FinOps! karpenterによるk8sコスト削減

動画: https://event.cloudnativedays.jp/cndt2023/talks/1966

Transcript

1. CLOUD NATIVE DAYS TOKYO 2023 AYA IGARASHI @LADICLE FINOPS! KARPENTERʹΑΔ K8Sίετ࡟ݮͷಓ
2. FinOps! 72% OF COMPANIES HAVE A DEDICATED FINOPS TEAM. SOURCE: FLEXERA 2023 STATE OF THE CLOUD REPORT
3. Source: Flexera 2023 State of the Cloud Report “WASTED AVG. 28% OF CLOUD COSTS”
4. FinOpsʹ͓͚Δίετͷ࠷ దԽ͸୯ʹҰ൪͍҆Πϯϑ ϥΛ࢖͏ͱ͍͏ҙຯ͚ͩͰ ͸ͳ͍ɻ ෆཁϦιʔε࡟আ౳ͷ Πϯϑϥίετ࡟ݮ WHAT IS COST OPTIMIZATION? IT DOES NOT JUST MEAN SAVING INFRASTRUCTURE EXPENSES ӡ༻ޮ཰Խ(ࣗಈԽ)ʹΑΔ ਓతίετ࡟ݮ ΠϯϑϥͷίετΛ҆͘Ͱ ͖ͨͱͯ͠΋ɺͦΕΛҡ࣋ ͢ΔͨΊʹଟ͘ͷखಈΦϖ Ϩʔγϣϯ͕ඞཁʹͳΔͱ ϏδωεʹूதͰ͖ͳ͍ɻ Մ༻ੑ޲্ʹΑΔ ػձଛࣦͷ௿ݮ (≒ ച্૿Ճ) Πϯϑϥͷ͚҆ͩ͞Λ௥ٻ ͯ͠ɺαʔϏε͕ఀࢭͨ͠ Γɺ඼࣭͕ྼԽͯ͠͠·͏ ͱɺϏδωεͷػձଛࣦ͕ ੜ·Εͯ͠·͏ɻ
5. FinOpsʹ͓͚Δίετͷ࠷ దԽ͸୯ʹҰ൪͍҆Πϯϑ ϥΛ࢖͏ͱ͍͏ҙຯ͚ͩͰ ͸ͳ͍ɻ ෆཁϦιʔε࡟আ౳ͷ Πϯϑϥίετ࡟ݮ WHAT IS COST OPTIMIZATION? IT DOES NOT JUST MEAN SAVING INFRASTRUCTURE EXPENSES ӡ༻ޮ཰Խ(ࣗಈԽ)ʹΑΔ ਓతίετ࡟ݮ ΠϯϑϥͷίετΛ҆͘Ͱ ͖ͨͱͯ͠΋ɺͦΕΛҡ࣋ ͢ΔͨΊʹଟ͘ͷखಈΦϖ Ϩʔγϣϯ͕ඞཁʹͳΔͱ ϏδωεʹूதͰ͖ͳ͍ɻ Մ༻ੑ޲্ʹΑΔ ػձଛࣦͷ௿ݮ (≒ ച্૿Ճ) Πϯϑϥͷ͚҆ͩ͞Λ௥ٻ ͯ͠ɺαʔϏε͕ఀࢭͨ͠ Γɺ඼࣭͕ྼԽͯ͠͠·͏ ͱɺϏδωεͷػձଛࣦ͕ ੜ·Εͯ͠·͏ɻ ࠷దͳόϥϯεΛऔΔ
6. CLOUD NATIVE DAYS TOKYO 2023 AYA IGARASHI @LADICLE FINOPS! KARPENTERʹΑΔ K8Sίετ࡟ݮ ࠷దԽͷಓ
7. AGENDA Optimize Cost KarpenterʹΑΔίετ࠷ దԽͷ͘͠Έ 04 WHAT YOU CAN LEARN TODAY! Optimize More! KarpenterͰΑΓίετΛ ࠷దԽ͢Δʹ͸ 05 Karpenter K8SΫϥελΦʔτεέʔ ϥͷKarpenterͱ͸ 03 Autoscalers ίετ࠷దԽʹ͓͚ΔΦʔ τεέʔϥͷར఺ 02 Cost Optimization ίετ࠷దԽͱ͸? 01
