Files
paperclip/packages/plugins/sandbox-providers/kubernetes/src/lease-lifecycle.ts
T
Jannes Stubbemann 05ab45225a feat(plugin-kubernetes): self-hostable Kubernetes sandbox provider (stage 1/3: plugin package) (#5790)
## Thinking Path

> - Paperclip is the open source app people use to manage AI agents for
work
> - Sandbox providers are the seam that lets agent runs execute in
isolated environments; today the only first-party remote provider is
Daytona, a hosted third-party service
> - Self-hosters running Paperclip on their own infrastructure (often
Kubernetes already) have no first-party way to run agent sandboxes on a
cluster they control
> - That gap matters for teams with data-residency, sovereignty, or cost
constraints who cannot or will not send workloads to a hosted sandbox
service
> - This pull request adds a Kubernetes sandbox-provider plugin as a
standalone, workspace-excluded package: it implements every
SandboxProvider hook the Daytona provider does, on infrastructure the
operator owns
> - The benefit is that any Paperclip deployment with a Kubernetes
cluster gets multi-tenant, network-isolated, quota-bounded agent
sandboxes with zero new external dependencies

## Linked Issues or Issue Description

No existing issue. Following the feature template:

- **Problem:** Paperclip's remote sandbox execution requires a hosted
third-party provider. Self-hosters cannot run agent sandboxes on their
own Kubernetes clusters with a first-party provider.
- **Proposed solution:** A `@paperclipai/plugin-kubernetes`
sandbox-provider plugin with two backends: long-lived sandboxes via the
[kubernetes-sigs/agent-sandbox](https://github.com/kubernetes-sigs/agent-sandbox)
CRD (multi-command exec, adapter-install pattern) and one-shot
`batch/v1` Jobs (stable APIs only, no extra controllers).
- **Alternatives considered:** Driving kubectl from a generic shell
provider (no lifecycle/lease semantics), or requiring a hosted provider
(exactly the constraint this removes).

## What Changed

This is **stage 1 of 3** of a staged contribution (direction agreed with
maintainers): the plugin package alone. Stage 2 (server integration:
lease params, provider registration) and stage 3 (agent runtime images +
CI) are companion PRs that will be cross-linked from a comment here.

- New package `packages/plugins/sandbox-providers/kubernetes`
(workspace-excluded, like the path already carved out in
`pnpm-workspace.yaml`): src, unit + kind integration tests, operator
prerequisite manifests, README, smoke-test guide
- Implements the full SandboxProvider hook surface the Daytona provider
implements: `validateConfig`, `probe`, `acquireLease`, `resumeLease`,
`releaseLease`, `destroyLease`, `realizeWorkspace`, `execute`
- Two backends: `sandbox-cr` (default; long-lived pod via the
agent-sandbox `Sandbox` CR, supports multi-command exec) and `job`
(one-shot `batch/v1` Job; nothing beyond k8s 1.27+ required)
- Per-run adapter resolution: one environment serves mixed harnesses;
the per-run `adapterType` hint is read through a local optional type
extension, so the plugin typechecks and builds against the current
plugin SDK and simply falls back to the environment's configured default
adapter until stage 2 lands
- Exec-env wrapping: the Kubernetes exec API carries no environment, so
commands are wrapped to receive the run's env
- Fast-upload interception for workspace realization, scoped per lease
- Per-tenant isolation: derived namespace per company, RBAC,
ResourceQuota, restricted-PSS pod security (runAsNonRoot, drop ALL,
seccomp RuntimeDefault, no SA token automount)
- Network egress policy in two flavors: native `NetworkPolicy` and
`CiliumNetworkPolicy` (FQDN allowlists)
- Image allowlist with glob matching, registry override, and per-run
image override validation
- Per-run Kubernetes Secrets carrying agent credentials, ownerRef'd to
the Job or Sandbox CR for cascade GC

## Verification

- Standalone build, exactly as the README documents:
  ```bash
  cd packages/plugins/sandbox-providers/kubernetes
  pnpm install --ignore-workspace
  pnpm test        # 147 unit tests, 17 files, all green
  pnpm typecheck   # clean against the in-repo plugin SDK on master
pnpm build # dist/ emitted, manifest + worker entrypoints present
  ```
- A kind-cluster end-to-end integration test is included
(`RUN_K8S_INTEGRATION_TESTS=1 pnpm test
test/integration/end-to-end-run.test.ts`)
- Beyond CI: this provider has been verified in a production
multi-tenant deployment against five harnesses (opencode, pi, codex,
gemini, claude code) with real billed runs

