Kubernetes k\u00fcmelerinde kaynak y\u00f6netimi ve maliyet optimizasyonu, modern bulut altyap\u0131lar\u0131n\u0131n en kritik bile\u015fenlerinden biri haline geldi. Amazon EKS (Elastic Kubernetes Service) \u00fczerinde dinamik ve verimli bir node sa\u011flama mekanizmas\u0131 arayanlar i\u00e7in Karpenter, Terraform ile entegre edildi\u011finde g\u00fc\u00e7l\u00fc ve d\u00f6n\u00fc\u015ft\u00fcr\u00fcc\u00fc bir \u00e7\u00f6z\u00fcm sunar. Bu makalede, Karpenter’\u0131n ne oldu\u011funu, EKS ortam\u0131nda neden bu kadar \u00f6nemli oldu\u011funu ve altyap\u0131y\u0131 kod olarak y\u00f6netme (Infrastructure as Code – IaC) yakla\u015f\u0131m\u0131n\u0131n bir par\u00e7as\u0131 olarak Terraform kullanarak ad\u0131m ad\u0131m nas\u0131l kurulaca\u011f\u0131n\u0131 derinlemesine inceleyece\u011fiz. Amac\u0131m\u0131z, hem yeni ba\u015flayanlar\u0131n konuyu temelden anlamas\u0131n\u0131 sa\u011flamak hem de deneyimli kullan\u0131c\u0131lar i\u00e7in pratik kurulum ad\u0131mlar\u0131 ve geli\u015fmi\u015f ipu\u00e7lar\u0131 sunmakt\u0131r.<\/p>\n
G\u00fcn\u00fcm\u00fcz\u00fcn h\u0131zla de\u011fi\u015fen dijital d\u00fcnyas\u0131nda, uygulamalar\u0131n anl\u0131k taleplere g\u00f6re \u00f6l\u00e7eklenebilmesi hayati \u00f6nem ta\u015f\u0131r. Geleneksel y\u00f6ntemlerle bu \u00f6l\u00e7eklenebilirli\u011fi sa\u011flamak hem maliyetli hem de operasyonel olarak karma\u015f\u0131k olabilir. \u0130\u015fte tam da bu noktada Karpenter gibi ak\u0131ll\u0131 bir node sa\u011flay\u0131c\u0131 devreye giriyor. \u00d6zellikle AWS ekosisteminde, EKS k\u00fcmelerinizdeki pod’lar\u0131n ihtiya\u00e7lar\u0131na g\u00f6re en uygun EC2 instance’lar\u0131n\u0131 otomatik ve h\u0131zl\u0131 bir \u015fekilde temin ederek, hem performans art\u0131\u015f\u0131 hem de \u00f6nemli \u00f6l\u00e7\u00fcde maliyet tasarrufu vaat ediyor. Terraform ile bu kurulumu otomatize etmek ise, altyap\u0131n\u0131z\u0131n tutarl\u0131l\u0131\u011f\u0131n\u0131 ve y\u00f6netilebilirli\u011fini art\u0131rarak size b\u00fcy\u00fck bir kolayl\u0131k sunar.<\/p>\n
Karpenter, AWS taraf\u0131ndan geli\u015ftirilen, Kubernetes k\u00fcmeleri i\u00e7in y\u00fcksek performansl\u0131, esnek ve maliyet optimize edilmi\u015f bir node sa\u011flay\u0131c\u0131d\u0131r. Geleneksel Kubernetes Cluster Autoscaler (K\u00fcme Otomatik \u00d6l\u00e7eklendiricisi) ile kar\u015f\u0131la\u015ft\u0131r\u0131ld\u0131\u011f\u0131nda, Karpenter’\u0131n \u00e7al\u0131\u015fma mant\u0131\u011f\u0131 ve sundu\u011fu avantajlar onu \u00f6ne \u00e7\u0131kar\u0131r. Cluster Autoscaler, pod’lar\u0131n planlanamad\u0131\u011f\u0131 durumlarda mevcut node gruplar\u0131ndaki node say\u0131s\u0131n\u0131 art\u0131rarak \u00e7al\u0131\u015f\u0131r. Bu, genellikle \u00f6nceden tan\u0131mlanm\u0131\u015f node gruplar\u0131na ba\u011fl\u0131d\u0131r ve yeni bir node’un ba\u015flat\u0131lmas\u0131 i\u00e7in belirli bir bekleme s\u00fcresi gerektirebilir. Ayr\u0131ca, farkl\u0131 i\u015f y\u00fckleri i\u00e7in farkl\u0131 node gruplar\u0131 tan\u0131mlamak operasyonel karma\u015f\u0131kl\u0131\u011f\u0131 art\u0131r\u0131r. Karpenter ise bu s\u00fcreci \u00e7ok daha dinamik ve verimli bir \u015fekilde ele al\u0131r.<\/p>\n
Karpenter’\u0131n temel fark\u0131, pod’lar\u0131n bekleyen durumlar\u0131n\u0131 do\u011frudan g\u00f6zlemlemesi ve bu pod’lar\u0131n gereksinimlerini (CPU, bellek, GPU, depolama, mimari vb.) kar\u015f\u0131layacak en uygun EC2 instance’\u0131n\u0131 AWS API’leri arac\u0131l\u0131\u011f\u0131yla an\u0131nda sa\u011flamas\u0131d\u0131r. Bu “tam zaman\u0131nda (just-in-time)” provisioning (kaynak sa\u011flama) yakla\u015f\u0131m\u0131, node’lar\u0131n gereksiz yere bekletilmesinin veya yanl\u0131\u015f boyutland\u0131r\u0131lmas\u0131n\u0131n \u00f6n\u00fcne ge\u00e7er. \u00d6rne\u011fin, bir pod sadece 1 CPU ve 2GB bellek istiyorsa, Karpenter tam da bu gereksinimleri kar\u015f\u0131layacak en k\u00fc\u00e7\u00fck ve uygun maliyetli instance’\u0131 bulmaya \u00e7al\u0131\u015f\u0131r. Bu, \u00f6zellikle Spot Instance’lar (kesintiye u\u011frayabilen, daha uygun maliyetli EC2 instance’lar\u0131) ve farkl\u0131 mimariler (Graviton gibi) kullan\u0131ld\u0131\u011f\u0131nda b\u00fcy\u00fck maliyet avantajlar\u0131 sa\u011flar.<\/p>\n
Karpenter’\u0131n sundu\u011fu ba\u015fl\u0131ca faydalar \u015funlard\u0131r:<\/p>\n
\u00d6zetle, Karpenter, EKS k\u00fcmelerinizde otomasyonu bir ad\u0131m \u00f6teye ta\u015f\u0131yarak, daha \u00e7evik, daha maliyet-etkin ve daha az operasyonel y\u00fckle y\u00f6netilebilir bir altyap\u0131 sunar. Geleneksel Cluster Autoscaler’\u0131n aksine, node gruplar\u0131na ba\u011fl\u0131 kalmadan do\u011frudan pod taleplerine yan\u0131t vermesi, onu modern Kubernetes ortamlar\u0131 i\u00e7in vazge\u00e7ilmez bir ara\u00e7 haline getirir.<\/p>\n
