feat: add managed SSH execution spaces
Add managed SSH Hosts and projects with signed side-by-side GoodBuddy Agent and OpenCode Runtime installation, Ask and Execute workspace access, a Main-owned model bridge, detached Agent lifecycle, durable ACP output recovery, and project activation UI. Use protocol v2 with bounded FIFO frames over SSH instead of custom credit, priority, batching, or model-message chunking. Mismatched Host Agents now upgrade automatically, an exactly verified Node runtime is reused on-host, and local Portable builds can embed strictly verified Agent artifacts. Release note: 新增托管 SSH 远程项目，可自动安装或升级签名 Agent/OpenCode Runtime，并在 SSH 断线后继续 Agent 工作和同步未确认输出。
This commit is contained in:
mesalogo
@@ -0,0 +1,588 @@
|
||||
import { describe, expect, it, vi } from 'vitest'
|
||||
import {
|
||||
AGENT_PROTOCOL_VERSION,
|
||||
type AttachWelcome,
|
||||
type DaemonCapabilities
|
||||
} from '../../shared/agent-protocol'
|
||||
import type {
|
||||
SshConnectionLease,
|
||||
SshConnectionPoolTarget
|
||||
} from '../ssh/ssh-connection-pool'
|
||||
import { verifyAgentInstallationId } from '../ssh/ssh-agent-command'
|
||||
import type { AgentAttachTransport } from './agent-attach-transport'
|
||||
import type {
|
||||
AgentProtocolClient,
|
||||
AgentProtocolMethod
|
||||
} from './agent-protocol-client'
|
||||
import {
|
||||
ControllerStateStore,
|
||||
type ControllerStateFile,
|
||||
type PersistedControllerState
|
||||
} from './controller-state-store'
|
||||
import {
|
||||
RemoteAgentConnectionManager,
|
||||
type RemoteAgentSshPool
|
||||
} from './remote-agent-connection-manager'
|
||||
|
||||
class MemoryStateFile implements ControllerStateFile {
|
||||
value?: unknown
|
||||
|
||||
async read(): Promise<unknown | undefined> {
|
||||
return structuredClone(this.value)
|
||||
}
|
||||
|
||||
async write(value: PersistedControllerState): Promise<void> {
|
||||
this.value = structuredClone(value)
|
||||
}
|
||||
}
|
||||
|
||||
class FakeProtocolClient {
|
||||
readonly connectionId: string
|
||||
readonly generation: number
|
||||
capabilitiesGate?: Promise<void>
|
||||
resumeGate?: Promise<void>
|
||||
capabilities: DaemonCapabilities = {
|
||||
generation: 1,
|
||||
capabilities: [
|
||||
{ name: 'workspace/read', version: 1, critical: true }
|
||||
],
|
||||
runtimes: []
|
||||
}
|
||||
disposed = false
|
||||
readonly requests: Array<{ method: AgentProtocolMethod; params: unknown }> =
|
||||
[]
|
||||
#closeListeners = new Set<(error: Error) => void>()
|
||||
|
||||
constructor(index = 0) {
|
||||
this.connectionId = `connection-${index + 1}`
|
||||
this.generation = index + 2
|
||||
}
|
||||
|
||||
async request(
|
||||
method: AgentProtocolMethod,
|
||||
params: unknown
|
||||
): Promise<unknown> {
|
||||
this.requests.push({ method, params })
|
||||
if (method === 'agent/status') {
|
||||
return {
|
||||
state: 'ready',
|
||||
installationId: 'agent-v1',
|
||||
binaryDigest: `sha256:${'a'.repeat(64)}`,
|
||||
daemonBootId: 'boot-1',
|
||||
agentVersion: '1.0.0',
|
||||
protocol: AGENT_PROTOCOL_VERSION,
|
||||
platform: 'linux',
|
||||
architecture: 'x64',
|
||||
supervisor: 'detached-on-demand',
|
||||
remoteUserIdentity: 'user-1',
|
||||
draining: false
