fix(gateway): auto-repair unloaded LaunchAgent on start/restart
When a macOS LaunchAgent silently unloads after sleep or extended idle, `gateway start` and `gateway restart` would bail with 'not loaded' hints instead of attempting recovery — even though the plist still exists on disk and `repairLaunchAgentBootstrap()` can re-register it. Changes: - Add optional `repairNotLoaded` method to `GatewayService` interface - Wire darwin implementation: checks plist exists, calls `repairLaunchAgentBootstrap()` (enable + bootstrap + kickstart) - `runServiceStart()`: when service is not loaded but repair succeeds, proceed with normal start flow instead of printing install hints - `runServiceRestart()`: when `onNotLoaded` returns null (no running process to signal) but repair succeeds, proceed with restart flow - Both paths are best-effort: if repair fails or throws, falls through to existing not-loaded behavior (no regression) Fixes #43602
This commit is contained in:
parent
bcbfbb831e
commit
cdcb9464fb
@ -190,4 +190,141 @@ describe("runServiceRestart token drift", () => {
|
||||
expect(payload.result).toBe("scheduled");
|
||||
expect(payload.message).toBe("restart scheduled, gateway will restart momentarily");
|
||||
});
|
||||
|
||||
describe("repairNotLoaded (#43602)", () => {
|
||||
it("start: repairs unloaded service when repairNotLoaded succeeds", async () => {
|
||||
service.isLoaded.mockResolvedValue(false);
|
||||
const repairNotLoaded = vi.fn().mockResolvedValue({ ok: true });
|
||||
const serviceWithRepair = { ...service, repairNotLoaded };
|
||||
|
||||
await runServiceStart({
|
||||
serviceNoun: "Gateway",
|
||||
service: serviceWithRepair,
|
||||
renderStartHints: () => [],
|
||||
opts: { json: true },
|
||||
});
|
||||
|
||||
expect(repairNotLoaded).toHaveBeenCalledTimes(1);
|
||||
// After successful repair, start should proceed to restart the service.
|
||||
expect(service.restart).toHaveBeenCalledTimes(1);
|
||||
const jsonLine = runtimeLogs.find((line) => line.trim().startsWith("{"));
|
||||
const payload = JSON.parse(jsonLine ?? "{}") as { result?: string };
|
||||
expect(payload.result).toBe("started");
|
||||
});
|
||||
|
||||
it("start: falls through to hints when repairNotLoaded returns ok:false", async () => {
|
||||
service.isLoaded.mockResolvedValue(false);
|
||||
const repairNotLoaded = vi.fn().mockResolvedValue({ ok: false });
|
||||
const serviceWithRepair = { ...service, repairNotLoaded };
|
||||
|
||||
await runServiceStart({
|
||||
serviceNoun: "Gateway",
|
||||
service: serviceWithRepair,
|
||||
renderStartHints: () => ["openclaw gateway install"],
|
||||
opts: { json: true },
|
||||
});
|
||||
|
||||
expect(repairNotLoaded).toHaveBeenCalledTimes(1);
|
||||
expect(service.restart).not.toHaveBeenCalled();
|
||||
const jsonLine = runtimeLogs.find((line) => line.trim().startsWith("{"));
|
||||
const payload = JSON.parse(jsonLine ?? "{}") as { result?: string; hints?: string[] };
|
||||
expect(payload.result).toBe("not-loaded");
|
||||
expect(payload.hints).toContain("openclaw gateway install");
|
||||
});
|
||||
|
||||
it("start: falls through to hints when repairNotLoaded throws", async () => {
|
||||
service.isLoaded.mockResolvedValue(false);
|
||||
const repairNotLoaded = vi.fn().mockRejectedValue(new Error("launchctl failed"));
|
||||
const serviceWithRepair = { ...service, repairNotLoaded };
|
||||
|
||||
await runServiceStart({
|
||||
serviceNoun: "Gateway",
|
||||
service: serviceWithRepair,
|
||||
renderStartHints: () => ["openclaw gateway install"],
|
||||
opts: { json: true },
|
||||
});
|
||||
|
||||
expect(repairNotLoaded).toHaveBeenCalledTimes(1);
|
||||
expect(service.restart).not.toHaveBeenCalled();
|
||||
const jsonLine = runtimeLogs.find((line) => line.trim().startsWith("{"));
|
||||
const payload = JSON.parse(jsonLine ?? "{}") as { result?: string };
|
||||
expect(payload.result).toBe("not-loaded");
|
||||
});