Amazon EKS, i\u015fletmelere esneklik ve \u00f6l\u00e7eklenebilirlik sunsa da, kaynak y\u00f6netimi ve maliyet kontrol\u00fc her zaman bir meydan okuma olmu\u015ftur. Karpenter, EKS ortam\u0131n\u0131n do\u011fas\u0131nda bulunan bu zorluklara do\u011frudan yan\u0131t vererek, bulut altyap\u0131n\u0131zdan en iyi \u015fekilde yararlanman\u0131z\u0131 sa\u011flar. \u00d6zellikle dinamik i\u015f y\u00fcklerine sahip uygulamalar i\u00e7in Karpenter’\u0131n sa\u011flad\u0131\u011f\u0131 avantajlar paha bi\u00e7ilmezdir.<\/p>\n
Karpenter’\u0131n EKS’teki en b\u00fcy\u00fck de\u011fer \u00f6nerilerinden biri, maliyet optimizasyonudur. Geleneksel y\u00f6ntemlerde, genellikle en k\u00f6t\u00fc senaryoyu d\u00fc\u015f\u00fcnerek fazla kaynak ay\u0131rma e\u011filimi vard\u0131r. Karpenter ise bu israf\u0131 ortadan kald\u0131r\u0131r:<\/p>\n
Maliyetin yan\u0131 s\u0131ra, performans ve h\u0131z da Karpenter’\u0131n EKS’e katt\u0131\u011f\u0131 \u00f6nemli de\u011ferlerdir:<\/p>\n
Karpenter, EKS’teki operasyonel karma\u015f\u0131kl\u0131\u011f\u0131 azaltarak DevOps ekiplerinin hayat\u0131n\u0131 kolayla\u015ft\u0131r\u0131r:<\/p>\n
Ger\u00e7ek D\u00fcnya Senaryosu: E-ticaret Sitesinin Anl\u0131k Kampanya D\u00f6nemleri<\/strong><\/p>\n Bir e-ticaret sitesinin Kara Cuma veya Sevgililer G\u00fcn\u00fc gibi \u00f6zel kampanya d\u00f6nemlerinde trafik yo\u011funlu\u011funda %500’e varan art\u0131\u015flar ya\u015fad\u0131\u011f\u0131n\u0131 d\u00fc\u015f\u00fcnelim. Geleneksel bir EKS kurulumunda, bu ani y\u00fck art\u0131\u015f\u0131n\u0131 kar\u015f\u0131lamak i\u00e7in ya \u00f6nceden \u00e7ok fazla kaynak ay\u0131rmak (ve kampanya d\u0131\u015f\u0131nda bo\u015fta b\u0131rakmak) ya da Cluster Autoscaler’\u0131n yava\u015f \u00f6l\u00e7eklenmesi nedeniyle performans sorunlar\u0131 ya\u015famak gerekebilir. Karpenter ile bu sorun ortadan kalkar. Kampanya ba\u015flad\u0131\u011f\u0131nda, yeni gelen pod’lar i\u00e7in Karpenter an\u0131nda uygun Spot Instance’lar\u0131 sa\u011flayarak, sitenin kesintisiz ve y\u00fcksek performansla \u00e7al\u0131\u015fmas\u0131n\u0131 garanti eder. Kampanya bitti\u011finde ise, kullan\u0131lmayan node’lar\u0131 otomatik olarak sonland\u0131rarak maliyetleri optimize eder. Bu sayede, hem m\u00fc\u015fteri memnuniyeti artar hem de gereksiz bulut harcamalar\u0131n\u0131n \u00f6n\u00fcne ge\u00e7ilir.<\/p>\n Karpenter’\u0131 Amazon EKS k\u00fcmenize entegre etmek, do\u011fru bir planlama ve \u00f6n haz\u0131rl\u0131k gerektirir. Kuruluma ba\u015flamadan \u00f6nce belirli \u00f6n ko\u015fullar\u0131 anlamak ve gerekli ara\u00e7lar\u0131 haz\u0131rlamak, sorunsuz bir ge\u00e7i\u015f i\u00e7in kritik \u00f6neme sahiptir. Terraform ile bu s\u00fcreci otomatize etmek, altyap\u0131n\u0131z\u0131n tutarl\u0131l\u0131\u011f\u0131n\u0131 ve y\u00f6netilebilirli\u011fini sa\u011flarken, ayn\u0131 zamanda tekrarlanabilir bir da\u011f\u0131t\u0131m modeli sunar.<\/p>\n Karpenter’\u0131n EKS \u00fczerinde verimli bir \u015fekilde \u00e7al\u0131\u015fabilmesi i\u00e7in baz\u0131 temel gereksinimler bulunmaktad\u0131r:<\/p>\n Terraform, Karpenter’\u0131n EKS \u00fczerindeki kurulumunu otomatize etmek i\u00e7in ideal bir ara\u00e7t\u0131r. Altyap\u0131y\u0131 Kod Olarak (IaC) y\u00f6netme yakla\u015f\u0131m\u0131n\u0131 benimseyerek bir\u00e7ok avantaj sunar:<\/p>\n Kuruluma ba\u015flamadan \u00f6nce sisteminizde a\u015fa\u011f\u0131daki ara\u00e7lar\u0131n y\u00fckl\u00fc ve yap\u0131land\u0131r\u0131lm\u0131\u015f oldu\u011fundan emin olun:<\/p>\n Bu \u00f6n ko\u015fullar\u0131 yerine getirerek ve gerekli ara\u00e7lar\u0131 haz\u0131rlayarak, Karpenter’\u0131 EKS k\u00fcmenize Terraform ile ba\u015far\u0131l\u0131 bir \u015fekilde entegre etmek i\u00e7in sa\u011flam bir temel olu\u015fturmu\u015f olursunuz. Bir sonraki b\u00f6l\u00fcmde, bu ad\u0131mlar\u0131 pratik kod \u00f6rnekleriyle detayland\u0131raca\u011f\u0131z.<\/p>\n Karpenter’\u0131 EKS k\u00fcmenize Terraform ile entegre etmek, birka\u00e7 ana ad\u0131mdan olu\u015fur: gerekli IAM rollerini ve politikalar\u0131n\u0131 olu\u015fturmak, VPC ve subnet’lerinizi etiketlemek, Karpenter Helm chart’\u0131n\u0131 da\u011f\u0131tmak ve son olarak Karpenter Provisioner kaynaklar\u0131n\u0131 tan\u0131mlamak. Bu ad\u0131mlar\u0131 s\u0131ras\u0131yla uygulayarak, otomatik \u00f6l\u00e7eklenen ve maliyet-etkin bir Kubernetes altyap\u0131s\u0131 kurabilirsiniz.<\/p>\n Bu b\u00f6l\u00fcmde, ad\u0131m ad\u0131m Terraform kod bloklar\u0131 ile Karpenter kurulumunu ger\u00e7ekle\u015ftirece\u011fiz. L\u00fctfen Karpenter’\u0131n AWS kaynaklar\u0131yla etkile\u015fim kurabilmesi i\u00e7in iki ana IAM rol\u00fcne ihtiyac\u0131m\u0131z var: Karpenter Controller i\u00e7in bir rol ve Karpenter taraf\u0131ndan ba\u015flat\u0131lacak EC2 node’lar\u0131 i\u00e7in bir instance profili. Bu rol ve profiller, gerekli izinleri i\u00e7ermelidir.