Skip to main content

Release notes

Apache Druid 34.0.0 contains over 270 new features, bug fixes, performance enhancements, documentation improvements, and additional test coverage from 48 contributors.

See the complete set of changes for additional details, including bug fixes.

Review the upgrade notes and incompatible changes before you upgrade to Druid 34.0.0. If you are upgrading across multiple versions, see the Upgrade notes page, which lists upgrade notes for the most recent Druid versions.

Important features, changes, and deprecations​

This section contains important information about new and existing features.

Java 11 support​

Java 11 support has been deprecated since Druid 32.0, and official support will be removed as early as Druid 35.0.0

Hadoop-based ingestion​

Hadoop-based ingestion has been deprecated since Druid 32.0 and will be removed as early as Druid 35.0.0. We recommend one of Druid's other supported ingestion methods, such as SQL-based ingestion or MiddleManager-less ingestion using Kubernetes.

As part of this change, you must now opt-in to using the deprecated index_hadoop task type. If you don't do this, your Hadoop-based ingestion tasks will fail.

To opt-in, set druid.indexer.task.allowHadoopTaskExecution to true in your common.runtime.properties file.

#18239

Use SET statements for query context parameters​

You can now use SET statements to define query context parameters for a query through the Druid console or the API.

#17894 #17974

SET statements in the Druid console​

The web console now supports using SET statements to specify query context parameters. For example, if you include SET timeout = 20000; in your query, the timeout query context parameter is set:

SET timeout = 20000;
SELECT "channel", "page", sum("added") from "wikipedia" GROUP BY 1, 2

#17966

SET statements with the API​

SQL queries issued to /druid/v2/sql can now include multiple SET statements to build up context for the final statement. For example, the following SQL query results includes the timeout, useCache, populateCache, vectorize, and engine query context parameters:

SET timeout = 20000;
SET useCache = false;
SET populateCache = false;
SET vectorize = 'force';
SET engine = 'msq-dart'
SELECT "channel", "page", sum("added") from "wikipedia" GROUP BY 1, 2

The API call for this query looks like the following:

curl --location 'http://HOST:PORT/druid/v2/sql' \
--header 'Content-Type: application/json' \
--data '{
  "query": "SET timeout=20000; SET useCache=false; SET populateCache=false; SET engine='\''msq-dart'\'';SELECT  user,  commentLength,COUNT(*) AS \"COUNT\" FROM wikipedia GROUP BY 1, 2 ORDER BY 2 DESC",
  "resultFormat": "array",
  "header": true,
  "typesHeader": true,
  "sqlTypesHeader": true
}'

This improvement also works for INSERT and REPLACE queries using the MSQ task engine. Note that JDBC isn't supported.

Improved HTTP endpoints​

You can now use raw SQL in the HTTP body for /druid/v2/sql endpoints. You can set Content-Type to text/plain instead of application/json, so you can provide raw text that isn't escaped.

#17937

Cloning Historicals (experimental)​

You can now configure clones for Historicals using the dynamic Coordinator configuration cloneServers. Cloned Historicals are useful for situations such as rolling updates where you want to launch a new Historical as a replacement for an existing one.

Set the config to a map from the target Historical server to the source Historical:

  "cloneServers": {"historicalClone":"historicalOriginal"}

The clone doesn't participate in regular segment assignment or balancing. Instead, the Coordinator mirrors any segment assignment made to the original Historical onto the clone, so that the clone becomes an exact copy of the source. Segments on the clone Historical do not count towards replica counts either. If the original Historical disappears, the clone remains in the last known state of the source server until removed from the cloneServers config.

When you query your data using the native query engine, you can prefer (preferClones), exclude (excludeClones), or include (includeClones) clones by setting the query context parameter cloneQueryMode. By default, clones are excluded.

As part of this change, new Coordinator APIs are available. For more information, see Coordinator APIs for clones.

#17863 #17899 #17956

Embedded kill tasks on the Overlord (Experimental)​

You can now run kill tasks directly on the Overlord itself. Embedded kill tasks provide several benefits; they:

  • Kill segments as soon as they're eligible
  • Don't take up tasks slot
  • finish faster since they use optimized metadata queries and don't launch a new JVM
  • Kill a small number of segments per task, ensuring locks on an interval aren't held for too long
  • Skip locked intervals to avoid head-of-line blocking
  • Require minimal configuration
  • Can keep up with a large number of unused segments in the cluster

This feature is controlled by the following configs:

  • druid.manager.segments.killUnused.enabled - Whether the feature is enabled or not (Defaults to false)
  • druid.manager.segments.killUnused.bufferPeriod - The amount of time that a segment must be unused before it is able to be permanently removed from metadata and deep storage. This can serve as a buffer period to prevent data loss if data ends up being needed after being marked unused (Defaults to P30D)

To use embedded kill tasks, you need to have segment metadata cache enabled.

