Object storage connector

Class: object_storage — Sink-only in the UI: subscribes to a stream and writes objects to S3-compatible storage as Parquet or JSON Lines, with optional compression and batching for large uploads.

Create and edit under Sinks only. Advanced Settings may expose more runtime options depending on deployment and permissions.

Source and sink behavior

Role	Behavior
Source	Not supported in the UI for this class (type must be sink).
Sink	Consumes the subscribed stream, batches events, and uploads objects using bucket, region, optional custom endpoint, and format settings.
Streams	Upstream tasks / pipelines must publish to the stream id this sink consumes (Streams).

Required fields

Every connector row

Field	Required	Notes
name	Yes	Display name; id derived from it.
class	Yes	Must be object_storage.
stream	Yes	Resolved stream id (sink reads here).
type	Yes	Must be sink (UI rejects source for this class).
config	Yes	Class-specific object; see below.

Class object_storage — required configuration

Setting	Required	Notes
bucket	Yes	Target bucket (UI errors if empty).
region	Conditional	Required for default AWS-style endpoints; optional when a custom endpoint is set—confirm for your deployment.

UI validation

• emit_verification_manifest requires integrity features to be enabled.
• max_batch_bytes ≤ max_object_bytes when both are set (same relationship for nested upload.batch.max_bytes vs upload.object.max_bytes when present).
• format: parquet must not combine with outer compression: gzip (Core + UI).

Create connector

1. Open Sinks → Create.
2. Set Class to Object storage (S3-compatible), set Sink Name, stream behavior, and Enabled.
3. Enter bucket and region (or custom endpoint per form).
4. Choose format, compression, batch/object size limits, and credentials.
5. Save, then ensure tasks publish into the subscribed stream.

Sink (UI)

The Object storage sink connector form.

UI area	Connector settings (typical)
Bucket / region	bucket, region
Endpoint	object_storage_endpoint, force_path_style
Prefix / layout	prefix
Format	object_storage_format, parquet_compression, …
Batch / size	max_batch_*, max_object_bytes
Credentials	role_arn, keys, or instance metadata patterns per build

Configuration

Bucket and layout

• bucket, region, prefix, object_storage_endpoint, force_path_style — Namespace and S3-compatible API targets (MinIO, ECS, etc.).

Format

• object_storage_format — json_lines vs Parquet plus Parquet row-group and compression options.

Batching and size limits

• max_batch_events, max_batch_bytes, max_batch_age_ms, max_object_bytes — Flush cadence and object caps.

Credentials and IAM

• role_arn, role_session_name, external_id — Assume-role style access when supported.

Reliability

• max_retries, retry_initial_backoff_ms, integrity_enabled, emit_verification_manifest — Retries and optional integrity features when present.

Timestamps

• timestamp — How event time is reflected in object metadata or columns.

Runtime behavior

• The sink runs after deployment when Enabled; it drains the stream and uploads asynchronously according to batch settings.
• With delivery.mode best_effort (the default), the sink acknowledges when events are appended to the in-memory object. With at_least_once, it acknowledges when PutObject succeeds for the object that holds that upload.batch. Bytes still only in memory are not delivered.
• Disabled sinks do not write objects.

Performance and operational notes

• Right-size batch thresholds to object size limits and storage rate policies.
• Prefer integrity manifests when compliance requires end-to-end object verification.

Related pages

• Sinks
• Core concepts — Sink connectors (Object Storage row)