<\/p>\n Yukar\u0131daki kod blo\u011fu, Karpenter Controller i\u00e7in bir IAM rol\u00fc ve bir de Karpenter taraf\u0131ndan ba\u015flat\u0131lacak EC2 instance’lar\u0131 i\u00e7in bir IAM rol\u00fc (instance profile) tan\u0131mlar. Controller rol\u00fc, OIDC sa\u011flay\u0131c\u0131s\u0131 \u00fczerinden Kubernetes Service Account’\u0131 ile ili\u015fkilendirilirken, node rol\u00fc EC2 instance’lar\u0131n\u0131n \u00fczerine atan\u0131r.<\/p>\n Karpenter’\u0131n do\u011fru VPC ve subnet’leri bulabilmesi i\u00e7in bu kaynaklara belirli etiketler eklememiz gerekiyor. Bu etiketler, Karpenter’\u0131n a\u011f yap\u0131land\u0131rman\u0131z\u0131 otomatik olarak ke\u015ffetmesini sa\u011flar.<\/p>\n Bu etiketler, Karpenter’\u0131n hangi VPC ve subnet’leri kullanarak EC2 instance’lar\u0131n\u0131 ba\u015flataca\u011f\u0131n\u0131 belirlemesine yard\u0131mc\u0131 olur. \u00d6zellikle Karpenter, Kubernetes k\u00fcmesine bir Helm chart olarak da\u011f\u0131t\u0131l\u0131r. Terraform’un Helm provider’\u0131n\u0131 kullanarak bu da\u011f\u0131t\u0131m\u0131 otomatize edebiliriz. Bu ad\u0131mda, \u00f6nceki ad\u0131mlarda olu\u015fturdu\u011fumuz IAM rol\u00fcn\u00fc ve di\u011fer k\u00fcme bilgilerini Helm chart’\u0131na parametre olarak ge\u00e7irece\u011fiz.<\/p>\n Bu Helm chart da\u011f\u0131t\u0131m\u0131, Karpenter Controller’\u0131 EKS k\u00fcmenize kurar ve Controller’\u0131n hangi IAM rol\u00fcn\u00fc kullanaca\u011f\u0131n\u0131, hangi k\u00fcme ad\u0131 i\u00e7in \u00e7al\u0131\u015ft\u0131\u011f\u0131n\u0131 ve node’lar\u0131 ba\u015flat\u0131rken hangi instance profilini kullanaca\u011f\u0131n\u0131 belirtir.<\/p>\n Karpenter’\u0131n nas\u0131l node sa\u011flayaca\u011f\u0131n\u0131 belirleyen ana konfig\u00fcrasyon, bir Custom Resource Definition (CRD) olan Bu T\u00fcm bu Terraform dosyalar\u0131n\u0131 olu\u015fturduktan sonra, projenizin k\u00f6k dizininde a\u015fa\u011f\u0131daki komutlar\u0131 \u00e7al\u0131\u015ft\u0131rarak Karpenter’\u0131 EKS k\u00fcmenize da\u011f\u0131tabilirsiniz:<\/p>\n Bu ad\u0131mlar tamamland\u0131\u011f\u0131nda, Karpenter EKS k\u00fcmenizde etkinle\u015ftirilmi\u015f ve pod’lar\u0131n\u0131z\u0131n ihtiya\u00e7lar\u0131na g\u00f6re dinamik olarak node sa\u011flamaya haz\u0131r hale gelmi\u015f olacakt\u0131r.<\/p>\n Karpenter’\u0131n kurulumunu tamamlad\u0131ktan sonra, sistemin beklendi\u011fi gibi \u00e7al\u0131\u015ft\u0131\u011f\u0131ndan emin olmak i\u00e7in test etmek ve do\u011frulamak \u00f6nemlidir. Bu b\u00f6l\u00fcm, Karpenter’\u0131n otomatik \u00f6l\u00e7eklenme yeteneklerini nas\u0131l g\u00f6zlemleyece\u011finizi ve node azaltma (deprovisioning) mekanizmas\u0131n\u0131n nas\u0131l \u00e7al\u0131\u015ft\u0131\u011f\u0131n\u0131 a\u00e7\u0131klar.<\/p>\n Karpenter’\u0131n \u00e7al\u0131\u015f\u0131p \u00e7al\u0131\u015fmad\u0131\u011f\u0131n\u0131 test etmenin en iyi yolu, mevcut node’lar\u0131n\u0131zda yer bulamayan, y\u00fcksek kaynak talebi olan bir Kubernetes Deployment (Da\u011f\u0131t\u0131m) olu\u015fturmakt\u0131r. Bu, Karpenter’\u0131n yeni bir node ba\u015flatmas\u0131n\u0131 tetikleyecektir.<\/p>\n \u00d6ncelikle, k\u00fcmenizde yeterli kaynak olmad\u0131\u011f\u0131ndan emin olun. Ard\u0131ndan, a\u015fa\u011f\u0131daki gibi bir Deployment tan\u0131m\u0131 olu\u015fturun. Bu \u00f6rnek, her biri 1 CPU ve 1GB bellek isteyen 50 adet NGINX pod’u ba\u015flatmaya \u00e7al\u0131\u015facakt\u0131r. E\u011fer k\u00fcmenizde bu kadar kayna\u011f\u0131 kar\u015f\u0131layacak bo\u015f bir node yoksa, Karpenter devreye girecektir.<\/p>\n Bu Deployment’\u0131 k\u00fcmenize uygulay\u0131n:<\/p>\n \u015eimdi Karpenter’\u0131n davran\u0131\u015f\u0131n\u0131 g\u00f6zlemleyin:<\/p>\n Baz\u0131 pod’lar\u0131n Karpenter’\u0131n yeni EC2 instance’lar\u0131n\u0131 ba\u015flat\u0131p Kubernetes k\u00fcmesine ekledi\u011fini g\u00f6rmelisiniz. Yeni node’lar genellikle Bu komut, Karpenter Controller’\u0131n loglar\u0131n\u0131 g\u00f6sterir. Loglarda “provisioning” veya “launching instance” gibi mesajlar g\u00f6rmelisiniz, bu da Karpenter’\u0131n aktif olarak \u00e7al\u0131\u015ft\u0131\u011f\u0131n\u0131 g\u00f6sterir.