フィードバック制御によるGoroutine起動数の最適化 / optimization-of-goroutine-numbers-by-feedback-control

ࡾ୐༔հ / Pepabo R&D Institute, GMO Pepabo, Inc.
2018.11.25 Go Conference 2018 Autumn
ϑΟʔυόοΫ੍ޚʹΑΔ
Goroutineىಈ਺ͷ࠷దԽ

View Slide

ϓϦϯγύϧΤϯδχΞ
ࡾ୐༔հ!NPOPDISPNFHBOF
(.0ϖύϘגࣜձࣾϖύϘݚڀॴ
IUUQTCMPHNPOPDISPNFHBOFDPN

View Slide

3

View Slide

1. ฒߦॲཧͱνϡʔχϯάʹ͓͚Δ՝୊
2. KaburayaΞʔΩςΫνϟ
3. Goݴޠ΁ͷద༻
4. ධՁ
5. ·ͱΊ
4
໨࣍

View Slide

1.
ฒߦॲཧͱνϡʔχϯάʹ͓͚Δ՝୊

View Slide

ΦϨ͸ΰϧʔνϯͰߦ͘ʂ

View Slide

• ϚϧνίΞ࣌୅ʹ͓͍ͯɼ୯ҰͷϚγϯͰॲཧੑೳΛ࠷େԽ͢ΔͨΊॲཧͷฒ
ߦԽ͕ߦΘΕΔɽ
• ࠷దͳฒߦ਺ΛٻΊ͍ͨɽ
• ϊϯϒϩοΩϯάͳॲཧํࣜͰ͸࠷దͳฒߦ਺͸CPUίΞ਺ͱ౳͍͠
• ϒϩοΩϯά͕Ұ෦෼Ͱ΋ൃੜ͢Δ৔߹͸CPUίΞ਺Ҏ্͕࠷దʹͳΔ৔
߹͕ଟ͍
7
ฒߦॲཧͱ࠷దͳฒߦ਺

View Slide

• ϒϩοΩϯά͕ൃੜ͢Δঢ়گʹ͓͍ͯ͸ɼ࠷దͳฒߦ਺ͷܾఆ͸ґવͱͯ͠։
ൃऀͷܦݧͱ஍ಓͳνϡʔχϯάʹґଘ
• nginx, unicornͷworker_processesɼSidekiqͷconcurrencyɼGoroutineͷ
ىಈ਺ͳͲ
• ࣮ߦ͞ΕΔॲཧͷಛੑͱϥϯλΠϜ΍εέδϡʔϦϯάͷಛੑΛߟྀͨ͠
νϡʔχϯά͕ඞཁ
• ͜ΕΒΛશͯߟྀͨ͠ීวతͳνϡʔχϯάख๏ͷൃݟ͸ࠔ೉
8
࠷దͳฒߦ਺ΛٻΊΔ

View Slide

• ద੾ͳࢦඪ͕બఆͰ͖Ε͹୯७ͳࢦඪͰ΋ޮՌ͕͋Δ
• Ծ૝αʔόͷ༧ଌతΦʔτεέʔϦϯά (*1)Ͱ͸ɼΞϓϦέʔγϣϯॲཧ಺
༰Ͱ͸ͳ͘ɼӡ༻ܦݧ͔Βಘͨαʔό୯ҐͷεϧʔϓοτΛࢦඪͱͨ͠ɽ
• ʮશࣗಈύϥϝʔλνϡʔχϯά͞Μʯ(*2)Ͱ͸ϝτϦΫεΛҰ఺ʹߜΓε
ϧʔϓοτ͕࠷େԽ͢Δ஋Λ୳ࡧతʹٻΊͨɽ
9
ैདྷख๏
*1. ࡾ୐ ༔հ, দຊ ྄հ, ྗ෢ ݈࣍, ܀ྛ ݈ଠ࿠, ΞΫηεස౓༧ଌʹجͮ͘Ծ૝αʔόͷܭըతΦʔτεέʔϦϯά, FIT 2018 ୈ17ճ৘ใ
Պֶٕज़ϑΥʔϥϜ, CL-002, Sep 2018.
*2. https://github.com/mirakui/tuningsan

View Slide

• ༧ଌతɼ୳ࡧతͳํࣜ͸WebαʔόʔϓϩηεͷΑ͏ʹ௕ظՔಈ͢Δ৔߹ʹ
ޮՌత͕ͩɼCLIͷΑ͏ͳ୹ظՔಈͰଈ࣌݁Ռ͕ٻΊΒΕΔ৔߹ʹద༻͕೉͠
͍
10
ैདྷख๏
• ࢦඪͷݶఆʹՃ͑ɼ൓Ԡత͔ͭޡ͕ࠩগͳ͘ࢦඪ΁௥ै͢Δख๏͕ٻΊΒΕΔ

