Search docs.molo17.com

    MySQL CDC logo MySQL agent changelog

    Release notes

    2.3.2.2

    Released: November 13, 2026

    • GSSD-1487: On an entity keyed by nullable columns, the row by row writer matches a NULL key with IS NULL, so an update no longer inserts a second copy of the row and a delete no longer leaves it in place;

    • GSSD-1487: On an entity keyed by nullable columns, an insert of a row the target already holds with a NULL in its key is looked up first and handled by the insert conflict strategy (UPSERT, SKIP, FAIL) instead of being written as a second copy after a resume, a Core Hub failover or a snapshot overlapping the changes;

    2.3.2.1

    Released: November 11, 2026

    • A region of the cache whose flush to the disk fails stays dirty and is flushed again, instead of being forgotten after that one failed attempt;

    2.3.2.0

    Released: November 10, 2026

    • CDC: the binlog reader starts when a snapshot starts, or restarts from its checkpoint with the entity’s tables, instead of reading a table only after its snapshot, which lost every change made to the table while it was copied (a pipeline started under load, or an entity added to a running pipeline);

    • CDC: a restart of the reader rolls back what it had staged of the transaction it was in the middle of, instead of committing that part when the transaction was read again and delivering those rows twice;

    • CDC: a checkpoint is taken as purged when its file is not among those SHOW BINARY LOGS lists, instead of comparing file names as text, which past binlog.999999 took a file still on the server for purged and jumped to the newest position, skipping everything between; the jump is now logged;

    • GSSD-1487: bulk load reads its CSV as UTF-8 (CHARACTER SET utf8mb4), instead of the default character set of the database, which stored Ù as Ù in a utf8mb4 column on a latin1 database;

    • GSSD-1487: bulk load writes a boolean into a BIT column as hex, instead of true, which arrived as 0, so every BIT set to 1 on a SQL Server source was written as 0;

    • GSSD-1487: bulk load writes a negative, decimal or floating-point number into a BIT column as its bits, instead of -128 into BIT(8) storing 255 and 10 storing 16;

    • Metrics: the source lag of a quiet pipeline no longer shows a date tens of thousands of years ahead, the read checkpoint age no longer shows about 55 years, and the cache lag is measured from the source event time;

    • The GS_TRANSACTION_TIMESTAMP technical field carries the source event time in epoch milliseconds instead of the agent’s nanosecond clock;

    • GSSD-1346: Target: bulk delete and old-values fill match nullable keys with <=>, so the key stays an index key: one staged key matching nothing reads 0 index rows instead of 402 on delete and 1,998 on fill;

    • GSSD-1494: Enable Recursion Protection takes effect: with it on, every transaction the target commits carries a marker, a row of <Gluesync schema>.gs_recursion_marker inserted and deleted again, and the binlog source discards the transactions that carry it, so in a bidirectional setup the rows Gluesync writes no longer come back. Before, nothing read the setting and every row a Gluesync target wrote came back through the binlog. The setup creates the marker table on InnoDB, refuses a server whose binlog_do_db or binlog_ignore_db leave the marker’s schema out of the binary log and an existing marker table that is not on a transactional engine, and names in a warning the target tables on a non-transactional engine;

    2.3.1.4

    Released: November 2, 2026

    • GSSD-1478 aside, transactions written with binlog_transaction_compression are captured, instead of the payload event being dropped with no change captured, no error raised and the read checkpoint never moving;

    2.3.1.2

    Released: October 28, 2026

    • CDC: the binlog position is saved as soon as CDC first starts. Before, it was saved only after a transaction on a captured table, so restarting the pipeline before then resumed from the current binlog head and the changes written in between never reached the target;

    • CDC: binlog ingestion rises from about 2.2k to about 11k single-row commits per second; the read checkpoint is written at most every 200 ms instead of on every commit, and is still flushed on reconnect and on stop;

    • CDC: the fractional seconds of a DATETIME, which were dropped, and the bytes of a BIT column are kept;

    • GSSD-1445: with Trust server certificate off, the uploaded trust store verifies the server and the key store is sent as the client certificate. A trust store alone, a PEM trust store such as Cloud SQL’s server-ca.pem and the binlog reader no longer fail the TLS handshake, and store passwords containing &, =, # or ; no longer break the connection;

    • Skipped binlog events (other databases, tables not under CDC, QUERY, GTID and FORMAT_DESCRIPTION events) are logged at DEBUG instead of WARN;

    • Snapshot: a time in the hour the agent’s time zone skips at the daylight saving change is replicated as the wall clock the source holds instead of an hour later, and a DATETIME in that hour no longer fails the read with HOUR_OF_DAY: 2 → 3;

    2.3.1.1

    Released: October 27, 2026

    • CDC: the local cache is drained up to its head in one poll instead of one message per poll, so a backlog reaches the target faster;

    • GSSD-1456: resetting the entities deletes the shared read checkpoint, so the reader no longer resumes from the old position;

    • Target: an object field from a NoSQL source is written as its JSON text instead of {name=Ada, age=36}, and an array that holds a map or a nested array is written as a JSON array instead of [Ljava.lang.Object;@…​;

    2.3.0.1

    Released: October 15, 2026

    • Bulk operations: an insert and a delete, or two updates, of a row keyed by binary or array values in one batch are merged instead of leaving the row on the target or failing with a duplicate key;

    • The agent hands its pooled database back when its pipeline is deleted or detached and when its initialization fails. Before, the next pipeline writing to the same database could fail in a loop with NoClassDefFoundError;

    2.3.0

    Released: October 13, 2026

    • Agent compatibility upgrade to Gluesync 2.3;

    • The source cache saves the corrected read position after a recovery scan, so a restart no longer starts reading mid-record or discards records written after a torn record;

    • Discovery returns schemas and tables in alphabetical order, so Query Studio pages no longer skip or repeat objects;