|
||||
|
||||
it("start: does not call repairNotLoaded when service is already loaded", async () => {
|
||||
service.isLoaded.mockResolvedValue(true);
|
||||
const repairNotLoaded = vi.fn().mockResolvedValue({ ok: true });
|
||||
const serviceWithRepair = { ...service, repairNotLoaded };
|
||||
|
||||
await runServiceStart({
|
||||
serviceNoun: "Gateway",
|
||||
service: serviceWithRepair,
|
||||
renderStartHints: () => [],
|
||||
opts: { json: true },
|
||||
});
|
||||
|
||||
expect(repairNotLoaded).not.toHaveBeenCalled();
|
||||
});
|
||||
|
||||
it("restart: repairs unloaded service when onNotLoaded returns null", async () => {
|
||||
service.isLoaded.mockResolvedValue(false);
|
||||
const repairNotLoaded = vi.fn().mockResolvedValue({ ok: true });
|
||||
const serviceWithRepair = { ...service, repairNotLoaded };
|
||||
|
||||
const result = await runServiceRestart({
|
||||
serviceNoun: "Gateway",
|
||||
service: serviceWithRepair,
|
||||
renderStartHints: () => [],
|
||||
opts: { json: true },
|
||||
onNotLoaded: async () => null,
|
||||
});
|
||||
|
||||
expect(result).toBe(true);
|
||||
expect(repairNotLoaded).toHaveBeenCalledTimes(1);
|
||||
const jsonLine = runtimeLogs.find((line) => line.trim().startsWith("{"));
|
||||
const payload = JSON.parse(jsonLine ?? "{}") as { result?: string; message?: string };
|
||||
expect(payload.result).toBe("restarted");
|
||||
expect(payload.message).toContain("re-registered");
|
||||
});
|
||||
|
||||
it("restart: skips repair when onNotLoaded handles it", async () => {
|
||||
service.isLoaded.mockResolvedValue(false);
|
||||
const repairNotLoaded = vi.fn().mockResolvedValue({ ok: true });
|
||||
const serviceWithRepair = { ...service, repairNotLoaded };
|
||||
|
||||
const result = await runServiceRestart({
|
||||
serviceNoun: "Gateway",
|
||||
service: serviceWithRepair,
|
||||
renderStartHints: () => [],
|
||||
opts: { json: true },
|
||||
onNotLoaded: async () => ({
|
||||
result: "restarted" as const,
|
||||
message: "handled by SIGUSR1",
|
||||
}),
|
||||
});
|
||||
|
||||
expect(result).toBe(true);
|
||||
expect(repairNotLoaded).not.toHaveBeenCalled();
|
||||
});
|
||||
|
||||
it("restart: falls through to hints when repair returns ok:false", async () => {
|
||||
service.isLoaded.mockResolvedValue(false);
|
||||
const repairNotLoaded = vi.fn().mockResolvedValue({ ok: false });
|
||||
const serviceWithRepair = { ...service, repairNotLoaded };
|
||||
|
||||
const result = await runServiceRestart({
|
||||
serviceNoun: "Gateway",
|
||||
service: serviceWithRepair,
|
||||
renderStartHints: () => ["openclaw gateway install"],
|
||||
opts: { json: true },
|
||||
onNotLoaded: async () => null,
|
||||
});
|
||||
|
||||
expect(result).toBe(false);
|
||||
expect(repairNotLoaded).toHaveBeenCalledTimes(1);
|
||||
const jsonLine = runtimeLogs.find((line) => line.trim().startsWith("{"));
|
||||
const payload = JSON.parse(jsonLine ?? "{}") as { result?: string };
|
||||
expect(payload.result).toBe("not-loaded");
|
||||
});
|
||||
});
|
||||
});
|
||||
|
||||
@ -194,7 +194,7 @@ export async function runServiceStart(params: {
|
||||
const json = Boolean(params.opts?.json);
|
||||
const { stdout, emit, fail } = createActionIO({ action: "start", json });
|
||||
|
||||
const loaded = await resolveServiceLoadedOrFail({
|
||||
let loaded = await resolveServiceLoadedOrFail({
|
||||
serviceNoun: params.serviceNoun,
|
||||
service: params.service,
|
||||
fail,
|
||||
@ -202,6 +202,25 @@ export async function runServiceStart(params: {
|
||||
if (loaded === null) {
|
||||
return;
|
||||
}
|
||||
if (!loaded && params.service.repairNotLoaded) {
|
||||
// The service was previously installed but is no longer loaded (e.g.
|
||||
// macOS LaunchAgent silently unloaded after sleep/idle). Attempt to
|
||||
// re-register the existing service definition before falling through
|
||||
// to the "not loaded" install hints. See #43602.
