Workers

Workers provide the compute capacity that runs e6 Ingestion Engine pipelines. The controller assigns pipeline work to available workers, while each worker processes data through sources, SQL transformations, stateful operators, and sinks.

How work is distributed

A pipeline is divided into parallel tasks. These tasks are distributed across workers according to the pipeline's configured parallelism and the capacity available in the cluster.

Increasing parallelism can improve throughput when the source and sink support parallel processing. Worker CPU and memory should be sized for the number and type of tasks they run.

Processing data

Workers perform the data-plane operations of a pipeline:

• Read records from configured sources
• Apply SQL transformations
• Maintain state for joins, windows, and aggregations
• Exchange data between parallel tasks when required
• Write results to configured sinks
• Participate in checkpoints and recovery

The controller manages task placement and pipeline lifecycle; users do not need to assign individual tasks to workers.

Scaling

In Kubernetes deployments, e6 Ingestion Engine creates and manages worker pods according to pipeline demand and cluster configuration.

Two settings affect capacity:

• Pipeline parallelism controls how many tasks process the pipeline concurrently.
• Worker resources control the CPU and memory available to those tasks.

Some connectors impose additional scaling constraints. Consult the connector documentation before increasing parallelism.

Failure recovery

If a worker becomes unavailable, e6 Ingestion Engine reschedules its work on available capacity. Stateful tasks restore from the latest successful checkpoint, while supported sources resume from their recorded positions.

Recovery time depends on checkpoint size, available cluster capacity, and source or sink connectivity.

Networking and security

Workers communicate with the controller, other workers, and configured external systems. Worker communication should remain on private cluster networks.

Access to sources, sinks, and object storage should use workload identity or secret references rather than credentials embedded in pipeline configuration.

Monitoring workers

Use the worker metrics and logs to monitor:

• Records and bytes processed
• Source lag
• Sink throughput and failures
• Checkpoint progress
• Backpressure and queue utilization
• Task failures and restarts

See Metrics for the available monitoring signals.