What debezium/debezium shipped
Written by FoxPlug from the repository's own releases; not affiliated with Debezium. Paste your own repository in the box below and FoxPlug writes the same weekly update for your product. This page is an automatic summary of the releases, pull requests and commits in github.com/debezium/debezium; Debezium did not write it and does not use or endorse FoxPlug. Every line links to the change it describes.
Get a weekly update like this for your product, free
Week of November 2, 2026
What shipped
- PartitionRouting can now read configured fields from the record key, so tombstones with a null value can be routed consistently. Pull request #8274
- SQL Server change events no longer replace a NULL after-image with the unavailable value placeholder for large text and binary columns after an update. Pull request #8359
- Oracle columns with extended string sizes no longer emit null for new VARCHAR2 and NVARCHAR2 values during streaming when the declared length is small. Pull request #8350
- The PostgreSQL connector now logs a warning at startup when the replication slot position or stored offset suggests the WAL search may stall. Pull request #8367
- The MySQL connector now validates a hex encoded unavailable.value.placeholder value instead of accepting an invalid setting silently. Pull request #8381
- The SQL Server connector now avoids storing an empty or stale change position after a snapshot, so the
when_neededsnapshot is no longer skipped. Pull request #8373 - DefaultOffsetStorageReader now reads integral offset values as Long, matching how Kafka Connect offsets are stored. Pull request #8353
- The ActivateTracingSpan SMT gains a tracing.span.context.format option, with json allowing a JSON object to hold the span context. Pull request #8399
- With decoderbufs, a BEGIN sharing a WAL position with already processed rows no longer consumes a skip and leaves the connector replaying incorrectly. Pull request #8355
- The MariaDB Connector/J dependency is bumped to 3.5.11, covering the reported vulnerabilities in the previous version. Pull request #8388
Why it matters
This week delivered a 3.7.1.Final maintenance release and the 3.8.0.Alpha1 release, alongside fixes that affect day to day data correctness and connector stability. Several changes address silently wrong or missing values in SQL Server and Oracle events, and others improve offset handling and startup diagnostics that users encounter after restarts.
Changelog entry
- PartitionRouting reads configured fields from the record key, allowing tombstones to be routed consistently. Pull request #8274
- SQL Server connector no longer skips the
when_neededsnapshot after a snapshot stored a null commit LSN. Pull request #8373 - SQL Server events keep NULL in the after state of large text and binary columns instead of using the unavailable value placeholder. Pull request #8359
- Oracle extended string columns emit their new VARCHAR2 and NVARCHAR2 values instead of null during streaming. Pull request #8350
- PostgreSQL connector logs WARN messages when the replication slot or stored offset indicates a stalled WAL search. Pull request #8367
- MySQL connector validates hex encoded unavailable.value.placeholder settings. Pull request #8381
- DefaultOffsetStorageReader reads integral offset values as Long. Pull request #8353
- ActivateTracingSpan supports JSON serialized span context through tracing.span.context.format. Pull request #8399
- Decoderbufs no longer uses up a skip for an already processed row when BEGIN shares a WAL position with its first rows. Pull request #8355
- MariaDB Connector/J bumped to 3.5.11. Pull request #8388
- SQL Server JDBC driver bumped to 13.6.2.jre11. Pull request #8409
- Changelog and version updates prepared for the 3.7.1.Final release. Pull request #8415
Debezium 3.7.1.Final and 3.8.0.Alpha1 are out. This week also brought fixes for SQL Server null handling after updates, Oracle extended string values, connector offset reading and PostgreSQL WAL search warnings.
Debezium 3.7.1.Final and 3.8.0.Alpha1 are now available. Alongside the releases, this week's changes focus on data correctness and operational clarity: SQL Server no longer substitutes the unavailable value placeholder for a real NULL after an update, Oracle keeps new values for columns with extended string sizes, and integral offsets are read as Long across the JDBC, Redis, NATS and ConfigMap stores. The PostgreSQL connector warns at startup when the replication slot or stored offset suggests the WAL search will stall, and the MySQL connector now rejects an invalid hex encoded unavailable value placeholder.
