Kubernetes at a glance
WARN! Kubernetes is NOT for you!
Extracted from @[https://pythonspeed.com/articles/dont-need-kubernetes/]
"""...
BºMicroservices are an organizational scaling technique: when you have º
Bº500 developers working on one live website, it makes sense to pay the º
Bºcost of a large-scale distributed system if it means the developer º
Bºteams can work independently. So you give each team of 5 developers a º
Bºsingle microservice, and that team pretends the rest of the º
Bºmicroservices are external services they can’t trust. º
RºIf you’re a team of 5 and you have 20 microservices, and you º
Rºdon’t have a very compelling need for a distributed system, º
Rºyou’re doing it wrong. Instead of 5 people per service like the big º
Rºcompany has, you have 0.25 people per service. º
"""
- Extracted from Local Persistent Volume documentation:
""" Because of these constraints, Rºit’s best to exclude nodes with º
Rºlocal volumes from automatic upgrades or repairsº, and in fact some
cloud providers explicitly mention this as a best practice. """
Basically most of the benefits of k8s are lost for apps managing
storage at App level. This is the case with most modern DDBBs and stream
architectures (PostgreSQL, MySQL, kafka, p2p-alike Ethereum, ....).
- As a reference Google, original creator of K8s, launches
Bº2 billion containers per week!!!º in its infrastructures.
This is not excatly a normal IT load.
- See also:
@[https://en.wikipedia.org/wiki/Fallacies_of_distributed_computing]
1.- "The network is reliable"
2.- "Latency is zero"
3.- "Bandwidth is infinite"
4.- "The network is secure"
5.- "Topology doesn't change"
6.- "There is one administrator"
7.- "Transport cost is zero"
8.- "The network is homogeneous"
WARN! Kubernetes is for you!
- You plan to have a running infrastructure for years,
and you know in advance that you will need to automate
lot of RESTfull deployments in an standard way.
- You want to bet for a technology that is well known
by software industry, so you can hire people that already
are familiar with it.
- Distributed systems are complex are is better to skip,
but you can not avoid them. Then k8s is the most familiar
approach.
- Your company has different isolated teams with random knowledge
and skills, from finances, to science, to marketing, .....
They all want a common base and future-resistent virtual
computing infrastructure.
They all want to share computer resources, and balance
computation when needed, as well as reuse DevOps knowledge
for databases, web servers, networking, ...
- Some k8s operator automate all the life-cycle of software
management for a given database, CMS, ... You want to profit
from such existing operator and the knowledge and experience
of the operator developers.
External Links
- About Desired State management @[https://www.youtube.com/watch?v=PH-2FfFD2PU]
- Online training:
@[https://www.katacoda.com/courses/kubernetes/playground]
@[https://labs.play-with-k8s.com/]
- @[https://kubernetes.io/]
- @[https://kubectl.docs.kubernetes.io/] ← Book
- @[https://kubernetes.io/docs/reference/] (Generated from source)
- @[https://kubernetes.io/docs/reference/glossary/]
- @[https://kubernetes.io/docs/reference/tools/]
- Most voted questions k8s on serverfault:
@[https://serverfault.com/questions/tagged/kubernetes?sort=votes]
- Most voted questions k8s on stackoverflow:
@[https://stackoverflow.com/questions/tagged/kubernetes?sort=votes]
- Real app Examples:
@[https://github.com/kubernetes/examples]
ºk8s Issues:º
- @[https://kubernetes.io/docs/reference/issues-security/issues/]
- @[https://kubernetes.io/docs/reference/issues-security/security/]
ºk8s Enhancements/Backlogº
- @[https://github.com/kubernetes/enhancements]
ºSIG-Appsº
- @[https://github.com/kubernetes/community/blob/master/sig-apps/README.md]
Special Interest Group for deploying and operating apps in Kubernetes.
- They meet each week to demo and discuss tools and projects.
""" Covers deploying and operating applications in Kubernetes. We focus
on the developer and devops experience of running applications in
Kubernetes. We discuss how to define and run apps in Kubernetes, demo
relevant tools and projects, and discuss areas of friction that can lead
to suggesting improvements or feature requests """
- Yaml Tips:
@[https://www.redhat.com/sysadmin/yaml-tips]
Summary
ref:@[https://kubernetes.io/docs/concepts/overview/components/]
@[https://kubernetes.io/docs/reference/#config-reference]
- k8s orchestates pools of CPUs, storage and networks
Kubernetes Cluster Components:
│ºetcdº: │ ºMasterº:
│- High availability key/value│ (Or cluster of masters)
│ddbb used to save the cluster│ - manages the cluster.
│metadata, service registrat. │ - keeps tracks of:
│and discovery │ - desired state.
