Connect your OpenTelemetry Collector

Tayga ingests OTLP traces and logs, over gRPC and over HTTP. It does not ingest metrics. Any OpenTelemetry Collector can send to it: add one exporter and list it in the traces and logs pipelines, next to the exporters you already have. Your other backends keep getting the same data.

Where to send

Install	gRPC	HTTP
Docker Compose, Collector on the host	localhost:4317	http://localhost:4318
Docker Compose, Collector in a container on tayga_default	tayga-ingest:4317	http://tayga-ingest:4318
Helm, release tayga in namespace tayga	tayga-ingest.tayga.svc:4317	http://tayga-ingest.tayga.svc:4318
The OTel demo	already wired: the demo’s Collector sends to tayga-ingest:4317

The host ports follow TAYGA_OTLP_GRPC_PORT and TAYGA_OTLP_HTTP_PORT (or --ports) in the Compose install.

The exporter

gRPC

otel-collector.yaml

exporters:
  otlp_grpc/tayga:
    endpoint: localhost:4317        # host:port, no scheme
    tls:
      insecure: true                # tayga-ingest serves plain gRPC
    compression: gzip
    timeout: 30s                    # above ingest's 25 s Kafka delivery timeout
    sending_queue:
      enabled: true
    retry_on_failure:
      enabled: true
      max_elapsed_time: 300s

service:
  pipelines:
    traces:
      receivers: [otlp]
      processors: [batch]
      exporters: [otlp_grpc/tayga]  # keep your other exporters in the list
    logs:
      receivers: [otlp]
      processors: [batch]
      exporters: [otlp_grpc/tayga]

HTTP

otel-collector.yaml

exporters:
  otlp_http/tayga:
    endpoint: http://localhost:4318   # the exporter appends /v1/traces and /v1/logs
    compression: gzip
    timeout: 30s
    sending_queue:
      enabled: true
    retry_on_failure:
      enabled: true
      max_elapsed_time: 300s

service:
  pipelines:
    traces:
      receivers: [otlp]
      processors: [batch]
      exporters: [otlp_http/tayga]
    logs:
      receivers: [otlp]
      processors: [batch]
      exporters: [otlp_http/tayga]

Both configs were checked with otelcol-contrib validate (Collector contrib 0.162.0). The bundle’s otel-collector.yaml is the gRPC one, complete with an OTLP receiver.

Exporter names

Recent Collector releases name the OTLP exporters otlp_grpc and otlp_http. Older releases call them otlp and otlphttp; 0.162.0 accepts both spellings. Use the one your Collector’s documentation shows.

Collector config merging replaces lists. If you add Tayga through an extra config file (as the demo does with deploy/otelcol-config-tayga.yml), repeat every existing exporter in the exporters list of each pipeline, or they stop receiving data.

gRPC or HTTP

Both carry the same data and land in the same place. Choose by what fits your network:

	gRPC (4317)	HTTP (4318)
Paths	the OTLP trace and log services	POST /v1/traces, POST /v1/logs
Encoding	protobuf; gzip accepted	protobuf or JSON; gzip accepted
Message size	up to 64 MiB per message	up to 64 MiB per body
Good for	Collector to Tayga inside one network: the default	proxies and load balancers that only speak HTTP/1.1, and SDKs in browsers or serverless runtimes

Tayga’s ingest also serves /metrics on the HTTP port.

Steps to connect

1. Add the exporter above to your Collector config, with the endpoint from Where to send.

2. Append it to the exporters of your traces and logs pipelines. Leave metrics pipelines alone; Tayga does not ingest metrics.

3. Restart the Collector, then check that data arrives: Tayga’s Pipeline page shows ingest records per second rising, and the Traces page lists the new traces within seconds. The Collector’s own logs show export errors if it cannot reach Tayga.

Network and security

• No authentication, no TLS on the OTLP ports. tayga-ingest accepts data from anyone who reaches it. Keep it on a trusted network: the Compose install binds to 127.0.0.1, and the Helm chart’s ingest Service is a ClusterIP by default. If the traffic crosses an untrusted network, put a TLS-terminating proxy or a Collector in front.
• A Collector in front is the safer shape. Your services send to a Collector you control, and only that Collector talks to Tayga. It also batches, compresses, and retries through a short Tayga restart.
• Sampling. Tayga builds stories from the traces it receives. Tail sampling that drops successful traces would skew the baselines that slow stories are measured against; head sampling that drops a failing trace means no story for it.

Sending without a Collector

SDKs can send to Tayga directly with the standard OTLP variables:

export OTEL_EXPORTER_OTLP_ENDPOINT=http://localhost:4318
export OTEL_EXPORTER_OTLP_PROTOCOL=http/protobuf    # or grpc, with port 4317

Each SDK then sends traces and logs (if its log signal is enabled) to Tayga only. To send to Tayga and an existing backend at once, use a Collector.