Backpressure Watermark Stall
criticalOperators experiencing backpressure prevent watermarks from advancing, causing time-based operations (windows, temporal joins) to produce no results despite incoming data.
When flink_operator_currentinputwatermark or flink_task_currentinputwatermark stops advancing while flink_operator_recordsin continues to increase, check the JobGraph backpressure tab. High backpressure combined with stalled watermarks indicates downstream operators cannot keep pace. This manifests as queries producing zero results despite active data ingestion.
Identify the bottleneck operator using the Flink UI backpressure tab. For sink backpressure, verify sink permissions and throughput capacity. For processing backpressure, check for data skew (uneven key distribution) using per-subtask metrics. Consider increasing parallelism for the bottleneck operator or setting an idle-timeout to allow watermarks to advance despite slow partitions.