Broker

• [fix][broker] Key_Shared subscription doesn't always deliver messages from the replay queue after a consumer disconnects and leaves a backlog (#24736)
• [fix][broker] Ensure KeyShared sticky mode consumer respects assigned ranges (#24730)
• [fix][broker] PIP-428: Fix corrupted topic policies issues with sequential topic policy updates (#24427)
• [fix][broker][branch-4.0]Can not access topic policies if topic partitions have not been created (#24680)
• [fix][broker] Add double-check for non-durable cursor creation (#24643)
• [fix][broker] First entry will be skipped if opening NonDurableCursor while trimmed ledger is adding first entry. (#24738)
• [fix][broker] Fix cannot shutdown broker gracefully by admin api (#24731)
• [fix][broker] Fix duplicate watcher registration after SessionReestablished (#24621)
• [fix][broker] Fix memory leak when metrics are updated in a thread other than FastThreadLocalThread (#24719)
• [fix][broker] Fix race condition in MetadataStoreCacheLoader causing inconsistent availableBroker list caching (#24639)
• [fix][broker] Fix REST API to produce messages to single-partitioned topics (#24450)
• [fix][broker] Invalid regex in PulsarLedgerManager causes zk data notification to be ignored (#23977)
• [fix][broker] Prevent unexpected recycle failure in dispatcher's read callback (#24741)
• [fix][broker]Fix never recovered metadata store bad version issue if received a large response from ZK (#24580)
• [fix][ml]Fix EOFException after enabled topics offloading (#24753)
• [fix][broker] Fix incorrect AuthData passed to AuthorizationService in proxy scenarios (#24593)
• [fix][broker] Fix namespace deletion TLS URL selection for geo-replication (#24591)
• [fix][broker] Fix NPE and annotate nullable return values for ManagedCursorContainer (#24706)
• [fix][broker] Fix NPE being logged if load manager class name is blank (#24570)
• [fix][broker]Dispatcher did unnecessary sort for recentlyJoinedConsumers and printed noisy error logs (#24634)
• [fix][broker]Failed to create partitions after the partitions were deleted because topic GC (#24651)
• [fix][broker]Fix dirty reading of namespace level offload thresholds (#24696)
• [fix][broker]Fix thread safety issues in BucketDelayedDeliveryTracker with StampedLock optimistic reads (#24542)
• [fix][broker]User topic failed to delete after removed cluster because of failed delete data from transaction buffer topic (#24648)
• [fix][ml] Negative backlog & acked positions does not exist & message lost when concurrently occupying topic owner (#24722)
• [fix][meta] Use getChildrenFromStore to read children data to avoid lost data (#24665)
• [improve][admin] PIP-422 part 1: Support global topic-level replicated clusters policy (#24390)
• [improve][broker]Part-2 Add Admin API to delete topic policies (#24602)
• [improve][broker] If there is a deadlock in the service, the probe should return a failure because the service may be unavailable (#23634)
• [improve][ml] Optimize ledger opening by skipping fully acknowledged ledgers (#24655)
• [improve][broker] Allow deletion of empty persistent topics regardless of retention policy (#24733)
• [improve][broker] Extract duplication in AbstractTopic#incrementTopicEpochIfNeeded (#24520)
• [improve][broker]Find the target position at most once, during expiring messages for a topic, even though there are many subscriptions (#24622)
• [improve][broker]Improve the anti-concurrency mechanism expirationCheckInProgress (#24607)
• [improve][broker]Remove block calling that named cursor.asyncGetNth when expiring messages (#24606)

