Reliability of Distributed Systems

Transcript

1. 1.

   Reliability of Distributed Systems - Piyush Verma

2. 2.

   Every product either dies a hero or lives long enough

   to hit Reliability issues. 2
3. 3.

   Customer Empathy No Chooran. Cost to Everything Architectures adapt to

   $ Priority 01 02 03 04 3 Take it and Go
4. 4.

   Service receives SMS User sends SMS Remind me to buy

   milk at 6:30 PM to 53308 Cron gets Activated when time is right. Call the User 4 Sample Product
5. 5.

   Service receives SMS User sends SMS Remind me to buy

   milk at 6:30 PM to 53308 Cron gets Activated when time is right. Call the User 5 Sample Product: Inbound
6. 6.

   Cron gets Activated when time is right. Call the User

   6 Sample Product: Outbound Service receives SMS User sends SMS Remind me to buy milk at 6:30 PM to 53308
7. 7.

   7 Inbound Connection

8. 8.

   — Leslie Lamport https://www.microsoft.com/en-us/research/uploads/prod/2016 /12/Distribution.pdf “A distributed system is one

   in which the failure of a computer you didn’t even know existed can render your own computer unusable” 8
9. 9.

   Four Flavors of Failure 9 Disk Network CPU Memory

10. 10.

    Network is Reliable Intra-LAN latency is ~ Zero Network is

    Homogeneous Network cost is Zero 01 02 03 04 10
11. 11.

    11 Scope of Failures: Again

12. 12.

    At-least one server is online All servers are below 100%

    All servers are responding within x ms. All of the above. 01 02 03 04 12
13. 13.

    13 #1 Server is Unavailable

14. 14.

    14 Replication Available

15. 15.

    15 Replication Available

16. 16.

    16 Available != Load Balanced

17. 17.

    17 Load Balanced

18. 18.

    18 Architecture of a Balancer

19. 19.

    GCP AWS On-Prem Azure Who uses What? 19

20. 20.

    Trilemma 20

21. 21.

    Trilemma 21 Available Economical Endurable

22. 22.

    22 Available + Load Balanced

23. 23.

    Load Balancing 23

24. 24.

    24 Monty Hall Problem: Was Marilyn vos Savant, right?

25. 25.

    25 Server-side Load Balancing Example: Fabio

26. 26.

    26 Look-aside Load Balancing Example: Consul/ DNS

27. 27.

    27 Client-side Load Balancing Example: Ribbon

28. 28.

    28 Client-side Load Balancing Example: Ribbon + Curator

29. 29.

    29 Problems

30. 30.

    30 Load Shedding

31. 31.

    — Tyler McMullen https://www.infoq.com/presentations/load-balancing/ “Load Balancing is almost Impossible” 31

32. 32.

    Alternate Reliability 32

33. 33.

    33 Asynchronous Architectures

34. 34.

    34 Asynchronous Architectures Example: RabbitMQ Kafka Kinesis SQS

35. 35.

    Cron gets Activated when time is right. Call the User

    35 Sample Product: Outbound Service receives SMS User sends SMS Remind me to buy milk at 6:30 PM to 53308 Part 1
36. 36.

    36 Outbound

37. 37.

    37 Scope of Failure: Outbound

38. 38.

    38 Retries

39. 39.

    39 Retries: Transient Failures

40. 40.

    40 Exponential Backoff: Short term Transient Failures ✋ ✋ ✋

    ✋ ✋
41. 41.

    41 Circuit Breaking: Long Term Transient Failures

42. 42.

    42 Revisited

43. 43.

    Dilemma 43 At-least Once Exactly-Once At-most Once

44. 44.

    44 At-most once delivery

45. 45.

    45 At-least once delivery

46. 46.

    46 Exactly once delivery

47. 47.

    47 Exactly once delivery = At-least-once Delivery + Exactly-once Processing

48. 48.

    Keys to Only-Once delivery 48 Atomic Window Idempotent

49. 49.

    Out-of-Order delivery 49

50. 50.

    Revisit 50

51. 51.

    Service receives SMS User sends SMS Remind me to buy

    milk at 6:30 PM to 53308 Cron gets Activated when time is right. Call the User 51 Sample Product
52. 52.

    52 Problems of State

53. 53.

    53 Problems of State

54. 54.

    54 Locked /Serialization

55. 55.

    55 Master/ Master/ Slave

56. 56.

    56 Clustering

57. 57.

    Scalability 57 Data Replication Reduced Communication Logic/Data Decentralization

58. 58.

    CAP Theorem [Sab topi pehna rahe] 58

59. 59.

    59 Trilemma Available Partition Consistent

60. 60.

    PACeLC Theorem 60

61. 61.

    Dilemma 61 Consistency Latency

62. 62.

    62 Revised Flow

63. 63.

    63 What about Spanner? What about Calvin?

64. 64.

    Reliable System 64 Scalable Correct Transparent

65. 65.

    Access Transparency Location Transparency Concurrency Transparency Failure Transparency 01 02

    03 04 65
66. 66.

    Size Scalability Geographical Scalability 01 02 66

67. 67.

    Summary 67 Consistent Available Economical Low Latency

68. 68.

    68 All Put Together

69. 69.

    Embrace your Bugs No Silver Bullet Cost to Everything Product

    First 01 02 03 04 69
70. 70.

    Does anyone have any questions? piyush@piyushverma.net Thanks 70