rollbridge 0.1.19 → 0.1.22

package/README.md

@@ -412,6 +412,15 @@ rollbridge daemon --config /srv/ticket-server/rollbridge.js \
   --boot-attestation sha256:0123456789abcdef0123456789abcdef0123456789abcdef0123456789abcdef
 ```
 
+External supervisors that need to replace a foreground owner without waiting
+for its workers to drain add `--takeover-owner`. The candidate starts and
+health-checks the exact release first. Only then does it retire the accepted
+owner's proxy/control listeners; the retired daemon keeps its already-accepted
+workers until their ordinary drain completes while the attested replacement
+binds the stable listeners. Candidate bootstrap failure leaves the accepted
+owner untouched. This is opt-in; ordinary daemon bootstrap and `shutdown` keep
+their existing behavior.
+
 The four bootstrap inputs are all-or-nothing and use absolute config/release
 paths. Rollbridge binds its proxy, activates the release through the normal
 deploy path, then exposes the control socket and stays foreground. A failed

package/docs/cli.md

@@ -23,7 +23,7 @@ process-policy details.
 ## `daemon`
 
 ```
-rollbridge daemon [--config <path>]
+rollbridge daemon [--config <path>] [--takeover-owner]
                   [--release-path <path> --release-id <id> --revision <sha>
                    [--boot-attestation <sha256:digest>]]
 ```
@@ -64,6 +64,13 @@ identity. Rollbridge does not calculate or interpret the digest.
 With no release options, daemon behavior is unchanged: it starts listener-only
 and waits for control-socket deployments.
 
+`--takeover-owner` requires the complete bootstrap tuple. It bootstraps and
+health-checks the replacement before sending the current daemon the private
+retirement command. Retirement stops stable listeners and new work promptly,
+but does not wait for owned workers' normal drain before the replacement binds
+the proxy and control socket. A bootstrap failure occurs before retirement, so
+the previously accepted owner remains available.
+
 ## `ensure-daemon`
 
 ```

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "rollbridge",
-  "version": "0.1.19",
+  "version": "0.1.22",
   "description": "Zero-downtime process supervisor and local traffic switcher for deploy-managed apps.",
   "keywords": [
     "deploy",

package/src/cli.js

@@ -36,6 +36,7 @@ export async function runCli(argv) {
     .option("--release-id <id>", "Bootstrap release id (requires --config, --release-path, and --revision)")
     .option("--revision <sha>", "Bootstrap revision (requires --config, --release-path, and --release-id)")
     .option("--boot-attestation <digest>", "Opaque bootstrap ownership attestation (requires the complete bootstrap release tuple)")
+    .option("--takeover-owner", "Boot and health-check before retiring the current external owner")
     .action(async (options) => {
       const bootstrap = await validateDaemonBootstrapOptions(options)
       const configPath = await resolveConfigPath(options.config)
@@ -43,7 +44,9 @@ export async function runCli(argv) {
       const runtime = await loadDaemonRuntimeIdentity(process.env.ROLLBRIDGE_DAEMON_RUNTIME_MANIFEST)
       const daemon = new RollbridgeDaemon({bootstrap, config, configPath, runtime})
 
-      await daemon.start({exposeControl: !bootstrap})
+      if (options.takeoverOwner && (!bootstrap || !bootstrap.attestation)) throw new Error("Daemon --takeover-owner requires the complete bootstrap release tuple and --boot-attestation.")
+
+      if (!options.takeoverOwner) await daemon.start({exposeControl: !bootstrap})
 
       const shutdown = async () => {
         await daemon.shutdown()
@@ -56,10 +59,25 @@ export async function runCli(argv) {
       if (bootstrap) {
         try {
           await daemon.deploy(bootstrap)
-          await daemon.exposeControl()
-        } catch {
-          daemon.logger("bootstrap activation failed", {releaseId: bootstrap.releaseId, status: "error"})
-          await daemon.shutdown()
+          if (options.takeoverOwner) {
+            await sendControlCommand({command: {attestation: bootstrap.attestation, command: "retire-owner"}, path: config.control.path})
+            await daemon.start({reportOrphans: false})
+          } else {
+            await daemon.exposeControl()
+          }
+        } catch (error) {
+          const failure = error instanceof Error ? error : String(error)
+
+          daemon.logger("bootstrap activation failed", {releaseId: bootstrap.releaseId, status: "error", ...errorLogData(failure)})
+
+          try {
+            await daemon.shutdown()
+          } catch (shutdownError) {
+            const shutdownFailure = shutdownError instanceof Error ? shutdownError : String(shutdownError)
+
+            daemon.logger("bootstrap shutdown failed", {releaseId: bootstrap.releaseId, status: "error", ...errorLogData(shutdownFailure)})
+          }
+
           process.exitCode = 1
           return
         }
@@ -952,3 +970,18 @@ function isMissingDaemonError(error) {
 
   return error.code === "ENOENT" || error.code === "ECONNREFUSED"
 }
+
+/**
+ * Converts any thrown value into JSON-safe diagnostics without losing an Error stack.
+ * @param {Error | string} error - Thrown value.
+ * @returns {{error: string, stack: string}} Safe structured log fields.
+ */
+function errorLogData(error) {
+  if (error instanceof Error) {
+    return {error: error.message, stack: error.stack || `${error.name}: ${error.message}`}
+  }
+
+  const message = String(error)
+
+  return {error: message, stack: message}
+}

package/src/daemon.js

@@ -56,13 +56,17 @@ export default class RollbridgeDaemon {
     this.proxy = httpProxy.createProxyServer({ws: true, xfwd: true})
     this.proxyServer = /** @type {http.Server | undefined} */ (undefined)
     this.controlServer = /** @type {net.Server | undefined} */ (undefined)
+    this.controlSocketOwned = false
     this.controlSockets = /** @type {Set<net.Socket>} */ (new Set())
     this.proxyPort = /** @type {number | undefined} */ (undefined)
     this.stopping = false
     this.statePath = config.statePath
     this.persistTimer = /** @type {ReturnType<typeof setInterval> | undefined} */ (undefined)
+    this.persistenceEnabled = false
     this.pendingWrite = /** @type {Promise<void> | undefined} */ (undefined)
+    this.stateCleanupEnabled = false
     this.shutdownPromise = /** @type {Promise<void> | undefined} */ (undefined)
+    this.retirementPromise = /** @type {Promise<void> | undefined} */ (undefined)
     this.controlClosePromise = /** @type {Promise<void> | undefined} */ (undefined)
     this.startingReleases = /** @type {Set<ReleaseGroup>} */ (new Set())
     // Still-alive managed processes left by a previous daemon (from statePath), captured at
@@ -74,11 +78,11 @@ export default class RollbridgeDaemon {
 
