Oracle LogMiner Agent Changelog

GSSD-1245: An incomplete row image is re-read while the COMMIT is processed, instead of as soon as its DML is mined, which replicated a sparse UPDATE with the values it replaced or dropped it as "Row not found";

GSSD-1245: A row whose ROWID cannot find it is read by its primary key, so replaying changes older than a TRUNCATE, an ALTER TABLE MOVE or a SHRINK SPACE no longer leaves the reader stuck on ORA-01410;

LOB columns are now replicated, so an UPDATE of another column no longer nulls the LOB on the target, an out-of-row INSERT no longer lands as the literal EMPTY_CLOB(), and a change made through the locator is captured, in a PDB too;

An entity with no stored position resumes from the database’s current position, instead of falling back to a reader checkpoint that can sit hours behind and replaying those hours on top of the snapshot;

The rewind after a lost change cache is armed once instead of at every reader start, so a stop and start no longer replays everything back to the oldest entity position;

GSSD-1374: Added support for "Enable TLS" and the trust store and key store of the connection, so reaching Oracle over TLS no longer requires a wallet;

GSSD-1374: Added a LogMiner instance name so the mining connections go through the wallet too, instead of opening in clear text and being refused by a TCPS-only listener;

Arena batches are handed over in the order they were written instead of being re-sorted by transaction id, which on Oracle LogMiner inverted two changes sharing an SCN and wrote the older value to the target;

Booleans and maps keep their type through the arena cache, instead of reaching the target as "true"/"false" and as a non-JSON string;

A failing step of the reader stop no longer skips the rest of it, which left the loop running and turned every later stop into a no-op;

Added the source lag probe, reporting the unread change backlog and how far behind the last read transaction is;

GSSD-1357: The two unchecked arena length fields are now bounded, so a damaged arena is reported as a corrupted record instead of surfacing as a NumberFormatException or advancing the recovery scan into garbage;

GSSD-1357: An arena is no longer mapped at a size the filesystem refused to give it, which on a full disk wrote into a sparse region with no blocks behind it;

ArenaCache now splits across multiple files, one per entity instead of a single shared file. See Source agent caching layer;

Optimized regex usage on the CDC hot path;

GSSD-1245: Fixed a CDC restart rewinding to the position of an idle table, which could cause a reported data gap;

GSSD-1245/1306: Fixed data loss when the redo logs rotate;

GSSD-1306: Fixed log window rotation can cause data loss;

GSSD-1245: Fixed commit check;

Target character columns: a value longer than its target column is now refused instead of being stored truncated with the transaction reported as written. Affects the IBM i series, Informix, MS SQL Server and Oracle targets, on the direct and the bulk write path alike (GSSD-1292, GSSD-1267);

Oracle target: a source value wider than its target column now fails the write even when the excess is only the padding of a fixed-length source; apply a TrimString field function on the target column to shorten a value on purpose;

MS SQL Server and Oracle targets: a failed insert batch is retried as an upsert only when the target already holds the row, so a rejection from the server is no longer turned into a truncation;

Oracle LogMiner: redo ages are now measured against the database clock instead of the agent host one, so on a database that is not on UTC transactions are no longer abandoned before their deadline and cached changes no longer expire early (GSSD-1245);

GSSD-1144: LogMiner no longer discards redo over a log file list registered by another connection, dropping the stale registrations and retrying the session start once instead;

GSSD-1144: a LogMiner session failure now reports the actual Oracle error code instead of the list of candidate ones;

GSSD-1144: corrected the warning logged for a DELETE with no undo image, now naming the table and the supplemental logging remedy;

GSSD-1144: a DELETE with no usable before-image now dumps its raw payload at error level, with the per-table occurrence count;

GSSD-1245: Fixed pending changes being silently skipped and acknowledged when LogMiner emits records out of order, causing missing rows on the target and entities stuck reporting no changes to sync;

Avoid blocking low watermark when pending uncommitted transactions are holding it back;

GSSD-1144: Fixed Oracle LogMiner data loss and low throughput;

GSSD-1245: Improved arena cache recover and shutdown barrier;

GSSD-1144: Improve Arena Cache corruption barrier to improve restart with corrupted cache file;

GSSD-1144: Fixed localised dates parsing;

GSSD-1144: Fixed Arena Cache transaction management when primary keys modified;

GSSD-1160: Fixed escape single quotes in Oracle staging table column names;

GSSD-1144: Fixed date parsing;

GSSD-1144: Fixed retrieve of current SCN when checkpoint not exist consuming too much DB CPU;

