Upgrade to Pro
— share decks privately, control downloads, hide ads and more …
Speaker Deck
Features
Speaker Deck
PRO
Sign in
Sign up for free
Search
Search
Formal Specifications and Other People's Tech
Search
Sponsored
·
SiteGround - Reliable hosting with speed, security, and support you can count on.
→
Marianne Bellotti
September 13, 2019
Technology
160
0
Share
Embed
Copy iframe code
Copy JS code
Copy link
Start on current slide
Formal Specifications and Other People's Tech
How do you use formal specification like TLA+ to model real world engineering challenges?
Marianne Bellotti
September 13, 2019
More Decks by Marianne Bellotti
See All by Marianne Bellotti
Building Safety Critical Systems
mbellotti
0
200
Killer Robots and Rogue AI
mbellotti
0
34
Five Lies of Modernization
mbellotti
0
42
Interdisciplinary Engineering
mbellotti
0
190
Worst Case Scenario in the Database
mbellotti
0
110
Death of the Trusted Internet
mbellotti
0
150
Debunking Other People's Data Science
mbellotti
0
140
Other Decks in Technology
See All in Technology
ゴールデンパスは敷いただけでは道にならない ─ 企画部門のエンジニアが技術標準を事業価値に変えるまで
mhrtech
1
230
タスクの複雑さでモデルを選ぶ ── Thompson Samplingで動かす“トークン/コスト最適化
satohy0323
0
570
人を動かすのは時間ではなく、納得感 〜新任EMが入社3ヶ月、組織を2回変えた話〜
kakehashi
PRO
3
270
AIと共生する開発者プラットフォーム:バクラクのモノレポ×マイクロサービス基盤
sakajunquality
2
3.9k
型は壁、Rustでもバグを直すな、表現できなくせよ
nwiizo
14
2.1k
[2026-07-15] AI Ready なはずだったアーキテクチャと、見えてきた課題・次に目指す状態
wxyzzz
9
4k
ソニー銀行におけるビジネスアジリティ向上のためのクラウドシフト戦略
srenext
0
820
“それは自分の仕事じゃない"を越えて行け
yuukiyo
1
480
「最後に責任を取るのはチーム」— 人間のPRレビューを最小化してアップデートしたメンタルモデル
jnishime_dresscode
0
920
「AIに依存している」と 「AIを使いこなしている」の違い
k8yasuma
0
120
実践!既存 Project への AI-Driven Development 適用〜 一ヶ月で Project 唯一のフロントエンドエンジニアを作り出せ〜
lycorptech_jp
PRO
0
240
Control Planeで育てるBtoB SaaSの認証基盤 - SRE NEXT 2026
pokohide
1
2.6k
Featured
See All Featured
Google's AI Overviews - The New Search
badams
0
1.1k
Put a Button on it: Removing Barriers to Going Fast.
kastner
60
4.4k
Context Engineering - Making Every Token Count
addyosmani
9
1k
How Fast Is Fast Enough? [PerfNow 2025]
tammyeverts
3
660
Bridging the Design Gap: How Collaborative Modelling removes blockers to flow between stakeholders and teams @FastFlow conf
baasie
0
610
Chrome DevTools: State of the Union 2024 - Debugging React & Beyond
addyosmani
10
1.3k
Producing Creativity
orderedlist
PRO
348
40k
Writing Fast Ruby
sferik
630
63k
The Illustrated Guide to Node.js - THAT Conference 2024
reverentgeek
1
410
Taking LLMs out of the black box: A practical guide to human-in-the-loop distillation
inesmontani
PRO
3
2.3k
Primal Persuasion: How to Engage the Brain for Learning That Lasts
tmiket
0
390
svc-hook: hooking system calls on ARM64 by binary rewriting
retrage
2
340
Transcript
auth0.com Formal Specification Taming Other People's Tech &
SPECS! BY BEGINNERS FOR BEGINNERS @bellmar bit.ly/bellmar-auth0
TLA+ Conf Finding Bugs Without Running or Even Looking at
Code Correctness Proofs of Distributed Systems with Isabelle
WHAT IS FORMAL SPECIFICATION? @bellmar bit.ly/bellmar-auth0
-Inputs -Expected behavior -Legal outputs -Illegal behavior
Tell me the future, wise computer @bellmar bit.ly/bellmar-auth0
@bellmar bit.ly/bellmar-auth0
!! @bellmar bit.ly/bellmar-auth0
@bellmar bit.ly/bellmar-auth0 Mostly I write bad specs, and this talk
is about how that still adds a lot of value The more people write specs the faster we all learn
S3 Design Session @bellmar bit.ly/bellmar-auth0
MOST ENGINEERS WILL NEVER BUILD A MESSAGE QUEUE You’ll buy
one from AWS
BUT SPECIFICATION CAN BE APPLIED TO MANY DIFFERENT THINGS! @bellmar
bit.ly/bellmar-auth0
@bellmar bit.ly/bellmar-auth0 Identity as a Service
@bellmar bit.ly/bellmar-auth0
Built by other teams
In taking over a service: - Defining SLOs - Evaluating
debt - Studying existing problems and rethinking architecture
EMAILS - validate email - password reset - principal mode
of communication We do logins not emails!
