% Using Java and Jython with OpenStack % Jim Baker % [email protected]

Run OpenStack on the JVM via Jython

Jython is Python for the Java platform:

Compiles to Java bytecode
Very good concurrency support - no GIL, so computational threads actually work
import java - can readily import and use any Java package
Jython 2.7.0 recently released, corresponds to CPython 2.7
Support for Python C Extension API is under rapid development

OpenStack on the JVM:

Feasible now for components like Keystone that don't use Eventlet
Can readily work with transition away from Eventlet
Or port Eventlet to Jython

Java components for OpenStack

JVM ecosystem is extensive, performant, and scales:

Apache Cassandra, Kafka, Spark, Storm, ZooKeeper, ...
Repose - high performance servlet filters, can frontend OpenStack

But also different devops challenges:

JVM-based services have different best practices - setup, tuning, logging, monitoring
Python APIs generally lag official Java APIs

Mixing it up: Python and Java

Best of both?

PySpark - use Py4J to embed CPython via JNI
Storm bolts/spouts, written in Python - Pyleus or Parse.ly or Jython
Keystone backend against Spring Data, using Jython
Repose Python - build mixed pipeline of Java servlet filters and any (PEP 3333 compliant) WSGI middleware using Falcon, Pecan, WebOb, Werkzeug, ...

Advantages:

Keep the flexibility of Python development
While maintaining good performance
Can also eliminate IPC, pickling/serialization costs