└─────────────────────────────┘ ─ application scalation
- rolling updates
- Raft consensus used in
multimaster mode
(requires 1,3,5,... masters
@[https://raft.github.io/])
└────────┬─────────────────┘
│
┌───────────────┬────────────┬────────┴────────────────┬──────────┬─────────────────────────────┐
ºkube─apiserver:º │ºkube─controller─managerº │ºkube-schedulerº ºcloud─controller─managerº
─ (REST) Wrapper around│ ─Oº:EMBEDS K8S CORE CONTROL LOOPS!!!º│ ─ Assigns workloads to nodes ─ k8s 1.6Rºalpha featureº
k8s objects │ ─ºhandles a number of controllers:º │ ─ tracks host-resource ussage ─ runs controllers
- Listen for management│ ─ regulating the state of cluster │ ─ tracks total resources interacting with underlying
tools (kubectl, │ ─ perform routine tasks │ available per server and cloud provider
dashboard,...) │ ─ ensures number of replicas for │ resources allocated to ─ affected controllers
│ a service, │ existing workloads assigned ─ Node Controller
│ ─ ... │ to each server ─ Route Controller
│ │ ─ Manages availability, ─ Service Controller
│ │ performance and capacity ─ Volume ControllerCluster
│ │
ºfederation-apiserveRº ºfederation-controller-managerº
- API server for federated - embeds the core control loops shipped
clusters with Kubernetes federation.
│ºnodeº (Pool):
│- (VM) computer serving as
│ a worker machine where pods
│ are scheduled for execution. takes a set of PodSpecs and ensures
│- Contains k8s services: thate described containers are
│ - Kubelet agent ←------ running and healthy.
│ - Container runtime Iface
│ (Docker or rkt)
│- Get External IPs of all nodes:
│$º$ kubectl get nodes -o jsonpath=\ º
│$º '{.items[*].status.addresses[?(@.type=="ExternalIP")].address}'º
│
└────────┬─────────────────┘
│
┌────────┴──────────────────┬─────────────────────────────┐
ºkubelet agentº ºContainer Runtimeº ºkube-proxyº
- Gº"SDN enabler"º
- simple or round-robin TCP/UDP
stream forwarding across set
of back-ends.
- Listen for new Service/Service.EndPoint
events from ºMasterº to update
Service Virtual IPs to
real EndPoint IPs with the help of
proxy-ports, iptables,....
K8S
CLUSTER 1 ←···→ N Namespace 1 ←─┬─→ N Deployment 1 ←·····┐
─────── ========= ==========
─ etcd "Virtual │ "Application" ·
─ Master cluster" │ ·
─ Node Pool Can be assigned │ · Load balancing ·
max Memory/CPU │ with(out) session ·
Quotas │ affinity ·
│ · Persistence storage ·
│ · Scheduing of pods into ·
│ node pool ·
│ · Rolling deployments ·
│ · Ex: ·
│ 1 DDBB pod ·
│ + 2 Front End pods ·
│ + 1 ESB Pod ·
│ ·
└── N "Other" Controllers ·
(ReplicaSet, ...) · ·
·
┌································································┘
│
└→ N Pods 1 ←················→ 1+ Container
==== =========
· Minimum Unit of Exec/Sched. · executable ("Docker") image
· At creation time resource .ºLifecycle Hooksº
limits for CPU/memory are - PostStart Hook
defined (min and max-opt) - PreStop Hook
· Ussually Pod 1←→1 Container
(rarely 2 when such type Container struct {
containers are closely Name string
related to make Pod work) Image string
·RºPods are ephemeral.º Command []string
Deploy Controller+ Args []string
Pods is high available. WorkingDir string
Ports []ContainerPort
· Define shared resources EnvFrom []EnvFromSource
for its container/s: Env []EnvVar
o Volumes: (Shared Storage) Resources ResourceRequirements
o Networking: (Sharedºuniqueº VolumeMounts []VolumeMount
ºcluster IPº) VolumeDevices []VolumeDevice
o Metadata: (Container image LivenessProbe *Probe ←······· kubelet exec the probes
version, ports to use,...) ReadinessProbe *Probe - ExecAction
Lifecycle *Lifecycle - TCPSocketAction
· Represent a "logical host" TerminationMessagePath string - HTTPGetAction
TerminationMessagePolicy TerminationMessagePolicy
· A Pod is tied to the Node where ImagePullPolicy PullPolicy
it has been scheduler to run, SecurityContext *SecurityContext
and remains there until Stdin bool
termination. }
· In case of failure, controllers
can be scheduled to run on other · containers in a Pod are
available Nodes in the cluster. co-located and co-scheduled
on the same cluster node.
· Controllers takes charge of
scalling /"healing" pods.
REF: @[https://kubernetes.io/docs/Architecture/architecture_map.html?query=service+registration+discovery]
┌───────────┐
─ Describes the resources available ─┐ │Node Status│
─ CPU │ ├───────────┤
─ memory │ │Addresses ←─(HostName, ExternalIP, InternalIP)
─ max.number of pods supported └───────→Capacity │ ^^^^^^^^^^
┌───────────────────────────────────────────→Condition │ Typically visible
status ofºallºRunning nodes. │Info ← General info. like within cluster
Node Description └───────────┘ kernel/k8s/docker ver,
----------------- ------------------------------------- OS name,...