GSSD-1138: Fixed transaction Id comparison logic;

GSSD-1138: Fixed wrong start checkpoint of LogMiner reader after a snapshot;

GSSD-1144: Fixed infinite loop on parse failures with online logs;

Removed active transaction checkpoint block;

Fixed issue with JDBC that writes empty string as NULL value;

GSSD-1138: Fixed ROWID technical field column emitted by the source even if not selected;

GSSD-1144: Added error ORA-00310 handling;

Fixed Oracle write query not using table name with quotes;

GSSD-1138: Fixed wrong description for source change retention causing ambiguity and loss of data when an entity is paused;

GSSD-1049: Fixed sparse deletion not reconcilied;

GSSD-1049: Added LogMiner start error handling restarting session;

Fixed DECIMAL parsing;

GSSD-1056: Added new date formats parsing support;

Enhanced discovery of logging type (both at db level and table level);

GSSD-1056: Fixed missing column values with min logging support;

GSSD-1032: Improved a specific code path that was rising unecessary ERRORs for transactions that were supposed to be discarded instead;

Added logic to recreate staging table for bulk operations to keep it aligned with entity schema;

GSSD-1004: Improved handling of ORA-00308 error;

Replaced Chronicle Queue in-flight transaction spill store with a custom off-heap memory-mapped file implementation (OffHeapMmf) modelled after the oracdc project. Each uncommitted transaction is persisted to a dedicated per-XID mmap file; full rollback is now O(1) (file close + delete). No more per-row Chronicle appender churn or 80 MB pre-allocated roll-cycle files.

Committed transaction cache for the Oracle LogMiner source now backed by InMemoryDatabaseCacheImpl, eliminating Chronicle Queue I/O and file locking on the committed side.

Added socket-level read timeout on the LogMiner JDBC connection (logminerReadTimeout, default 60000 ms). Prevents executeQuery() from blocking indefinitely when Oracle becomes unreachable.

Added Connection.isValid() probe before each LogMiner polling window to detect zombie connections before they stall a query.

In-place recovery on Oracle host restart or ORA-17800: worker drops orphaned in-flight transaction state, reconnects, and resumes from the persisted SCN checkpoint — no software restart required.

Fixed self-cancellation bug in the error-recovery path: safeRestartService was cancelling its own coroutine scope from within the loop, logging Job was cancelled and leaving the worker permanently dead.

Fixed FileNotFoundException after in-place recovery: cleanupAll() was deleting the spill root directory which was only created once via by lazy.

GSSD-990: Added support for trimming trailing spaces for text columns;

GSSD-972: Fixed incorrect equals and compareTo logic causing a false positive when comparing transaction IDs, leading Gluesync to perform unwanted and unnecessary reads from the cache.

GSSD-972: fixed Oracle data types migration;

GSSD-966: fixed Oracle LogMiner reader blocked by missing commit operation;

Fixed Oracle number data type wrong management;

GSSD-862: Improved Oracle CDC transaction id comparison and added logging;

Improved Oracle LogMiner reliability to avoid losing transactions in edge cases.

Added DLQ handling for SQL parse errors (LOGMNR_PARSE_DLQ) with configurable behavior (skipTransactionsWithParseErrors).

Improved checkpoint + transaction ordering logic to prevent replay loops / stuck progress scenarios.

Added spill-to-disk buffering for very large transactions to reduce memory pressure and improve stability.

Improved handling of CSF/fragmented SQL rows and log fast-forward/read behavior.

Added stronger regression test coverage (including full LogMiner dump replay) for safer releases.

Improved support for enhaced columns discovery (precision, lenght, scale);

Fixed possible duplicate key error using bulk operations for target agent.

Snapshot performance enhancements;

Palallel redo logs processing;

Fixes and enhacements;

Fixed wrong transaction sorting;

GSSD-781: Implemented support for minimal supplemental logging;

GSSD-774: Added missing RESETLOGS (Incarnation Change) handle;

GSSD-781: Fixed transaction logs sort problem and improved error notifications;

Database pooling technology;

GSSD-760: Fixes to Batch Update errors and Type modifier is not allowed for type 'text';

GSSD-717: Fixed issue with white spaces truncation in VARCHAR primary key columns;

Added backoff delay retry when connection restart is required;

Improved reader stop process;

Improved transaction logs processing;

Support for bulk loading both for CDC and snapshot operations;

Optimized metadata queries;

GSSD-714: Fixed LogMiner stop logic;

Agent compatibility upgrade to Gluesync 2.2;