8. HOW TO OPTIMIZE K8S COST?
9. HOW TO OPTIMIZE K8S COST? Autoscaler
10. ݱࡏඞཁͱ͢ΔϦιʔεͩ ͚Λ֬อ͢ΔͷͰɺແବͳ ϦιʔεͷίετΛ࡟ݮͰ ͖Δɻ ෆཁϦιʔε࡟আ౳ͷ Πϯϑϥίετ࡟ݮ AUTOSCALER OPTIMIZE COST THE BENEFITS OF AUTOSCALER ӡ༻ޮ཰Խ(ࣗಈԽ)ʹΑΔ ਓతίετ࡟ݮ ඞཁͳϦιʔεྔ͕มԽͯ͠ ΋ࣗಈ௥ैͰ͖ΔͷͰɺӡ ༻ऀ͕εέʔϦϯάͷͨΊ ͷϦιʔε؂ࢹ΍࡞ۀΛ࡟ ݮͰ͖Δɻ Մ༻ੑ޲্ʹΑΔ ػձଛࣦͷ௿ݮ (≒ ച্૿Ճ) ਓͷखΛհͣ͞ʹεέʔϦ ϯάͰ͖ΔͷͰɺਓతϛε ΍஗ԆͷϦεΫΛܰݮ͠ɺ Մ༻ੑΛ޲্Ͱ͖Δɻ
11. ݱࡏඞཁͱ͢ΔϦιʔεͩ ͚Λ֬อ͢ΔͷͰɺແବͳ ϦιʔεͷίετΛ࡟ݮͰ ͖Δɻ ෆཁϦιʔε࡟আ౳ͷ Πϯϑϥίετ࡟ݮ AUTOSCALER OPTIMIZE COST THE BENEFITS OF AUTOSCALER ӡ༻ޮ཰Խ(ࣗಈԽ)ʹΑΔ ਓతίετ࡟ݮ ඞཁͳϦιʔεྔ͕มԽͯ͠ ΋ࣗಈ௥ैͰ͖ΔͷͰɺӡ ༻ऀ͕εέʔϦϯάͷͨΊ ͷϦιʔε؂ࢹ΍࡞ۀΛ࡟ ݮͰ͖Δɻ Մ༻ੑ޲্ʹΑΔ ػձଛࣦͷ௿ݮ (≒ ച্૿Ճ) ਓͷखΛհͣ͞ʹεέʔϦ ϯάͰ͖ΔͷͰɺਓతϛε ΍஗ԆͷϦεΫΛܰݮ͠ɺ Մ༻ੑΛ޲্Ͱ͖Δɻ ֤ίετΛ࡟ݮͰ͖Δ!!
12. AGENDA Optimize Cost KarpenterʹΑΔίετ࠷ దԽͷ͘͠Έ 04 WHAT YOU CAN LEARN TODAY! Karpenter K8SΫϥελΦʔτεέʔ ϥͷKarpenterͱ͸ 03 Autoscalers ίετ࠷దԽʹ͓͚ΔΦʔ τεέʔϥͷར఺ 02 Optimize More! KarpenterͰΑΓίετΛ ࠷దԽ͢Δʹ͸ 05 Cost Optimization ίετ࠷దԽͱ͸? 01
13. K8S AUTOSCALERS Quick Recap
14. THE TYPES OF AUTOSCALERS KUBERNETES AUTOSCALING Workload(/Pod) Autoscaling Cluster(/Node) Autoscaling Pod Node scale Pod Pod Node scale Pod Pod Node scale Pod New! Node
15. K8S CLUSTER AUTOSCALER
16. K8S CLUSTER AUTOSCALER Karpenter
17. AWS͕։ൃͨ͠k8sͷΫϥελΦʔτεέʔϥ https://karpenter.sh/ https://github.com/aws/karpenter AWSҎ֎ʹ΋AzureʹରԠ͍ͯ͠Δ https://github.com/Azure/karpenter ίΞػೳ͸k8sίϛϡχςΟͰ؅ཧ͞Ε͍ͯΔ https://github.com/kubernetes-sigs/karpenter ONE OF KUBERNETES CLUSTER AUTOSCALER WHAT IS KARPENTER? 2021/11/26: Ұൠఏڙ։࢝ 01 02 03 2023/11/06: AzureͷαϙʔτΛൃද 2023/11/23: sig-autoscalerʹدଃ
18. AWS͕։ൃͨ͠k8sͷΫϥελΦʔτεέʔϥ https://karpenter.sh/ https://github.com/aws/karpenter AWSҎ֎ʹ΋AzureʹରԠ͍ͯ͠Δ https://github.com/Azure/karpenter ίΞػೳ͸k8sίϛϡχςΟͰ؅ཧ͞Ε͍ͯΔ https://github.com/kubernetes-sigs/karpenter ONE OF KUBERNETES CLUSTER AUTOSCALER WHAT IS KARPENTER? aws/karpenter Azure/karpenter kuberentes-sigs/ karpenter v0.33.0 v0.2.0 v0.33.0
19. VS. CLUSTER AUTOSCALER (CAS)
