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
Lessons from 6 Months of using Luigi
Search
peteowlett
May 07, 2016
Technology
4
910
Lessons from 6 Months of using Luigi
AKA Why it's better to be woken up by you cat than by the server alarm
peteowlett
May 07, 2016
Tweet
Share
More Decks by peteowlett
See All by peteowlett
Takeaway Tales
peteowlett
1
200
Send More Riders
peteowlett
4
930
Other Decks in Technology
See All in Technology
モバイルゲームの開発を支える基盤の歩み ~再現性のある開発ラインを量産する秘訣~
qualiarts
0
600
Shadow DOMとセキュリティ - 光と影の境界を探る / Shibuya.XSS techtalk #13
masatokinugawa
0
310
AIエージェントを支える設計
tkikuchi1002
11
2.3k
分散トレーシングによる コネクティッドカーのデータ処理見える化の試み
thatsdone
0
270
「育てる」サーバーレス 〜チーム開発研修で学んだ、小さく始めて大きく拡張するAWS設計〜
yu_kod
1
170
興味の胞子を育て 業務と技術に広がる”きのこ力”
fumiyasac0921
0
290
東京海上日動におけるセキュアな開発プロセスの取り組み
miyabit
0
200
2025-07-25 NOT A HOTEL TECH TALK ━ スマートホーム開発の最前線 ━ SOFTWARE
wakinchan
0
170
新規事業におけるAIリサーチの活用例
ranxxx
0
170
増え続ける脆弱性に立ち向かう: 事前対策と優先度づけによる 持続可能な脆弱性管理 / Confronting the Rise of Vulnerabilities: Sustainable Management Through Proactive Measures and Prioritization
nttcom
1
210
With Devin -AIの自律とメンバーの自立
kotanin0
2
780
機械学習を「社会実装」するということ 2025年夏版 / Social Implementation of Machine Learning July 2025 Version
moepy_stats
1
1.4k
Featured
See All Featured
How STYLIGHT went responsive
nonsquared
100
5.7k
Principles of Awesome APIs and How to Build Them.
keavy
126
17k
Automating Front-end Workflow
addyosmani
1370
200k
Why Our Code Smells
bkeepers
PRO
337
57k
Designing Dashboards & Data Visualisations in Web Apps
destraynor
231
53k
I Don’t Have Time: Getting Over the Fear to Launch Your Podcast
jcasabona
33
2.4k
The World Runs on Bad Software
bkeepers
PRO
70
11k
Improving Core Web Vitals using Speculation Rules API
sergeychernyshev
18
1k
The Psychology of Web Performance [Beyond Tellerrand 2023]
tammyeverts
48
2.9k
Code Reviewing Like a Champion
maltzj
524
40k
Visualization
eitanlees
146
16k
Practical Tips for Bootstrapping Information Extraction Pipelines
honnibal
PRO
21
1.4k
Transcript
Lessons from 6 months of using Luigi in production @peterowlett
@deliveroo
Hello! I’m Pete
None
I work for these folks
WE DO THIS
Why it’s better to be woken up by your cat
than by the server alarm A BETTER TITLE
This is Kitty
NO RESPECT FOR PERSONAL SPACE
This is PagerDuty
Even less respect for personal space
Let’s Compare! - Goes off at any time, day or
night - Loud ring tone, text messages, answer phone messages and flashing - Resolution can take hours - Goes off only once at precisely 6am - Cute batting motion to wake - Resolved in time it takes to open cat food packet
I think we can all agree with my premise Kitty
>> PagerDuty
Lets get to it
Chapter 1 “The Model”
Let’s build a model
I’m ready, where’s the data?
“Just pg_dump the prod db”
OH PLS PLS NO DON’T DO THAT
Lets spin up a read slave and ETL the data
to a warehouse …
… then train our models from that
How do we ensure tasks run in Order?
I want them to run one after the other 2
1 3 4 5 6 7 8
Directed Acyclic Graph
None
Enter Stage Left …
Simple Task
Postgres Loader Task
We string these together to make DAGs CHECK MAX ROW
ID LOAD DATA MOD DATA MAKE MODEL CHECK MAX ROW ID LOAD DATA TABLE1 TABLE2
DAGs solve the dependency problem
Bung it all on EC2
Define an entry point
Run the scheduler
Kick it all off with CRON
With luigi we were up and running in a few
hours
Chapter 2 “The Nuclear Option”
A few weeks later, something happened …
None
Schema can change anytime without warning HAS THE SCHEMA CHANGED?
