10 Traumas in 10 minutes

10 Traumas in 10 minutes

A lightening talk - walking through 10 trauma's for developing applications for the cloud

B83d5b590577969ee79a5ad845411a7d?s=128

Amy Palamountain

October 26, 2012
Tweet

Transcript

  1. Hi

  2. @ammeep

  3. None
  4. 10 E Traumas in 10 minutes

  5. 1 E Deployments are painful

  6. automated deployments are hard in the cloud

  7. automated deployments are hard in the cloud dirty lies

  8. automated build automated test automated deploy

  9. automate provisioning API

  10. automate network setup API

  11. automate deployment API

  12. automate a the thing

  13. 2 E i’m married to my cloud

  14. AbstractProxyBooleanProvider avoid a ma iage of inconvenience just don’t commit

    be abstract
  15. abstract message queues abstract storage providers abstract auto scaling

  16. abstract away the gory details if you don’t own it

  17. 3 E insanity the definition of

  18. Oh HAI please may i retrieve my data? sure ...

    i’ll hold...
  19. hello? ... ok holding ... may i have my data?

  20. data? what? ........

  21. None
  22. Are we there yet? Are we there yet? Are we

    there yet? Are we there yet? Are we there yet? Are we there yet? Are we there yet? Are we there yet? not a are born equal transient faults
  23. exponential the nature of the transient fault Unde tand perhaps

    try back off
  24. 4 E storage the limitations of

  25. Transactions can be capped but can we still have infinite

    storage capacity infinite throughput and
  26. many ways to skin a cat When is it accessed

    Wh is accessing it What is accessed
  27. 111001001100101011100110110010101101110011101000110000101110100011010010110111101 011111100100000010000010110110100100000011010010010000001110100011010000110000101 0010000001100010011011110111001001101001011011100110011100111111 110011001110101011000110110101101101001011011100110011100100000011000110111010101 111010000100000011000100110000101101100011011000111001100100000011110010110111101 010000001100011011000010110111000100000011100100110010101100001011001000010000001 110100001101001011100110011111100100000011101110110100001111001001000000111010001 110010100100000011001100111010101100011011010110010000001100001011100100110111001 010000001111001011011110111010100100000011011000110100101110011011101000110010101 110100101101110011001110010000001110100011011110010000001101101011110010010000001

    111001001100101011100110110010101101110011101000110000101110100011010010110111101 011111100100000010000010110110100100000011010010010000001110100011010000110000101 010000001100010011011110111001001101001101110011001110011111110011100100000011000 101011011100111010000100000011000100110000101101100011011000111001100100000011110 111011101010010000001100011011000010110111000100000011100100110010101100001011001 000011101000110100001101001011100110011111100100000011101110110100001111001001000 100011010000110010100100000011001100111010101100011011010110010000001100001011100 110011101000010000001111001011011110111010100100000011011000110100101110011011101 101011011100110100101101110011001110010000001110100011011110010000001101101011110 000011100000111001001100101011100110110010101101110011101000110000101110100011010 111011011100011111100100000010000010110110100100000011010010010000001110100011010 100011001000010111010000100000011000100110111101110010011010010110111001100111001 110011001110101011000110110101101101001011011100110011100100000011000110111010101 111010000100000011000100110000101101100011011000111001100100000011110010110111101 010000001100011011000010110111000100000011100100110010101100001011001000010000001 110100001101001011100110011111100100000011101110110100001111001001000000111010001 110010100100000011001100111010101100011011010110010000001100001011100100110111001 Partition you data Spread out you data Cache you data
  28. 5 E web scale we’ve got to be

  29. web scale cloud scale what does that even mean?

  30. twitter uses it but... it must be g d

  31. choice of technology is not a function of fashion

  32. it ok t be relational if it works for you

    & your data
  33. 6 E go daddy goes down

  34. DNS is a single point of failure avoiding this is

    a tricky problem
  35. have a simple back up plan Use more than one

    domain
  36. 7 E slow our app is

  37. hogging resources 100% CPU blocking locking opening cursors never closing

    interrupt obstructing stalling no connection pooling n+1 leaking memory performance ++ deadlocks
  38. Just scale Up not optimised performance for node

  39. Just scale out not optimised for cost

  40. Bo leneck cost performance tuning is important

  41. 8 E chaos monkey

  42. patche upgrade VM shuffling DNS changes shared hardware slo node

    Storage Transient failur queue Transient failures physical location time Storage Transient failures patches o upgrade queue Transient failures provisioning failure autoscaling capacity latency
  43. patche upgrade VM shuffling DNS changes shared hardware Slo node

    Storage Transient failur queue Transient failures physical location time Storage Transient failures patches o upgrade queue Transient failures provisioning failure autoscaling capacity latency VOLATILE
  44. a ept that It will fail and plan fo it

  45. simulate chao get that monkey working for you

  46. 9 E you have lupus

  47. it’s 2 am the phone rings your app is crashing

    servers are unresponsive
  48. and it gets worse YOU CAN’T TELL WHAT IS GOING

    ON
  49. centrally store logs

  50. user activity correlate be able to

  51. diagnostics n d t be a first class concern

  52. 10 E you are going to fail

  53. accept it You’re not going to get it right not

    the first time and maybe not the time after that
  54. change i coming

  55. your designs responsibly to adapt be prepared

  56. 10 E phew thats it

  57. for listening thanks @ammeep