Auto-merge for PR #483 via VersionBot

Resumable deltas

CHANGELOG.md

@@ -4,6 +4,10 @@ All notable changes to this project will be documented in this file
 automatically by Versionist. DO NOT EDIT THIS FILE MANUALLY!
 This project adheres to [Semantic Versioning](http://semver.org/).
 
+## v6.2.0 - 2017-08-16
+
+* Try to resume the download of a delta if it fails due to flaky network #483 [Akis Kesoglou]
+
 ## v6.1.4 - 2017-08-07
 
 * Fix references in deploy-to-resin.js and use github credentials when pushing in pr-to-meta-resin.sh #481 [Pablo Carranza Velez]

package.json

@@ -1,7 +1,7 @@
 {
   "name": "resin-supervisor",
   "description": "This is resin.io's Supervisor, a program that runs on IoT devices and has the task of running user Apps (which are Docker containers), and updating them as Resin's API informs it to.",
-  "version": "6.1.4",
+  "version": "6.2.0",
   "license": "Apache-2.0",
   "repository": {
     "type": "git",
@@ -29,7 +29,7 @@
     "buffer-equal-constant-time": "^1.0.1",
     "coffee-loader": "^0.7.3",
     "coffee-script": "~1.11.0",
-    "docker-delta": "1.1.1",
+    "docker-delta": "^2.0.1",
     "docker-progress": "^2.6.0",
     "docker-toolbelt": "^3.0.1",
     "event-stream": "^3.0.20",
@@ -46,7 +46,7 @@
     "pinejs-client": "^2.4.0",
     "pubnub": "^3.7.13",
     "request": "^2.51.0",
-    "request-progress": "^2.0.1",
+    "resumable-request": "^1.0.0",
     "resin-lint": "^1.3.1",
     "resin-register-device": "^3.0.0",
     "rimraf": "^2.5.4",

src/application.coffee

@@ -19,6 +19,8 @@ proxyvisor = require './proxyvisor'
 osRelease = require './lib/os-release'
 deviceConfig = require './device-config'
 
+DEFAULT_DELTA_APPLY_TIMEOUT = 300 * 1000 # 6 minutes
+
 class UpdatesLockedError extends TypedError
 ImageNotFoundError = (err) ->
 	return "#{err.statusCode}" is '404'
@@ -218,9 +220,15 @@ fetch = (app, setDeviceUpdateState = true) ->
 			Promise.join utils.getConfig('apiKey'), utils.getConfig('uuid'), (apiKey, uuid) ->
 				if conf['RESIN_SUPERVISOR_DELTA'] == '1'
 					logSystemEvent(logTypes.downloadAppDelta, app)
-					requestTimeout = checkInt(conf['RESIN_SUPERVISOR_DELTA_REQUEST_TIMEOUT'], positive: true) ? 30 * 60 * 1000
-					totalTimeout = checkInt(conf['RESIN_SUPERVISOR_DELTA_TOTAL_TIMEOUT'], positive: true) ? 24 * 60 * 60 * 1000
-					dockerUtils.rsyncImageWithProgress(app.imageId, { requestTimeout, totalTimeout, uuid, apiKey }, onProgress)
+					deltaOpts = {
+						uuid, apiKey
+						# use user-defined timeouts, but fallback to defaults if none is provided.
+						requestTimeout: checkInt(conf['RESIN_SUPERVISOR_DELTA_REQUEST_TIMEOUT'], positive: true)
+						applyTimeout: checkInt(conf['RESIN_SUPERVISOR_DELTA_APPLY_TIMEOUT'], positive: true) ? DEFAULT_DELTA_APPLY_TIMEOUT
+						retryCount: checkInt(conf['RESIN_SUPERVISOR_DELTA_RETRY_COUNT'], positive: true)
+						retryInterval: checkInt(conf['RESIN_SUPERVISOR_DELTA_RETRY_INTERVAL'], positive: true)
+					}
+					dockerUtils.rsyncImageWithProgress(app.imageId, deltaOpts, onProgress)
 				else
 					logSystemEvent(logTypes.downloadApp, app)
 					dockerUtils.fetchImageWithProgress(app.imageId, onProgress, { uuid, apiKey })

src/docker-utils.coffee

@@ -4,12 +4,11 @@ process.env.DOCKER_HOST ?= "unix://#{config.dockerSocket}"
 Docker = require 'docker-toolbelt'
 { DockerProgress } = require 'docker-progress'
 Promise = require 'bluebird'
-progress = require 'request-progress'
 dockerDelta = require 'docker-delta'
 
 _ = require 'lodash'
 knex = require './db'
-{ request } = require './request'
+{ request, resumable } = require './request'
 Lock = require 'rwlock'
 utils = require './utils'
 rimraf = Promise.promisify(require('rimraf'))
@@ -53,6 +52,23 @@ getRepoAndTag = (image) ->
 			registry = ''
 		return { repo: "#{registry}#{imageName}", tag: tagName }
 
