10.0.8: Simba on a schedule, a CDC wave, and streaming nested mappings

10.0.8 shipped on October 22, two weeks after 10.0.0. It is the first feature release on the new platform, and most of it is about what happens when nobody is watching: an agent that runs on its own, CDC that tells you where it is struggling, and workflows that recover from a failed step.

An agent that runs on its own

Run Simba Prompt is a new flow type. It takes a saved task for Simba, our AI agent, and runs it unattended like any other flow: on a schedule, from an Etlworks webhook, when a listener receives an event from another system, or as a step in a nested flow or in Composer, our visual workflow builder. The report lands in execution history and can be emailed as formatted HTML.

The fastest way in is the calendar icon in the Simba chat panel. It turns a conversation you already had into a reusable prompt with a schedule, in one dialog.

An unattended run cannot stop to ask for approval, so permissions are granted on the flow. Allow changes and Allow destructive and operational actions are both off by default, and every grant writes an audit event. In a workflow, Simba can return a typed true/false answer, and Use Simba result as condition branches on it. Simba tasks run on the app, not on Integration Agents. Run Simba Prompt has the details, and the AI agent page has the overview.

A wave of CDC options

Every one of the seven CDC connectors gained something, and all of it is opt-in:

• SQL Server: Streaming Fetch Size bounds buffering during streaming reads.
• Oracle: compact event metadata, an Optimized Memory Transaction Buffer that releases rolled-back payloads earlier, and SCN-consistent chunk planning.
• PostgreSQL: failover-capable slots on 17 and read-only incremental snapshots on 13 and later.
• MongoDB: post-image document retrieval on 6.0 and later, and filtered incremental snapshots.
• IBM i: online incremental snapshots in bounded chunks while journal streaming continues.
• MySQL: tagged GTIDs and binlogs larger than 4 GiB.

All relational connectors can also keep CDC metadata on disk behind a bounded cache, for capture sets too large to hold in memory.

You can now see the pressure, too. Running Tasks, the view of everything executing right now, shows live queue occupancy, estimated bytes, backpressure, and destination flush timing for each CDC flow, and every CDC connection can write a latency summary to the log every five minutes. Monitor CDC buffer pressure and latency.

Streaming reaches nested mappings

10.0.0 streamed large JSON, XML, and HTTP responses. 10.0.8 extends that to nested mappings: eligible JSON, MongoDB, and queue sources map one outer record at a time, including across paginated API responses read with Stream JSON Response. Memory follows the size of one record, not the whole source. Per-record nested mapping.

Seeing what the engine did

Each transformation's statistics now include a runtime plan: streaming or buffered, bind variables, and the configured batch size next to the effective one, with the reason for any adjustment. The mapping editor predicts the same plan before you run.

Resource Usage, the view that ranks flows by the CPU and memory they consume, now keeps per-flow allocation and CPU history. Administrators can capture a heap dump and analyze it from the CLI. Resource Usage · Heap dumps.

Workflows that recover

Go to on exception lets a failed step in a nested flow or Composer workflow route to another step once its retries and exception handlers are exhausted. The failure stays in the record and the workflow keeps going. Go to on exception.

Smaller things

The Amazon S3 (SDK) connector can authenticate through the AWS default credential chain, so EKS Pod Identity, IRSA, and instance roles work. Column names with literal dots such as Column.1 stay one identifier when quoting is on. SFTP connections take a connect timeout.

On security, Log4j moves to 2.25.5 (CVE-2021-45046) and the Spring Framework is upgraded (CVE-2025-41249).

Upgrading

The CDC changes are opt-in and the defaults preserve existing behavior, but the new options need an updated engine or Integration Agent. The full release notes list every change and fix.