Week of October 26, 2026
What shipped
- An opt-in binlog-metadata-based schema mode was added for the MySQL and MariaDB connectors, reconstructing streaming table schema from TABLE_MAP metadata instead of a schema history topic. Pull request #7593
- The sink framework became the default mode for the JDBC sink connector, with CI moved to sink framework mode only and daily tests covering legacy mode. Pull request #8234
- The connector validate() step now checks signal.data.collection existence, fully qualified name shape and effective column count across MySQL, MariaDB, PostgreSQL, SQL Server, Oracle and MongoDB. Pull request #8025
- The JDBC sink connector now supports primitive record keys when primary.key.mode is set to
record_key. Pull request #8310 - The JDBC sink connector now converts connection.pool.timeout in seconds rather than milliseconds for the Agroal pool, so the documented 1800 second default is no longer applied as 1.8 seconds. Pull request #8337
- The Oracle connector now uses the row SCN for source.scn with OpenLogReplicator and falls back to
c_scnwhen absent, while the offset still usesc_scn. Pull request #8302 - SQL Server datetime chunk bounds in incremental snapshots are now cast correctly, accounting for the type's 1/300 second tick storage. Pull request #8279
- Oracle INTERVAL DAY TO SECOND fractional seconds are now scaled correctly instead of being read as microseconds. Pull request #8375
- MongoDB connection string masking now handles percent-encoded credentials. Pull request #8368
- A CI check was added for issue to PR links. Pull request #8336
Why it matters
This week brings a new opt-in schema mode for MySQL and MariaDB that removes reliance on a persisted schema history topic, plus a default switch to the sink framework for the JDBC sink connector. Several fixes address values that were silently wrong or unreliable: JDBC pool timeouts, SQL Server chunk bounds, Oracle intervals, MongoDB credential masking and PostgreSQL idle connections. Validation of signal.data.collection at connector validate() time also moves several misconfiguration failures earlier.
Changelog entry
- MySQL and MariaDB: add opt-in binlog-metadata-based schema mode Pull request #7593
- JDBC sink: make the sink framework the default mode Pull request #8234
- Connectors: validate signal.data.collection at validate() time for MySQL, MariaDB, PostgreSQL, SQL Server, Oracle and MongoDB Pull request #8025
- JDBC sink: support primitive record keys with primary.key.mode=record_key Pull request #8310
- JDBC sink: treat connection.pool.timeout as seconds when configuring the Agroal pool Pull request #8337
- Oracle: use the row SCN for source.scn with OpenLogReplicator Pull request #8302
- SQL Server: cast datetime chunk bounds correctly for incremental snapshots Pull request #8279
- Oracle: scale INTERVAL DAY TO SECOND fractions correctly Pull request #8375
- MongoDB: mask percent-encoded credentials in connection strings Pull request #8368
- PostgreSQL: avoid opening an idle bean registry connection Pull request #8330
- Core: widen MathOps.add long sums to BigInteger on overflow Pull request #8326
- JDBC sink: recover schema evolution after a lost alter Pull request #8284
This week: an opt-in binlog-metadata schema mode for MySQL and MariaDB, the sink framework becomes the default for the JDBC sink connector, signal.data.collection is validated at connector validate() time, and fixes for JDBC pool timeouts, SQL Server datetime chunk bounds and Ora
This week in Debezium: an opt-in binlog-metadata-based schema mode for the MySQL and MariaDB connectors lets table schema be reconstructed from TABLE_MAP events instead of a schema history topic. The sink framework is now the default mode for the JDBC sink connector, with daily tests covering the legacy mode. Connector validate() now checks signal.data.collection existence, name shape and column count across MySQL, MariaDB, PostgreSQL, SQL Server, Oracle and MongoDB. Fixes include JDBC sink support for primitive record keys with record_key mode, correct seconds conversion for connection.pool.timeout, correct Oracle INTERVAL DAY TO SECOND fractions, SQL Server datetime chunk bound casting, percent-encoded credential masking in MongoDB connection strings, and avoiding idle bean registry conn
Week of October 19, 2026
What shipped
- The EnforceRecordSize SMT gains an opt-in
claim_checkstrategy with an OversizedRecordStorage SPI in debezium-storage-api and an S3 implementation, keeping proportional truncation as the default. Pull request #7932 - A new debezium-storage-nats module provides NATS JetStream-backed storage for connector offsets and internal schema history. Pull request #7904
- MySQL and MariaDB connectors now fail on fatal binlog error 1236 instead of retrying the stream forever. Pull request #8296
- MySQL and MariaDB snapshots read DATETIME values without a time zone and handle invalid dates, matching streaming output under the default time.precision.mode. Pull request #8314
- The pgoutput decoder resolves tuple column types by OID, fixing columns whose unqualified type name collides with a built-in type in another schema. Pull request #8311
