Outdated JobStatusChanged messages no longer trigger a RemoveJob message but are
logged and ignored. This has the advantage, that an outdated JobStatusChanged message
cannot interfere with a recovered job which can have the same job id.

Updated the Mesos dependency, to unlock some new features (notably the
ability to fetch into sandbox sub-directories).

Shaded the protobuf dependency because the new Mesos library depends on
a newer version than does akka-remoting.

This closes #2827.

This closes #2758.

- Fix injects a union with a null record before the global aggregation.

This closes #2840

This closes #2894

This closes #2884
This closes #2874 (closing PR with Public API breaking changes)

This closes #2823.

This closes #2875

[hotfix] [tests] Harden timeout logic for TaskManager registration in AbstractTaskManagerProcessFailureRecoveryTest

This transitive dependency has an incompatible license.

This closes #2824

Allowing Trigger.onMerge() to return a TriggerResult is not necessary
since an onMerge() call will always be followed by an onElement() call
when adding the element that caused the merging to the merged window.
Having this complicates the internal logic of the WindowOperator and
makes writing Triggers more confusing than it has to be.

Integrates the RollingSink with the new state abstractions so
that its parallelism can change after resuming execution from
a savepoint.

This closes #2845.

This closes #2862.

This closes #2852.

[FLINK-5112] [ExecutionGraph] Remove unused accumulator aggregation code from ArchivedExecutionJobVertex

The ArchivedExecutionJobVertex calculated for its ExecutionVertices the aggregated accumulator
value. However, the result was nowhere stored. This indicates that this code is no longer used
and can be removed.

This closes #2846.

The new Kerberos authentication code in Flink assumed that it's running
against vanilla Hadoop. Original Hadoop's behavior is to skip a secure
login if security is not configured. This is different for other
distributions, e.g. the MapR Hadoop distribution of Hadoop.

Thus, we need to make sure we don't perform any login action if security
is not configured.

This also performs minor code cleanup.

This closes #2864.

Closes #2094

This is a followup to 0d3ff88b

This closes #2856.

This closes #2822.

This changes the ContinuousEventTimeTrigger to behave like the
EventTimeTrigger in the sense that it also triggers at the end of the
window.

This prevents the trigger from not firing at all in case the first
trigger interval is after the window end.

This closes #2860.

[typo] fix toString() of ContinuousEventTimeTrigger

This closes #2854.

This closes #2755.

This closes #2792.

This closes #2853.

- Fixes invalid shipping strategy between consecutive unions.

This closes #2848.

Allows via a setting to deploy a base image on that a task manager
runs.

This closes #2703.

This removes "managed" from the OperatorStateStore and KeyedStateStore
access methods. There is no "un-managed" state and users might be
wondering what "managed" means here.

Set the akka deathwatch interval to 10s, the akka deathwatch pause to 60s and the tcp
connection timeout to 20s per default.

This closes #2831.

[FLINK-2608] Updated test with Java collections.

[FLINK-2608] Updated Chill and Kryo dependencies.

[FLINK-2608] Added collections serialization test.

This closes #2623.

This closes #2851.

If a prior execution attempt cannot be retrieved because it has been evicted before,
the request handler will now throw a meaningful exception to notify the requester
about the evicted execution attempt.

This closes #2837.

This closes #2691.

The CheckpointCoordinator is now given an Executor which is used to execute the state discard
calls asynchronously. This will prevent blocking operations to be executed from within the
calling thread.

Shut down ExecutorServices gracefully

This closes #2825.