カスタムメトリクスでサービスの拡張を行い、Eコマースのトラフィックスパイクを効率的に管理するためのKEDA / KEDA for scaling services with custom metrics to manage e-commerce traffic spikes efficiently.

このセッションでは、KEDA（Kubernetes Event-driven Autoscaler）を使用してカスタムメトリクスに基づいてKubernetesのポッドをスケールし、eコマースシステムでのトラフィックスパイクを効率的に管理する方法を探ります。従来のHorizontal Pod Autoscaler（HPA）の限界を強調し、KEDAの利点を紹介し、eコマース環境での実装事例を提示します。迅速なスケーリングとコスト削減の利点についても議論します。

Transcript

1. ,&%"
   GPS
   TDBMJOH
   TFSWJDFT
   XJUI
   DVTUPN
   NFUSJDT
   UP
   NBOBHF
   FDPNNFSDF
   USBGGJD
   TQJLFT
   FGGJDJFOUMZ -*/&
   $PNQBOZ
   5IBJMBOE $IPOLJBUUJQPPN
   "SOJXBUU
2.  /BNF  $PNQBOZ   3PMF  1SPEVDU *OUSPEVDUJPO "CPVU
   .F $POUBDU  &NBJM BSOJXBUUD!MJOFDPSQDPN  (JUIVC
   
   
   
   
   
   
   
   QSVFLL  -JOLFEJO
   
   
   
   
   JOBSOJXBUUDIPOLJBUUJQPPN "SOJXBUU
   1SVFL
   $IPOLJBUUJQPPN -JOF
   $PNQBOZ
   5IBJMBOE 4JUF
   3FMJBCJMJUZ
   &OHJOFFS &DPNNFSDF
   5IBJMBOE
3.  5IF
   $IBMMFOHF
   PG
   &DPNNFSDF
   5SBGGJD
   4QJLFT  ,&%"
   5IF
   'VOEBNFOUBMT  ,&%"
   %FFQ
   %JWF  *NQMFNFOUBUJPO
   
   #FTU
   1SBDUJDFT  $PODMVTJPO "HFOEB
4. 5IF
   $IBMMFOHF
   PG
   &DPNNFSDF
   5SBGGJD
   4QJLFT ⎯ 'MBTI
   4BMFT
   
   1SPNPUJPOT ⎯ )PMJEBZ
   4FBTPOT ⎯ .BSLFUJOH
   $BNQBJHOT ⎯ (FPHSBQIJD
   -PBE
   4IJGUT 5SBGGJD
   JT
   PGUFO
   VOQSFEJDUBCMF ⎯ 
   TFDPOE
   MPBE
   UJNF
   
