Merge pull request #2225 from balena-os/cache-last-reported-current-state

Cache last reported current state

src/api-binder/report.ts

@@ -2,11 +2,8 @@ import * as url from 'url';
 import * as _ from 'lodash';
 import { CoreOptions } from 'request';
 import { performance } from 'perf_hooks';
-
+import { setTimeout } from 'timers/promises';
-import { withBackoff, OnFailureInfo } from '../lib/backoff';
+import { readFile } from 'fs/promises';
-import { log } from '../lib/supervisor-console';
-import { InternalInconsistencyError, StatusError } from '../lib/errors';
-import { getRequestInstance } from '../lib/request';
 
 import { DeviceState } from '../types';
 import * as config from '../config';
@@ -14,8 +11,13 @@ import { SchemaTypeKey, SchemaReturn } from '../config/schema-type';
 import * as eventTracker from '../event-tracker';
 import * as deviceState from '../device-state';
 
+import { withBackoff, OnFailureInfo } from '../lib/backoff';
+import { log } from '../lib/supervisor-console';
+import { InternalInconsistencyError, StatusError } from '../lib/errors';
+import { getRequestInstance } from '../lib/request';
 import { shallowDiff, prune, empty } from '../lib/json';
-import { setTimeout } from 'timers/promises';
+import { pathOnRoot } from '../lib/host-utils';
+import { touch, writeAndSyncFile } from '../lib/fs-utils';
 
 let lastReport: DeviceState = {};
 let lastReportTime: number = -Infinity;
@@ -26,6 +28,8 @@ const maxReportFrequency = 10 * 1000;
 // How often can we report metrics to the server in ms; mirrors server setting.
 // Metrics are low priority, so less frequent than maxReportFrequency.
 const maxMetricsFrequency = 300 * 1000;
+// Path of the cache for last reported state
+const CACHE_PATH = pathOnRoot('/tmp/balena-supervisor/state-report-cache');
 
 // TODO: This counter is read by the healthcheck to see if the
 // supervisor is having issues to connect. We have removed the
@@ -109,6 +113,12 @@ async function reportCurrentState(opts: StateReportOpts, uuid: string) {
 		await report({ body: stateDiff, opts });
 		lastReportTime = performance.now();
 		lastReport = currentState;
+
+		// Cache last report so it survives Supervisor restart.
+		// On Supervisor startup, Supervisor will be able to diff between the
+		// cached report and thereby report less unnecessary data.
+		await cache(currentState);
+
 		log.info('Reported current state to the cloud');
 	};
 
@@ -128,6 +138,43 @@ async function reportCurrentState(opts: StateReportOpts, uuid: string) {
 	}
 }
 
+/**
+ * Cache last reported current state to CACHE_PATH
+ */
+async function cache(state: DeviceState) {
+	try {
+		await writeAndSyncFile(CACHE_PATH, JSON.stringify(state));
+	} catch (e: unknown) {
+		log.debug(`Failed to cache last reported state: ${(e as Error).message}`);
+	}
+}
+
+/**
+ * Get last cached state report from CACHE_PATH
+ */
+async function getCache(): Promise<DeviceState> {
+	try {
+		// Touch the file, which will create it if it doesn't exist
+		await touch(CACHE_PATH);
+
+		// Get last reported current state
+		const rawStateCache = await readFile(CACHE_PATH, 'utf-8');
+		const state = JSON.parse(rawStateCache);
+
+		// Return current state cache if valid
+		if (!DeviceState.is(state)) {
+			throw new Error();
+		}
+		log.debug('Retrieved last reported state from cache');
+		return state;
+	} catch {
+		log.debug(
+			'Could not retrieve last reported state from cache, proceeding with empty cache',
+		);
+		return {};
+	}
+}
+
 function handleRetry(retryInfo: OnFailureInfo) {
 	if (retryInfo.error instanceof StatusError) {
 		// We don't want these errors to be classed as a report error, as this will cause
@@ -166,6 +213,9 @@ export async function startReporting() {
 		throw new InternalInconsistencyError('No uuid found for local device');
 	}
 
+	// Get last reported state from cache
+	lastReport = await getCache();
+
 	let reportPending = false;
 	// Reports current state if not already sending and prevents a state change
 	// from exceeding report frequency. Returns true if sent; otherwise false.

