rollbridge 0.1.42 → 0.1.44

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "rollbridge",
-  "version": "0.1.42",
+  "version": "0.1.44",
   "description": "Zero-downtime process supervisor and local traffic switcher for deploy-managed apps.",
   "keywords": [
     "deploy",

package/src/daemon.js

@@ -505,7 +505,10 @@ export default class RollbridgeDaemon {
       if (!transfer?.config || !transfer.snapshot) throw new Error("Committed owner published incomplete replacement state")
       const unresolvedTransition = transfer.snapshot.generationTransition
 
-      if (unresolvedTransition && unresolvedTransition.phase !== "committed" && unresolvedTransition.configDigest !== this.ownerRecoveryConfigDigest()) {
+      const transferredTransitionConfigDigest = transfer.config ? ownerConfigDigest(transfer.config) : undefined
+      if (unresolvedTransition && unresolvedTransition.phase !== "committed" &&
+        unresolvedTransition.configDigest !== this.ownerRecoveryConfigDigest() &&
+        unresolvedTransition.configDigest !== transferredTransitionConfigDigest) {
         throw new Error(`Owner replacement cannot change config authority while unresolved generation transition ${unresolvedTransition.candidateReleaseId} remains at ${unresolvedTransition.phase}`)
       }
       const registeredProcesses = new Map((await this.guardian.inventory()).map(({key, provenance}) => [key, provenance]))
@@ -1814,7 +1817,7 @@ export default class RollbridgeDaemon {
     }
 
     if (transition.phase === "restoring_previous") {
-      if (candidate.state !== "draining") throw new Error(`Failed candidate ${candidate.releaseId} is not retired before incumbent restoration`)
+      if (candidate.state !== "draining" && candidate.state !== "stopped") throw new Error(`Failed candidate ${candidate.releaseId} is not retired before incumbent restoration`)
       try {
         await previous.reactivateGeneration()
       } catch (error) {
@@ -1914,7 +1917,14 @@ export default class RollbridgeDaemon {
    * @param {{config: import("./config.js").RollbridgeConfig, releaseId: string, releasePath: string, revision: string}} candidate - Requested identity.
    */
   assertExactGenerationTransition(transition, candidate) {
-    if (transition.candidateReleaseId !== candidate.releaseId || transition.candidateReleasePath !== candidate.releasePath || transition.candidateRevision !== candidate.revision || transition.configDigest !== ownerConfigDigest(candidate.config)) {
+    const retainedCandidate = this.releases.get(transition.candidateReleaseId)
+    const retainedAuthorityMatches = transition.phase !== "committed" &&
+      retainedCandidate?.releasePath === transition.candidateReleasePath &&
+      retainedCandidate.revision === transition.candidateRevision &&
+      ownerConfigDigest(retainedCandidate.config) === transition.configDigest
+    const requestedAuthorityMatches = ownerConfigDigest(candidate.config) === transition.configDigest
+
+    if (transition.candidateReleaseId !== candidate.releaseId || transition.candidateReleasePath !== candidate.releasePath || transition.candidateRevision !== candidate.revision || (!requestedAuthorityMatches && !retainedAuthorityMatches)) {
       throw new Error(`Release generation transition for ${transition.candidateReleaseId} is unresolved at ${transition.phase}; only the exact same release, path, revision, and config authority may resume it`)
     }
   }

package/src/release-group.js

@@ -290,10 +290,22 @@ export default class ReleaseGroup extends EventEmitter {
     if (!coordinator) throw new Error(`Generation activation process ${processConfig.id} is not retained for release ${this.releaseId}`)
     const generationIds = new Set([...this.handoffServiceIds, ...this.nonBlockingDrainIds])
 
+    /** @param {import("./managed-process.js").default} processInstance - Retained generation process. */
+    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")
+      else throw new Error(`Generation process is ${state}; refusing active-role restoration`)
+    }
+
     for (const [id, processInstance] of this.processes) {
-      if (generationIds.has(id) && processInstance !== coordinator.process) await processInstance.reactivateStrict()
+      if (generationIds.has(id) && processInstance !== coordinator.process) await reactivate(processInstance)
     }
-    await coordinator.process.reactivateStrict()
+    await reactivate(coordinator.process)
     this.state = "active"
     this.activatedAt = new Date().toISOString()
     this.drainStartedAt = undefined

package/test/rollbridge.test.js

@@ -696,8 +696,16 @@ test("candidate activation failure reports restoration failure and exact recover
     assert.match(String(restorationEvent?.data.error), /incumbent restoration rejected/)
     assert.deepEqual(await lifecycleEvents(fixture.lifecycleLogPath), ["activate:v1", "retire:v1", "retire:v2"])
     await assert.rejects(() => daemon.deploy({releaseId: "v3", releasePath: fixture.root, revision: "v3"}), /transition.*v2.*unresolved/i)
+    const failedCandidate = daemon.releases.get("v2")
 
+    assert.ok(failedCandidate)
+    await failedCandidate.stop()
     incumbentCoordinator.reactivateStrict = reactivate
+    await incumbentCoordinator.stop()
+    assert.equal(failedCandidate.state, "stopped")
+    assert.equal(incumbentCoordinator.status().state, "stopped")
+    daemon.config = structuredClone(daemon.config)
+    daemon.config.processes[0].lifecycle.activateTimeoutMs = (daemon.config.processes[0].lifecycle.activateTimeoutMs ?? 30000) + 1
     const recovery = await sendControlCommand({
       command: {
         command: "recover-generation-transition",