   /**
    * Starts daemon listeners.
-   * @param {{exposeControl?: boolean}} [options] - Whether to expose the control socket immediately.
+   * @param {{exposeControl?: boolean, reportOrphans?: boolean}} [options] - Listener startup options.
    * @returns {Promise<void>} Resolves when the requested listeners are ready.
    */
-  async start({exposeControl = true} = {}) {
-    await this.reportOrphans()
+  async start({exposeControl = true, reportOrphans = true} = {}) {
+    if (reportOrphans) await this.reportOrphans()
     await this.startProxy()
     if (exposeControl) await this.exposeControl()
   }
@@ -126,6 +130,7 @@ export default class RollbridgeDaemon {
     await new Promise((resolve, reject) => {
       server.once("error", reject)
       server.listen(this.config.control.path, () => {
+        this.controlSocketOwned = true
         this.logger("control socket listening", {path: this.config.control.path})
         resolve(undefined)
       })
@@ -358,6 +363,13 @@ export default class RollbridgeDaemon {
       return {message: "shutdown"}
     }
 
+    if (commandName === "retire-owner") {
+      const attestation = requiredString(data.attestation, "attestation")
+      if (!/^sha256:[a-f0-9]{64}$/.test(attestation)) throw new Error("Owner retirement attestation must use the canonical sha256:<64 lowercase hex> format")
+      await this.retireOwner({attestation, completionSocket: controlSocket})
+      return {message: "owner retired"}
+    }
+
     throw new Error(`Unknown command: ${String(commandName)}`)
   }
 
@@ -732,6 +744,8 @@ export default class RollbridgeDaemon {
   startStatePersistence() {
     if (!this.statePath) return
 
+    this.stateCleanupEnabled = true
+    this.persistenceEnabled = true
     this.persistState()
     this.persistTimer = setInterval(() => this.persistState(), STATE_PERSIST_INTERVAL_MS)
     this.persistTimer.unref?.()
@@ -743,7 +757,7 @@ export default class RollbridgeDaemon {
    * @returns {void}
    */
   persistState() {
-    if (!this.statePath || this.stopping) return
+    if (!this.statePath || !this.persistenceEnabled || this.stopping) return
 
     const statePath = this.statePath
     const status = /** @type {Record<string, JsonValue>} */ (secretSafeStateValue(this.status()))
@@ -770,6 +784,7 @@ export default class RollbridgeDaemon {
   async reportOrphans() {
     if (!this.statePath) return
 
+    this.stateCleanupEnabled = true
     const orphans = liveProcesses(await readState(this.statePath))
 
     // Keep them for status() so `rollbridge status` reflects still-running children after a
@@ -797,6 +812,50 @@ export default class RollbridgeDaemon {
     if (waitForControlConnections && this.controlClosePromise) await this.controlClosePromise
   }
 
+  /**
+   * Relinquishes stable listeners promptly while retaining draining children under
+   * this daemon until their normal stop contract completes.
+   * @param {{attestation: string, completionSocket?: net.Socket}} options - Attested handoff request.
+   * @returns {Promise<void>} Resolves once a replacement can exclusively bind listeners.
+   */
+  async retireOwner({attestation, completionSocket}) {
+    if (this.retirementPromise) return await this.retirementPromise
+    this.retirementPromise = this.performOwnerRetirement(attestation, completionSocket)
+    return await this.retirementPromise
+  }
+
+  /**
+   * @param {string} attestation - Replacement boot attestation.
+   * @param {net.Socket | undefined} completionSocket - Requesting handoff connection.
+   * @returns {Promise<void>} Resolves after quiesce and listener release.
+   */
+  async performOwnerRetirement(attestation, completionSocket) {
+    this.stopping = true
+    if (this.persistTimer) {
+      clearInterval(this.persistTimer)
+      this.persistTimer = undefined
+    }
+    this.persistenceEnabled = false
+    if (this.pendingWrite) await this.pendingWrite
+    this.controlClosePromise = this.closeServer(this.controlServer)
+    for (const socket of this.controlSockets) if (socket !== completionSocket) socket.destroy()
+    await Promise.all([
+      ...[...this.services.values()].map((processInstance) => processInstance.quiesce()),
+      ...[...this.singletons.values()].map((processInstance) => processInstance.quiesce()),
+      ...[...this.startingReleases].map((release) => release.quiesce()),
+      ...[...this.releases.values()].map((release) => release.quiesce())
+    ])
+    await this.removeControlSocket()
+    void this.closeServer(this.proxyServer)
+    void Promise.allSettled([
+      ...[...this.services.values()].map((processInstance) => processInstance.stop()),
+      ...[...this.singletons.values()].map((processInstance) => processInstance.stop()),
+      ...[...this.startingReleases].map((release) => release.stop()),
+      ...[...this.releases.values()].map((release) => release.stop())
+    ])
+    this.logger("external owner retired", {attestation, status: "draining"})
+  }
+
   /**
    * @param {net.Socket | undefined} completionSocket - Requester retained for the final response.
    * @returns {Promise<void>} Retires listeners and cleans up every daemon-owned resource.
@@ -834,7 +893,7 @@ export default class RollbridgeDaemon {
     // new writes start: stopping is set and the persist timer is cleared above). Prior-daemon
     // orphans are not owned by this daemon, so retain their records until they are confirmed gone.
     await captureShutdownError(cleanupErrors, "persistent state cleanup", async () => {
-      if (!this.statePath) return
+      if (!this.statePath || !this.stateCleanupEnabled) return
       if (this.pendingWrite) await this.pendingWrite
       const orphans = this.orphans.filter((orphan) => isProcessAlive(orphan.pid))
 