+applyDelta = (imgSrc, deltaUrl, { requestTimeout, applyTimeout, retryCount, retryInterval }, onProgress) ->
+	new Promise (resolve, reject) ->
+		resumable(request, { url: deltaUrl, timeout: requestTimeout })
+		.on('progress', onProgress)
+		.on('retry', onProgress)
+		.on('error', reject)
+		.on 'response', (res) ->
+			if res.statusCode isnt 200
+				reject(new Error("Got #{res.statusCode} when requesting delta from storage."))
+			else if parseInt(res.headers['content-length']) is 0
+				reject(new Error('Invalid delta URL.'))
+			else
+				deltaStream = dockerDelta.applyDelta(imgSrc, timeout: applyTimeout)
+				res.pipe(deltaStream)
+				.on('id', resolve)
+				.on('error', reject)
+
 do ->
 	_lock = new Lock()
 	_writeLock = Promise.promisify(_lock.async.writeLock)
@@ -66,7 +82,7 @@ do ->
 		.disposer (release) ->
 			release()
 
-	exports.rsyncImageWithProgress = (imgDest, { requestTimeout, totalTimeout, uuid, apiKey, startFromEmpty = false }, onProgress) ->
+	exports.rsyncImageWithProgress = (imgDest, { requestTimeout, applyTimeout, retryCount, retryInterval, uuid, apiKey, startFromEmpty = false }, onProgress) ->
 		Promise.using readLockImages(), ->
 			Promise.try ->
 				if startFromEmpty
@@ -87,36 +103,30 @@ do ->
 					.get(1)
 					.then (b) ->
 						opts =
+							followRedirect: false
 							timeout: requestTimeout
 
 						if b?.token?
-							deltaAuthOpts =
-								auth:
-									bearer: b?.token
+							opts.auth =
+								bearer: b.token
 								sendImmediately: true
-							opts = _.merge(opts, deltaAuthOpts)
 						new Promise (resolve, reject) ->
-							progress request.get("#{config.deltaHost}/api/v2/delta?src=#{imgSrc}&dest=#{imgDest}", opts)
-							.on 'progress', (progress) ->
-								# In request-progress ^2.0.1, "percentage" is a ratio from 0 to 1
-								onProgress(percentage: progress.percentage * 100)
-							.on 'end', ->
-								onProgress(percentage: 100)
+							request.get("#{config.deltaHost}/api/v2/delta?src=#{imgSrc}&dest=#{imgDest}", opts)
 							.on 'response', (res) ->
-								if res.statusCode is 504
+								res.resume() # discard response body -- we only care about response headers
+								if res.statusCode in [ 502, 504 ]
 									reject(new Error('Delta server is still processing the delta, will retry'))
-								else if res.statusCode isnt 200
+								else if not (300 <= res.statusCode < 400 and res.headers['location']?)
 									reject(new Error("Got #{res.statusCode} when requesting image from delta server."))
 								else
+									deltaUrl = res.headers['location']
 									if imgSrc is 'resin/scratch'
 										deltaSrc = null
 									else
 										deltaSrc = imgSrc
-									res.pipe(dockerDelta.applyDelta(deltaSrc, imgDest))
-									.on('id', resolve)
-									.on('error', reject)
+									deltaOpts = { requestTimeout, applyTimeout, retryCount, retryInterval }
+									resolve(applyDelta(deltaSrc, deltaUrl, deltaOpts, onProgress))
 							.on 'error', reject
-				.timeout(totalTimeout)
 			.then (id) ->
 				getRepoAndTag(imgDest)
 				.then ({ repo, tag }) ->

src/request.coffee

@@ -2,6 +2,7 @@ config = require './config'
 PlatformAPI = require 'pinejs-client'
 Promise = require 'bluebird'
 request = require 'request'
+resumable = require 'resumable-request'
 url = require 'url'
 osRelease = require './lib/os-release'
 
@@ -16,12 +17,23 @@ if osVersion?
 	else
 		userAgent += " (Linux; #{osVersion})"
 
+# With these settings, the device must be unable to receive a single byte
+# from the network for a continuous period of 20 minutes before we give up.
+# (reqTimeout + retryInterval) * retryCount / 1000ms / 60sec ~> minutes
+DEFAULT_REQUEST_TIMEOUT = 30000 # ms
+DEFAULT_REQUEST_RETRY_INTERVAL = 10000 # ms
+DEFAULT_REQUEST_RETRY_COUNT = 30
+
 requestOpts =
 	gzip: true
-	timeout: 30000
+	timeout: DEFAULT_REQUEST_TIMEOUT
 	headers:
 		'User-Agent': userAgent
 
+resumableOpts =
+	maxRetries: DEFAULT_REQUEST_RETRY_COUNT
+	retryInterval: DEFAULT_REQUEST_RETRY_INTERVAL
+
 try
 	PLATFORM_ENDPOINT = url.resolve(config.apiEndpoint, '/v2/')
 	exports.resinApi = resinApi = new PlatformAPI
@@ -35,3 +47,5 @@ catch
 request = request.defaults(requestOpts)
 
 exports.request = Promise.promisifyAll(request, multiArgs: true)
+
+exports.resumable = resumable.defaults(resumableOpts)