<\/p>\n<\/li>\n<\/ul>\n Karpenter, sadece \u00f6l\u00e7eklenmekle kalmaz, ayn\u0131 zamanda kaynak kullan\u0131m\u0131 d\u00fc\u015ft\u00fc\u011f\u00fcnde gereksiz node’lar\u0131 ak\u0131ll\u0131ca sonland\u0131rarak maliyet tasarrufu sa\u011flar. Bu s\u00fcreci g\u00f6zlemlemek i\u00e7in olu\u015fturdu\u011fumuz Deployment’\u0131 silebiliriz:<\/p>\n Deployment silindikten sonra, Karpenter’\u0131n node’lar\u0131 nas\u0131l bo\u015faltt\u0131\u011f\u0131n\u0131 ve sonland\u0131rd\u0131\u011f\u0131n\u0131 g\u00f6zlemleyin:<\/p>\n Karpenter’\u0131n ba\u015flatt\u0131\u011f\u0131 node’lar\u0131n bir s\u00fcre sonra Bu testler, Karpenter’\u0131n hem yukar\u0131 (scale-up) hem de a\u015fa\u011f\u0131 (scale-down) y\u00f6nde otomatik \u00f6l\u00e7eklenme yeteneklerini ba\u015far\u0131yla do\u011frulaman\u0131za yard\u0131mc\u0131 olacakt\u0131r. Art\u0131k EKS k\u00fcmeniz, dinamik i\u015f y\u00fcklerine \u00e7ok daha esnek ve maliyet-etkin bir \u015fekilde yan\u0131t verebilir.<\/p>\n Karpenter’\u0131n temel kurulumu ve do\u011frulamas\u0131 \u00f6nemli bir ad\u0131m olsa da, ger\u00e7ek d\u00fcnya senaryolar\u0131nda daha karma\u015f\u0131k ihtiya\u00e7lar\u0131 kar\u015f\u0131lamak i\u00e7in geli\u015fmi\u015f \u00f6zelliklerini kullanmak, maliyet ve performans avantajlar\u0131n\u0131 daha da art\u0131rabilir. Bu b\u00f6l\u00fcmde, deneyimli kullan\u0131c\u0131lar i\u00e7in Karpenter’\u0131n daha ileri d\u00fczey kullan\u0131m senaryolar\u0131n\u0131 ve ipu\u00e7lar\u0131n\u0131 ele alaca\u011f\u0131z.<\/p>\n Tek bir Karpenter’\u0131n maliyet optimizasyon yeteneklerinden tam olarak yararland\u0131\u011f\u0131n\u0131zdan emin olmak i\u00e7in izleme ve analiz ara\u00e7lar\u0131n\u0131 kullanmak \u00f6nemlidir:<\/p>\n Vaka Analizi: B\u00fcy\u00fck Bir Veri \u0130\u015fleme Platformunun Maliyet Optimizasyonu<\/strong><\/p>\n B\u00fcy\u00fck bir veri analizi \u015firketi, g\u00fcnde terabaytlarca veriyi i\u015fleyen Spark tabanl\u0131 bir platformu EKS \u00fczerinde \u00e7al\u0131\u015ft\u0131r\u0131yordu. Geleneksel Cluster Autoscaler ile, Spark i\u015flerinin ani kaynak talepleri nedeniyle genellikle b\u00fcy\u00fck ve pahal\u0131 On-Demand node’lar \u00f6nceden sa\u011flanmak zorunda kal\u0131n\u0131yordu. Bu da \u00f6nemli \u00f6l\u00e7\u00fcde bo\u015fta kaynak ve y\u00fcksek bulut faturalar\u0131 anlam\u0131na geliyordu. \u015eirket, Karpenter’\u0131 devreye alarak, Spark i\u015flerinin taleplerine g\u00f6re an\u0131nda Spot Instance’lar\u0131 temin etmeye ba\u015flad\u0131. Karpenter’\u0131n ak\u0131ll\u0131 konsolidasyon mekanizmas\u0131 sayesinde, i\u015fler tamamland\u0131\u011f\u0131nda node’lar h\u0131zla sonland\u0131r\u0131ld\u0131. Sonu\u00e7 olarak, \u015firketin EC2 maliyetlerinde %30’un \u00fczerinde bir d\u00fc\u015f\u00fc\u015f ya\u015fand\u0131 ve Spark i\u015flerinin ba\u015flang\u0131\u00e7 s\u00fcreleri de \u00f6nemli \u00f6l\u00e7\u00fcde k\u0131sald\u0131. Bu, Karpenter’\u0131n hem maliyet hem de performans a\u00e7\u0131s\u0131ndan ne kadar d\u00f6n\u00fc\u015ft\u00fcr\u00fcc\u00fc olabilece\u011finin somut bir \u00f6rne\u011fidir.<\/p>\n Karpenter’\u0131 Amazon EKS k\u00fcmenize Terraform ile entegre etmek, bulut altyap\u0131n\u0131z\u0131n y\u00f6netimini k\u00f6kten de\u011fi\u015ftiren g\u00fc\u00e7l\u00fc bir ad\u0131md\u0131r. Bu makalede, Karpenter’\u0131n ne oldu\u011funu, geleneksel otomatik \u00f6l\u00e7eklendirme \u00e7\u00f6z\u00fcmlerinden farklar\u0131n\u0131, EKS ortam\u0131nda neden bu kadar de\u011ferli oldu\u011funu ve Terraform kullanarak ad\u0131m ad\u0131m nas\u0131l kurulaca\u011f\u0131n\u0131 detayl\u0131 bir \u015fekilde ele ald\u0131k. Geli\u015fmi\u015f kullan\u0131m senaryolar\u0131 ve en iyi uygulamalarla, Karpenter’\u0131n t\u00fcm potansiyelini nas\u0131l ortaya \u00e7\u0131karabilece\u011finizi g\u00f6sterdik.<\/p>\n Sonu\u00e7 olarak, Karpenter, EKS k\u00fcmelerinizde h\u0131zl\u0131, maliyet-etkin ve operasyonel y\u00fck\u00fc azaltan bir node sa\u011flama mekanizmas\u0131 sunar. Terraform ile bu kurulumu otomatize etmek, altyap\u0131n\u0131z\u0131n tutarl\u0131l\u0131\u011f\u0131n\u0131, tekrarlanabilirli\u011fini ve s\u00fcr\u00fcm kontrol\u00fcn\u00fc sa\u011flayarak DevOps s\u00fcre\u00e7lerinizi g\u00fc\u00e7lendirir. Modern bulut uygulamalar\u0131n\u0131n dinamik do\u011fas\u0131na uyum sa\u011flamak ve bulut harcamalar\u0131n\u0131z\u0131 optimize etmek isteyen her kurulu\u015f i\u00e7in Karpenter, d\u00fc\u015f\u00fcn\u00fclmesi gereken kritik bir ara\u00e7t\u0131r.<\/p>\n Evet, \u00e7o\u011fu senaryoda Karpenter, Cluster Autoscaler’\u0131n yerini almak \u00fczere tasarlanm\u0131\u015ft\u0131r. Cluster Autoscaler’\u0131n aksine, Karpenter node gruplar\u0131na ba\u011fl\u0131 kalmaz ve pod taleplerine g\u00f6re do\u011frudan EC2 instance’lar\u0131n\u0131 sa\u011flar, bu da daha h\u0131zl\u0131 \u00f6l\u00e7eklenme ve daha iyi maliyet optimizasyonu sunar. Ancak, ge\u00e7i\u015f senaryolar\u0131nda bir s\u00fcre paralel olarak kullan\u0131labilirler.<\/p>\n Karpenter, AWS taraf\u0131ndan geli\u015ftirildi\u011fi i\u00e7in \u00f6ncelikli olarak AWS EKS ile entegre olacak \u015fekilde tasarlanm\u0131\u015ft\u0131r ve AWS’in EC2, Spot Instances gibi hizmetlerinden tam olarak yararlan\u0131r. \u015eu an i\u00e7in sadece AWS ekosisteminde desteklenmektedir.