Introduce workaround that fixes windows dns issue on balena push using .local device names.

Improve error handling in deployToDevice so that versionErrors don't mask other errors.

Resolves:#1518
Change-type:patch
Signed-off-by:Scott Lowe <scott@balena.io>

lib/utils/device/deploy.ts

@@ -30,9 +30,11 @@ import { Readable } from 'stream';
 
 import { BALENA_ENGINE_TMP_PATH } from '../../config';
 import { checkBuildSecretsRequirements, makeBuildTasks } from '../compose_ts';
+import { workaroundWindowsDnsIssue } from '../helpers';
 import Logger = require('../logger');
 import { DeviceAPI, DeviceInfo } from './api';
 import * as LocalPushErrors from './errors';
+import { DeviceAPIError } from './errors';
 import LivepushManager from './live';
 import { displayBuildLog } from './logs';
 
@@ -74,7 +76,7 @@ async function environmentFromInput(
 	const varRegex = /^(?:([^\s:]+):)?([^\s]+?)=(.*)$/;
 
 	const ret: ParsedEnvironment = {};
-	// Propolulate the object with the servicenames, as it
+	// Populate the object with the servicenames, as it
 	// also means that we can do a fast lookup of whether a
 	// service exists
 	for (const service of serviceNames) {
@@ -139,6 +141,8 @@ export async function deployToDevice(opts: DeviceDeployOptions): Promise<void> {
 		);
 	}
 
+	await workaroundWindowsDnsIssue(opts.deviceHost);
+
 	const versionError = new Error(
 		'The supervisor version on this remote device does not support multicontainer local mode. ' +
 			'Please update your device to balenaOS v2.20.0 or greater from the dashboard.',
@@ -156,10 +160,18 @@ export async function deployToDevice(opts: DeviceDeployOptions): Promise<void> {
 			);
 			opts.nolive = true;
 		}
-	} catch {
+	} catch (e) {
-		exitWithExpectedError(versionError);
+		// Very old supervisor versions do not support /version endpoint
+		// a DeviceAPIError is expected in this case
+		if (e instanceof DeviceAPIError) {
+			exitWithExpectedError(versionError);
+		} else {
+			throw e;
+		}
 	}
 
+	await workaroundWindowsDnsIssue(opts.deviceHost);
+
 	globalLogger.logInfo(`Starting build on device ${opts.deviceHost}`);
 
 	const project = await loadProject(
@@ -183,6 +195,8 @@ export async function deployToDevice(opts: DeviceDeployOptions): Promise<void> {
 	// Try to detect the device information
 	const deviceInfo = await api.getDeviceInformation();
 
+	await workaroundWindowsDnsIssue(opts.deviceHost);
+
 	let buildLogs: Dictionary<string> | undefined;
 	if (!opts.nolive) {
 		buildLogs = {};
@@ -218,6 +232,8 @@ export async function deployToDevice(opts: DeviceDeployOptions): Promise<void> {
 
 	await api.setTargetState(targetState);
 
+	await workaroundWindowsDnsIssue(opts.deviceHost);
+
 	// Now that we've set the target state, the device will do it's thing
 	// so we can either just display the logs, or start a livepush session
 	// (whilst also display logs)

lib/utils/helpers.ts

@@ -338,3 +338,21 @@ function windowsCmdExeEscapeArg(arg: string): string {
 	// duplicate internal double quotes, and double quote overall
 	return `"${arg.replace(/["]/g, '""')}"`;
 }
+
+/*
+ * Workaround a window system bug which causes multiple rapid DNS lookups
+ * to fail for mDNS.
+ *
+ * It introduces a simple pause, and should be used between operations that
+ * trigger mDNS resolutions.
+ *
+ * Windows bug: https://support.microsoft.com/en-gb/help/4057932/getaddrinfo-failed-with-wsahost-not-found-11001-error
+ */
+export async function workaroundWindowsDnsIssue(ipOrHostname: string) {
+	// 300ms seemed to be the smallest delay that worked reliably but may
+	// vary between systems.
+	const delay = 500;
+	if (process.platform === 'win32' && ipOrHostname.includes('.local')) {
+		await new Promise(r => setTimeout(r, delay));
+	}
+}