src/types/state.ts

@@ -54,63 +54,6 @@ export interface DeviceLegacyState {
 	commit?: string;
 }
 
-export type ServiceState = {
-	image: string;
-	status: string;
-	download_progress?: number | null;
-};
-
-export type ReleaseState = {
-	services: {
-		[serviceName: string]: ServiceState;
-	};
-};
-
-export type ReleasesState = {
-	[releaseUuid: string]: ReleaseState;
-};
-
-export type AppState = {
-	release_uuid?: string;
-	releases: ReleasesState;
-};
-
-export type DeviceReport = {
-	name?: string;
-	status?: string;
-	os_version?: string | null; // TODO: Should these purely come from the os app?
-	os_variant?: string | null; // TODO: Should these purely come from the os app?
-	supervisor_version?: string; // TODO: Should this purely come from the supervisor app?
-	provisioning_progress?: number | null; // TODO: should this be reported as part of the os app?
-	provisioning_state?: string; // TODO: should this be reported as part of the os app?
-	ip_address?: string;
-	mac_address?: string | null;
-	api_port?: number; // TODO: should this be reported as part of the supervisor app?
-	api_secret?: string | null; // TODO: should this be reported as part of the supervisor app?
-	logs_channel?: string | null; // TODO: should this be reported as part of the supervisor app? or should it not be reported anymore at all?
-	memory_usage?: number;
-	memory_total?: number;
-	storage_block_device?: string;
-	storage_usage?: number;
-	storage_total?: number;
-	cpu_temp?: number;
-	cpu_usage?: number;
-	cpu_id?: string;
-	is_undervolted?: boolean;
-	// TODO: these are ignored by the API but are used by supervisor local API
-	update_failed?: boolean;
-	update_pending?: boolean;
-	update_downloaded?: boolean;
-};
-
-export type DeviceState = {
-	[deviceUuid: string]: DeviceReport & {
-		apps?: {
-			[appUuid: string]: AppState;
-		};
-	};
-};
-
 // Return a type with a default value
 export const withDefault = <T extends t.Any>(
 	type: T,
@@ -162,6 +105,80 @@ const fromType = <T extends object>(name: string) =>
 // Alias short string to UUID so code reads more clearly
 export const UUID = ShortString;
 
+/*****************
+ * Current state *
+ *****************/
+const ServiceState = t.intersection([
+	t.type({
+		image: t.string,
+		status: t.string,
+	}),
+	t.partial({
+		download_progress: t.union([t.number, t.null]),
+	}),
+]);
+export type ServiceState = t.TypeOf<typeof ServiceState>;
+
+const ReleaseState = t.type({
+	services: t.record(DockerName, ServiceState),
+});
+export type ReleaseState = t.TypeOf<typeof ReleaseState>;
+
+const ReleasesState = t.record(UUID, ReleaseState);
+export type ReleasesState = t.TypeOf<typeof ReleasesState>;
+
+const AppState = t.intersection([
+	t.type({
+		releases: ReleasesState,
+	}),
+	t.partial({
+		release_uuid: UUID,
+	}),
+]);
+export type AppState = t.TypeOf<typeof AppState>;
+
+const DeviceReport = t.partial({
+	name: t.string,
+	status: t.string,
+	os_version: t.union([t.string, t.null]),
+	os_variant: t.union([t.string, t.null]),
+	supervisor_version: t.string,
+	provisioning_progress: t.union([t.number, t.null]),
+	provisioning_state: t.string,
+	ip_address: t.string,
+	mac_address: t.union([t.string, t.null]),
+	api_port: t.number,
+	api_secret: t.union([t.string, t.null]),
+	logs_channel: t.union([t.string, t.null]),
+	memory_usage: t.number,
+	memory_total: t.number,
+	storage_block_device: t.string,
+	storage_usage: t.number,
+	storage_total: t.number,
+	cpu_temp: t.number,
+	cpu_usage: t.number,
+	cpu_id: t.string,
+	is_undervolted: t.boolean,
+	update_failed: t.boolean,
+	update_pending: t.boolean,
+	update_downloaded: t.boolean,
+});
+export type DeviceReport = t.TypeOf<typeof DeviceReport>;
+
+export const DeviceState = t.record(
+	UUID,
+	t.intersection([
+		DeviceReport,
+		t.partial({
+			apps: t.record(UUID, AppState),
+		}),
+	]),
+);
+export type DeviceState = t.TypeOf<typeof DeviceState>;
+
+/****************
+ * Target state *
+ ****************/
 /**
  * A target service has docker image, a set of environment variables
  * and labels as well as one or more configurations