<\/p>\n Karpenter, EC2 instance katalo\u011fundaki neredeyse t\u00fcm instance tiplerini (C, M, R, T, G serileri dahil) kullanabilir. Provisioner kaynaklar\u0131n\u0131zda Karpenter, node’lar\u0131 iki ana durumda sonland\u0131r\u0131r:<\/p>\n Karpenter’\u0131n maliyet optimizasyonunu takip etmek i\u00e7in AWS Cost Explorer’\u0131 kullanabilirsiniz. Karpenter, ba\u015flatt\u0131\u011f\u0131 EC2 instance’lar\u0131na otomatik olarak #AWS #EKS #Karpenter #Terraform #Kubernetes #CloudMaliyetOptimizasyonu #DevOps<\/p>\n","protected":false},"excerpt":{"rendered":"G\u00fcn\u00fcm\u00fcz\u00fcn h\u0131zla de\u011fi\u015fen dijital d\u00fcnyas\u0131nda, uygulamalar\u0131n anl\u0131k taleplere g\u00f6re \u00f6l\u00e7eklenebilmesi hayati \u00f6nem ta\u015f\u0131r.","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"csco_page_header_type":"","csco_page_load_nextpost":"","csco_page_subscribe_form":"","csco_page_contact_form":"","footnotes":""},"categories":[1],"tags":[],"class_list":{"0":"post-44537","1":"post","2":"type-post","3":"status-publish","4":"format-standard","6":"category-genel","7":"cs-entry","8":"cs-video-wrap"},"yoast_head":"\nKarpenter’\u0131 EKS’e Kurmadan \u00d6nce Neleri Bilmelisiniz?<\/h2>\n
\u00d6n Ko\u015fullar ve Gerekli AWS \u0130zinleri<\/h3>\n
\n
kubectl<\/code> ile eri\u015filebilir durumda olmal\u0131d\u0131r.<\/li>\nkubernetes.io\/cluster\/<cluster-ad\u0131>: owned<\/code> ve karpenter.sh\/discovery: <cluster-ad\u0131><\/code> gibi etiketler kullan\u0131l\u0131r.<\/li>\nkubeconfig<\/code> dosyas\u0131na veya uygun kimlik do\u011frulama mekanizmalar\u0131na sahip olmas\u0131 gerekir.<\/li>\n<\/ul>\nTerraform’un Rol\u00fc ve Avantajlar\u0131<\/h3>\n
\n
Gerekli Ara\u00e7lar<\/h3>\n
\n
Karpenter’\u0131 Terraform ile EKS \u00dczerinde Ad\u0131m Ad\u0131m Kurulumu<\/h2>\n
<YOUR_CLUSTER_NAME><\/code>, <YOUR_AWS_REGION><\/code> ve <YOUR_ACCOUNT_ID><\/code> gibi yer tutucular\u0131 kendi ortam\u0131n\u0131za g\u00f6re g\u00fcncellemeyi unutmay\u0131n.<\/p>\n1. IAM Rolleri ve Politikalar\u0131 Olu\u015fturma<\/h3>\n
\n# main.tf\n# Karpenter Controller i\u00e7in IAM Rol\u00fc ve Politikas\u0131\nresource \"aws_iam_role\" \"karpenter_controller_role\" {\n name = \"KarpenterControllerRole-${var.cluster_name}\"\n\n assume_role_policy = jsonencode({\n Version = \"2012-10-17\",\n Statement = [\n {\n Effect = \"Allow\",\n Principal = {\n Federated = \"arn:aws:iam::${data.aws_caller_identity.current.account_id}:oidc-provider\/${replace(data.aws_eks_cluster.cluster.identity[0].oidc[0].issuer, \"https:\/\/\", \"\")}\"\n },\n Action = \"sts:AssumeRoleWithWebIdentity\",\n Condition = {\n StringEquals = {\n \"${replace(data.aws_eks_cluster.cluster.identity[0].oidc[0].issuer, \"https:\/\/\", \"\")}:sub\" = \"system:serviceaccount:karpenter:karpenter\"\n }\n }\n }\n ]\n })\n}\n\nresource \"aws_iam_role_policy_attachment\" \"karpenter_controller_policy_attach\" {\n policy_arn = \"arn:aws:iam::aws:policy\/KarpenterControllerPolicy\" # Bu politikay\u0131 manuel olarak veya ba\u015fka bir Terraform resource ile olu\u015fturman\u0131z gerekebilir.\n role = aws_iam_role.karpenter_controller_role.name\n}\n\n# Karpenter Controller i\u00e7in gerekli \u00f6zel izinler (\u00f6rnek)\nresource \"aws_iam_policy\" \"karpenter_controller_custom_policy\" {\n name = \"KarpenterControllerCustomPolicy-${var.cluster_name}\"\n description = \"Karpenter Controller i\u00e7in \u00f6zel izinler\"\n\n policy = jsonencode({\n Version = \"2012-10-17\",\n Statement = [\n {\n Effect = \"Allow\",\n Action = [\n \"ec2:CreateLaunchTemplate\",\n \"ec2:CreateFleet\",\n \"ec2:RunInstances\",\n \"ec2:CreateTags\",\n \"ec2:TerminateInstances\",\n \"ec2:DeleteLaunchTemplate\",\n \"ec2:DescribeLaunchTemplates\",\n \"ec2:DescribeInstances\",\n \"ec2:DescribeImages\",\n \"ec2:DescribeSubnets\",\n \"ec2:DescribeSecurityGroups\",\n \"ec2:DescribeInstanceTypes\",\n \"ec2:DescribeInstanceTypeOfferings\",\n \"ec2:DescribeAvailabilityZones\",\n \"ssm:GetParameter\" # E\u011fer SSM parametrelerinden AMI \u00e7ekiyorsan\u0131z\n ],\n Resource = \"*\"\n },\n {\n Effect = \"Allow\",\n Action = \"iam:PassRole\",\n Resource = aws_iam_role.karpenter_node_role.arn # Node Instance Profile rol\u00fcn\u00fc burada referans veriyoruz\n }\n ]\n })\n}\n\nresource \"aws_iam_role_policy_attachment\" \"karpenter_controller_custom_policy_attach\" {\n policy_arn = aws_iam_policy.karpenter_controller_custom_policy.arn\n role = aws_iam_role.karpenter_controller_role.name\n}\n\n# Karpenter Node'lar\u0131 i\u00e7in IAM Rol\u00fc (Instance Profile)\nresource \"aws_iam_role\" \"karpenter_node_role\" {\n name = \"KarpenterNodeRole-${var.cluster_name}\"\n\n assume_role_policy = jsonencode({\n Version = \"2012-10-17\",\n Statement = [\n {\n Effect = \"Allow\",\n Principal = {\n Service = \"ec2.amazonaws.com\"\n },\n Action = \"sts:AssumeRole\"\n }\n ]\n })\n}\n\nresource \"aws_iam_instance_profile\" \"karpenter_node_instance_profile\" {\n name = \"KarpenterNodeInstanceProfile-${var.cluster_name}\"\n role = aws_iam_role.karpenter_node_role.name\n}\n\n# Node'lar i\u00e7in gerekli politikalar (\u00f6rnek)\nresource \"aws_iam_role_policy_attachment\" \"karpenter_node_amazon_eks_worker_node_policy\" {\n policy_arn = \"arn:aws:iam::aws:policy\/AmazonEKSWorkerNodePolicy\"\n role = aws_iam_role.karpenter_node_role.name\n}\n\nresource \"aws_iam_role_policy_attachment\" \"karpenter_node_amazon_eks_cni_policy\" {\n policy_arn = \"arn:aws:iam::aws:policy\/AmazonEKS_CNI_Policy\"\n role = aws_iam_role.karpenter_node_role.name\n}\n\nresource \"aws_iam_role_policy_attachment\" \"karpenter_node_amazon_ec2_container_registry_readonly_policy\" {\n policy_arn = \"arn:aws:iam::aws:policy\/AmazonEC2ContainerRegistryReadOnly\"\n role = aws_iam_role.karpenter_node_role.name\n}\n\n# EKS Cluster ve Caller Identity verilerini \u00e7ekme\ndata \"aws_eks_cluster\" \"cluster\" {\n name = var.cluster_name\n}\n\ndata \"aws_caller_identity\" \"current\" {}\n<\/code><\/pre>\n<\/div>\n2. VPC ve Subnet Etiketlemesi<\/h3>\n
\n# main.tf\n# VPC Etiketlemesi\nresource \"aws_vpc\" \"eks_vpc\" {\n # ... mevcut VPC tan\u0131m\u0131n\u0131z veya yeni VPC olu\u015fturma ...\n tags = {\n \"kubernetes.io\/cluster\/${var.cluster_name}\" = \"owned\"\n }\n}\n\n# Subnet Etiketlemesi (\u00d6rnek olarak public subnet'ler)\nresource \"aws_subnet\" \"public_subnet_a\" {\n # ... mevcut subnet tan\u0131m\u0131n\u0131z veya yeni subnet olu\u015fturma ...\n vpc_id = aws_vpc.eks_vpc.id\n cidr_block = \"10.0.1.0\/24\"\n availability_zone = \"${var.aws_region}a\"\n tags = {\n \"kubernetes.io\/cluster\/${var.cluster_name}\" = \"owned\"\n \"karpenter.sh\/discovery\" = var.cluster_name\n }\n}\n\nresource \"aws_subnet\" \"public_subnet_b\" {\n # ... mevcut subnet tan\u0131m\u0131n\u0131z veya yeni subnet olu\u015fturma ...\n vpc_id = aws_vpc.eks_vpc.id\n cidr_block = \"10.0.2.0\/24\"\n availability_zone = \"${var.aws_region}b\"\n tags = {\n \"kubernetes.io\/cluster\/${var.cluster_name}\" = \"owned\"\n \"karpenter.sh\/discovery\" = var.cluster_name\n }\n}\n# \u00d6zel subnet'ler i\u00e7in de benzer etiketlemeler yap\u0131lmal\u0131d\u0131r.\n# \u00d6rne\u011fin, \u00f6zel subnet'ler i\u00e7in \"kubernetes.io\/role\/internal-elb\" etiketi ekleyebilirsiniz.\n<\/code><\/pre>\n<\/div>\nkarpenter.sh\/discovery<\/code> etiketi, Karpenter i\u00e7in kritik \u00f6neme sahiptir.<\/p>\n3. Karpenter Helm Chart’\u0131 Kurulumu<\/h3>\n
\n# main.tf\n# Helm provider tan\u0131m\u0131 (e\u011fer daha \u00f6nce yapmad\u0131ysan\u0131z)\nprovider \"helm\" {\n kubernetes {\n host = data.aws_eks_cluster.cluster.endpoint\n cluster_ca_certificate = base64decode(data.aws_eks_cluster.cluster.certificate_authority[0].data)\n exec {\n api_version = \"client.authentication.k8s.io\/v1beta1\"\n command = \"aws\"\n args = [\"eks\", \"get-token\", \"--cluster-name\", var.cluster_name, \"--region\", var.aws_region]\n }\n }\n}\n\n# Karpenter Helm Chart Kurulumu\nresource \"helm_release\" \"karpenter\" {\n name = \"karpenter\"\n repository = \"oci:\/\/public.ecr.aws\/karpenter\/karpenter\"\n chart = \"karpenter\"\n namespace = \"karpenter\"\n create_namespace = true\n version = \"0.33.0\" # Kulland\u0131\u011f\u0131n\u0131z Karpenter versiyonunu belirtin\n\n set {\n name = \"serviceAccount.annotations.eks\\\\.amazonaws\\\\.com\/role-arn\"\n value = aws_iam_role.karpenter_controller_role.arn\n type = \"string\"\n }\n\n set {\n name = \"settings.aws.clusterName\"\n value = var.cluster_name\n type = \"string\"\n }\n\n set {\n name = \"settings.aws.defaultInstanceProfile\"\n value = aws_iam_instance_profile.karpenter_node_instance_profile.name\n type = \"string\"\n }\n\n set {\n name = \"settings.aws.interruptionQueueName\"\n value = \"karpenter-${var.cluster_name}\" # Opsiyonel: Spot kesinti bildirimleri i\u00e7in SQS kuyru\u011fu\n type = \"string\"\n }\n}\n<\/code><\/pre>\n<\/div>\n4. Provisioner Kayna\u011f\u0131 Tan\u0131mlama<\/h3>\n