View Slide

2.
KaburayaΞʔΩςΫνϟ

View Slide

• ॲཧ΍ϥϯλΠϜͷಛੑʹґଘͤͣʹɼϚγϯͷෛՙʹԠͯ͡ 
൓Ԡత͔ͭܧଓతʹɼฒߦ਺Λ࠷దԽ͢Δ͜ͱ͸Մೳ͔ʁ
12
ຊݚڀͷʮ໰͍ʯ

View Slide

• ॲཧ΍ϥϯλΠϜͷಛੑʹґଘͤͣʹɼϚγϯͷෛՙʹԠͯ͡൓Ԡత͔ͭܧଓ
తʹɼฒߦ਺Λ࠷దԽ͢Δ
• 1. ॲཧ΍ϥϯλΠϜΛϒϥοΫϘοΫεͱͯ͠ѻ͍ɼܧଓతʹಈ࡞Λ੍ޚ͢
Δ࢓૊ΈͰ͋ΔϑΟʔυόοΫ੍ޚΛ༻͍ͯ࠷దͳฒߦ਺ΛٻΊΔ
• 2. ฒߦ਺ͷ੍ޚʹηϚϑΥΛ༻͍Δ
• Kaburaya is “ధ໼”. It is japanese arrow with a whistle.
13
Kaburaya ΞʔΩςΫνϟ

View Slide

• ໨ඪ஋ͱग़ྗͷϑΟʔυόοΫͷޡࠩΛ༻͍ͯγεςϜͷೖྗΛܾఆ͠ɼग़ྗ
Λ໨ඪ஋ʹ͚ۙͮΔ
• ภࠩΛॲཧ͢Δ੍ޚثɼͦͯ͠ೖྗͱग़ྗʹԿΛ༻͍Δ͔ͷઃܭ͕ॏཁ
14
ϑΟʔυόοΫ੍ޚ
੍ޚث γεςϜ
ೖྗ ग़ྗ
ภࠩ
ϑΟʔυόοΫ
໨ඪ஋

View Slide

• ໨ඪ஋ͱͷࠩΛޮ཰తʹղফ͢ΔͨΊͷݹయత੍ޚख๏
• ൺྫ੍ޚ(P)ɼੵ෼੍ޚ(I)ɼඍ෼੍ޚ(D)ͷ࿨Λೖྗ஋ͱ͢Δ
15
PID੍ޚ(1/2)
੍ޚث
ภࠩ
໨ඪ஋ 1 Kpe(t)
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
* Ki
Z t
0
e(r)dr
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
% Kd
de(t)
dt
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

r
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
e
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

γεςϜ

ϑΟʔυόοΫ
ೖྗ ग़ྗ

View Slide

• ൺྫ੍ޚ(P)͸ภࠩʹൺྫ͢Δɽภ͕ࠩେ͖͍৔߹ʹେ͖͘मਖ਼ɽ
• ੵ෼੍ޚ(I)͸ภࠩͷ૯࿨Λ༻͍ΔɽPͷ࢒ཹภࠩΛղফ͢Δɽ
• ඍ෼੍ޚ(D)͸ภࠩͷඍ෼Λ༻͍Δɽ֎ཚ౳ʹΑΔٸมಈʹػහʹରॲ͢Δɽ
16
PID੍ޚ(2/2)
੍ޚث
ภࠩ
໨ඪ஋ 1 Kpe(t)
* Ki
Z t
0
e(r)dr
% Kd
de(t)
dt