   DVTUPNFS
   BCBOEPONFOU ⎯ 0VUBHF
   
   3FWFOVF
   -PTT ⎯ &UD *NQBDU
5. 5IF
   $IBMMFOHF
   PG
   &DPNNFSDF
   5SBGGJD
   4QJLFT $BNQBJHO
   TUBSU
   BU
    16x
6. 5IF
   $IBMMFOHF
   PG
   &DPNNFSDF
   5SBGGJD
   4QJLFT ⎯ 3FBDUJWF
   OPU
   1SPBDUJWF ⎯ .JTMFBEJOH
   .FUSJDT ⎯ $VTUPN
   #VTJOFTT
   -PHJD -JNJUBUJPOT
   PG
   5SBEJUJPOBM
   "VUPTDBMJOH 17x 6x
7. ,&%"
   5IF
   'VOEBNFOUBMT ,&%"
   ,VCFSOFUFT
   &WFOUESJWFO
   "VUPTDBMJOH ,&%"
   0QFSBUPS 4DBMFST .FUSJDT 4FSWFS
8. ,&%"
   )PX
   JU
   XPSLT ,VCFSOFUFT
   DMVTUFS KEDA Operator Metrics Server Scalers HPA Trigger Scale Other Backends Deployment Pod 1 Pod 2 Pod N
9. ,&%"
   $VTUPN
   .FUSJDT ⎯ .FTTBHF
   2VFVF
   -FOHUI ⎯ "DUJWF
   6TFS
   4FTTJPOT ⎯ 1FOEJOH
   1BZNFOUT ⎯ "1*
   (BUFXBZ
   3FRVFTU
   3BUF ⎯ %BUBCBTF
   $POOFDUJPO
   1PPM
   6UJMJ[BUJPO ⎯ *OWFOUPSZ
   )PME
   $PVOU $VTUPN
   .FUSJDT
   "SF
   /FFEFE
10. &DPNNFSDF
    TZTUFN ,VCFSOFUFT
    DMVTUFS Scale Users LB Pull Metrics Server
11. ,&%"
    %FFQ
    %JWF apiVersion: autoscaling/v2beta1 kind: HorizontalPodAutoscaler metadata: name: order-processor spec: scaleTargetRef: apiVersion: apps/v1 kind: Deployment name: order-processor minReplicas: 2 maxReplicas: 20 ... 5SBEJUJPOBM
    "VUPTDBMJOH HPA metrics: - type: Resource resource: name: memory targetAverageUtilization: 100 - type: Resource resource: name: cpu targetAverageUtilization: 100
12. ,&%"
    %FFQ
    %JWF 5SBEJUJPOBM
    "VUPTDBMJOH metrics: - type: Resource resource: name: cpu targetAverageUtilization: 100 desiredReplicas = ceil(1 x 11.4 / 100) = ceil(0.114) = 1 desiredReplicas = ceil(1 x 241 / 100) = ceil(2.41) = 3
13. ,&%"
    %FFQ
    %JWF pollingInterval: 30 # How often KEDA checks the metric (seconds) cooldownPeriod: 300 # How long to wait before scaling down (seconds) triggers: - type: memory metricType: Utilization metadata: value: ”100" - type: cpu metricType: Utilization metadata: value: ”100" )1"
    UP
    4DBMFE0CKFDU Scalers apiVersion: keda.sh/v1alpha1 kind: ScaledObject metadata: name: order-processor-scaler spec: scaleTargetRef: apiVersion: apps/v1 kind: Deployment name: order-processor-deployment minReplicaCount: 2 maxReplicaCount: 20 ...
14. triggers: - type: memory metricType: Utilization metadata: value: ”100" - type: cpu metricType: Utilization metadata: value: ”100” - type: prometheus metadata: serverAddress: http://prometheus.cattle-monitoring-system:9090 metricName: istio_requests_total query: sum(increase(istio_requests_total{destination_app=”order-processor"}[1m])) threshold: "50.00" ,&%"
    %FFQ
    %JWF 4DBMFE0CKFDU XJUI
    DVTUPN
    TDBMFST Scalers
15. ,&%"
    %FFQ
    %JWF 4DBMFE0CKFDU XJUI
    DVTUPN
    TDBMFST - type: prometheus metadata: metricName: istio_requests_total threshold: "5000" desiredReplicas = ceil(1 x 4510 / 5000) = ceil(0.902) = 1 desiredReplicas = ceil(1 x 16000 / 5000) = ceil(3.2) = 4
16. ,&%"
    %FFQ
    %JWF $PNQBSJOH
    UIF
    NFUSJDT Requests increased ~300% CPU increased ~180%
17.  #VTJOFTT0SJFOUFE -FBEJOH
    *OEJDBUPST (SBOVMBSJUZ 4UBCJMJUZ *NQMFNFOUBUJPO
    
    #FTU
    1SBDUJDFT $IPPTF
    UIF
    3JHIU
    .FUSJDT
18.  QPMMJOH*OUFSWBM DPPMEPXO1FSJPE NJO3FQMJDB$PVOU NBY3FQMJDB$PVOU *NQMFNFOUBUJPO
    
    #FTU
    1SBDUJDFT 'JOFUVOF
    4DBMJOH
    1BSBNFUFST
19.  .POJUPS
    ,&%"
    JUTFMG  .POJUPS
    4DBMFE
    .FUSJDT  .POJUPS
    )1"
    &WFOUT  4FU
    "MFSUT *NQMFNFOUBUJPO
    
    #FTU
    1SBDUJDFT .POJUPSJOH
    
    "MFSUJOH
20. *NQMFNFOUBUJPO
    
    #FTU
    1SBDUJDFT  4JNVMBUF
    4QJLFT  0CTFSWF
    #FIBWJPS  *UFSBUF 5FTUJOH
    "VUPTDBMJOH
    1PMJDJFT
    VOEFS
    -PBE
21. $PODMVTJPO  0QUJNJ[FE
    3FTPVSDF
    6UJMJ[BUJPO  &OIBODFE
    3FTQPOTJWFOFTT  (SFBUFS
    'MFYJCJMJUZ  1SFEJDUJWF
    $BQBCJMJUJFT &NQPXFSJOH
    :PVS
    &DPNNFSDF
    1MBUGPSN