Slide 1

Slide 1 text

Development was the easy part

Slide 2

Slide 2 text

Development was the easy part

Slide 3

Slide 3 text

André Arko @indirect

Slide 4

Slide 4 text

No content

Slide 5

Slide 5 text

The Ruby Way therubyway.io

Slide 6

Slide 6 text

stickers!

Slide 7

Slide 7 text

No content

Slide 8

Slide 8 text

stickers!

Slide 9

Slide 9 text

No content

Slide 10

Slide 10 text

stickers!

Slide 11

Slide 11 text

Development is very different

Slide 12

Slide 12 text

from Production

Slide 13

Slide 13 text

you rn →

Slide 14

Slide 14 text

you later →

Slide 15

Slide 15 text

Metrics

Slide 16

Slide 16 text

Metrics are important

Slide 17

Slide 17 text

Metrics tell you what is happening

Slide 18

Slide 18 text

Metrics convince you you understand

Slide 19

Slide 19 text

Averages convince you you understand

Slide 20

Slide 20 text

Averages are lie-candy for your brain

Slide 21

Slide 21 text

Averages 5 -5 -4 -3 -2 -1 0 1 2 3 4 0 0.1 0.2 0.3 0.4

Slide 22

Slide 22 text

Averages 5 -5 -4 -3 -2 -1 0 1 2 3 4 0 0.1 0.2 0.3 0.4

Slide 23

Slide 23 text

No content

Slide 24

Slide 24 text

No content

Slide 25

Slide 25 text

just heard “we have a great average” →

Slide 26

Slide 26 text

Averages mask problems

Slide 27

Slide 27 text

10 0 1 2 3 4 5 6 7 8 9 250 0 50 100 150 200

Slide 28

Slide 28 text

Graph the median

Slide 29

Slide 29 text

10 0 1 2 3 4 5 6 7 8 9 250 0 50 100 150 200

Slide 30

Slide 30 text

Graph 95th percentile

Slide 31

Slide 31 text

10 0 1 2 3 4 5 6 7 8 9 250 0 50 100 150 200

Slide 32

Slide 32 text

Graph 99th percentile

Slide 33

Slide 33 text

10 0 1 2 3 4 5 6 7 8 9 1000 0 250 500 750

Slide 34

Slide 34 text

Aggregate graphs another kind of average

Slide 35

Slide 35 text

No content

Slide 36

Slide 36 text

Breakout graphs see individual variations

Slide 37

Slide 37 text

No content

Slide 38

Slide 38 text

No content

Slide 39

Slide 39 text

Aggregate alerts more dead servers than alive servers

Slide 40

Slide 40 text

site’s up if any servers are up!

Slide 41

Slide 41 text

Breakout alerts first dead server not all the servers

Slide 42

Slide 42 text

Servers

Slide 43

Slide 43 text

you have no idea what is going on Servers

Slide 44

Slide 44 text

really.

Slide 45

Slide 45 text

Routing

Slide 46

Slide 46 text

your app has this Routing

Slide 47

Slide 47 text

how does it work? Routing

Slide 48

Slide 48 text

Development App You

Slide 49

Slide 49 text

People Router Server App App Router Server App App Router Production

Slide 50

Slide 50 text

how slow is it? Routing

Slide 51

Slide 51 text

does it back up? Routing

Slide 52

Slide 52 text

Request

Slide 53

Slide 53 text

not the time you measure Request

Slide 54

Slide 54 text

wall-clock time from real clients Request

Slide 55

Slide 55 text

make requests from around the world Request

Slide 56

Slide 56 text

graph them Request

Slide 57

Slide 57 text

graph them alert on them Request

Slide 58

Slide 58 text

graph them alert on them thank me later Request

Slide 59

Slide 59 text

VM lag

Slide 60

Slide 60 text

do you have it? VM lag

Slide 61

Slide 61 text

do you check for it? VM lag

Slide 62

Slide 62 text

do you know how to check for it? VM lag

Slide 63

Slide 63 text

Runtime

Slide 64

Slide 64 text

how do you tell you lost consciousness? Runtime

Slide 65

Slide 65 text

do you have it? Runtime

Slide 66

Slide 66 text

do you have it? you have it. Runtime

Slide 67

Slide 67 text

do you have it? you have it. how bad is it? Runtime

Slide 68

Slide 68 text

Data Stores

Slide 69

Slide 69 text

in production Data Stores

Slide 70

Slide 70 text

in production are distributed Data Stores

Slide 71

Slide 71 text

what does that mean?

Slide 72

Slide 72 text

your experience (so far) is wrong

Slide 73

Slide 73 text

Saving data

Slide 74

Slide 74 text

tries to save your data Saving data

Slide 75

Slide 75 text

might save your data Saving data

Slide 76

Slide 76 text

Replication

Slide 77

Slide 77 text

is not data- saving magic Replication

Slide 78

Slide 78 text

tries to save your data… Replication

Slide 79

Slide 79 text

tries to save your data… repeatedly Replication

Slide 80

Slide 80 text

Postgres

Slide 81

Slide 81 text

totally safe, right? Postgres

Slide 82

Slide 82 text

async replication Postgres

Slide 83

Slide 83 text

network failures lose “saved” data Postgres

Slide 84

Slide 84 text

Redis

Slide 85

Slide 85 text

is single-threaded Redis

Slide 86

Slide 86 text

has no failover Redis

Slide 87

Slide 87 text

Mongo replication sets default to one write

Slide 88

Slide 88 text

Mongo demand N copies survive N-1 failures

Slide 89

Slide 89 text

trust no one

Slide 90

Slide 90 text

if you didn’t try it yourself you are guessing

Slide 91

Slide 91 text

test it yourself

Slide 92

Slide 92 text

So, in the end what did we learn?

Slide 93

Slide 93 text

Production is fundamentally

Slide 94

Slide 94 text

Production is fundamentally systemically

Slide 95

Slide 95 text

Production is fundamentally systemically different

Slide 96

Slide 96 text

Failures will happen

Slide 97

Slide 97 text

Failures can be resisted

Slide 98

Slide 98 text

Failures should not result in one-off patches

Slide 99

Slide 99 text

Survival requires systematic trials & testing

Slide 100

Slide 100 text

Good luck! %☁