Using Druid for low latency timeseries charts

Hi all,

I have the task to provide timeseries data for a dashboard which comes in the form (timestamp, source ID, value … plus some meta information like value type and additional timestamps). Data is published to a Kafka cluster that I can consume from. The dashboard should display timeseries charts e.g. for the last hour for a certain source ID. Also I want to mention that data points are not arriving with increasing timestamps and that there up to a couple hundred source IDs.

Now my first question is, when configuring a Kafka Listener to stream insert to Druid, will the values be a dimension or a metric? I do not necessarily do aggregations.

My second question is, after registering such a druid datasource as Hive external table, is there a proper way to “pivot” the data such that a JDBC Frontend like Tableau could access the data from a wide table where source IDs are in the columns?

My third question is, has anybody done a dashboard on this kind of data with superset? So far I have only managed to create timeseries charts in superset e.g. for a “count” metric, but from a datasource as above, filtered by dimension, and then displaying the “value”.

Happy to discuss and appreciating any comment!

Best

J