OutOfDisk True → insufficient free space on the
node for adding new pods
Ready True → node is healthy and ready to accept pods
if not True after ºpod─eviction─timeoutº
(5 minutes by default)
an argument is passed to the kube-controller-manager
and all Pods
MemoryPressure True → pressure exists on the node memory
PIDPressure True → pressure exists on the processes
DiskPressure True → pressure exists on the disk size
NetworkUnavailable True → node network not correctly configured
Network
REF:@[https://kubernetes.io/docs/concepts/architecture/cloud-controller/]
NETWORK SCHEMA WITHOUT CLOUD-CONTROLLER | NETWORK SCHEMAºWITHº CLOUD-CONTROLLER:
o Cloud connector _ _ | o Cloud connector _ _
___| | ___ _ _ __| | | ___| | ___ _ _ __| |
kube─controller─managero──→ / __| |/ _ \| | | |/ _` | | cloud-controller─managero──→ / __| |/ _ \| | | |/ _` |
^ | (__| | (_) | |_| | (_| | | ^ | (__| | (_) | |_| | (_| |
│ \___|_|\___/ \__,_|\__,_| | kube─controller─manager │ \___|_|\___/ \__,_|\__,_|
│ ^ | ^ │
│ │ | │ │
│ │ | └─────────────────┤
v │ node | │
etcd ←───→ kube─apiserver ←──┬─┐ ┌─o────────┐ | v node
│ └──→kubelet ←───┐ | etcd ←───────→ kube─apiserver ←──┬─┐ ┌──────────┐
^ │ │ │ │ | │ └─→kubelet ←───┐
│ │ │ │ │ | ^ │ │ │ │
v │ │ │ │ | │ │ │ │ │
kube─scheduler └────→kube─proxy│ │ | v │ │ │ │
└──────────┘ │ | kube─scheduler └───→kube─proxy│ │
│ | └──────────┘ │
│ ┌────────┐ │
└─→Image ←───────────────────────────────────────────────┘
│Registry│
└────────┘
Object Management
REF:
- @[https://kubernetes.io/docs/tasks/tools/install-kubectl/]
- @[https://kubernetes.io/docs/concepts/overview/object-management-kubectl/overview/]
- @[https://kubernetes.io/docs/reference/kubectl/cheatsheet/]
º┌──────────┐º ┌───────────┐
º│k8s object│º ┌─→│metadata │
º├──────────┤º │ ├───────────┤
º│kind │º one of the GºResource typesº│ │name │← maps to /api/v1/pods/name
*│metadata │←─────────────────────────────┘ │UID │← Distinguish between historical
º│spec │← desired stateº │namespace │ occurrences of similar object
º│state │← present stateº │labels │
º└──────────┘º │annotations│
└───────────┘
metadata is organized
around the concept of an
application. k8s does NOT
enforce a formal notion of
application. Apps are
described thorugh metadata
in a loose definition.
etcd kubernetes (/openapi/v2) cli management:
(serialized ←────── objects ←────────────────→ API ←───→ ºkubectlº'action' Gº'resource'º
API resource - Represent API resources Server *1 ^^^^^^
states) (persistent entities ºget º: list resources
in a cluster) ºdescribeº: show details (and events for pods)
- The can describe: ºlogs º: print container
- apps running on nodes logs
- resources available to ºexec º: exec command on
a given app container
- policies around apps ºapply º: creates and updates resources
(restart, upgrades, ...) ...
common kubectl flags:
--all-namespaces
-o wide
--include-uninitialized
--sort-by=.metadata.name
--sort-by='.status.containerStatuses[0].restartCount'
--selector=app=cassandra
*1: @[https://kubernetes.io/docs/conceptsverview/kubernetes-api/]
Resource Types
GºRESOURCE TYPES SUMMARYº
clusters │podtemplates │statefulsets
(cs)componentstatuses │(rs)replicasets │(pvc)persistentvolumeclaims
(cm)configmaps │(rc)replicationcontrollers │(pv) persistentvolumes
(ds)daemonsets │(quota)resourcequotas │(po) pods
(deploy)deployments │cronjob │(psp)podsecuritypolicies
(ep)endpoints │jobs │secrets
(ev)event │(limits)limitranges │(sa)serviceaccount
(hpa)horizon...oscalers │(ns)namespaces │(svc)services
(ing)ingresses │networkpolicies │storageclasses
│(no)nodes │thirdpartyresources
GºRESOURCE TYPES EXTENDEDº @[https://github.com/kubernetes/kubernetes/blob/master/pkg/apis/core/types.go]
001 Volume 021 FlexPersistentVolumeSource 041 DownwardAPIProjection
002 VolumeSource 022 FlexVolumeSource 042 AzureFileVolumeSource
003 PersistentVolumeSource 023 AWSElasticBlockStoreVolumeSource 043 AzureFilePersistentVolumeSource
