AsyncCassandra is a shim for using Cassandra as a backend for OpenTSDB instead of the default HBase backend. Because you can configure Cassandra to operate similar to HBase, albeit without the consistency guarantees, swapping out the backend is pretty easy.

Enforcing the TSD schema on Cassandra requires using the Thrift interface and this library uses Netflix's Astyanax library for a higher level API. Also note that because this is a shim and not a low level client, performance may not match that of AsyncHBase. Additionally, because Cassandra doesn't support atomic mutations, locking for atomic increments is implemented by writing a special lock column and checking the timestamps to see who won with retries if acquisition failed. That means UID assignments will be messy.

To use TSDB with Cassandra, your cluster must use the ByteOrderedPartitioner:

partitioner:  org.apache.cassandra.dht.ByteOrderedPartitioner

Maven has been adopted as the building tool for this project. To produce jar file, run

mvn clean package

For OpenTSDB, make sure you're using version 2.3 from GitHub (as of 11/2/2015 the "put" branch) and compile with

sh build-cassandra.sh

In your opentsdb.conf file, set asynccassandra.seeds to the proper value for your Cassandra cluster. E.g. asynccassandra.seeds=127.0.0.1:9160 Also set tsd.storage.hbase.uid_table = tsdbuid and tsd.storage.hbase.data_table = tsdb