Upgrade to Pro — share decks privately, control downloads, hide ads and more …

Understanding distributed processing in Python

chie8842
June 16, 2019
Technology
2
2k

Understanding distributed processing in Python

#pyconth2019
https://th.pycon.org/talks/#row-47
Slides of my talk at PyConTh2019

chie8842

June 16, 2019
Tweet

More Decks by chie8842

See All by chie8842
MongoDB Atlas Search のご紹介
chie8842
2
1.5k
MongoDB Atlas Vectorsearchではじめる生成AIアプリ開発
chie8842
3
1.5k
AWS GlueとAWS Lake Formationではじめるデータマネジメント
chie8842
0
1k
Distributed Processing in Python
chie8842
2
680
クックパッドにおける推薦（と検索）の取り組み
chie8842
20
8k
Performance Tuning Tips of TensorFlow Inference
chie8842
1
730
クックパッドにおけるCloud AutoML事例
chie8842
9
7.8k
Cookpad_Internship_MLOps_Lecture_2018
chie8842
35
16k
機械学習デプロイを支えるコンテナ技術(Machine Learning on Docker)
chie8842
14
8.3k

Other Decks in Technology

See All in Technology
Cracking the Coding Interview 6th Edition
gdplabs
14
28k
AWSを活用したIoTにおけるセキュリティ対策のご紹介
kwskyk
0
280
PHPで印刷所に入稿できる名札データを作る / Generating Print-Ready Name Tag Data with PHP
tomzoh
0
180
依存パッケージの更新はコツコツが勝つコツ！ / phpcon_nagoya2025
blue_goheimochi
3
200
Amazon Aurora のバージョンアップ手法について
smt7174
1
110
Raycast AI APIを使ってちょっと便利な拡張機能を作ってみた / created-a-handy-extension-using-the-raycast-ai-api
kawamataryo
0
180
わたしのOSS活動
kazupon
2
330
ESXi で仮想化した ARM 環境で LLM を動作させてみるぞ
unnowataru
0
150
コンピュータビジョンの社会実装について考えていたらゲームを作っていた話
takmin
1
550
IAMポリシーのAllow/Denyについて、改めて理解する
smt7174
2
180
プロダクトエンジニア構想を立ち上げ、プロダクト志向な組織への成長を続けている話 / grow into a product-oriented organization
hiro_torii
1
330
Goで作って学ぶWebSocket
ryuichi1208
3
2.4k

Featured

See All Featured
Responsive Adventures: Dirty Tricks From The Dark Corners of Front-End
smashingmag
251
21k
Code Review Best Practice
trishagee
67
18k
CSS Pre-Processors: Stylus, Less & Sass
bermonpainter
356
29k
Side Projects
sachag
452
42k
The Art of Delivering Value - GDevCon NA Keynote
reverentgeek
10
1.3k
Build The Right Thing And Hit Your Dates
maggiecrowley
34
2.5k
Designing Experiences People Love
moore
140
23k
Practical Tips for Bootstrapping Information Extraction Pipelines
honnibal
PRO
12
980
Raft: Consensus for Rubyists
vanstee
137
6.8k
Templates, Plugins, & Blocks: Oh My! Creating the theme that thinks of everything
marktimemedia
30
2.2k
How to train your dragon (web standard)
notwaldorf
91
5.9k
Product Roadmaps are Hard
iamctodd
PRO
50
11k

Transcript

  1. Understanding distributed processing in Python Chie Hayashida

  2. $IJF)BZBTIJEB !DIJF) • 4PGUXBSF&OHJOFFSBU$PPLQBE GSPN+BQBO! • EFWFMPQBOEDPOUSJCVUFUPTPNF044 XSJUFBSUJDMFT BOENBLF QSFTFOUBUJPOT

    • 8POGJSTUQSJ[FBUSF*OWFOU (BNF%BZ

  3. $PPLQBE ‒ worldʼs largest recipe sharing platform Try it

  4. $PPLQBE JTHMPCBMTFSWJDF

  5. "HFOEB •CBTJDLOPXMFEHFPG DPODVSSFOUQBSBMMFMEJTUSJCVUFEQSPDFTTJOH •QBSBMMFMQSPDFTTJOHJO1ZUIPO •EJTUSJCVUFEQSPDFTTJOHJO1ZUIPO

