# Bidirectional Replication

K2K can be configured for bidirectional replication, enabling data to flow between two Kafka clusters in both directions while preventing infinite replication loops. When enabled, K2K uses header-based filtering to identify and exclude records that have already been replicated from the opposite direction.

### Deployment

Bidirectional replication requires **two separate K2K instances** to be deployed:

1. **Instance 1**: Replicates from Cluster A → Cluster B
2. **Instance 2**: Replicates from Cluster B → Cluster A

Both instances must:

* Use the **same pipeline name** (if `pipeline.mode` is set to `pipeline`)
* Use the **same bidirectional configuration** (same `mode` and `headers.bidirectional` value)

### Example

The following example shows the required properties set across the features and bidirectional configuration blocks to enable bidirectional replication. 

This configuration must be used by both K2K instances:

```yaml
# ---------------------------------------------------
# K2K Configuration for Bidirectional Replication
# ---------------------------------------------------

# 1. Global feature flag to enable bidirectional replication.
features:
  bidirectionalReplication: enabled

# 2. Bidirectional configuration block (optional).
bidirectional:
  # Mode determines the identifier used in the bidirectional header.
  # "pipeline" uses the pipeline name; "cluster" uses the source cluster ID.
  mode: cluster
  
  headers:
    # Name of the header used to mark bidirectional records.
    # This header is added to replicated records and checked when filtering.
    bidirectional: "__k2k_bidirectional"

# 3. Pipeline name.
# Must be the same for both instances when bidirectional.mode is set to pipeline
name: bidirectional-pipeline
```

### Bidirectional Modes

The bidirectional mode The choice between `pipeline` and `cluster` mode is only relevant when multiple pipelines target the same Kafka cluster pairs. 

The mode determines the scope of filtering: 

* in `pipeline` mode, records are only filtered if they were replicated by the same pipeline (identified by pipeline name), allowing different pipelines between the same clusters to operate independently. 
* in `cluster` mode, records are filtered if they originated from the same cluster (identified by cluster ID), regardless of which pipeline replicated them.