Use container name instead of id to identify apps, and avoid duplicated containers

By storing the container name before creating the container, we avoid problems
if the supervisor crashes or the device reboots between creating a container and storing its id.

Change-Type: patch
Signed-off-by: Pablo Carranza Velez <pablo@resin.io>

src/api.coffee

@@ -154,10 +154,12 @@ module.exports = (application) ->
 			return res.status(400).send('Missing app id')
 		Promise.using application.lockUpdates(appId, force), ->
 			utils.getKnexApp(appId)
-			.tap (app) ->
+			.then (app) ->
+				application.getContainerId(app)
+				.then (containerId) ->
 					application.kill(app, removeContainer: false)
 					.then (app) ->
-				res.json(_.pick(app, 'containerId'))
+						res.json({ containerId })
 		.catch utils.AppNotFoundError, (e) ->
 			return res.status(400).send(e.message)
 		.catch (err) ->
@@ -172,8 +174,10 @@ module.exports = (application) ->
 			utils.getKnexApp(appId)
 			.tap (app) ->
 				application.start(app)
-			.then (app) ->
-				res.json(_.pick(app, 'containerId'))
+				.then ->
+					application.getContainerId(app)
+				.then (containerId) ->
+					res.json({ containerId })
 		.catch utils.AppNotFoundError, (e) ->
 			return res.status(400).send(e.message)
 		.catch (err) ->
@@ -185,13 +189,16 @@ module.exports = (application) ->
 		if !appId?
 			return res.status(400).send('Missing app id')
 		Promise.using application.lockUpdates(appId, true), ->
-			columns = [ 'appId', 'containerId', 'commit', 'imageId', 'env' ]
+			columns = [ 'appId', 'containerName', 'commit', 'imageId', 'env' ]
 			utils.getKnexApp(appId, columns)
 			.then (app) ->
+				application.getContainerId(app)
+				.then (containerId) ->
 					# Don't return keys on the endpoint
 					app.env = _.omit(JSON.parse(app.env), config.privateAppEnvVars)
+					app.containerId = containerId
 					# Don't return data that will be of no use to the user
-				res.json(app)
+					res.json(_.omit(app, [ 'containerName' ]))
 		.catch utils.AppNotFoundError, (e) ->
 			return res.status(400).send(e.message)
 		.catch (err) ->

src/application.coffee

@@ -18,6 +18,7 @@ proxyvisor = require './proxyvisor'
 { checkInt, checkTruthy } = require './lib/validation'
 osRelease = require './lib/os-release'
 deviceConfig = require './device-config'
+randomHexString = require './lib/random-hex-string'
 
 class UpdatesLockedError extends TypedError
 ImageNotFoundError = (err) ->
@@ -155,7 +156,7 @@ logSpecialAction = (action, value, success) ->
 application.kill = kill = (app, { updateDB = true, removeContainer = true } = {}) ->
 	logSystemEvent(logTypes.stopApp, app)
 	device.updateState(status: 'Stopping')
-	container = docker.getContainer(app.containerId)
+	container = docker.getContainer(app.containerName)
 	container.stop(t: 10)
 	.then ->
 		container.remove(v: true) if removeContainer
@@ -183,7 +184,7 @@ application.kill = kill = (app, { updateDB = true, removeContainer = true } = {}
 	.tap ->
 		logSystemEvent(logTypes.stopAppSuccess, app)
 		if removeContainer && updateDB
-			app.containerId = null
+			app.containerName = null
 			knex('app').update(app).where(appId: app.appId)
 	.catch (err) ->
 		logSystemEvent(logTypes.stopAppError, app, err)
@@ -255,6 +256,12 @@ isExecFormatError = (err) ->
 		message = err.json.trim()
 	/exec format error$/.test(message)
 