- PostgreSQL text[] and varchar[] column defaults are now parsed into the generated Connect schema instead of being dropped. Pull request #8282
- SQL Server function-mode CDC changes are ordered by command id, so a delete and insert sharing a sequence value are no longer emitted in the wrong order. Pull request #8275
- SQL Server schema recovery treats a missing change LSN in a heartbeat offset correctly, so post-snapshot history records are no longer all seen as newer. Pull request #8287
- GeometryFormatTransformer now uses the configured format as its target instead of always converting to the other format. Pull request #8290
- A StringifyFields transformation in debezium-connect-plugins serializes selected record-value fields to JSON strings while other fields stay strongly typed. Pull request #7782
Why it matters
This week fixes several correctness problems that surface after restarts or reconnects: schema history comments, SQL Server LSN handling and ordering, and snapshot versus streaming date output. It also adds new storage options for offsets and schema history plus new transformations, and removes configuration that no longer did anything.
Changelog entry
- Add an opt-in
claim_checkstrategy and OversizedRecordStorage SPI to the EnforceRecordSize SMT. Pull request #7932 - Add debezium-storage-nats for NATS JetStream offset and schema history storage. Pull request #7904
- Fail on fatal binlog error 1236 in MySQL and MariaDB instead of retrying. Pull request #8296
- Read MySQL and MariaDB snapshot DATETIME values without a time zone and handle invalid dates. Pull request #8314
- Resolve pgoutput tuple column types by OID. Pull request #8311
- Parse PostgreSQL text[] and varchar[] defaults into the Connect schema. Pull request #8282
- Order SQL Server function-mode CDC changes by command id. Pull request #8275
- Recover SQL Server schema changes when the offset has no change LSN. Pull request #8287
- Fix schema history serde for table-level comments. Pull request #8233
- Use the configured format as the target in GeometryFormatTransformer. Pull request #8290
- Stop MathOps.add from narrowing double sums to Float. Pull request #8319
- Remove the unused database.server.id.offset property from MySQL and MariaDB connectors. Pull request #8295
This week: pluggable claim-check storage in EnforceRecordSize, NATS JetStream offset and schema history storage, fatal binlog errors no longer retried, PostgreSQL array defaults in the Connect schema, and SQL Server CDC ordering fixes.
This week brings new storage options and a set of correctness fixes. The EnforceRecordSize SMT gains an opt-in claim_check strategy with a storage-neutral SPI and an S3 implementation, and a new module stores offsets and internal schema history in NATS JetStream. MySQL and MariaDB now fail on fatal binlog error 1236 instead of retrying forever, snapshots read DATETIME values to match streaming, and PostgreSQL text[] and varchar[] defaults now appear in the Connect schema. Fixes also cover pgoutput type resolution by OID, SQL Server CDC ordering and schema recovery, geometry format conversion, and table-level comment serde in schema history.
Week of October 12, 2026
What shipped
- Fixed an Oracle data loss race condition caused by incorrect use of the log file's next system change number at log boundaries. Pull request #8253
- Oracle snapshot locks are now held for the whole initial snapshot instead of only during schema read. Pull request #8172
- MySQL streaming can now resume from an offset that has a binlog position but no GTID set. Pull request #8244
- MongoDB pre- and post-images can be configured independently, including requiring them so the connector fails visibly when images are unavailable. Pull request #8087
- Incremental snapshot state is now cleared when a snapshot is aborted, so it no longer leaks into a later incremental snapshot. Pull request #7201
- MySQL and MariaDB REAL columns now map to DOUBLE, matching how the server treats REAL unless REAL_AS_FLOAT is enabled. Pull request #7641
- SQL Server datetime, datetime2 and smalldatetime values without a timezone are preserved during DST gaps instead of being shifted. Pull request #8240
- TIME values are now decoded correctly in cursor-based MySQL snapshots, including fractional seconds. Pull request #8232
- String truncation in TruncateColumn no longer splits surrogate pairs, which had produced invalid Unicode and broke strict JSON parsing. Pull request #8261
- Added initial documentation for the SQLite source connector, including how it captures changes through triggers into an internal log table. Pull request #8038
Why it matters
This week covers several correctness fixes for MySQL, MariaDB, Oracle and SQL Server that could affect the values and defaults users see in change events, plus resume and locking fixes for Oracle and MySQL. The MongoDB, SQLite documentation and schema handling changes broaden what users can configure and rely on. Several of these fix silent data problems rather than adding new surface area.