Provisioner<\/code> kayna\u011f\u0131d\u0131r. Bu kaynak, hangi instance tiplerinin kullan\u0131labilece\u011fi, hangi Availability Zone’larda \u00e7al\u0131\u015f\u0131laca\u011f\u0131, Spot veya On-Demand tercihi gibi kurallar\u0131 tan\u0131mlar. Terraform’da bu kayna\u011f\u0131 kubernetes_manifest<\/code> kayna\u011f\u0131n\u0131 kullanarak tan\u0131mlayabiliriz.<\/p>\n\n# provisioner.tf\nresource \"kubernetes_manifest\" \"default_provisioner\" {\n provider = kubernetes\n\n manifest = {\n apiVersion = \"karpenter.sh\/v1beta1\"\n kind = \"Provisioner\"\n metadata = {\n name = \"default\"\n }\n spec = {\n requirements = [\n {\n key = \"karpenter.k8s.aws\/instance-category\"\n operator = \"In\"\n values = [\"c\", \"m\", \"r\"] # Compute, Memory, General Purpose instance kategorileri\n },\n {\n key = \"topology.kubernetes.io\/zone\"\n operator = \"In\"\n values = [\"${var.aws_region}a\", \"${var.aws_region}b\", \"${var.aws_region}c\"] # Kullan\u0131lacak AZ'ler\n },\n {\n key = \"kubernetes.io\/arch\"\n operator = \"In\"\n values = [\"amd64\"] # Veya [\"arm64\"] Graviton i\u00e7in\n },\n {\n key = \"karpenter.sh\/capacity-type\"\n operator = \"In\"\n values = [\"spot\", \"on-demand\"] # Spot veya On-Demand tercihleri\n },\n ]\n limits = {\n resources = {\n cpu = \"1000\" # K\u00fcmenin toplam CPU limiti\n memory = \"1000Gi\" # K\u00fcmenin toplam bellek limiti\n }\n }\n providerRef = {\n name = \"default\" # AWSNodeTemplate i\u00e7in referans\n }\n ttlSecondsAfterEmpty = 30 # Node bo\u015fald\u0131ktan 30 saniye sonra sonland\u0131r\n ttlSecondsUntilExpired = 2592000 # 30 g\u00fcn sonra node'lar\u0131 yenile (g\u00fcvenlik g\u00fcncellemeleri i\u00e7in)\n }\n }\n}\n\nresource \"kubernetes_manifest\" \"aws_node_template\" {\n provider = kubernetes\n\n manifest = {\n apiVersion = \"karpenter.k8s.aws\/v1beta1\"\n kind = \"AWSNodeTemplate\"\n metadata = {\n name = \"default\"\n }\n spec = {\n subnetSelector = {\n \"karpenter.sh\/discovery\" = var.cluster_name\n }\n securityGroupSelector = {\n \"kubernetes.io\/cluster\/${var.cluster_name}\" = \"owned\"\n }\n instanceProfile = aws_iam_instance_profile.karpenter_node_instance_profile.name\n # E\u011fer \u00f6zel bir AMI kullanmak isterseniz:\n # amiSelector = {\n # \"karpenter.k8s.aws\/instance-family\" = \"al2023\" # Veya \"bottlerocket\", \"ubuntu\"\n # }\n }\n }\n}\n<\/code><\/pre>\n<\/div>\nProvisioner<\/code> ve AWSNodeTemplate<\/code> kaynaklar\u0131, Karpenter’a hangi t\u00fcr EC2 instance’lar\u0131n\u0131, hangi k\u0131s\u0131tlamalarla ve hangi a\u011f ayarlar\u0131yla ba\u015flatmas\u0131 gerekti\u011fini s\u00f6yler. \u00d6rne\u011fin, requirements<\/code> b\u00f6l\u00fcm\u00fcnde instance kategorilerini, Availability Zone’lar\u0131 ve kapasite tiplerini belirleyebilirsiniz. ttlSecondsAfterEmpty<\/code> gibi parametreler ise node’lar\u0131n ne zaman sonland\u0131r\u0131laca\u011f\u0131n\u0131 kontrol eder.<\/p>\n\nterraform init\nterraform plan\nterraform apply --auto-approve\n<\/code><\/pre>\n<\/div>\nKarpenter’\u0131 Test Etme ve Do\u011frulama<\/h2>\n
\u0130lk Pod’u Da\u011f\u0131tma ve \u00d6l\u00e7eklenmeyi G\u00f6zlemleme<\/h3>\n
\n# test-deployment.yaml\napiVersion: apps\/v1\nkind: Deployment\nmetadata:\n name: high-resource-app\nspec:\n replicas: 50\n selector:\n matchLabels:\n app: high-resource-app\n template:\n metadata:\n labels:\n app: high-resource-app\n spec:\n containers:\n - name: nginx\n image: nginx\n resources:\n requests:\n cpu: \"1\"\n memory: \"1Gi\"\n limits:\n cpu: \"1\"\n memory: \"1Gi\"\n<\/code><\/pre>\n<\/div>\n\nkubectl apply -f test-deployment.yaml\n<\/code><\/pre>\n<\/div>\n\n
\n kubectl get pods -w\n <\/code><\/pre>\n<\/p><\/div>\nPending<\/code> (Beklemede) durumunda oldu\u011funu g\u00f6rmelisiniz. K\u0131sa s\u00fcre sonra, Karpenter yeni node’lar\u0131 sa\u011flamaya ba\u015flad\u0131k\u00e7a bu pod’lar Running<\/code> (\u00c7al\u0131\u015f\u0131yor) durumuna ge\u00e7ecektir.<\/p>\n<\/li>\n\n kubectl get nodes -w\n <\/code><\/pre>\n<\/p><\/div>\nkarpenter.sh\/provisioner-name: default<\/code> gibi etiketlere sahip olacakt\u0131r.<\/p>\n<\/li>\n\n kubectl logs -f -n karpenter $(kubectl get pods -n karpenter -l app.kubernetes.io\/name=karpenter -o name)\n <\/code><\/pre>\n<\/p><\/div>\nNode Azaltma (Deprovisioning) Mekanizmas\u0131<\/h3>\n
\nkubectl delete -f test-deployment.yaml\n<\/code><\/pre>\n<\/div>\n\n
high-resource-app<\/code> pod’lar\u0131n\u0131n sonland\u0131\u011f\u0131n\u0131 g\u00f6rmelisiniz.<\/li>\n\n kubectl get nodes -w\n <\/code><\/pre>\n<\/p><\/div>\nNotReady<\/code> durumuna ge\u00e7ip daha sonra listeden kayboldu\u011funu g\u00f6rmelisiniz. Karpenter, ttlSecondsAfterEmpty<\/code> (bo\u015fald\u0131ktan sonra ya\u015fam s\u00fcresi) ve ttlSecondsUntilExpired<\/code> (s\u00fcresi dolana kadar ya\u015fam s\u00fcresi) gibi Provisioner ayarlar\u0131n\u0131za g\u00f6re node’lar\u0131 sonland\u0131racakt\u0131r. Konsolidasyon mekanizmas\u0131 sayesinde, birden fazla k\u00fc\u00e7\u00fck node yerine tek bir b\u00fcy\u00fck node’a ge\u00e7i\u015f gibi optimizasyonlar da yapabilir.<\/p>\n<\/li>\nGeli\u015fmi\u015f Karpenter Kullan\u0131m Senaryolar\u0131 ve \u0130pu\u00e7lar\u0131<\/h2>\n
Farkl\u0131 Provisioner Profilleri ile Esneklik<\/h3>\n
