groupbyattrsprocessor

Group by Attributes processor

Status
Stability	beta: traces, metrics, logs
Distributions	contrib
Issues
Code Owners	@rnishtala-sumo

Description

This processor re-associates spans, log records and metric datapoints to a Resource that matches with the specified attributes. As a result, all spans, log records or metric datapoints with the same values for the specified attributes are "grouped" under the same Resource.

Typical use cases:

• extract resources from "flat" data formats, such as Fluentbit logs or Prometheus metrics
• associate Prometheus metrics to a Resource that describes the relevant host, based on label present on all metrics
• optimize data packaging by extracting common attributes
• compacting multiple records that share the same Resource and InstrumentationLibrary attributes but are under multiple ResourceSpans/ResourceMetrics/ResourceLogs, into a single ResourceSpans/ResourceMetrics/ResourceLogs (when empty list of keys is being provided). This might happen e.g. when groupbytrace processor is being used or data comes in multiple requests. By compacting data, it takes less memory, is more efficiently processed, serialized and the number of export requests is reduced.

It is recommended to use the groupbyattrs processor together with batch processor, as a consecutive step, as this will reduce the fragmentation of data (by grouping records together under matching Resource/Instrumentation Library)

Examples

Grouping metrics

Consider the below metrics, all originally associated to the same Resource:

Resource {host.name="localhost",source="prom"}
  Metric "gauge-1" (GAUGE)
    DataPoint {host.name="host-A",id="eth0"}
    DataPoint {host.name="host-A",id="eth0"}
    DataPoint {host.name="host-B",id="eth0"}
  Metric "gauge-1" (GAUGE) // Identical to previous Metric
    DataPoint {host.name="host-A",id="eth0"}
    DataPoint {host.name="host-A",id="eth0"}
    DataPoint {host.name="host-B",id="eth0"}
  Metric "mixed-type" (GAUGE)
    DataPoint {host.name="host-A",id="eth0"}
    DataPoint {host.name="host-A",id="eth0"}
    DataPoint {host.name="host-B",id="eth0"}
  Metric "mixed-type" (SUM)
    DataPoint {host.name="host-A",id="eth0"}
    DataPoint {host.name="host-A",id="eth0"}
  Metric "dont-move" (Gauge)
    DataPoint {id="eth0"}

With the below configuration, the groupbyattrs will re-associate the metrics with either host-A or host-B, based on the value of the host.name attribute.

processors:
  groupbyattrs:
    keys:
      - host.name

The output of the processor will therefore be:

Resource {host.name="localhost",source="prom"}
  Metric "dont-move" (Gauge)
    DataPoint {id="eth0"}

Resource {host.name="host-A",source="prom"}
  Metric "gauge-1"
    DataPoint {id="eth0"}
    DataPoint {id="eth0"}
    DataPoint {id="eth0"}
    DataPoint {id="eth0"}
  Metric "mixed-type" (GAUGE)
    DataPoint {id="eth0"}
    DataPoint {id="eth0"}
  Metric "mixed-type" (SUM)
    DataPoint {id="eth0"}
    DataPoint {id="eth0"}

Resource {host.name="host-B",source="prom"}
  Metric "gauge-1"
    DataPoint {id="eth0"}
    DataPoint {id="eth0"}
  Metric "mixed-type" (GAUGE)
    DataPoint {id="eth0"}

Notes:

• The DataPoints for the gauge-1 (GAUGE) metric were originally split under 2 Metric instances and have been merged in the output
• The DataPoints of the mixed-type (GAUGE) and mixed-type (SUM) metrics have not been merged under the same Metric, because their DataType is different
• The dont-move metric DataPoints don't have a host.name attribute and therefore remained under the original Resource
• The new Resources inherited the attributes from the original Resource (source="prom"), plus the specified attributes from the processed metrics (host.name="host-A" or host.name="host-B")
• The specified "grouping" attributes that are set on the new Resources are also removed from the metric DataPoints
• While not shown in the above example, the processor also merges collections of records under matching InstrumentationLibrary

Compaction

In some cases, the data might come in single requests to the collector or become fragmented due to use of groupbytrace processor. Even after batching there might be multiple duplicated ResourceSpans/ResourceLogs/ResourceMetrics objects, which leads to additional memory consumption, increased processing costs, inefficient serialization and increase of the export requests. As a remedy, groupbyattrs processor might be used to compact the data with matching Resource and InstrumentationLibrary properties.

For example, consider the following input:

Resource {host.name="localhost"}
  InstrumentationLibrary {name="MyLibrary"}
  Spans
    Span {span_id=1, ...}
  InstrumentationLibrary {name="OtherLibrary"}
  Spans
    Span {span_id=2, ...}
    
Resource {host.name="localhost"}
  InstrumentationLibrary {name="MyLibrary"}
  Spans
    Span {span_id=3, ...}
    
Resource {host.name="localhost"}
  InstrumentationLibrary {name="MyLibrary"}
  Spans
    Span {span_id=4, ...}
    
Resource {host.name="otherhost"}
  InstrumentationLibrary {name="MyLibrary"}
  Spans
    Span {span_id=5, ...}

With the below configuration, the groupbyattrs will re-associate the spans with matching Resource and InstrumentationLibrary.

processors:
  batch:
  groupbyattrs:

pipelines:
  traces:
    processors: [batch, groupbyattrs/grouping]
    ...

The output of the processor will therefore be:

Resource {host.name="localhost"}
  InstrumentationLibrary {name="MyLibrary"}
  Spans
    Span {span_id=1, ...}
    Span {span_id=3, ...}
    Span {span_id=4, ...}
  InstrumentationLibrary {name="OtherLibrary"}
  Spans
    Span {span_id=2, ...}

Resource {host.name="otherhost"}
  InstrumentationLibrary {name="MyLibrary"}
  Spans
    Span {span_id=5, ...}

Configuration

The configuration is very simple, as you only need to specify an array of attribute keys that will be used to "group" spans, log records or metric data points together, as in the below example:

processors:
  groupbyattrs:
    keys:
      - foo
      - bar

The keys property describes which attribute keys will be considered for grouping:

• If the processed span, log record and metric data point has at least one of the specified attributes key, it will be moved to a Resource with the same value for these attributes. The Resource will be created if none exists with the same attributes.
• If none of the specified attributes key is present in the processed span, log record or metric data point, it remains associated to the same Resource (no change), with multiple instances of the same Resource still compacted.

Please refer to:

• config.go for the config spec
• config.yaml for detailed examples on using the processor

Internal Metrics

The following internal metrics are recorded by this processor: