled": true,
  "source": "message",
  "samples": [],
  "grok": {
    "match_rules": "rule \\{\"_aws\":\\{\"Timestamp\":%{INT:timestamp},\"CloudWatchMetrics\":\\[\\{\"Namespace\":\"%{WORD:namespace}\",\"Dimensions\":\\[\\[\"%{WORD:dimension}\"\\]\\],\"Metrics\":\\[\\{\"Name\":\"%{WORD:metric_name}\",\"Unit\":\"%{WORD:unit}\"\\}\\]\\}\\]\\},\"%{WORD:dim_key}\":\"%{NOTSPACE:dim_val}\",\"%{WORD:metric_key}\":%{NUMBER:metric_value}\\}",
    "support_rules": ""
  }
}

# Target Fields Created:
# timestamp (integer), namespace (string), dimension (string), metric_name (string), unit (string), dim_key (string), dim_val (string), metric_key (string), metric_value (float)

OpenTelemetry Collector (transform processor)

regex_parser

# ==============================================================================
# OpenTelemetry Collector Configuration (otel-collector-config.yaml)
# Option 1: Filelog Receiver with regex_parser Operator
# ==============================================================================
receivers:
  filelog:
    include: [ /var/log/**/*.log ]
    start_at: beginning
    operators:
      - type: regex_parser
        id: logpipeline_regex_parser
        regex: '^\{"_aws":\{"Timestamp":(?<timestamp>(?:[+-]?(?:[0-9]+))),"CloudWatchMetrics":\[\{"Namespace":"(?<namespace>\b\w+\b)","Dimensions":\[\["(?<dimension>\b\w+\b)"\]\],"Metrics":\[\{"Name":"(?<metric_name>\b\w+\b)","Unit":"(?<unit>\b\w+\b)"\}\]\}\]\},"(?<dim_key>\b\w+\b)":"(?<dim_val>\S+)","(?<metric_key>\b\w+\b)":(?<metric_value>(?:(?:(?:[+-]?(?:[0-9]+(?:\.[0-9]+)?|\.[0-9]+)))))\}$'
        timestamp:
          parse_from: attributes.timestamp
          layout: '%Y-%m-%dT%H:%M:%S%z'

# ==============================================================================
# Option 2: Transform Processor (OTel Transformation Language - OTTL)
# ==============================================================================
processors:
  transform:
    error_mode: ignore
    log_statements:
      - context: log
        statements:
          - merge_maps(attributes, extract_patterns(body, "^\\{\"_aws\":\\{\"Timestamp\":(?<timestamp>(?:[+-]?(?:[0-9]+))),\"CloudWatchMetrics\":\\[\\{\"Namespace\":\"(?<namespace>\\b\\w+\\b)\",\"Dimensions\":\\[\\[\"(?<dimension>\\b\\w+\\b)\"\\]\\],\"Metrics\":\\[\\{\"Name\":\"(?<metric_name>\\b\\w+\\b)\",\"Unit\":\"(?<unit>\\b\\w+\\b)\"\\}\\]\\}\\]\\},\"(?<dim_key>\\b\\w+\\b)\":\"(?<dim_val>\\S+)\",\"(?<metric_key>\\b\\w+\\b)\":(?<metric_value>(?:(?:(?:[+-]?(?:[0-9]+(?:\\.[0-9]+)?|\\.[0-9]+)))))\\}$"), "insert")

service:
  pipelines:
    logs:
      receivers: [filelog]
      processors: [transform]
      exporters: [otlp]

Logstash Filter Configuration

filter.grok

# ==============================================================================
# Logstash Pipeline Configuration (/etc/logstash/conf.d/logpipeline.conf)
# ==============================================================================
filter {
  grok {
    match => { "message" => "\{\"_aws\":\{\"Timestamp\":%{INT:timestamp:integer},\"CloudWatchMetrics\":\[\{\"Namespace\":\"%{WORD:namespace}\",\"Dimensions\":\[\[\"%{WORD:dimension}\"\]\],\"Metrics\":\[\{\"Name\":\"%{WORD:metric_name}\",\"Unit\":\"%{WORD:unit}\"\}\]\}\]\},\"%{WORD:dim_key}\":\"%{NOTSPACE:dim_val}\",\"%{WORD:metric_key}\":%{NUMBER:metric_value:float}\}" }
    tag_on_failure => [ "_grokparsefailure" ]
  }

  date {
    match => [ "timestamp", "ISO8601", "dd/MMM/yyyy:HH:mm:ss Z" ]
    target => "@timestamp"
    remove_field => [ "timestamp" ]
  }
}