From acec5e4367392bab2dbfad33a88f2c3b5a0853ef Mon Sep 17 00:00:00 2001
From: npc0-hue <github@npc0.com>
Date: Thu, 6 Aug 2026 18:18:20 +0800
Subject: [PATCH] Allow restarting failed server instances

---
 platform/service/server_lifecycle.go          |  1 +
 platform/service/server_lifecycle_test.go     | 17 +++++++++++++++++
 .../contracts/serverManagement.test.ts        | 19 +++++++++++++++++++
 platform_web/contracts/serverManagement.ts    |  2 +-
 4 files changed, 38 insertions(+), 1 deletion(-)
 create mode 100644 platform_web/contracts/serverManagement.test.ts

diff --git a/platform/service/server_lifecycle.go b/platform/service/server_lifecycle.go
index 9fcb707..3386487 100644
--- a/platform/service/server_lifecycle.go
+++ b/platform/service/server_lifecycle.go
@@ -173,6 +173,7 @@ func (svc *CoreService) StartServerInstance(command domain.ServerLifecycleComman
 	return svc.dispatchExistingServerLifecycle(command, domain.ServerLifecycleActionStart, []domain.ServerInstanceState{
 		domain.ServerInstanceStateReady,
 		domain.ServerInstanceStateStopped,
+		domain.ServerInstanceStateFailed,
 	})
 }
 
diff --git a/platform/service/server_lifecycle_test.go b/platform/service/server_lifecycle_test.go
index e76238b..150f1d4 100644
--- a/platform/service/server_lifecycle_test.go
+++ b/platform/service/server_lifecycle_test.go
@@ -283,6 +283,23 @@ func TestCoreServiceLifecycleFailureProjectsFailedState(t *testing.T) {
 	if instance.State != domain.ServerInstanceStateFailed {
 		t.Fatalf("expected failed install result to mark failed, got %+v", instance)
 	}
+
+	started, err := svc.StartServerInstance(domain.ServerLifecycleCommand{
+		ServerInstanceID:      instance.ID,
+		ExpectedConfigVersion: instance.ConfigVersion,
+		IdempotencyKey:        "idem-start-after-failure",
+	})
+	if err != nil || started.Job.Capability != domain.LifecycleCapabilityStart || started.Job.TargetKey != "actions/start.json" {
+		t.Fatalf("failed server should allow explicit restart dispatch, result=%+v err=%v", started, err)
+	}
+	claimAndCompleteLifecycleJob(t, svc, sessionToken, domain.LifecycleCapabilityStart, domain.JobStateSucceeded)
+	recovered, err := svc.GetServerInstance("server-1")
+	if err != nil {
+		t.Fatalf("get recovered instance: %v", err)
+	}
+	if recovered.State != domain.ServerInstanceStateRunning {
+		t.Fatalf("expected explicit restart to recover failed instance, got %+v", recovered)
+	}
 }
 
 func TestCoreServicePluginLifecycleManagesMultipleInstancesIndependently(t *testing.T) {
diff --git a/platform_web/contracts/serverManagement.test.ts b/platform_web/contracts/serverManagement.test.ts
new file mode 100644
index 0000000..58e426b
--- /dev/null
+++ b/platform_web/contracts/serverManagement.test.ts
@@ -0,0 +1,19 @@
+import { describe, expect, it } from "vitest";
+
+import { canStartServer, canStopServer } from "./serverManagement";
+import type { ServerInstanceState } from "../api/types";
+
+describe("server management lifecycle contracts", () => {
+  it("allows explicit starts from recoverable non-running states", () => {
+    const startable: ServerInstanceState[] = ["ready", "stopped", "failed"];
+    const blocked: ServerInstanceState[] = ["draft", "installing", "running", "deleted"];
+
+    expect(startable.every(canStartServer)).toBe(true);
+    expect(blocked.some(canStartServer)).toBe(false);
+  });
+
+  it("keeps stop limited to running servers", () => {
+    expect(canStopServer("running")).toBe(true);
+    expect(canStopServer("failed")).toBe(false);
+  });
+});
diff --git a/platform_web/contracts/serverManagement.ts b/platform_web/contracts/serverManagement.ts
index cf5a29e..bb41b9a 100644
--- a/platform_web/contracts/serverManagement.ts
+++ b/platform_web/contracts/serverManagement.ts
@@ -108,7 +108,7 @@ export function endpointLabel(endpoint: RunEndpointResponse | undefined, runEndp
 }
 
 export function canStartServer(state: ServerInstanceState): boolean {
-  return state === "ready" || state === "stopped";
+	return state === "ready" || state === "stopped" || state === "failed";
 }
 
 export function canStopServer(state: ServerInstanceState): boolean {