+generateContainerName = (app) ->
+	randomHexString.generate()
+	.then (randomString) ->
+		sanitisedAppName = app.name.replace(/[^a-zA-Z0-9]+/g, '_')
+		return "#{sanitisedAppName}_#{randomString}"
+
 application.start = start = (app) ->
 	device.isResinOSv1().then (isV1) ->
 		volumes = utils.defaultVolumes(isV1)
@@ -270,9 +277,9 @@ application.start = start = (app) ->
 				.map((port) -> port.trim())
 				.filter(isValidPort)
 
-			if app.containerId?
+			if app.containerName?
 				# If we have a container id then check it exists and if so use it.
-				container = docker.getContainer(app.containerId)
+				container = docker.getContainer(app.containerName)
 				containerPromise = container.inspect().return(container)
 			else
 				containerPromise = Promise.rejected()
@@ -283,7 +290,13 @@ application.start = start = (app) ->
 				.then (imageInfo) ->
 					logSystemEvent(logTypes.installApp, app)
 					device.updateState(status: 'Installing')
-
+					generateContainerName(app)
+					.tap (name) ->
+						app.containerName = name
+						knex('app').update(app).where(appId: app.appId)
+						.then (affectedRows) ->
+							knex('app').insert(app) if affectedRows == 0
+					.then (name) ->
 						ports = {}
 						portBindings = {}
 						if portList?
@@ -301,6 +314,7 @@ application.start = start = (app) ->
 						.then (shouldMount) ->
 							binds.push('/bin/kmod:/bin/kmod:ro') if shouldMount
 							docker.createContainer(
+								name: name
 								Image: app.imageId
 								Cmd: cmd
 								Tty: true
@@ -341,21 +355,20 @@ application.start = start = (app) ->
 				.catch (err) ->
 					# If starting the container failed, we remove it so that it doesn't litter
 					container.remove(v: true)
+					.finally ->
+						throw err
 				.then ->
-						app.containerId = null
+					device.updateState(commit: app.commit)
+					logger.attach(app)
+		.catch (err) ->
+			# If creating or starting the app failed, we soft-delete it so that
+			# the next update cycle doesn't think the app is up to date
+			app.containerName = null
+			app.markedForDeletion = true
 			knex('app').update(app).where(appId: app.appId)
 			.finally ->
 				logSystemEvent(logTypes.startAppError, app, err)
 				throw err
-				.then ->
-					app.containerId = container.id
-					device.updateState(commit: app.commit)
-					logger.attach(app)
-			.tap (container) ->
-				# Update the app info, only if starting the container worked.
-				knex('app').update(app).where(appId: app.appId)
-				.then (affectedRows) ->
-					knex('app').insert(app) if affectedRows == 0
 		.tap ->
 			if alreadyStarted
 				logSystemEvent(logTypes.startAppNoop, app)
@@ -765,7 +778,7 @@ application.update = update = (force, scheduled = false) ->
 							app = localApps[appId]
 							Promise.using lockUpdates(app, force), ->
 								# We get the app from the DB again in case someone restarted it
-								# (which would have changed its containerId)
+								# (which would have changed its containerName)
 								utils.getKnexApp(appId)
 								.then(kill)
 								.then ->
@@ -848,6 +861,8 @@ application.update = update = (force, scheduled = false) ->
 			device.updateState(status: 'Idle')
 			return
 