## Risks

- **Zero behavior change for any existing deployment.** The package is
workspace-excluded; nothing in the server imports or loads it until
stage 2's integration lands. No existing code paths are touched.
- The default `sandbox-cr` backend depends on an alpha CRD
(`agents.x-k8s.io/v1alpha1`); the README flags this and the `job`
backend uses only stable APIs as a fallback.
- Risk surface is confined to deployments that explicitly install and
configure the plugin.
- The default runtime images (`ghcr.io/paperclipai/agent-runtime-*`) are
published by the stage 3 companion PR (#7934); until that lands,
deployments must point `runtimeImage` at their own images.

## Model Used

Claude Opus 4.8 (1M context), extended thinking, with tool use (Claude
Code).

## Checklist

- [x] I have included a thinking path that traces from project context
to this change
- [x] I have specified the model used (with version and capability
details)
- [x] I have checked ROADMAP.md and confirmed this PR does not duplicate
planned core work
- [x] I have searched GitHub for duplicate or related PRs and linked
them above
- [x] I have either (a) linked existing issues with `Fixes: #` / `Closes
#` / `Refs #` OR (b) described the issue in-PR following the relevant
issue template
- [x] I have run tests locally and they pass
- [x] I have added or updated tests where applicable
- [ ] If this change affects the UI, I have included before/after
screenshots (no UI changes)
- [x] I have updated relevant documentation to reflect my changes
- [x] I have considered and documented any risks above
- [ ] All Paperclip CI gates are green (pending this push)
- [ ] Greptile is 5/5 with no open P2s, recommendations, or follow-ups
- [x] I will address all Greptile and reviewer comments before
requesting merge

