Dedupe events Transform

The Vector dedupe transform deduplicates events to reduce data volume by eliminating copies of data.

Configuration

• Common
• Advanced

• vector.toml
• vector.yaml
• vector.json

[transforms.my_transform_id]
  # General
  type = "dedupe" # required
  inputs = ["my-source-or-transform-id", "prefix-*"] # required

  # Fields
  fields.match = ["timestamp", "host", "message"] # optional, default

• optionaltable

  cache

  Options controlling how we cache recent Events for future duplicate checking.

  • commonoptionaluint

    num_events

    The number of recent Events to cache and compare new incoming Events against.

    • Default: 5000
• commonrequiredtable

  fields

  Options controlling what fields to match against.

  • optional[string]

    ignore

    The field names to ignore when deciding if an Event is a duplicate. Incompatible with the fields.match option.

    • View examples
  • commonoptional[string]

    match

    The field names considered when deciding if an Event is a duplicate. This can also be globally set via the global log_schema options. Incompatible with the fields.ignore option.

    • Default: ["timestamp","host","message"]
    • View examples

Telemetry

This component provides the following metrics that can be retrieved through the internal_metrics source. See the metrics section in the monitoring page for more info.

• counter

  events_discarded_total

  The total number of events discarded by this component. This metric includes the following tags:

  • instance - The Vector instance identified by host and port.

  • job - The name of the job producing Vector metrics.

• counter

  events_in_total

  The total number of events accepted by this component. This metric includes the following tags:

  • component_kind - The Vector component kind.

  • component_name - The Vector component ID.

  • component_type - The Vector component type.

  • instance - The Vector instance identified by host and port.

  • job - The name of the job producing Vector metrics.

• counter

  processed_events_total

  The total number of events processed by this component. This metric includes the following tags:

  • component_kind - The Vector component kind.

  • component_name - The Vector component ID.

  • component_type - The Vector component type.

  • file - The file that produced the error

  • instance - The Vector instance identified by host and port.

  • job - The name of the job producing Vector metrics.

• counter

  events_out_total

  The total number of events emitted by this component. This metric includes the following tags:

  • component_kind - The Vector component kind.

  • component_name - The Vector component ID.

  • component_type - The Vector component type.

  • instance - The Vector instance identified by host and port.

  • job - The name of the job producing Vector metrics.

• counter

  processed_bytes_total

  The total number of bytes processed by the component. This metric includes the following tags:

  • component_kind - The Vector component kind.

  • component_name - The Vector component ID.

  • component_type - The Vector component type.

  • instance - The Vector instance identified by host and port.

  • job - The name of the job producing Vector metrics.

How It Works

Cache Behavior

This transform is backed by an LRU cache of size cache.num_events. That means that this transform will cache information in memory for the last cache.num_events Events that it has processed. Entries will be removed from the cache in the order they were inserted. If an Event is received that is considered a duplicate of an Event already in the cache that will put that event back to the head of the cache and reset its place in line, making it once again last entry in line to be evicted.

Memory Usage Details

Each entry in the cache corresponds to an incoming Event and contains a copy of the 'value' data for all fields in the Event being considered for matching. When using fields.match this will be the list of fields specified in that configuration option. When using fields.ignore that will include all fields present in the incoming event except those specified in fields.ignore. Each entry also uses a single byte per field to store the type information of that field. When using fields.ignore each cache entry additionally stores a copy of each field name being considered for matching. When using fields.match storing the field names is not necessary.

Memory Utilization Estimation

If you want to estimate the memory requirements of this transform for your dataset, you can do so with these formulas:

When using fields.match:

Sum(the average size of the *data* (but not including the field name) for each field in `fields.match`) * `cache.num_events`

When using fields.ignore:

(Sum(the average size of each incoming Event) - (the average size of the field name *and* value for each field in `fields.ignore`)) * `cache.num_events`

Missing Fields

Fields with explicit null values will always be considered different than if that field was omitted entirely. For example, if you run this transform with fields.match = ["a"], the event "{a: null, b:5}" will be considered different to the event "{b:5}".

State

This component is stateful, meaning its behavior changes based on previous inputs (events). State is not preserved across restarts, therefore state-dependent behavior will reset between restarts and depend on the inputs (events) received since the most recent restart.