How would you handle errors when data is corrupted while working on streaming data?

In a streaming pipeline, corrupted records can be handled by schema validation and a dead‑letter queue (DLQ). When a record fails validation, it is sent to the DLQ for later inspection instead of blocking the stream. The pipeline should be idempotent so that retries do not duplicate work. Use a watermark or checkpointing mechanism to track progress and avoid reprocessing corrupted data. Implement monitoring and alerting on DLQ size and error rates. For transient corruption, retry with exponential backoff. If corruption is due to schema evolution, maintain backward‑compatible schemas or use a schema registry. Logging detailed error context (timestamp, key, payload) helps root‑cause analysis.