004 PersistentVolumeClaimVolumeSource 024 GitRepoVolumeSource 044 VsphereVirtualDiskVolumeSource
005 PersistentVolume 025 SecretVolumeSource 045 PhotonPersistentDiskVolumeSource
006 PersistentVolumeSpec 026 SecretProjection 046 PortworxVolumeSource
007 VolumeNodeAffinity 027 NFSVolumeSource 047 AzureDiskVolumeSource
008 PersistentVolumeStatus 028 QuobyteVolumeSource 048 ScaleIOVolumeSource
009 PersistentVolumeList 029 GlusterfsVolumeSource 049 ScaleIOPersistentVolumeSource
010 PersistentVolumeClaim 030 GlusterfsPersistentVolumeSource 050 StorageOSVolumeSource
011 PersistentVolumeClaimList 031 RBDVolumeSource 051 StorageOSPersistentVolumeSource
012 PersistentVolumeClaimSpec 032 RBDPersistentVolumeSource 052 ConfigMapVolumeSource
013 PersistentVolumeClaimCondition 033 CinderVolumeSource 053 ConfigMapProjection
014 PersistentVolumeClaimStatus 034 CinderPersistentVolumeSource 054 ServiceAccountTokenProjection
015 HostPathVolumeSource 035 CephFSVolumeSource 055 ProjectedVolumeSource
016 EmptyDirVolumeSource 036 SecretReference 056 VolumeProjection
017 GCEPersistentDiskVolumeSource 037 CephFSPersistentVolumeSource 057 KeyToPath
018 ISCSIVolumeSource 038 FlockerVolumeSource 058 LocalVolumeSource
019 ISCSIPersistentVolumeSource 039 DownwardAPIVolumeSource 059 CSIPersistentVolumeSource
020 FCVolumeSource 040 DownwardAPIVolumeFile 060 CSIVolumeSource
061 ContainerPort 081 Handler 101 PreferredSchedulingTerm
062 VolumeMount 082 Lifecycle 102 Taint
063 VolumeDevice 083 ContainerStateWaiting 103 Toleration
064 EnvVar 084 ContainerStateRunning 104 PodReadinessGate
065 EnvVarSource 085 ContainerStateTerminated 105 PodSpec
066 ObjectFieldSelector 086 ContainerState 106 HostAlias
067 ResourceFieldSelector 087 ContainerStatus 107 Sysctl
068 ConfigMapKeySelector 088 PodCondition 108 PodSecurityContext
069 SecretKeySelector 089 PodList 109 PodDNSConfig
070 EnvFromSource 090 NodeSelector 110 PodDNSConfigOption
071 ConfigMapEnvSource 091 NodeSelectorTerm 111 PodStatus
072 SecretEnvSource 092 NodeSelectorRequirement 112 PodStatusResult
073 HTTPHeader 093 TopologySelectorTerm 113 Pod
074 HTTPGetAction 094 TopologySelectorLabelRequirement 114 PodTemplateSpec
075 TCPSocketAction 095 Affinity 115 PodTemplate
076 ExecAction 096 PodAffinity 116 PodTemplateList
077 Probe 097 PodAntiAffinity 117 ReplicationControllerSpec
078 Capabilities 098 WeightedPodAffinityTerm 118 ReplicationControllerStatus
079 ResourceRequirements 099 PodAffinityTerm 119 ReplicationControllerCondition
080 Container 100 NodeAffinity 120 ReplicationController
121 ReplicationControllerList 141 DaemonEndpoint 161 Preconditions 181 ResourceQuotaSpec
122 ServiceList 142 NodeDaemonEndpoints 162 PodLogOptions 182 ScopeSelector
123 SessionAffinityConfig 143 NodeSystemInfo 163 PodAttachOptions 183 ScopedResourceSelerRequirement
124 ClientIPConfig 144 NodeConfigStatus 164 PodExecOptions 184 ResourceQuotaStatus
125 ServiceStatus 145 NodeStatus 165 PodPortForwardOptions 185 ResourceQuota
126 LoadBalancerStatus 146 AttachedVolume 166 PodProxyOptions 186 ResourceQuotaList
127 LoadBalancerIngress 147 AvoidPods 167 NodeProxyOptions 187 Secret
128 ServiceSpec 148 PreferAvoidPodsEntry 168 ServiceProxyOptions 188 SecretList
129 ServicePort 149 PodSignature 169 ObjectReference 189 ConfigMap
130 Service 150 ContainerImage 170 LocalObjectReference 190 ConfigMapList
131 ServiceAccount 151 NodeCondition 171 TypedLocalObjectReference 191 ComponentCondition
132 ServiceAccountList 152 NodeAddress 172 SerializedReference 192 ComponentStatus
133 Endpoints 153 NodeResources 173 EventSource 193 ComponentStatusList
134 EndpointSubset 154 Node 174 Event 194 SecurityContext
135 EndpointAddress 155 NodeList 175 EventSeries 195 SELinuxOptions
136 EndpointPort 156 NamespaceSpec 176 EventList 196 WindowsSecurityContextOptions
137 EndpointsList 157 NamespaceStatus 177 LimitRangeItem 197 RangeAllocation
138 NodeSpec 158 Namespace 178 LimitRangeSpec
139 NodeConfigSource 159 NamespaceList 179 LimitRange