As part of this feature, new metrics have been added.

#18028 #18124

Preferred tier selection​

You can now configure the Broker service to prefer Historicals on a specific tier. This is useful for across availability zone deployment. Brokers in one AZ select historicals in the same AZ by default but still keeps the ability to select historical nodes in another AZ if historicals in the same AZ are not available.

To enable, set property druid.broker.select.tier to perferred in Broker runtime properties. You can then configure druid.broker.select.tier.preferred.tier to the tier you want each broker to prefer (i.e. for brokers in AZ1, you could set this to the tier name of your AZ1 historical servers).

#18136

Dart improvements​

The Dart query engine now uses the /druid/v2/sql endpoint like other SQL query engines. The former Dart specific endpoint is no longer supported. To use Dart for a query, include the engine query context parameter and set it to msq-dart.

#18003 #18003

Enabling Dart remains the same, add the following line to your broker/runtime.properties and historical/runtime.properties files:

druid.msq.dart.enabled = true

Additionally, Dart now queries real-time tasks by default. You can control this behavior by setting the query context parameter includeSegmentSource to REALTIME (default) or NONE, in a similar way to MSQ tasks. You can also run synchronous or asynchronous queries.

#18076 #18241

SegmentMetadataCache on the Coordinator​

#17996 #17935

This section contains detailed release notes separated by areas.

Web console​

  • You can now assign tiered replications to tiers that aren't currently online #18050
  • You can now filter tasks by the error in the Task view #18057
  • Improved SQL autocomplete and added JSON autocomplete #18126
  • Changed how the web console determines what functions are available, improving things like auto-completion #18214
  • Updated the web console to use the Overlord APIs instead of Coordinator APIs when managing segments, such as marking them as unused #18172

Ingestion​

  • Improved concurrency for batch and streaming ingestion tasks #17828
  • Removed the useMaxMemoryEstimates config. When set to false, Druid used a much more accurate memory estimate that was introduced in Druid 0.23.0. That more accurate method is the only available method now. The config has defaulted to false for several releases #17936

Streaming ingestion​

Multi-stream supervisors (experimental)​

You can now use more than one supervisor to ingest data into the same datasource. Use the id field to distinguish between supervisors ingesting into the same datasource (identified by spec.dataSchema.dataSource for streaming supervisors).

When using this feature, make sure you set useConcurrentLocks to true for the context field in the supervisor spec.

#18149 #18082

Supervisors and the underlying input stream​

Seekable stream supervisors (Kafka, Kinesis, and Rabbit) can no longer be updated to ingest from a different input stream (such as a topic for Kafka). Since such a change is not fully supported by the underlying system, a request to make such a change will result in a 400 error.

#17955 #17975

Other streaming ingestion improvements​
  • Improved streaming ingestion so that it automatically determine the maximum number of columns to merge #17917

Querying​

Metadata query for segments​

You can use a segment metadata query to find the list of projections attached to a segment.

#18119 #18223

json_merge() improvement​

json_merge() is now SQL-compliant when arguments are null. The function now returns null if any argument is null. For example, queries like SELECT JSON_MERGE(null, null) and SELECT JSON_MERGE(null, '') will return null instead of throwing an error.

#17983

Other querying improvements​

  • You can now perform big decimal aggregations using the MSQ task engine #18164
  • Changed MV_OVERLAP and MV_CONTAINS functions now aligns more closely with the native inType filter #18084
  • Improved query handling when segments are temporarily missing on Historicals but not detected by Brokers. Druid doesn't return partial results incorrectly in such cases. #18025

Cluster management​

Configurable timeout for subtasks​

You can now configure a timeout for index_parallel and compact type tasks. Set the context parameter subTaskTimeoutMillis to the maximum time in milliseconds you want to wait before a subtask gets canceled. By default, there's no timeout.

Using this config helps parent tasks fail sooner instead of being stuck running zombie sub-tasks.