  6. %JTDMBJNFS • 8IBUEP*UBML • CBTJDBSDIJUFDUVSFPGQBSBMMFMEJTUSJCVUFE QSPDFTTJOH DPNQVUJOH MJCSBSJFTJO1ZUIPO • 8IBUEP*OPU

    UBML • IPXUPJOTUBMMBOETFUVQFBDIMJCSBSJFT • VTBHFTPGEFUBJMFE"1*TPGXIBUXFJOUSPEVDF

  7. 8IFOXFVTFQBSBMMFMEJTUSJCVUFE QSPDFTTJOH FH •%PXOMPBEMBSHFEBUBJOTIPSUUJNF •6TFMBSHFEBUBGPS%BUBBOBMZTJTPS .BDIJOF-FBSOJOH •)BOEMFBMPUPGSFRVFTUTTJNVMUBOFPVTMZ BU8FC4JUF

  8. 28IBUJT UIFEJGGFSFODFPGCFMPX •$PODVSSFOUQSPDFTTJOH •1BSBMMFMQSPDFTTJOH

  9. l$PODVSSFOUQSPDFTTJOHzBOE l1BSBMMFM1SPDFTTJOHz $PODVSSFODZJTBCPVU EFBMJOHXJUI MPUTPG UIJOHTBUPODF 1BSBMMFMJTNJTBCPVU EPJOH MPUTPGUIJOHTBUPODF 

    3PC1JLFIUUQTUBMLTHPMBOHPSHXB[BTMJEF

  10. l$PODVSSFOUQSPDFTTJOHzBOE l1BSBMMFM1SPDFTTJOHz )FSFJT$PODVSSFOU )FSFJTQBSBMMFMBOEBMTPDPODVSSFOU

  11. $PODVSSFOUBOEQBSBMMFMQSPHSBNTDIFNF •$PODVSSFOUQSPHSBNTDIFNF $PODVSSFOU 1SPHSBN 1SPDFTT 1SPDFTT TDIFEVMFS $16 1SPDFTT 2VFVF

    1BSBMMFM 1SPHSBN 1SPDFTT 1SPDFTT $16 $16 •1BSBMMFMQSPHSBNTDIFNF

  12. 28IBUJT UIFEJGGFSFODFPGCFMPX •1BSBMMFMQSPDFTTJOH •%JTUSJCVUFEQSPDFTTJOH

  13. l1BSBMMFMQSPDFTTJOHzBOEl%JTUSJCVUFEQSPDFTTJOHz • 1BSBMMFMQSPDFTTJOH 1SPDFTTPS T .FNPSZ 1SPDFTTPS T .FNPSZ 1SPDFTTPS

    T .FNPSZ 1SPDFTTPS 1SPDFTTPS 1SPDFTTPS .FNPSZ • %JTUSJCVUFEQSPDFTTJOH /8 /8 /8 QSPDFTTPSTIBSFTNFNPSZ*OQBSBMMFMQSPDFTTJOH POUIFPUIFSIBOEQSPDFTTPSTIBTPXONFNPSJFTJOEJTUSJCVUFEQSPDFTTJOH

  14. 8IFOBOEIPXTIPVMEXF VTFQBSBMMFMEJTUSJCVUFE BSDIJUFDUVSFJO1ZUIPO

  15. UISFBEJOHBOENVMUJQSPDFTTJOH •1ZUIPOIBTPGGJDJBMNPEVMFTGPSQBSBMMFMQSPDFTTJOH • l#BUUFSZJODMVEFEz •8FDBOQBSBMMFMJ[FPVSQSPHSBNXJUIPVUBOZUIJSE QBSUZMJCSBSJFT

  16. UISFBEJOH • UISFBEJOHNPEVMFEJWJEFTPOFQSPDFTTJOUPNVMUJQMF UISFBET .BJO5ISFBE 5ISFBE 5ISFBE 1SPDFTT

  17. UISFBEJOH • (*- (MPCBM *OUFSQSFUFS-PDL MJNJUTPOMZPOFUISFBEDBOCF JOBTUBUFPGFYFDVUJPOBUBOZQPJOUJOUJNFJO$1ZUIPO 1SPDFTT 5ISFBE 

    5ISFBE  5ISFBE  SVO BXBJU BXBJU SVO BXBJU SVO SVO SVO BXBJU BXBJU SVO "DRVJSF-PDL

  18. 8IZ(*- • 5ISFBETJOQSPDFTTVTFTTBNFNFNPSZTQBDF • (*-XBTDIPTFOUPQSPUFDUUIFJOUFSOBMNFNPSZVTFECZ UIFJOUFSQSFUFS QSPDFTT UISFBE UISFBE .FNPSZ4QBDF