EMAILS - validate email - password reset - principal mode
of communication @bellmar bit.ly/bellmar-auth0
X per hour @bellmar bit.ly/bellmar-auth0
@bellmar bit.ly/bellmar-auth0
@bellmar bit.ly/bellmar-auth0
Email all my users! Password Reset Anomaly Detection Welcome Email
Validation @bellmar bit.ly/bellmar-auth0
@bellmar bit.ly/bellmar-auth0
HOW DO WE KNOW IF THIS CHANGE IS SAFE? @bellmar
bit.ly/bellmar-auth0
LET’S WRITE A SPEC! @bellmar bit.ly/bellmar-auth0
STEPS WITH SPECIFICATIONS YOU THINK IN STATES WITH ALGORITHMS YOU
THINK IN @bellmar bit.ly/bellmar-auth0
@bellmar bit.ly/bellmar-auth0
@bellmar bit.ly/bellmar-auth0
\ 40% → ✅ 41% → ✅ 42% → ✅
43% → ✅ 44% → ✅ 45% → ✅ 46% → ✅ 47% → ✅ 48% → ✅ 49% → ✅ 50% → ✅
VM 50% VM 20% VM 70%
VM 50% VM 20% VM 70% Test all possible states
to make sure the load balancer does the right thing
40% → ✅ 41% → ✅ 42% → ✅ 43%
→ ✅ 44% → ✅ 45% → ✅ 46% → ✅ 47% → ✅ 48% → ✅ 49% → ✅ 50% → ✅ 40% → ✅ 50% → ✅ 60% → ✅ 70% → ❌ ✅ ❌
Normal Idle Unhealthy @bellmar bit.ly/bellmar-auth0
Must always be true: 0 < servers > 10 Goto!
Party like it’s 1979…
LEAVE YOUR PROBABILITY AT HOME
Sometimes we kill a server, sometimes we create one Literally
the one time we just keep killing servers @bellmar bit.ly/bellmar-auth0
Sometimes we kill a server, sometimes we create one Literally
the one time we just keep killing servers ARGH! BUT WHY? THAT WILL NEVER HAPPEN!!
