Inconsistent results in pull queries with distributed KsqlDB setup about ksql HOT 1 OPEN

Some addition.

I tried to create a table with only one partition:

CREATE TABLE EVENTS_HOURLY_COUNTS WITH (PARTITIONS=1)
AS SELECT EVENTS.ROW_PARTITION AS PARTITION,
       COUNT(*)
FROM EVENTS
WINDOW TUMBLING ( SIZE 1 HOURS )
GROUP EVENTS.BY ROW_PARTITION
EMIT CHANGES;

So there is only one partition but it still collects the keys from 0 to 59. And it's the same behavior. When I run pull query for 20 hours I expect to receive 1200 rows in results. Most times it is 1200 rows but from time to time it could be 1199, 936 or even 1201 rows!

from ksql.