AWS IoT 是一個全託管的雲平台，讓裝置能夠輕易且安全與雲上的應用程式及其他裝置互動。身為 IoT 工程師，您會希望使用 AWS 的服務如 Kinesis、 Lambda、Amazon Machine Learning 進行互動以讓您的 IoT 應用發揮最大的效益。在本場演講，我們會深入探討如何在 Rule engine 裡定義 rules，或是使用device shadow 存取您裝置的最後可知狀態，了解 Greengrass的優點與場景，您的 IoT 應用程式也可利用雲平台將資料從裝置帶回AWS上。

1. © 2017, Amazon Web Services, Inc. or its Affiliates. All rights reserved.
Ivan Cheng, Solutions Architect
June 7, 2017
Deep Dive on AWS IoT
Shadows, Rules, and More

2. AWS IoT
Data storage
& analytics
Administration
Sensors
Actuators
Connected Farm
Control
automation

3. Agenda
• Introduction to AWS IoT
• Telemetry & analytics
• Cloud control
• Mobile control
• Managing software updates
• Wrap up
• Appendix: Lifecycle management

4. AWS IoT
DEVICE SDK
Set of client libraries to
connect, authenticate and
exchange messages
DEVICE GATEWAY
Communicate with devices via
MQTT and HTTP
AUTHENTICATION
AUTHORIZATION
Secure with mutual
authentication and encryption
RULES ENGINE
Transform messages
based on rules and
route to AWS services
AWS services
- - - - -
3P services
DEVICE SHADOW
Persistent thing state
during intermittent
connections
APPLICATIONS
AWS IoT API
DEVICE REGISTRY
Identity and management of
your things

5. Key takeaways
• Messaging
• Be careful with wide fan out
• No message ordering guarantees
• Avoid large fan in
• WebSockets for Cognito authentication
• Rules
• Send data to multiple data stores at the same time
• Manage device lifecycle events
• Shadows
• Designed for the real world: poor connectivity, out of order messages
• Fine-grained control over software rollouts
• Not ideal for storing time-series analytics data
• Security
• One cert per device
• Set fine-grained permissions for devices and Cognito users
• Naming conventions can simplify policy management

6. Telemetry & Analytics

7. Administration
Actuators
Control
automation
AWS IoT
Data storage
& analytics
Sensors
Connected Farm

8. AWS IoT Telemetry & Analytics
1. Connect devices
2. Send data
3. Collect & store the data
4. Do something with the data

9. AWS IoT Telemetry & Analytics
DEVICE GATEWAY
Communicate with devices via
MQTT and HTTP
AUTHENTICATION
AUTHORIZATION
Secure with mutual
authentication and encryption
RULES ENGINE
Transform messages
based on rules and
route to AWS services
AWS services
- - - - -
3P Services

10. 1) Connect the devices
1. Provision a certificate
2. Attach policy
3. Connect over MQTT

11. 2) Send data
PUBLISH macdonald/sensors/123 (qos: 0)
{
"timestamp": "2016-01-29T10:00:00",
"temperature": 55
"humidity": 39,
"ph": 6.7
}

12. 3) Collect the data
AWS IoT
Data storage
& analytics
Sensors
?

13. Single consumer (don’t do this)
AWS IoT instance database
PUBLISH sensors/123
PUBLISH sensors/456
SUBSCRIBE sensors/+
PUBLISH sensors/789

14. Don’t do this: scalability
AWS IoT instance
SUBSCRIBE #

15. Don’t do this: availability
AWS IoT instance

16. Don’t do this: maintainability
AWS IoT

17. Store it in the device shadow (don’t do this)
Sensors
DEVICE SHADOWS

18. 1. AWS Services
(Direct Integration)
Rules Engine
Actions
AWS IoT Rules Engine
LambdaSNS SQS
S3
Amazon
KinesisDynamoDB RDS
Amazon
Redshift
Amazon Glacier
EC2
3. External Endpoints
(via Lambda and SNS)
Rules Engine connects AWS
IoT to external endpoints and
AWS services.
2. Rest of AWS
(via Amazon Kinesis,
Lambda, S3, and more)

