Page History

...

• Managed memory is allocated to operators with keyed inputs.

• A general purpose keyed state backend which does not assume inputs sorted by key (e.g. EmbeddedRocksDBStateBackend) is instantiated by Flink runtime.

• Before source operator emits isBacklog=false (during backlog processing), the input to the downstream operator is sorted by the key. Operator can leverage the sorted input to optimize the performanceBefore source operator emits isBacklog=false (during backlog processing), the keyed input to one input operator is sorted by the key. Operator can leverage the sorted input to optimize the performance. Multiple input operator can buffer and sort the inputs internally.
  

• At the point when isBacklog switches to false:
  • Source operator emits RecordAttributes(isBacklog=false).
  • The downstream operator should make sure that flush and process all the buffered inputs and update the keyed state backend contains to the latest value.
• The operator continues to process records using the keyed state backend instantiated by Flink runtime, which now contains the full state obtained from records received while isBacklog=true.

...