Slide 1

Slide 1 text

Peter Baumgartner Just Enough Ops for Developers DjangoCon US 2022 @ipmb lincolnloop.com

Slide 2

Slide 2 text

About Me • Founder at Lincoln Loop — lincolnloop.com • Co-author of High Performance Django — highperformancedjango.com • Building AppPack — apppack.io

Slide 3

Slide 3 text

Prepping your project for production Watch my talk at DjangoCon 2019

Slide 4

Slide 4 text

Why just enough Ops?

Slide 5

Slide 5 text

PaaS & Managed Services are really good

Slide 6

Slide 6 text

Forget about 😴 • System security/hardening • Routing/networking • Secrets management • Deployments • Scaling how to, not when to • Process Management (systemd, docker, etc.) • Hardware failures

Slide 7

Slide 7 text

But! You need to understand the basics • CPU • RAM • I/O

Slide 8

Slide 8 text

No content

Slide 9

Slide 9 text

No content

Slide 10

Slide 10 text

CompSci 101

Slide 11

Slide 11 text

No content

Slide 12

Slide 12 text

Disk (HDD, SSD) Persistent File Storage

Slide 13

Slide 13 text

• Disk access is slow • Usually ephemeral Disk (HDD, SSD) Persistent File Storage

Slide 14

Slide 14 text

CPU (processor) Code execution

Slide 15

Slide 15 text

How many CPUs do I need?

Slide 16

Slide 16 text

How many CPUs do I need? It depends

Slide 17

Slide 17 text

1 request/1 process/1 CPU

Slide 18

Slide 18 text

https://fastapi.tiangolo.com/async/ Cook = CPU Cashier = Python Process Customer = Request

Slide 19

Slide 19 text

How many CPUs do I need? It depends

Slide 20

Slide 20 text

How many cooks do I need? It depends

Slide 21

Slide 21 text

More Users → More CPU Rule of 👍

Slide 22

Slide 22 text

2 requests/1 process/1 CPU

Slide 23

Slide 23 text

2 requests/2 process/2 CPU

Slide 24

Slide 24 text

💸

Slide 25

Slide 25 text

How do I minimize cost?

Slide 26

Slide 26 text

How do I maximize CPU usage?

Slide 27

Slide 27 text

Multiple Processes per CPU How do I maximize CPU usage?

Slide 28

Slide 28 text

2 requests/2 processes/1 CPU

Slide 29

Slide 29 text

2 requests/2 processes/1 CPU

Slide 30

Slide 30 text

2 requests/2 processes/1 CPU

Slide 31

Slide 31 text

gunicorn \ --workers=3 \ …

Slide 32

Slide 32 text

Tuning Worker Count Each app is different • Start at double your CPU count • You can experiment with adding workers until • Not enough memory • Response times plateau or degrade

Slide 33

Slide 33 text

Improve Application Performance How do I maximize CPU usage?

Slide 34

Slide 34 text

1 CPU / 1 minute 100ms response time = 6000 requests 1s response time = 60 requests

Slide 35

Slide 35 text

⏳Slow responses ⏱More CPU time 💰Higher cost

Slide 36

Slide 36 text

I/O CompSci 101

Slide 37

Slide 37 text

I/O (Input/Output) Examples • Reading/writing files • Database queries • Object storage (S3) • Search index queries • Third-party APIs

Slide 38

Slide 38 text

Everything waits for input/output

Slide 39

Slide 39 text

No content

Slide 40

Slide 40 text

Synchronous I/O is blocking 🤓

Slide 41

Slide 41 text

🤓 Asynchronous I/O is non-blocking

Slide 42

Slide 42 text

Async Django Views since Django 3.1 / ORM since Django 4.1

Slide 43

Slide 43 text

Asynchronous programming is hard 🤯

Slide 44

Slide 44 text

Jeff Atwood “Hardware is cheap, Programmers are expensive

Slide 45

Slide 45 text

gevent How do I maximize CPU usage?

Slide 46

Slide 46 text

🪄 magic How do I maximize CPU usage?

Slide 47

Slide 47 text

2 workers with gevent

Slide 48

Slide 48 text

gunicorn \ --worker-class=gevent \ --worker-connections=50 \ …

Slide 49

Slide 49 text

Not Enough CPU

Slide 50

Slide 50 text

No content

Slide 51

Slide 51 text

Backlog

Slide 52

Slide 52 text

Error Codes • 502 Bad Gateway • 503 Service Unavailable • 504 Gateway Timeout

Slide 53

Slide 53 text

Memory (RAM) Ephemeral cache

Slide 54

Slide 54 text

• “Warm” code cache • Python objects • File processing/manipulation • Network sockets • File descriptors Memory (RAM) Ephemeral cache

Slide 55

Slide 55 text

• Fast • Limited Memory (RAM) Ephemeral cache

Slide 56

Slide 56 text

Memory usage for a typical Django app

Slide 57

Slide 57 text

128 - 512+ MB Memory usage for a typical Django app

Slide 58

Slide 58 text

Memory usage for a typical Django app 128 - 512+ MB per process

Slide 59

Slide 59 text

👷 4 gunicorn workers 📊 512 MB per process 🟰 2 GB of memory

Slide 60

Slide 60 text

Over 100% memory → ☠

Slide 61

Slide 61 text

Not enough memory? • ⬆ Increase allowed memory • ⬇ Reduce processes/workers

Slide 62

Slide 62 text

Memory usage should be stable

Slide 63

Slide 63 text

Stable Memory Usage

Slide 64

Slide 64 text

Variable (but Stable) Memory Usage

Slide 65

Slide 65 text

Django memory tips • Don’t read huge files into a string/byte object • Don’t process a huge queryset • Use Model.objects.iterator() • Use .values() to avoid creating a model instance • Use .only() to avoid loading large text fields

Slide 66

Slide 66 text

Memory Leak

Slide 67

Slide 67 text

Memory Leak Causes • Bug in C extension (no garbage collection) • Leaving file descriptors or network sockets open (use context managers) • Global objects (sometimes accidental)

Slide 68

Slide 68 text

gunicorn \ --max-requests=10000 \ --max-requests-jitter=500 \ …

Slide 69

Slide 69 text

Databases

Slide 70

Slide 70 text

Same rules apply

Slide 71

Slide 71 text

Remember: disk is slow 🌋 Database should fit in RAM Rule of 👍

Slide 72

Slide 72 text

Scaling

Slide 73

Slide 73 text

No content

Slide 74

Slide 74 text

Horizontal Vertical

Slide 75

Slide 75 text

Autoscaling is great 🎉

Slide 76

Slide 76 text

Autoscaling is great 🎉

Slide 77

Slide 77 text

…but it’s not magic 🪄

Slide 78

Slide 78 text

Serverless

Slide 79

Slide 79 text

1 request/1 process/1 CPU Serverless

Slide 80

Slide 80 text

Pricing Per request + Allocated resources per millisecond of response time Serverless

Slide 81

Slide 81 text

• Scaling • CPU Allocation • Application server (workers, gevent, etc.) Forget about 😴 Serverless

Slide 82

Slide 82 text

Worry about 😥 • Budgeting/variable costs • Cold starts • Database connections • Limitations (upload size, max duration) • Remote shell access Serverless

Slide 83

Slide 83 text

Final Thoughts 🧐 • Get to know your application — “observability” • CPU usage • Memory usage • Response times • Error rates/uptime

Slide 84

Slide 84 text

Thanks! 👋 @ipmb lincolnloop.com apppack.io