|
||||
}
|
||||
}
|
||||
if (method === 'agent/capabilities') {
|
||||
await this.capabilitiesGate
|
||||
return structuredClone(this.capabilities)
|
||||
}
|
||||
if (method === 'controller/resume') {
|
||||
await this.resumeGate
|
||||
const request = params as {
|
||||
daemonBootId: string
|
||||
capabilityGeneration: number
|
||||
}
|
||||
return {
|
||||
resumed: true,
|
||||
generation: this.generation,
|
||||
daemonBootId: request.daemonBootId,
|
||||
capabilityGeneration: request.capabilityGeneration,
|
||||
leaseDeadlineAt: '2030-01-01T00:00:00.000Z'
|
||||
}
|
||||
}
|
||||
throw new Error(`Unexpected method ${method}`)
|
||||
}
|
||||
|
||||
onClose(listener: (error: Error) => void): () => void {
|
||||
this.#closeListeners.add(listener)
|
||||
return () => this.#closeListeners.delete(listener)
|
||||
}
|
||||
|
||||
transportClose(): void {
|
||||
for (const listener of this.#closeListeners) {
|
||||
listener(new Error('transport closed'))
|
||||
}
|
||||
}
|
||||
|
||||
dispose(): void {
|
||||
this.disposed = true
|
||||
}
|
||||
}
|
||||
|
||||
function target(revision = 1): SshConnectionPoolTarget {
|
||||
return {
|
||||
host: {
|
||||
id: 'host-1',
|
||||
name: 'Builder',
|
||||
hostname: 'builder.example',
|
||||
port: 22,
|
||||
username: 'goodbuddy',
|
||||
authentication: 'password',
|
||||
password: 'not-persisted',
|
||||
hostKey: {
|
||||
algorithm: 'ssh-ed25519',
|
||||
publicKeyBase64: 'a2V5',
|
||||
fingerprintSha256: `SHA256:${'a'.repeat(43)}`,
|
||||
generation: revision
|
||||
}
|
||||
},
|
||||
hostRevision: revision,
|
||||
hostKeyGeneration: revision
|
||||
}
|
||||
}
|
||||
|
||||
const installation = {
|
||||
installationId: verifyAgentInstallationId('agent-v1'),
|
||||
binaryDigest: `sha256:${'a'.repeat(64)}`,
|
||||
agentVersion: '1.0.0',
|
||||
protocol: AGENT_PROTOCOL_VERSION,
|
||||
platform: 'linux' as const,
|
||||
architecture: 'x64' as const,
|
||||
supervisor: 'detached-on-demand' as const,
|
||||
requiredCapabilities: [
|
||||
{ name: 'workspace/read', minimumVersion: 1, critical: true }
|
||||
]
|
||||
}
|
||||
|
||||
function harness(
|
||||
connectGate?: Promise<void>,
|
||||
reconnectResumeGate?: Promise<void>
|
||||
) {
|
||||
let currentTarget = target()
|
||||
const release = vi.fn()
|
||||
const sshLease = {
|
||||
identity: {
|
||||
hostId: 'host-1',
|
||||
hostRevision: 1,
|
||||
hostKeyGeneration: 1,
|
||||
authenticationIdentity: 'b'.repeat(64)
|
||||
},
|
||||
isUsable: () => true,
|
||||
release
|
||||
} as unknown as SshConnectionLease
|
||||
const sshPool: RemoteAgentSshPool = {
|
||||
acquire: vi.fn(async () => sshLease),
|
||||
disposeHost: vi.fn()
|
||||
}
|
||||
const resolver = {
|
||||
resolve: vi.fn(async () => currentTarget)
|
||||
}
|
||||
const clients: FakeProtocolClient[] = []
|
||||
const stateFile = new MemoryStateFile()
|
||||
const connectTransport = vi.fn(async () => {
|
||||
await connectGate
|
||||
const welcome: AttachWelcome = {
|
||||
type: 'goodbuddy-agent-welcome',
|
||||
protocol: AGENT_PROTOCOL_VERSION,
|
||||
connectionId: `connection-${connectTransport.mock.calls.length}`,
|
||||
generation: connectTransport.mock.calls.length + 1,
|
||||
installationId: 'agent-v1',
|
||||