  19. *OXIBUTJUVBUJPOTJTUISFBEJOH FGGFDUJWF • 5BTLTXIJDIJT*0 CPVOETVDIBT • SFBEXSJUFGJMFT • %#DPOOFDUJPO •

    %PXOMPBEEBUBGSPN/8 1SPDFTT 5ISFBE  5ISFBE  5ISFBE  SVO BXBJU BXBJU SVO BXBJU SVO SVO SVO BXBJU BXBJU SVO "DRVJSF-PDL

  20. *OXIBUTJUVBUJPOTJTUISFBEJOH FGGFDUJWF • 5BTLTXIJDIJT*0 CPVOETVDIBT • SFBEXSJUFGJMFT • %#DPOOFDUJPO •

    %PXOMPBEEBUBGSPN/8 1SPDFTT 5ISFBE  5ISFBE  5ISFBE  SVO BXBJU BXBJU SVO BXBJU SVO SVO SVO *0 BXBJU SVO "DRVJSF -PDL *0 *0 *0 VTF$16 EPOPUVTF$16

  21. NVMUJQSPDFTTJOH •NVMUJQSPDFTTJOHTQBXOTQSPDFTTFTGPS QBSBMMFMJ[BUJPO •QBSBMMFMJ[FJOCPUIMPDBMBOESFNPUF .BJO1SPDFTT $IJME1SPDFTT $IJME1SPDFTT

  22. NVMUJQSPDFTTJOH • FBDIQSPDFTTFTIBTJUTPXONFNPSZTQBDF • /PUBGGFDUFECZ(*- • -FTTMJLFMZUPDBVTFFSSPSTEVFUPEBUBDPSSVQUJPOPS EFBEMPDL .BJO1SPDFTT $IJME1SPDFTT

    $IJME1SPDFTT PTGPSL .FNPSZ4QBDF .FNPSZ4QBDF $PQZ .FNPSZ4QBDF $PQZ

  23. *OXIBUTJUVBUJPOTJTNVMUJUISFBEJOH FGGFDUJWF • 5BTLTXIJDIJT$16CPVOETVDIBT • /VNFSJDBMDBMDVSBUJPO • 3FHVMBSFYQSFTTJPOQSPDFTTJOH • #VUUIFQFSGPSNBODFJO1ZUIPO`TNVMUJQSPDFTTJOHEPFTO`U

    JODSFBTFMJOFBSMZ

  24. NVMUJQSPDFTTJOH • 5IFSFJTPWFSIFBEEVFUPDPNNVOJDBUJPO CFUXFFOQBSFOUQSPDFTTBOEDIJMEQSPDFTT .BJO1SPDFTT $IJME1SPDFTT $IJME1SPDFTT PTGPSL .FNPSZ4QBDF .FNPSZ4QBDF

    $PQZ .FNPSZ4QBDF $PQZ

  25. NVMUJQSPDFTTJOH •UIFNBYOVNCFSPGUBTLTXIJDIDBOCFSVO TJNVMUBOFPVTMZJTSFTUSJDUFEUP$16OVNCFS 5BTL 5BTL 5BTL 5BTL 5BTL 5BTL $16

    $16

  26. +PCMJC • UIJSEQBSUZMJCSBSZGPSQBSBMMFMQSPDFTTJOHJO1ZUIPO • +PCMJC DBOVTF UISFBEJOHPSNVMUJQSPDFTTJOHGPSCBDLFOE • MB[ZFWBMVBUJPOVTJOHUSBOTQBSFOUEJTLDBDIJOHPGUIF PVUQVU

    • FBTZBOETJNQMFDPEJOH • MPHHJOHBOEUSBDJOHPGUIFFYFDVUJPO https://joblib.readthedocs.io/

  27. 4VNNBSZVQUPIFSF • 1ZUIPOIBTPGGJDJBMNPEVMFTGPSQBSBMMFMQSPDFTTJOH • UISFBEJOHJTFGGFDUJWFUP*0CPVOEQSPDFTTJOH • NVMUJQSPDFTTJOHJTFGGFDUJWFUP$16CPVOEQSPDFTTJOH • CPUIPGUXPBSFNBJOMZGPSPOMZGPSQSPDFTTFTSVOBUPOMZ POFNBDIJOF

    • TPNFPGEBUBMJCSBSJFTUSBOTQBSFOUMZVTFUIFTFQBSBMMFM QSPDFTTJOHMJCSBSJFT

  28. %JTUSJCVUFEQSPDFTTJOHJO1ZUIPO • 8IFOXFVTFMBSHFEBUBXIJDIDPVMEO`UQSPDFTTJOPOF NBDIJOF XFIBWFUPDPOTJEFSBCPVUEJTUSJCVUFE QSPDFTTJOH • QSPQFSUJFTXIJDITIPVMECFJODMVEFEJOEJTUSJCVUFE GSBNFXPSL •

