Gilmour: GopherCon London

Transcript

1. GILMOUR
   DESIGN PATTERNS IN MICRO-SERVICES
   ARCHITECTURES &
   * THIS TALK IS BUZZWORD COMPLIANT.
   

   View Slide

2. MICROSERVICES & GILMOUR?
   

   View Slide

3. ABOUT
   WHO AM I?
   ▸ Twitter: @meson10
   ▸ mailto: piyush@piyushverma.net
   ▸ piyushverma.net
   

   View Slide

4. SOA (VS SERVERS)
   WHAT ARE SERVICES?
   ▸ Logical Entities.
   ▸ Independent.
   ▸ Comprised of more services.
   ▸ Multiple instances.
   ▸ Scalable.
   ▸ Flexible.
   

   View Slide

5. WHODUNNIT
   WHY ARE SERVICES?
   ▸ Software Problem
   ▸ Heterogeneity
   ▸ Flexibility
   ▸ Scalability
   ▸ Management Problem
   ▸ Authority
   ▸ Responsibility.
   

   View Slide

6. DESIGN PATTERNS
   

   View Slide

7. REQUEST RESPONSE
   Boomerang
   

   View Slide

8. REQUEST RESPONSE
   CALLER SERVICE
   HELLO?
   ALOHA!
   WRONG NUMBER
   SENDER_ID
   CODE
   

   View Slide

9. REQUEST
   req := g.NewRequest(“message.echo”)
   data := G.NewMessage().SetData("hello?")
   resp, err := req.Execute(data)
   fmt.Println("Echoclient: received", resp)
   g.ReplyTo(“message.echo", func(req *G.Request, resp *G.Message) {
   var msg string
   req.Data(&msg) //Parse data from incoming request.
   resp.SetData(“Pong") //Send response.
   }, opts)
   RECEIVER
   REQUEST
   

   View Slide

10. ASYNC / SIGNAL-SLOTS
    

    View Slide

11. ASYNC: SIGNALS AND SLOTS
    Slots
    Signal
    EVENT.POWER_UP
    EVENT.*
    EVENT.*
    GROUP: REVENUE
    FP SCALING
    PUSH MESSAGE
    SERVER
    ADS/
    MARKETINGSERVER
    ANALYTICS/
    LOGS
    BADGES/
    UNLOCKS
    EVENT.*
    ADS/MARKETING
    GAME GILMOUR
    EVENT
    EVENT.BRIDGE_FALL
    FALL FROM
    BRIDGE
    FALL FROM
    BRIDGE
    FALL FROM
    BRIDGE
    GROUP: BRIDGE_FALL
    

    View Slide

12. ASYNC: SLOTS
    g.Slot("example.log", func(req *G.Request) {
    var msg string
    req.Data(&msg)
    log.Println(req.Sender(), "->", msg)
    })
    msg := G.NewMessage().SetData(“log message emitted”)
    g.Signal("example.log", msg)
    SIGNAL
    RECEIVER / SLOT
    

    View Slide

13. SIGNAL-SLOT
    VS.
    REQUEST-RESPONSE
    

    View Slide

14. ASYNC VS SYNC: WHAT TO USE WHEN?
    REQUEST-RESPONSE
    ▸ Confirmed delivery.
    ▸ Response or Error; Guaranteed.
    ▸ HTTP-like.
    ▸ Sender responsible for Errors.
    

    View Slide

15. View Slide

16. ASYNC VS SYNC: WHAT TO USE WHEN?
    SIGNAL-SLOTS
    ▸ Broadcasting.
    ▸ Wildcard topics.
    ▸ UDP-Like.
    ▸ Ok with unreliable delivery.
    ▸ Receiver responsible for Errors.
    

    View Slide

17. ERRORS: REQUEST - RESPONSE
    CALLER SERVICE
    HELLO?
    WRONG NUMBER
    SENDER_ID
    CODE
    ERROR
    REPORTER
    HYSTRIX
    PAGER DUTY
    HONEY BADGER
    EMAIL ALERTS
    

    View Slide

18. ERRORS: SIGNAL - SLOTS
    WEB GILMOUR
    EMAIL
    NOTIFIER
    SMS
    NOTIFIER
    PUSH
    MESSAGE
    Slots
    Signal
    PAGER DUTY
    HONEY BADGER
    EMAIL ALERTS
    ERROR
    REPORTER
    GILMOUR.ERROR
    

