Databases On Kubernetes: Why You should care

Developers always expected databases to work out-of-the-box, but historically it is the exact opposite.

With the rise of Kubernetes StatefulSets and CRDs, we started thinking about running databases on it. But why should I do that in the first place? How hard is it? Which are the challenges? Is it production already? All those questions will be answered during a live demo where we will deploy a database, deploy an operator, fail nodes, scale up and down with almost no manual intervention.

Transcript

1. Confidential and Proprietary. Do not distribute without Couchbase consent. © Couchbase 2017. All rights reserved.
   DATABASES ON KUBERNETES
   Denis Rosa | Developer Advocate
   @deniswsrosa
   @deniswsrosa
   WHY YOU SHOULD CARE
   

   View Slide

2. Confidential and Proprietary. Do not distribute without Couchbase consent. © Couchbase 2017. All rights reserved. 2
   2016
   

   View Slide

3. Confidential and Proprietary. Do not distribute without Couchbase consent. © Couchbase 2017. All rights reserved.
   @deniswsrosa
   2016
   Containers are designed to be
   stateless
   

   View Slide

4. Confidential and Proprietary. Do not distribute without Couchbase consent. © Couchbase 2017. All rights reserved.
   @deniswsrosa
   Fast forward to 2020…
   

   View Slide

5. Confidential and Proprietary. Do not distribute without Couchbase consent. © Couchbase 2017. All rights reserved. 5
   2020
   

   View Slide

6. Confidential and Proprietary. Do not distribute without Couchbase consent. © Couchbase 2017. All rights reserved.
   @deniswsrosa
   What have changed?
   

   View Slide

7. Confidential and Proprietary. Do not distribute without Couchbase consent. © Couchbase 2017. All rights reserved.
   @deniswsrosa
   • Production-Grade Images
   • Easier to setup and configure
   • Separate storage from compute
   • Kubernetes and StatefulSets …
   

   View Slide

8. Confidential and Proprietary. Do not distribute without Couchbase consent. © Couchbase 2017. All rights reserved.
   @deniswsrosa
   Databases typically have numerous tuning parameters,
   a new immutable container image for every possible
   database configuration can quickly result in image
   sprawl.
   

   View Slide

9. Confidential and Proprietary. Do not distribute without Couchbase consent. © Couchbase 2017. All rights reserved.
   @deniswsrosa
   Using configuration management tools can
   help to solve part of the problem, but shouldn’t
   it be part of the solution?
   

   View Slide

10. Confidential and Proprietary. Do not distribute without Couchbase consent. © Couchbase 2017. All rights reserved.
    DEMO1 – A K8S TODO LIST
    

    View Slide

11. Confidential and Proprietary. Do not distribute
    without Couchbase consent. © Couchbase 2017.
    All rights reserved. 11
    K8 CRDs
    

    View Slide

12. Confidential and Proprietary. Do not distribute without Couchbase consent. © Couchbase 2017. All rights reserved. 12
    CRDs + K8s Operators
    

    View Slide

13. Confidential and Proprietary. Do not distribute without Couchbase consent. © Couchbase 2017. All rights reserved. 13
    Java Operator - Example
    Source: https://developers.redhat.com/blog/2019/10/07/write-a-simple-kubernetes-operator-in-java-using-the-fabric8-kubernetes-client/
    

    View Slide

14. Confidential and Proprietary. Do not distribute without Couchbase consent. © Couchbase 2017. All rights reserved. 14
    Quick Operator Revision
    Operator
    Knowledge from
    app experts
    Deployments
    Autoscaling
    StatefulSets
    Configs
    

    View Slide

15. Confidential and Proprietary. Do not distribute without Couchbase consent. © Couchbase 2017. All rights reserved. 15
    K8s Database Operators
    Provide a cloud agnostic
    Database-as-a-Service solution
    

    View Slide

16. Confidential and Proprietary. Do not distribute without Couchbase consent. © Couchbase 2017. All rights reserved.
    DEMO 2 – LET’S PLAY WITH A
    DATABASE ON KUBERNETES
    

    View Slide

17. Confidential and Proprietary. Do not distribute without Couchbase consent. © Couchbase 2017. All rights reserved. 17
    Other Possibilities
    Automated
    Backup
    XDCR
    Prometheus
    Metrics
    

    View Slide

18. Confidential and Proprietary. Do not distribute
    without Couchbase consent. © Couchbase 2017.
    All rights reserved. 18
    Standard
    Deployment
    

    View Slide

19. Confidential and Proprietary. Do not distribute without Couchbase consent. © Couchbase 2017. All rights reserved. 19
    MySQL Operator
    https://github.com/presslabs/mysql-operator
    

    View Slide

20. Confidential and Proprietary. Do not distribute without Couchbase consent. © Couchbase 2017. All rights reserved. 20
    DBaaS vs Databases on K8S
    • Easy;
    • Low TCO for small/medium workloads*;
    • Lacks flexibility*;
    • Average performance;
    DBaaS DB on K8S
    • Still requires some K8s knowledge;
    • Same flexibility as bare metal;
    • Operators can automate most of the
    hard work;
    • One of the best options for databases
    with large number of nodes;
    • Potentially the default option in the
    near future;
    

