Sinks

A sink is where aggregate sends its results. Every sink is generated from the same facts files, so you can turn on several at once and add another later without scanning anything again.

Sink	Use it for	Writes	Tutorial
openmetrics	Prometheus dashboards and alerts	metrics.prom, or a push to a Pushgateway	Prometheus
otlp	Datadog, Honeycomb, Grafana Cloud and other OpenTelemetry backends	Gauges over OTLP gRPC	OpenTelemetry
parquet	History, trends and SQL in a warehouse	modules.parquet, module_calls.parquet	Parquet
csv	Spreadsheets and quick looks	modules.csv, module_calls.csv	CSV
json	Archiving a run, your own tooling	metrics.json and a copy of every facts file	JSON
markdown	PR comments and job summaries	One report per repo	Markdown
sarif	Code scanning alerts and PR annotations	One SARIF log per repo	SARIF

Choose sinks on the command line

--sink names the sinks for one run. Repeat it for more than one:

$ tfmetrics aggregate facts/ --offline --sink openmetrics --sink parquet --out metrics/
aggregated 2 repos (2 with Terraform) into 45 series

File sinks write into --out (default metrics/) unless their path option says otherwise. With no --sink and no sinks in the config, aggregate writes the json sink.

scan and org take --sink too, to scan and aggregate in one step. scan never looks up module versions; org does, unless you run aggregate separately with --offline.

Configure sinks in tfmetrics.yaml

For anything you run more than once, list the sinks under sinks. Each item is a one-key map from the sink's name to its options:

tfmetrics.yaml

sinks:
  - openmetrics:
      path: out/prom/metrics.prom
  - parquet:
      path: out/tables
  - markdown:
      path: out/report.md

aggregate with no --sink then writes all three:

$ tfmetrics aggregate facts/ --offline
aggregated 2 repos (2 with Terraform) into 45 series
$ find out -type f | sort
out/prom/metrics.prom
out/report.md
out/tables/module_calls.parquet
out/tables/modules.parquet

--sink narrows the run to the sinks you name, and each one still takes its options from the matching entry in the config:

$ tfmetrics aggregate facts/ --offline --sink parquet
aggregated 2 repos (2 with Terraform) into 45 series
$ find out -type f | sort
out/tables/module_calls.parquet
out/tables/modules.parquet

Things every sink shares

• Version drift needs the network. aggregate looks up module versions for tf_module_versions_behind. --offline skips the lookups and leaves that metric out; every other metric is unchanged.
• Labels are shaped before any sink runs. labels.properties adds custom properties as labels, and labels.drop removes high-cardinality labels and sums the series that merge. Every gauge sink sees the same series.
• Local paths only. File sinks refuse s3:// and other object-store URLs. Write locally, then upload with aws s3 cp, gcloud storage cp or similar; the Parquet tutorial shows the pattern.
• Re-aggregate any time. Facts files are the source of truth. Changing a sink, a label or a weight and running aggregate again over saved facts regenerates the output without a scan.