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
[PyCon JP 2019] 新米Pythonistaが贈るAirflow入門&活用事例紹介
Search
Naoki Matsuda
September 17, 2019
Technology
7k
2
Share
Embed
Copy iframe code
Copy JS code
Copy link
Start on current slide
[PyCon JP 2019] 新米Pythonistaが贈るAirflow入門&活用事例紹介
PyCon JP 2019の発表資料です。
Naoki Matsuda
September 17, 2019
More Decks by Naoki Matsuda
See All by Naoki Matsuda
Tech x Marketing #4 Airflowでもサブワークフロー単位で分割開発したい!
matsudan
0
220
Other Decks in Technology
See All in Technology
Multicaで30個のミニプロジェクトをAIエージェント運用して見えてきたこと
eiei114
0
550
CDKで書くECSのベストプラクティス、 改めて考え直す2026 #cdkconf2026
makies
3
880
AIレビューはどこまで任せられるのか?自動化と人が背負うレビューの境界
sansantech
PRO
3
1.2k
生成AIワークショップ / Custom-AI-Agents-workshop
databricksjapan
0
110
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
240
“それは自分の仕事じゃない"を越えて行け
yuukiyo
1
490
壊して学ぶAWS CDK: そのcdk deployで消えるもの、残るもの
k_adachi_01
1
440
Alphaモジュール使っていいのかい!?いけないのかい!?どっちなんだいっ!?
watany
1
300
AIと1000本ノックしてたどり着いた、最速のプロダクト開発 ~toC向けAIエージェントUXを、動く選択肢とAIキャパシティで設計する~
lycorptech_jp
PRO
1
100
複数プロダクト組織のAIネイティブ化における戦略 / AICon2026_kude
rakus_dev
0
260
発表と総括 / Presentations and Summary
ks91
PRO
0
150
AIコード生成×サプライチェーン攻撃 — PHPが直面する“二重の信頼問題
shinyasaita
0
340
Featured
See All Featured
How to train your dragon (web standard)
notwaldorf
97
6.7k
The Psychology of Web Performance [Beyond Tellerrand 2023]
tammyeverts
49
3.5k
Avoiding the “Bad Training, Faster” Trap in the Age of AI
tmiket
0
190
RailsConf & Balkan Ruby 2019: The Past, Present, and Future of Rails at GitHub
eileencodes
141
35k
Intergalactic Javascript Robots from Outer Space
tanoku
273
27k
Gemini Prompt Engineering: Practical Techniques for Tangible AI Outcomes
