feat(memfault): fetch reboots for devices on button press

If a device publishes a button press for the button 42 the backend will poll every 15 seconds for up to two minutes for reboots on the Memfault API. If a reboot is found that is newer than the button press it is sent to the frontend.
NordicPlayground · Mar 21, 2024 · 1847a11 · 1847a11
1 parent 40178f6
commit 1847a11
Show file tree

Hide file tree

Showing 9 changed files with 321 additions and 79 deletions.
diff --git a/cdk/BackendLambdas.d.ts b/cdk/BackendLambdas.d.ts
@@ -15,4 +15,5 @@ type BackendLambdas = {
 	updatesToLwM2M: PackedLambda
 	publishLwM2MShadowsToJSON: PackedLambda
 	memfault: PackedLambda
+	memfaultPollForReboots: PackedLambda
 }
diff --git a/cdk/backend.ts b/cdk/backend.ts
@@ -15,6 +15,7 @@ const packagesInLayer: string[] = [
 	'jsonata',
 	'mqtt',
 	'@protobuf-ts/runtime',
+	'p-retry',
 ]
 const pack = async (
 	id: string,
@@ -53,6 +54,7 @@ new BackendApp({
 		nrplusGatewayScan: await pack('nrplusGatewayScan'),
 		updatesToLwM2M: await pack('updatesToLwM2M'),
 		memfault: await pack('memfault'),
+		memfaultPollForReboots: await pack('memfaultPollForReboots'),
 		// For hello.nrfcloud.com/map
 		publishLwM2MShadowsToJSON: await pack('publishLwM2MShadowsToJSON'),
 	},

