OSS Performance Tuning Tips

Transcript

1. OSS Performance Tuning Tips #gocon #gocon_hall GoCon 2019 Autumn @orisano

2. ໨త ̎ͭͷ͜ͱΛڞ༗͍ͨ͠

3. νϡʔχϯάΛͲ͏ਐΊΔͷ͔ Ҿ͖ग़͠Λ૿΍͢

4. νϡʔχϯάΛͲ͏ਐΊΔͷ͔ Ҿ͖ग़͠Λ૿΍͢

5. Ͳ͏ਐΊ͍ͯΔͷ͔

6. 0. ෆຬͷϋʔυϧΛԼ͛Δ

7. ෆຬ͕ վળͷ͖͔͚ͬʹͳΓ·͢

8. ීஈ࢖͍ͬͯΔ ϥΠϒϥϦɺιϑτ΢ΣΞ͕ ஗͍΋ͷͩͱࢥͬͯΈΔ

9. ͖͔͚ͬͱ ϞνϕʔγϣϯΛ ࣗΒ࡞Γग़͠·͠ΐ͏

10. 1. ܭଌͷϋʔυϧΛԼ͛Δ

11. ஗͍ݪҼΛ ؾܰʹ ௐ΂ΒΕΔೳྗΛ਎ʹ͚ͭΔ

12. ख͕͔͔ؒΔߦҝΛ ਓؒ͸͋·Γ͠ͳ͍

13. ͨ͘͞Μܦݧ͢Δ

14. ؾܰʹ ௐࠪͰ͖ΔΑ͏ͳ࢓૊ΈΛ ೖΕΔ

15. ͜Ε͸OSSΛެ։͢Δଆͷ࿩ ؀ڥม਺Ͱࢦఆͨ͠Γ σϑΥϧτͰprofile͕औΕΔ ࢓૊ΈΛఏڙ͢Δͱ վળ͠΍͍͢/͞Ε΍͍͢

16. ໰୊ʹૺ۰ͨ͠ਓ͕ ݪҼௐࠪ͠΍͍͢ͱخ͍͠

17. 2. ໰୊ՕॴΛಛఆ͢Δ

18. ஗͍ͱײͯ͡ ໰୊Օॴ͕ؔ਺ϨϕϧͰ Θ͔Βͳ͍ͷ͸ cliΛ࣮ߦ͍ͯ͠Δͱ͖

19. mainʹ github.com/pkg/profile Λ௥Ճͯ͠ΈΔ

20. None

21. ͜Ε͚ͩ

22. cliͷ࣮ߦ͕1౓ʹ 1࣌ؒҎ্͔͔Δ৔߹ͳͲ͸ net/http/pprofͷ΄͏͕ ྑ͍৔߹΋

23. profileΛݟͯ ؔ਺ϨϕϧͰಛఆ͢Δ

24. pprofͷweb൛ʹ׳ΕΔ ͪΌΜͱҙຯ͕Θ͔Δ

25. ͍ΖΜͳݟํͰݟͯΈΔ top, graph, flame-graph source, disasm

26. 3. BenchmarkΛॻ͘

27. ͢Ͱʹॻ͍ͯ͋Δ৔߹͸ ϥοΩʔ ஗͘ͳ͍ͬͯͳ͍৔߹͸ ஗͘ͳΔέʔεΛ௥Ճ͢Δ

28. BenchmarkΛॻ࣌͘͸ ࠷దԽͰফ͍͑ͯͳ͍͔֬ೝ

29. BenchmarkΛॻ͘ͱ ؆୯ʹprofile͕औΕΔ

30. go test -cpuprofile go test -memprofile Λ֮͑Α͏ (go help testflagΛಡ΋͏)

31. 4. ղܾࡦΛߟ͑Δ

32. Ͳ͏ղܾ͢Δ͔ʹ͍ͭͯ͸ “Ҿ͖ग़͠Λ૿΍͢”Ͱ

33. OSSͷ ύϑΥʔϚϯενϡʔχϯά ͸Ϛʔδ͞Εͯ׬ྃ

34. Ͳ͏΍Ε͹Ϛʔδ͞ΕΔ ύονʹͳΔ͔

35. Ϛʔδ͢Δͱ͍͏͜ͱ͸ ૬ख͕ϝϯςφϯε͢Δ ͱ͍͏͜ͱ

36. มߋ͕গͳ͍ ૬खͷίετΛ૿΍͞ͳ͍

37. ഁյతมߋ͕ͳ͍ ޿͘࢖ΘΕ͍ͯΔ΋ͷ΄Ͳ ॏཁͳͱ͜Ζ

38. ҟৗͳίετ͕ͳ͍ มߋ, อक͕೉͘͠ͳΔ࣮૷ (ΞηϯϒϥΛ࢖ͬͨΓ)