---------

Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-06-10 21:07:00 -07:00

194 lines
6.8 KiB
TypeScript

/**
* Resume + destroy lifecycle helpers for Kubernetes sandbox leases.
*
* Resume semantics: a lease is resumable only while its workload resource
* (Sandbox CR or Job) still exists and its pod is Running/Ready (or becomes
* Ready within a short bounded wait). Unlike Daytona — where a stopped sandbox
* can be started again by ID — Kubernetes pods are NOT restartable: once the
* pod backing a lease is gone or terminally failed, the lease can never be
* revived in place. That asymmetry is intentional; the plugin reports the
* lease as expired and the server falls back to a fresh acquireLease, which
* provisions a new pod.
*
* Destroy semantics: the forced cleanup path. Deletes every resource
* acquireLease created (Sandbox CR / Job, its pod, the per-run Secret),
* treating 404s as success so it is idempotent and safe to call against
* half-deleted leases.
*/
import type { KubeClients } from "./kube-client.js";
import { deleteJob, findPodForJob, getJobStatus } from "./job-orchestrator.js";
import {
deleteSandboxCr,
findPodForSandbox,
waitForSandboxReady,
} from "./sandbox-cr-orchestrator.js";
/** True when a Kubernetes API error means "resource not found" (HTTP 404). */
export function isKubeNotFoundError(err: unknown): boolean {
const code = (err as { code?: number; statusCode?: number }).code
?? (err as { code?: number; statusCode?: number }).statusCode;
return code === 404;
}
async function ignoreNotFound(promise: Promise<unknown>): Promise<void> {
try {
await promise;
} catch (err) {
if (!isKubeNotFoundError(err)) throw err;
}
}
export type ResumeCheckResult =
| { resumable: true; podName: string | null; phase: "Pending" | "Running" }
| { resumable: false; reason: string };
export interface ResumeCheckInput {
namespace: string;
/** Workload resource name (Sandbox CR name or Job name) == providerLeaseId. */
name: string;
backend: "sandbox-cr" | "job";
/** Bounded wait for an existing Sandbox pod to report Ready. */
readyTimeoutMs?: number;
pollMs?: number;
}
/**
* Check whether the workload behind a lease is still alive and exec-able.
* Returns `resumable: false` (never throws "expected" states) when the
* resource is gone (404), terminally failed, terminating, or doesn't become
* Ready within the bounded wait — all of which mean the caller should fall
* back to a fresh acquireLease.
*/
export async function checkLeaseResumable(
clients: KubeClients,
input: ResumeCheckInput,
): Promise<ResumeCheckResult> {
if (input.backend === "sandbox-cr") {
// Bounded wait for the Sandbox to report Ready. waitForSandboxReady fails
// fast on Failed/Terminating; a timeout means the pod never came up. None
// of those states are resumable — k8s pods cannot be restarted in place.
try {
await waitForSandboxReady(clients, input.namespace, input.name, {
timeoutMs: input.readyTimeoutMs ?? 30_000,
pollMs: input.pollMs ?? 1_000,
});
} catch (err) {
if (isKubeNotFoundError(err)) {
return { resumable: false, reason: "Sandbox CR no longer exists" };
}
return {
resumable: false,
reason: err instanceof Error ? err.message : String(err),
};
}
let podName: string | null;
try {
podName = await findPodForSandbox(clients, input.namespace, input.name);
} catch (err) {
// CR deleted between the readiness check and the pod lookup.
if (isKubeNotFoundError(err)) {
return { resumable: false, reason: "Sandbox CR no longer exists" };
}
throw err;
}
if (!podName) {
return {
resumable: false,
reason: "Sandbox is Ready but no backing pod was found",
};
}
// Confirm the pod itself is Running and not being torn down — the CR
// status can lag pod deletion.
let pod: { metadata?: { deletionTimestamp?: unknown }; status?: { phase?: string } };
try {
pod = await clients.core.readNamespacedPod({
namespace: input.namespace,
name: podName,
}) as typeof pod;
} catch (err) {
if (isKubeNotFoundError(err)) {
return { resumable: false, reason: `Pod ${podName} no longer exists` };
}
throw err;
}
const podPhase = pod.status?.phase;
const terminating = Boolean(pod.metadata?.deletionTimestamp);
if (podPhase !== "Running" || terminating) {
return {
resumable: false,
reason: `Pod ${podName} is ${terminating ? "terminating" : podPhase ?? "in an unknown phase"}`,
};
}
return { resumable: true, podName, phase: "Running" };
}
// ── Job backend ───────────────────────────────────────────────────────────
let status;
try {
status = await getJobStatus(clients, input.namespace, input.name);
} catch (err) {
if (isKubeNotFoundError(err)) {
return { resumable: false, reason: "Job no longer exists" };
}
throw err;
}
if (status.phase === "Succeeded" || status.phase === "Failed") {
// Terminal Jobs cannot be re-run in place.
return { resumable: false, reason: `Job is ${status.phase}` };
}
// Pending/Running Jobs are resumable: execute() waits for completion
// itself, so a not-yet-scheduled pod (podName null) is fine here.
const podName = await findPodForJob(clients, input.namespace, input.name);
return {
resumable: true,
podName,
phase: status.phase === "Running" ? "Running" : "Pending",
};
}
export interface DestroyLeaseInput {
namespace: string;
/** Workload resource name (Sandbox CR name or Job name) == providerLeaseId. */
name: string;
backend: "sandbox-cr" | "job";
podName: string | null;
secretName: string | null;
}
/**
* Forcibly delete every resource acquireLease created for this lease.
* Workload first (its deletion cascades to the pod and, via ownerReferences,
* the per-run Secret in the normal case); then the pod and Secret explicitly
* so a wedged controller or broken ownerRef cannot strand them. Every delete
* treats 404 as success — destroy is idempotent.
*/
export async function destroyLeaseResources(
clients: KubeClients,
input: DestroyLeaseInput,
): Promise<void> {
if (input.backend === "sandbox-cr") {
await ignoreNotFound(deleteSandboxCr(clients, input.namespace, input.name));
} else {
await ignoreNotFound(deleteJob(clients, input.namespace, input.name));
}
if (input.podName) {
await ignoreNotFound(
clients.core.deleteNamespacedPod({
namespace: input.namespace,
name: input.podName,
}),
);
}
if (input.secretName) {
await ignoreNotFound(
clients.core.deleteNamespacedSecret({
namespace: input.namespace,
name: input.secretName,
}),
);
}
}