Changelog entry
- Oracle: fix data loss race condition on log boundaries. Pull request #8253
- Oracle: hold snapshot locks until the end of the snapshot. Pull request #8172
- MySQL: resume streaming from an offset without a GTID set. Pull request #8244
- MongoDB: support required and independently configured pre- and post-images. Pull request #8087
- Clear incremental snapshot state when a snapshot is aborted. Pull request #7201
- MySQL and MariaDB: map REAL columns to DOUBLE. Pull request #7641
- SQL Server: preserve timezone-less timestamps during DST gaps. Pull request #8240
- MySQL: fix TIME decoding in cursor-based snapshots. Pull request #8232
- Don't split surrogate pairs when truncating strings. Pull request #8261
- Oracle: fix NUMBER scale resolution with asterisk length. Pull request #8248
- Add documentation for the SQLite connector. Pull request #8038
- Fix schema merging of JSON array elements. Pull request #8260
This week: Oracle snapshot locks now cover the full snapshot, an Oracle log boundary race that could lose data is fixed, and MySQL streaming can resume from an offset without a GTID set. Also MongoDB pre/post-image modes, SQL Server DST handling and SQLite docs.
This week in Debezium: an Oracle data loss race at log boundaries was fixed, and Oracle snapshot locks are now held for the full initial snapshot rather than only during schema read. MySQL streaming can resume from an offset that has a binlog position but no GTID set, incremental snapshot state is cleared when a snapshot aborts, and MySQL/MariaDB REAL columns now map to DOUBLE. SQL Server timezone-less timestamps are preserved during DST gaps, MongoDB pre- and post-images can be configured independently, time decoding in cursor-based MySQL snapshots is corrected, and string truncation no longer splits surrogate pairs. Initial SQLite connector documentation was also added.
Week of October 5, 2026
What shipped
- Adds the Neo4j Debezium Converter, a single message transform that converts Neo4j CDC change events into the Debezium envelope. Pull request #8143
- Applies primary key field filtering when primary key mode is record header and primary key fields are configured. Pull request #8181
- Fixes MongoDB connection and authentication resource cleanup by giving authentication providers a connector-managed cleanup callback. Pull request #8067
- Validates MongoDB connector access against the configured capture scope and target instead of only listing database names. Pull request #8123
- Preserves NULL elements in PostgreSQL ltree arrays so an array containing a SQL NULL no longer drops the whole array or stops the connector. Pull request #8199
- Emits heartbeats for quiet unbuffered Oracle LogMiner streams after each completed mining query. Pull request #8169
- Handles XA START and XA PREPARE as transaction boundaries. Pull request #8139
- Runs connector cleanup when coordinator shutdown fails by calling stop from a finally block. Pull request #8126
- Applies table and database charset changes in DDL parsers so string columns added later are decoded with the correct character set. Pull request #8227
- Rejects a change stream stage in the MongoDB cursor pipeline instead of retrying without capturing events. Pull request #8229
Why it matters
This week brings a new Neo4j converter transform, several MongoDB connector fixes around authentication cleanup, access validation and cursor pipeline handling, and Oracle changes covering heartbeats on quiet streams, XA transaction boundaries and cleanup on shutdown failure. PostgreSQL array handling and character set changes in DDL parsing also address data correctness issues that connector users can hit during streaming. Documentation updates cover Oracle downstream mining and standby capture, Informix and MongoDB references, and a new reference chapter for the Debezium command-line interface.
