PyConDE 2016 - Building Data   Pipelines with Python

Slide 1

Slide 1 text

Building Data   Pipelines with Python Data Engineer @ TY  @mfcabrera  [email protected] Miguel Cabrera  PyCon Deutschland 30.10.2016

Slide 2

Slide 2 text

Agenda

Slide 3

Slide 3 text

Agenda Context Data Pipelines with Luigi Tips and Tricks Examples

Slide 4

Slide 4 text

Data Processing Pipelines

Slide 5

Slide 5 text

cat file.txt | wc -‐ l |   mail -‐s “hello” [email protected]

Slide 6

Slide 6 text

ETL

Slide 7

Slide 7 text

ETL • Extract data from a data source • Transform the data • Load into a sink

Slide 8

Slide 8 text

No content

Slide 9

Slide 9 text

Feature   Extraction Parameter Estimation Model Training Feature   Extraction Model Predict Visualize/ Format

Slide 10

Slide 10 text

Steps in diﬀerent technologies

Slide 11

Slide 11 text

Steps can be run in parallel

Slide 12

Slide 12 text

Steps have complex dependencies among them

Slide 13

Slide 13 text

Workﬂows • Repeat • Parametrize • Resume • Schedule it

Slide 14

Slide 14 text

No content

Slide 15

Slide 15 text

No content

Slide 16

Slide 16 text

“A Python framework for data ﬂow deﬁnition and execution” Luigi

Slide 17

Slide 17 text

Concepts

Slide 18

Slide 18 text

Concepts Tasks Parameters Targets Scheduler & Workers

Slide 19

Slide 19 text

Tasks

Slide 20

Slide 20 text

No content

Slide 21

Slide 21 text

Slide 22

Slide 22 text

Slide 23

Slide 23 text

Slide 24

Slide 24 text

Slide 25

Slide 25 text

WordCountTask ﬁle.txt wc.txt

Slide 26

Slide 26 text

WordCountTask ﬁle.txt wc.txt ToJsonTask wc.json

Slide 27

Slide 27 text

No content

Slide 28

Slide 28 text

Parameters

Slide 29

Slide 29 text

No content

Slide 30

Slide 30 text

Parameters Used to idenNfy the task From arguments or from conﬁguraNon Many types of Parameters (int, date, boolean, date range, Nme delta, dict, enum)

Slide 31

Slide 31 text

Targets

Slide 32

Slide 32 text

Targets Resources produced by a Task Typically Local files or files distributed file system (HDFS) Must implement the method exists() Many targets available

Slide 33

Slide 33 text

No content

Slide 34

Slide 34 text

Scheduler & Workers

Slide 35

Slide 35 text

No content

Slide 36

Slide 36 text

Source: h@p:/ /www.arashrouhani.com/luigid-‐basics-‐jun-‐2015

Slide 37

Slide 37 text

BaVeries Included

Slide 38

Slide 38 text

Batteries Included Package contrib ﬁlled with goodies Good support for Hadoop Diﬀerent Targets Extensible

Slide 39

Slide 39 text

Task Types Task -‐ Local Hadoop MR, Pig, Spark, etc SalesForce, ElasNcsearch, etc. ExternalProgram check luigi.contrib !

Slide 40

Slide 40 text

Target LocalTarget HDFS, S3, FTP, SSH, WebHDFS, etc. ESTarget, MySQLTarget, MSQL, Hive, SQLAlchemy, etc.

Slide 41

Slide 41 text

No content

Slide 42

Slide 42 text

Tips & Tricks

Slide 43

Slide 43 text

Separate pipeline and logic

Slide 44

Slide 44 text

Extend to avoid boilerplate code

Slide 45

Slide 45 text

DRY

Slide 46

Slide 46 text

Conclusion Luigi is a mature, baVeries-‐included alternaNve for building data pipelines Lacks of powerful visualizaNon of the pipelines Requires a external way of launching jobs (i.e. cron). Hard to debug MR Jobs

Slide 47

Slide 47 text

Lear More hVps:/ /github.com/spoNfy/luigi hVp:/ /luigi.readthedocs.io/en/stable/

Slide 48

Slide 48 text

Thanks!

Slide 49

Slide 49 text

Credits • pipe icon by Oliviu Stoian from the Noun Project • Photo Credit: (CC) h@ps:/ /www.flickr.com/photos/ 47244853@N03/29988510886 from hb.s via Compfight • Concrete Mixer: (CC) h@ps:/ /www.flickr.com/photos/ 145708285@N03/30138453986 by MasLabor via Compfight