Slide 1

Slide 1 text

3/20/15 Ensuring success during disaster SRECON 2015

Slide 2

Slide 2 text

3/20/15 ENSURING SUCCESS DURING DISASTER @dougbarth

Slide 3

Slide 3 text

3/20/15 ENSURING SUCCESS DURING DISASTER

Slide 4

Slide 4 text

3/20/15 ENSURING SUCCESS DURING DISASTER Dev Ops Consider killing this

Slide 5

Slide 5 text

3/20/15 ENSURING SUCCESS DURING DISASTER How is babby PagerDuty formed?

Slide 6

Slide 6 text

3/20/15 ENSURING SUCCESS DURING DISASTER

Slide 7

Slide 7 text

3/20/15 ENSURING SUCCESS DURING DISASTER Reliability

Slide 8

Slide 8 text

3/20/15 ENSURING SUCCESS DURING DISASTER Agenda

Slide 9

Slide 9 text

3/20/15 ENSURING SUCCESS DURING DISASTER DR vs. HA Data DR Failover Active/Active Legacy Systems Q&A

Slide 10

Slide 10 text

3/20/15 ENSURING SUCCESS DURING DISASTER Disaster Recovery

Slide 11

Slide 11 text

3/20/15 ENSURING SUCCESS DURING DISASTER A plan for surviving rare failure events that threaten our ability to continue operating

Slide 12

Slide 12 text

3/20/15 ENSURING SUCCESS DURING DISASTER High availability

Slide 13

Slide 13 text

3/20/15 ENSURING SUCCESS DURING DISASTER DR HA Rare failures Common failures Slow recovery Fast recovery

Slide 14

Slide 14 text

3/20/15 ENSURING SUCCESS DURING DISASTER Latency

Slide 15

Slide 15 text

3/20/15 ENSURING SUCCESS DURING DISASTER DR needs to exist

Slide 16

Slide 16 text

3/20/15 ENSURING SUCCESS DURING DISASTER DR needs to be tested

Slide 17

Slide 17 text

3/20/15 ENSURING SUCCESS DURING DISASTER DR needs to be tested for correctness

Slide 18

Slide 18 text

3/20/15 ENSURING SUCCESS DURING DISASTER DR needs to be tested for capacity

Slide 19

Slide 19 text

3/20/15 ENSURING SUCCESS DURING DISASTER DR needs to be tested for execution

Slide 20

Slide 20 text

3/20/15 ENSURING SUCCESS DURING DISASTER Data DR

Slide 21

Slide 21 text

3/20/15 ENSURING SUCCESS DURING DISASTER You can never have too many copies

Slide 22

Slide 22 text

3/20/15 ENSURING SUCCESS DURING DISASTER Lose a disk? Lose a server?

Slide 23

Slide 23 text

3/20/15 ENSURING SUCCESS DURING DISASTER Secondary Primary Secondary

Slide 24

Slide 24 text

3/20/15 ENSURING SUCCESS DURING DISASTER Data corruption?

Slide 25

Slide 25 text

3/20/15 ENSURING SUCCESS DURING DISASTER Backups

Slide 26

Slide 26 text

3/20/15 ENSURING SUCCESS DURING DISASTER Test restorations

Slide 27

Slide 27 text

3/20/15 ENSURING SUCCESS DURING DISASTER DROP TABLE USERS; DELETE FROM USERS;

Slide 28

Slide 28 text

3/20/15 ENSURING SUCCESS DURING DISASTER Delayed secondary Primary Secondary (delayed 2 hrs)

Slide 29

Slide 29 text

3/20/15 ENSURING SUCCESS DURING DISASTER Failover

Slide 30

Slide 30 text

3/20/15 ENSURING SUCCESS DURING DISASTER Primary DC & Secondary DC

Slide 31

Slide 31 text

3/20/15 ENSURING SUCCESS DURING DISASTER The flip

Slide 32

Slide 32 text

3/20/15 ENSURING SUCCESS DURING DISASTER Testing == outage

Slide 33

Slide 33 text

3/20/15 ENSURING SUCCESS DURING DISASTER So it goes untested

Slide 34

Slide 34 text

3/20/15 ENSURING SUCCESS DURING DISASTER Tested infrequently

Slide 35

Slide 35 text

3/20/15 ENSURING SUCCESS DURING DISASTER Flip test Flip test Breaking change Failure window

