Upgrade to Pro — share decks privately, control downloads, hide ads and more …

KubeCon 2019 NAから見るServerless × k8sの現状 / State of Serverless x k8s in KubeCon 2019 NA

KubeCon 2019 NAから見るServerless × k8sの現状 / State of Serverless x k8s in KubeCon 2019 NA

2019/12/14に開催されたServerlessDays Fukuoka 2019での萬治による登壇資料です

LINE Developers
PRO

December 14, 2019
Tweet

More Decks by LINE Developers

Other Decks in Programming

Transcript

  1. KubeCon 2019 NAから見る
    Serverless × k8sの現状
    Wataru Manji - Verda Reliability Engineering, LINE Corp.
    2019.12 Serverless Day Fukuoka

    View Slide

  2. Overview: KubeCon 2019 NA
    Topics: Serverless × k8s
    Sessions overview: Knative and Kubeflow
    Impressions
    Knative Usecase on Verda
    01
    02
    03
    04
    05
    Agenda
    2

    View Slide

  3. 3
    Wataru Manji
    Engineer - LINE Corp.
    2019.06~: Verda Reliability Engineering
    Roles:
    - OpenStack Service Monitoring
    - Networking Monitoring
    - DevOps Design / Management
    Verda: LINE’s Private Infra Platform
    About me

    View Slide

  4. SanDiego: 2019/11/18 ~ 11/21
    Overview of KubeCon 2019 NA
    4
    - The world's largest event for Kubernetes and CloudNative
    - Attendees: over 10,000 (really?)
    - Sessions: about 400
    - Major Topics: Scaling, Security, Networking, Devops, Monitoring, ...

    View Slide

  5. View Slide

  6. Knative, Kubeflow and KEDA
    Topic: Serverless × k8s
    6
    - Quite few sessions that tagged serverless (about 1%)
    - Most of the k8s users are using services such as GKE, and
    aren't thinking to manage the serverless service on their own.
    - The following 3 products were talked about:
    - Knative
    - Kubeflow
    - KEDA
    - Knative, Kubeflow sessions had many practical contents

    View Slide

  7. K8s-based platform to build, deploy, and manage modern serverless workloads
    Knative
    7
    - Provides a FaaS platform that runs on k8s
    - Used as the backend of Cloud Run on GCP
    - Managed by Google

    View Slide

  8. Knative
    Sessions Summary
    8
    - Cloud Native Smart Contract with Knative Jay Guo & Ying Chun Guo, IBM
    - As the title
    - Serverless Platform for Large Scale Mini-Apps:
    From Knative to Production Yitao Dong & Ke Wang, Ant Financial
    - Accelerate containers running on Knative with hot stanby

    View Slide

  9. Deploy and manage an ML stack on Kubernetes
    Kubeflow
    9
    - Builds a convenient env set on k8s
    for machine learning operators
    (likes TensorFlowExtended)
    - Provides multi-step ML workflow
    - Abstract model handling

    View Slide

  10. Kubeflow
    Sessions Summary
    10
    - Nuclio : KubeFlow Serverless’s component Orit Nissan-Messing, Iguazio
    - In addition to Kubeflow, examples and demonstrations of
    performing processing at high speed using Serverless data I/O
    component called Nuclio were shown.

    View Slide

  11. K8s-based Event Driven Auto-scaling component
    KEDA
    11
    - Provides a FaaS platform that runs on k8s
    (likes Knative)
    - Supports events issued by Kafka and etc..
    - Developed by Microsoft

    View Slide

  12. KEDA
    Sessions Summary
    12
    - KEDA: Event Driven and Serverless Containers in Kubernetes Jeff Hollan, Microsoft
    - KEDA function introduction

    View Slide

  13. Serverless Platform for Large Scale Mini-Apps: From Knative to Production
    https://kccncna19.sched.com/event/Uad9/serverless-platform-for-large-scale-mini-apps-from-knative-to-production-yitao-dong-ke-wang-ant-financial
    Interested Session:
    13
    - Reduced the time from event issuance to pod activation to 1/10
    - Implemented pool based pre-warming
    - Reduce costs per tenant by sharing Knative controllers among tenants
    - Implemented multitenancy in cluster by namespace and RBAC
    - Scalability for large clusters
    - Virtually sharding a Cluster and setting up a each controller

    View Slide

  14. Serverless × k8s
    Impressions
    14
    - Why bother to implement Serverless on theirs k8s?
    → I think there are two reasons:
    - They want to easily manage workloads that require a lot of
    resources.
    - They want to make effective use of resources in their cluster.
    - I think it is still immature but is developing at high speed
    - If you use both Serverless and k8s on Public Cloud etc.,
    I think it is worth paying attention.

    View Slide

  15. What is Verda
    Knative Usecase on Verda
    15
    Verda is LINE’s Private Infra Platform
    ・about 45,000 VMs
    ・about 10,000 baremetal server
    ・k8s as a service
    ・function as a service
    ・various managed services

    View Slide

  16. What is Verda's motivation
    Knative Usecase on Verda
    16
    IaaS functions are enriched
    but there are few functions for automating operations with event-base
    → Provide FaaS for centrally managing events
    and promoting operational automation

    View Slide

  17. Side-effect
    Knative Usecase on Verda
    17
    VMs are heavily used, but the average load is not high…
    (The cost per VM is not cheap!)
    → VMs just for executing batches should be replaced with FaaS
    more detail: https://speakerdeck.com/line_devday2019/cloud-native-challenges-in-private-cloud-with-k8s-knative

    View Slide

  18. Currently Status of VFS (Verda Function Service)
    Knative Usecase on Verda
    18
    - Recently released in Production
    - Supported events:
    - HTTP, Cron
    - Change server and network status
    - Change project and member information
    - etc
    - Active in event-driven operations and batch processing

    View Slide

  19. Next challenges
    Knative Usecase on Verda
    19
    - Increase support runtime
    - Increase support events
    - Maintenance notice
    - Monitoring alert
    - etc
    - Function Pipeline

    View Slide

  20. Interested in LINE's Infra platform?
    We are Hiring!!!
    20
    Verda in LINE Fukuoka:
    - Network
    > Software L7 load balancer development
    - Infra software
    > Monitoring and provisioning automation development
    - SRE
    > Operation, automation and user support

    View Slide

  21. THANK YOU

    View Slide