    "VUPNBUJDSFDPWFSZ GSPNFSSPS • 5BTLEJWJTJPO KPCTDIFEVMJOH • 5PPMTGPS.POJUPSJOH %FCVHHBCJMJUZ

  29. %JTUSJCVUFEQSPDFTTJOHJO1ZUIPO *JOUSPEVDFUISFFGBNPVTMJCSBSJFTGPS EJTUSJCVUFEQSPDFTTJOHJO1ZUIPO •$FMFSZ •%BTL •1Z4QBSL

  30. $FMFSZ •EJTUSJCVUFEUBTLRVFVFCVJMUJO1ZUIPO •TJNQMFBOEGMFYJCMFBSDIJUFDUVSF • &WFSZQBSUTDBOCFFYUFOE • IFBWJMZVTFEGPSUBTLCBTFEXPSLMPBETJO 1ZUIPO • XJUI%KBOHP

    • "JSGMPXCBDLFOE

  31. $FMFSZ"SDIJUFDUVSF $MJFOU CSPLFS 3BCCJU.2 3FEJT  424 FUD 2VFVF 2VFVF

    2VFVF 8PSLFS $POTVNFS 8PSLFS $POTVNFS #BDLFOE 3FTVMUT 3FEJT FUD 8PSLFS $POTVNFS 8PSLFS $POTVNFS

  32. *OXIBUTJUVBUJPOTJTDFMFSZFGGFDUJWF  • 8IFOUIFUBTLJTTJNQMFCVUUIFBNPVOUPGEBUBJTUPP NVDIGPSPOFNBDIJOF • 8IFOZPVXBOUUPVTFJUXJUIBOPUIFSGSBNFXPSLTVDIBT %KBOHP

  33. %BTL • EJTUSJCVUFEQSPDFTTJOHGSBNFXPSLCVJMUJO1ZUIPO • CFBCMFUPVTF/VN1Z1BOEBT-JTUTMJLFQBSBMMFM PCKFDUT EBTLBSSBZ EBTLEBUBGSBNF EBTLCBH BOE"1*T

    • 4DIFEVMFSJTDVTUPNJ[BCMFUPHFUHPPEQFSGPSNBODFGPS CPUIGPSQBSBMMFMQSPDFTTJOHPOMPDBMNBDIJOFBOEGPS EJTUSJCVUFEQSPDFTTJOHPODMVTUFS • )BTHPPE8FC*OUFSGBDFGPSSFBMUJNFKPCNPOJUPSJOH

  34. %BTL TDIFEVMFS http://docs.dask.org/en/latest/scheduling.html

  35. %BTL 8FC*OUFSGBDF

  36. *OXIBUTJUVBUJPOTJT%BTL FGGFDUJWF  • 8IFOXFXBOUUPQBSBMMFMJ[F/VN1ZPSQBOEBTQSPDFTTJOH • 8IFOUIFEBUBTJ[FGJUTPOFUP NBDIJOFT

  37. 1Z4QBSL • CVJMEJO+BWBBOEIBT1ZUIPO*OUFSGBDF • DBOCFNPSFTDBMFPVUUIBODFMFSZBOEEBTL • 0SJHJOBMMZCVJMUUPSVOPOB)BEPPQDMVTUFS • GBTUBOEDPTUFGGJDJFOUQSPDFTTJOH •

    )BTHPPE8FC*OUFSGBDFGPSSFBMUJNFKPCNPOJUPSJOH • NVMUJGVODUJPOBM • TUSFBN "1* • NBDIJOFMFBSOJOH"1* • DBOCFVTFJONBOBHFNFOUTFSWJDFJODMPVETFSWJDF

  38. 1Z4QBSL • CVJMEJO+BWBBOEIBT1ZUIPO*OUFSGBDF only with partial API

  39. 1Z4QBSL • DBOCFNPSFTDBMFPVUUIBODFMFSZBOE%BTL node number 1 4-5 thousands

  40. *OXIBUTJUVBUJPOTJT1Z4QBSL FGGFDUJWF  • 8IFOXFXBOUUPSVODPNQMFYUBTLTXJUIBMBSHFBNPVOU PGEBUB • 8IFOXFIBWF)BEPPQDMVTUFS

  41. 4VNNBSZ • &YQMBJOFEBCPVUUIFEJGGFSFODFPG DPODVSSFOUQBSBMMFMEJTUSJCVUFEQSPDFTTJOH •*OUSPEVDFEUIFBSDIJUFDUVSFPG QBSBMMFMEJTUSJCVUFE QSPDFTTJOHMJCSBSJFT

  42. DPVMEO`UJOUSPEVDF • BTZODJP • DPODVSSFOUGVUVSF • "OPUIFSUIJSEQBSUZMJCSBSJFT

  43. %PObUCFBGSBJEUPVTFQBSBMMFM QSPDFTTJOHNPSF

  44. 2VFTUJPO