Changelog entry
- Add Neo4j Debezium Converter transform for Neo4j CDC change events. Pull request #8143
- Apply primary key fields to record header key mode. Pull request #8181
- Fix MongoDB connection and authentication resource cleanup. Pull request #8067
- Validate MongoDB connector access against the configured capture scope. Pull request #8123
- Preserve NULL elements in PostgreSQL ltree arrays. Pull request #8199
- Emit heartbeats for quiet unbuffered Oracle LogMiner streams. Pull request #8169
- Handle XA START and XA PREPARE as transaction boundaries. Pull request #8139
- Run connector cleanup when coordinator shutdown fails. Pull request #8126
- Apply table and database charset changes in DDL parsers. Pull request #8227
- Reject change stream stage in MongoDB cursor pipeline. Pull request #8229
- Preserve explicit null values in TimezoneConverter. Pull request #8128
- Add reference documentation for the Debezium command-line interface. Pull request #8144
New this week: a Neo4j Debezium Converter transform that turns Neo4j CDC changes into the Debezium envelope, MongoDB access validation against the configured capture scope, PostgreSQL ltree array NULL handling, and heartbeats for quiet Oracle LogMiner streams.
This week the project adds the Neo4j Debezium Converter, a single message transform that converts Neo4j CDC change events into the Debezium envelope for graph-to-relational pipelines. MongoDB gets fixes for authentication resource cleanup and access validation that now checks the configured capture scope and target, and it rejects a change stream stage in the cursor pipeline rather than retrying silently. PostgreSQL ltree arrays keep NULL elements instead of dropping the array or stopping the connector, and DDL parsers now apply table and database charset changes. Oracle gains heartbeats for quiet unbuffered LogMiner streams, XA START and XA PREPARE treated as transaction boundaries, and connector cleanup when coordinator shutdown fails.
Week of September 28, 2026
What shipped
- Version 3.7.0.Final released. Release
- MySQL and MariaDB integer schema defaults now preserve precision for large BIGINT values and correctly round negative fractional defaults. Pull request #8158
- MongoDB incremental snapshots now replay unfinished snapshot windows after restart instead of skipping undelivered keys. Pull request #8117
- PostgreSQL snapshot reads of date columns and temporal arrays now preserve era information instead of silently changing values. Pull request #8085
- MongoDB connector adds capture.start.timestamp setting to accept readable Unix seconds and ISO-8601 timestamps instead of requiring packed BSON. Pull request #8007
- MongoDB change events now expose resume tokens in optional source.resume_token metadata. Pull request #8056
- SQL Server queries reading cdc.lsn_time_mapping now use NOLOCK hint to avoid blocking locks on every streaming batch. Pull request #8120
- TimescaleDB 2.29+ transforms are now supported. Pull request #7746
- ExtractNewDocumentState SMT adds lossless BSON Timestamp handling mode to preserve both time and increment components. Pull request #7975
- Locale.ROOT is now used consistently in SerializerType.withName to fix casing issues in locales with special rules like Turkish. Pull request #8138
Why it matters
Version 3.7.0 ships numerous correctness fixes across multiple connectors: MySQL and MariaDB now preserve numeric precision, MongoDB snapshots no longer skip keys on restart, PostgreSQL temporal types maintain era information, and SQL Server reduces locking contention. The release also adds new configuration options for MongoDB timestamps and improves BSON handling losslessness.
Changelog entry
- Preserve MySQL and MariaDB integer schema defaults without precision loss Pull request #8158
- Replay unfinished MongoDB incremental snapshot windows after restart Pull request #8117
- Preserve era for PostgreSQL DATE and temporal array types Pull request #8085
- Add capture.start.timestamp to MongoDB connector for readable timestamp formats Pull request #8007
- Expose MongoDB resume tokens in source.resume_token metadata Pull request #8056
- Apply NOLOCK hint to SQL Server cdc.lsn_time_mapping queries Pull request #8120
- Preserve SQL NULL defaults in MySQL ALTER COLUMN Pull request #8153
- Add lossless BSON Timestamp handling mode to ExtractNewDocumentState Pull request #7975
- Add support for TimescaleDB 2.29+ transform Pull request #7746
- Use Locale.ROOT in SerializerType.withName for consistent casing Pull request #8138
- Shut down MongoDB incremental snapshot thread pool properly Pull request #8070
- Wake MongoDB streaming consumers for buffered events instead of waiting for backoff Pull request #8151
Debezium 3.7.0 released. Fixes for MySQL integer defaults, MongoDB snapshot window replay, PostgreSQL temporal types, SQL Server locking, and more.