Client

• [fix][client] Avoid recycling the same ConcurrentBitSetRecyclable among different threads (#24725)
• [fix][client] fix ArrayIndexOutOfBoundsException in SameAuthParamsLookupAutoClusterFailover (#24662)
• [fix][client] Fix ArrayIndexOutOfBoundsException when using SameAuthParamsLookupAutoClusterFailover (#23336)
• [fix][client] Fix receiver queue auto-scale without memory limit (#24743)
• [fix][client] Retry for unknown exceptions when creating a producer or consumer (#24599)
• [fix][client] Skip schema validation when sending messages to DLQ to avoid infinite loop when schema validation fails on an incoming message (#24663)
• [fix][client]Prevent ZeroQueueConsumer from receiving batch messages when using MessagePayloadProcessor (#24610)
• [fix][client]TopicListWatcher not closed when calling PatternMultiTopicsConsumerImpl.closeAsync() method (#24698)
• [fix][client] rollback TopicListWatcher retry behavior (#24752)
• [improve][client] Support load RSA PKCS#8 private key (#24582)
• [fix][client] Exclude io.prometheus:simpleclient_caffeine from client-side dependencies (#24761)
• [improve][client] Add OpenTelemetry metrics for client memory buffer usage (#24647)
• [improve][client] RawReader support pause and resume (#24597)
• [fix][client] Refactor AttributeWrappedMeasurement to use a class instead of a record to compatible with lower version JDK

Pulsar IO and Pulsar Functions

• [fix][io] Improve Kafka Connect source offset flushing logic (#24654)
• [improve][io] Add dependency file name information to error message when .nar file validation fails with ZipException (#24604)
• [improve][io] Support specifying Kinesis KPL native binary path with 1.0 version specific path (#24669)
• [improve][io] Add support for the complete KinesisProducerConfiguration in KinesisSinkConfig (#24489)
• [feat][fn] Fallback to using STATE_STORAGE_SERVICE_URL in PulsarMetadataStateStoreProviderImpl.init (#24721)

Others

• [fix][proxy] Fix TooLongFrameException with Pulsar Proxy (#24626)
• [fix][misc] Fix compareTo contract violation for NamespaceBundleStats, TimeAverageMessageData and ResourceUnitRanking (#24772)
• [improve][doc] Improve the JavaDocs of sendAsync to avoid improper use (#24601)
• [fix][offload] Exclude unnecessary dependencies from tiered storage provider / offloader nar files (#24649)
• [fix][ws] Allow websocket principals to specify originalPrincipal without proxy role (#24642)
• [fix][ws] Fix WebSocket authentication with authenticateOriginalAuthData enabled (#24615)
• [fix][ws] Fix WebSocket proxy originalPrincipal for HTTP admin API calls (#24613)
• [fix] Exclude commons-lang dep from bookkeeper (#24749)

Tests & CI

• [improve][build] Use org.apache.nifi:nifi-nar-maven-plugin:2.1.0 with skipDocGeneration=true (#24668)
• [fix][ci] Fix code coverage metrics in Pulsar CI (#24595)
• [fix][test] Fix ConcurrentModificationException in Ipv4Proxy (#24632)
• [fix][test] Flaky-test: BrokerServiceTest.testShutDownWithMaxConcurrentUnload (#24769)
• [fix][test]fix flaky ZeroQueueSizeTest.testZeroQueueGetExceptionWhenReceiveBatchMessage (#24633)
• [fix][test]fix flaky ZeroQueueSizeTest.testZeroQueueGetExceptionWhenReceiveBatchMessage (#24630)
• [fix][broker] Fix flaky testReplicatorsInflightTaskListIsEmptyAfterReplicationFinished (#24590)
• [fix][broker] Fix testServiceConfigurationRetentionPolicy unit test (#24756)
• [fix][broker]Fix flaky test PartitionCreationTest.testCreateMissedPartitions (#24679)
• [fix][broker]Fix the wrong logic of the test PartitionCreationTest.testCreateMissedPartitions (#24683)
• [improve][build] Disable javadoc build failure (#24594)
• [improve][build] Increase maven resolver's sync context timeout (#24666)
• [improve][test] Add test for dead letter topic with max unacked messages blocking (#24535)
• [improve][test] Refactor the way way pulsar-io-debezium-oracle nar file is patched when building the test image (#24586)
• [fix][ci] Fix CI for Java 25 including upgrade of Gradle Develocity Maven extension (#24767)
• [fix][build][branch-4.0] Fix checkstyle in BrokerServiceAutoTopicCreationTest

For the complete list, check the full changelog.