Roadmap

Project Roadmap

Enhancements in upcoming releases without a schedule

These improvements could happen at ANY time in any of the releases. We’d welcome input on additional features that people would like to see from the project!

• Make dynamic accounts the default account-management experience. The add/remove accounts UI landed in 2026.3.0; we still want it enabled by default going forward.
• Replace the “kubectl” based execution of kubernetes operations to native kubernetes SDK operations.
• AWS Lambda to async for invocations. Currently lambda operations (primarily invocations) are “SYNCHRONOUS” calls. The plan is to change these to an ASYNC which captures the execution from AWS & polls for completion instead of keeping the network connection open. This is critical for long running operations.
• CI: Add more CI providers to enable the UI to show build information. GitLab CI trigger/cancel support landed in 2026.3.0; more providers are welcome.
• Bring in changes from forks. There’s a few forks that have nice features we could bring in at anytime. Azure fixes and some UI tweaks.
• Move igor to use SQL instead of redis
• Move rosco to use SQL instead of redis

Recently completed: 2026.3.0

The primary goal of 2026.3.0 was removing and upgrading core libraries, and it shipped a large batch of deprecations along with it. See the full 2026.3.0 changelog for details. Highlights:

• Angular fully removed from Deck in favor of React
• Halyard and Edda removed from the codebase
• SAML configuration moved to native Spring configuration
• Front50 non-SQL storage (S3/GCS/etc.), Redis-backed Orca execution storage, and Titus are all marked deprecated for removal in 2027.0.0
• GCE migrated from the Compute beta API to stable v1
• Spin CLI API token support, a UI for adding/removing accounts, and a native Spinnaker MCP server all shipped
• GitHub App authentication for git/repo and github/file artifact accounts
• An alpha pub/sub-based cache scheduler (cats.pubsub) as a first step away from the Redis caching scheduler
• The AWS provider moved substantially onto the SDK v2 (EC2/ASG, ECS, SNS/SQS/SWF/CloudFormation, Lambda)

2026.4.0

Java 25, Spring Boot 4, and a set of security fixes have already landed on main ahead of this release; see the next release preview for the running list as it’s cut. Still in flight for this release or shortly after:

• Virtual thread support for better performance
• Merge deck-kayenta into deck. A separate UI for kayenta at this point doesn’t make sense, so collapse it down to simplify the builds and project
• Better pub/sub integration. A multi-phase plan is underway to add broadcast (fan-out) and single-delivery (competing-consumer) semantics to kork-pubsub, backed by Redis Streams and native Redis Pub/Sub, so account-change notifications and eventually operations no longer rely on polling. Phases 0-2 are up as draft PRs ( #8029 , #8030 , #8031 ); SNS/SQS broadcast support and the clouddriver account-refresh wiring that motivates all of this are still to come.
• GCP “direct” operations: a proposed V2 for the GCP provider that exposes more of the SDK’s available options directly by API type, rather than the current atomic, heavily-masked operations — closer to how the Kubernetes provider works today.

Cleanup of past deprecations:

• Remove S3 storage from Front50 (Front50 becomes SQL-only)
• Keep S3 (and the new SQL option, see below) storage for the artifact/entity store used by Orca and Clouddriver — that’s a separate concern from Front50’s own storage
• Finish removing AWS SDK v1 support — the last real dependency path (awsobjectmapper) was removed ahead of this release; a bridging shim for third-party plugins that still expect v1 types may remain longer

In discussion, not yet scheduled

These are active proposals and in-progress PRs without a committed release. Feedback and review are welcome on all of them.

• Fiat removal. There’s a large, actively-discussed draft PR to replace Fiat with token-based service-to-service authorization, decoupling permission checks behind a pluggable policy-decision-point seam. Fiat is known to have scale issues with large numbers of accounts/projects and roles, and is another service to operate. Reviewers have asked for a proper RFC and a staged rollout (decouple first, remove last) before this lands, so it’s not officially planned for a specific release — but plan long-term on Fiat eventually going away.
• Artifact account permissions. A draft PR would let artifact accounts (Helm, git/repo, github/file, etc.) carry READ/WRITE role permissions like cluster and CI accounts already can, so the UI can filter artifact lists by what a user is authorized to see ( issue #6592 ). It’s intentionally waiting on the Fiat-removal discussion above to settle, since it would need rework depending on that outcome.
• Spectator to Micrometer. An active PR removes Netflix-specific Spectator from kork in favor of the standard Micrometer abstraction, which plays much better with off-the-shelf observability tooling (including OpenTelemetry auto-instrumentation).
• Remote runner architecture. An early RFC-stage PR proposes a more modern, push-based execution model to eventually replace how Rosco bakes and how Kubernetes “Run Job” operations are handled, aimed at safer remote/distributed execution (motivated partly by the helmfile security hardening in 2026.4.0) instead of today’s polling model.
• Standardized canary query template. A PR in progress collapses Kayenta’s per-provider query UIs (Prometheus in particular has been a persistent source of user confusion) into a single template field across all canary providers, with backward compatibility for existing custom-templated configs.
• New cloud providers. Early-stage work on a Proxmox provider (caching agents already pulling VM data) and a companion HAProxy load-balancer provider for on-prem deployments alongside it.