mfonobong
2
460
A Guide to Academic Writing Using Generative AI - A Workshop
ks91
PRO
1
350
The Mindset for Success: Future Career Progression
greggifford
PRO
0
430
The browser strikes back
jonoalderson
0
1.4k
[RailsConf 2023 Opening Keynote] The Magic of Rails
eileencodes
31
10k
Building a Scalable Design System with Sketch
lauravandoore
463
34k
Money Talks: Using Revenue to Get Sh*t Done
nikkihalliwell
0
400
Transcript
৽ถPythonista͕ଃΔAirflowೖ & ׆༻ࣄྫհ PyCon JP 2019 2019.9.17 Naoki Matsuda
Agenda 0. ࣗݾհ 1. Airflowͷ֓ཁ 2. Airflowͷࣾࣄྫհ - ։ൃϓϩμΫτ֓ཁͱ՝, Airflowͷڥ
3. AirflowͰͭ·͍ͮͨ - λεΫؒͷσʔλͷΓͱΓ - DAGಈ࡞֬ೝ ~ dockerͰϩʔΧϧ։ൃڥߏங
ࣗݾհ দా थ (·ͭͩ ͳ͓͖) - ॴଐɿגࣜձࣾ ి௨σδλϧ - ۀɿόοΫΤϯυαʔϏεɺETLपΓͷ։ൃ
- 2018ೖࣾ
1. Airflowͷ֓ཁ
Apache Airflow֓ཁ όονॲཧ͔ΒͳΔϫʔΫϑϩʔͷεέδϡʔϦϯάˍϞχλ Ϧϯά͕ՄೳͳϓϥοτϑΥʔϜ - Airbnbࣾ - Φʔϓϯιʔε (Apache software
foundationͷincubation project)1,2 - PythonͰ࣮͞Ε͍ͯΔ3 - ։ൃίϛϡχςΟ͕׆ൃ3 IUUQTJODVCBUPSBQBDIFPSHQSPKFDUTBJSGMPXIUNM IUUQTBJSGMPXBQBDIFPSHMJDFOTFIUNM IUUQTHJUIVCDPNBQBDIFBJSGMPX
։ൃίϛϡχςΟͷ׆ൃ͞(2019.9.9࣌)
Apache AirflowͰͰ͖Δ͜ͱ - PythonίʔυͰϫʔΫϑϩʔ(DAG)Λఆٛ - ґଘؔʹج͍ͮͨλεΫͷ࣮ߦ - ϫʔΫϑϩʔͷεέδϡʔϦϯά - ϦονͳWeb
UI - DAG࣮ߦεςʔλεͷϞχλϦϯά - λεΫͷϩά֬ೝ - ґଘؔͷՄࢹԽ ͳͲ
PythonίʔυͰϫʔΫϑϩʔఆٛ(DAGͷ࡞) λεΫؒґଘؔͷఆٛ λεΫ1 λεΫ2 DAGͷڞ௨ઃఆ ࣮ߦස, ࣮ߦظؒ, λΠϜΞτ࣌ؒͳͲ
ϫʔΫϑϩʔΛߏ͢ΔλεΫͷ࡞ ≈ - ϫʔΫϑϩʔOperatorͱݺΕΔλεΫʹΑΓߏ͞ΕΔ1 - 1ͭͷOperatorͰ1ͭͷλεΫΛهड़ OperatorͷҾɻ ֤Operator͕ԿͷҾΛ ͱΔ͔υΩϡϝϯτࢀর2 IUUQTBJSGMPXBQBDIFPSHDPODFQUTIUNMPQFSBUPST
IUUQTBJSGMPXBQBDIFPSH@BQJBJSGMPXPQFSBUPSTJOEFYIUNM
ϫʔΫϑϩʔΛߏ͢ΔλεΫͷ࡞ - BashίϚϯυ࣮ߦ: BashOperator - Python࣮ؔߦ: PythonOperator - SQL࣮ߦ: MySqlOperator,