20. LESS MANUAL OPERATION, FASTER SCALE (AWS) WHAT IS DIFFERENT FROM CAS? Kubernetes Cluster Autoscaler (CAS) Karpenter CAS Pod Node Pod Pod 1. Unschedulable Node 2. Create a new node from NG Node Group AWS ASG
21. LESS MANUAL OPERATION, FASTER SCALE (AWS) WHAT IS DIFFERENT FROM CAS? Kubernetes Cluster Autoscaler (CAS) Karpenter CAS Pod Node Pod Pod Node Node Group AWS ASG Large Node Group AWS ASG Small 1. Unschedulable 2. Create a new node from NG (small)
22. LESS MANUAL OPERATION, FASTER SCALE (AWS) WHAT IS DIFFERENT FROM CAS? Kubernetes Cluster Autoscaler (CAS) Karpenter CAS Pod Node Pod Pod Node Karpenter Pod Node Pod Node AWS Fleet Node Pool Node Group AWS ASG Pod 1. Unschedulable 2. Create a new Node
23. LESS MANUAL OPERATION, FASTER SCALE (AWS) WHAT IS DIFFERENT FROM CAS? Kubernetes Cluster Autoscaler (CAS) Karpenter CAS Pod Node Pod Pod Node Karpenter Pod Node Pod Pod Node 1. Unschedulable 2. Create a new (medium) Node AWS Fleet Node Pool Node Group AWS ASG
24. LESS MANUAL OPERATION, FASTER SCALE (AWS) WHAT IS DIFFERENT FROM CAS? Kubernetes Cluster Autoscaler (CAS) Karpenter • Node͸CASͷNodeGroup(NG)ઃఆʹैͬ ͯ௥Ճ͞ΕΔ (AWSͰ͸EC2 ASGʹରԠ) • ௥Ճ͢ΔNodeͷγϛϡϨʔτʹ͸NG͝ͱ ʹ1छྨͷNodeαΠζ͕࢖ΘΕΔ • ৽͍͠ΠϯελϯελΠϓͷNodeΛCAS͔ Βੜ੒͢Δʹ͸NGͷ௥Ճ͕ඞཁͰɺ·ͨɺ NG͕ଟ͗͢Δͱεέʔϧʹ͕͔͔࣌ؒΔ • Node͸NodePoolʹॻ͔Εͨ࡞੒͢Δϊʔ ͷ੍໿ࣄ߲ΛݩʹKarpenter͕બ୒͢Δ • ௥Ճ͢ΔNodeͷγϛϡϨʔτ͸NodePool Λݩʹ༷ʑͳαΠζͷNode͕࢖ΘΕΔ • AWSͷ৔߹ɺΠϯελϯεΛEC2 ASGΛܦ ༝ͤͣʹEC2 Fleet͔Β௚઀্ཱͪ͛ΔͷͰ CASʹൺ΂ΔͱΠϯελϯεͷىಈ͕ૣ͍
25. LESS MANUAL OPERATION, FASTER SCALE (AWS) WHAT IS DIFFERENT FROM CAS? Kubernetes Cluster Autoscaler (CAS) Karpenter • Node͸CASͷNodeGroup(NG)ઃఆʹैͬ ͯ௥Ճ͞ΕΔ (AWSͰ͸EC2 ASGʹରԠ) • ௥Ճ͢ΔNodeͷγϛϡϨʔτʹ͸NG͝ͱ ʹ1छྨͷNodeαΠζ͕࢖ΘΕΔ • ৽͍͠ΠϯελϯελΠϓͷNodeΛCAS͔ Βੜ੒͢Δʹ͸NGͷ௥Ճ͕ඞཁͰɺ·ͨɺ NG͕ଟ͗͢Δͱεέʔϧʹ͕͔͔࣌ؒΔ • Node͸NodePoolʹॻ͔Εͨ࡞੒͢Δϊʔ ͷ੍໿ࣄ߲ΛݩʹKarpenter͕બ୒͢Δ • ௥Ճ͢ΔNodeͷγϛϡϨʔτ͸NodePool Λݩʹ༷ʑͳαΠζͷNode͕࢖ΘΕΔ • AWSͷ৔߹ɺΠϯελϯεΛEC2 ASGΛܦ ༝ͤͣʹEC2 Fleet͔Β௚઀্ཱͪ͛ΔͷͰ CASʹൺ΂ΔͱΠϯελϯεͷىಈ͕ૣ͍ ख࡞ۀ͕ݮΓɺ ΑΓૣ͘εέʔϧͰ͖Δ!