140 ConfigMapNodeConfigSource 160 Binding 180 LimitRangeList
k8s (def.)TCP Ports
REF @[https://kubernetes.io/docs/tasks/tools/install-kubeadm/]
┌───────┬───────┐
│Master │Worker │
│node(s)│node(s)│
┌──────────────────────────────────────┼───────┼───────┤
│Port Range │ Purpose │ X │ │
├──────────────────────────────────────┼───────┼───────┤
│6443* │ Kubernetes API server │ X │ │
├──────────────────────────────────────┼───────┼───────┤
│2379─2380 │ etcd server client API │ X │ │
├──────────────────────────────────────┼───────┼───────┤
│10250 │ Kubelet API │ X │ X │
├──────────────────────────────────────┼───────┼───────┤
│10251 │ kube─scheduler │ X │ │
├──────────────────────────────────────┼───────┼───────┤
│10252 │ kube─controller─manager│ X │ │
├──────────────────────────────────────┼───────┼───────┤
│10255 │ Read─only Kubelet API │ X │ X │
├──────────────────────────────────────┼───────┼───────┤
│ 30000─32767 │ NodePort Services │ │ X │
└──────────────────────────────────────┴───────┴───────┘
Application Troubleshooting
REF: @[https://learnk8s.io/]
- @[https://kubernetes.io/docs/tasks/debug-application-cluster/debug-application/]
$º$ kubectl get pods º
└─────┬────────┘
QºQ: Is there any pending Pod?º
└─────────┬────────────────┘
┌─────────┘
├ YES → $º$ kubectl describe º←QºQ: Is the cluster full?º
│ $º pod $pod_name º └────────┬────────────┘
│ ┌─────────────────────────────┘
│ ├ NO →QºQ: Are you hitting ResourceQuotaLimits?º
│ │ └────────────────────┬────────────────┘
│ │ ┌─────────────────────┘
│ │ ├ NO →QºQ: Are you mounting a PENDINGº
│ │ │ Qº PersistentVolumeClaim?º
│ │ │ └───────────┬────────────────┘
│ │ │ ┌────────────┘
│ │ │ ├ NO →$º$ kubectl get pods º←QºQ: Is the Pod assignedº
│ │ │ │ $º -o wide º Qº to the Node?º
│ │ │ │ └────┬───────────────┘
│ │ │ │ ┌─────────────────────────────┘
│ │ │ │ ├ YES →RºThere is an issue with the Kubeletº
│ │ │ │ │
│ │ │ │ └ NO →RºThere is an issue with the Schedulerº
│ │ │ │
│ │ │ └ YES →BºFix the PersistentVolumeClaimº
│ │ │
│ │ └ YES →BºRelax Quota Limitsº
│ │
│ └ YES →BºProvision a bigger clusterº
│
└ NO →QºQ: Are the Pods Running?º
└──────┬───────────────┘
┌───────────────┘
├ NO → $º$ kubectl logs $pod_name º←QºQ: Can you see the logsº
│ ↑ Qº for the App?º
│ │ └─────────┬───────────┘
│ │ ┌────────────────────────────────────┘
│ │ ├ Yes →BºFix the issue in the Appº
│ │ └ NO →QºQ: Did the container died too Quickly?º
│ │ └──────┬─────────────────────────────┘
│ │ ┌───────────┘
│ │ ├ NO → $º$ kubectl describe º←QºQ: Is the Pod in statusº
│ │ │ $º pod $pod_name º Qº ImagePullBackOff? º
│ │ │ └──┬──────────────────┘
│ │ │ ┌────────────────────────────────┘
│ │ │ ├ NO →QºQ: Is the Pode Status CrashLoopBackOff?º
│ │ │ │ └─────────┬─────────────────────────┘
│ │ │ │ ┌────────────────┘
│ │ │ │ ├ NO →QºQ: Is the Pod status RunContainerError?º
│ │ │ │ │ └─────────┬───────────────────────────┘
│ │ │ │ │ ┌───────────────┘
│ │ │ │ │ ├ NO →BºConsult StackOverflowº
│ │ │ │ │ │
│ │ │ │ │ └ YES →BºThe issue is likely to be withº
│ │ │ │ │ Bºmounting volumesº
│ │ │ │ │
│ │ │ │ └ YES →QºQ: Did you inspect the logs and fixedº
│ │ │ │ Qº the crashes?º
│ │ │ │ └─────────┬─────────────────────────┘
│ │ │ │ ┌───────────────┘
│ │ │ │ ├ NO →BºFix the app crahsesº
│ │ │ │ │
│ │ │ │ └ YES →QºQ: Did you forget the 'CMD' instructionº
│ │ │ │ Qº in the Dockerfile? º
│ │ │ │ └──────────┬──────────────────────────┘
│ │ │ │ ┌────────────────┘
│ │ │ │ ├ YES →BºFix the Dockerfileº
│ │ │ │ │
│ │ │ │ └ NO →QºQ: Is the Pod restarting frequently?º
│ │ │ │ Qº Cycling between Running and º
│ │ │ │ Qº CrashLoopBackoff?º
│ │ │ │ └──────────┬────────────────────────┘
│ │ │ │ ┌──────────────────┘
│ │ │ │ ├ YES →BºFix the liveness probeº
│ │ │ │ │
│ │ │ │ └ NO →RºUnknown Stateº
│ │ │ │
│ │ │ └ YES →QºQ: Is the name of the image correct?º
│ │ │ └────┬─────────────────────────────┘
│ │ │ ┌────────────┘
│ │ │ ├ NO →BºFix the image nameº
│ │ │ │
│ │ │ └ YES →QºQ: Is the image tag valid?º
