Sinks

Sinks export telemetry to backends. Each sink declares its upstream sources (sources or processors) and only receives signals it supports.

---

otlp — OTLP Export

Exports all three signals to any OTLP-compatible backend over gRPC or HTTP. This is the recommended sink for sending data to InfraSage.

Signals: Logs, Metrics, Traces

sinks:
  infrasage:
    type: otlp
    endpoint: "https://api.infrasage.dev"
    protocol: http                                         # http or grpc (default: grpc)
    auth_header: "Bearer ${INFRASAGE_API_KEY}"
    sources: [batch_main]

gRPC mode (default):

sinks:
  otel_collector:
    type: otlp
    endpoint: "http://otel-collector:4317"
    # protocol defaults to grpc
    sources: [batch_main]

With custom headers:

sinks:
  custom_backend:
    type: otlp
    endpoint: "https://ingest.example.com"
    protocol: http
    headers:
      x-team-id: "platform"
      x-datacenter: "${DC}"
    sources: [batch_main]

Option	Default	Description
endpoint	required	Backend URL
protocol	grpc	Transport: grpc or http
auth_header	—	Shortcut for Authorization header
headers	{}	Additional HTTP headers or gRPC metadata

TLS is automatically enabled when the endpoint uses https:// or port 443.

---

clickhouse — ClickHouse

Writes to ClickHouse using the native TCP protocol with LZ4 compression. Creates tables automatically on first start.

Signals: Logs, Metrics, Traces

sinks:
  clickhouse_out:
    type: clickhouse
    endpoint: "tcp://clickhouse:9000"
    database: infrasage
    username: default
    password: "${CH_PASSWORD}"
    compression: lz4
    async_insert: true
    sources: [batch_main, trace_sample]

Option	Default	Description
endpoint	required	ClickHouse native TCP URL
database	default	Database name
username	default	Auth username
password	—	Auth password
compression	lz4	Compression: lz4, zstd, or none
async_insert	true	Use ClickHouse async insert mode

Tables created automatically:

Table	Contents
logs	Log records with resource attrs, severity, body, trace/span IDs
metrics_gauge	Gauge data points
metrics_sum	Monotonic and non-monotonic sum data points
traces	Spans with resource attrs, timing, status, attributes

---

prometheus_rw — Prometheus Remote Write

Exports metrics via the Prometheus Remote Write protocol (snappy-compressed protobuf).

Signals: Metrics

sinks:
  prometheus:
    type: prometheus_rw
    endpoint: "https://prometheus:9090/api/v1/write"
    sources: [batch_main]

Compatible with Prometheus, Grafana Cloud, Thanos, Cortex, Mimir, and VictoriaMetrics.

Option	Default	Description
endpoint	required	Remote Write endpoint URL

---

kafka — Kafka Producer

Publishes OTLP-encoded messages to Kafka topics, one topic per signal type.

Signals: Logs, Metrics, Traces

sinks:
  kafka_out:
    type: kafka
    brokers: ["kafka:9092"]
    topics:
      logs: "otlp-logs"
      metrics: "otlp-metrics"
      traces: "otlp-traces"
    encoding: otlp_proto      # or: otlp_json
    compression: snappy
    sources: [batch_main]

Option	Default	Description
brokers	required	Kafka broker addresses
topics	required	Topic name per signal type
encoding	otlp_proto	otlp_proto or otlp_json
compression	snappy	snappy, gzip, lz4, or none

---

s3 — Amazon S3

Writes logs as NDJSON and metrics as Parquet to S3-compatible object storage. Files are partitioned by signal type and date.

Signals: Logs, Metrics

sinks:
  s3_archive:
    type: s3
    bucket: "my-telemetry-archive"
    prefix: "infrasagent"
    region: "us-east-1"
    flush_interval: 300s       # write a new object every 5 minutes
    sources: [batch_main]

Option	Default	Description
bucket	required	S3 bucket name
prefix	infrasagent	Object key prefix
region	from env	AWS region
endpoint	—	Custom endpoint for S3-compatible stores (MinIO, etc.)
flush_interval	300s	Maximum time before writing a new object

Credentials are read from the standard AWS credential chain (AWS_ACCESS_KEY_ID, ~/.aws/credentials, instance metadata).

---

loki — Grafana Loki

Pushes log streams to the Loki HTTP push API.

Signals: Logs

sinks:
  loki_out:
    type: loki
    endpoint: "http://loki:3100"
    labels:
      job: "infrasagent"
      env: "${ENV:-production}"
    sources: [batch_main]

Option	Default	Description
endpoint	required	Loki base URL
labels	{}	Static labels to attach to every stream
tenant_id	—	X-Scope-OrgID header value for multi-tenant Loki

Resource attributes from each log record are included as stream labels.

---

elasticsearch — Elasticsearch

Writes log records to Elasticsearch via the Bulk API.

Signals: Logs

sinks:
  es_out:
    type: elasticsearch
    addresses: ["https://elasticsearch:9200"]
    index: "infrasage-logs"
    username: elastic
    password: "${ES_PASSWORD}"
    sources: [batch_main]

Option	Default	Description
addresses	required	Elasticsearch node URLs
index	infrasage-logs	Target index name
username	—	Basic auth username
password	—	Basic auth password

---

http — Generic HTTP Webhook

Posts telemetry as OTLP JSON to any HTTP endpoint. Useful for custom ingestion pipelines or webhook-based backends.

Signals: Logs, Metrics, Traces

sinks:
  webhook:
    type: http
    endpoint: "https://hooks.example.com/ingest"
    method: POST
    headers:
      Authorization: "Bearer ${WEBHOOK_TOKEN}"
      X-Source: "infrasagent"
    timeout: 10s
    sources: [batch_main]

Option	Default	Description
endpoint	required	Target URL
method	POST	HTTP method
headers	{}	Extra request headers
timeout	10s	Per-request timeout

---

stdout — Standard Output

Prints telemetry as JSON to stdout. Useful during development and debugging.

Signals: Logs, Metrics, Traces

sinks:
  debug:
    type: stdout
    pretty: true     # pretty-print JSON
    sources: [otlp_in]

Option	Default	Description
pretty	false	Indent JSON output