26. AGENDA Optimize Cost KarpenterʹΑΔίετ࠷ దԽͷ͘͠Έ 04 WHAT YOU CAN LEARN TODAY! Karpenter K8SΫϥελΦʔτεέʔ ϥͷKarpenterͱ͸ 03 Autoscalers ίετ࠷దԽʹ͓͚ΔΦʔ τεέʔϥͷར఺ 02 Optimize More! KarpenterͰΑΓίετΛ ࠷దԽ͢Δʹ͸ 05 Cost Optimization ίετ࠷దԽͱ͸? 01
27. CREATE THE RIGHT SIZE INSTANCE AND DELETE THE UNNECESSARY INSTANCE KARPENTER OPTIMIZES COSTS! Provisioning Karpenter Node AWS Fleet Node Pool Pod Pod Pod Pod Pod Pod Pod Pod 1. Grouping pending Pods and Finding f itting candidates. 3. Create an optimal instance based on strategy • Pending PodΛ·ͱΊͯϑΟοτ͢ΔNode ͷछྨΛ୳͢(BinPacking, FFD Algorithm) • ϑΟοτ͢Δ΋ͷΑΓେ͖͍max 60ީิͷ ͏ͪɺઓུʹԠͨ͡࠷దͳΠϯελϯε͕ ىಈͰ͖Δ on-demand: lowest-price (࠷΋͍҆΋ͷ) spot: price-capacity-optimized (҆͘தஅՄೳੑͷ௿͍΋ͷ) 2. Send max 60 Candidates Save Costs!
28. CREATE THE RIGHT SIZE INSTANCE AND DELETE THE UNNECESSARY INSTANCE KARPENTER OPTIMIZES COSTS! Disruption Controllers • Expiration: ಛఆ͕࣌ؒܦաͨ͠Β࡟আ • Consolidation: Node্ͷશͯͷPod͕ଞͷNode ʹҠಈՄೳͳΒ࡟আɺ·ͨ͸ɺݱࡏͷΠϯελϯ εΑΓ͍҆Πϯελϯε͕ར༻ՄೳͳΒஔ׵ • Drift: NodePoolͷઃఆ͕มߋ͞ΕɺطଘͷΠϯε λϯε͕๬·͍͠ঢ়ଶͱͷࠩ෼͕ͰͨΒ࡟আ͢Δ • Interruption: SpotΠϯελϯεͷఀࢭ΍ UnhealthyʹͳͬͨΒஔ׵͢Δ Karpenter Pod Node Pod Node Pod Node Delete Unnecessary or Costly Nodes Save Costs!
29. Node A Pod Node A Pod Pod CREATE THE RIGHT SIZE INSTANCE AND DELETE THE UNNECESSARY INSTANCE KARPENTER OPTIMIZES COSTS! Disruption Controllers • Expiration: ಛఆ͕࣌ؒܦաͨ͠Β࡟আ • Consolidation: Node্ͷશͯͷPod͕ଞͷNode ʹҠಈՄೳͳΒ࡟আɺ·ͨ͸ɺݱࡏͷΠϯελϯ εΑΓ͍҆Πϯελϯε͕ར༻ՄೳͳΒஔ׵ • Drift: NodePoolͷઃఆ͕มߋ͞ΕɺطଘͷΠϯε λϯε͕๬·͍͠ঢ়ଶͱͷࠩ෼͕ͰͨΒ࡟আ͢Δ • Interruption: SpotΠϯελϯεͷఀࢭ΍ UnhealthyʹͳͬͨΒஔ׵͢Δ Pod Expire 2h Delete after 2 hours