@@ -856,7 +915,10 @@ export default class RollbridgeDaemon {
 
   /** @returns {Promise<void>} Removes the configured control socket path. */
   async removeControlSocket() {
+    if (!this.controlSocketOwned) return
+
     await fs.rm(this.config.control.path, {force: true})
+    this.controlSocketOwned = false
   }
 
   /**
@@ -925,7 +987,7 @@ function isShutdownControlLine(line) {
   try {
     const command = JSON.parse(line)
 
-    return Boolean(command && typeof command === "object" && command.command === "shutdown")
+    return Boolean(command && typeof command === "object" && ["retire-owner", "shutdown"].includes(command.command))
   } catch {
     return false
   }

package/src/managed-process.js

@@ -62,6 +62,8 @@ export default class ManagedProcess extends EventEmitter {
     this.memoryWarned = false
     this.startedAtMs = /** @type {number | undefined} */ (undefined)
     this.intentionalStop = false
+    this.intentionalStopSignal = /** @type {ProcessExitSignal | undefined} */ (undefined)
+    this.quiescePromise = /** @type {Promise<void> | undefined} */ (undefined)
     this.restartTimer = undefined
     this.child = undefined
     this.exitPromise = undefined
@@ -78,6 +80,8 @@ export default class ManagedProcess extends EventEmitter {
     if (this.child) return
 
     this.intentionalStop = false
+    this.intentionalStopSignal = undefined
+    this.quiescePromise = undefined
     this.exitCode = undefined
     this.exitSignal = undefined
     this.state = "starting"
@@ -165,8 +169,8 @@ export default class ManagedProcess extends EventEmitter {
   onExit(code, signal) {
     const wasIntentional = this.intentionalStop
 
-    this.exitCode = code
-    this.exitSignal = signal
+    this.exitCode = this.intentionalStopSignal ? null : code
+    this.exitSignal = signal ?? this.intentionalStopSignal
     this.child = undefined
     this.pid = undefined
     this.exitPromise = undefined
@@ -338,36 +342,22 @@ export default class ManagedProcess extends EventEmitter {
    * @returns {Promise<void>} Resolves when stopped.
    */
   async stop(options = {}) {
-    this.intentionalStop = true
-    this.clearMemoryMonitor()
-
-    if (this.restartTimer) {
-      clearTimeout(this.restartTimer)
-      this.restartTimer = undefined
-    }
-
-    const child = this.child
+    const pgid = this.child?.pid ?? this.pid
+    const exitPromise = this.exitPromise
+    await this.quiesce()
 
-    if (!child || !child.pid) {
+    if (!pgid) {
       this.state = "stopped"
       return
     }
 
-    const pgid = child.pid
-    const exitPromise = this.exitPromise
-
-    this.state = "stopping"
-
-    const {drainCommand, drainTimeoutMs, quietCommand, stopCommand} = this.lifecycle
+    const {drainCommand, drainTimeoutMs, stopCommand} = this.lifecycle
 
     const hookTimeoutMs = this.hookTimeoutMs()
 
-    // 1. Quiesce: tell the process to stop accepting new work.
-    if (quietCommand) await this.runHook(quietCommand, hookTimeoutMs, "quiet command")
-
     // 2. Drain: let in-flight work finish, bounded by drainTimeoutMs (0 skips the step). A
     //    drainCommand blocks until drained; otherwise wait for the process to exit on its own.
-    if (this.child && drainTimeoutMs > 0) {
+    if (this.processGroupExists(pgid) && drainTimeoutMs > 0) {
       if (drainCommand) await this.runHook(drainCommand, drainTimeoutMs, "drain command")
       else await this.waitForExit(drainTimeoutMs)
     }
@@ -375,13 +365,16 @@ export default class ManagedProcess extends EventEmitter {
     // 3. Stop whatever is still running, then SIGKILL if it outlasts the graceful window.
     if (this.processGroupExists(pgid)) {
       if (stopCommand) await this.runHook(stopCommand, hookTimeoutMs, "stop command", pgid)
-      else this.killProcessGroup(this.stopSignal, pgid)
+      else {
+        this.intentionalStopSignal = /** @type {ProcessExitSignal} */ (this.stopSignal)
+        await this.signalProcessGroup(this.stopSignal, pgid, options.timeoutMs ?? this.stopTimeoutMs)
+      }
 
       const timeoutMs = options.timeoutMs ?? this.stopTimeoutMs
 
       if (!(await this.waitForProcessGroupExit(pgid, timeoutMs))) {
         this.logger("process stop timed out; sending SIGKILL", {id: this.id, pid: pgid})
-        this.killProcessGroup("SIGKILL", pgid)
+        await this.signalProcessGroup("SIGKILL", pgid, 5000)
         await this.waitForProcessGroupExit(pgid, 5000)
       }
     }
@@ -391,6 +384,26 @@ export default class ManagedProcess extends EventEmitter {
     this.state = "stopped"
   }
 