binaryDigest: `sha256:${'a'.repeat(64)}`,
|
||||
daemonBootId: 'boot-1',
|
||||
serverNonce: 'server-1'
|
||||
}
|
||||
return {
|
||||
welcome,
|
||||
dispose: vi.fn()
|
||||
} as unknown as AgentAttachTransport
|
||||
})
|
||||
const manager = new RemoteAgentConnectionManager({
|
||||
resolver,
|
||||
sshPool,
|
||||
controllerState: new ControllerStateStore(stateFile),
|
||||
goodBuddyVersion: '0.11.0',
|
||||
idleTimeoutMs: 0,
|
||||
connectTransport,
|
||||
createProtocolClient: () => {
|
||||
const client = new FakeProtocolClient(clients.length)
|
||||
if (clients.length > 0) {
|
||||
client.resumeGate = reconnectResumeGate
|
||||
}
|
||||
clients.push(client)
|
||||
return client as unknown as AgentProtocolClient
|
||||
}
|
||||
})
|
||||
return {
|
||||
manager,
|
||||
resolver,
|
||||
sshPool,
|
||||
release,
|
||||
connectTransport,
|
||||
clients,
|
||||
stateFile,
|
||||
changeRevision: (revision: number) => {
|
||||
currentTarget = target(revision)
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
describe('RemoteAgentConnectionManager', () => {
|
||||
it('deduplicates concurrent connects and ref-counts the shared connection', async () => {
|
||||
const test = harness()
|
||||
const [first, second] = await Promise.all([
|
||||
test.manager.acquire('host-1', installation),
|
||||
test.manager.acquire('host-1', installation)
|
||||
])
|
||||
expect(first.client).toBe(second.client)
|
||||
expect(test.connectTransport).toHaveBeenCalledTimes(1)
|
||||
expect(test.sshPool.acquire).toHaveBeenCalledTimes(1)
|
||||
first.release()
|
||||
expect(test.release).not.toHaveBeenCalled()
|
||||
second.release()
|
||||
expect(test.release).toHaveBeenCalledTimes(1)
|
||||
await test.manager.dispose()
|
||||
})
|
||||
|
||||
it('isolates cache entries across Host revisions', async () => {
|
||||
const test = harness()
|
||||
const first = await test.manager.acquire('host-1', installation)
|
||||
test.changeRevision(2)
|
||||
const second = await test.manager.acquire('host-1', installation)
|
||||
expect(first.identity.cacheKey).not.toBe(second.identity.cacheKey)
|
||||
expect(test.connectTransport).toHaveBeenCalledTimes(2)
|
||||
first.release()
|
||||
second.release()
|
||||
await test.manager.dispose()
|
||||
})
|
||||
|
||||
it('cancels one waiter without canceling a shared connect', async () => {
|
||||
let openTransport!: () => void
|
||||
const gate = new Promise<void>((resolve) => {
|
||||
openTransport = resolve
|
||||
})
|
||||
const test = harness(gate)
|
||||
const controller = new AbortController()
|
||||
const canceled = test.manager.acquire(
|
||||
'host-1',
|
||||
installation,
|
||||
controller.signal
|
||||
)
|
||||
const retained = test.manager.acquire('host-1', installation)
|
||||
await vi.waitFor(() =>
|
||||
expect(test.connectTransport).toHaveBeenCalledTimes(1)
|
||||
)
|
||||
controller.abort(new Error('waiter canceled'))
|
||||
await expect(canceled).rejects.toThrow('waiter canceled')
|
||||
openTransport()
|
||||
const lease = await retained
|
||||
expect(lease.state).toBe('ready')
|
||||
lease.release()
|
||||
await test.manager.dispose()
|
||||
})
|
||||
|
||||
it('invalidates every connection for a Host without stopping its daemon', async () => {
|
||||
const test = harness()