30. CREATE THE RIGHT SIZE INSTANCE AND DELETE THE UNNECESSARY INSTANCE KARPENTER OPTIMIZES COSTS! Disruption Controllers Node A Pod Node B Pod Node A’ Pod Pod $0.05 $0.02 Recreate a new node Reallocate a Pod • Expiration: ಛఆ͕࣌ؒܦաͨ͠Β࡟আ • Consolidation: Node্ͷશͯͷPod͕ଞͷNode ʹҠಈՄೳͳΒ࡟আɺ·ͨ͸ɺݱࡏͷΠϯελϯ εΑΓ͍҆Πϯελϯε͕ར༻ՄೳͳΒஔ׵ • Drift: NodePoolͷઃఆ͕มߋ͞ΕɺطଘͷΠϯε λϯε͕๬·͍͠ঢ়ଶͱͷࠩ෼͕ͰͨΒ࡟আ͢Δ • Interruption: SpotΠϯελϯεͷఀࢭ΍ UnhealthyʹͳͬͨΒஔ׵͢Δ Pod
31. CREATE THE RIGHT SIZE INSTANCE AND DELETE THE UNNECESSARY INSTANCE KARPENTER OPTIMIZES COSTS! Disruption Controllers • Expiration: ಛఆ͕࣌ؒܦաͨ͠Β࡟আ • Consolidation: Node্ͷશͯͷPod͕ଞͷNode ʹҠಈՄೳͳΒ࡟আɺ·ͨ͸ɺݱࡏͷΠϯελϯ εΑΓ͍҆Πϯελϯε͕ར༻ՄೳͳΒஔ׵ • Drift: NodePoolͷઃఆ͕มߋ͞ΕɺطଘͷΠϯε λϯε͕๬·͍͠ঢ়ଶͱͷࠩ෼͕ͰͨΒ࡟আ͢Δ • Interruption: SpotΠϯελϯεͷఀࢭ΍ UnhealthyʹͳͬͨΒஔ׵͢Δ Node A Pod Node A’ Pod Recreate Im age v1 Im age v2 Desired state of machine image was changed from v1 to v2. Node Pool v2
32. Node A CREATE THE RIGHT SIZE INSTANCE AND DELETE THE UNNECESSARY INSTANCE KARPENTER OPTIMIZES COSTS! Disruption Controllers • Expiration: ಛఆ͕࣌ؒܦաͨ͠Β࡟আ • Consolidation: Node্ͷશͯͷPod͕ଞͷNode ʹҠಈՄೳͳΒ࡟আɺ·ͨ͸ɺݱࡏͷΠϯελϯ εΑΓ͍҆Πϯελϯε͕ར༻ՄೳͳΒஔ׵ • Drift: NodePoolͷઃఆ͕มߋ͞ΕɺطଘͷΠϯε λϯε͕๬·͍͠ঢ়ଶͱͷࠩ෼͕ͰͨΒ࡟আ͢Δ • Interruption: SpotΠϯελϯεͷఀࢭ΍ UnhealthyʹͳͬͨΒஔ׵͢Δ Spot Spot interruptions have a 2-minute notice Node A’ Pod Pod Recreate
33. CREATE THE RIGHT SIZE INSTANCE AND DELETE THE UNNECESSARY INSTANCE KARPENTER OPTIMIZES COSTS! Provisioning Disruption Controllers • Pending PodΛ·ͱΊͯϑΟοτ͢ΔNode ͷछྨΛ୳͢(BinPacking, FFD Algorithm) • ϑΟοτ͢Δ΋ͷΑΓେ͖͍max 60ީิͷ ͏ͪɺઓུʹԠͨ͡࠷దͳΠϯελϯε͕ ىಈͰ͖Δ on-demand: lowest-price (࠷΋͍҆΋ͷ) spot: price-capacity-optimized (҆͘தஅՄೳੑͷ௿͍΋ͷ) • Expiration: ಛఆ͕࣌ؒܦաͨ͠Β࡟আ • Consolidation: Node্ͷશͯͷPod͕ଞͷNode ʹҠಈՄೳͳΒ࡟আɺ·ͨ͸ɺݱࡏͷΠϯελϯ εΑΓ͍҆Πϯελϯε͕ར༻ՄೳͳΒஔ׵ • Drift: NodePoolͷઃఆ͕มߋ͞ΕɺطଘͷΠϯε λϯε͕๬·͍͠ঢ়ଶͱͷࠩ෼͕ͰͨΒ࡟আ͢Δ • Interruption: spotͷΠϯελϯεͷఀࢭ΍ UnhealthyʹͳͬͨΒஔ׵͢Δ CASΑΓΞάϨογϒ ͳͷͰ஫ҙ!
