High-performance Jackson #渋谷Java

第17回 #渋谷Java の発表資料です
http://shibuya-java.connpass.com/event/38674/

Transcript

1. High-performance
   Jackson
   ौ୩Java #17 2016-10-01
   KOMIYA Atsushi (@komiya_atsushi)
   

   View Slide

2. KOMIYA Atsushi
   @komiya_atsushi
   

   View Slide

3. Today’s topic
   

   View Slide

4. Jackson
   

   View Slide

5. = JSON parser / generator
   Ͱ͸ͳ͍
   (ݫີʹ͸)
   

   View Slide

6. “Jackson is a suite of data-
   processing tools for Java”
   https://github.com/FasterXML/jackson/blob/master/README.md
   

   View Slide

7. JSON parer / generator

   ʰ΋ʱඪ४αϙʔτ͍ͯ͠Δ
   σʔλॲཧϥΠϒϥϦ
   

   View Slide

8. Jackson
   • JSON processor
   • Streaming API
   • Data binding (POJO to and from JSON)
   • Other formats
   • XML, YAML
   • Avro, CBOR, CSV, Smile, Protobuf
   

   View Slide

9. Popular JSON Libraries
   • 4,180 usages: Jackson Databind
   • 2,820 usages: Gson
   • 1,238 usages: Json (org.json)
   • 516 usages: JSON.simple
   • 494 usages: Jersey Json
   ※mvnrepository.com ௐ΂ɺ2016-10-01 ࣌఺
   

   View Slide

10. Jackson as JSON processor
    • Pros
    • γϦΞϥΠζɾσγϦΞϥΠζͱ΋ʹ଎͍
    • ࠷଎Ͱ͸ͳͦ͞͏͚ͩͲ…
    • ػೳ͕๛෋Ͱɺᙱ͍ͱ͜Ζʹख͕ಧ͖΍͍͢
    • ৼΔ෣͍ͷΧελϚΠζ͸·͋·͋͠΍͍͢
    • ͍͍ҙຯͰރΕ͍ͯΔ
    ※ݸਓͷײ૝Ͱ͋Γɺੑೳɾੑ࣭Λอূ͢Δ΋ͷͰ͸͋Γ·ͤΜ
    

    View Slide

11. Jackson as JSON processor
    • Cons
    • ମܥతʹهड़͞Εͨྑ࣭ͳ೔ຊޠυΩϡϝϯτ͸ͳ͍
    • ػೳ͕๛෋ա͗Δ͕Ώ͑ʹɺ࢖͍͜ͳ͢ͷ͸೉͍͠
    • ʮ͜͏͍͏͜ͱ͸Ͱ͖Δͷ͔ͳʁʯ

    ʮͲ͏΍ͬͯઃఆ͢Ε͹͍͍Μͩʁʯ
    • ࣮૷͸៉ྷ͚ͩͲɺΫϥε਺͕ଟͯ͘ॳݟࡴ͠ײ͕͋
    Δ
    ※ݸਓͷײ૝Ͱ͋Γɺੑೳɾੑ࣭Λอূ͢Δ΋ͷͰ͸͋Γ·ͤΜ
    

    View Slide

12. High-performance
    Jackson
    

    View Slide

13. (JSON prosessor ͱͯ͠͸
    ࠷଎Ͱ͸ͳ͍͔΋͠Εͳ͍͕)

    Jackson Λ࠷଎ʹར༻͍ͨ͠❗
    

    View Slide

14. Shortest path to
    high-performance
    • Presentation: Jackson Performance
    • https://github.com/FasterXML/jackson-docs/
    wiki/Presentation:-Jackson-Performance
    • Jackson Best Practices: Performance
    • http://wiki.fasterxml.com/
    JacksonBestPracticesPerformance
    

    View Slide

15. ৄ͘͠͸ͪ͜ΒΛ͝ࢀরԼ͍͞
    http://qiita.com/komiya_atsushi/items/803f69b51426ed476a75
    

