[BEAM-10984] Dataflow's UnboundedReaderIterator should support overriding default bundle size targets

• Type: Improvement
• Status: Open
• Priority: P3
• Resolution: Unresolved
• Component/s: runner-dataflow
• Assignee: Unassigned
• Reporter: Daniel Mills
• Created: 29/Sep/20 17:20
• Updated: 04/Jun/22 18:16

Description

Currently, the UnboundedReaderIterator will read until 10k elements have been read or 10s have passed.  This works for most pipelines, but is insufficient for pipelines that either require very high throughput or require low latency.  We should make these values controllable via a PipelineOption, probably under DataflowPipelineDebugOptions.

The constants are defined here: https://github.com/apache/beam/blob/master/runners/google-cloud-dataflow-java/worker/src/main/java/org/apache/beam/runners/dataflow/worker/WorkerCustomSources.java#L757