9.8.3: Execution trees, and a faster CDC snapshot
The Metrics tab can show the runtime tree of a nested execution, CDC snapshots get an experimental fast path, and CDC events can land straight in Parquet.
A release about finding out what happened, and about loading faster.
Hierarchical execution metrics. The Metrics tab of the Flow Statistics dashboard now has a Hierarchy view alongside the flat one. It draws the runtime tree of an execution: the main flow, nested flows, executed steps, loop iterations, retries, on-error handlers, and the transformations each flow produced, with an Execution Path on every row. Large loops collapse into expandable ranges, and conditional branches that were evaluated but never ran show up as Skipped. A failed step can be traced back to its parent flow and loop iteration without opening the log.
Fast CDC Snapshot. An experimental snapshot modifier that gives snapshot rows a more direct path to the CDC destination, cutting the per-row overhead that can make a snapshot slower than the equivalent ETL load. It covers full initial snapshots and blocking ad-hoc snapshots on SQL Server, Oracle, MySQL, PostgreSQL, DB2 LUW, AS400 and IBM i, and modern MongoDB CDC. It is off by default, and we mean the word experimental: benchmark it before you turn it on for a pipeline that matters.
CDC to Parquet. Change events can be written directly to Parquet, so a change stream lands in a lake without a conversion step in between.
MCP protocol revision 2026-07-28. The MCP server is now dual-era. The same endpoint answers both the legacy initialize handshake and the stateless per-request form, which replaces the handshake with server/discover and carries the protocol version and client capabilities on every request. Existing clients keep working unchanged.
A fallback model for Simba, our AI agent. Configure a lower-tier model and a turn that fails on the primary, through a rate limit or an unavailable model, retries on the fallback automatically. It also keeps the agent answering after a monthly usage cap is reached.
Also in this release: wildcard COPY INTO combined with MERGE for Snowflake bulk load, autocomplete in the Macro Editor with a per-user preference under Profile, snapshot-only CDC flows that stop on their own when the snapshot finishes, multi-select in Etlworks Explorer, our built-in view of connections and files, editable save messages in artifact change history, MySQL 8.4 CDC through SHOW BINARY LOG STATUS, an Integration Agent installer with the WMIC dependency removed for current Windows builds, a code-signed update service, macros that synchronize across nodes, and parentheses accepted in Data Type Overrides so NUMBER(18,2) can be written as it is.