Sander Ploegsma created BEAM-5307:
-------------------------------------
Summary: Allow injection of state in DoFn.FinishBundle
Key: BEAM-5307
URL: https://issues.apache.org/jira/browse/BEAM-5307
Project: Beam
Issue Type: Improvement
Components: sdk-java-core
Reporter: Sander Ploegsma
Assignee: Kenneth Knowles
Example use case: a stateful {{DoFn}} that requires persisting its state to an
external database. Instead of writing to the external database for each
element, it is much more efficient to flush the state every once in a while, or
when cleaning up.
Currently, this is not possible because the {{@StateId}} injection is only
available in processing functions and timers.
[This|https://stackoverflow.com/questions/51789776/calculating-deltas-in-apache-beam-using-stateful-processing]
might be a workaround, but I'm not even sure if it works.
Instead, by allowing the use of {{@StateId}} inside a {{@FinishBundle}}
function, we can make sure the internal state is persisted in all scenarios:
{code:java}
@FinishBundle
public void flush(FinishBundleContext context, @StateId("myState")
ValueState<StateObj> state) {
repository.save(state.read());
}
{code}
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