Debezium 3.7.0 is now available. This release includes critical correctness fixes: MySQL/MariaDB preserve integer precision, MongoDB snapshots properly replay unfinished windows, PostgreSQL temporal arrays maintain era information, and SQL Server reduces lock contention. New features include readable MongoDB timestamps and lossless BSON Timestamp handling in ExtractNewDocumentState SMT.
Week of September 21, 2026
What shipped
- Release v3.7.0.CR1 shipped with bug fixes across Oracle, PostgreSQL, MySQL, SQL Server, MongoDB, and other connectors. Release
- MongoDB relational mapping SMT added to convert JSON strings in before/after fields into Kafka Connect Structs while retaining the Debezium envelope. Pull request #8008
- HTTP notification channel introduced to send snapshot notifications via HTTP POST to configurable endpoints. Pull request #8050
- CappedLogFileSessionSelector now eligible for
dictionary_from_filemining strategies in Oracle standby streams. Pull request #8063 - Oracle mining window unit now derives from configured log.mining.log.count.min across collected logs rather than relying on online redo log sizes. Pull request #8071
- Oracle snapshot.include.collection.list snapshot behavior fixed. Pull request #8091
- SQL Server nvarchar(max) detection improved to accept either length the metadata reports. Pull request #8078
- Initial snapshot STARTED notification now includes the list of data collections captured in the snapshot. Pull request #7978
- PostgreSQL temporal infinity constants fixed for epoch scale and year sign. Pull request #7927
- Strings.mask fixed to use the replacement mask parameter instead of hardcoding and to handle null inputs without throwing NullPointerException. Pull request #8121
Why it matters
Version 3.7.0.CR1 includes improvements across multiple database connectors addressing snapshot behavior, mining strategies, temporal handling, and notification capabilities. The MongoDB relational mapping SMT and HTTP notification channel represent new functionality that expands Debezium's flexibility for different use cases. Multiple connector-specific fixes address edge cases in Oracle, PostgreSQL, MySQL, and SQL Server that improve reliability and correctness.
Changelog entry
- Strings.mask now uses the provided replacement mask parameter and handles null original strings without throwing NullPointerException Pull request #8121
- Oracle log mining window unit now derives from configured log.mining.log.count.min across collected logs Pull request #8071
- Fixed Oracle snapshot.include.collection.list snapshot behavior Pull request #8091
- TimescaleDB SMT can now deduce hypertable ID from chunk names when recently dropped chunks are not found Pull request #8109
- MongoDB primary election test stabilized to retry transient election failures Pull request #8111
- Schema history validation submission and executor shutdown now synchronized to prevent source task startup failures Pull request #8104
- TimescaleDB SMT now retries on EOF exceptions when connection to metadata is closed by proxy Pull request #8099
- IsAutonomous() made resilient on databases that do not support CLOUD_SERVICE Pull request #8106
- PartitionRouting now handles null intermediate structs gracefully when configured with nested payload field paths Pull request #8101
- Fixed regression with internal.log.mining.relaxed.quote.detection configuration Pull request #8102
- Added MongoDB relational mapping SMT (MongoToRelationalMapper) to convert JSON strings in before/after into Kafka Connect Structs Pull request #8008
- ExtractNewDocumentState now emits null for fields removed by $unset operations Pull request #7974
- PostgreSQL JDBC sink now properly binds Connect ARRAY values to PostgreSQL bytea[] columns Pull request #7972
- MariaDB standalone DDL GTID events now propagated correctly with binlog buffering enabled Pull request #7862
- SQL Server nvarchar(max) detection improved to accept either length reported by metadata Pull request #8078
- JsonNode method matching improved to handle class loader isolation in Kafka Connect plugins Pull request #8100
- TIME snapshot now supports midnight value Pull request #8064
- Added HTTP notification channel to send snapshot notifications via HTTP POST Pull request #8050
- Initial snapshot STARTED notification now includes list of data collections captured in snapshot Pull request #7978
Debezium 3.7.0.CR1 released with Oracle mining improvements, MongoDB relational mapping SMT, HTTP notifications, and fixes across PostgreSQL, MySQL, SQL Server connectors.