39. ґଘؔ܎Λ௥Ճ͢Δ৔߹͸ ৻ॏʹબ୒͢Δ ΞΫςΟϒʹϝϯςφϯε͞ Ε͍ͯΔ΋ͷ͔ ຊ౰ʹඞཁ͔?

40. ͘͢͝ύϑΥʔϚϯε͕ վળ͢Δ ܶతͳվળ͸ड͚ೖΕΒΕΔ

41. ςετ͕ॻ͍ͯ͋Δ ͢Ͱʹॻ͍ͯ͋Δ৔߹͸डཧ ͞Ε΍͍͢

42. 5. ࢼߦࡨޡ͢Δ

43. ·ͣॳظঢ়ଶͰ े෼ͳճ਺ timeoutʹͳΒͳ͍Α͏ʹ Benchmark݁ՌΛऔ͓ͬͯ͘

44. go test -bench . -count=10 -timeout=30000s | tee old.txt

45. ύονΛૹΔͱ͖ʹ benchstatͷ݁ՌΛૹΔͷͰ ϕʔεϥΠϯ͸े෼ͳճ਺Ͱ

46. timeout͸ σϑΥϧτͩͱ10mͰ ҙ֎ͱΦʔόʔ͢Δ

47. ղܾࡦΛࢥ͍͍ͭͨΒ ·ͣ͸1౓Benchmark

48. ے͕ྑͦ͞͏ͳΒ े෼ͳճ਺Benchmark

49. ଟ͘ࢼߦࡨޡ͍ͯ͠Δͱ profileͱsourceͷؔ܎͕ Θ͔Βͳ͘ͳΔ

50. pprofʹ͸ sourceΛݟΔػೳ͕͋Δ͕ อ͍࣋ͯ͠ΔΘ͚Ͱ͸ͳ͍ pathͷΈ

51. git add . git commit -m “$2” REV=$(git rev-parse HEAD)

    go test -bench $1 -benchmem -cpuprofile cpu.${REV}.pb.gz -memprofile mem.$ {REV}.pb.gz | tee ${REV}.txt rm ./${REV}.* && git reset HEAD^