diff --git a/cdk/resources/Memfault.ts b/cdk/resources/Memfault.ts
@@ -5,6 +5,7 @@ import {
 	aws_s3 as S3,
 	aws_iam as IAM,
 	aws_lambda as Lambda,
+	aws_iot as IoT,
 	Stack,
 	RemovalPolicy,
 } from 'aws-cdk-lib'
@@ -29,6 +30,7 @@ export class Memfault extends Construct {
 		}: {
 			lambdaSources: {
 				memfault: PackedLambda
+				memfaultPollForReboots: PackedLambda
 			}
 			baseLayer: Lambda.ILayerVersion
 			assetTrackerStackName: string
@@ -72,8 +74,7 @@ export class Memfault extends Construct {
 				ASSET_TRACKER_STACK_NAME: assetTrackerStackName,
 				NODE_NO_WARNINGS: '1',
 				BUCKET: this.bucket.bucketName,
-				WEBSOCKET_CONNECTIONS_TABLE_NAME:
-					websocketAPI.connectionsTable.tableName,
+				CONNECTIONS_TABLE_NAME: websocketAPI.connectionsTable.tableName,
 			},
 			initialPolicy: [
 				new IAM.PolicyStatement({
@@ -104,5 +105,98 @@ export class Memfault extends Construct {
 			principal: new IAM.ServicePrincipal('events.amazonaws.com') as IPrincipal,
 			sourceArn: rule.ruleArn,
 		})
+
+		// When a devices publishes button press 42, poll the Memfault API for an update
+		const pollForRebootsFn = new Lambda.Function(this, 'pollForRebootsFn', {
+			handler: lambdaSources.memfaultPollForReboots.handler,
+			architecture: Lambda.Architecture.ARM_64,
+			runtime: Lambda.Runtime.NODEJS_20_X,
+			timeout: Duration.seconds(120),
+			memorySize: 1792,
+			code: Lambda.Code.fromAsset(
+				lambdaSources.memfaultPollForReboots.lambdaZipFile,
+			),
+			description:
+				'Poll the Memfault API for an update after a device publishes a button event for button 42',
+			layers: [baseLayer],
+			environment: {
+				VERSION: this.node.tryGetContext('version'),
+				STACK_NAME: Stack.of(this).stackName,
+				ASSET_TRACKER_STACK_NAME: assetTrackerStackName,
+				NODE_NO_WARNINGS: '1',
+				BUCKET: this.bucket.bucketName,
+				CONNECTIONS_TABLE_NAME: websocketAPI.connectionsTable.tableName,
+				WEBSOCKET_MANAGEMENT_API_URL: websocketAPI.websocketManagementAPIURL,
+			},
+			initialPolicy: [
+				new IAM.PolicyStatement({
+					actions: ['ssm:GetParametersByPath'],
+					resources: [
+						`arn:aws:ssm:${Stack.of(this).region}:${Stack.of(this).account}:parameter/${Stack.of(this).stackName}/memfault/*`,
+					],
+				}),
+				new IAM.PolicyStatement({
+					actions: ['execute-api:ManageConnections'],
+					resources: [websocketAPI.websocketAPIArn],
+				}),
+				new IAM.PolicyStatement({
+					actions: ['iot:DescribeThing'],
+					resources: ['*'],
+				}),
+			],
+			...new LambdaLogGroup(this, 'pollForRebootsFnLogs'),
+		})
+
+		websocketAPI.connectionsTable.grantFullAccess(pollForRebootsFn)
+
+		const button42RuleRole = new IAM.Role(this, 'button42RuleRole', {
+			assumedBy: new IAM.ServicePrincipal('iot.amazonaws.com'),
+			inlinePolicies: {
+				rootPermissions: new IAM.PolicyDocument({
+					statements: [
+						new IAM.PolicyStatement({
+							actions: ['iot:Publish'],
+							resources: [
+								`arn:aws:iot:${Stack.of(this).region}:${Stack.of(this).account}:topic/errors`,
+							],
+						}),
+					],
+				}),
+			},
+		})
+
+		const button42Rule = new IoT.CfnTopicRule(this, 'button42Rule', {
+			topicRulePayload: {
+				awsIotSqlVersion: '2016-03-23',
+				description:
+					'Trigger a fetch of the Memfault data when a device publishes a button event for button 42',
+				ruleDisabled: false,
+				sql: [
+					'SELECT topic(1) as deviceId,',
+					'btn.ts as ts,',
+					`parse_time("yyyy-MM-dd'T'HH:mm:ss.S'Z'", timestamp()) as timestamp`,
+					"FROM '+/messages'",
+					'WHERE btn.v = 42',
+				].join(' '),
+				actions: [
+					{
+						lambda: {
+							functionArn: pollForRebootsFn.functionArn,
+						},
+					},
+				],
+				errorAction: {
+					republish: {
+						roleArn: button42RuleRole.roleArn,
+						topic: 'errors',
+					},
+				},
+			},
+		})
+
+		pollForRebootsFn.addPermission('storeMessagesRule', {
+			principal: new IAM.ServicePrincipal('iot.amazonaws.com'),
+			sourceArn: button42Rule.attrArn,
+		})
 	}
 }
diff --git a/lambda/memfault.ts b/lambda/memfault.ts
@@ -1,88 +1,28 @@
+import { DynamoDBClient } from '@aws-sdk/client-dynamodb'
 import { IoTClient } from '@aws-sdk/client-iot'
-import { GetParametersByPathCommand, SSMClient } from '@aws-sdk/client-ssm'
+import { PutObjectCommand, S3Client } from '@aws-sdk/client-s3'
+import { SSMClient } from '@aws-sdk/client-ssm'
 import { fromEnv } from '@nordicsemiconductor/from-env'
-import { S3Client, PutObjectCommand } from '@aws-sdk/client-s3'
-import { getActiveConnections } from './notifyClients.js'
-import { DynamoDBClient } from '@aws-sdk/client-dynamodb'
 import { listThingsInGroup } from './listThingsInGroup.js'
+import { createAPIClient, type Reboot } from './memfault/api.js'
+import { getActiveConnections } from './notifyClients.js'
 
-const ssm = new SSMClient({})
+export const ssm = new SSMClient({})
 const iot = new IoTClient({})
 const s3 = new S3Client({})
 const db = new DynamoDBClient({})
 
-const {
-	stackName,
-	nrfAssetTrackerStackName,
-	bucket,
-	websocketConnectionsTableName,
-} = fromEnv({
-	stackName: 'STACK_NAME',
-	nrfAssetTrackerStackName: 'ASSET_TRACKER_STACK_NAME',
-	bucket: 'BUCKET',
-	websocketConnectionsTableName: 'WEBSOCKET_CONNECTIONS_TABLE_NAME',
-})(process.env)
-
-const Prefix = `/${stackName}/memfault/`
-const { organizationAuthToken, organizationId, projectId } = (
-	(
-		await ssm.send(
-			new GetParametersByPathCommand({
-				Path: Prefix,
-			}),
-		)
-	)?.Parameters ?? []
-).reduce(
-	(params, p) => ({
-		...params,
-		[(p.Name ?? '').replace(Prefix, '')]: p.Value ?? '',
-	}),
-	{} as Record<string, string>,
-)
-
-if (
-	organizationAuthToken === undefined ||
-	organizationId === undefined ||
-	projectId === undefined
-)
-	throw new Error(`Memfault settings not configured!`)
-
-const getActive = getActiveConnections(db, websocketConnectionsTableName)
+const { stackName, nrfAssetTrackerStackName, bucket, connectionsTableName } =
+	fromEnv({
+		stackName: 'STACK_NAME',
+		nrfAssetTrackerStackName: 'ASSET_TRACKER_STACK_NAME',
+		bucket: 'BUCKET',
+		connectionsTableName: 'CONNECTIONS_TABLE_NAME',
+	})(process.env)
 