PostgresOperator, … - HTTPϦΫΤετૹ৴: SimpleHttpOperator - ͦͷଞΫϥυܥͳͲ: BigQueryOperator, AWSAthenaOperator, … - ಛఆ݅Ληϯγϯά: Sensor IUUQTBJSGMPXBQBDIFPSHDPODFQUTIUNMPQFSBUPST IUUQTBJSGMPXBQBDIFPSH@BQJBJSGMPXPQFSBUPSTJOEFYIUNM
2. ࣾࣄྫհ - ։ൃϓϩμΫτ֓ཁͱ՝, Airflowͷߏ
։ൃ৫ͱϓϩμΫτʹ͍ͭͯ ɾɾɾ ࠂ৴ σʔλ ϓϥϯφʔ Ӧۀ - ڈ7~9݄, - GoogleDispla
y - ҿྉۀք imp 100000 clicks 5000 cv 700 cost ɾɾɾ ϝσΟΞ ։ൃ৫ νʔϜنɿσʔλΤϯδχΞ ໊ σʔλαΠΤϯςΟετͳͲ໊ d ϓϩμΫτ ͚ࣾσδλϧࠂϓϥϯχϯάπʔϧ ࣾͰѻ͏ϝσΟΞɾΫϥΠΞϯτͷࠂ৴࣮σʔλΛՄࢹԽˍ༧ଌ ϓϩμΫτ (PPHMF :BIPP 5XJUUFS 'BDFCPPL -*/&
։ൃϓϩμΫτʹ͓͚Δ՝ - σʔλ͕RDBʹೖͬͯͳ͍ ৴Ϩϙʔτσʔλ͕ੳ༻ͷྻࢤσʔλϕʔεʹ ͋ͬͨΓɺϚελσʔλ͕εϓϨουγʔτʹ͋ͬͨΓ… - ඞཁͳใΛՃ͢ΔͨΊʹଟ͘ͷϦϨʔγϣϯΛͨͲΔ
։ൃϓϩμΫτʹ͓͚Δ՝ - σʔλ͕RDBʹೖͬͯͳ͍ ৴Ϩϙʔτσʔλ͕ੳ༻ͷྻࢤσʔλϕʔεʹ ͋ͬͨΓɺϚελσʔλ͕εϓϨουγʔτʹ͋ͬͨΓ… - ඞཁͳใΛՃ͢ΔͨΊʹଟ͘ͷϦϨʔγϣϯΛͨͲΔ → ։ൃϓϩμΫτ༻ʹσʔλϚʔτ࡞ RDBʹϑΝΫτ,
σΟϝϯγϣϯςʔϒϧΛETLͰ࡞
Airflowߏ apache-airflow 1.10.2 web worker scheduler Amazon S3 Amazon RDS
Airflow AWS Fargate Amazon ElastiCache Redis Elastic Load Balancing flower DAGs - AWS FargateʹAirflowΛσϓϩΠ ≈ ≈
docker-airflow https://github.com/puckel/docker-airflow
ߏஙͨ͠σʔλϑϩʔ ֤ϝσΟΞࠂ৴σʔλ ϦϨʔγϣϯςʔϒϧܥ JOIN ΫϥΠΞϯτใܥ ʜ ΧϥϜ໊ دͤͳͲ ≈ Amazon
Athena Backend service INSERT "JSGMPX͕࣮ߦ͢ΔλεΫ INSERT
3. AirflowͰͭ·͍ͮͨ - λεΫؒͷσʔλͷΓͱΓ - DAGಈ࡞֬ೝ ~dockerͰϩʔΧϧ։ൃڥߏங
λεΫؒͷσʔλͷΓͱΓ λεΫؒͷσʔλΓͱΓXComΛ͏ - XComͷ͍ํ - XComσʔλΛpush - ؔͰreturn - ؔͰkwargs['task_instance’].
xcom_push(value=hoge, key=‘huga’) - Λฦ͢Operator ྫ: BigqueryGetDataOperator - XCom͔ΒσʔλΛpull - kwargs['task_instance’].xcom_pull() metadata database
λεΫؒͷσʔλͷΓͱΓ BigQuery͔ΒςʔϒϧσʔλΛऔಘͯͦ͠ͷσʔλΛՃ͢Δྫ
λεΫؒͷσʔλͷΓͱΓ BQςʔϒϧ͔Βσʔλऔಘ # XComʹpush͞ΕΔ BigQuery͔ΒςʔϒϧσʔλΛऔಘͯͦ͠ͷσʔλΛՃ͢Δྫ
λεΫؒͷσʔλͷΓͱΓ BQςʔϒϧ͔Βσʔλऔಘ # XComʹpush͞ΕΔ transpose_dataؔΛ࣮ߦ BigQuery͔ΒςʔϒϧσʔλΛऔಘͯͦ͠ͷσʔλΛՃ͢Δྫ
λεΫؒͷσʔλͷΓͱΓ 1. task1Ͱpush͞ΕͨXcomͷ σʔλΛpullͯ͠ 2. ςʔϒϧͷσʔλΛసஔ BQςʔϒϧ͔Βσʔλऔಘ # XComʹpush͞ΕΔ transpose_dataؔΛ࣮ߦ