RELOAD JUST NEW ROWS DROP AND CREATE WHOLE SCHEMA RELOAD ALL TABLES NO! YES!
None
Handle schema changes robustly
Let’s test our pipeline before we deploy it. But how?
Two new operating modes TEST MODE Run the whole pipeline
but only write to a test schema UNIT MODE Run the current task, ignoring its dependencies
Configure these modes in the pipeline using luigi.Parameter
Now nothing will ever go wrong, ever again …
Make your testing comprehensive
Make your testing fast
Adding in external API services
Build Loaders for each API
Loading Schedules
Plumbing them in
Keep def rows as short as possible
Be consistent in loader design pattern
Expect external API services to misbehave
Expect external API services to misbehave X
Trust external API services as if they actively want to
hurt you
Hey cool, all our data is in one place, we
might as well use it for BI Reporting
Chapter 3 “The Management Report”
This happened
And then your ad hoc database is now supporting global
business critical apps
None
Stuff that was happening • Irrelevant upstream failures • Low
priority upstream failures • Flakey Data (but it worked!)
Our DAG looked like this: START LOAD2 LOAD1 LOAD3 LOAD
DONE MAKE1 MAKE2 MAKE3 END
And this was happening START LOAD2 LOAD1 LOAD3 LOAD DONE
MAKE1 MAKE2 MAKE3 END
This one doesn’t need LOAD3 START LOAD2 LOAD1 LOAD3 LOAD
DONE MAKE1 MAKE2 MAKE3 END
So we changed it to this START LOAD2 LOAD1 LOAD3
LOAD ALL MAKE1 MAKE2 MAKE3 END
Now when 3 fails: START LOAD2 LOAD1 LOAD3 LOAD ALL
MAKE1 MAKE2 MAKE3 END
Decide about what can be allowed to fail, and what
can’t
Isolate the path to critical ops jobs
Loading tables more reliably 5AM SATURDAY
The currency table failed to update
Loading tables more reliably DROP TABLE CREATE TABLE LOAD DATA
Task 1 Task 2 Task 3
Loading tables more reliably DROP TABLE CREATE TABLE THIS CAN
GO WRONG LOAD DATA Task 1 Task 2 Task 3 THIS CAN GO WRONG THIS CAN GO WRONG
Expect Failure, Rollback Transaction CREATE TEMP TABLE Task 1 (There
is no task 2) LOAD DATA RENAME OLD TABLE RENAME NEW TABLE ROLLBACK
None
Encapsulate logic in bigger chunks
None
Anticipating problems early
Going beyond system monitoring
Defined Monitoring Tests
Measuring outcomes directly
And get gentler alerts in slack
Monitor (and alert on) outcomes as well as system metrics
Try / Except / Slack Alert low priority tasks
Chapter 4 “Hi, this is Australia calling …”
In the beginning there was the UK YAY! DOWNTIME!! Midnight
Midnight Midday UK Ops
Then Europe Ok cool still loads of downtime Midnight Midnight
Midday
Then Some Other Places No such thing as downtime anymore
Midnight Midnight Midday
Table Loading - Take 2 HASH THE TABLE SCHEMA COMPARE
TO LAST HASH SAME! CHANGED! DROP AND REBUILD JUST LOAD ROWS
Get rid of the nuclear option
SORRY RIPLEY
Chapter 5 “Moving to Scale”
When it comes to BI, Old School Rules Still Apply
Configuration Management (Docker + ECS)
Distributed workers make some pain go away
Protobuf3 on Message Bus
Final Thoughts
I regret nothing!
Everything is defined in code
Two people, tiny budget
Time spent speeding up the build process is time well
spent
Think carefully about what dependencies *mean*
To finish …
None
We’re hiring! Grab me after :) https://roo.it/peteo Also £5 off
your first order!
Sleep Well! @peterowlett @deliveroo Sleep Well!