Rejected records

Learn how RDI stores records that cannot be processed.

Redis Data Integration (RDI) sends records that it cannot process to a dead letter queue (DLQ). In the Redis Cloud UI, these records are called rejected records.

Rejected records help you understand which source tables are affected and why specific records could not continue through the pipeline. They are intended for troubleshooting and support, not for normal pipeline operation.

When records are rejected

RDI can reject a record when it cannot safely transform or write the change event. Common causes include:

• The incoming change event is malformed or missing required metadata.
• A transformation job fails while processing the record.
• A target write fails, for example because the target key already has an incompatible data type.

By default, RDI stores rejected records instead of silently dropping them. You can change this behavior with the processors.error_handling setting. See the pipeline configuration file for more information.

How RDI stores rejected records

RDI stores rejected records in the RDI database as capped Redis streams. Each DLQ stream corresponds to a source table and tracks the records rejected for that table.

DLQ stream names use the dlq: prefix followed by the source data stream name. In current RDI versions, the stream name is typically:

dlq:data:{rdi}:<schema_or_database>.<table>

For example, rejected records for the public.users table are stored in:

dlq:data:{rdi}:public.users

For sources that include the source name in the stream qualifier, the final part can contain three components:

dlq:data:{rdi}:<source>.<schema_or_database>.<table>

Some RDI versions or configurations can use a hash-tagged variant such as dlq:{data:rdi:<schema_or_database>.<table>}. To find all DLQ streams in the RDI database, scan for stream keys that start with dlq:.

The maximum number of records stored per DLQ stream is controlled by processors.dlq_max_messages. When the stream reaches the configured limit, older entries are evicted as newer entries are added.

What to inspect

Start with the rejected count for each affected table, then inspect a sample record from the table with the highest count or the table that matters most to your application.

Useful fields include:

• The affected table.
• The rejection time.
• The rejected operation. RDI stores this as an opcode value such as c, u, d, r, t, or m. See Using the operation code for the operation labels.
• The rejection reason.
• The transformation job or operation, when the failure happened during transformation.

Resolve rejected records

Use the rejection reason to identify the likely fix:

• If a transformation job failed, update the job configuration and deploy the pipeline change.
• If target writes failed because of incompatible existing keys, update the target data or key mapping.
• If records are malformed, inspect the source connector and source database change data capture configuration.

RDI does not automatically replay records from the DLQ after you fix the cause. If you need existing source data to be processed again, reset the pipeline after applying the fix. See Reset data pipeline for Redis Cloud, or use the appropriate self-managed RDI reset workflow.

CLI and API access

For self-managed RDI, use the redis-di list-dlqs command to see the dead-letter queues and the redis-di list-dlq-records command (also available as redis-di get-rejected) to inspect the rejected records of a queue.

For Redis Cloud RDI, connect to the RDI database and inspect the corresponding DLQ streams directly when you need details that are not shown in the Redis Cloud UI.

RDI API v2 also includes DLQ inspection endpoints. See the API reference for endpoint details.