#18039

Coordinator APIs for clones​

The following Coordinator APIs are now available:

  • /druid/coordinator/v1/cloneStatus to get information about ongoing cloning operations.
  • /druid/coordinator/v1/brokerConfigurationStatus which returns the broker sync status for coordinator dynamic configs.

#17899

Other cluster management improvements​

  • Added the optional taskCountStart property to the lag based auto scaler. Use it to specify the initial task count for the supervisor to be submitted with #17900
  • Added audit logs for the following BasicAuthorizerResource update methods: authorizerUserUpdateListener, authorizerGroupMappingUpdateListener, authorizerUpdateListener (deprecated) #17916
  • Added support for streaming task logs to Indexers #18170
  • Improved how MSQ task engine tasks get canceled, speeding it up and freeing up resources sooner #18095

Metrics and monitoring​

Metrics for Historical cloning​

The following metrics for Historical cloning have been added:

  • config/brokerSync/time
  • config/brokerSync/total/time
  • config/brokerSync/error

Real-time ingestion metrics​

The following metrics for streaming ingestion have been added:

MetricDescriptionDimensions
ingest/events/maxMessageGapMaximum seen time gap in milliseconds between each ingested event timestamp and the current system timestamp of metrics emission. This metric is reset every emission period.dataSource, taskId, taskType, groupId, tags
ingest/events/minMessageGapMinimum seen time gap in milliseconds between each ingested event timestamp and the current system timestamp of metrics emission. This metric is reset every emission period.dataSource, taskId, taskType, groupId, tags
ingest/events/avgMessageGapAverage time gap in milliseconds between each ingested event timestamp and the current system timestamp of metrics emission. This metric is reset every emission period.dataSource, taskId, taskType, groupId, tags

#17847

Kafka consumer metrics​

The following metrics that correspond to Kafka metrics have been added:

Kafka metricDruid metric
bytes-consumed-totalkafka/consumer/bytesConsumed
records-consumed-totalkafka/consumer/recordsConsumed
fetch-totalkafka/consumer/fetch
fetch-ratekafka/consumer/fetchRate
fetch-latency-avgkafka/consumer/fetchLatencyAvg
fetch-latency-maxkafka/consumer/fetchLatencyMax
fetch-size-avgkafka/consumer/fetchSizeAvg
fetch-size-maxkafka/consumer/fetchSizeMax
records-lagkafka/consumer/recordsLag
records-per-request-avgkafka/consumer/recordsPerRequestAvg
outgoing-byte-totalkafka/consumer/outgoingBytes
incoming-byte-totalkafka/consumer/incomingBytes

#17919

Overlord kill task metrics​

MetricDescriptionDimensions
segment/killed/metadataStore/countNumber of segments permanently deleted from metadata storedataSource, taskId, taskType, groupId, tags
segment/killed/deepStorage/countNumber of segments permanently deleted from deep storagedataSource, taskId, taskType, groupId, tags
segment/kill/queueReset/timeTime taken to reset the kill queue on the Overlord. This metric is emitted only if druid.manager.segments.killUnused.enabled is true.
segment/kill/queueProcess/timeTime taken to fully process all the jobs in the kill queue on the Overlord. This metric is emitted only if druid.manager.segments.killUnused.enabled is true.
segment/kill/jobsProcessed/countNumber of jobs processed from the kill queue on the Overlord. This metric is emitted only if druid.manager.segments.killUnused.enabled is true.
segment/kill/skippedIntervals/countNumber of intervals skipped from kill due to being already locked. This metric is emitted only if druid.manager.segments.killUnused.enabled is true.dataSource, taskId

#18028

New task metrics​

The MSQ task engine and Dart now support the following metrics:

  • query/time: Reported by controller and worker at the end of the query.
  • query/cpu/time: Reported by each worker at the end of the query.

Additionally, MSQ task engine metrics now include the following dimensions:

  • queryId
  • sqlQueryId
  • engine: Denotes the engine used for the query, msq-dart or msq-task.
  • dartQueryId: (Dart only)
  • type: Always msq
  • dataSource
  • interval
  • duration
  • success

