Azure Cosmos DB agent changelog
Release notes
2.3.2.2
Released: November 13, 2026
-
A snapshot continued after a pause or a restart resumes from the documents its checkpoint counts instead of reading the container again from the start, so documents are no longer copied twice and the progress stays within 100%;
2.3.2.0
Released: November 10, 2026
-
Snapshot: partitions update the shared checkpoint and hand their page on in that order, instead of a page counting another partition’s rows being stored first, which made a stop or a write error resume past documents the target never got, or closed the snapshot before the other pages arrived;
2.3.1.2
Released: October 28, 2026
-
Snapshot: a partitioned snapshot reads the documents whose split field is null, missing or of another type than the bounds, and the partitions can be computed on a field that holds null values;
2.3.1.1
Released: October 27, 2026
-
Metrics: a page’s read time no longer includes the wait for the target to consume the previous page;
2.3.0.1
Released: October 15, 2026
-
Breaking change: change data capture now reads the Azure Cosmos DB change feed in all versions and deletes mode. Inserts, updates and deletes (including TTL expiry deletes) are captured natively and every intermediate change is delivered in order;
-
Breaking change: soft-delete support is removed. The Soft delete document delete field setting is no longer available (and it is ignored if present in a restored configuration), and the Azure Function that marked documents as deleted is no longer needed. Documents that still carry the old
deleted: trueflag are replicated as normal updates; -
The agent requires the Azure Cosmos DB for NoSQL API, continuous backup (7-day or 30-day) and the All versions and deletes change feed mode feature enabled on the account. See Upgrading from Gluesync 2.2.
-
Change data capture delivers every change at least once: the change feed position is saved only after the target confirms the changes, and it is kept across pipeline restarts. See Delivery guarantees;
-
The change feed is started before the snapshot, so no change made while the snapshot runs is lost;
-
The agent stops with a clear error when the account does not meet the change feed requirements. See Troubleshooting;
-
The lease container is always used with the
/idpartition key. The Lease container partition key field setting is removed (and it is ignored if present in a restored configuration); -
Fixed: explicit
nullvalues of a document were replicated as0,falseor the textnull; -
Fixed: decimal numbers lost their fractional part;
-
Fixed: elements of arrays were not replicated as plain values;
-
As a target, a delete of a document that no longer exists is now ignored instead of failing;
-
The source account must not use partition merge, and the client connects in gateway mode (HTTPS on port 443);
-
An entity must be keyed on
idplus the partition key fields for its deletes to replicate; otherwise they are skipped and reported; -
Resuming works only within the continuous backup retention window;
-
As a target, deletes work in containers partitioned by a field other than
id; -
Discovery lists the containers, scopes, and types of every page;
-
Stopping an entity while its change feed is starting no longer reports a fetch error;
-
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; -
Azure Cosmos DB Java SDK upgraded to 4.83.0;