r
e

γεςϜ

ϑΟʔυόοΫ
ೖྗ ग़ྗ

View Slide

1. ॳظ஋͔ΒϫʔΧʔ਺Λม͑ͳ͍FixController (ൺֱ༻)
2. CPU࢖༻཰100%Λ໨ඪ஋ͱ͠ɼෆ଍ͨ͠৔߹͸ workerΛ1ͣͭ૿Ճͤ͞Δ SimpleController
3. CPU࢖༻཰100%Λ໨ඪͱ͠ɼ໨ඪͱͷࠩ෼ͷKഒΛՃ͑Δ PController(ൺྫ੍ޚ)
4. 3.ͷCPU࢖༻཰ͷ໨ඪ஋Λ90%ͱͨ͠΋ͷ
5. ௚ۙ3؍ଌ఺ͷฏۉΛ໨ඪ஋ͱͯ͠3؍ଌ͝ͱʹPControllerͷ໨ඪ஋ΛมԽͤ͞ΔDynamicController
6. ҰఆظؒͷCPU࢖༻཰ͷඪ४ภࠩΛͱͬͯͦΕ͕Ұఆͷ஋ҎԼͩͬͨΒ҆ఆͨ͠ͱΈͳͯ͠ɼworkerΛݮΒ͍ͯ͠
͘StabilityController
7. CPU࢖༻཰Ͱ͸ͳͯ͘CPU࢖༻཰ͷมԽ཰Λݩʹ੍ޚ͢ΔRateController
8. 5.ͷDynamicControllerΛݩʹఆظͰ͸ͳ͘େ͖ͳมಈ͝ͱʹ໨ඪ஋Λݟ௚͢ํࣜ
9. 8.ͷDynamicControllerΛݩʹఆৗ࣌ͷෆཁͳworkerΛ࡟ݮ͢Δํࣜ
10. 9.ͷDynamicControllerΛݩʹมಈͷਫ਼౓ͱ଎౓Λ޲্ͤ͞ΔͨΊʹੵ෼ඍ෼੍ޚΛಋೖ͢Δํࣜ
17
࠷దͳ੍ޚثΛٻΊͯ

View Slide

• ઃܭ੍ͨ͠ޚثΛฒߦԽ͞ΕͨλεΫΛܭ
ࢉࢿݯ্ݶ·Ͱฒྻॲཧ͢Δ؆қతͳγϛϡ
ϨʔλͰධՁ
18
؆қධՁ
• λεΫಛੑʹΑͬͯCPU্ݶ͸ҟͳΔ
• ࠷దͳฒߦ਺͸࣮ߦ࣌ͷΈ൑அͰ͖ΔͨΊ
CPU࢖༻཰͕ҰఆͱͳΔΑ͏ʹ૿ݮ͕ඞཁ
• CPU࢖༻཰ͷ૿ݮΛܖػʹ໨ඪ஋Λݟ௚͢
͜ͱ͕௥ै଎౓ͷ޲্ʹޮՌ͋Γ

View Slide

19
Kaburaya: DynamicTargetController
ภࠩ
໨ඪ஋
1
*
%

r
e

(PBQQ

ϑΟʔυόοΫ
4FNBQIPSF $166TBHF
มԽ཰
• ฒྻॲཧͷݶ౓ͱͳΔCPU࢖༻཰Λ࣮ߦ࣌ʹଊ͑ΔͨΊ໨ඪ஋ΛมԽ཰͔Βߋ
৽͢Δೋॏϧʔϓ੍ޚثΛ࣮૷ɽ
• ಺ଆͷϧʔϓ͸ฒߦ਺ΛೖྗɼCPU࢖༻཰Λग़ྗͱ͢ΔPID੍ޚث

View Slide

• ༧ߘʮKaburaya: CPUෛՙʹԠͯ͡ܧଓతʹ্ݶ஋Λ࠷దԽ͢ΔಈతηϚ
ϑΥʯͱͯ͠ୈ3ճWeb System Architectureݚڀձˏ౦ژ #wsaݚͰൃද
• https://blog.monochromegane.com/blog/2018/11/25/wsa_3_kaburaya/
• https://speakerdeck.com/monochromegane/wsa-3-kaburaya
20
༧ߘ

View Slide

3.
Goݴޠ΁ͷద༻

View Slide

22
• CPUό΢ϯυͳॲཧͱͯ͠ϑΟϘφον(35)ͱ(10)ΛҰఆճ਺ٻΊΔॲཧ࣌ؒΛܭଌ
• CPUΛ100%࢖͍੾ΔΑ͏ͳ৔߹͸ίΞ਺Ҏ্ͷฒߦ਺͸ੑೳ޲্͠ͳ͍
• CPUʹ༨༟͕͋Δ৔߹͸ίΞ਺Ҏ্ͷฒߦ਺Ͱ΋ੑೳ޲্ͷ༨஍͕͋Δ

View Slide

23
• ϒϩοΩϯάॲཧͱͯ͠Sleep΍NetworkΛҰఆճ਺ߦ͏ॲཧ࣌ؒΛܭଌ
• ඇৗʹ୹͍SleepͰ͸ίΞ਺Ҏ্ͷฒߦ਺Ͱ΋ੑೳ޲্ͷ༨஍͕͋Δ
• ωοτϫʔΫॲཧͰ͸଴ͪͷ௕͕͞ฒྻԽͷԸܙΛͳͨ͘͢Ίੑೳ޲্͕ฒߦ਺ʹґଘ