    View Slide

16. ϕετϓϥΫςΟεʹै͏͜ͱͰ
    ͲΕ͘Β͍଎͘ͳΔͷ͔ʁ
    

    View Slide

17. ࠓճର৅ͱ͍ͯ͠ΔλεΫ
    • 1 ߦ͝ͱʹ JSON ͕ه࿥͞ΕͨϑΝΠϧ

    (= Line delimited JSON) ͕༩͑ΒΕ͍ͯΔ
    • JSON 1 ݅͝ͱʹಛఆͷॲཧΛ͍ͨ͠
    • 㱺 JSON ͷσγϦΞϥΠζΛ࠷଎ʹ❗
    

    View Slide

18. ۩ମతͳαϯϓϧλεΫ
    • Twitter Streaming API Ͱ 1,000 ݅ͷ tweet Λऩूͯ͠ɺ̍ͭ
    ͷϑΝΠϧʹ·ͱΊ͓ͯ͘
    ǖ DVSMHFUIUUQTTUSFBNUXJUUFSDPNTUBUVTFTTBNQMFKTPOǘ
    • શ෦Ͱ໿ 4MB = JSON 1 ͭ͋ͨΓ 4KB ͙Β͍
    • ্هϑΝΠϧΛಡΈࠐΈɺ೔ຊޠ tweet ͷ݅਺Λ਺্͑͛Δ
    • ೔ຊޠ͔൱͔͸ “lang” ଐੑͷ஋Ͱ൑ఆ͢Δ
    • ͜ͷλεΫͷεϧʔϓοτ (ops/s) Λ JMH Ͱଌఆ͢Δ
    

    View Slide

19. Baseline implementation
    • Best practice ʹࢥ͍͖ͬΓ൓͢Δɺ͋͑ͯ

    ඇޮ཰ͳ࣮૷͔Β࢝ΊΔ
    • ϑΝΠϧ͔Β 1 ߦͣͭจࣈྻͱͯ͠ಡΈࠐΉ
    • Data-binding Ͱ JSON → POJO ʹσγϦΞϥ
    Πζ͢Δ
    • ͜ͷॲཧͷ౎౓ɺObjectMapper Λੜ੒͢Δ
    

    View Slide

20. Baseline implementation
    2.596 ops/s
    ͜Ε͕࠷΋ඇޮ཰
    

    View Slide

21. 1. ObjectMapper Λ࠶ར༻͢Δ
    • POJO from JSON ͷ data-binding Λ͢Δͷʹ
    ຖճੜ੒͍ͯͨ͠ ObjectMapper Λ࢖͍·Θ͢
    • ObjectMapper ͷΦϒδΣΫτੜ੒͸ಛʹॏ
    ͍ॲཧ
    • ObjectMapper ͸εϨουηʔϑ ☺
    

    View Slide

22. 1. ObjectMapper Λ࠶ར༻͢Δ
    32.533 ops/s (12.53x)
    

    View Slide

23. 2. ObjectReader Λར༻͢Δ
    • ObjectMapper ͷ୅ΘΓʹ ObjectReader Λ
    ࢖͏
    • ObjectMapper Λ࢖͏ΑΓͪΐͬͱ͚ͩ

    ޮ཰͕͍͍
    

    View Slide

24. 2. ObjectReader Λར༻͢Δ
    33.306 ops/s (1.024x)
    

    View Slide

25. 3. ObjectReader

    #readValues(InputStream)
    • 1 ߦͣͭจࣈྻͱͯ͠ಡΈࠐΉͷΛ΍Ίɺ
    InputStream ͷ··Ͱ Jackson ʹ༩͑Δ
    • Jackson ͸ byte[] ΍ InputSteam ͳͲͷ

    ೖྗιʔεʹରͯ͠࠷దԽ͍ͯ͠Δ
    • String ͷೖྗιʔεʹ͸ͦΜͳʹ࠷దԽ͍ͯ͠
    ΔΘ͚Ͱ͸ͳ͍
    

    View Slide

26. ࢀߟهࣄ (InfoQ)
    https://www.infoq.com/jp/news/2014/05/jackson-founder-responds
    

