Apache Samza 1.5 [Docs]
IMPORTANT NOTE: As noted in the last release, this release contains backward incompatible changes regarding samza job submission. Details can be found on SEP-23: Simplify Job Runner
We are thrilled to announce the release of Apache Samza 1.5.0.
Today, Samza forms the backbone of hundreds of real-time production applications across a multitude of companies, such as LinkedIn, Slack, and Redfin, among many others. Samza provides leading support for large-scale stateful stream processing with:
First class support for local states (with RocksDB store). This allows a stateful application to scale up to 1.1 Million events/sec on a single machine with SSD.
Support for incremental checkpointing of state instead of full snapshots. This enables Samza to scale to applications with very large states.
A fully asynchronous programming model that makes parallelizing remote calls efficient and effortless.
High level API for expressing complex stream processing pipelines in a few lines of code.
Beam Samza Runner that marries Beam’s best in class support for EventTime based windowed processing and sophisticated triggering with Samza’s stable and scalable stateful processing model.
A fully pluggable model for input sources (e.g. Kafka, Kinesis, DynamoDB streams etc.) and output systems (HDFS, Kafka, ElastiCache etc.).
A Table API that provides a common abstraction for accessing remote or local databases and allows developers to “join” an input event stream with such a Table.
Flexible deployment model for running the applications in any hosting environment and with cluster managers other than YARN.
New Features, Upgrades and Bug Fixes:
This release brings the following features, upgrades, and capabilities (highlights):
Samza Container Placement
Container Placements API gives you the ability to move / restart one or more containers (either active or standby) of your cluster based applications from one host to another without restarting your application. You can use these api to build maintenance, balancing & remediation tools.
Simplify Job Runner & Configs
Job Runner will now simply submit Samza job to Yarn RM without executing any user code and job planning will happen on ClusterBasedJobCoordinator instead. This simplified workflow addresses security requirements where job submissions need to be isolated in order to execute user code as well as operational pain points where deployment failure could happen at multiple places.
Full list of the jiras addressed in this release can be found here.
Upgrading your application to Apache Samza 1.5.0
ConfigFactory is deprecated as Job Runner does not load full job config anymore. Instead, ConfigLoaderFactory is introduced to be executed on ClusterBasedJobCoordinator to fetch full job config. If you are using the default PropertiesConfigFactory, simply switching to use the default PropertiesConfigLoaderFactory will work, otherwise if you are using a custom ConfigFactory, kindly creates its new counterpart following ConfigLoaderFactory.
Configs related to job submission must be explicitly provided to Job Runner as it is no longer loading full job config anymore. These configs include
- Configs directly related to job submission, such as yarn.package.path, job.name etc.
- Configs needed by the config loader on AM to fetch job config, such as path to the property file in the tarball, all of such configs will have a job.config.loader.properties prefix.
- Configs that users would like to override
Full list of the job submission configurations can be found here
Alternative way when submitting job,
can be simplified to
where submission.properties contains
In case of a problem in Samza 1.5, users can rollback to Samza 1.4 and keep the old start up flow using config-path & config-factory.
Simplify Job Runner
SAMZA-2488 Add JobCoordinatorLaunchUtil to handle common logic when launching job coordinator
SAMZA-2471 Simplify CommandLine
SAMZA-2458 Update ProcessJobFactory and ThreadJobFactory to load full job config
SAMZA-2453 Update ClusterBasedJobCoordinator to support Beam jobs
SAMZA-2441 Update ApplicationRunnerMain#ApplicationRunnerCommandLine not to load local file
SAMZA-2420 Update CommandLine to use config loader for local config file
Container Placement API
SAMZA-2402 Tie Container placement service and Container placement handler and validate placement requests
SAMZA-2379 Support Container Placements for job running in degraded state
SAMZA-2378 Container Placements support for Standby containers enabled jobs
SAMZA-2515 Thread safety for Kafka consumer in KafkaConsumerProxy
SAMZA-2511 Handle container-stop-fail in case of standby container failover
SAMZA-2510 Incorrect shutdown status due to race between runloop thread and process callback thread
SAMZA-2506 Inconsistent end of stream semantics in SystemStreamPartitionMetadata
SAMZA-2464 Container shuts down when task fails to remove old state checkpoint dirs
SAMZA-2468 Standby container needs to respond to shutdown request
SAMZA-2519 Support duplicate timer registration
SAMZA-2508 Use cytodynamics classloader to launch job container
SAMZA-2478 Add new metrics to track key and value size of records written to RocksDb
SAMZA-2462 Adding metric for container thread pool size