+  /** @returns {Promise<void>} Stops restarts and waits until the process has stopped accepting new work. */
+  async quiesce() {
+    if (this.quiescePromise) return await this.quiescePromise
+    this.quiescePromise = (async () => {
+      this.intentionalStop = true
+      this.clearMemoryMonitor()
+      if (this.restartTimer) {
+        clearTimeout(this.restartTimer)
+        this.restartTimer = undefined
+      }
+      if (!this.child?.pid) {
+        this.state = "stopped"
+        return
+      }
+      this.state = "stopping"
+      if (this.lifecycle.quietCommand) await this.runHook(this.lifecycle.quietCommand, this.hookTimeoutMs(), "quiet command")
+    })()
+    return await this.quiescePromise
+  }
+
   /** @returns {number} Timeout used for lifecycle hooks. */
   hookTimeoutMs() {
     if (this.stopTimeoutMs === "indefinite") return 30000
@@ -481,6 +494,49 @@ export default class ManagedProcess extends EventEmitter {
     }
   }
 
+  /**
+   * Signals descendants before their shell leader so the leader can reap them before it exits.
+   * Falls back to the portable group signal when procfs cannot identify group members.
+   * @param {string} signal - Signal name.
+   * @param {number} pgid - Process group id.
+   * @param {StopTimeoutMs} timeoutMs - Maximum time to wait for descendant reaping.
+   * @returns {Promise<void>} Resolves after the leader has been signalled.
+   */
+  async signalProcessGroup(signal, pgid, timeoutMs) {
+    const members = processGroupMembers(pgid)
+    const descendants = members.filter((member) => member.pid !== pgid)
+
+    if (descendants.length === 0) {
+      this.killProcessGroup(signal, pgid)
+      return
+    }
+
+    for (const descendant of descendants) this.killProcess(descendant.pid, signal)
+    const deadline = timeoutMs === "indefinite" ? undefined : Date.now() + timeoutMs
+
+    while (processGroupMembers(pgid).some((member) => member.pid !== pgid)) {
+      if (deadline !== undefined && Date.now() >= deadline) break
+      await new Promise((resolve) => setTimeout(resolve, 25))
+    }
+
+    this.killProcess(pgid, signal)
+  }
+
+  /**
+   * Signals one verified member of the owned process group.
+   * @param {number} pid - Process id.
+   * @param {string} signal - Signal name.
+   * @returns {void}
+   */
+  killProcess(pid, signal) {
+    try {
+      process.kill(pid, signal)
+    } catch (error) {
+      if (error && typeof error === "object" && "code" in error && error.code === "ESRCH") return
+      throw error
+    }
+  }
+
   /**
    * @param {number} pgid - Process group id.
    * @returns {boolean} True until the process group no longer exists.

package/src/release-group.js

@@ -403,6 +403,11 @@ export default class ReleaseGroup extends EventEmitter {
     this.stoppedAt = new Date().toISOString()
   }
 
+  /** @returns {Promise<void>} Quiesces every release process without waiting for its drain. */
+  async quiesce() {
+    await Promise.all([...this.processes.values()].map((processInstance) => processInstance.quiesce()))
+  }
+
   /** @returns {ReleaseStatus} Status payload. */
   status() {
     return {

package/test/daemon-bootstrap.test.js

@@ -4,6 +4,7 @@ import assert from "node:assert/strict"
 import {spawn} from "node:child_process"
 import {once} from "node:events"
 import fs from "node:fs/promises"
+import net from "node:net"
 import os from "node:os"
 import path from "node:path"
 import test from "node:test"
@@ -14,9 +15,12 @@ import {isProcessAlive, liveProcesses, readState, writeState} from "../src/state
 const currentDir = path.dirname(fileURLToPath(import.meta.url))
 const binPath = path.join(currentDir, "..", "bin", "rollbridge")
 const dummyAppPath = path.join(currentDir, "fixtures", "dummy-app.js")
+const ownedChildPath = path.join(currentDir, "fixtures", "owned-child.js")
 const firstAttestation = `sha256:${"a".repeat(64)}`
 const secondAttestation = `sha256:${"b".repeat(64)}`
 
+/** @typedef {{data?: Record<string, import("../src/json.js").JsonValue>, message?: string}} StructuredRecord */
+
 test("daemon bootstrap requires complete, safe, absolute inputs before binding listeners", async (t) => {
   const cases = [
     {args: ["--config", "CONFIG", "--release-path", "RELEASE", "--release-id", "v1"], message: /must be provided together/},
@@ -85,6 +89,49 @@ test("daemon bootstrap activates the exact release through the foreground daemon
   }
 })
 
+test("failed takeover bootstrap preserves the previously accepted owner", async () => {
+  const fixture = await createFixture({persistState: true})
+  const accepted = spawnDaemon(fixture, {attestation: firstAttestation, releaseId: "accepted", revision: "accepted123"})
+
+  try {
+    await waitForLog(accepted, "control socket listening")
+    await waitForFile(fixture.statePath)
+    const badConfig = JSON.parse((await fs.readFile(fixture.configPath, "utf8")).replace(/^module\.exports = /, ""))
+    badConfig.processes[0].health.path = "/never-ready"
+    badConfig.processes[0].health.timeoutMs = 100
+    await fs.writeFile(fixture.configPath, `module.exports = ${JSON.stringify(badConfig, null, 2)}\n`)
+
+    const result = await runDaemon([
+      "--config", fixture.configPath,
+      "--release-path", fixture.root,
+      "--release-id", "candidate",
+      "--revision", "candidate123",
+      "--boot-attestation", secondAttestation,
+      "--takeover-owner"
+    ])
+
+    assert.notEqual(result.code, 0)
+    const records = parseStructuredOutput(result.output)
+    const failure = records.find((record) => record.message === "bootstrap activation failed")
+    const candidatePid = Number(await fs.readFile(fixture.startedPath, "utf8"))
+
+    assert.match(String(failure?.data?.error), /Health check failed/)
+    assert.match(String(failure?.data?.stack), /Error: Health check failed/)
+    assert.equal(isProcessAlive(candidatePid), false, "the failed candidate process must be stopped")
+    const status = await sendControlCommand({command: {command: "status"}, path: fixture.socketPath})
+    assert.equal(status.activeReleaseId, "accepted")
+    assert.ok(status.bootstrap && typeof status.bootstrap === "object" && !Array.isArray(status.bootstrap))
+    assert.equal(status.bootstrap.attestation, firstAttestation)
+    const priorState = await readState(fixture.statePath)
+
+    assert.ok(priorState && typeof priorState === "object" && !Array.isArray(priorState) && priorState.activeReleaseId === "accepted", "candidate cleanup must preserve the prior owner's state")
+  } finally {
+    accepted.kill("SIGTERM")
+    if (accepted.exitCode === null) await once(accepted, "exit")
+    await fs.rm(fixture.root, {force: true, recursive: true})
+  }
+})
+
 test("daemon bootstrap does not expose control deploys until activation completes", async () => {
   const fixture = await createFixture({healthGate: true, healthTimeoutMs: 60000})
   const started = waitForFile(fixture.startedPath)
@@ -235,24 +282,56 @@ test("SIGTERM during multi-process bootstrap owns every process started after sh
   }
 })
 
