It is likely that they are being dropped because of "windowTime". If
you are working with older data, it is currently recommended to ingest
data via hadoop jobs instead of direct from kafka.
If you don't run hadoop and don't want to set it up, some people have
successfully created setups where they ingest data direct from kafka
topics. This requires setting the "rejectionPolicy" to "messageTime"
and ensuring that your data is being delivered in time order
You should also look into enabling metrics to be logged out, this will
provide some log lines that will tell you if messages are being
ingested or dropped on the floor (events/processed, events/thrownAway,
events/unparseable). This can be done by setting
(http://druid.io/docs/latest/Configuration.html). Also, in order to
avoid lots of log spew, make sure you don't have the other loggers
configured (set druid.monitoring.monitors=).