    View Slide

19. DISCOVERY & LOAD-BALANCING
    

    View Slide

20. LOAD BALANCING - TYPICAL
    NOTIFICATION
    SERVER 1
    REPLICATED
    LOAD
    BALANCER
    X.Y.Y.Z?
    CALLER
    SERVICE
    REGISTER
    I, AM
    HEALTHY?
    NOTIFICATION
    SERVER 2
    NOTIFICATION
    SERVER 3
    

    View Slide

21. LOAD BALANCING : GILMOUR
    NOTIFICATION
    SERVER 2
    GILMOUR
    MANAGER.NOTIFICATION
    NOTIFICATION
    SERVER 1
    NOTIFICATION
    SERVER 3
    MANAGER.NOTIFICATION
    MANAGER.NOTIFICATION
    CALLER
    SERVICE
    MANAGER.NOTIFICATION
    {“NOTIFICATION”: “DATA”}
    

    View Slide

22. DISCOVERY AND LOAD BALANCING
    LOOK MA, NO SERVICE DISCOVERY
    ▸ Gilmour works on Subscriber Publisher.
    ▸ No querying discovery process.
    ▸ Call service and not servers.
    

    View Slide

23. DISCOVERY AND LOAD BALANCING
    LOOK MA, NO LOAD BALANCING
    ▸ Has Capacity, Will Serve.
    ▸ Message delivered to one and all.
    ▸ Fittest node acquires lock first.
    

    View Slide

24. ERRORS:
    DETECTION & HANDLING
    

    View Slide

25. GILMOUR ERRORS
    ERROR DETECTION
    ▸ Timeout:
    ▸ Server-side - Execution expired.
    ▸ Client-side - Response too late.
    ▸ Confirm Subscribers or 404.
    ▸ Distinguished network failures.
    

    View Slide

26. GILMOUR ERRORS
    ERROR HANDLING
    ▸ Ignore
    ▸ Publish
    ▸ Queue
    func main() {
    redis := backends.MakeRedis("127.0.0.1:6379", "pass")
    g := G.Get(redis)
    //g.SetErrorPolicy(protocol.ErrorPolicyQueue)
    //g.SetErrorPolicy(protocol.ErrorPolicyIgnore)
    g.SetErrorPolicy(protocol.ErrorPolicyPublish)
    }
    

    View Slide

27. ERRORS:
    type GilmourError interface {
    Marshal() ([]byte, error)
    GetTopic() string
    GetSender() string
    GetCode() int
    GetTimestamp() string
    GetRequestData() string
    GetUserData() string
    GetBacktrace() interface{}
    }
    type gError struct {
    topic string //topic to which the request was sent
    requestData string //request payload
    userData string //implementation dependent information
    sender string //sender from the request
    timestamp string //timestamp of error
    backtrace interface{} //backtrace of the error
    code int //response code (non-200)
    }
    ERROR
    IMPLEMENTATION
    

    View Slide

28. GILMOUR ERRORS
    ERROR SAMPLE
    

    View Slide

29. OPS: MONITORING
    

    View Slide

30. OPS: MONITORING
    SERVERS
    ▸ Logging
    ▸ Error Reporting
    ▸ Health
    

    View Slide

31. OPS: MONITORING
    SERVICES
    ▸ Are the services servicing?
    ▸ Do services have spare capacity?
    ▸ Do services respond in time?
    ▸ Health Bulletin: github.com/gilmour-libs/health-bulletin
    

    View Slide

32. OPS: MONITORING
    

    View Slide

33. OPS: MONITORING
    

    View Slide

34. OPS: MONITORING
    Feb 23 12:25:14 qa-next-backend-manager0.datascale.io gilmour_health: D, [2016-02-23T06:55:12.930435
    #22111] DEBUG -- : Ensuring topics have at least one subscriber
    Feb 23 12:25:14 qa-next-backend-manager0.datascale.io gilmour_health: D, [2016-02-23T06:55:12.931160
    #22111] DEBUG -- : Ensuring that gilmour servers respond to health ping
    Feb 23 12:26:14 qa-next-backend-manager0.datascale.io backend_manager: D, [2016-02-23T06:56:12.932775
    #9605] DEBUG -- : Executing handler for gilmour.health.ip-10-0-0-146-pid-9605-uuid-9d0484dd-08e5-4ec3-
    a3e4-9a109634712b from 2881867d41f24228b462b9eff0b7e88e
    2016-04-03 00:06:56 d0e4ea929645600e2fa97657091613a4 -> Cleared VPC jenkins-backend-manager-pipeline-qa-17
    2016-04-03 00:06:56 d0e4ea929645600e2fa97657091613a4 -> Clearing databag customer_jenkins-backend-manager-
    pipeline-qa-17_config::vpc_config
    2016-04-03 00:06:57 d0e4ea929645600e2fa97657091613a4 -> Deleted data_bag[customer_jenkins-backend-manager-
    pipeline-qa-17_config]
    2016-04-03 00:06:57 d0e4ea929645600e2fa97657091613a4 -> Clearing DNS entries
    2016-04-03 00:07:00 d0e4ea929645600e2fa97657091613a4 -> Cleaning Chef environment
    2016-04-03 00:07:01 d0e4ea929645600e2fa97657091613a4 -> Deleted customer_jenkins-backend-manager-pipeline-
    qa-17
    REQUEST LOGGING
    

