- The deprecated sources convert each OpenTelemetry record into a Mezmo-specific event. The main payload (log body, metric, or span) goes in
message, and the OpenTelemetry context (resource, scope, attributes) goes inmetadata. - The OpenTelemetry source passes each record through in its original OpenTelemetry JSON form. Everything, including the resource and scope, is inside
message.
At a glance
The OpenTelemetry source envelope
All three outputs share the same envelope.message.type is log, metric, or trace, and message.record holds the signal-specific fields:
annotations.appis set from theservice.nameresource attribute, ork8s.container.nameif that isn’t set.annotations.hostis set from thek8s.pod.nameresource attribute, orhost.nameif that isn’t set.timestampis set from the record’s time:
Logs
In the deprecated Logs source,
metadata.resource holds resource attributes directly. In the deprecated Metrics and Traces sources, they’re nested under metadata.resource.attributes. In the OpenTelemetry source, they’re always at message.resource.attributes.Log Records With an Empty Body
Some OpenTelemetry exporters, such as Codex, put all of a log record’s data in its attributes and send no body. The OpenTelemetry source keeps these records and setsmessage.record.body to an empty string (""). If your processors match on the log body, also check message.record.attributes so these records aren’t skipped.
The source drops a log record only when it has both an empty body and no attributes.
Metrics
Which fields appear in
dataPoint depends on the metric type. The example above is a gauge. Sums, histograms, and summaries can carry other fields.Sending OpenTelemetry Metrics to Metric Destinations
You can connect the Metrics output directly to a metric destination such as New Relic or Datadog Metrics. The pipeline converts each OpenTelemetry metric to the Mezmo metric format before the destination sends it, so you don’t need to remap the fields yourself. The conversion works like this:- Gauge, sum, histogram, exponential histogram, and summary data points are converted. Exponential histograms are approximated as regular histograms.
- Resource attributes and data point attributes become metric tags.
- Metric names and tag keys keep their original OpenTelemetry form. For example,
k8s.pod.memory.major_page_faultskeeps its dots and gets no_totalsuffix. - If a required field is missing or invalid, the event is dropped and the error names the field, for example
record.metric.name.
Traces
Migrating from the deprecated sources
When you replace the deprecated sources with the OpenTelemetry source:- Add one OpenTelemetry source, and connect each downstream processor to its Logs, Metrics, or Traces output.
- Update the field paths your processors use, following the tables above. For example:
- Logs:
messagebecomesmessage.record.body - Metrics:
message.namebecomesmessage.record.metric.name - Traces:
message.trace_idbecomesmessage.record.traceId - All signals:
metadata.resource…becomesmessage.resource.attributes…
- Logs:
- Convert values before doing math or numeric comparisons: metric values,
*UnixNanotimes, and many attribute values arrive as strings. - Check for anything that relies on fields the new source doesn’t set, such as
annotations.levelon logs ormetadata.levelon traces. - Check dashboards and alerts that query metrics by name or tag. Metrics from the OpenTelemetry source keep their original names and attribute keys, while the deprecated source normalized them. See Sending OpenTelemetry Metrics to Metric Destinations.