34. CREATE THE RIGHT SIZE INSTANCE AND DELETE THE UNNECESSARY INSTANCE KARPENTER OPTIMIZES COSTS! Provisioning Disruption Controllers • Pending PodΛ·ͱΊͯϑΟοτ͢ΔNode ͷछྨΛ୳͢(BinPacking, FFD Algorithm) • ϑΟοτ͢Δ΋ͷΑΓେ͖͍max 60ީิͷ ͏ͪઓུʹԠͨ͡࠷దͳΠϯελϯε͕ى ಈ͞ΕΔ on-demand: lowest-price (࠷΋͍҆΋ͷ) spot: price-capacity-optimized (҆͘தஅՄೳੑͷ௿͍΋ͷ) • Expiration: ಛఆ͕࣌ؒܦաͨ͠Β࡟আ • Consolidation: Node্ͷશͯͷPod͕ଞͷNode ʹҠಈՄೳͳΒ࡟আɺ·ͨ͸ɺݱࡏͷΠϯελϯ εΑΓ͍҆Πϯελϯε͕ར༻ՄೳͳΒஔ׵ • Drift: NodePoolͷઃఆ͕มߋ͞ΕɺطଘͷΠϯε λϯε͕๬·͍͠ঢ়ଶͱͷࠩ෼͕ͰͨΒ࡟আ͢Δ • Interruption: spotͷΠϯελϯεͷఀࢭ΍ UnhealthyʹͳͬͨΒஔ׵͢Δ CASΑΓΞάϨογϒ ͳͷͰ஫ҙ! ίετΛ࠷దԽ! ඞཁͳϦιʔε͚ͩ࡞੒ ෆཁͳϦιʔε͸࡟আ ✕ ࣗಈԽ
35. AGENDA Optimize Cost KarpenterʹΑΔίετ࠷ దԽͷ͘͠Έ 04 WHAT YOU CAN LEARN TODAY! Karpenter K8SΫϥελΦʔτεέʔ ϥͷKarpenterͱ͸ 03 Autoscalers ίετ࠷దԽʹ͓͚ΔΦʔ τεέʔϥͷར఺ 02 Optimize More! KarpenterͰΑΓίετΛ ࠷దԽ͢Δʹ͸ 05 Cost Optimization ίετ࠷దԽͱ͸? 01
36. EC2 INSTANCE PURCHASE TYPES Quick Recap
37. CHARACTERISTICS OF PURCHASE TYPES AND COST EC2 INSTANCE PURCHASE TYPES On-Demand On-Demand Instance EC2ΠϯελϯεͷఆՁɻ Spot Instance धཁͱڙڅʹԠͯ͡Ձ͕֨ௐ੔͞ΕΔɻΠϯελϯ εͷ࡟আ͕͋Δ͕ɺେ෯ʹׂҾ͔ΕΔ͜ͱ͕ଟ͍ɻ SavingsPlans / Reserved Instance 1-3೥ͷOn-Demandͷ௕ظίϛοτΛࣄલܖ໿͢Δ ͜ͱͰɺେ෯ͳׂҾΛड͚Δϓϥϯɻ࡟আ͸ͳ͍ɻ
38. CHARACTERISTICS OF PURCHASE TYPES AND COST EC2 INSTANCE PURCHASE TYPES On-Demand Instance EC2ΠϯελϯεͷఆՁɻ Spot Instance धཁͱڙڅʹԠͯ͡Ձ͕֨ௐ੔͞ΕΔɻΠϯελϯ εͷ࡟আ͕͋Δ͕ɺେ෯ʹׂҾ͔ΕΔ͜ͱ͕ଟ͍ɻ SavingsPlans / Reserved Instance 1-3೥ͷOn-Demandͷ௕ظίϛοτΛࣄલܖ໿͢Δ ͜ͱͰɺେ෯ͳׂҾΛड͚Δϓϥϯɻ࡟আ͸ͳ͍ɻ m5.large t2.medium c4.xlarge spot spot pool spot pool
39. CHARACTERISTICS OF PURCHASE TYPES AND COST EC2 INSTANCE PURCHASE TYPES On-Demand Instance EC2ΠϯελϯεͷఆՁɻ Spot Instance धཁͱڙڅʹԠͯ͡Ձ͕֨ௐ੔͞ΕΔɻΠϯελϯ εͷ࡟আ͕͋Δ͕ɺେ෯ʹׂҾ͔ΕΔ͜ͱ͕ଟ͍ɻ SavingsPlans / Reserved Instance 1-3೥ͷOn-Demandͷ௕ظίϛοτΛࣄલܖ໿͢Δ ͜ͱͰɺେ෯ͳׂҾΛड͚Δϓϥϯɻ࡟আ͸ͳ͍ɻ 1-3 years max 72% off
