This closes #3132

This uses Flakka (a custom Akka 2.3 build) to resolve the issue that
the bind address needs to be matching the external address of the
JobManager. With the changes applied, we can now bind to all
interfaces, e.g. via 0.0.0.0 (IPv4) or :: (IPv6).

For this to work properly, the configuration entry
JOB_MANAGER_IPC_ADDRESS now represents the external address of the
JobManager. Consequently, it should not be resolved to an IP address
anymore because it may not be resolvable from within containered
environments. Akka treats this address as the logical address. Any
messages which are not tagged with this address will be received by
the Actor System (because we listen on all interfaces) but will be
dropped subsequently. In addition, we need the external address for
the JobManager to be able to publish it to Zookeeper for HA setups.

Flakka: https://github.com/mxm/flakka
Patch applied: https://github.com/akka/akka/pull/15610

- convert host to lower case
- use consistent format for IPv6 address
- adapt config and test cases
- adapt documentation to clarify the address config entry
- TaskManager: resolve the initial hostname of the StandaloneLeaderRetrievalService

This closes #2917.

This closes #2850

This closes #2324

- Add YarnFlinkResourceManager test to reaccept task manager registrations from a re-elected job manager

- Remove unnecessary sync logic between JobManager and ResourceManager

- Avoid duplicate reigstration attempts in case of a refused registration

- Add test case to check that not an excessive amount of RegisterTaskManager messages are sent

- Remove containersLaunched from YarnFlinkResourceManager and instead not clearing registeredWorkers when JobManager loses leadership

- Let YarnFlinkResourceManagerTest extend TestLogger

- Harden YarnFlinkResourceManager.getContainersFromPreviousAttempts

- Add FatalErrorOccurred message handler to FlinkResourceManager;
  Increase timeout for YarnFlinkResourceManagerTest;
  Add additional constructor to TestingYarnFlinkResourceManager for tests

- Rename registeredWorkers field into startedWorkers
Additionally, the RegisterResource message is renamed into NotifyResourceStarted which
tells the RM that a resource has been started. This reflects the current semantics of
the startedWorkers map in the resource manager.

- Fix concurrency issues in TestingLeaderRetrievalService

This closes #2257

This closes #1915

Consolidation is done via plugin management.

Closes #1753

- add suffixes
- adapt change-scala-version script

This closes #1529.

0.10-SNAPSHOT continues on branch release-0.10

Tests would fail because of version conflicts because the tests include
the original curator dependencies even though we shade them away in the
final build result.

This also fixes dependency management entries for curator dependencies.
We shade it, therefore it cannot be in the dependency management
section in the root pom.

Closes #851

This closes #689

This closes #454

- users can now also deploy Flink on YARN for executing a single job.
- The flink-yarn project has been moved out of the flink-addons module
- the MiniYARNCluster is used for testing Flink on YARN
- There is now a (undocumented) Java interface Flink's YARN client, allowing users to manually control the Yarn session.
- ALL ports used by Flink when running on YARN are automatically determined. In the past users reported problems with blocked ports (YARN is telling the client the RPC address of the application master)
- The checks before deployment have been improved to give better error messages if the user is requesting too many resources for a YARN session

to HadoopFileSystem wrapper, which supports all subclasses of org.apache.hadoop.fs.FileSystem.
This allows us to let users use all file systems with support for HDFS.
The change has been tested with Tachyon, Google Cloud Storage Hadoop Adapter and HDFS.

The change also cleans up the Hadoop dependency exclusions.

Removed Disclaimer file
Eliminated unnecessary incubating substrings
Bumped version to 0.9-SNAPSHOT

Replacing the EventCollector and the MemoryArchivist.

Finished EventCollector and adjusted ExecutionGraph and ExecutionVertex to register actors as listeners.

This closes #232

- Clean dependencies in flink-streaming connectors

We'll still ship the Logback configuration file and pass them their path to the JVM so that
users can use Flink with Logback as well.

This closes #125

I excluded all slf4j-log4j12 transitive dependencies to get rid off the warning that slf4j finds multiple slf4j bindings in the classpath. Additionally, I added log4j-over-slf4j and excluded all log4j transitive dependencies so that slf4j handles also legacy log4j logging messages from dependencies.

Author: Till Rohrmann <trohrmann@apache.org>

Closes #121 from tillrohrmann/FLINK-1097 and squashes the following commits:

b9f76d5 [Till Rohrmann] Excluded slf4j-log4j12 as transitive dependency in order to not have multiple slf4j bindings in the class path. Added log4j-overslf4j and excluded log4j transitive dependencies to collect log4j outputs.

- update _config.yaml of integrated documentation
- quick check on the quickstarts / build instructions to match the latest release
- quickstart.sh scrips now point to the correct versions.

Fix #584
Fix #568