|
||||
const lease = await test.manager.acquire('host-1', installation)
|
||||
await test.manager.invalidateHost('host-1')
|
||||
expect(test.sshPool.disposeHost).toHaveBeenCalledWith('host-1')
|
||||
expect(() => lease.client).toThrow()
|
||||
lease.release()
|
||||
await test.manager.dispose()
|
||||
})
|
||||
|
||||
it('does not publish a connection that finishes after Host invalidation', async () => {
|
||||
let finishTransport!: () => void
|
||||
const gate = new Promise<void>((resolve) => {
|
||||
finishTransport = resolve
|
||||
})
|
||||
const test = harness(gate)
|
||||
const acquiring = test.manager.acquire('host-1', installation)
|
||||
await vi.waitFor(() =>
|
||||
expect(test.connectTransport).toHaveBeenCalledOnce()
|
||||
)
|
||||
|
||||
await test.manager.invalidateHost('host-1')
|
||||
finishTransport()
|
||||
|
||||
await expect(acquiring).rejects.toMatchObject({
|
||||
disposition: 'transient',
|
||||
reason: 'network'
|
||||
})
|
||||
expect(test.clients[0]?.disposed).toBe(true)
|
||||
expect(test.release).toHaveBeenCalledOnce()
|
||||
await test.manager.dispose()
|
||||
})
|
||||
|
||||
it('refreshes and validates capabilities on the current client generation', async () => {
|
||||
const test = harness()
|
||||
const lease = await test.manager.acquire('host-1', installation)
|
||||
const refreshed: DaemonCapabilities = {
|
||||
generation: 2,
|
||||
capabilities: [
|
||||
{ name: 'workspace/read', version: 1, critical: true },
|
||||
{ name: 'runtime/acp', version: 3, critical: true }
|
||||
],
|
||||
runtimes: [
|
||||
{
|
||||
runtimeId: 'opencode',
|
||||
version: '1.0.0',
|
||||
bundleDigest: `sha256:${'b'.repeat(64)}`,
|
||||
acpCapabilitiesDigest: `sha256:${'c'.repeat(64)}`,
|
||||
sessionLoad: true,
|
||||
sessionResume: true
|
||||
}
|
||||
]
|
||||
}
|
||||
test.clients[0]!.capabilities = refreshed
|
||||
|
||||
await expect(lease.refreshCapabilities()).resolves.toEqual(
|
||||
refreshed
|
||||
)
|
||||
expect(lease.capabilities).toEqual(refreshed)
|
||||
|
||||
test.clients[0]!.capabilities = {
|
||||
generation: 3,
|
||||
capabilities: [
|
||||
{ name: 'runtime/acp', version: 3, critical: true }
|
||||
],
|
||||
runtimes: refreshed.runtimes
|
||||
}
|
||||
await expect(lease.refreshCapabilities()).rejects.toMatchObject({
|
||||
reason: 'capability',
|
||||
disposition: 'terminal'
|
||||
})
|
||||
expect(lease.capabilities).toEqual(refreshed)
|
||||
lease.release()
|
||||
await test.manager.dispose()
|
||||
})
|
||||
|
||||
it('rejects versions above an exact capability requirement', async () => {
|
||||
const test = harness()
|
||||
const lease = await test.manager.acquire('host-1', {
|
||||
...installation,
|
||||
requiredCapabilities: [
|
||||
{
|
||||
name: 'workspace/read',
|
||||
exactVersion: 1,
|
||||
critical: true
|
||||
}
|
||||
]
|
||||
})
|
||||
test.clients[0]!.capabilities = {
|
||||
generation: 2,
|
||||
capabilities: [
|
||||
{ name: 'workspace/read', version: 2, critical: true }
|
||||
],
|
||||
runtimes: []
|
||||
}
|
||||
|
||||
await expect(lease.refreshCapabilities()).rejects.toMatchObject({
|
||||
reason: 'capability',
|
||||
disposition: 'terminal'
|
||||
})
|
||||
lease.release()
|
||||
await test.manager.dispose()