+sanitiseContainerName = (name) -> name.replace(/^\//, '')
+
 listenToEvents = do ->
 	appHasDied = {}
 	return ->
@@ -859,14 +874,14 @@ listenToEvents = do ->
 			parser.on 'error', (err) ->
 				console.error('Error on docker events JSON stream:', err, err.stack)
 			parser.on 'data', (data) ->
-				if data?.Type? && data.Type == 'container' && data.status in ['die', 'start']
-					knex('app').select().where({ containerId: data.id })
+				if data?.Type? && data.Type == 'container' && data.status in ['die', 'start'] && data.Actor?.Attributes?.Name?
+					knex('app').select().where({ containerName: sanitiseContainerName(data.Actor.Attributes.Name) })
 					.then ([ app ]) ->
 						if app?
 							if data.status == 'die'
 								logSystemEvent(logTypes.appExit, app)
-								appHasDied[app.containerId] = true
-							else if data.status == 'start' and appHasDied[app.containerId]
+								appHasDied[app.containerName] = true
+							else if data.status == 'start' and appHasDied[app.containerName]
 								logSystemEvent(logTypes.appRestart, app)
 								logger.attach(app)
 					.catch (err) ->
@@ -878,8 +893,34 @@ listenToEvents = do ->
 		.catch (err) ->
 			console.error('Error listening to events:', err, err.stack)
 
+migrateContainerIdApps = ->
+	knex.schema.hasColumn('app', 'containerId')
+	.then (exists) ->
+		return if not exists
+		knex('app').whereNotNull('containerId').select()
+		.then (apps) ->
+			return if !apps? or apps.length == 0
+			Promise.map apps, (app) ->
+				docker.getContainer(app.containerId).inspect()
+				.catchReturn({ Name: null })
+				.then (container) ->
+					app.containerName = sanitiseContainerName(container.Name)
+					app.containerId = null
+					knex('app').update(app).where({ id: app.id })
+		.then ->
+			knex.schema.table 'app', (table) ->
+				table.dropColumn('containerId')
+
+application.getContainerId = (app) ->
+	docker.getContainer(app.containerName).inspect()
+	.catchReturn({ Id: null })
+	.then (container) ->
+		return container.Id
+
 application.initialize = ->
 	listenToEvents()
+	migrateContainerIdApps()
+	.then ->
 		getAndApplyDeviceConfig()
 	.then ->
 		knex('app').whereNot(markedForDeletion: true).orWhereNull('markedForDeletion').select()

src/db.coffee

@@ -40,7 +40,7 @@ knex.init = Promise.all([
 			knex.schema.createTable 'app', (t) ->
 				t.increments('id').primary()
 				t.string('name')
-				t.string('containerId')
+				t.string('containerName')
 				t.string('commit')
 				t.string('imageId')
 				t.string('appId')
@@ -52,6 +52,7 @@ knex.init = Promise.all([
 			Promise.all [
 				addColumn('app', 'commit', 'string')
 				addColumn('app', 'appId', 'string')
+				addColumn('app', 'containerName', 'string')
 				addColumn('app', 'config', 'json')
 				addColumn('app', 'markedForDeletion', 'boolean')
 			]

src/lib/logger.coffee

@@ -80,19 +80,19 @@ exports.log = ->
 do ->
 	_lock = new Lock()
 	_writeLock = Promise.promisify(_lock.async.writeLock)
-	loggerLock = (containerId) ->
-		_writeLock(containerId)
+	loggerLock = (containerName) ->
+		_writeLock(containerName)
 		.disposer (release) ->
 			release()
 
 	attached = {}
 	exports.attach = (app) ->
-		Promise.using loggerLock(app.containerId), ->
-			if !attached[app.containerId]
-				docker.getContainer(app.containerId)
+		Promise.using loggerLock(app.containerName), ->
+			if !attached[app.containerName]
+				docker.getContainer(app.containerName)
 				.logs({ follow: true, stdout: true, stderr: true, timestamps: true })
 				.then (stream) ->
-					attached[app.containerId] = true
+					attached[app.containerName] = true
 					stream.pipe(es.split())
 					.on 'data', (logLine) ->
 						space = logLine.indexOf(' ')
@@ -101,6 +101,6 @@ do ->
 							publish(msg)
 					.on 'error', (err) ->
 						console.error('Error on container logs', err, err.stack)
-						attached[app.containerId] = false
+						attached[app.containerName] = false
 					.on 'end', ->
-						attached[app.containerId] = false
+						attached[app.containerName] = false