    View Slide

35. 

    COMPOSITIONS
    

    View Slide

36. COMPOSITION
    ▸ compose - service1 | service2 | composition3
    ▸ andand - service1 && service2 && composition3
    ▸ oror - service1 || composition2 || service3
    ▸ batch
    ▸ (service1; service2; service3) > out
    ▸ service1; composition2; service3 > out
    ▸ parallel
    ▸ lambda functions
    

    View Slide

37. COMPOSITION
    popular := g.NewParallel(
    g.NewRequestComposition("example.popular3"),
    g.NewRequestComposition("example.popular4"),
    )
    batch := g.NewPipe(
    g.NewRequestComposition("example.fetch"),
    g.NewRequestComposition("example.words"),
    g.NewRequestComposition("example.count"),
    popular,
    )
    data := G.NewMessage()
    data.SetData("https://s3.amazonaws.com/ds-data-sample/test.txt")
    msg := <-batch.Execute(data)
    

    View Slide

38. COMPOSITION
    popular := g.NewParallel(
    g.NewRequestComposition("example.popular3"),
    g.NewRequestComposition(“example.popular4"),
    g.NewRequestComposition("example.popular5"),
    )
    batch := g.NewPipe(
    g.NewRequestComposition("example.fetch"),
    g.NewRequestComposition("example.words"),
    g.NewRequestComposition("example.stopfilter"),
    g.NewRequestComposition("example.count"),
    popular,
    &MyComposer{},
    )
    

    data := G.NewMessage()
    data.SetData("https://s3.amazonaws.com/ds-data-sample/test.txt")
    msg := <-batch.Execute(data)
    

    View Slide

39. COMPOSITION
    batch := e.NewPipe(
    e.NewRequest("weather.fetch"),
    e.NewRequest("weather.group"),
    e.NewParallel(
    e.NewPipe(
    e.NewLambda(monthLambda("jan")),
    e.NewParallel(
    e.NewRequest("weather.min"),
    e.NewRequest("weather.max"),
    ),
    ),
    e.NewPipe(
    e.NewLambda(monthLambda("feb")),
    e.NewParallel(
    e.NewRequest("weather.min"),
    e.NewRequest("weather.max"),
    ),
    ),
    ),
    )
    

    View Slide

40. DEMO
    Please contact to
    advertise here.
    

    View Slide

41. SAMPLE TEXT
    

    View Slide

42. EXAMPLES: POPULAR WORD COUNT
    ▸ Given
    ▸ A S3 URL pointing to a file that has (lot of) random text.
    ▸ Goal
    ▸ Eliminate stop words from the file.
    ▸ Find most popular 3, 4, and 5 letter words.
    ▸ Find these in parallel.
    

    View Slide

43. EXAMPLES: WEATHER AGGREGATION
    ▸ Output
    ▸ Popular 3 letter words: [way, for]
    ▸ Popular 4 letter words: [Text, copy]
    ▸ Popular 5 letter words: [Blind, blind]
    

    View Slide

44. EXAMPLES: COMPOSITION
    

    View Slide

45. GILMOUR
    RECAP
    ▸ No Load Balancer
    ▸ No Service Discovery
    ▸ Signal, Slots.
    ▸ Broadcast using Wildcard Topics.
    ▸ Exclusion groups.
    ▸ Sync - Request, Reply
    ▸ Error and Health Monitoring - HealthBulletin.
    ▸ Failure Detection
    ▸ Composition - Real micro-services.
    ▸ No message persistence.
    

    View Slide

46. GILMOUR
    CREDITS & LINKS
    ▸ Aditya Godbole
    ▸ Amit Sethi
    ▸ Mohan Dutt
    ▸ Poorva Mahajan
    ▸ Datascale.IO datascale.io
    ▸ Repository github.com/gilmour-libs
    

    View Slide