-test("failed daemon bootstrap reports a structured failure, cleans its processes, and exits non-zero", async () => {
-  const fixture = await createFixture({healthPath: "/never-ready", healthTimeoutMs: 100})
+test("failed ordinary bootstrap completely shuts down attempt-owned resources and exposes the cause", async () => {
+  const fixture = await createFixture({attemptOwnedProcesses: true, fixedPorts: true, missingControlParent: true})
 
   try {
     const result = await runDaemon([
       "--config", fixture.configPath,
       "--release-path", fixture.root,
-      "--release-id", "bad-release",
-      "--revision", "bad123"
+      "--release-id", "ordinary-failure",
+      "--revision", "ordinary123"
     ])
-    const records = result.output.split("\n").filter(Boolean).map((line) => JSON.parse(line))
+    const records = parseStructuredOutput(result.output)
     const failure = records.find((record) => record.message === "bootstrap activation failed")
 
     assert.notEqual(result.code, 0)
-    assert.deepEqual(failure?.data, {releaseId: "bad-release", status: "error"})
-    assert.ok(records.some((record) => record.message === "release startup process status" && record.data?.phase === "after cleanup" && record.data?.state === "stopped"))
+    assert.equal(failure?.data?.releaseId, "ordinary-failure")
+    assert.equal(failure?.data?.status, "error")
+    assert.match(String(failure?.data?.error), /listen (?:EACCES|ENOENT)/)
+    assert.match(String(failure?.data?.stack), /Error: listen (?:EACCES|ENOENT)/)
+    await assertAttemptResourcesStopped(fixture, records)
     await assert.rejects(() => fs.stat(fixture.socketPath), {code: "ENOENT"})
   } finally {
+    await killAttemptProcesses(fixture.lifecyclePath)
+    await fs.rm(fixture.root, {force: true, recursive: true})
+  }
+})
+
+test("failed takeover retirement completely shuts down and exits non-zero instead of lingering", async () => {
+  const fixture = await createFixture({attemptOwnedProcesses: true, fixedPorts: true})
+
+  try {
+    const result = await runDaemon([
+      "--config", fixture.configPath,
+      "--release-path", fixture.root,
+      "--release-id", "orphaned-candidate",
+      "--revision", "candidate123",
+      "--boot-attestation", secondAttestation,
+      "--takeover-owner"
+    ], {timeoutMs: 2000})
+    const records = parseStructuredOutput(result.output)
+    const failure = records.find((record) => record.message === "bootstrap activation failed")
+
+    assert.notEqual(result.code, 0)
+    assert.equal(failure?.data?.releaseId, "orphaned-candidate")
+    assert.equal(failure?.data?.status, "error")
+    assert.match(String(failure?.data?.error), /connect ENOENT/)
+    assert.match(String(failure?.data?.stack), /Error: connect ENOENT/)
+    await assertAttemptResourcesStopped(fixture, records)
+    await assert.rejects(() => fs.stat(fixture.socketPath), {code: "ENOENT"})
+  } finally {
+    await killAttemptProcesses(fixture.lifecyclePath)
     await fs.rm(fixture.root, {force: true, recursive: true})
   }
 })
@@ -318,12 +397,12 @@ test("failed daemon bootstrap preserves prior live process records in statePath"
 })
 
