Kubernetes
Architecture
Big Picture View
Masters
Overvie Nodes
w Pods
Services
Deployments
The API and API
Server Recap
Up Next:
Kubernetes: Big Picture
Overview
Kubernetes: Big Picture
Overview
Tea
m
Tea
m
Tea
m
Manage
r
(coach)
Tea
m
RW HM
CH
S RB
Manage LW
r
(coach) CH AM
LB
S
S GK
1
S S
2 3
S S
Tea
m
AM
LW RW
Manage HM
r
(coach)
CH CH
LB RB
S
1 GK
S S
2 3
HTTPS HTTPS
Search Auth
K/V store MySQL
Log Load
balancer
Load
balancer
HTTPS
HTTPS
Auth
Search
Log
K/V store
MySQL
Load
balancer
HTTPS
HTTPS
Auth
Search Log
Node 1
Node 2
Node 3 K/V store
MySQL
K8s cluster
K8s cluster
k k
KV
apiserver{
k k
}
k k
K8s cluster
k k
KV
apiserver{
k k
}
k k
K8s cluster
k k
KV
<app>
apiserver{
k k
}
<YAML>
k k
Up
Next:
Masters
Master
s
Master
s
Also known as
head nodes
or the
control plane
Multi-master H/A control plane
Failure domain Failure domain Failure domain
A B C
H/A
Sweet spot
Desig
n
…
Follower Follower Leader
Hosted K8s control plane
Hosted K8s control plane
Performance
Availability
Updates/Upgrades
API Server Flags
…
Hidden
Exposed
Highly available API
Endpoint
K8s cluster
Cluster ops User/business apps
kube-apiserve
r KV
{}
Fe
kube-apiserve
r• Front-end to the KV
control plane
• Exposes the API
(REST)
apiserver{}
• Consumes JSON/YAML
Fe
Cluster
Store
• Persists cluster state Cluster store
and config KV
• Based on
etcd
• Performance is apiserver{}
critical
• Have recovery plans Fe
in place
Kube-controller-manager
• Controller of controllers Cluster store
• Node controller KV
• Deployment controller Controllers
• Endpoints/EndpointSlice
controller…
apiserver{}
• Watch loops
Fe
• Reconciles observed state
with desired state
Kube-scheduler
• Watches API Server for Cluster store
new work tasks KV
Scheduler Controllers
• Assigns work to cluster
nodes
• Affinity/Anti-affinity apiserver{}
• Constraints
• Taints Fe
• Resources…
Cluster store
KV
Scheduler Controllers
$kubectl apiserver{}
Fe
Up
Next:
Nodes
Nodes
k
Fe
k
Kubelet
• Main Kubernetes
agent
• Registers node with cluster
• Watches API Server for
k
work tasks (Pods)
Fe
• Executes Pods
• Reports back to Masters
Container runtime
• Can be
Docker
• Pluggable: Container
Runtime Interface (CRI)
k
• Docker, containerd, CRI-O, Fe
Kata…
• Low-level container intelligence
Kube-proxy
• Networking component
• Pod IP k
addresses
Fe
Kube-proxy
Stable Name and
IP
Kube-proxy
• Networking component k
• Pod IP addresses
• Basic Fe
load-balancing
k
k
Fe
k Kubelet
Main K8s agent
Container runtime
Docker, containerd, CRI-O, k
more…
Kube-proxy
Vital role in networking Fe
Nodeless Kubernetes
Pods run on cloud’s hosted container back-end
Up Next:
The Declarative Model & Desired
State
The Declarative Model and Desired
State
Declarative
model
Describe what you
want (desired state)
in a manifest
file
KV
Manifest
Desired apiserver{}
state
Fe
KV
apiserver{}
Fe
kind: Kitchen
spec:
type: New
location: Rear
style: OpenPlan
heating:
type: Underfloor
medium: Water
windows:
type: FloorToCeiling
aspect: South
doors:
type: FireDoor
accessTo: Garage
island: Yes
roofGarden: Yes
…
apiVersion: apps/v1
kind: Deployment
metadata:
name: test
spec:
replicas: 6
selector:
matchLabels: Declarative
app: ps-test (Declaring what you want)
template:
spec:
containers:
- name: c1
image: web1:1.3
ports: 8080
- containerPort
...
name: web
Desired state: 3
replicas: 3
Observed state:
...
3
Web Web Web
name: web
Desired state: 3
replicas: 3
Observed state:
...
