Skip to main content

Send2Bucket vs the OpenTelemetry Collector S3 exporter

The OpenTelemetry Collector's awss3exporter writes OTLP data from a Collector you run to an S3 bucket. Send2Bucket is a hosted OTLP endpoint that does the buffering, retries and delivery for you, and can also write to tables.

Short answer: if you already operate Collectors and only need OTLP files in S3, the exporter is free and works. Choose Send2Bucket when you do not want to run, scale and monitor the write path, or when you want Iceberg, S3 Tables, DuckLake or a non-AWS bucket.

At a glance​

Send2BucketCollector awss3exporter
What you runNothing: point an OTLP exporter at Send2BucketCollector instances, their queues and scaling
Protocols inOTLP over HTTP and gRPC, plus JSON eventsWhatever receivers the Collector has
Outputgzip NDJSON with OTLP JSON in each record, or table rowsOTLP JSON or OTLP protobuf files, and other marshalers
DestinationsS3, Google Cloud Storage, S3-compatible, Iceberg REST, S3 Tables, Supabase Analytics, DuckLakeS3 and S3-compatible endpoints
Durability200 only after the batch is stored durably; retries until the destination confirmsDepends on how you configure the Collector's queue and storage
PricePlans and billingFree software; you pay for the machines

When the exporter fits better​

  • You already run Collectors per cluster and have their queues and alerts in place.
  • You want OTLP protobuf files or a custom file layout.
  • You only need raw files in S3, not tables.

When Send2Bucket fits better​

  • No Collector to run. SDKs and existing Collectors export straight to the hosted endpoint. See the API reference.
  • Tables, not only files. Logs, spans and metrics can be appended to an Iceberg, S3 Tables or DuckLake table as they arrive.
  • One place for events too. Product events over JSON go to the same destinations with the same key.

Keeping your Collectors​

You can keep Collectors for sampling and enrichment and replace only the S3 exporter: add an otlphttp exporter with the Send2Bucket endpoint and API key header. The Quick start shows the configuration. Delivered records can be read with DuckDB or any engine in Querying.

Sources​