I'm trying to write a PyFlink (1.17) pipeline with...
# troubleshooting
d
I'm trying to write a PyFlink (1.17) pipeline with a Kafka source that uses SlidingEventTimeWindows, but I can't get it to properly use event time (windows never seem to trigger). Processing time works fine. I even tried setting a custom TimestampAssigner to assign a timestamp from data in my events...which I know is being called because I put logging in it...but still never triggers on event time. Any ideas on what to try or how to debug this?
r
Hi @Dave Voutila as you have specified that processing time windows are working fine, my next approach would be to get around to Timestamp Assigner and validate whether the event time assignment is being done properly. Also have a look at the watermark configs. Try to run without watermarks first and see if that succeeds.
d
@Dave Voutila Have you seen the watermark advancing normally? This could be seen from the Flink web UI. Usually this problem is caused when there is very little data during test phase and the watermark isn’t advancing normally. If this is the case, you could work around it simply by setting the parallelism of the job to 1.
a
Have you enabled checkpointing? Flink doesn't seem to write from Kafka to Parque on S3 without it.
d
@Dian Fu I haven't tried looking at the web ui. @Ari Huttunen I've tried checkpointing...did not help.