Kuna kan kira don sabis na yanar gizo: load balancer a gaban rundunar na'urori, cache a gaban database, da manufar samuwa ta 99.9 %. Kowane minti samfurin yana lissafin jinkirin layi, ƙarewar lokaci, cache hits da nauyin database daga dabaru na littafin karatu — da abin da shawarwarinku ke kashewa.
Ƙasan rundunar na'urori. Ɗaga shi yana ƙaddamar da na'urori nan take — har yanzu suna buƙatar jinkirin tashi kafin su yi hidima.
Bin manufa a kan amfani; babu sabon ƙara waje yayin da na'urori har yanzu suna tashi. A kashe = daidai mafi ƙaranci.
Ƙasa = ƙarin fili da ƙarin kuɗi. Amfani da aka auna ba zai iya wuce 100 % ba, don haka rundunar da ta cika tana girma mataki-mataki kawai.
TTL mai tsawo = ƙarin hits, amma amsoshi na iya zama tsofaffi (matsakaicin shekaru ≈ TTL/2).
Yana loda maɓallan masu zafi tsawon mintuna 6 (+12 % na cache a minti) a farashin tambayoyin database 600 na ƙari a daƙiƙa.
Kason 30 % mai ƙarancin fifiko (prefetch, batch, crawlers) da aka ƙi a load balancer da "a sake gwadawa daga baya".
Fasali mai nauyi yana ƙara 20 ms na CPU da tambayar database ɗaya ga kowace buƙata. A kashe = raguwa mai laushi.
Yana sake tura ginin ƙarshe mai kyau a sabon rukunin na'urori (mintuna 5, ana biyan kuɗi sau biyu), sannan ya canja zirga-zirga. Danna sau biyu yana sake fara shiri; idan babu mummunan gini mai aiki, kuɗi kawai yake kashewa.
| Ƙimar ƙonewa (awa 1) | 0.0 × |
|---|---|
| Buƙatu | 1125 req/s |
| Na'urori masu yin hidima | 10 |
| Na'urori masu tashi | 0 |
| Ƙimar cache hit | 77 % |
| Amfani da database | 19 % |
| Zirga-zirgar da aka zubar | 0 % |
| Kuɗin runduna | 4.00 $/h |
| Kuɗin da aka kashe zuwa yanzu | 0.00 $ |
| Matsakaicin shekarun amsoshin da aka ajiye a cache | 30 s |
| Zirga-zirga a mummunan ginin | 0 % |
| Shawarwari da ake da su | 100 % |
Sabon gini ya fito a 09:10. Watan ya riga ya yi wahala: kashi 20 % kawai na kasafin kuskure ya rage. Mintuna bayan turawa faɗakarwar ƙimar ƙonewa tana ƙara. Ku kare kasafin.
Wata hanyar haɗi zuwa sabis ɗin tana yaɗuwa da sauri kuma ana tsammanin tsalle na zirga-zirga wani lokaci a safiyar nan — babu wanda ya san lokacin, ko girmansa. Sababbin na'urori suna buƙatar minti 8 don farawa a yau. Idan ya zo, riƙe kurakurai da jinkiri ƙasa ba tare da ƙona kuɗi kan ƙarfin da ba a amfani da shi ba.
A kololuwar rana rubutun gyara ya share dukan cache. Yanzu kowace buƙata tana tafiya zuwa database, wanda aka auna shi don ƙimar hit ta yau da kullum ta 85 %. Ku dawo da sabis ba tare da ɗora wa database nauyi fiye da kima ba.
Kowace alaƙa da na'urar koyi ke amfani da ita, tare da tushenta. Ƙayyadaddun ƙimomin da aka yi wa alama a matsayin zato daidaitawa ne na misali.
λ(t) = base × (1 + 0.25·sin(2π(t + clock − 6 h)/24 h)) × surge(t), clock = time of day at the start (peak at 12:00); count/min ≈ N(60λ, √(60λ)) × (1 + N(0, 0.02))[6]Zato: yawan workers, lokutan hidima, ƙarfin database, girman cache, saurin cikawa, farashi da ƙimar kuskuren mummunan ginin ƙimomi ne na misali don matsakaicin sabis na yanar gizo.N = instances × 16 workers, a = λ·S; C(a, N) = B / (1 − (a/N)(1 − B)), B = Erlang B[3][6]P(W > t) = C·e^(−(N/S − λ)·t); timeouts = P(W > 2 s); a ≥ N ⇒ failed share = 1 − N/a[3]p99 ≈ S·ln 100 + ln(C/0.01)/(N/S − λ) (service + waiting quantile, an approximation)[3][6]Kimantawa: jimlar quantile na hidima da jira ba ainihin p99 na jimlarsu ba ne (yana iya zama ɗan sama ko ƙasa); ana ɗaukar jerin gwano a matsayin mai tsayayye a cikin kowane minti saboda buƙatu suna ɗaukar millisecond.busy workers L = λ·S ⇒ utilization = λ·S / N[4]hit = warm × rT/(1 + rT), r = λ / 20,000 objects; mean age of a cached answer ≈ T/2[5]DB load = λ·q·(1 − hit); query time = 5 ms/(1 − ρ_db) (≤ 250 ms); S = S_app + (1 − hit)·q·query time[6]Zato: yawan workers, lokutan hidima, ƙarfin database, girman cache, saurin cikawa, farashi da ƙimar kuskuren mummunan ginin ƙimomi ne na misali don matsakaicin sabis na yanar gizo.budget = 1 − SLO = 0.1 %; burn = error rate / 0.1 %; Δbudget per min = burn / 43,200; burn (1 h) = mean error rate over the last 60 min / 0.1 % (window pre-filled with the opening minute)[1][2]desired = ⌈serving × utilization / target⌉ (utilization saturates at 100 %); new instances serve after the boot delay[6]Zato: yawan workers, lokutan hidima, ƙarfin database, girman cache, saurin cikawa, farashi da ƙimar kuskuren mummunan ginin ƙimomi ne na misali don matsakaicin sabis na yanar gizo.16 workers/instance · app time 50 ms (+20 ms and +1 query with the feature on) · 2 queries/request · DB 4,000 queries/s · 20,000 hot objects · cache refill τ = 30 min (slower while the DB is saturated) · warm-up job +12 %/min for 6 min, +600 queries/s · timeout 2 s · 30 % low-priority traffic · bad build +5 % errors, ×1.25 CPU · rollback 5 min · $0.40 per instance-hour · scale-in by ≤ 20 % of the fleet after 10 quiet minutes · up to 40 instancesZato: yawan workers, lokutan hidima, ƙarfin database, girman cache, saurin cikawa, farashi da ƙimar kuskuren mummunan ginin ƙimomi ne na misali don matsakaicin sabis na yanar gizo.Rashin tsari: mai samar da mulberry32 mai iri; rarrabawar da aka yi amfani da su — uniform, exponential (inverse CDF), normal (Box–Muller), Poisson (Knuth). Ana nuna iri kuma ana iya raba shi.
Samfurin ilimi — ba don yanke shawarar aiki ba. Wuraren gaske suna daidaita kowace ƙayyadaddiyar ƙima da kayan aiki da bayanansu.