|
||||
})
|
||||
|
||||
it('refuses capability refresh after its lease is released', async () => {
|
||||
const test = harness()
|
||||
const lease = await test.manager.acquire('host-1', installation)
|
||||
lease.release()
|
||||
await expect(lease.refreshCapabilities()).rejects.toThrow(
|
||||
'lease is released'
|
||||
)
|
||||
await test.manager.dispose()
|
||||
})
|
||||
|
||||
it('does not publish a capability refresh after its lease is released in flight', async () => {
|
||||
const test = harness()
|
||||
const [refreshing, retained] = await Promise.all([
|
||||
test.manager.acquire('host-1', installation),
|
||||
test.manager.acquire('host-1', installation)
|
||||
])
|
||||
let finishRefresh!: () => void
|
||||
test.clients[0]!.capabilitiesGate = new Promise<void>((resolve) => {
|
||||
finishRefresh = resolve
|
||||
})
|
||||
test.clients[0]!.capabilities = {
|
||||
generation: 2,
|
||||
capabilities: [
|
||||
{ name: 'workspace/read', version: 1, critical: true }
|
||||
],
|
||||
runtimes: []
|
||||
}
|
||||
|
||||
const refresh = refreshing.refreshCapabilities()
|
||||
refreshing.release()
|
||||
finishRefresh()
|
||||
|
||||
await expect(refresh).rejects.toThrow('lease is released')
|
||||
expect(retained.capabilities.generation).toBe(1)
|
||||
retained.release()
|
||||
await test.manager.dispose()
|
||||
})
|
||||
|
||||
it('proves the exact predecessor tuple and notifies retained leases after a client swap', async () => {
|
||||
const test = harness()
|
||||
const lease = await test.manager.acquire('host-1', installation)
|
||||
const originalClient = lease.client
|
||||
await lease.updateAcpBinding('binding-1', {
|
||||
bindingId: 'binding-1',
|
||||
channelId: 'binding-1',
|
||||
channelEpoch: '9',
|
||||
cursors: {
|
||||
lastOutboundJournaledSequence: '2',
|
||||
lastOutboundDeliveredSequence: '2',
|
||||
lastInboundJournaledSequence: '4',
|
||||
lastMainAckSequence: '3'
|
||||
}
|
||||
})
|
||||
const changed = vi.fn()
|
||||
lease.onClientChange?.(changed)
|
||||
|
||||
test.clients[0]!.transportClose()
|
||||
expect(lease.state).toBe('offline')
|
||||
await lease.reconnect()
|
||||
|
||||
expect(lease.client).not.toBe(originalClient)
|
||||
expect(lease.client.generation).toBe(3)
|
||||
expect(changed).toHaveBeenCalledOnce()
|
||||
expect(
|
||||
test.clients[1]!.requests.find(
|
||||
(request) => request.method === 'controller/resume'
|
||||
)
|
||||
).toEqual({
|
||||
method: 'controller/resume',
|
||||
params: {
|
||||
previousGeneration: 2,
|
||||
previousConnectionId: 'connection-1',
|
||||
daemonBootId: 'boot-1',
|
||||
capabilityGeneration: 1
|
||||
}
|
||||
})
|
||||
lease.release()
|
||||
await test.manager.dispose()
|
||||
})
|
||||
|
||||
it('persists ACP recovery cursors while the retained transport is offline', async () => {
|
||||
const test = harness()
|
||||
const lease = await test.manager.acquire('host-1', installation)
|
||||
await lease.updateAcpBinding('binding-1', {
|
||||
bindingId: 'binding-1',
|
||||
channelId: 'binding-1',
|
||||
channelEpoch: '9',
|
||||
cursors: {
|
||||
lastOutboundJournaledSequence: '2',
|
||||
lastOutboundDeliveredSequence: '2',
|
||||
lastInboundJournaledSequence: '3',
|
||||
lastMainAckSequence: '1'
|
||||
