The declarative programming model of Kubernetes is markedly different from what most developers are used to. That the API is a set of resources rather than a list of methods on objects is a bit mind bending. But this programming model is not entirely new – rather, it smacks quite heavily of functional programming. Functional programming had mostly been relegated to academic endeavors until recently. What’s changed that is that our apps are now distributed systems and are simply too complex for us to reason about without help. Kubernetes helps. In order to effectively use Kubernetes to deploy and manage your workloads you need to understand some of the principles of functional programming and how they surface in K8s. In this session I will cover these underlying principles of the K8s programming model so that you can up the robustness and manageability of your application deployments. Presented at KubeCon Barcelona, May 2019

1. You Might Just be a
Functional Programmer Now
Cornelia Davis
VP of Technology
Pivotal
@cdavisafc
Kubecon Barcelona
May 2019

2. Me?
❖ Cal State Northridge: Computer Science
❖ Indiana University: Computer Science
(ABD)
❖ Pivotal: Product Strategy (PCF, PCC, PKS)
❖ And now…
Ops
as much as
Dev
now
Discount code 40% off!: ctwkubeconeur19

3. Assembly Language
Fortran
C
C++
Java
Golang
Clojure
Scala
Kotlin
Functional
Programming
Imperative
Programming
F#

4. [(0, 5), (5, 17), (17, 20), (20, 33)]
Merge short Segments together
[(0, 17), (17, 33)]

5. def mergeSegments(s_list, threshold):
list_len = len(s_list)
i = 0
while i < list_len:
seg_length = s_list[i][1]-s_list[i][0]
if seg_length < threshold:
s_list[i+1] =
(s_list[i][0], s_list[i+1][1])
del s_list[i]
list_len -= 1
else:
i += 1

6. 0 5 5 17 17 20 20 33
i
0 1 2 3
4
List Length
0
0 17 17 20 20 33
17
i
3
0 17 17 20 20 33
i
3
0 17 17 33
i
2

7. Hallmarks of Imperative Programs
❖ Sequential - we control every step in the process.
❖ Variables … with side effects
❖ Hairy edge cases
❖ Difficult to parallelize
@cdavisafc

8. (define (mergeSegments segs threshold)
(if
(or (null? segs) (null? (cdr segs)))
segs
(let
((rest (mergeSegments (cdr segs) threshold)))
(if
(< (currSegLen segs) threshold)
(cons (cons (start segs) (end rest))
(cdr rest))
(cons (car segs) rest)))))

9. 0 5 5 17 17 20 20 33
5 17 17 33
0 17
Merge short Segments together
17 33

10. Hallmarks of Functional Programs
❖ Declarative
❖ NO side effects
❖ Recursive
❖ Easier to program in a way that allows for parallelization
❖ Far fewer edge cases
❖ Provably correct
@cdavisafc

11. Gene Kim
3000 Lines of Objective C
To
1500 Lines of Typescript & React
To
500 Lines of Clojure

12. What is the
model
we use to reason with?

13. Imperative Functional
Iterative Declarative
Mutable State Immutability
Looping Recursive
Libraries (compiled in) Aspect Oriented Prog.
Maybe not functional, but modern
programming practice
*
*
Programming

14. Imperative Functional
Iterative Declarative
Mutable State Immutability
Looping Recursive
Libraries (compiled in) Aspect Oriented Prog.*
Programming

15. We cannot reason about every detail so we need
a model
that allows machines to reason for us.

16. Bash
Puppet
Chef
Systems
Programming
Ansible
Salt
Imperative

17. Kubernetes
BOSH
CF/PAS
Systems
Programming
Functional

18. Scripted Deployments Declarative Deployments
ssh Immutable Infrastructure
Long Running Context (Ephemeral) Containers
Middleware Sidecars
We cannot reason about every detail so we need
a model
that allows machines to reason for us.
Imperative Functional
Systems Programming

19. Declarative Deployments
@cdavisafc

20. My app:
- image: posts
type: pod
replicas: 3
- image: connections
type: pod
replicas: 5
- image: aggregator
type: pod
replicas: 2
Compliance:
- image: audit
type: daemonset
P
P
P
CC
C
C
C
A
A
Audit
Audit Audit
Audit

21. My app:
- image: posts
type: pod
replicas: 3
- image: connections
type: pod
replicas: 5
- image: aggregator
type: pod
replicas: 2
Compliance:
- image: audit
type: daemonset
P
P
P
CC
C
C
A
Audit
Audit Audit
C
A
AuditA
C
Control Loop