3
Web Web
name: web
Desired state: 3
replicas: 3
Observed state:
...
2
Web Web
name: web
Desired state: 3
replicas: 3
Observed state:
...
2
Web Web Web
name: web
Desired state: 3
replicas: 3
Observed state:
...
3
Web Web Web
apiserver{}
name: web apiserver{}
replicas: 3
...
apiserver{}
apiserver{}
Web
apiserver{}
Web Web
apiserver{}
Web Web Web
apiserver{}
Desired state: 3
Observed state:
3
Web Web Web
apiserver{}
Desired state: 3
Observed state:
3
Watch
loops
Web Web Web
Up Next:
The Mighty Pod
The Mighty Pod
Virtual Machine Container Pod
Atomic units of scheduling
<app>
<hlpr> <app>
<hlpr> <app>
<hlpr> <app>
Kubernetes for Developers:
Integrating Volumes and Using Multi-container Pods
[S h a r e d e x e c u t i o n e n v]
Shared
access
[S h a r e d e x e c u t i o n e n v]
Shared
access
[S h a r e d e x e c u t i o n e n v]
8080 9090
10.0.0.17
8080 9090
8080 9090
localhost
8080 9090
localhost
Shared
access Network access
Tightly coupled Loosely coupled
When two containers (app When two containers (app
absolutely need don’t
services) to services) absolutely need
share vols, memory etc. to share resources
Shared
access Network access
Tightly coupled Loosely coupled
When two containers (app When two containers (app
absolutely need don’t
services) to services) absolutely need
share vols, memory etc. to share resources
Scaling
App ContaMinesrh
Container
Net
App Container Mesh Container
Main app ctr Helper ctr
App Container Mesh Container
Pod scheduled to single node
Fe
Phase: Phase: Phase: succeeded/failed
pending running
<app>
<app>
Annotations
Labels
Policies
Resources
Co-scheduling containers
…
Up Next:
Stable Networking with
Services
Stable Networking with
Services
<app>
10.0.0.5
<app>
10.0.0.7
5
<app>
10.0.0.7
<app>
10.0.0.7
<app> <app> <app> <app> <app>
10.0.0. 10.0.0. 10.0.0. 10.0.0.1 10.0.0.1
7 8 9 0 1
<app> <app> <app>
10.0.0. 10.0.0. 10.0.0.
7 8 9
10.0.0. 10.0.0. 10.0.0. 10.0.0.1
7 8 9 0
<app> <app> <app> <app>
App/
client
10.0.0. 10.0.0. 10.0.0. 10.0.0.1
7 8 9 0
<app> <app> <app> <app>
Unreliable Pod IPs
10.0.0. 10.0.0. 10.0.0. 10.0.0.1
7 8 9 0
<app> <app> <app> <app>
Unreliable Pod IPs
10.0.0. 10.0.0. 10.0.0. 10.0.0.1
7 8 9 0
<app> <app> <app> <app>
web web
Stable name and
IP
Load balancing
10.0.0. 1100..00
8 ..00..19
<app> 0 <app>
web web
Stable name and
IP
Load balancing
10.0.0. 10.0.0. 1100..00 10.0.0.1
7 8 ..00..19 1
<app> <app> 0 <app> <app>
web web
Stable name and
IP
Load balancing
1100..00 1100..00 10.0.0.1 10.0.0.1
..00..17 ..00..18 15
4
2 <app> 3 <app> 0 <app> <app>
Labels Labels are very simple and very
powerful!