    View Slide

21. Confidential and Proprietary. Do not distribute without Couchbase consent. © Couchbase 2017. All rights reserved. 21
    DBaaS
    

    View Slide

22. Confidential and Proprietary. Do not distribute without Couchbase consent. © Couchbase 2017. All rights reserved. 22
    Kubernetes Storage
    https://github.com/brianfrankcooper/YCSB/wiki/Core-Workloads
    Ephemeral Remote
    Persistent
    Local
    Persistent
    Easiest one, but you lose
    all data if the pod gets
    killed.
    Latency limits the
    performance of
    applications with intensive
    read/write profiles or
    distributed datastores
    Not a general purpose
    storage solution
    

    View Slide

23. Confidential and Proprietary. Do not distribute without Couchbase consent. © Couchbase 2017. All rights reserved. 23
    Local Persistent Storage
    https://github.com/brianfrankcooper/YCSB/wiki/Core-Workloads
    

    View Slide

24. Confidential and Proprietary. Do not distribute without Couchbase consent. © Couchbase 2017. All rights reserved. 24
    Local Persistent Storage - Trade offs
    https://github.com/brianfrankcooper/YCSB/wiki/Core-Workloads
    • Inflexible placements
    • Lower availability
    • Lower data durability (not replicated)
    • Requires Node/disk preparation
    

    View Slide

25. Confidential and Proprietary. Do not distribute without Couchbase consent. © Couchbase 2017. All rights reserved. 25
    Local Persistent Storage - Trade offs
    https://github.com/brianfrankcooper/YCSB/wiki/Core-Workloads
    It is ok for databases as they already have data
    replication internally
    

    View Slide

26. Confidential and Proprietary. Do not distribute without Couchbase consent. © Couchbase 2017. All rights reserved. 26
    Host Path vs Local Persistent Storage
    https://github.com/brianfrankcooper/YCSB/wiki/Core-Workloads
    Allows your pod to specify any path in
    your disk
    Not portable
    Not disk accountable
    Hard to use at scale
    Host Path
    

    View Slide

27. Confidential and Proprietary. Do not distribute without Couchbase consent. © Couchbase 2017. All rights reserved. 27
    Host Path vs Local Persistent Storage
    https://github.com/brianfrankcooper/YCSB/wiki/Core-Workloads
    Leverages the persistent volumes
    interface
    Security: Only adms can create/define
    the content of the PVs
    Portability
    Disk accountability
    StatefulSets
    Local Persistent Storage
    

    View Slide

28. Confidential and Proprietary. Do not distribute without Couchbase consent. © Couchbase 2017. All rights reserved. 28
    Bare Metal Vs Containers
    https://github.com/brianfrankcooper/YCSB/wiki/Core-Workloads
    Yahoo! Cloud Serving Benchmark
    YCSB includes a set of core workloads that define a basic benchmark for cloud systems
    

    View Slide

29. Confidential and Proprietary. Do not distribute without Couchbase consent. © Couchbase 2017. All rights reserved. 29
    Workload A
    https://blog.couchbase.com/performance-certification-of-couchbase-server-5-5-on-kubernetes-platform/
    Workload A
    Update heavy workload: Mix of 50/50 reads and writes
    Workload E
    Short ranges: Short ranges of records are queried, instead of
    individual records.
    ~0%
    No substantial difference
    <10%
    Performance penalty
    

    View Slide

30. Confidential and Proprietary. Do not distribute without Couchbase consent. © Couchbase 2017. All rights reserved. 30
    Workload A
    https://blog.couchbase.com/performance-certification-of-couchbase-server-5-5-on-kubernetes-platform/
    

    View Slide

31. Confidential and Proprietary. Do not distribute without Couchbase consent. © Couchbase 2017. All rights reserved. 31
    Workload E
    https://blog.couchbase.com/performance-certification-of-couchbase-server-5-5-on-kubernetes-platform/
    

    View Slide

32. Confidential and Proprietary. Do not distribute without Couchbase consent. © Couchbase 2017. All rights reserved. 32
    Bare Metal Vs Containers
    https://blog.couchbase.com/performance-certification-of-couchbase-server-5-5-on-kubernetes-platform/
    Worst Case: Add an extra node to compensate for
    potential performance penalties
    

    View Slide

33. Confidential and Proprietary. Do not distribute without Couchbase consent. © Couchbase 2017. All rights reserved. 33
    Other Operators
    @deniswsrosa
    

    View Slide

34. Confidential and Proprietary. Do not distribute without Couchbase consent. © Couchbase 2017. All rights reserved. 34
    Other Operators
    https://www.youtube.com/watch?v=J7h0F34iBx0 https://www.youtube.com/watch?v=Zn1vd7sQ_bc
    @deniswsrosa
    

    View Slide

35. Confidential and Proprietary. Do not distribute without Couchbase consent. © Couchbase 2017. All rights reserved.
    THANKS!
    @DENISWSROSA
    

    View Slide