Releases: druid-io/tranquility
Releases · druid-io/tranquility
v0.8.2
v0.8.1
v0.8.0
Major changes this release include support for Druid 0.9.0 input formats and extensions, performance improvements due to timestamp caching, and an adjustment to the Beam trait that enables more fine grained error reporting.
Upgrading from earlier releases:
- The Beam trait changed in #140. The change is source-compatible for users of Beams and BeamFactories (all existing methods still work), but requires adjustments for implementers of custom Beams (there is a new
sendAll
method to implement). See http://static.druid.io/tranquility/api/v0.8.0/#com.metamx.tranquility.beam.Beam for the updated trait.
New features:
- Support for flattenSpec. (#113)
- Support for formats other than JSON. (#119)
- Support all Druid formats by using InputRows internally. (#140)
- Allow setting lingerMillis to -1 to always wait for complete batches. (#147)
Improvements:
- DruidBeamMaker: Use availabilityGroup no longer tied to baseFirehoseId. (#152)
- Update to druid-0.9.0. (#155)
- Remove dimension sorting workaround. (#156)
- Add some additional hourly cycling to minutely cycleBuckets. (#171)
Bug fixes:
- Respect druid.discovery.curator.path in fromConfig. (#151)
- Spark: Share Tranquilizers using registry in BeamRDD. (#168)
Other:
v0.7.4
New features:
- Flink: Add Flink accumulators (#124)
- Optionally locate Druid tasks through "runningTasks" endpoint on the overlord. (#126)
- Samza: Include property for setting lingerMillis. (#127)
- Support full tuningConfig by handling it in a more loosely typed way. (#129)
- Server: Asynchronous API option. (#130)
Improvements:
- Disable Finagle fail-fast behavior by default. (#123)
- Cache timestamps when using DruidBeams.fromConfig. (#128)
- Flink: Update to Flink 1.0.0. (#135)
- DruidTaskResolver: Demote polling log message to DEBUG. (#139)
- Additional logging for non-retryable responses. (#141)
- Server: Change asynchronous route to an ?async option. (#146)
Bug fixes:
- Server: Raise default http.threads. (#132)
- Fix 'Loss of Druid redundancy' alert 'status' field. (#133)
- Fix objectWriter, automatic partitioning for MessageHolder. (#137)
- Kafka: Avoid List.sort(Comparator), it doesn't exist in java 7. (#144)
Other:
v0.7.3
New features:
- Flink module (#92)
- Add buildV9Directly to tuning object, defaults to false (#99)
- Tranquilizer.flush() to make batch / micro-batch oriented workflows easier (#110)
- Standardize on configuration files and use of Tranquilizer (#112)
- Console sender (#117)
Improvements:
- Adjust spark/flink examples to use event-time-based Timestamper (#100)
- Changes to make configuration files easier to use (#101)
Bug fixes:
- Fix transient KafkaConsumerTest failures (#105)
- Fix DruidBeam exception at TRACE logging level (#106)
- Kafka: Switch to kafka 0.8.2.2 for compatibility (#109)
- IndexService: close should return a future, and should not attempt to close a client that has not been created yet (#111)
- Spark: Recommend using singleton Beam (#115)
- Target Java 7 (#118)
- Fix javadoc generation (#120)
v0.7.2
v0.7.1
v0.7.0
New features:
- Tranquility Server (#57, #79)
- Tranquilizer and Simple Tranquilizer APIs (#57, #77)
- Add Beam.sendBatch (#75)
Improvements:
- Changed default overlord service name (#63)
- Update jackson, finagle, twitter util, netty (#66)
- Avoid materializing Spark RDDs when sending (#68)
- Update to Druid 0.8.2 (#73)
- Update source headers (#74)
- Update to Spark 1.6.0 (#82)
Bug fixes:
Other notes:
This release contains a backwards incompatible change for those of you that have extended the Beam trait. The "propagate" method is now final, and instead you should implement the "sendBatch" method. See #75 for more information.