Slide 36

Slide 36 text

3/20/15 ENSURING SUCCESS DURING DISASTER Forgotten roles

Slide 37

Slide 37 text

3/20/15 ENSURING SUCCESS DURING DISASTER Low capacity

Slide 38

Slide 38 text

3/20/15 ENSURING SUCCESS DURING DISASTER Expensive

Slide 39

Slide 39 text

3/20/15 ENSURING SUCCESS DURING DISASTER Active/Active

Slide 40

Slide 40 text

3/20/15 ENSURING SUCCESS DURING DISASTER PagerDuty Active/Active

Slide 41

Slide 41 text

3/20/15 ENSURING SUCCESS DURING DISASTER Multiple regions Use all the AZs!

Slide 42

Slide 42 text

3/20/15 ENSURING SUCCESS DURING DISASTER

Slide 43

Slide 43 text

3/20/15 ENSURING SUCCESS DURING DISASTER Constant validation

Slide 44

Slide 44 text

3/20/15 ENSURING SUCCESS DURING DISASTER Failure Friday HTTPS://WWW.PAGERDUTY.COM/BLOG/FAILURE-FRIDAY-AT-PAGERDUTY/

Slide 45

Slide 45 text

3/20/15 ENSURING SUCCESS DURING DISASTER Trade latency for reliability

Slide 46

Slide 46 text

3/20/15 ENSURING SUCCESS DURING DISASTER 30ms RTT

Slide 47

Slide 47 text

3/20/15 ENSURING SUCCESS DURING DISASTER Legacy Systems

Slide 48

Slide 48 text

3/20/15 ENSURING SUCCESS DURING DISASTER Older systems still use failover

Slide 49

Slide 49 text

3/20/15 ENSURING SUCCESS DURING DISASTER Correctness Capacity Execution

Slide 50

Slide 50 text

3/20/15 ENSURING SUCCESS DURING DISASTER Correctness Capacity Execution

Slide 51

Slide 51 text

3/20/15 ENSURING SUCCESS DURING DISASTER 1% of requests go to DR site

Slide 52

Slide 52 text

3/20/15 ENSURING SUCCESS DURING DISASTER Finds issues immediately

Slide 53

Slide 53 text

3/20/15 ENSURING SUCCESS DURING DISASTER Requires DR uses production DB DR Prod App App App App App App

Slide 54

Slide 54 text

3/20/15 ENSURING SUCCESS DURING DISASTER 30ms per DB call

Slide 55

Slide 55 text

3/20/15 ENSURING SUCCESS DURING DISASTER Trade latency to test correctness

Slide 56

Slide 56 text

3/20/15 ENSURING SUCCESS DURING DISASTER Correctness Capacity Execution

Slide 57

Slide 57 text

3/20/15 ENSURING SUCCESS DURING DISASTER DR is a mirror of production

Slide 58

Slide 58 text

3/20/15 ENSURING SUCCESS DURING DISASTER

Slide 59

Slide 59 text

3/20/15 ENSURING SUCCESS DURING DISASTER dr-compare.rb

Slide 60

Slide 60 text

3/20/15 ENSURING SUCCESS DURING DISASTER Checks that DR footprint is the same as production

Slide 61

Slide 61 text

3/20/15 ENSURING SUCCESS DURING DISASTER Surfaces missing roles

Slide 62

Slide 62 text

3/20/15 ENSURING SUCCESS DURING DISASTER Correctness Capacity Execution

Slide 63

Slide 63 text

3/20/15 ENSURING SUCCESS DURING DISASTER The flip

Slide 64

Slide 64 text

3/20/15 ENSURING SUCCESS DURING DISASTER Production impact

Slide 65

Slide 65 text

3/20/15 ENSURING SUCCESS DURING DISASTER Scale up staging

Slide 66

Slide 66 text

3/20/15 ENSURING SUCCESS DURING DISASTER Practice flip there

Slide 67

Slide 67 text

3/20/15 ENSURING SUCCESS DURING DISASTER Tested infrequently

Slide 68

Slide 68 text

3/20/15 ENSURING SUCCESS DURING DISASTER PagerDuty Active/Active

Slide 69

Slide 69 text

3/20/15 ENSURING SUCCESS DURING DISASTER Correctness Capacity Execution

Slide 70

Slide 70 text

3/20/15 ENSURING SUCCESS DURING DISASTER pagerduty.com/jobs [email protected]

Slide 71

Slide 71 text

3/20/15 [email protected] Questions?