}
|
||||
})
|
||||
test.clients[0]!.transportClose()
|
||||
expect(lease.state).toBe('offline')
|
||||
|
||||
await expect(
|
||||
lease.updateAcpBinding('binding-1', {
|
||||
bindingId: 'binding-1',
|
||||
channelId: 'binding-1',
|
||||
channelEpoch: '9',
|
||||
cursors: {
|
||||
lastOutboundJournaledSequence: '2',
|
||||
lastOutboundDeliveredSequence: '2',
|
||||
lastInboundJournaledSequence: '3',
|
||||
lastMainAckSequence: '2'
|
||||
}
|
||||
})
|
||||
).resolves.toBeUndefined()
|
||||
await expect(lease.flushAcpBindings()).resolves.toBeUndefined()
|
||||
expect(test.stateFile.value).toMatchObject({
|
||||
connections: [
|
||||
{
|
||||
acpBindings: [
|
||||
{
|
||||
bindingId: 'binding-1',
|
||||
cursors: {
|
||||
lastMainAckSequence: '2'
|
||||
}
|
||||
}
|
||||
]
|
||||
}
|
||||
]
|
||||
})
|
||||
|
||||
lease.release()
|
||||
await expect(
|
||||
lease.flushAcpBindings()
|
||||
).rejects.toThrow(/released/iu)
|
||||
await test.manager.dispose()
|
||||
})
|
||||
|
||||
it('preserves ACP cursors persisted while reconnect resume is in flight', async () => {
|
||||
let finishResume!: () => void
|
||||
const resumeGate = new Promise<void>((resolve) => {
|
||||
finishResume = resolve
|
||||
})
|
||||
const test = harness(undefined, resumeGate)
|
||||
const lease = await test.manager.acquire('host-1', installation)
|
||||
await lease.updateAcpBinding('binding-1', {
|
||||
bindingId: 'binding-1',
|
||||
channelId: 'binding-1',
|
||||
channelEpoch: '9',
|
||||
cursors: {
|
||||
lastOutboundJournaledSequence: '2',
|
||||
lastOutboundDeliveredSequence: '2',
|
||||
lastInboundJournaledSequence: '3',
|
||||
lastMainAckSequence: '1'
|
||||
}
|
||||
})
|
||||
await lease.flushAcpBindings()
|
||||
test.clients[0]!.transportClose()
|
||||
const reconnect = lease.reconnect()
|
||||
await vi.waitFor(() =>
|
||||
expect(
|
||||
test.clients[1]?.requests.some(
|
||||
(request) => request.method === 'controller/resume'
|
||||
)
|
||||
).toBe(true)
|
||||
)
|
||||
|
||||
await lease.updateAcpBinding('binding-1', {
|
||||
bindingId: 'binding-1',
|
||||
channelId: 'binding-1',
|
||||
channelEpoch: '9',
|
||||
cursors: {
|
||||
lastOutboundJournaledSequence: '2',
|
||||
lastOutboundDeliveredSequence: '2',
|
||||
lastInboundJournaledSequence: '3',
|
||||
lastMainAckSequence: '2'
|
||||
}
|
||||
})
|
||||
await lease.flushAcpBindings()
|
||||
finishResume()
|
||||
await reconnect
|
||||
|
||||
expect(test.stateFile.value).toMatchObject({
|
||||
connections: [
|
||||
{
|
||||
acpBindings: [
|
||||
{
|
||||
bindingId: 'binding-1',
|
||||
cursors: {
|
||||
lastMainAckSequence: '2'
|
||||
}
|
||||
}
|
||||
]
|
||||
}
|
||||
]
|
||||
})
|
||||
lease.release()
|
||||
await test.manager.dispose()
|
||||
})
|
||||
|
||||
it('shuts down channels and refuses later acquisition', async () => {
|
||||
const test = harness()
|
||||
const lease = await test.manager.acquire('host-1', installation)
|
||||
await test.manager.dispose()
|
||||
expect(test.clients[0]!.disposed).toBe(true)
|
||||
expect(() => lease.client).toThrow()
|
||||
await expect(
|
||||
test.manager.acquire('host-1', installation)
|
||||
).rejects.toMatchObject({ reason: 'shutdown' })
|
||||
})
|
||||
})
|
||||
Reference in New Issue
Block a user