 /**
- * @param {{healthGate?: boolean, healthPath?: string, healthTimeoutMs?: number, multiProcessSignal?: boolean, persistState?: boolean}} [options] - Fixture options.
- * @returns {Promise<{configPath: string, gatePath: string, healthGatePath: string, lifecyclePath: string, root: string, socketPath: string, startedPath: string, statePath: string, stoppedPath: string}>} Fixture paths.
+ * @param {{attemptOwnedProcesses?: boolean, fixedPorts?: boolean, healthGate?: boolean, healthPath?: string, healthTimeoutMs?: number, missingControlParent?: boolean, multiProcessSignal?: boolean, persistState?: boolean}} [options] - Fixture options.
+ * @returns {Promise<{configPath: string, gatePath: string, healthGatePath: string, lifecyclePath: string, processPort: number, proxyPort: number, root: string, socketPath: string, startedPath: string, statePath: string, stoppedPath: string}>} Fixture paths.
  */
-async function createFixture({healthGate = false, healthPath = "/ping", healthTimeoutMs = 1000, multiProcessSignal = false, persistState = false} = {}) {
+async function createFixture({attemptOwnedProcesses = false, fixedPorts = false, healthGate = false, healthPath = "/ping", healthTimeoutMs = 1000, missingControlParent = false, multiProcessSignal = false, persistState = false} = {}) {
   const root = await fs.mkdtemp(path.join(os.tmpdir(), "rollbridge-bootstrap-"))
-  const socketPath = path.join(root, "control.sock")
+  const socketPath = missingControlParent ? path.join(root, "missing", "control.sock") : path.join(root, "control.sock")
   const statePath = path.join(root, "state.json")
   const startedPath = path.join(root, "started.pid")
   const stoppedPath = path.join(root, "stopped.pid")
@@ -332,8 +411,12 @@ async function createFixture({healthGate = false, healthPath = "/ping", healthTi
   const healthGatePath = path.join(root, "health-ready")
   const configPath = path.join(root, "rollbridge.js")
   const command = `${JSON.stringify(process.execPath)} ${JSON.stringify(dummyAppPath)}`
+  const ownedCommand = `exec ${JSON.stringify(process.execPath)} ${JSON.stringify(ownedChildPath)}`
+  const ownedWebCommand = `exec ${command}`
+  const [processPort, proxyPort] = fixedPorts ? await availablePorts(2) : [0, 0]
   const lifecycleEnv = {ROLLBRIDGE_TEST_LIFECYCLE_PATH: lifecyclePath}
   const webEnv = {
+    ...(attemptOwnedProcesses ? lifecycleEnv : {}),
     ROLLBRIDGE_TEST_STARTED_PATH: startedPath,
     ROLLBRIDGE_TEST_STOPPED_PATH: stoppedPath,
     ...(healthGate ? {ROLLBRIDGE_TEST_HEALTH_GATE_PATH: healthGatePath} : {})
@@ -345,8 +428,13 @@ async function createFixture({healthGate = false, healthPath = "/ping", healthTi
       {command: `trap '' TERM; printf '%s\\n' '{"event":"shutdown"}' >> ${JSON.stringify(lifecyclePath)}; kill -TERM "$ROLLBRIDGE_TEST_DAEMON_PID"; printf '{"event":"started","pid":%s,"processId":"database","replicaIndex":"0"}\\n' "$$" >> ${JSON.stringify(lifecyclePath)}; read ignored < ${JSON.stringify(gatePath)}`, env: lifecycleEnv, id: "database", policy: "service"},
       {command: `exec ${command}`, env: lifecycleEnv, id: "worker", policy: "companion", replicas: 2},
       {command: `exec ${command}`, env: lifecycleEnv, health: {path: healthPath, timeoutMs: healthTimeoutMs}, id: "web", policy: "proxied", port: {from: 0, to: 0}}
+    ] : attemptOwnedProcesses ? [
+      {command: ownedCommand, env: lifecycleEnv, id: "database", policy: "service"},
+      {command: ownedCommand, env: lifecycleEnv, id: "worker", policy: "companion"},
+      {command: ownedCommand, env: lifecycleEnv, id: "scheduler", policy: "singleton"},
+      {command: ownedWebCommand, env: webEnv, health: {path: healthPath, timeoutMs: healthTimeoutMs}, id: "web", policy: "proxied", port: {from: processPort, to: processPort}}
     ] : [{command, env: webEnv, health: {path: healthPath, timeoutMs: healthTimeoutMs}, id: "web", policy: "proxied", port: {from: 0, to: 0}}],
-    proxy: {host: "127.0.0.1", port: 0},
+    proxy: {host: "127.0.0.1", port: proxyPort},
     ...(persistState ? {statePath} : {})
   }
 
@@ -360,7 +448,7 @@ async function createFixture({healthGate = false, healthPath = "/ping", healthTi
   }
 
   await fs.writeFile(configPath, `${setup}module.exports = ${JSON.stringify(config, null, 2)}\n`)
-  return {configPath, gatePath, healthGatePath, lifecyclePath, root, socketPath, startedPath, statePath, stoppedPath}
+  return {configPath, gatePath, healthGatePath, lifecyclePath, processPort, proxyPort, root, socketPath, startedPath, statePath, stoppedPath}
 }
 
 /**
@@ -395,6 +483,12 @@ async function waitForFile(filePath) {
   const watcher = fs.watch(path.dirname(filePath))
 
   try {
+    try {
+      return await fs.readFile(filePath, "utf8")
+    } catch (error) {
+      if (!error || typeof error !== "object" || !("code" in error) || error.code !== "ENOENT") throw error
+    }
+
     for await (const event of watcher) {
       if (event.filename === path.basename(filePath)) return await fs.readFile(filePath, "utf8")
     }
@@ -420,26 +514,143 @@ function spawnDaemon(fixture, release) {
 
 /**
  * @param {string[]} args - Daemon arguments.
+ * @param {{timeoutMs?: number}} [options] - Process execution options.
  * @returns {Promise<{code: number | null, output: string, stderr: string}>} Completed process result.
  */
-async function runDaemon(args) {
-  return await runRollbridge(["daemon", ...args])
+async function runDaemon(args, options) {
+  return await runRollbridge(["daemon", ...args], options)
 }
 
 /**
  * @param {string[]} args - Rollbridge command and arguments.
+ * @param {{timeoutMs?: number}} [options] - Process execution options.
  * @returns {Promise<{code: number | null, output: string, stderr: string}>} Completed process result.
  */
-async function runRollbridge(args) {
+async function runRollbridge(args, {timeoutMs} = {}) {
   const child = spawn(process.execPath, [binPath, ...args], {stdio: ["ignore", "pipe", "pipe"]})
   let output = ""
   let stderr = ""
 