View Slide

24
• ϥϯλΠϜ͸ٖࣅతͳฒྻԽ΋͘͠͸ϚϧνίΞΛར༻͢ΔฒྻԽΛఏڙ͢Δ͕ɼδϣϒͷಛ
ੑʹΑͬͯɼλεΫॲཧ࣌ͷCPU࢖༻཰͸CPU্ݶ΋͘͠͸ಛఆͷ࣌఺ͰมԽ͠ͳ͘ͳΔ

View Slide

• λεΫͷಛੑɼϥϯλΠϜͷಛੑʹΑͬͯ࠷దͳฒߦ਺͸ҟͳΔɽ
• ฒྻॲཧͷ݁ՌɼCPU࢖༻཰͕CPU্ݶ·ͨ͸Ұఆ஋Ͱۉߧ͢Δɽ
25
ฒߦ਺ͷࢉग़
• KaburayaΞʔΩςΫνϟͷલఏʹԊ͏͜ͱ͔Βɼ੍ޚثͱͯ͠
DynamicTargetControllerΛར༻

View Slide

Goݴޠʹ͓͚ΔόοϑΝ෇͖νϟϯωϧΛ
ηϚϑΥͱ͢Δಉ࣮࣌ߦ਺੍ޚ 26
ฒߦ਺ͷ੍ޚ
func main() {
var wg sync.WaitGroup
sem := make(chan struct{}, 3)
for i := 0; i < 10; i++ {
wg.Add(1)
sem <- struct{}{}
go func() {
defer wg.Done()
defer func() { <-sem }()
task()
}()
}
wg.Wait()
close(sem)
}
• ηϚϑΥͷ্ݶ஋͕ಈతʹมߋՄೳͰ
͋Δ͜ͱ
• ௨ৗͷηϚϑΥͱಉ༷ʹPૢ࡞ʹ͓͍
ͯ஋͕ෛʹͳΔ৔߹ʹ࣮ߦ͕ϒϩοΫ
͞ΕΔ͜ͱ
• ͜ΕΒͷ஋ͷมߋ͕ΞτϛοΫʹߦΘ
ΕΔ͜ͱ
• ্ݶ஋ΛมߋՄೳͳόοϑΝ෇͖νϟ
ϯωϧ͕ඞཁ

View Slide

• ೖྗͱग़ྗͷೋͭͷνϟϯωϧͱΩϡʔΛ΋ͭߏ଄ମ
• Ωϡʔͷঢ়ଶʹԠͯ͡ೖྗͱग़ྗͷՄ൱Λ੍ޚ͢Δ
• Ref: https://github.com/npat-efault/musings/wiki/Elastic-channels
• Kaburaya༻్Ͱ͸Ωϡʔͷ஋Λར༻͢Δ͜ͱ͸ͳ͍ͷͰΧ΢ϯλͰ୅༻
27
όοϑΝ্ݶ஋ΛมߋՄೳͳElasticChannel
DIBO DIBO
&MBTUJD$IBOOFM
2VFVF
&NQUZ*OPOMZ
'VMM0VUPOMZ
0UIFS*O0VU
$IBOHFMJNJU

View Slide

• Kaburaya optimize the number of goroutines by feedback control.
28
monochromegane/kaburaya
func main() {
sem := kaburaya.NewSem(100 * time.Millisecond)
var wg sync.WaitGroup
for i := 0; i < 10; i++ {
wg.Add(1)
sem.Wait()
go func() {
defer sem.Signal()
defer wg.Done()
task()
}()
}
wg.Wait()
sem.Stop()
}
Create semaphore and
specify time span for
update the semaphore and
CPU target.
Wait for decrement operation.
Increment operation.
Stop update the semaphore and CPU target.

