Github user rtudoran commented on the issue:
https://github.com/apache/flink/pull/3590
@fhueske @sunjincheng121
I have done another commit (actually i did it last evening but the network
of the machine was down and did not push it). What i did was:
1)address the formatting issues (i hope i did not miss (too) many)
2) added some new tests including for processingtime. I did not find the
reference to what @fhueske mentioned but i created a custom source that emits
events 1 second apart. This gives us a good framework about events arriving
with processing time 1 second apart, so we can run any tests and validate. I
think this is quite relevant
3) I did update the processing functions. However, i am still using the
ValueState[Queue[JTuple2[Long,Row]]]= _ for buffering the events. As per my
previosu mentioning - i strongly believe this is a better approach. Having the
events sorted by their order is an advantage that we should not loose.
Otherwise we will pay a high price later (sorting, limit, top, distinct...). We
pay some price for serialziation, but as we do this over a full object it
should be relatively ok rather than serializing independent objects (like in
hash map...IMHO)
4) I kept the partitioned/non partitioned cases separately ... as per
initial argument of Fabian (i think it is worth paying the price of having an
extra class) not to add extra operators that need to be maintained, dedicate
resource, monitored for liveness and redeploy in case of failures.
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at [email protected] or file a JIRA ticket
with INFRA.
---
