Release 26.09

Sep 29, 2026 - 4 min read

Technical Release Details: https://github.com/esnet/StackV/releases/tag/26.09

Release 26.09 is our Q3 quarterly release, introducing integrated Project tracking, per-task access controls, and a large expansion of IRI and WAN Data Fabric orchestration. And as this release is part of our year-of-production sprint, we naturally also have major improvements to orchestrator model stability and observability.

⚠️ Breaking Changes

Task endpoints now enforce individual task access. Reading a task requires READ access, updating its status or message requires WRITE access, and archiving requires OWNER access. API clients that previously relied on broad access to arbitrary tasks may now receive 401, 403, or 404 responses.
- To remedy this, you may need to adjust the task’s tag or establish access control policies as needed.

✨ New Features

Projects are now a first-class concept in SENSE-O. Administrators can record project definitions and associate them with users and groups, which will then be associated with any licenses or instances they create. A new project association mode controls how those associations are applied along a new set of /project API endpoints.

Project Management Dialog

Task Access Controls bring the same ownership model used elsewhere in SENSE-O to Tasks. Owners can delegate read, write, or owner access to other users directly from the task UI, and task rows now include a quick link to navigate to their parent.

Task Access Controls

IRI and WAN Data Fabric orchestration has been significantly expanded:
- L3VPN support for the WAN Data Fabric, including optional BGP MD5 authentication keys, also available as an IRI Facility Space network parameter.
- L3 routing over facility interconnects, with local and remote addressing defined from the facility’s perspective.
- A new facility endpoint type for WAN Data Fabric connections, and improved subnet resolution for L2VPN and job placement.
- New IRI facility routing and compute switching services.
- Support for the IRI v2 API, with continued support for v1.

Observability improvements across the portal and backend:
- Instances now have clearer observability and audit tracing in the details view.
- System Health now reports more comprehensive monitoring information, along with diagnostic model metrics.
- The Driver table now shows overall health metrics across the board.
- Model-cache resets are faster, and the cache layer is more resilient under unstable conditions.

Minor Additions

The orchestrator container now ships with a built-in healthcheck against the version endpoint, and our Docker Compose templates have been updated to match. Operators maintaining their own Compose or Helm definitions should adopt the new check.
The JVM now writes a heap dump and exits on an OutOfMemoryError instead of stalling, keeps a rotating GC log in the server log directory, and warns at startup if JAVA_MEMORY is set below the recommended floor.
Container images now include SBOMs and better multi-platform support.

🐛 Bug Fixes

Fixed HTTP client starvation when dozens of authentication challenges ran simultaneously.
Fixed address pools allocating subnets outside their configured range.
Fixed several model-cache miss and lock issues that could stall orchestration.
Fixed slow portal loading and redundant requests when Keycloak is under load.
Fixed a client leak in the user service.
Fixed a stale layout state that could break the visualization when nodes were missing.
Fixed soft-capped QoS not always releasing bandwidth.
Timekeeping and date rendering are now centralized across SENSE-P for consistent display.

🚧 Known Issues

Project synchronization from external sources is not yet available, but is being actively developed.