rollbridge 0.1.44 → 0.1.45

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "rollbridge",
-  "version": "0.1.44",
+  "version": "0.1.45",
   "description": "Zero-downtime process supervisor and local traffic switcher for deploy-managed apps.",
   "keywords": [
     "deploy",

package/src/managed-process.js

@@ -531,7 +531,7 @@ export default class ManagedProcess extends EventEmitter {
     const child = this.child
     const pid = this.pid
 
-    if (!child?.pid || child.pid !== pid || (this.state !== "quiesced" && this.state !== "running")) {
+    if (!child?.pid || child.pid !== pid || (this.state !== "failed" && this.state !== "quiesced" && this.state !== "running")) {
       throw new Error(`Process ${this.id} is not retained for reactivation`)
     }
     await this.runReactivationHook(pid)

package/src/release-group.js

@@ -294,11 +294,8 @@ export default class ReleaseGroup extends EventEmitter {
     const reactivate = async (processInstance) => {
       const {pid, state} = processInstance.status()
 
-      if (state === "failed") {
-        await processInstance.stop()
-        await processInstance.start("crash", "active")
-      } else if (pid !== undefined) await processInstance.reactivateStrict()
-      else if (state === "stopped") await processInstance.start("crash", "active")
+      if (pid !== undefined) await processInstance.reactivateStrict()
+      else if (state === "stopped" || state === "failed") await processInstance.start("crash", "active")
       else throw new Error(`Generation process is ${state}; refusing active-role restoration`)
     }
 

package/test/managed-process.test.js

@@ -532,6 +532,31 @@ test("reactivateStrict restores a retained quiesced process only after activatio
   }
 })
 
+test("reactivateStrict retries a failed active-role startup against the retained child", async () => {
+  const managed = buildLongLived(() => false)
+  let attempts = 0
+
+  managed.lifecycle = {activateCommand: "jobs activate", drainTimeoutMs: 0}
+  managed.runHook = async () => {
+    attempts += 1
+    return attempts === 1 ? new Error("startup activation raced readiness") : undefined
+  }
+
+  try {
+    await assert.rejects(() => managed.start("deploy", "active"), /startup activation raced readiness/)
+    const failed = managed.status()
+
+    assert.equal(failed.state, "failed")
+    assert.ok(failed.pid)
+    await managed.reactivateStrict()
+    assert.equal(managed.status().state, "running")
+    assert.equal(managed.status().lifecycleRole, "active")
+    assert.equal(managed.status().pid, failed.pid)
+  } finally {
+    await managed.stop()
+  }
+})
+
 test("quiesce waits for active-role restoration before retiring a restarted process", async () => {
   const managed = buildLongLived(() => false)
   const hooks = /** @type {string[]} */ ([])