│ │ │ Qº Does it exists?º
│ │ │ └──┬──────────────────────┘
│ │ │ ┌───────────┘
│ │ │ ├ NO →BºFix the tagº
│ │ │ │
│ │ │ └ YES →QºQ: Are you pulling images from aº
│ │ │ Qº private registry?º
│ │ │ └─────────┬────────────────────┘
│ │ │ ┌─────────────────┘
│ │ │ ├ NO →BºThe Issue could be with CRI|Kubeletº
│ │ │ │
│ │ │ └ YES →BºConfigure pulling images from aº
│ │ │ Bºprivate registryº
│ │ │
│ │ └ YES → $º$ kubectl logs $pod_name --previous º
│ │ └────────────┬──────────────────┘
│ └───────────────────────────────┘
└ YES →QºQ: Are ther Pods READY?º
└──┬──────────────────┘
┌───────────┘
├ NO → $º$ kubectl describe º←QºQ: Is the Readiness probeº
│ $º pod $pod_name º Qº failing?º
│ └──┬────────────────────┘
│ ┌───────────────────────┘
│ ├ YES →BºFix the Readiness Probeº
│ │
│ └ NO →RºUnknown Stateº
│
└ YES → $º$ kubectl port-forward \ º ←QºQ: Can you access the app?º
$º $pod_name 8080:$pod_port º └──────────┬─────────────┘
┌───────────────────────────────────────────────────────┘
├ NO →QºQ: Is the port exposed by container correctº
│ Qº and listening on 0.0.0.0?º
│ └────────────────────┬────────────────────┘
│ ┌─────────────────────┘
│ ├ NO →BºFix the app. It should listen onº
│ │ Bº0.0.0.0.º
│ │ BºUpdate the containerPortº
│ │
│ └ YES →RºUnknown Stateº
↓
YES
↓
Bº***************************º
Bº*POD ARE RUNNING CORRECTLY*º
Bº***************************º
└──────────┬────────────┘
↓
$º$ kubectl describe \ º ←QºQ: Can you see a listº
$º service $SERVICE_NAME º Qº of endpoints?º
└──────┬────────────┘
┌───────────────────────────────────┘
├ NO →QºQ: Is the Selector matching the rightº
│ Qº Pod label?º
│ └──┬────────────────────────────────┘
│ ┌─────┘
│ ├ NO → Fix the Service selector. It has
│ │ to match the Pod labels
│ │
│ └ YES →QºQ: Does the Pod have an IP address assigned?º
│ └──┬───────────────────────────────────────┘
│ ┌─────┘
│ ├ NO → There is an issue with
│ │ the Controller Manager
│ │
│ └ YES → There is an issue with the Kubelet
│
└ YES →$º$ kubectl port-forward \ º←QºQ: Can you visit the app?º
$º service/$SERVICE_NAME \ º └──┬────────────────────┘
$º 8080:$SERVICE_PORT º │
│
┌───────────────────────────────────────────┘
├ NO →QºQ: Is the targetPort on the Service º
│ Qº matching the containerPort in theº
│ Qº Pod?º
│ └──┬────────────────────────────────┘
│ ┌─────┘
│ ├ NO →BºFix the Service targetPort andº
│ │ Bºthe containerPodº
│ │
│ └ YES →BºThe issue could be with Kube Proxyº
↓
YES
↓
Bº******************************º
Bº*SERVICE IS RUNNING CORRECTLY*º
Bº******************************º
↓
$º$ kubectl describe \ º←QºQ: Can you see a list of Backends?º
$º ingress $INGRESS_NAME º └──┬─────────────────────────────┘
┌──────────────────────────────┘
├ NO →QºQ: Are the serviceName and servicePortº
│ Qº mathcing the service?º
│ └──┬─────────────────────────────────┘
│ ┌───┘
│ ├ NO →BºFix the ingress serviceName and servicePortº
│ │
│ └ YES →BºThe issue is specific to the Ingress Controllerº
│ BºConsult the docs for your Ingressº
↓
YES
↓
Bº**********************************º
Bº*THE INGRESS IS RUNNING CORRECTLY*º
Bº**********************************º
Bº(The app should be working now!!!)º
│
↓ ┌ NO →RºThe issue is likely to be with the º
│ RºInfrastructure and how the cluster isº
QºQ: Can you visit app º→ ─┤ Rºexposedº
Qº from the Internet?º │
│ Bº*****º
└ YES → Bº*END*º
Bº*****º
Kubectl
- @[https://kubernetes.io/docs/reference/kubectl/jsonpath/]
@[https://kubernetes.io/docs/reference/kubectl/overview/]
@[https://kubernetes.io/docs/reference/kubectl/kubectl/]
@[https://kubernetes.io/docs/reference/kubectl/kubectl-cmds/]
@[https://kubernetes.io/docs/reference/kubectl/conventions/]
@[https://kubernetes.io/docs/reference/kubectl/docker-cli-to-kubectl/]
$ KUBE_EDITOR="vim"
$ source ˂(kubectl completion bash) # kubectl autocomple
$ source ˂(kubectl completion zsh)
# use multiple kubeconfig files
$ KUBECONFIG=~/.kube/config:~/.kube/kubconfig2 \
kubectl config view # Show Merged kubeconfig settings.