Debezium 3.7.0.CR1 is now available. This release includes significant enhancements: a new MongoDB relational mapping SMT for schema-aware transformations, HTTP-based notification channels for snapshot events, improved Oracle log mining strategies for standby streams, and numerous connector-specific fixes for PostgreSQL temporal handling, SQL Server nvarchar(max) detection, and MySQL/MariaDB snapshot stability.
Week of September 14, 2026
What shipped
- Released v3.6.3.Final with JBang handling and version calculation fixes, changelog updates, and contributor file sorting in release scripts. Release
- Released v3.7.0.Beta2 with changelog and version updates. Release
- Added TotalAbandonedTransactionCount JMX metric for Oracle LogMiner. Pull request #8054
- Added Oracle LogMiner log-pending-transactions signal with new synchronous signal action hook. Pull request #7991
- Added per-table row count estimation and progress reporting to incremental snapshots. Pull request #8016
- Extended Oracle
index_attributesgrammar rule to support ONLINE keyword. Pull request #8055 - Added documentation for chunk-based initial snapshots across SQL-based connectors including Db2, Informix, Ingres, MariaDB, MySQL, Oracle, PostgreSQL, and SQL Server. Pull request #8021
- Fixed ExtractChangedRecordState binary comparison and schema evolution issues affecting unchanged binary and array fields. Pull request #8030
- Added SEQUENCE# to XML begin and end events and moved last enqueued event map to transaction cache. Pull request #8024
- Added handling for Oracle XStream LOB_WRITE, LOB_TRIM, and LOB_ERASE events in the XStream adapter. Pull request #7374
Why it matters
Two releases shipped this week with Oracle LogMiner enhancements, improved incremental snapshot visibility, and fixes for binary field comparisons and XStream LOB operations. Documentation expanded for chunk-based snapshots across multiple SQL connectors, improving user guidance for initial snapshot configuration.
Changelog entry
- Released version 3.6.3.Final Release
- Released version 3.7.0.Beta2 Release
- Oracle LogMiner: Added TotalAbandonedTransactionCount JMX metric Pull request #8054
- Oracle LogMiner: Added log-pending-transactions signal with synchronous signal action hook Pull request #7991
- Oracle: Extended
index_attributesgrammar rule to support ONLINE keyword Pull request #8055 - Incremental snapshots: Added per-table row count estimation and progress reporting Pull request #8016
- ExtractChangedRecordState: Fixed binary comparison and schema evolution issues Pull request #8030
- Oracle LogMiner: Added SEQUENCE# to XML begin/end events, moved last enqueued event map to transaction cache Pull request #8024
- Oracle XStream: Added handling for LOB_WRITE, LOB_TRIM, and LOB_ERASE events Pull request #7374
- Documentation: Added chunk-based initial snapshots documentation for SQL-based connectors Pull request #8021
- Oracle: Added ORA-02002 as retriable exception Pull request #8027
- MySQL/MariaDB: Fixed null throwable handling in error retry logic Pull request #8009
- MongoDB: Fixed validation of split-event resume tokens Pull request #8014
- MongoDB: Fixed read preference handling after replica set elections Pull request #7950
- MongoDB: Made source.collection field optional to fix DataException on connector startup Pull request #7980
- JDBC storage: Removed unnecessary validation blocking passwordless authentication for offsets and schema history Pull request #8010
- Oracle LogMiner: Improved active/pending transaction debug logging Pull request #7992
V3.6.3.Final and v3.7.0.Beta2 released. New Oracle LogMiner metrics, improved incremental snapshot progress reporting, XStream LOB event handling, and expanded chunk-based snapshot documentation.
Debezium shipped v3.6.3.Final and v3.7.0.Beta2 this week. Highlights include Oracle LogMiner enhancements with new JMX metrics and pending-transaction signals, per-table row count progress reporting in incremental snapshots, XStream LOB operation support, and comprehensive documentation for chunk-based snapshots across Db2, Informix, MariaDB, MySQL, Oracle, PostgreSQL, and SQL Server.