   child.stdout.setEncoding("utf8").on("data", (chunk) => { output += chunk })
   child.stderr.setEncoding("utf8").on("data", (chunk) => { stderr += chunk })
-  const [code] = await once(child, "exit")
+  let timer
+
+  const exited = once(child, "exit")
+  const result = timeoutMs === undefined ? await exited : await Promise.race([
+    exited,
+    new Promise((resolve) => {
+      timer = setTimeout(() => {
+        child.kill("SIGKILL")
+        resolve(["timeout"])
+      }, timeoutMs)
+    })
+  ])
+
+  if (timer) clearTimeout(timer)
+  if (result[0] === "timeout") {
+    await exited
+    throw new Error(`Rollbridge did not terminate within ${timeoutMs}ms`)
+  }
+
+  const [code] = result
+
+  return {code: typeof code === "number" || code === null ? code : null, output, stderr}
+}
+
+/**
+ * @param {string} output - JSON-lines daemon output.
+ * @returns {StructuredRecord[]} Parsed records.
+ */
+function parseStructuredOutput(output) {
+  return output.split("\n").filter(Boolean).map((line) => JSON.parse(line))
+}
+
+/**
+ * @param {{lifecyclePath: string, processPort: number, proxyPort: number}} fixture - Attempt fixture.
+ * @param {StructuredRecord[]} records - Structured daemon records.
+ * @returns {Promise<void>} Resolves after all owned resources are verified stopped.
+ */
+async function assertAttemptResourcesStopped(fixture, records) {
+  const events = await readLifecycleEvents(fixture.lifecyclePath)
+  const started = events.filter((event) => event.event === "started")
+  const stoppedPids = new Set(events.filter((event) => event.event === "stopped").map((event) => event.pid))
+  const expectedProcessIds = new Set(["database", "scheduler", "web", "worker"])
+  const managedStarts = new Set(records.filter((record) => record.message === "process started").map((record) => record.data?.processId))
+  const managedExits = new Set(records.filter((record) => record.message === "process exited").map((record) => record.data?.processId))
+
+  assert.deepEqual(managedStarts, expectedProcessIds)
+  assert.deepEqual(managedExits, expectedProcessIds)
+  for (const event of started) {
+    assert.equal(stoppedPids.has(event.pid), true, `${event.processId} must receive graceful shutdown`)
+    assert.equal(isProcessAlive(Number(event.pid)), false, `${event.processId} pid ${event.pid} must be gone before daemon exit`)
+  }
+
+  await assertPortAvailable(fixture.processPort)
+  await assertPortAvailable(fixture.proxyPort)
+}
+
+/**
+ * @param {string} lifecyclePath - Fixture lifecycle log.
+ * @returns {Promise<Record<string, import("../src/json.js").JsonValue>[]>} Parsed lifecycle events.
+ */
+async function readLifecycleEvents(lifecyclePath) {
+  return (await fs.readFile(lifecyclePath, "utf8")).trim().split("\n").filter(Boolean).map((line) => JSON.parse(line))
+}
+
+/**
+ * Best-effort cleanup for a deliberately failing lingering-process regression.
+ * @param {string} lifecyclePath - Fixture lifecycle log.
+ * @returns {Promise<void>} Resolves after known fixture process groups are stopped.
+ */
+async function killAttemptProcesses(lifecyclePath) {
+  try {
+    const events = await readLifecycleEvents(lifecyclePath)
+
+    for (const event of events) {
+      const pid = Number(event.pid)
+
+      if (event.event === "started" && isProcessAlive(pid)) process.kill(-pid, "SIGKILL")
+    }
+  } catch {
+    // The attempt can fail before creating the lifecycle file.
+  }
+}
+
+/**
+ * @param {number} port - Local port expected to be free.
+ * @returns {Promise<void>} Resolves after binding and releasing the port.
+ */
+async function assertPortAvailable(port) {
+  const server = net.createServer()
+
+  await new Promise((resolve, reject) => {
+    server.once("error", reject)
+    server.listen(port, "127.0.0.1", () => resolve(undefined))
+  })
+  await new Promise((resolve) => server.close(() => resolve(undefined)))
+}
+
+/**
+ * Reserves distinct ephemeral ports until all have been observed, then releases them.
+ * @param {number} count - Number of ports.
+ * @returns {Promise<number[]>} Available port numbers.
+ */
+async function availablePorts(count) {
+  const servers = Array.from({length: count}, () => net.createServer())
+
+  await Promise.all(servers.map((server) => new Promise((resolve, reject) => {
+    server.once("error", reject)
+    server.listen(0, "127.0.0.1", () => resolve(undefined))
+  })))
+  const ports = servers.map((server) => {
+    const address = server.address()
+
+    if (!address || typeof address === "string") throw new Error("Expected a TCP server address")
+    return address.port
+  })
 
-  return {code, output, stderr}
+  await Promise.all(servers.map((server) => new Promise((resolve) => server.close(() => resolve(undefined)))))
+  return ports
 }
 
 /**

package/test/fixtures/owned-child.js

@@ -0,0 +1,19 @@
+// @ts-check
+
+import fs from "node:fs"
+
+const lifecyclePath = process.env.ROLLBRIDGE_TEST_LIFECYCLE_PATH
+
+if (!lifecyclePath) throw new Error("ROLLBRIDGE_TEST_LIFECYCLE_PATH is required")
+
+/** @param {string} event - Lifecycle event. */
+const record = (event) => {
+  fs.appendFileSync(lifecyclePath, `${JSON.stringify({event, pid: process.pid, processId: process.env.ROLLBRIDGE_PROCESS_ID})}\n`)
+}
+
+record("started")
+process.on("SIGTERM", () => {
+  record("stopped")
+  process.exit(0)
+})
+setInterval(() => {}, 1000)

package/test/managed-process.test.js

@@ -401,18 +401,18 @@ test("sends the configured stopSignal as the graceful stop signal", async () =>
 
   /** @type {string[]} */
   const signals = []
-  const killProcessGroup = managed.killProcessGroup.bind(managed)
+  const killProcess = managed.killProcess.bind(managed)
 
-  managed.killProcessGroup = (signal) => {
+  managed.killProcess = (pid, signal) => {
     signals.push(signal)
-    killProcessGroup(signal)
+    killProcess(pid, signal)
   }
 
   await managed.start()
   await managed.stop()
 
   // The graceful stop uses the configured signal (a SIGKILL fallback, if any, comes after).
-  assert.equal(signals[0], "SIGINT")
+  assert.deepEqual(signals, ["SIGINT", "SIGINT"])
   assert.equal(managed.status().state, "stopped")
 })
 
@@ -431,11 +431,11 @@ test("indefinite stop waits for the process to exit without SIGKILL", async () =
   })
   /** @type {string[]} */
   const signals = []
-  const killProcessGroup = managed.killProcessGroup.bind(managed)
+  const killProcess = managed.killProcess.bind(managed)
 
