Evenfire is an intelligence layer platform for running LLM agents that take real actions, on your infrastructure
TypeScript
/**
* E7.1–E7.5: Recipe lifecycle E2E tests (Phase 7).
*
* Validates the full WorkflowRecipe state machine transitions in minikube:
* - Create recipe reaches deploying phase
* - Status subresource updates correctly
* - Delete cleans up all child resources
* - PVC retention after recipe delete
*
* Prerequisites: Run scripts/minikube-setup.sh before these tests.
* These tests run AFTER delegation.test.ts (sequential mode).
*/
import { afterAll, beforeAll, describe, expect, it } from 'vitest'
import {
MCP_SERVER_NAMESPACE,
RECIPE_NAMESPACE,
SANDBOX_NAMESPACE,
kubectl,
kubectlJson,
sleep,
waitForResource,
} from './helpers'
const FILE_BROWSER_RECIPE = 'mcp-file-browser'
const FILE_BROWSER_FILE = `${__dirname}/../../samples/mcp-file-browser.yaml`
const BATCH_RECIPE = 'mcp-batch-processor'
const BATCH_FILE = `${__dirname}/../../samples/mcp-batch-processor.yaml`
beforeAll(async () => {
// Clean up leftovers from previous runs
for (const name of [FILE_BROWSER_RECIPE, BATCH_RECIPE]) {
try {
kubectl(
`delete workflowrecipe ${name} -n ${RECIPE_NAMESPACE} --ignore-not-found --timeout=20s`
)
} catch {
/* ignore */
}
}
// Clean up PVCs from previous runs
try {
kubectl(`delete pvc data-storage -n ${MCP_SERVER_NAMESPACE} --ignore-not-found`)
} catch {
/* ignore */
}
await sleep(3_000)
})
afterAll(() => {
for (const name of [FILE_BROWSER_RECIPE, BATCH_RECIPE]) {
try {
kubectl(`delete workflowrecipe ${name} -n ${RECIPE_NAMESPACE} --ignore-not-found`)
} catch {
/* ignore */
}
}
try {
kubectl(`delete pvc data-storage -n ${MCP_SERVER_NAMESPACE} --ignore-not-found`)
} catch {
/* ignore */
}
})
describe('Recipe Lifecycle E2E', () => {
// E7.1: Apply recipe → reaches deploying
it('E7.1 — Apply file-browser recipe creates Deployment', async () => {
const result = kubectl(`apply -f ${FILE_BROWSER_FILE}`)
expect(result).toMatch(/created|configured/)
// Wait for Deployment to appear
// Pre-deploy handshake (Option C) adds up to 30s before Deployment creation
await waitForResource(
`deploy -l clerum.io/recipe=${FILE_BROWSER_RECIPE}`,
MCP_SERVER_NAMESPACE,
{
shouldExist: true,
timeoutMs: 60_000,
}
)
const deploy = kubectlJson<{
items: Array<{ metadata: { name: string } }>
}>(`get deploy -l clerum.io/recipe=${FILE_BROWSER_RECIPE} -n ${MCP_SERVER_NAMESPACE}`)
expect(deploy.items.length).toBeGreaterThanOrEqual(1)
expect(deploy.items[0].metadata.name).toBe('file-browser')
})
// E7.2: PVC is created for file-browser recipe
it('E7.2 — PVC is created for file-browser recipe', async () => {
await waitForResource(`pvc data-storage`, MCP_SERVER_NAMESPACE, {
shouldExist: true,
timeoutMs: 15_000,
})
const pvc = kubectlJson<{
metadata: { labels: Record<string, string>; ownerReferences?: unknown[] }
spec: { resources: { requests: { storage: string } } }
}>(`get pvc data-storage -n ${MCP_SERVER_NAMESPACE}`)
expect(pvc.metadata.labels['clerum.io/recipe']).toBe(FILE_BROWSER_RECIPE)
// PVC must NOT have ownerReferences (Risk 3.7: data retention)
expect(pvc.metadata.ownerReferences).toBeUndefined()
expect(pvc.spec.resources.requests.storage).toBe('5Gi')
})
// E7.3: Batch processor Job is created (non-MCP → sandbox-recipes via namespace splitting)
it('E7.3 — Batch processor creates K8s Job with backoffLimit', async () => {
const result = kubectl(`apply -f ${BATCH_FILE}`)
expect(result).toMatch(/created|configured/)
await waitForResource(`job -l clerum.io/recipe=${BATCH_RECIPE}`, SANDBOX_NAMESPACE, {
shouldExist: true,
timeoutMs: 30_000,
})
const job = kubectlJson<{
items: Array<{
metadata: { name: string; namespace: string }
spec: { backoffLimit: number }
}>
}>(`get job -l clerum.io/recipe=${BATCH_RECIPE} -n ${SANDBOX_NAMESPACE}`)
expect(job.items.length).toBe(1)
expect(job.items[0].metadata.name).toBe('batch-processor')
expect(job.items[0].metadata.namespace).toBe('sandbox-recipes')
expect(job.items[0].spec.backoffLimit).toBe(2)
})
// E7.4: Delete recipe removes workloads (Job was in sandbox-recipes)
it('E7.4 — Delete batch-processor removes Job', async () => {
kubectl(`delete workflowrecipe ${BATCH_RECIPE} -n ${RECIPE_NAMESPACE}`)
await waitForResource(`job -l clerum.io/recipe=${BATCH_RECIPE}`, SANDBOX_NAMESPACE, {
shouldExist: false,
timeoutMs: 30_000,
})
})
// E7.5: PVC retained after recipe delete (CRITICAL)
it('E7.5 — PVC retained after file-browser recipe delete', async () => {
kubectl(`delete workflowrecipe ${FILE_BROWSER_RECIPE} -n ${RECIPE_NAMESPACE}`)
// Wait for Deployment to be deleted
await waitForResource(
`deploy -l clerum.io/recipe=${FILE_BROWSER_RECIPE}`,
MCP_SERVER_NAMESPACE,
{
shouldExist: false,
timeoutMs: 30_000,
}
)
// PVC must still exist (data retention)
const pvcExists = (() => {
try {
kubectl(`get pvc data-storage -n ${MCP_SERVER_NAMESPACE}`)
return true
} catch {
return false
}
})()
expect(pvcExists).toBe(true)
})
})