singer_sdk.RecordSink

class singer_sdk.RecordSink

Base class for singleton record writers.

This class supports both legacy and modern loading patterns:

Legacy pattern (deprecated but supported):

Override process_record(record: dict, context: dict) to load each record

Modern pattern (recommended):

Override load_record(record: dict) for cleaner, simpler immediate loading

Example (modern pattern):

class MyRecordSink(RecordSink):

def load_record(self, record: dict) -> None:

# Clear, simple loading self.api_client.post(record)

Example (legacy pattern - still works):

class MyLegacySink(RecordSink):

def process_record(self, record: dict, context: dict) -> None:

# Old style - context is unused for RecordSink self.api_client.post(record)

load_record(record)

Load a single record immediately to the target.

Modern pattern (recommended): Override this method for cleaner, simpler immediate record loading.

This method provides a cleaner alternative to process_record(). It takes only the record (no context dict), making the intent clear and reducing complexity.

Example

class MyRecordSink(RecordSink):

def load_record(self, record: dict) -> None:

# Simple, clear loading self.api_client.post(“/endpoint”, json=record)

Parameters:

record (dict) – The record to load immediately.

Return type:

None

Note

If you override this method, you do NOT need to override process_record(). The framework will automatically call this method instead.

final process_batch(context)

Do nothing and return immediately.

The RecordSink class does not support batching.

This method may not be overridden.

Parameters:

context (dict) – Stream partition or context dictionary.

Return type:

None

process_record(record, context)

Load the latest record from the stream.

Legacy pattern (deprecated but supported): Override this method for dict-based record loading with context.

Modern pattern (recommended): Override load_record() instead for simpler, clearer immediate loading.

If you override load_record(), you do NOT need to override this method. The framework will automatically route to your load_record() implementation.

Implementations should permanently serialize each record to the target prior to returning.

If duplicates are merged/skipped instead of being loaded, merges can be tracked via tally_duplicate_merged().

Parameters:

• record (dict) – Individual record in the stream.

• context (dict) – Stream partition or context dictionary.

Raises:

NotImplementedError – If neither process_record nor load_record is overridden.

Return type:

None

final start_batch(context)

Do nothing and return immediately.

The RecordSink class does not support batching.

This method may not be overridden.

Parameters:

context (dict) – Stream partition or context dictionary.

Return type:

None