> ## Documentation Index
> Fetch the complete documentation index at: https://docs.chronosphere.io/llms.txt
> Use this file to discover all available pages before exploring further.

# Use OpenTelemetry protocol endpoints

export const MyTenant = () => <>
    Replace <em><code>TENANT</code></em> with the name of your Observability Platform tenant.
  </>;

Chronosphere Observability Platform supports ingestion of metrics and traces using
OpenTelemetry protocol (OTLP) gRPC and HTTP ingestion endpoints.

<Note>
  For best results, ingest telemetry through a local OpenTelemetry Collector
  for processing and batching. Sending telemetry directly from an OpenTelemetry SDK
  can be less reliable.
</Note>

## Requirements

Your OTLP configuration must meet certain requirements.

* **`API-Token` header:** All requests must include the `API-Token` HTTP header,
  and its value must be the API token of a valid
  [service account](/administer/accounts-teams/service-accounts)
  with write permissions.
* **OTLP version support:** Only v1.0 and higher are supported.
* **Supported compression methods:** Chronosphere supports `gzip`, `snappy`,
  and `zstd` compression, and recommends `zstd`.
* **Encryption:** All requests must be encrypted using TLS.

- **Payload requirements:**
  * Compressed payloads should be smaller than 16 MB.
  * Metric payloads must contain fewer than 10,000 items.
  * For details, see [Optimize the batch processor](#optimize-the-batch-processor).

### Recommendations

Chronosphere recommends setting the timeout value for a metrics or trace request
to 30 seconds, and to enable retries to prevent transient network errors from causing
data loss.

Chronosphere recommends using gRPC instead of HTTP when possible, because gRPC is a more efficient
protocol for sending telemetry to the Observability Platform.

### Optimize the batch processor

The
[OpenTelemetry Batch Processor](https://github.com/open-telemetry/opentelemetry-collector/tree/main/processor/batchprocessor)
defines batch limits by the quantity of items, not the size of the batch in bytes.
Configure the batch processor as follows to prevent batches from exceeding the maximum
payload size:

* **`send_batch_size`:** Set to `1000` to trigger sending a batch when it reaches
  1,000 items.
* **`send_batch_max_size`:** Set to `2000` to force splitting the batch if it exceeds
  more than 2,000 items.
* **`timeout`:** Set to `1s` to send any items in the buffer after 1 second.

## OTLP endpoint URLs

Observability Platform provides gRPC and HTTP endpoints for metrics and traces.

| Telemetry type | gRPC Endpoint | HTTP endpoint URL                              |
| -------------- | ------------- | ---------------------------------------------- |
| Metrics        | `TENANT:443`  | `https://TENANT/data/opentelemetry/v1/metrics` |
| Traces         | `TENANT:443`  | `https://TENANT/data/opentelemetry/v1/traces`  |

<MyTenant />

OpenTelemetry exporters complete the path based on the telemetry type. For the `otlphttp`
exporter, set the endpoint base path as follows:

```yaml theme={null}
exporters:
  otlphttp/chronosphere:
    endpoint: https://TENANT/data/opentelemetry/
    ...
```

For an example OTLP exporter configuration, see
[Configure the OpenTelemetry Collector](/ingest/metrics-traces/otel/otel-ingest#configure-the-opentelemetry-collector).

## Metrics partial success responses

If a metrics ingestion request succeeds but any series are rejected, the ingester
returns an [OTLP partial success](https://opentelemetry.io/docs/specs/otlp/#partial-success)
response that includes a count (`rejected_data_points`) and a description (`error_message`).
The verbosity of the description is controlled by the `Chronosphere-Metrics-Validation-Response`
header, which can be set to `SHORT`, `SUMMARY`, or `DETAILED` in the
[OTel Collector configuration](/ingest/metrics-traces/otel/otel-ingest#configure-the-opentelemetry-collector).

When a partial success response is returned, the client must not retry the request.
All valid series from the request have been accepted, and the rejected series are
dropped.

For details about the response's design, see the
[OTLP Partial Success Specification](https://opentelemetry.io/docs/specs/otlp/#otlpgrpc-response).