19. Example rule
{
"rule": {
"sql": "SELECT * AS message FROM 'sensors/#'",
"description": "Store all sensor data into dynamodb and firehose",
"actions": [{
"dynamoDB": {
"tableName": "sensor_data",
"roleArn": "arn:aws:iam::123456789012:role/aws_iot_dynamoDB",
"hashKeyField": "sensor_id",
"hashKeyValue": "${topic(2)}",
"rangeKeyField": "timestamp“
"rangeKeyValue": "${timestamp()}",
}
}, {
"firehose": {
"roleArn": "arn:aws:iam::123456789012:role/aws_iot_firehose",
"deliveryStreamName": "my_firehose_stream"
}
}]
}
}

20. Different Data Scenarios
Want to run a lot of queries constantly?
Use Amazon Kinesis Firehose to write into
Amazon Redshift
Need fast lookups, e.g. in Rules or Lambda functions?
Write into DynamoDB, add indexes if necessary
Have a need for heavy queries but not always-on?
Use Firehose & S3, process with Amazon EMR.

21. Takeaways
• Avoid single “firehose” MQTT consumer architecture
• Rules scalably route data into the rest of AWS
• Fork data into multiple data stores simultaneously
• Avoid the device shadow for analytics

22. Cloud Control

23. Administration
AWS IoT
Data storage
& analytics
Sensors
Connected Farm
Actuators
Control
automation

24. Automated Sprinkler Service
Amazon
Kinesis
Amazon Machine
Learning
Amazon
Redshift
Rules
Engine
Device
Gateway
Sensor
Sprinkler
Amazon Kinesis–
enabledapp

25. Talking back to the sprinklers
Amazon
Kinesis
Amazon Machine
Learning
Amazon
Redshift
Rules
Engine
Sensor
Device
Gateway
Sprinkler
Amazon Kinesis–
enabledapp

26. Publish on/off to the sprinkler (don’t do this)
Device
Gateway
Sprinkler
Control
logic
SUBSCRIBE
macdonald/sprinkler-456

27. Publish on/off to the sprinkler (don’t do this)
Device
Gateway
Sprinkler
Control
logic
PUBLISH
macdonald/sprinkler-456
{ "water": "on" }

28. Direct publishing: why not?
Device
Gateway
Sprinkler
(offline) Control
logic
PUBLISH
macdonald/sprinkler-456
{ "water": "on" }

29. Direct publishing: why not?
Sprinkler
Control
logic
on
off
Device
Gateway
off
on

30. Direct publishing: why not?
• Messages aren’t ordered
• Connection blips
So then what?

31. Device Shadows
Shadow
State
Apps
offline

32. Device Shadows
Device Controller
reported
state
desired
state

33. Device Shadows
Device Controller
reported
state
desired
state
HTTP/REST
WebSockets
MQTT

34. AWS IoT Shadow - Simple Yet Powerful
{
"state" : {
“desired" : {
"lights": { "color": "RED" },
"engine" : "ON"
},
"reported" : {
"lights" : { "color": "GREEN" },
"engine" : "ON"
},
"delta" : {
"lights" : { "color": "RED" }
} },
"version" : 10
}
Thing
Report its current state to one or multiple shadows
Retrieve its desired state from shadow
Mobile App
Set the desired state of a device
Get the last reported state of the device
Delete the shadow
Shadow
Shadow reports delta, desired and reported
states along with metadata and version

35. Device Shadows and versioning
Sprinkler
Control
logic
on (version=1)
off (version=2)
Device
Gateway
off (version=2)
on (version=1)
(old message ignored by device)

36. Takeaways
• Plan for devices losing connectivity
• Send devices commands through shadows
• Query device state through shadows
• Version numbers control concurrency

37. Mobile Control

38. Data storage
& analytics
Sensors
Talking back to the sprinklers: manual override
Control
automation
AWS IoT
Administration
Actuators

39. AWS IoT
DEVICE SHADOW
Persistent thing state
during intermittent
connections
APPLICATIONS

40. Using Cognito with IoT
DEVICE SHADOW
Persistent thing state
during intermittent
connections
APPLICATIONS
AMAZON
COGNITO
PERMISSIONS APIs
Configure device and
Cognito User permissions
end-user
(farmer)

41. end-user
(farmer)
Using Cognito with IoT
DEVICE SHADOW
Persistent thing state
during intermittent
connections
APPLICATIONS
AMAZON
COGNITO
PERMISSIONS APIs
Configure device and
Cognito User permissions

