Trigger Firing for Late Window Elements

Posted by Scott Kidder on Oct 19, 2018; 5:14pm
URL: http://deprecated-apache-flink-user-mailing-list-archive.369.s1.nabble.com/Trigger-Firing-for-Late-Window-Elements-tp23971.html

I'm using event-time windows of 1 hour that have an allowed lateness of several hours. This supports the processing of access logs that can be delayed by several hours. The windows aggregate data over the 1 hour period and write to a database sink. Pretty straightforward.

Will the event-time trigger lead to the window trigger firing for every single late element? Suppose thousands of late elements arrive simultaneously; I'd like to avoid having that lead to thousands of database updates in a short period of time. Ideally, I could batch up the late window changes and have it trigger when the window is finally closed or some processing-time duration passes (e.g. once per minute).

For reference, here's what the aggregate window definition looks like with Flink 1.5.3:

        chunkSource.keyBy(record -> record.getRecord().getEnvironmentId())

                .timeWindow(Time.hours(1))

                .allowedLateness(Time.hours(3))

                .aggregate(new EnvironmentAggregateWatchTimeFunction())

                .uid("env-watchtime-stats")

                .name("Env Watch-Time Stats")

                .addSink(new EnvironmentWatchTimeDBSink());

Thank you,

--

Scott Kidder