-type Reboot = {
-	type: 'memfault'
-	mcu_reason_register: null
-	time: string // e.g. '2024-03-14T07:26:37.270000+00:00'
-	reason: number // e.g. 7
-	software_version: {
-		version: string // e.g. '1.11.1+thingy91.low-power.memfault'
-		id: number // e.g.504765
-		software_type: {
-			id: number //e.g. 32069;
-			name: string // e.g. 'thingy_world'
-		}
-		archived: boolean
-	}
-}
+const getActive = getActiveConnections(db, connectionsTableName)
 
-const api = {
-	getLastReboots: async (deviceId: string): Promise<null | Array<Reboot>> => {
-		const res = await fetch(
-			`https://api.memfault.com/api/v0/organizations/${organizationId}/projects/${projectId}/devices/${deviceId}/reboots?${new URLSearchParams(
-				{
-					since: new Date(Date.now() - 24 * 60 * 60 * 1000).toISOString(),
-				},
-			).toString()}`,
-			{
-				headers: new Headers({
-					Authorization: `Basic ${Buffer.from(`:${organizationAuthToken}`).toString('base64')}`,
-				}),
-			},
-		)
-		if (!res.ok) return null
-		return (await res.json()).data
-	},
-}
+const api = await createAPIClient(ssm, stackName)
 
 const listThings = listThingsInGroup(iot)
 

diff --git a/lambda/memfault/api.ts b/lambda/memfault/api.ts
@@ -0,0 +1,68 @@
+import { GetParametersByPathCommand, SSMClient } from '@aws-sdk/client-ssm'
+
+export const createAPIClient = async (
+	ssm: SSMClient,
+	stackName: string,
+): Promise<{
+	getLastReboots: (deviceId: string) => Promise<null | Array<Reboot>>
+}> => {
+	const Prefix = `/${stackName}/memfault/`
+
+	const { organizationAuthToken, organizationId, projectId } = (
+		(
+			await ssm.send(
+				new GetParametersByPathCommand({
+					Path: Prefix,
+				}),
+			)
+		)?.Parameters ?? []
+	).reduce(
+		(params, p) => ({
+			...params,
+			[(p.Name ?? '').replace(Prefix, '')]: p.Value ?? '',
+		}),
+		{} as Record<string, string>,
+	)
+
+	if (
+		organizationAuthToken === undefined ||
+		organizationId === undefined ||
+		projectId === undefined
+	)
+		throw new Error(`Memfault settings not configured!`)
+
+	return {
+		getLastReboots: async (deviceId) => {
+			const res = await fetch(
+				`https://api.memfault.com/api/v0/organizations/${organizationId}/projects/${projectId}/devices/${deviceId}/reboots?${new URLSearchParams(
+					{
+						since: new Date(Date.now() - 24 * 60 * 60 * 1000).toISOString(),
+					},
+				).toString()}`,
+				{
+					headers: new Headers({
+						Authorization: `Basic ${Buffer.from(`:${organizationAuthToken}`).toString('base64')}`,
+					}),
+				},
+			)
+			if (!res.ok) return null
+			return (await res.json()).data
+		},
+	}
+}
+
+export type Reboot = {
+	type: 'memfault'
+	mcu_reason_register: null
+	time: string // e.g. '2024-03-14T07:26:37.270000+00:00'
+	reason: number // e.g. 7
+	software_version: {
+		version: string // e.g. '1.11.1+thingy91.low-power.memfault'
+		id: number // e.g.504765
+		software_type: {
+			id: number //e.g. 32069;
+			name: string // e.g. 'thingy_world'
+		}
+		archived: boolean
+	}
+}