|
||||
try {
|
||||
const repair = await params.service.repairNotLoaded({ env: process.env });
|
||||
if (repair.ok) {
|
||||
loaded = true;
|
||||
if (!json) {
|
||||
defaultRuntime.log(
|
||||
`${params.serviceNoun} was not loaded — re-registered from existing service definition.`,
|
||||
);
|
||||
}
|
||||
}
|
||||
} catch {
|
||||
// Best-effort repair; fall through to normal not-loaded handling.
|
||||
}
|
||||
}
|
||||
if (!loaded) {
|
||||
await handleServiceNotLoaded({
|
||||
serviceNoun: params.serviceNoun,
|
||||
@ -356,7 +375,7 @@ export async function runServiceRestart(params: {
|
||||
return true;
|
||||
};
|
||||
|
||||
const loaded = await resolveServiceLoadedOrFail({
|
||||
let loaded = await resolveServiceLoadedOrFail({
|
||||
serviceNoun: params.serviceNoun,
|
||||
service: params.service,
|
||||
fail,
|
||||
@ -384,6 +403,26 @@ export async function runServiceRestart(params: {
|
||||
fail(`${params.serviceNoun} restart failed: ${String(err)}`);
|
||||
return false;
|
||||
}
|
||||
if (!handledNotLoaded && params.service.repairNotLoaded) {
|
||||
// No running process to signal, but the service definition may still
|
||||
// exist on disk (e.g. macOS LaunchAgent unloaded after sleep/idle).
|
||||
// Re-register it so `restart` can proceed normally. See #43602.
|
||||
try {
|
||||
const repair = await params.service.repairNotLoaded({ env: process.env });
|
||||
if (repair.ok) {
|
||||
loaded = true;
|
||||
handledNotLoaded = {
|
||||
result: "restarted",
|
||||
message: `${params.serviceNoun} was not loaded — re-registered from existing service definition.`,
|
||||
};
|
||||
if (!json) {
|
||||
defaultRuntime.log(handledNotLoaded.message);
|
||||
}
|
||||
}
|
||||
} catch {
|
||||
// Best-effort repair; fall through to normal not-loaded handling.
|
||||
}
|
||||
}
|
||||
if (!handledNotLoaded) {
|
||||
await handleServiceNotLoaded({
|
||||
serviceNoun: params.serviceNoun,
|
||||
|
||||
@ -1,8 +1,10 @@
|
||||
import {
|
||||
installLaunchAgent,
|
||||
isLaunchAgentLoaded,
|
||||
launchAgentPlistExists,
|
||||
readLaunchAgentProgramArguments,
|
||||
readLaunchAgentRuntime,
|
||||
repairLaunchAgentBootstrap,
|
||||
restartLaunchAgent,
|
||||
stopLaunchAgent,
|
||||
uninstallLaunchAgent,
|
||||
@ -64,6 +66,16 @@ export type GatewayService = {
|
||||
isLoaded: (args: GatewayServiceEnvArgs) => Promise<boolean>;
|
||||
readCommand: (env: GatewayServiceEnv) => Promise<GatewayServiceCommandConfig | null>;
|
||||
readRuntime: (env: GatewayServiceEnv) => Promise<GatewayServiceRuntime>;
|
||||
/**
|
||||
* Attempt to re-register and start a service that was previously installed
|
||||
* but is no longer loaded (e.g. after macOS sleep/idle unloads the
|
||||
* LaunchAgent). Returns `{ ok: true }` when the service was successfully
|
||||
* re-bootstrapped, `{ ok: false }` when the service definition does not
|
||||
* exist on disk (caller should fall through to install hints).
|
||||
*
|
||||
* Optional — platforms that do not experience silent unloads can omit this.
|
||||
*/
|
||||
repairNotLoaded?: (args: GatewayServiceEnvArgs) => Promise<{ ok: boolean; detail?: string }>;
|
||||
};
|
||||
|
||||
export function describeGatewayServiceRestart(
|
||||
@ -105,6 +117,14 @@ const GATEWAY_SERVICE_REGISTRY: Record<SupportedGatewayServicePlatform, GatewayS
|
||||
isLoaded: isLaunchAgentLoaded,
|
||||
readCommand: readLaunchAgentProgramArguments,
|
||||
readRuntime: readLaunchAgentRuntime,
|
||||
repairNotLoaded: async (args) => {
|
||||
const env = args.env ?? (process.env as Record<string, string | undefined>);
|
||||
const plistExists = await launchAgentPlistExists(env);
|
||||
if (!plistExists) {
|
||||
return { ok: false, detail: "plist not found on disk" };
|
||||
}
|
||||
return await repairLaunchAgentBootstrap({ env });
|
||||
},
|
||||
},
|
||||
linux: {
|
||||
label: "systemd",
|
||||
|
||||
Loading…
x
Reference in New Issue
Block a user