42. Policy for Cognito with IoT
Cognito identity pool policy:
{
"Effect": "Allow",
"Action": "iot:UpdateThingShadow",
"Resource": "*"
}
Specific policy for Old Macdonald Cognito user:
{
"Effect": "Allow",
"Action": "iot:UpdateThingShadow",
"Resource": "arn:aws:iot:…:thing/macdonald-sprinkler123"
}

43. Overall Cognito “pairing” workflow
1. Create a Cognito identity pool
2. Customer signs in using mobile app
3. Associate their user with their “farm”
4. Create a scope-down policy in IoT for their user
5. Attach that policy to their Cognito user in IoT

44. Managing fine-grained permissions
• One “farm owner” needs permissions to many shadows
• "arn:aws:iot:…:thing/sprinkler123abc"
• "arn:aws:iot:…:thing/sprinkler456def"
• …
• Listing each is tedious

45. Best practice: Thing name prefixing
• Prefix thing name with logical owner
• sensor123abc -> macdonald-sensor123abc
• Aspen policy supports wildcards
• "arn:aws:iot:…:thing/sensor123abc"
• "arn:aws:iot:…:thing/sensor123abc"
• "arn:aws:iot:…:thing/sensor456def"
• …
• "arn:aws:iot:…:thing/macdonald-*"

46. Takeaways: Cognito authorization
• Cognito enables secure human control over IoT devices
• IoT scope-down policy supports fine-grained control
• Naming conventions simplify policy management

47. Managing Software Updates

48. Data storage
& analytics
Managing software updates
Control
automation
AWS IoT
Administration
Actuators
Sensors

49. Firmware topic (don’t do this)
• Have all devices subscribe to a topic
• Publish updated binaries to this topic
SUBSCRIBE sensor/firmware
SUBSCRIBE sensor/firmware
SUBSCRIBE sensor/firmware
PUBLISH sensor/firmware
01100100 01101111 00100000
01101110 01101111 01110100
00100000 01100100 01101111
00100000 01110100 01101000
01101001 01110011

50. Firmware topic (don’t do this)
Pros:
• Sending an update is easy
Cons:
• Large messages not supported
• Offline devices miss updates
• No control over rollout

51. Firmware version shadow (don’t do this)
• One thing shadow for the current firmware version
• All devices subscribe to shadow updates
• Messages include a CloudFront download URL
SUBSCRIBE
$aws/shadow/firmware-thing
PUBLISH $aws/shadow/firmware-thing
{
"desired": {
"version": “123.45"
"url": “https://abc123.cloudfront.net/newversion"
}
}
SUBSCRIBE
$aws/shadow/firmware-thing

52. Firmware version shadow (don’t do this)
Pros:
• Sending an update is easy
• Offline devices eventually see updates
• Bulk download happens through CloudFront
Cons:
• No control over rollout
• Shadow protocol is chatty

53. Firmware in device shadows
• Set each device’s shadow to its desired firmware version
• Devices subscribe to their own shadow
• Messages include a CloudFront download URL

54. Firmware in device shadows
SUBSCRIBE
$aws/shadow/sensor-abc123
PUBLISH $aws/shadow/sensor-abc123
{
"desired": {
"version": “123.45"
"url": "https://abc123.cloudfront.net/newversion"
}
}
SUBSCRIBE
$aws/shadow/sensor-def456
PUBLISH $aws/shadow/sensor-def456
{
"desired": {
"version": “123.45"
"url": "https://abc123.cloudfront.net/newversion"
}
}

55. Firmware in device shadows
Pros:
• Full control over rollout / rollback
• Offline devices eventually see updates
• Bulk download happens through CloudFront
Cons:
• Sending updates requires sending multiple messages

56. Takeaway
• Be careful with wide fan out to millions of devices
• Wide fan out is supported, but slow
• Encourage safe device management

57. Wrap-up

58. AWS IoT
Data storage
& analytics
Administration
Sensors
Actuators
Connected Farm
Control
automation

59. AWS IoT
DEVICE SDK
Set of client libraries to
connect, authenticate, and
exchange messages
DEVICE GATEWAY
Communicate with devices via
MQTT and HTTP
AUTHENTICATION
AUTHORIZATION
Secure with mutual
authentication and encryption
RULES ENGINE
Transform messages
based on rules and
route to AWS services
AWS services
- - - - -
3P services
DEVICE SHADOW
Persistent thing state
during intermittent
connections
APPLICATIONS
AWS IoT API
DEVICE REGISTRY
Identity and management of
your things