$ kubectl config current-context
$ kubectl config use-context my-cluster-name
$ kubectl run nginx --image=nginx
$ kubectl explain pods,svc
$ kubectl edit svc/my-service-1 # ← Edit resource
ºkubectl Field Selectorsº
- filter k8s objects based on the value of one or more object fields.
- the possible field depends on each object type/kind but
all resource-types support 'metadata.name' and 'metadata.namespace'
Ex:
$ kubectl get pods --field-selector metadata.name==my-service,spec.restartPolicy=Always
$ kubectl get pods --field-selector metadata.namespace!=default,status.phase==Running
Building Blocks
NameSpaces
Namespace
@[https://kubernetes.io/docs/concepts/overview/working-with-objects/namespaces/]
@[https://kubernetes.io/docs/tasks/administer-cluster/namespaces/]
- Provide a scope for names
- Namespaces are intended for use in environments with many users spread
across multiple teams, or projects. For clusters with a few to tens of users,
you should not need to create or think about namespaces at all. Start
using namespaces when you need the features they provide.
- use labels, not namespaces, to distinguish resources within the same namespace
- Services are created with DNS entry
"service-name"."namespace".svc.cluster.local
$ kubectlºget namespacesº
NAME STATUS AGE
default Active 1d
kube-system Active 1d
$ kubectlºcreate namespaceºmy-namespace // ← create namespace:
$ kubectlº--namespace=my-namespaceº \ ← set NS for request
run nginx --image=nginx
$ kubectlºconfig set-contextº\ ← permanently save the NS
$(kubectl ºconfig current-contextº) \ ← shell syntax sugar: Exec command and take output
--namespace=MyFavouriteNS as input for next command
$ kubectl config view | \
grep namespace: # Validate
Secrets
@[https://kubernetes.io/docs/concepts/configuration/secret/]
@[https://github.com/kubernetes/community/blob/master/contributors/design-proposals/auth/secrets.md]
See also:
@[https://kubernetes.io/docs/tasks/inject-data-application/distribute-credentials-secure/]
- (Handling passwords, OAuth tokens, ssh keys, ... in k8s)
- Users (and system) create secrets
- At runtime Pods reference the secrets in three ways:
- as files in a mounted volume
- as ENVIRONMENT variables.
- Also used by kubelet when pulling images for the pod
ºBuilt-in Secretsº
- Service Accounts automatically create and attach secrets with API Credentials
- Kubernetes automatically creates secrets which contain credentials for
accessing the API and it automatically modifies your pods to use this type
of secret.
ºCreating User Secretsº
│ ALTERNATIVE 1 │ ALTERNATIVE 2 │ALTERNATIVE 3
│ ºSTEP 1º: Create un-encrypted secrets locally:│ºSTEP 1:ºCreate Secret with data │ºSTEP 1º: Create Secret with stringData
│ $ echo -n 'admin' ˃ ./username.txt │ cat ˂˂ EOF ˃ secret.yaml │apiVersion: v1
│ $ echo -n '1f2d1e2e67df' ˃ ./password.txt │ apiVersion: v1 │kind:ºSecretº
│ ^^^^^^^^^^^^ │ kind:ºSecretº │metadata:
│ special chars. must be │ metadata: │ name: mysecret
│ '\' escaped. │ name: mysecret │type: Opaque
│ │ type: Opaque │stringData:
│ │ data: │ config.yaml: |-
│ │ username: $(echo -n 'admin' | base64) │ apiUrl: "https://my.api.com/api/v1"
│ │ password: $(echo -n '1f2d1e2e67df' | base64) │ username: admin
│ │ EOF │ password: 1f2d1e2e67df
│ │ │
│ ºSTEP 2:º Package into a Secret k8s object │ºSTEP 2:º Apply │ºSTEP 2:ºApply
│ $ kubectlºcreate secretº\ │ $ kubectl apply -f ./secret.yaml │ $ kubectl apply -f ./secret.yaml
│ genericGºdb-user-passº \ │ │
│ --from-file=./username.txt \ │ │
│ --from-file=./password.txt │ │
│ºSTEP 3:º Check that Secret object has been ┌────────────────────────────────────────────
│ properly created. │ ºUsing the Secrets in a Podº
│ $ kubectl get secrets │ Control secret path with items: Consume as ENV.VARS