40. ARE WE OPTIMIZING COST ENOUGH?
41. ARE WE OPTIMIZING COST ENOUGH? Discount Plan NOT Supported https://github.com/aws/karpenter/issues/5163
42. KARPENTER USES PUBLIC ON-DEMAND AND SPOT INSTANCE RATE DISCOUNT PLAN NOT SUPPORTED KarpenterͱߪೖΦϓγϣϯ • αϙʔτ͍ͯ͠ΔߪೖΦϓγϣϯ͸ ެ։ϨʔτͷOn-DemandͱSpotͷ ΈͰɺׂҾϓϥϯʹ͍ͭͯ͸ߟྀ͞ Εͳ͍ • Reserved Instance΍Savings Plan ͸ɺSpotͷΑ͏ʹΠϯελϯε͕ಥ વ࡟আ͞ΕΔ͜ͱ΋ͳ͘ɺίετ࡟ ݮʹΑ͘࢖ΘΕ͍ͯΔΦϓγϣϯ Flexera 2023 State of the Cloud Report
43. IN CASE YOU HAVE A SAVINGS PLAN FOR $0.047/HR EXAMPLE: COST IS NOT OPTIMIZED ҎԼͷSavings PlansΛܖ໿ͨ͠৔߹… 3 years EC2 Savings Plan Compute Instance Max Savings 66% 72% t3.large Sample (ఆՁ: $0.1088) $0.062 (43%) $0.047 (57%) Locked Single Instance Family & Region No Yes t3.large, Tokyo Region, 3 years, No Upfront, Linux ※ 2023/11/15 21:00 ࣌఺ Savings Plans for K8S Nodes $0.047 t3.large t3.large 1 hour t3.large t3.large • Instance Saving PlanͷTokyo Region, t3.largeΛ3೥ؒ • 1࣌ؒ͋ͨΓ$0.047Λܖ໿ 57% off
44. IN CASE YOU HAVE A SAVINGS PLAN FOR $0.047/HR EXAMPLE: COST IS NOT OPTIMIZED c5.largeͷOn-Demand͕࡞੒͞Εͨ৔߹ Instance Type Purchase Type Price (Discount) c5.large On-demand $0.096 t3.large On-demand $0.1088 SavingsPlan $0.047 (57%) t3.large, Tokyo Region, 3 years, No Upfront, Linux ※ 2023/11/15 21:00 ࣌఺ EC2 Instance Pricing List • c5.largeͷOn-demand͸t3.largeΑΓ͍҆ • c5.large͕࡞੒͞ΕΔͱ௥ՃͰ$0.096ͷ௥Ճίετ c5.large t3.large +$0.096 $0.047 Karpenter 1 hour < Compare with c5.large and t3.large on-demand rate.
45. +$0.033 t3.large IN CASE YOU HAVE A SAVINGS PLAN FOR $0.047/HR EXAMPLE: COST IS NOT OPTIMIZED t3.largeͷSpot͕࡞੒͞Εͨ৔߹ Instance Type Purchase Type Price (Discount) t3.large On-demand $0.1088 Spot $0.033 (73%) SavingsPlan $0.047 (57%) t3.large, Tokyo Region, 3 years, No Upfront, Linux ※ 2023/11/15 21:00 ࣌఺ EC2 Instance Pricing List • t3.largeͷSpot͸t3.largeͷOn-demandΑΓ͍҆ • SpotͰ࡞੒͞ΕΔͱ௥ՃͰ$0.033ͷ௥Ճίετ t3.large $0.047 Karpenter 1 hour < Compare with t3.large spot and on-demand rate. spot
46. EXAMPLE: COST IS NOT OPTIMIZED IN CASE YOU HAVE A SAVINGS PLAN FOR $0.047/HR Unoptimized Optimized! 👍 Unused Savings Plans
47. Optimized! 👍 EXAMPLE: COST IS NOT OPTIMIZED IN CASE YOU HAVE A SAVINGS PLAN FOR $0.047/HR Unoptimized Unused Savings Plans ίετ࠷దԽʹ͸SavingsPlanͷ࢖༻཰্͕ ͕ΔΑ͏ͳߟྀ΋ඞཁ!