52. source͸gitʹ؅ཧͯ͠΋Β͏ ͪΌΜͱඥ෇͚ΒΕΔ

53. ଟ͘ͷprofile͕͋Δͱ ୯ମͷޮՌ͕Θ͔Βͳ͍ ͜ͱ͕͋Δ

54. go tool pprof -diff_base Ͱprofileؒͷࠩ෼͕ݟΕΔ

55. 6. ύονΛૹΔ

56. ࠷ऴͷBenchmarkΛ े෼ͳճ਺ߦ͏

57. ઐ༻ͷΠϯελϯεͰܭଌ ೉͍͠৔߹͸ ۃྗ֎෦ϓϩηεΛఀࢭ͢Δ

58. νϡʔχϯάΛͲ͏ਐΊΔͷ͔ Ҿ͖ग़͠Λ૿΍͢

59. ࣮ྫϕʔεͰ ͲͷΑ͏ʹղܾ͔ͨ͠ ࣗ෼ͳΒͲ͏ղܾ͢Δ͔ ߟ͑ͯ΋Β͍͍ͨ

60. 1. ࣮ࡍʹ౤͛ͨύον 2. ଎͍ϥΠϒϥϦ͕ॻ͖͍ͨ

61. 1. ࣮ࡍʹ౤͛ͨύον 2. ଎͍ϥΠϒϥϦ͕ॻ͖͍ͨ

62. src-d/go-git

63. pure-goͰ gitͷૢ࡞Λ͢ΔͨΊͷ ϥΠϒϥϦ

64. ύονΛૹͬͨഎܠ

65. aquasecurity/trivy Ͱ౰ॳ࢖͍͕ͬͯͨ େ͖ͳrepositoryͷcloneʹ ҟৗͳ࣌ؒ(10min~)͕͔͔Δ ໰୊͕͋ͬͨ

66. ݪҼ͸ gitͷIndexͷߏஙΛߦ͏Օॴ ͩͬͨ

67. Indexͷ಺෦දݱ(public)͕ sliceʹͳ͓ͬͯΓ nameͰҰҙʹ͢ΔͨΊʹ deleteͱappend͕ ϑΝΠϧ਺ճ࣮ߦ͞Ε͍ͯͨ

68. delete͸nameΛࢦఆͯ͠ ࡟আ͢ΔͷͰ sliceͷཁૉΛ͢΂ͯ૸ࠪ͢Δ

69. ౰ॳ͸ ಺෦දݱΛmapʹมߋ͢Ε͹ deleteͷܭࢉྔΛݮΒͤΔ ͷͰͦ͏͠Α͏ͱࢥͬͨ

70. ͔͠͠ publicͳϑΟʔϧυ internalͰͳ͍ύοέʔδ 4500 star௒͑ͷ஌໊౓

71. ౸ఈड͚ೖΕΒΕͳ͍ มߋͩͱࢥ͍ఘΊͨ

72. ޙ೔

73. ιʔείʔυΛ ݟ௚͍ͯ͠Δͱ

74. ໌ه͞Ε͍ͯΔ ϑΟʔϧυͷ࢓༷ (sliceͷॱং͸อূ͠ͳ͍) େྔͷIndexΛߏங͢Δ λΠϛϯά(clone࣌)