Provisioner<\/code> kayna\u011f\u0131 genellikle yeterli olsa da, farkl\u0131 i\u015f y\u00fckleri i\u00e7in \u00f6zelle\u015ftirilmi\u015f node sa\u011flama davran\u0131\u015flar\u0131 gerekebilir. \u00d6rne\u011fin, kesintiye toleransl\u0131 batch i\u015fleri i\u00e7in Spot Instance’lar\u0131 zorunlu k\u0131larken, kritik \u00f6neme sahip durum bilgisi olan uygulamalar i\u00e7in On-Demand Instance’lar\u0131 tercih edebilirsiniz.<\/p>\n\n
Provisioner<\/code> ve AWSNodeTemplate<\/code> tan\u0131mlayarak, farkl\u0131 kapasite tipleri (Spot\/On-Demand), instance aileleri (c5<\/code>, m5<\/code>, r5<\/code>) veya mimariler (amd64<\/code>, arm64<\/code> – Graviton) i\u00e7in ayr\u0131 profiller olu\u015fturabilirsiniz. Pod’lar\u0131n\u0131zda nodeSelector<\/code> veya affinity<\/code> kullanarak hangi Provisioner’\u0131n kullan\u0131laca\u011f\u0131n\u0131 belirleyebilirsiniz.<\/li>\nkarpenter.k8s.aws\/instance-gpu-count: \"1\"<\/code> gibi bir gereksinim ekleyebilirsiniz.<\/li>\n<\/ul>\n\n# gpu-provisioner.tf\nresource \"kubernetes_manifest\" \"gpu_provisioner\" {\n provider = kubernetes\n manifest = {\n apiVersion = \"karpenter.sh\/v1beta1\"\n kind = \"Provisioner\"\n metadata = {\n name = \"gpu-workloads\"\n }\n spec = {\n requirements = [\n { key = \"karpenter.k8s.aws\/instance-gpu-count\", operator = \"Gt\", values = [\"0\"] },\n { key = \"karpenter.sh\/capacity-type\", operator = \"In\", values = [\"on-demand\"] }, # GPU genellikle Spot'ta daha riskli olabilir\n # ... di\u011fer gereksinimler ...\n ]\n providerRef = { name = \"gpu-node-template\" }\n ttlSecondsAfterEmpty = 300 # Daha uzun s\u00fcre bo\u015fta kalabilir\n }\n }\n}\n\nresource \"kubernetes_manifest\" \"gpu_node_template\" {\n provider = kubernetes\n manifest = {\n apiVersion = \"karpenter.k8s.aws\/v1beta1\"\n kind = \"AWSNodeTemplate\"\n metadata = {\n name = \"gpu-node-template\"\n }\n spec = {\n instanceProfile = aws_iam_instance_profile.karpenter_node_instance_profile.name\n # ... subnetSelector, securityGroupSelector ...\n # amiSelector ile GPU destekli bir AMI belirleyebilirsiniz.\n }\n }\n}\n<\/code><\/pre>\n<\/div>\nMaliyet Y\u00f6netimi ve G\u00f6zetim<\/h3>\n
\n
karpenter.sh\/provisioner-name<\/code> ve karpenter.sh\/provisioner-family<\/code> gibi etiketler ekler, bu da maliyet analizinizi kolayla\u015ft\u0131r\u0131r.<\/li>\n<\/ul>\nEn \u0130yi Uygulamalar<\/h3>\n
\n
Provisioner<\/code> kaynaklar\u0131n\u0131zda limits.resources<\/code> alan\u0131n\u0131 kullanarak k\u00fcmenizin toplam CPU ve bellek limitlerini belirleyin. Bu, Karpenter’\u0131n \u00e7ok fazla kaynak sa\u011flamas\u0131n\u0131 veya k\u00fcmenizin beklenenden daha b\u00fcy\u00fck olmas\u0131n\u0131 engeller.<\/li>\nnodeSelector<\/code> ve toleration<\/code> kullanarak, Karpenter’\u0131n belirli i\u015f y\u00fckleri i\u00e7in en uygun node’lar\u0131 se\u00e7mesini sa\u011flay\u0131n. \u00d6rne\u011fin, kesintiye toleransl\u0131 pod’lar i\u00e7in Spot Instance’lara \u00f6zel tolerasyonlar ekleyin.<\/li>\nAWSNodeTemplate<\/code> i\u00e7inde amiSelector<\/code> alan\u0131n\u0131 kullanarak belirtebilirsiniz. AMI’lerinizi d\u00fczenli olarak g\u00fcncel tutmak, g\u00fcvenlik yamalar\u0131 ve performans iyile\u015ftirmeleri i\u00e7in \u00f6nemlidir.<\/li>\n<\/ul>\nSonu\u00e7 ve S\u0131k\u00e7a Sorulan Sorular<\/h2>\n
S\u0131k\u00e7a Sorulan Sorular<\/h3>\n
Karpenter Cluster Autoscaler’\u0131n yerine mi ge\u00e7iyor?<\/h3>\n
Karpenter sadece AWS EKS ile mi \u00e7al\u0131\u015f\u0131r?<\/h3>\n
Karpenter ile hangi instance tiplerini kullanabilirim?<\/h3>\n
requirements<\/code> alan\u0131n\u0131 kullanarak belirli instance ailelerini, mimarileri (amd64, arm64) ve kapasite tiplerini (spot, on-demand) tercih edebilirsiniz.<\/p>\nKarpenter node’lar\u0131 ne zaman sonland\u0131r\u0131r?<\/h3>\n
\n
ttlSecondsAfterEmpty<\/code> parametresinde belirtilen s\u00fcre (varsay\u0131lan 30 saniye) dolduktan sonra node’u sonland\u0131r\u0131r.<\/li>\nttlSecondsUntilExpired<\/code> parametresinde belirtilen ya\u015fam s\u00fcresi (varsay\u0131lan 7 g\u00fcn) doldu\u011funda, g\u00fcvenlik g\u00fcncellemeleri veya daha yeni AMI’lere ge\u00e7i\u015f i\u00e7in node’u sonland\u0131r\u0131r ve yerine yenisini sa\u011flar.<\/li>\n<\/ol>\nKarpenter’\u0131n maliyetleri nas\u0131l optimize etti\u011fini nas\u0131l takip edebilirim?<\/h3>\n
karpenter.sh\/provisioner-name<\/code> ve karpenter.sh\/provisioner-family<\/code> gibi etiketler ekler. Cost Explorer’da bu etiketlere g\u00f6re filtreleme yaparak, Karpenter’\u0131n sa\u011flad\u0131\u011f\u0131 node’lar\u0131n maliyetini ve tasarruflar\u0131n\u0131 detayl\u0131 bir \u015fekilde analiz edebilirsiniz. Ayr\u0131ca, Prometheus ve Grafana ile Karpenter metriklerini izleyerek operasyonel verimlili\u011fi g\u00f6zlemleyebilirsiniz.<\/p>\n