48. HOW DO WE IMPROVE SAVINGS PLAN UTILIZATION?
49. • Savings PlanΛߪೖͨ͠ΞΧ΢ϯτ(΋͘͠ ͸ڞ༗͍ͯ͠Δ૊৫)಺ͷଞͷk8sΫϥελ ΍ɺk8sҎ֎Ͱ࢖ΘΕ͍ͯΔΠϯελϯεͷ ߟྀ͕ඞཁ • Savings Plan͸όϥόϥͳ࣌ظʹෳ਺ߪೖ ͢Δ͜ͱ΋ଟ͍ͷͰɺͦΕͧΕͷϓϥϯͷ ঢ়گΛखಈͰ؅ཧ͢Δͷ΋େม • GetSavingsPlansUtilization API͸1೔୯Ґͳ ͷͰཻ౓͕͋Β͗͢Δ • ίετͱ࢖༻ঢ়گϨϙʔτ(AWS CUR) ͔Β1 ࣌ؒ୯Ґͷར༻཰ΛऔಘͰ͖Δ͕ɺجຊత ʹલ೔ͷσʔλʹͳΔ HOW CAN WE CALCULATE SAVINGS PLAN UTILIZATION? CHALLENGING PARTS Savings Planͷར༻཰ΛϦΞϧλΠϜʹऔ ಘ͢ΔAPI͕ଘࡏ͠ͳ͍ K8sΫϥελ಺ͷΠϯελϯε৘ใ͚ͩͰ ͸ར༻཰͕ܭࢉͰ͖ͳ͍
50. • Savings PlanΛߪೖͨ͠ΞΧ΢ϯτ(΋͘͠ ͸ڞ༗͍ͯ͠Δ૊৫)಺ͷଞͷk8sΫϥελ ΍ɺk8sҎ֎Ͱ࢖ΘΕ͍ͯΔΠϯελϯεͷ ߟྀ͕ඞཁ • Savings Plan͸όϥόϥͳ࣌ظʹෳ਺ߪೖ ͢Δ͜ͱ΋ଟ͍ͷͰɺͦΕͧΕͷϓϥϯͷ ঢ়گΛखಈͰ؅ཧ͢Δͷ΋େม • GetSavingsPlansUtilization API͸1೔୯Ґͳ ͷͰཻ౓͕͋Β͗͢Δ • ίετͱ࢖༻ঢ়گϨϙʔτ(AWS CUR) ͔Β1 ࣌ؒ୯Ґͷར༻཰ΛऔಘͰ͖Δ͕ɺجຊత ʹલ೔ͷσʔλʹͳΔ HOW CAN WE CALCULATE SAVINGS PLAN UTILIZATION? CHALLENGING PARTS Savings Planͷར༻཰ΛϦΞϧλΠϜʹऔ ಘ͢ΔAPI͕ଘࡏ͠ͳ͍ K8sΫϥελ಺ͷΠϯελϯε৘ใ͚ͩͰ ͸ར༻཰͕ܭࢉͰ͖ͳ͍ CUR͔ΒSavings Plan৘ใΛूܭ ✕ ݱࡏͷશΠϯελϯεঢ়گΛऩू → Savings Planͷར༻཰ΛγϛϡϨʔτ
51. PREDICT PLANS UTILIZATION FEEDBACK ON PREDICATED RESULTS TO KARPENTER S3 Agent Karpenter Node Pool Cost Usage Report Kubernetes Cluster 1. Process CUR & Predicate SavingsPlan Utilization Server Node Pool Node Pool 2. Send Instance Information & Get Predicate Result 3. Update NodePool according to predication Cost Usage Report Upload Create Optimal Node!
52. Demo
53. Optimize Cost KarpenterʹΑΔίετ࠷ దԽͷ͘͠Έ 04 Cost Optimization ίετ࠷దԽͱ͸? 01 Autoscalers ίετ࠷దԽʹ͓͚ΔΦʔ τεέʔϥͷར఺ 02 Optimize More! KarpenterͰΑΓίετΛ ࠷దԽ͢Δʹ͸ 05 🏆 AGENDA WHAT YOU CAN LEARN TODAY! Karpenter K8SΫϥελΦʔτεέʔ ϥͷKarpenterͱ͸ 03
54. WHAT YOU LEARNED TODAY! 1. FinOpsͰ͸ɺ୯ʹ҆͞ΛٻΊΔ͚ͩͰ͸ͳ͘ɺӡ༻ޮ཰Խ΍Մ༻ ੑͷόϥϯεΛऔͬͨίετ࠷దԽΛ໨ࢦ͢ 2. Φʔτεέʔϥ͸ɺk8sͷίετ࠷దԽͷ༗༻ͳखஈͰ͋Γɺ Karpenter͸ͦͷதͷબ୒ࢶͷҰͭ 3. Savings PlanʹରԠͨ͠Πϯελϯε͕࡞੒͞Εͳ͍ͳͲɺ୯ʹ πʔϧΛಋೖ͚ͨͩͩ͠ͱɺίετ࠷దԽͰ͖ͳ͍έʔε͕͋Δ RECAP
55. THANKS! AYA IGARASHI, @LADICLE CLOUDNATIX INC.