75. ݟ௚ͨ݁͠Ռ ഁյతมߋແ͠Ͱ मਖ਼͢Δํ๏Λࢥ͍͍ͭͨ

76. ଟ͘ݺͼग़͞ΕΔՕॴ ͷΈʹ͓͍ͯ mapͰอ࣋͠return͢Δࡍʹ sliceʹม׵͢Δ

77. mapΛprivateͳؔ਺Ͱ Ҿ͖ճ͢Α͏ʹ͢Δ͜ͱͰ ഁյతมߋΛճආͰ͖ͨ

78. None

79. ͦ͜Λղܾ͢Δͱ mallocgc͕ॏ͘ͳͬͨ

80. ϝϞϦ࢖༻ྔͷݪҼ͸ io.Copyʹ͋ΔΑ͏ͩͬͨ

81. ϑΝΠϧ͝ͱʹio.Copy͕ ݺ͹Ε͍ͯͨ

82. େྔʹݺ͹ΕΔՄೳੑ͕͋Δ ՕॴͰ͸io.CopyͰ͸ͳ͘ io.CopyBufferΛ࢖͏ͱ ϝϞϦ࢖༻ྔΛ੍ޚͰ͖Δ

83. ࢖͏bufferΛͲ͏ࢦఆ͢Δ͔ ֎෦͔Β༩͑ΒΕͳ͍ͱ ݁ہϝϞϦ࢖༻ྔมΘΒͳ͍

84. privateͳϝιουͱ͸͍͑ Ҿ਺ʹ௥Ճ͢Δͱ มߋൣғ͕େ͖͘ͳΔ

85. ղܾࡦͱͯ͠ globalʹsync.PoolΛ ஔ͘͜ͱʹͨ͠

86. None

87. https://github.com/src-d/ go-git/pull/1179

88. None

89. 605ඵ -> 249ඵ

90. ͔͠͠ ·ͩ·ͩϝϞϦ࢖༻ྔ͕ଟ͍ 56 GB/op

91. ϑΝΠϧʹࠩ෼Λద༻͢Δ ॲཧ͕ϝϞϦΛେྔʹ࢖༻

92. ؔ਺಺෦Ͱ͸ publicͷؔ਺Ͱ buffer͕֎෦͔Β౉ͤͳ͍ ΋ͷ͕࢖ΘΕ͍ͯͨ

93. গͳ͘ͱ΋಺෦͔Βͷ࢖༻ Ͱ͸bufferΛࢦఆ͍ͨ͠

94. None

95. ಺෦޲͚ʹbufferࢦఆͰ͖Δ Α͏ʹͯ͠ղܾ

96. https://github.com/src-d/ go-git/pull/1180

97. 56.1 GB -> 29.8 GB

98. None
99. None

100. image/png

101. ύονΛૹͬͨഎܠ

102. ࣾ಺ISUCONͰ QRίʔυΛߴ଎ʹੜ੒͢Δ ඞཁ͕͋ͬͨͨΊ

103. ࣾ಺ISUCONͷৼΓฦΓͰ ύϑΥʔϚϯενϡʔχϯά ΛҰਓͰָ͠ΜͰ͍ͨ

104. ऄ଍

105. ߴ଎ʹେྔͷpngΛ ग़ྗ͢ΔͨΊʹ͸

106. 1.9͔Βೖͬͨ png.EncoderBufferPool

107. CompressionLevelͷઃఆ

108. ಠࣗimage.Image࣮૷Λ ࢖Θͳ͍

109. നࠇը૾Ͱ͋Ε͹ image.GrayΛ࢖͏ (Opaque͕bypassͰ͖Δ)

110. ऄ଍ऴྃ

111. image/png͕ bottleneckʹͳ͖ͬͯͨ

112. ͢ͰʹBenchmark͕ ॻ͔Ε͍ͯͨͷͰ࣮ߦ

113. ࠓճ࢖͍ͬͯΔ ՕॴͰ͸ͳ͍͕࿐ࠎʹ஗͍ ෦෼͕͋ͬͨ

114. https://go- review.googlesource.com/ c/go/+/187417

115. None

116. ࣮ࡍॏ͔ͬͨՕॴ͸ compress/deflate

117. pprof͕ॏ͍ͱࣔͨ͠ͷ͸ for͕ॻ͔Ε͍ͯΔߦͩͬͨ

118. None

119. ౰֘forจͷasmΛݟͯΈΔͱ ແବͳϝϞϦΞΫηε͕

120. compilerʹregisterΛ ࢖ͬͯ΋Β͏ͨΊʹ ϩʔΧϧม਺Λఆٛ

121. None

122. https://go- review.googlesource.com/ c/go/+/187837

123. None
124. None

125. GoogleContainerTools/ kaniko

126. ύονΛૹͬͨഎܠ

127. ΞΠσΟΞ͕εΩͩͬͨ

128. kanikoΛCIͰ࢖ͬͯ Կʹ͕͔͔͍࣌ؒͬͯΔͷ͔ ؾʹͳͬͨ

129. kaniko͸ϝϞϦ্ʹ filesystemͷsnapshotΛ࣋ͭ

130. ίϚϯυΛ࣮ߦ͢Δͨͼʹ ࠩ෼͕ͳ͍͔ൺֱ͢Δ

131. md5Ͱൺֱ͢Δ

132. ͦͷmd5͕ॏ͔ͬͨ

133. ϑΝΠϧ͕ ಉҰ͔Ͳ͏͔͚ͩͰྑ͍ͷͰ md5Ͱ͋Δඞཁ͸ͳ͍

134. minio/HighwayHashʹมߋ (ຊ౰͸Α͘ͳ͍)

135. None
136. None

137. mount͞Ε͍ͯΔ σΟϨΫτϦ͸ snapshotର৅֎

138. ϑΝΠϧ͕whitelistʹ ؚ·ΕΔ൑ఆ͢ΔՕॴͰ strings.SplitΛ࢖༻

139. ϑΝΠϧ਺͕ଟ͔ͬͨΓ directory͕ਂ͍ͱ ແବʹϝϞϦΛ࢖༻͢Δ

140. ಛੑΛߟ͑ͯ strings.SplitNΛ࢖༻

141. None
142. None

143. https://github.com/ GoogleContainerTools/ kaniko/pull/694

144. 129.54s -> 88.29s

145. None

146. 1. ࣮ࡍʹ౤͛ͨύον 2. ଎͍ϥΠϒϥϦ͕ॻ͖͍ͨ

147. orisano/wyhash

148. kanikoͷύονΛ ॻ͍͍ͯΔͱ͖ʹ Կ͕ྑ͍hashͳͷͩΖ͏

149. Q. ଎͍hash͸?

150. Q. ଎͍hash͸? A. ܭଌ͠·͠ΐ͏

151. dgryski/trifles/hashbench खݩͰ૸ΒͤͯΈ·͠ΐ͏

152. (ݟ͔ͭΒͳ͍package͕ ͋ΔͷͰಈ͖·ͤΜ)

