ClickHouse® + Spark

jdbc

The trivial & natural way to talk to ClickHouse from Spark is using jdbc. There are 2 jdbc drivers:

ClickHouse-Native-JDBC has some hints about integration with Spark even in the main README file.

‘Official’ driver does support some conversion of complex data types (Roaring bitmaps) for Spark-ClickHouse integration: https://github.com/ClickHouse/clickhouse-jdbc/pull/596

But proper partitioning of the data (to spark partitions) may be tricky with jdbc.

Some example snippets:

Connectors

via Kafka

ClickHouse can produce / consume data from/to Kafka to exchange data with Spark.

via hdfs

You can load data into hadoop/hdfs using sequence of statements like INSERT INTO FUNCTION hdfs(...) SELECT ... FROM clickhouse_table later process the data from hdfs by spark and do the same in reverse direction.

via s3

Similar to above but using s3.

via shell calls

You can call other commands from Spark. Those commands can be clickhouse-client and/or clickhouse-local.

do you really need Spark? :)

In many cases you can do everything inside ClickHouse without Spark help :) Arrays, Higher-order functions, machine learning, integration with lot of different things including the possibility to run some external code using executable dictionaries or UDF.

More info + some unordered links (mostly in Chinese / Russian)

Last modified 2025.01.16: Streamlined page metadata, simplified directory structure (afe0f3c)