-  managed.killProcessGroup = (signal) => {
+  managed.killProcess = (pid, signal) => {
     signals.push(signal)
-    killProcessGroup(signal)
+    killProcess(pid, signal)
   }
 
   try {
@@ -443,7 +443,7 @@ test("indefinite stop waits for the process to exit without SIGKILL", async () =
     await managed.stop()
 
     assert.equal(managed.status().state, "stopped")
-    assert.deepEqual(signals, ["SIGTERM"])
+    assert.deepEqual(signals, ["SIGTERM", "SIGTERM"])
   } finally {
     await managed.stop()
   }
@@ -491,6 +491,42 @@ test("stop waits for process group descendants after the detached shell exits",
   }
 })
 
+test("stop does not return while a gracefully stopped descendant remains unreaped", async () => {
+  const dir = fs.mkdtempSync(path.join(os.tmpdir(), "rollbridge-process-group-"))
+  const pidPath = path.join(dir, "child.pid")
+  const child = [
+    "const fs = require('node:fs')",
+    `fs.writeFileSync(${JSON.stringify(pidPath)}, String(process.pid))`,
+    "process.on('SIGTERM', () => setTimeout(() => process.exit(0), 50))",
+    "setInterval(() => {}, 1000)"
+  ].join("; ")
+  const managed = new ManagedProcess({
+    command: `trap 'exit 0' TERM; ${JSON.stringify(process.execPath)} -e ${JSON.stringify(child)} & wait`,
+    cwd: undefined,
+    env: {},
+    id: "worker",
+    logger: () => {},
+    outputLines: 50,
+    restartDelayMs: 10,
+    shouldRestart: () => false,
+    stopSignal: "SIGTERM",
+    stopTimeoutMs: 2000
+  })
+
+  try {
+    await managed.start()
+    await waitFor(() => fs.existsSync(pidPath))
+    const childPid = Number(fs.readFileSync(pidPath, "utf8"))
+
+    await managed.stop()
+
+    assert.throws(() => process.kill(childPid, 0), {code: "ESRCH"})
+  } finally {
+    await managed.stop()
+    fs.rmSync(dir, {force: true, recursive: true})
+  }
+})
+
 test("a memory restart respawns and is counted when the supervisor still wants the process", async () => {
   const managed = buildLongLived(() => true)
 

package/test/rollbridge.test.js

@@ -855,6 +855,14 @@ test("a process over its memory limit is restarted with reason memory", {skip: p
     assert.equal(hog.lastStartReason, "memory")
     assert.equal(typeof hog.lastMemoryRestartAt, "string")
 
+    // Keep the replacement alive long enough to observe its next monitor sample. The fixture
+    // remains over the configured limit after every launch, otherwise it can restart again and
+    // clear rssBytes/children before this polling loop observes them on slower CI runners.
+    const hogProcess = daemon.activeRelease?.processes.get("hog")
+
+    assert.ok(hogProcess?.memory)
+    hogProcess.memory.limitBytes = Number.MAX_SAFE_INTEGER
+
     // rssBytes is sampled on the monitor's interval; wait for a measurement of the running process.
     await waitFor(() => {
       const rssBytes = activeProcessStatus(daemon, "hog")?.rssBytes

package/test/shutdown-completion.test.js

@@ -115,6 +115,43 @@ test("shutdown response waits for endpoint and owned-process cleanup before imme
   }
 })
 
+test("external-owner retirement releases listeners before a long-draining companion exits", async () => {
+  const root = await fs.mkdtemp(path.join(os.tmpdir(), "rollbridge-owner-retirement-"))
+  const socketPath = path.join(root, "rollbridge.sock")
+  const gatePath = path.join(root, "retire.fifo")
+  const stoppingPath = path.join(root, "stopping")
+  const gate = spawn("mkfifo", [gatePath])
+  assert.equal((await once(gate, "exit"))[0], 0)
+  const config = buildConfig(socketPath, {companion: {
+    command: `${JSON.stringify(process.execPath)} -e ${JSON.stringify("setInterval(() => {}, 1000)")}`,
+    id: "worker",
+    lifecycle: {drainCommand: `read released < ${JSON.stringify(gatePath)}`, drainTimeoutMs: 60000, quietCommand: `printf stopping > ${JSON.stringify(stoppingPath)}`},
+    policy: "companion"
+  }})
+  const daemon = new RollbridgeDaemon({config, logger: () => {}})
+  let replacement
+
+  try {
+    await daemon.start()
+    await daemon.deploy({releaseId: "old", releasePath: root, revision: "old"})
+    const retirement = sendControlCommand({command: {attestation: `sha256:${"a".repeat(64)}`, command: "retire-owner"}, path: socketPath})
+    const response = await retirement
+    assert.deepEqual(response, {message: "owner retired", status: "success"})
+    assert.equal(await fs.readFile(stoppingPath, "utf8"), "stopping", "old worker must stop accepting work before listener takeover")
+
+    replacement = new RollbridgeDaemon({config, logger: () => {}})
+    await replacement.start({reportOrphans: false})
+    assert.equal((await sendControlCommand({command: {command: "status"}, path: socketPath})).application, "shutdown-target")
+    assert.deepEqual(replacement.status().orphans, [], "intentional retired companions are not replacement orphans")
+    assert.equal(daemon.status().releases[0].processes[0].state, "stopping")
+  } finally {
+    await fs.writeFile(gatePath, "done\n").catch(() => {})
+    if (replacement) await replacement.shutdown()
+    await daemon.shutdown()
+    await fs.rm(root, {force: true, recursive: true})
+  }
+})
+
 test("control socket unlink failure is reported only after owned cleanup completes", async () => {
   const root = await fs.mkdtemp(path.join(os.tmpdir(), "rollbridge-shutdown-unlink-failure-"))
   const socketPath = path.join(root, "control.sock")