Switch to Mesos 1.5.x protobufs

Merge pull request #17 from tobilg/mesos-1.5

Switch to Mesos 1.5 protobufs

Bugfixes & improvements

The following bugfixes and improvements are contained in this version:

• Task loading improvements (loading just dynamic info)
• stringifyEnums() fixes to avoid modification to the source message and support for enums within arrays.
• Exporting the helpers.js module to avoid code duplication on the boilerplate side

Thanks a lot again @dudyk!

License change

• The license for version 0.5.2 (and all future releases) was changes from MIT to APL 2.0.

Bugfixes & improvements

The following things have been changed:

• Exponential subscription back-off
• Task health check fixes

Major bugfix release

The following things have been addressed in this release:

• More scheduler resubscribe fixes
• Sync will now reconcile all tasks known to the scheduler (to avoid stale task information)
• Task health check now includes the option for a function to check the body content to decide on the status
• Task health check now will delete the timer before setting a new one, to avoid duplicate timers
• Logging changes
• Removed cases in which logging output caused an exception
• Fixes in static ports feature
• Fixes and changes to Mesos HTTP health check support
• Use fake timers for tests

Added functionality and fixed bugs

The following functionality was added:

• Support for static ports definitions
• Added an option to remove serial number from tasks (serial numbers added for ZK persistency)
• Added a task launched event
• Support for task graceful restart (on the "boilerplate" side) by launching a duplicate and killing the task after the duplicate is healthy
• Added task health check support (both in Mesos and in the framework itself with events)

Fixes:

• Fixed issues when reconnecting to master
• Fixed an issue with large event messages not handled properly
• Fixed an issue with old tasks staying in the ZK and loaded.

Thanks @dudyk for contributing the code!

Tests

Thanks again for @dudyk and @jacob-koren-zooz for their contribution of the tests for this project!

ZooKeeper integration

Changes:

• Thanks to @dudyk for contributing the code for the ZooKeeper integration. Task information such as taskId and its state are now persisted to ZooKeeper, which makes it possible for the scheduler to recover its running tasks when it gets restarted
• Some small bugfixes
• Updated documentation

Improvements & bugfixes

The following things have been inplemented:

• Use consistent frameworkID upon scheduler reconnect (#2)
• Enable the scheduler to start without tasks at startup, and be able to add tasks later (#1)

Basic scheduler failover

Added re-registration capabilities in case the long-living Master connection is closed/times out.