22. My app:
- image: posts
type: pod
replicas: 3
- image: connections
type: pod
replicas: 5
- image: aggregator
type: pod
replicas: 2
Compliance:
- image: audit
type: daemonset
P
P
P
CC
C
C
A
Audit
Audit Audit
A
C
Independent Control Loops!!!
ReplicaSet Controller
DaemonSet Controller

23. (define sumToN
(lambda (n)
(if (= n 1)
1
(+ n (sumToN (- n 1)))
)))
A bit like Functional Programming
Base Case
General Case
(Bit of a leap of faith)

24. Immutable Infrastructure
@cdavisafc

25. provision
Infrastructure
configure deploy configure
Pipelines,
not humans…

26. Challenges Remain
Slow Shortcuts
ssh (eeghads!)
Complex Dependencies

27. Configured
PLATFORM
provision
Infrastructure
configure deploy configure

28. Infrastructure
Configured PLATFORM
deploy configure
MillisecondsDesigned for
self-service!

29. configureconfigureconfigureconfigure
deploydeploydeploydeploy
Infrastructure
Configured PLATFORM
deploy configure
Designed for
self-service!

30. configureconfigureconfigureconfigure
deploydeploydeploydeploy
Infrastructure
Configured PLATFORM
deploy
configuredeploy
Designed for
self-service!

31. configureconfigureconfigureconfigure
deploydeploydeploydeploy
Infrastructure
Configured PLATFORM
deploy
configuredeploy
Designed for
self-service!

32. But what does all of this have to do with
Imperative and Functional Models?
@cdavisafc

33. 0 5 5 17 17 20 20 33
i
0State:

34. 5 17 17 20 20 33
i
0
17
State:

35. 5 17 20 330
17
i
State:

36. 5 17 20 330
17
i
State:
def mergeSegments(s_list, threshold):
list_len = len(s_list)
i = 0
while i < list_len:
seg_length = s_list[i][1]-s_list[i][0]
if seg_length < threshold:
s_list[i+1] =
(s_list[i][0], s_list[i+1][1])
del s_list[i]
list_len -= 1
else:
i += 1
Correctness?
?
Other Influences?

37. The call stack captures a
series of events from
which state can be derived
Frames are
immutable!
0 17 17 33
Final
State:

38. Frames are
immutable!
0 17 17 33
Final
State:
The call stack captures a
series of events from
which state can be derived

39. Container images
are a bit like your
stack frames…
… with the
controller as your
interpreter

40. event store
… …
There is another place
we are seeing this
pattern…
Events are
immutable and
sequenced!
Legacy Store

41. (Ephemeral) Containers
@cdavisafc

42. HOST
Host OS
(Kernel)
OS Image
Runtime Layer
Application Layer

43. HOST
Host OS
(Kernel)
OS Image
Runtime Layer
Application Layer

44. HOST
Host OS
(Kernel)
OS Image
Runtime Layer
Application Layer
OS Image
Runtime Layer
Application Layer

45. OS Image
Runtime Layer
Application Layer
HOST
Host OS
(Kernel)
OS Image
Runtime Layer
Application Layer
HOST
Host OS
(Kernel)
You can
Repave
the entire
environment
VERY Often!!!
Like several
times per week
“often”!

46. Sidecars
@cdavisafc

47. Database
Web
Server
Messaging
Your Application Code
Virtualized
Infrastructure
Platform Team
Application Team
Iteratively building and
delivering digital offerings to
the consumer
Enabling the app teams all
while maintaining
Security
Compliance
Resilience
Cost Efficiency
47
Platform

49. POD
HOST
Host OS
(Kernel)
SidecarApp

52. POD
HOST
Host OS
(Kernel)
SidecarApp

53. POD
HOST
Host OS
(Kernel)
SidecarApp
POD
Daemon Set
(Sidecar)

54. D
D
D
D

55. It’s about injecting cross-cutting concerns
(As it was for AOP)
@cdavisafc

56. configureconfigureconfigureconfigure
deploydeploydeploydeploy
provision
Infrastructure
Configured PLATFORM
deploy
configuredeploy
Cross-cutting
Concerns
Cross-cutting
Concerns

57. Systems Programming
Scripted Deployments Declarative Deployments
ssh Immutable Infrastructure
Long Running Context (Ephemeral) Containers
Middleware Sidecars
(Declarative)
(Immutable)
(Recursive)
(AOP)

58. Systems ProgrammingFunctional
We cannot reason about every detail so we need
a model
that allows machines to reason for us.

59. Thank you
@cdavisafc