BigQuery͔ΒςʔϒϧσʔλΛऔಘͯͦ͠ͷσʔλΛՃ͢Δྫ
λεΫؒͷσʔλͷΓͱΓ provide_contextΛTrueʹ͠ͳ͍ͱkwargs[‘task_intance’]ͰKeyError - provide_context=False (default) kwargs : {} - provide_context=True
kwargs: { 'dag': <DAG: sample>, 'ds': '2019-09-10’, 'next_ds': '2019-09-10’, … 'task_instance’: <TaskInstance: sample.task1_2 …> … }
λεΫؒͷσʔλͷΓͱΓ - PythonOperatorͷҾͰTrue OR - default_argsͰઃఆ
DAGಈ࡞֬ೝ ~ϩʔΧϧ։ൃڥߏங - ࡞ͨ͠ϫʔΫϑϩʔ(DAG)ͷςετͲ͏Δʁ എܠɿ - Ϋϥυ্devڥͰͷDAGಈ࡞֬ೝͰS3upload͢Δखؒ - ଞͷਓ͕ಉ͡λΠϛϯάͰ։ൃ͍ͯ͠ΔͱΓͮΒ͍… →
ϩʔΧϧͰDAGͷಈ࡞֬ೝ͍ͨ͠ʂ
DAGಈ࡞֬ೝ ~ϩʔΧϧ։ൃڥߏங → dockerͰAirflowΛϩʔΧϧʹ্ཱͪ͛Δ - ࡞ͨ͠ϫʔΫϑϩʔ(DAG)ͷςετͲ͏Δʁ എܠɿ - Ϋϥυ্devڥͰͷDAGಈ࡞֬ೝͰS3upload͢Δखؒ -
ଞͷਓ͕ಉ͡λΠϛϯάͰ։ൃ͍ͯ͠ΔͱΓͮΒ͍… → ϩʔΧϧͰDAGͷಈ࡞֬ೝ͍ͨ͠ʂ
DAGಈ࡞֬ೝ ~ϩʔΧϧ։ൃڥߏங
DAGಈ࡞֬ೝ ~ϩʔΧϧ։ൃڥߏங LocalExecutorΛ༻
DAGಈ࡞֬ೝ ~ϩʔΧϧ։ൃڥߏங LocalExecutorΛ༻ dagsσΟϨΫτϦΛvolumeͱ͠ ͯϚϯτ
DAGಈ࡞֬ೝ ~ϩʔΧϧ։ൃڥߏங - dockerͷvolumeͱͯ͠dagsσΟϨΫτϦΛϚϯ τ͍ͯ͠ΔͷͰॻ͖͑ͨΒ͙͢ʹө - Web UIʹ͕ࣗ࡞ͨ͠DAGͷΈ͕දࣔ͞ΕΔ - ECR͔ΒimageΛऔͬͯ͘ΔΑ͏ʹͯ͠ຊ൪ͱಉ͡
ڥͰಈ࡞֬ೝͰ͖Δ
·ͱΊ - ETL͕ඞཁͳࣾ։ൃϓϩμΫτʹ͓͍ͯAirflowΛ ͍·ͨ͠ɻ - ຊ൪ڥͷAirflowECS FargateʹσϓϩΠ͠·ͨ͠ɻ - ϩʔΧϧ։ൃڥʹdockerΛ༻ͯ͠։ൃָ͕ʹͳΓ ·ͨ͠ɻ
We are hiring ! https://bit.ly/2UqWPGO
supplementary information
λεΫؒґଘؔͷఆٛ - Ϗοτγϑτԋࢉࢠ(>>, <<)Λ͍λεΫͷґଘؔΛද͢ - ޙଓλεΫͷ࣮ߦ݅શͯͷઌߦλεΫޭ͕σϑΥϧτઃఆ1 - શOperator͕࣋ͭtrigger_ruleҾͰ࣮ߦ݅ΛมߋՄೳ1 IUUQTBJSGMPXBQBDIFPSHDPODFQUTIUNMUSJHHFSSVMFT
- γϯϓϧͳґଘؔ task1 >> task2 - λεΫάϧʔϓ͕͋Δґଘؔ task1 >> [task2-1,task2-2] >> task3
Web UI: ϞχλϦϯά - Tree View - Gantt Chart
Web UI: Variable
ฒྻઃఆ Configuring parallelism in airflow.cfg - parallelism : ࢄॲཧΫϥελશମͰ࣮ߦՄೳͳϓϩηε -
dag_concurrency : ҰͭͷϫʔΧͰಉ࣮࣌ߦՄೳͳ࠷େϓϩηε - max_active_runs_per_dag : DAG෦Ͱಉ࣮࣌ߦՄೳͳ࠷େλε Ϋ - worker_concurrency : ҰͭͷCeleryϫʔΧͰಉ࣮࣌ߦՄೳͳ࠷େ ϓϩηε IUUQTBOBMZUJDTMJWFTFOTFDPKQFOUSZ