    View Slide

27. ࢀߟهࣄ (InfoQ)
    https://www.infoq.com/jp/news/2014/05/jackson-founder-responds
    “Jackson͕RESTͷج൫Ͱ͋Δ
    όΠτετϦʔϜɼ͋Δ͍͸
    ϑΝΠϧετϨʔδΛపఈత
    ʹ࠷దԽ͍ͯ͠Δ”
    

    View Slide

28. 3. ObjectReader

    #readValues(InputStream)
    43.641 ops/s (1.310x)
    

    View Slide

29. (ิ଍) ObjectMaper → JsonFactory
    → JsonParser ͱ͍͏ํ๏΋͋Δ
    

    View Slide

30. 4. Afterburner ϞδϡʔϧΛ࢖͏
    • POJO to and from JSON ͷ data-binding Λߴ଎Խ͢
    ΔϞδϡʔϧ
    • POJO ͷੜ੒΍ getter / setter ͷݺͼग़͠Λɺ

    ϦϑϨΫγϣϯͳ͠ʹόΠτίʔυੜ੒ͨ͠΋ͷͰ
    ࣮ݱ͢Δ
    • primitive ͳ஋ͷ boxing ΋ආ͚Δ͜ͱ͕Ͱ͖Δ
    • طଘͷίʔυ͸΄ͱΜͲมߋෆཁͰಋೖָ͕
    

    View Slide

31. 4. Afterburner ϞδϡʔϧΛ࢖͏
    49.764 ops/s (1.140x)
    ͜ΕΛ௥ه
    

    View Slide

32. 5. Streaming API Λར༻͢Δ
    • POJO to and from JSON ͸ɺͲ͏଍ૡ͍ͯ΋ॏ͍
    ॲཧͰ͋Δ͜ͱʹ͸มΘΒͳ͍
    • POJO ͷΦϒδΣΫτ͸ੜ੒͠ͳ͘ͱ΋ɺ
    Streaming API Λ࢖͑͹ JSON ͸ύʔεͰ͖Δ
    • Streaming API Ͱ໨తͷॲཧΛ࣮૷͢Δͷ͸େม
    ͕ͩɺ଎౓໘ʹ͓͍ͯ͸Ұ൪ޮՌత❗
    

    View Slide

33. 5. Streaming API Λར༻͢Δ
    77.969 ops/s (1.567x)
    ͜ͷ࣮૷ΛؤுΔ
    

    View Slide

34. Benchmark & result
    

    View Slide

35. Benchmark app
    https://github.com/komiya-atsushi/

    java-playground/tree/master/jackson-performance
    

    View Slide

36. Result: table
    4DPSF
    #BTFMJOF
    3FVTF0CKFDU.BQQFS
    6TF0CKFDU3FBEFS
    0CKFDU3FBEFSSFBE7BMVFT
    
    "GUFSCVSOFS
    4USFBNJOH"1*
    

    View Slide

37. Result: chart
    2x~ faster !
    

    View Slide

38. Conclusion
    

    View Slide

39. Conclusion
    • Jackson Ͱ POJO to and from JSON ͷγϦΞϥΠζɾσγϦΞ
    ϥΠζΛߴ଎ʹॲཧ͍ͨ͠ͷͰ͋Ε͹:
    • Jackson ͷΦϒδΣΫτ (ObjectMapper, ObjectReader ͳͲ)
    ͸ੵۃతʹ࠶ར༻͠Α͏
    • ͳΔ΂͘ InputStream ΦϒδΣΫτΛ Jackson ʹ༩͑ΔΑ͏
    ʹ͠Α͏
    • Afterburner ΋࢖͏͜ͱΛݕ౼͠Α͏
    • ΋ͬͱ଎౓ΛՔ͍͗ͨͷͰ͋Ε͹ɺStreaming API Λར༻͠Α͏
    

    View Slide

40. Thank you !
    

    View Slide