JPPF Press Kit

Content: Text of the original release - Where to download - License - Features - Documentation - Contacts

Press release: JPPF 6.2

Tasks dependencies

• tasks within the same job can now depend on each other
• JPPF guarantees that all dependencies of a task are executed before it can start
• traversal of the tasks graph can be performed on either server or client side
• built-in detection and handling of dependency cycles

JPPF in Docker containers and Kubernetes

• the JPPF drivers, nodes and web administration console are now available as Docker images, publicly available on Docker Hub
• JPPF grids can be deployed on Kubernetes clusters using the JPPF Helm chart
• JPPF grids can also be deployed on Docker swarm clusters as Docker service stacks

Performance improvement in the client

A new client-side cache of class definitions was implemented, to improve the overall dynamic class loading performance.

Job dependencies and job graphs

• JPPF jobs can now be organized into dependency graphs, where a job can start only when its dependencies have completed
• new convenient and intuitive APIs to define the dependency graphs
• automatic detection and handling of circular dependencies (i.e. cycles)
• by default, cancellation of a job is automatically cascaded to dependent jobs. This can be overriden for each job
• built-in management MBean to monitor the state of the job dependency graph

DSL for job selectors

The job selector API was extended to include complex boolean expressions and comparison tests, similarly to execution policies

• boolean opearators AND, OR, XOR, Negate
• comparison operators on Comparable values: more than, less than, at least, at most
• range comparison operators on Comparable values: between with inclusion or exclusion of upper and lower bounds
• "contains", "one of" and "regex" selectors
• job name selector
• already existing selectors: all jobs, job uuids, scripted and custom

Use of java.time.* classes for JPPF schedules

java.time.* classes can now be used to build JPPFSchedule instances to specify jobs start or expiration schedules

Inspection of the client-side jobs queue

New methods were added to the client API to inspect the jobs queue

Major refactoring of the management and monitoring MBeans

• the node forwarding MBean now provides compile-time typing of the results
• all built-in MBeans are annotated with metadata available at runtime
• the same metadata is used to generate static forwarding proxies (and their unit tests) for the built-in node MBeans
• the MBeans metadata is also used to generate the built-in MBeans reference documentation
• the management and monitoring documentation was refactored for greater clarity and readability

Administration and monitoring console improvements

• numerous performance hotspots were identified and fixed, both in the monitoring API and in the UI rendering code
• for license compatibility reasons, the JFreeChart charting library was replaced with XCharts, resulting in further performance improvements

Performance improvements in the driver and JMX remote connector

Several performance sinks were identified and fixed, in high stress scenarios where a very large number of job lifecyclle notifications were generated, causing high spikes in CPU and memory usage.