153. wyhash͕ GitHubͷTrendingͰ ্͕͖ͬͯͨ

154. ଎ͯ͘ϙʔλϒϧͰڧ͍ Β͍͠

155. ඇৗʹ୯७ͳͷͰ GoʹҠ২ͯ͠ΈΑ͏ͱࢥͬͨ

156. 2೔ҐͰҠ২͕ऴྃ

157. hashbenchʹ௥Ճ ֬ೝ͢ΔͱϘϩෛ͚͍ͯ͠Δ

158. ͜Μͳܭࢉ͔͠ͳ͍ॲཧΛ Ͳ͏΍ͬͯߴ଎Խ͢Δͷ͔

159. ྨࣅϥΠϒϥϦͷௐࠪ

160. ҉߸ܥ΍hashܥ͸ جຊతʹasm͕࢖ΘΕ͍ͯΔ

161. asmΛ࢖͏ͱ଎͍?

162. ॻ͍ͯΈΑ͏

163. Go asm͸ಠಛͳײ͡ ॻ͍͍ͯΔ೔ຊਓ͕ ΄ͱΜͲ͍ͳ͍? ͋·Γࢿྉ͕ͳ͍

164. ؤுͬͯ AVXΛ࢖ͬͯॏ͍ॲཧΛॻ͘

165. lldbΛ࢖ͬͯ bug(SEGV)Λमਖ਼͢Δ

166. Benchmark݁Ռ ஗͘ͳ͍ͬͯΔ

167. None

168. Կނ͔

169. asmͰॻ͍ͨؔ਺͸ inlineԽ͞Εͳ͍

170. math/bits encoding/binary ίϯύΠϥ͕ݡ͘࠷దԽ͢Δ https://dave.cheney.net/ 2019/08/20/go-compiler- intrinsics

171. inlineԽ͞ΕΔΑ͏ͳ খ͞ͳؔ਺͸ asmͷޮՌ͕ಘΒΕͳ͍

172. ࠓճͷΑ͏ͳ৔߹ͩͱ loop·ͰؚΊͯasmԽ͢΂͖

173. ෆ׳ΕͳasmͰ ଟ͘ͷίʔυΛॻ͖ͨ͘ͳ͍

174. mmcloughlin/avo Λ࢖͓͏

175. None

176. GoͰasmΛੜ੒͢Δ ϓϩάϥϜΛॻ͘Ξϓϩʔν

177. Կ͕ྑ͍͔?

178. Go asmͷ͓࡞๏Λ avo͕΍ͬͯ͘ΕΔ

179. Go IDEͰͷิ׬͕ޮ͘

180. avoΛ࢖ͬͯؤுͬͨ 5 GB/s -> 11 GB/s

181. ߴ଎ͳasmΛॻ͘ͷ͕೉͍͠

182. asmϨϕϧͰͳͥ஗͍ͷ͔ pprofͰ͸Θ͔Βͳ͍

183. ύΠϓϥΠχϯάΛҙࣝ͢Δ 11 GB/s -> 14 GB/s

184. ·ͱΊ

185. νϡʔχϯάΛͲ͏ਐΊΔͷ͔ • 0. ෆຬͷϋʔυϧΛԼ͛Δ • 1. ܭଌͷϋʔυϧΛԼ͛Δ • 2. ໰୊ՕॴΛಛఆ͢Δ

     • 3. BenchmarkΛॻ͘

186. νϡʔχϯάΛͲ͏ਐΊΔͷ͔ • 4. ղܾࡦΛߟ͑Δ • Ϛʔδ͞Ε΍͍͢ղܾࡦΛ໨ࢦ͢ • ϝϯςφϯείετΛ্͛ͳ͍΋ͷ • 5.

     ࢼߦࡨޡ͢Δ • 6. ύονΛૹΔ

187. Ҿ͖ग़͠Λ૿΍͢ • ϝϞϦ࢖༻ྔ͸໰୊ʹͳΓ͕ͪ • ֎෦͔ΒBuffer͕ड͚औΕΔAPIΛߟྀ͢Δ • มߋ͕༰қͰ͸ͳ͍৔߹sync.PoolΛߟྀ͢Δ

188. Ҿ͖ग़͠Λ૿΍͢ • ద੾ͳhashΛબ୒͢Δ • Ξηϯϒϥ͸ϝϯςφϯείετ͕ߴ͍ͷͰ ۃྗආ͚Δ • ॏ͍ॲཧͷ෦෼͚ͩσʔλͷ࣋ͪํΛม͑ͯ ΈΔ