SYSTEM NOT SYSTEM LGTM SAFETY @bellmar bit.ly/bellmar-auth0
Normal Idle Unhealthy Health Check @bellmar bit.ly/bellmar-auth0
- Several unhealthy servers == Create new - All servers
healthy and more than 3 == Shut one down - Else assume we kill and refresh unhealthy
- Several unhealthy servers == Create new - All servers
healthy and more than 3 == Shut one down - Else assume we kill and refresh unhealthy
WHAT STATES SHOULD BE IMPOSSIBLE? (NOT UNDESIRABLE?) @bellmar bit.ly/bellmar-auth0
Servers keep restarting Too many servers Too few servers Healthcheck
action inconsistency @bellmar bit.ly/bellmar-auth0
Servers keep restarting Too many servers Too few servers Healthcheck
action inconsistency Undesirable but not impossible states @bellmar bit.ly/bellmar-auth0
Maybe you should turn yourself rather than the puzzle
Anyway… @bellmar bit.ly/bellmar-auth0
x per hour @bellmar bit.ly/bellmar-auth0
One please! @bellmar bit.ly/bellmar-auth0
One please! Sorry no more tokens. Sit in the retry
queue and come back later @bellmar bit.ly/bellmar-auth0
One please! Sorry no more tokens. Sit in the retry
queue and come back later @bellmar bit.ly/bellmar-auth0
STATE HOW DO WE MAKE THIS LOOK LIKE A MACHINE
@bellmar bit.ly/bellmar-auth0
Queued → Worker → Sent Has Tokens → No Tokens
@bellmar bit.ly/bellmar-auth0
Queued Worker Sent @bellmar bit.ly/bellmar-auth0
Queued Worker Sent Dropped @bellmar bit.ly/bellmar-auth0
WHAT’S STATES SHOULD BE IMPOSSIBLE? @bellmar bit.ly/bellmar-auth0
Has Token → Deleted Sent → Deleted Deleted → Sent
No Token → Sent @bellmar bit.ly/bellmar-auth0
HOW DOES OUR IMPLEMENTATION PREVENT THESE STATES? @bellmar bit.ly/bellmar-auth0
JUST TRYING TO BUILD THE MODEL OFTEN REVEALS BUGS @bellmar
bit.ly/bellmar-auth0
Email all my users! Password Reset Anomaly Detection Welcome Email
Validation @bellmar bit.ly/bellmar-auth0
Email all my users! Password Reset Anomaly Detection Welcome Email
Validation @bellmar bit.ly/bellmar-auth0
Queued Worker @bellmar bit.ly/bellmar-auth0
GREAT, SO WHAT SHOULD OUR MODEL LOOK LIKE? @bellmar bit.ly/bellmar-auth0
Tenant Tenant Tenant @bellmar bit.ly/bellmar-auth0
Tenant @bellmar bit.ly/bellmar-auth0 For each tenant For each bucket
LOOPS WILL F— YOU UP LOOPS @bellmar bit.ly/bellmar-auth0
x in [a, b, c] print x “a” “b” “c”
@bellmar bit.ly/bellmar-auth0
x in [a, b, c] l=append(l, x) l = [
] print l @bellmar bit.ly/bellmar-auth0
x in [a, b, c] l=append(l, x) [“a”] [“a”, “b”]
[“a”, “b”, “c”] l = [ ] print l @bellmar bit.ly/bellmar-auth0
x in [a, b, c] l = [ ] [“a”]
[“b”] [“c”] l=append(l, x) print l @bellmar bit.ly/bellmar-auth0
A B C then then A B C or or
Steps States @bellmar bit.ly/bellmar-auth0
Tenant Tenant Tenant @bellmar bit.ly/bellmar-auth0
Tenant Tenant Tenant @bellmar bit.ly/bellmar-auth0
@bellmar bit.ly/bellmar-auth0
Alice Bob Oh… that’s easy!
Undesirable -vs- Impossible
Queue Publishers
“SHOULD THE PUBLISHERS BE PUTTING THINGS DIRECTLY IN THE QUEUE?”
@bellmar bit.ly/bellmar-auth0 WAIT…
1) Make it look like a state machine 2) Define
the impossible states 3) Model what makes them impossible To Summarize: @bellmar bit.ly/bellmar-auth0
Things that already look like state machines: @bellmar bit.ly/bellmar-auth0
Things that already look like state machines: - Redux @bellmar
bit.ly/bellmar-auth0
Things that already look like state machines: - Redux -
AWS Lambda @bellmar bit.ly/bellmar-auth0
Things that already look like state machines: - Redux -
AWS Lambda - Workflow Orchestration (Airflow, Conductor, Luigi) @bellmar bit.ly/bellmar-auth0
Bit.ly/bellmar-auth0