60. Key takeaways
• Messaging
• Be careful with wide fan out
• No message ordering guarantees
• Avoid large fan-in
• WebSockets for Cognito authentication
• Rules
• Send data to multiple data stores at the same time
• Manage device lifecycle events
• Shadows
• Designed for the real world: poor connectivity, out of order messages
• Fine-grained control over software rollouts
• Not ideal for storing time-series analytics data
• Security
• One cert per device
• Set fine-grained permissions for devices and Cognito users
• Naming conventions can simplify policy management

61. Thank you!
Kudos to Brett Frantzis, Olewale Oladehin,
and especially David Yanacek!

62. Appendix:
Lifecycle Management

63. Actuators
Data storage
& analytics
Device lifecycle management
Control
automation
AWS IoT
Sensors
Maintenance
1

64. Lifecycle workflow
Notify operator
1
Connected Disconnected Still
disconnected?

65. AWS IoT Rules Engine & Amazon SNS
Push Notifications
Apple APNS Endpoint, Google GCM Endpoint, Amazon ADM Endpoint, Windows
WNS
Amazon SNS -> HTTP Endpoint (or SMS or email)
Call HTTP based 3rd party endpoints through SNS with subscription and retry support
SNS
2

66. Detecting disconnects
DisconnectedConnected
Graceful disconnect
Crash
Back online

67. Lifecycle events
• Connect
• PUBLISH lifecycle/sensor-123
{"status": "online"}
• Disconnect (graceful)
• PUBLISH lifecycle/sensor-123
{"status": "offline"}
• Disconnect (crash)
• PUBLISH lifecycle/sensor-123
{"status": "offline", "isCrash": true}

68. AWS IoT Rules Engine’s Flexibility
SELECT *, clientId() as MQTTClientId
FROM 'one/rule'
WHERE
startsWith(topic(2), 'IME33') AND
(state = 'INIT' OR hydro_temp >
surface_temp)",
"actions":
[{
"republish": {
"topic":
"controllers/${substring(topic(3),
3, 5)}",
}]

69. Handling lifecycle events
SELECT
status,
topic(2) as deviceId,
timestamp() as time,
isCrash
FROM lifecycle/#
WHERE status='offline'
- Look up mobile push ID for device owner
- Send SNS mobile push
AWS Lambda Function

70. Delayed lifecycle events
SELECT
status,
topic(2) as deviceId,
timestamp() as time,
isCrash
FROM lifecycle/#
Device Status Time
sensor-123 connected 11:30
…
- Double-check the status in DynamoDB
- Send SNS push notification if still offline
- Store update device status in DynamoDB
- If offline: enqueue an SQS message with
DelaySeconds
AWS Lambda Function
SQS Message (15 minutes later)
Amazon
DynamoDB

71. Generating lifecycle events
• Connect
• PUBLISH lifecycle/sensor-123
{"status": "online"}
• Disconnect (graceful)
• PUBLISH lifecycle/sensor-123
{"status": "offline"}
• Disconnect (crash)
• PUBLISH lifecycle/sensor-123
{"status": "offline", "isCrash": true}

72. Lifecycle events: connecting
1. CONNECT
2. PUBLISH lifecycle/sensor-123
{"state": "online"}

73. Lifecycle events: disconnecting
1. PUBLISH lifecycle/sensor-123
{“state”: “offline”}
2. DISCONNECT

74. Last Will and Testament
CONNECT message parts:
Protocol: MQTT 3.1.1
ClientId: abc
KeepAlive: 60 seconds
LastWill PUBLISH message:
Topic: foo/bar
QoS: 1
Payload: {"foo": "bar"}

75. Lifecycle events: connecting
1. CONNECT with LWT:
PUBLISH lifecycle/sensor-123
{“crash”: true, “state”: “offline”}
2. PUBLISH lifecycle/sensor-123
{“state”:”online”}

76. Lifecycle events: simplified
• Automatic lifecycle PUBLISH messages
PUBLISH $aws/events/presence/connected/abc123
{
"ClientId": "abc123",
"Principal": "arn:aws:...",
"Timestamp": "2016-01-31T11:30",
"Status": "disconnected",
“GracefulDisconnect": true
}

77. Takeaways: lifecycle management
• Publish messages, use LWT for lifecycle events
• SQS delayed messages and DynamoDB can reduce
false positives
• Automatic lifecycle events