web web
Stable name and
IP 10.0.0.9
0
pro pod <app>
d
pro
be
Load balancing1.3 d
be
1.3
10.0.0.1 10.0.0.1 10.0.0.1 10.0.0.1
2 3 4 5
<app> <app> <app> <app>
pro pro pro pro
d d d d
be be be be
1.3 1.3 1.3 1.3
web web
Stable name and
IP
pro
d
be
1.3
Load balancing
10.0.0.2 10.0.0.1 10.0.0.1 10.0.0.2
0 3 4 1
pod <app> <app> <app> pod <app>
pro pro pro pro
d d d d
be be be be
1.4 1.3 1.3 1.4
web web
Stable name and
IP
pro
d
be
1.4
Load balancing
10.0.0.2 10.0.0.1 10.0.0.1 10.0.0.2
0 3 4 1
pod <app> <app> <app> pod <app>
pro pro pro pro
d d d d
be be be be
1.4 1.3 1.3 1.4
web web
Stable name and
IP
pro
d
be
1.4
Load balancing
10.0.0.2 10.0.0.1 10.0.0.1 10.0.0.2
0 3 4 1
pod <app> <app> <app> pod <app>
pro pro pro pro
d d d d
be be be be
1.4 1.3 1.3 1.4
Only sends traffic to healthy Pods
Can do session affinity
Can send traffic to endpoints outside the cluster
Can do TCP and UDP
Up Next:
Game-changing
Deployments
Game-changing
Deployments
Advanced features
No advanced features <app>
Infrastructure Masters & Nodes
Stateless Stateful apps…
apps
One instance on every Time-based short-lived jobs
node
Deployment Controller/Reconciliation
loop
Watches API Server for new
Deployments
Implements them
observed state
Constantly compares with
desired state
Control plane
Cluster
store
apiserver{}
Control plane
Cluster
store
4
apiserver{}
Control plane
Cluster
4 store
4
apiserver{}
Control plane
Cluster
4 store
4
apiserver{}
Node 1 Node 2 Node 3 Node 4
Control plane
Cluster
4 store
4
Desired state
V apiserver{}
Observed state
Node 1 Node 2 Node 3 Node 4
Control plane
DEFCON
Cluster
1
4 store
4
Desired state
V apiserver{}
Observed state
pod
Node 1 Node 2 Node 3 Node 4
Control plane
DEFCON
Cluster
5
4 store
4
Desired state
V apiserver{}
Observed state
You/me
Node 1 Node 2 Node 3 Node 4
Manages updates, rollbacks…
Replica count, self-healing, previous
versions…
Labels, annotations, co-scheduling…
<app>
Control plane
Cluster
store
apiserver{}
Control plane
Cluster
store
Desired
state
apiserver{}
Control plane
Cluster
Desired
store
state
Desired
state
apiserver{}
Control plane
Cluster
Desired
store
state
Desired
state
apiserver{}
Node 1 Node 2 Node 3 Node 4
Control plane
Cluster
Desired
store
state
Desired
state
Desired state
V apiserver{}
Observed state
Node 1 Node 2 Node 3 Node 4
Up Next:
The K8s API and API
Server
The K8s API and API
Server
Atomic unit of scheduling Replica count
Updates and rollbacks Stable network abstraction
K8s cluster
{ API
}
api
…
K8s cluster
{ API
} (catalog)
api
…
K8s cluster
{ API
} (catalog)
api
Networking
…
K8s cluster
{ API
} (catalog)
api
Networking
Stateful apps
…
K8s cluster
{ API
} (catalog)
api
…
K8s cluster
{ API
} (catalog)
api
K8s cluster
{ API
} (catalog)
api
Deployment object definition
• API sub-group: [apps/v1]
• replicas (integer)
• minReadySeconds (integer)
• progressDeadlineSeconds (integer)
• paused (boolean)
• revisionHistoryLimit (integer)
• selector (LabelSelector)
• strategy (DeploymentStrategy)
• template (PodTemplateSpec)
K8s cluster
{ API
} (catalog)
api
HTTP methods/verbs
apiserver{} $ kubectl
REST
HTTPS
…
K8s cluster
{ API
} (catalog)
api
core workloads/apps
HTTP methods/verbs
apiserver{} $ kubectl
storage.k8s.io networking.k8s.io REST
HTTPS
More…
Up Next:
Epic Recap
Epic Recap
Tea
m
Tea
m
Tea
m
Manage
r
(coach)
Tea
m
RW HM
CH
S RB
Manage LW
r
(coach) CH AM
LB
S
S GK
1
S S
2 3
S S
Tea
m
AM
LW RW
Manage HM
r
(coach)
CH CH
LB RB
S
1 GK
S S
2 3
K8s cluster
Control plane Worker nodes
K8s cluster
Control plane Worker nodes
H/A Masters Nodes
K8s cluster
Control plane Worker nodes
Gateway to API { { {
Nodes
} } }
api api api
Cluster store
(state)
Scheduler (assigns
work) Controllers (watch
loops)
H/A Masters
K8s cluster
Control plane Worker nodes
Gateway to API { { { k k k k Main K8s agent
} } } Container runtime
api api api
Cluster store
(state) Kube-proxy (networking)
Scheduler (assigns work)
Nodes
Controllers (watch loops)
H/A Masters
K8s cluster
Control plane
Worker nodes
{ { {
} } }
api api api
k k k k
H/A Masters Nodes
Up Next:
Getting
Kubernetes