View Slide

4.
ධՁ

View Slide

• ߴ଎Խ཰(Speedup): ஞ࣮࣍ߦ࣌ؒͱฒྻ࣮ߦ࣌ؒͷൺ཰
• ࣮ߦޮ཰(Efﬁciency): ฒߦ਺͋ͨΓͷߴ଎Խ཰ɽܭࢉࢿݯͷޮ཰ੑࢦඪɽ
30
࠷దͳฒߦ਺ͱ͸
$ ./score fib 40 64
[S] 41.20142734s
[1] 40.94367043s -> Speedup: 1.006301 Efficiency: 1.006301
• ͳ͓ɼKaburayaͰ͸ฒߦ਺͕૿ݮ͠ɼಛੑ্֎Ε஋ؚ͕·ΕΔ͜ͱ͔Βɼ࣮
ߦޮ཰ͷฒߦ਺ʹ͸ɼॲཧظؒͷฒߦ਺ͷதԝ஋Λ༻͍Δ

View Slide

network(1) * 8192
31
$ ./score network 1 8192 2> record.txt
[S] 18.906590961s
[16] 920.629201ms -> Speedup: 20.550000 Efficiency: 1.284375
[K] 738.116568ms -> Speedup: 25.617886 Efficiency: 0.711608
• K{p,i,d}=0.25ɼDuration=50ms
• CPU্ݶʹୡ͠ͳ͍λεΫʹରͯ͠ൺֱతଥ౰ͳฒߦ਺ͷ੍ޚ͕ߦ͑ͨɽ

View Slide

ﬁb(40) * 64
32
$ ./score fib 40 64 2> record.txt
[S] 42.498952641s
[K] 13.368050152s -> Speedup: 3.179084 Efficiency: 0.635817
• CPU্ݶʹୡ͢ΔλεΫʹରͯ͠ൺֱతଥ౰ͳฒߦ਺ͷ੍ޚ͕ߦ͑ͨɽ
• λεΫ׬ྃ·ͰͷσΟϨΠʹΑͬͯηϚϑΥ஋ͷ௿ݮ͕ਐΈ͗͢Δ܏޲͋Γ

View Slide

ﬁb(15) * 3276800
33
$ ./score fib 15 3276800 2> record.txt
[S] 13.069192052s
[K] 15.535130172s -> Speedup: 0.841262 Efficiency: 0.052579
• CPUෛՙͷগͳ͍େྔͷλεΫʹରͯ͠͸ߴ଎Խ཰ɼ࣮ߦޮ཰ͷ྆໘Ͱࣦഊ
• ݱ࣮૷Ͱ͸ElasticChannelͷWaitૢ࡞Ͱbuffer0ͷChannel͕ϘτϧωοΫ

View Slide

• ϑΟʔυόοΫ੍ޚҰൠʹର͢Δ՝୊
• ήΠϯ(K{p,i,d})ͷ࠷ద஋ΛٻΊΔඞཁ͕͋Δ
• σΟϨΠʹΑΔௐ੔ޡࠩΛݮΒ͢ඞཁ͕͋Δ
• ϑΟʔυόοΫ੍ޚ෼໺ͷϊ΢ϋ΢Λ༗ޮ׆༻ͯ͠ղܾ͍ͨ͠
• ࣮૷ʹର͢Δ՝୊
• GoݴޠʹΑΔಈతηϚϑΥͷੑೳ໰୊Λղܾ͢Δඞཁ͕͋Δ
• System V semaphoreͳͲ௿ϨΠϠ࣮૷Ͱղܾ͍ͨ͠
34
՝୊ͱରԠࢦ਑

View Slide

5.
·ͱΊ

View Slide

• ॲཧ΍ϥϯλΠϜͷಛੑʹґଘͤͣʹɼϚγϯͷෛՙʹԠͯ͡൓Ԡత͔ͭܧଓ
తʹɼฒߦ਺Λ࠷దԽ͢Δख๏ͱͯ͠ɼCPUෛՙʹԠͯ͡ܧଓతʹ্ݶ஋Λ࠷
దԽ͢ΔKaburayaΞʔΩςΫνϟΛGoݴޠʹద༻
• ฒߦ਺ͷ૿Ճ͕ੑೳʹد༩͢ΔλεΫʹ͓͍ͯ͸ɼଥ౰ͳฒߦ਺Λࢉग़Մೳ
• ࠓޙ͸՝୊Λղܾͯ͠ਫ਼౓ͱ଎౓Λ޲্ͤ͞ɼݴޠΛ௒͑ͨ൚༻ੑΛ໨ࢦ͢
• Go
36
·ͱΊ

View Slide

View Slide

フィードバック制御によるGoroutine起動数の最適化 / optimization-of-goroutine-numbers-by-feedback-control

フィードバック制御によるGoroutine起動数の最適化 / optimization-of-goroutine-numbers-by-feedback-control

monochromegane

More Decks by monochromegane

Other Decks in Programming

Featured

Transcript