#18121

Other metrics and monitoring improvements​

  • Added the description dimension for the task/run/time metric
  • Added a metric for how long it takes to complete an autoscale action: task/autoScaler/scaleActionTime #17971
  • Added a taskType dimension to Overlord-emitted task count metrics #18032
  • Added the following groupBy metrics to the Prometheus emitter: mergeBuffer/used, mergeBuffer/acquisitionTimeNs, mergeBuffer/acquisition, groupBy/spilledQueries, groupBy/spilledBytes, and groupBy/mergeDictionarySize #17929
  • Changed the logging level for query cancellation from warn to info to reduce noise #18046
  • Changed query logging so that SQL queries that can't be parsed are no longer logged and don't emit metrics #18102
  • Changed the logging level for lifecycle from debug to info #17884
  • Added groupId and tasks to Overlord logs #17980
  • You can now use the druid.request.logging.rollPeriod to configure the log rotation period (default 1 day) #17976
  • Improved metric emission on the Broker to include per-query result-level caching (query/resultCache/hit returning 1 means the cache was used) #18063

Extensions​

Kubernetes​

  • The task runner log now includes the task id for a job and includes a log before a job is created #18105

Upgrade notes and incompatible changes​

Upgrade notes​

Hadoop-based ingestion​

Hadoop-based ingestion has been deprecated since Druid 32.0 and will be removed as early as Druid 35.0.0.

We recommend one of Druid's other supported ingestion methods, such as SQL-based ingestion or MiddleManager-less ingestion using Kubernetes.

As part of this change, you must now opt-in to using the deprecated index_hadoop task type. If you don't do this, your Hadoop-based ingestion tasks will fail. To opt-in, set druid.indexer.task.allowHadoopTaskExecution to true in your common.runtime.properties file. #18239

groupBy and topN queries​

Druid now uses the groupBy native query type, rather than topN, for SQL queries that group by and order by the same column, have LIMIT, and don't have HAVING. This speeds up execution of such queries since groupBy is vectorized while topN is not.

You can restore the previous behavior by setting the query context parameter useLexicographicTopN to true. Behavior for useApproximateTopN is unchanged, and the default remains true.

#18074

IS_INCREMENTAL_HANDOFF_SUPPORTED config removed​

Removed the IS_INCREMENTAL_HANDOFF_SUPPORTED context reference from supervisors, as incremental publishing has been the default behavior since version 0.16.0. This context was originally introduced to support rollback to LegacyKafkaIndexTaskRunner in versions earlier than 0.16.0, which has since been removed.

useMaxMemoryEstimates config removed​

Removed the useMaxMemoryEstimates config. When set to false, Druid used a much more accurate memory estimate that was introduced in Druid 0.23.0. That more accurate method is the only available method now. The config has defaulted to false for several releases.

#17936

Incompatible changes​

Developer notes​

  • Some maven plugins no longer use hard-coded version numbers. Instead, they now pull from the Apache parent #18138

Dependency updates​

  • Added j2objc-annotations #18154

The following dependencies have had their versions bumped:

  • apache.kafka from 3.9.0 to 3.9.1 #18178
  • aws.sdk for Java from 1.12.638 to 1.12.784 #18068
  • fabric8 from 6.7.2 to 6.13.1. The updated fabric8 version uses Vert.x as an HTTP client instead of OkHttp #17913
  • Curator from 5.5.0 to 5.8.0 #17857
  • com.fasterxml.jackson.core from 2.12.7.1 to 2.18.4 #18013
  • fabric8 from 6.13.1 to 7.2.0
  • org.apache.parquet:parquet-avro from 1.15.1 to 1.15.2 #18131
  • commons-beanutils:commons-beanutils from 1.9.4 to 1.11.0 #18132
  • form-data from 4.0.0 to 4.0.4 18310
  • guava from 32.0.1 to 32.1.3 #18154
  • confluent from 6.2.12 to 6.2.15 #18154
  • netty4 from 4.1.118.Final to 4.1.122.Final #18154
  • slf4j from 1.7.36 to 2.0.16 #18154
  • commons-logging from 1.1.1 to 1.3.5 #18154
  • commons-lang3 to 3.17.0 #18154
  • commons-text to 1.13.1 #18154
  • json-smart to 2.5.2 #18154
  • kotlin-stdlib to 1.9.25 #18154
  • joda-time to 2.14.0 #18154
  • com.google.code.findbugs to 3.0.2 #18154
  • log4j-slf4j updated to log4j-slf4j2 #18154
  • snappy-java to 1.1.10.7 #18154
  • httpcore to 4.4.16 #18154
  • asm to 9.8 #18154
  • async-http-client to 3.0.2 #18154
  • plexus-utils to 3.1.0 #18154
  • equalsverifier to 3.15.8 #18154
  • value-annotations to 2.10.1 #18154
  • form-data to 4.0.4 #18310