Kurtosis for Persistent, Scalable Devnets: A Vision & Roadmap

1 Overview

Kurtosis began as a developer‑focused tool to spin up reproducible, ephemeral environments. Over the last two years the community has stretched those limits, running long‑lived devnets and complex integration testbeds on top of Kurtosis. We propose an evolution that promotes Kurtosis from purely ephemeral to persistent devnets, and eventually, a production‑grade orchestration platform capable of hosting stateful, durable, multi‑tenant networks in the cloud.

The Bloctopus team, in collaboration with Kurtosis maintainers, is already experimenting with StatefulSets, advanced log exporting, dynamic persistent storage, and REST‑API stability fixes. We look forward to collaborating with the community to upstream future improvements into the mainline project. We invite the current maintainers, power‑users, and new contributors to collaborate on this roadmap so Kurtosis can serve both ephemeral and always‑on workloads.

2 Motivation

  1. Long‑running environments need state. Ephemeral Pods are routinely garbage‑collected by Kubernetes after crashes or node rotations, wiping validator keys and chain data.
  2. CI/CD & Chaos require weeks of uptime. Many projects run soak tests, live‑network simulations, or fuzzing campaigns that last days to weeks.
  3. Package explosion. Flagship blockchain packages for Ethereum, Optimism, and Polygon are growing in size and complexity. Keeping everything in a single enclave and package is no longer practical.
  4. Decentralization and autoscaling are essential. Currently, enclaves and networks are confined to a single cluster or infrastructure. Ideally, systems should enable decentralised deployments and automatic scaling of services based on real-time demand. Decentralization is particularly critical for larger-scale devnets, testnets, and persistent dev environment to integrate ci/cd.
  5. Community demand is growing. Multiple users (see issue #2611, https://github.com/kurtosis-tech/kurtosis/issues/2710, https://github.com/kurtosis-tech/kurtosis/issues/2451) are already requesting multi‑enclave coordination and longer‑lived stacks. Infrastructure teams still rely on the interactive CLI but also need Helm, GitOps and APIs for automated workflows. Machine-to-machine interactions should be treated as equally important as developer experience.

3 Kurtosis Success Stories

Package Owner Stars Highlights
ethereum‑package ethPandaOps 327 Multi‑client PoS testnet with MEV‑Boost and observability stack (github.com)
optimism‑package ethPandaOps 184 Rolls out full L2 stack incl. op‑node, op‑geth, L1 deployment & Grafana dashboards (github.com)
kurtosis‑cdk Polygon Labs 60 Modular Polygon CDK devnet (L1 + L2) with customizable components (github.com)
polkadot‑kurtosis‑package HugoByte 22 Multiple parachain topologies, explorer & Grafana (github.com)
cosmos‑package Bloctopus 10 Coreum & Cosmos‑Hub devnets, staking & IBC ready (github.com)
hyperlane‑package Abacus Works 10 Multi‑domain bridge devnet with validators & relayers (github.com)
layerzero-package Bloctopus 6 Full implementation of DVN and executor for any EVM network for multi-chain messaging

These packages demonstrate Kurtosis’ power but also expose its limitations for persistent devnets: operators often hack around storage, logging and multi‑enclave needs

4 What We Are Building

All the capabilities described below are being tested in our public fork (github.com/brianle1301/kurtosis). Feedback is welcome while we harden these features for upstream PRs.

Area Change Status
Stateful workloads Added StatefulSet support so any user service can opt‑in to ordered, sticky identity Pods. PoC in fork; internal soak‑testing - will expose user‑config options (github.com)
Durable storage Automatic PVC provisioning & auto‑scaling PoC in fork; internal testing with hard‑coded 60 % → 2× rule - aiming for dynamic & customizable scaling (github.com)
Stable REST API Fixed enclave‑scoped connectivity bug that blocked programmatic clients Fixed in fork (github.com)
K8s‑native deploy Decoupled Engine & workloads from CLI - direct kubectl apply now works Fixed in fork
**Restarts and
snapshotting** Allow restarting of enclaves and exporting state for quicker start-up WIP

5 Proposed Roadmap

5.1 Helm Charts & GitOps‑Friendly Manifests

Current limitation: Today, the kurtosis engine and packages are launched primarily via the developer‑focused CLI. While perfect for local iteration, invoking the CLI from a jump‑box or scripting kurtosis run … in CI makes it hard to follow GitOps best‑practices and to integrate with platform‑operator tooling.

Introducing first‑class Helm charts solves this by letting operators manage Kurtosis like any other Kubernetes application through declarative manifests that can be version‑controlled, templated and promoted through environments.