│ → NAME TYPE DATA AGE │ apiVersion: v1 │ apiVersion: v1 | apiVersion: v1
│ → db-user-pass Opaque 2 51s │ kind:ºPodº │ kind:ºPodº | kind: Pod
│ │ metadata: │ metadata: | metadata:
│ $ kubectl describe secrets/db-user-pass │ name: mypod │ name: mypod | name: secret-env-pod
│ → Name: Gºdb-user-passº │ spec: │ spec: | spec:
│ → Namespace: default │ containers: │ containers: | containers:
│ → Labels: ˂none˃ │ ─ name: mypod │ - name: mypod | - name: mycontainer
│ → Annotations: ˂none˃ │ image: redis │ image: redis | image: redis
│ → │ volumeMounts: │ volumeMounts: | env:
│ → Type: Opaque │ ─ name:ºfoOº │ - name:ºfooº | - name: SECRET_USERNAME
│ → │ mountPath: "/etc/foo" │ mountPath: "/etc/foo" | valueFrom:
│ → Data │ readOnly:ºtrueº │ readOnly: true | secretKeyRef:
│ → ==== │ volumes: │ volumes: | name: Gºdb─user─passº
│ →ºpassword.txt:º 12 bytes │ ─ name:ºfoOº │ - name:ºfooº | key: username
│ →ºusername.txt:º 5 bytes │ secret: │ secret: | - name: SECRET_PASSWORD
│ secretName:Gºdb─user─passº│ secretName:Gºdb─user─passº | valueFrom:
│ defaultMode: 256 │ items: | secretKeyRef:
│ ^ │ - key: username | name: Gºdb─user─passº
│ | │ path: my-group/my-username | key: password
| ^^^^^^^^^^^^^^^^^^^^
JSON does NOT support · username will be seen in container as:
octal notation. /etc/foo/my-group/my-username
256 = 0400 · password secret is not projected
Ex 2:
SECRET CREATION | SECRET USSAGE
$ kubectl create secret generic \ | apiVersion: v1
Gºssh-key-secretº \ | kind: Pod
--from-file=ssh-privatekey=/path/to/.ssh/id_rsa \ | metadata:
--from-file=ssh-publickey=/path/to/.ssh/id_rsa.pub | name: secret-test-pod
| labels:
| name: secret-test
| spec:
| volumes:
| - name: secret-volume
| secret:
| secretName:Gºssh-key-secretº
| containers:
| - name: ssh-test-container
| image: mySshImage
| volumeMounts:
| - name: secret-volume
| readOnly: true
| mountPath: "/etc/secret-volume"
| ^^^^^^^^^^^^^^^^^^^^
| secret files available visible like:
| /etc/secret-volume/ssh-publickey
| /etc/secret-volume/ssh-privatekey
FEATURE STATE: Kubernetes v1.13 beta
You can enable encryption at rest for secret data, so that the secrets are not stored in the clear into etcd .
(Pod)ConfigMap
@[https://kubernetes.io/docs/tasks/configure-pod-container/configure-pod-configmap/]
ConfigMaps: decouple configuration artifacts from image content
ºCreate a ConfigMapº
$ kubectl create configmap \
'map-name' \
'data-source' ← data-source: directories, files or literal values
translates to the key-value pair in the ConfigMap where
key = file_name or key provided on the cli
value = file_contents or literal_value provided on the cli
You can use kubectl describe or kubectl get to retrieve information about a ConfigMap.
ConfigMapºfrom config-fileº:
┌───────────────────────────── ┌──────────────────────────────────────
│ºSTEP 0:º Input to config map │ºSTEP 1:º create ConfigMap object
│ ...configure─pod─container/ │ $ kubectl create configmap \
│ └─ configmap/ │ game-config
│ ├─ºgame.propertiesº │ --from-file=configmap/
│ └─º ui.propertiesº │ ^^^^^^^^^^
│ │ combines the contents of
│ │ all files in the directory
┌───────────────────────────────────────────────────────────────────────────────────────────────────────────
│ºSTEP 2:º check STEP 1
│$ kubectlºget configmapsºgame-config -o yaml │$ kubectlºdescribe configmapsºgame-config
│→ apiVersion: v1 │→
│→ kind: ConfigMap │→ Name: game-config
│→ metadata: │→ Namespace: default
│→ creationTimestamp: 2016-02-18T18:52:05Z │→ Labels:
│→ name: game-config │→ Annotations:
│→ namespace: default │→
│→ resourceVersion: "516" │→ Data
│→ selfLink: /api/v1/namespaces/default/configmaps/game-config │→ ====
│→ uid: b4952dc3-d670-11e5-8cd0-68f728db1985 │→ºgame.properties:º 158 bytes
│→ data: │→ºui.properties: º 83 bytes
│→ ºgame.properties:º|
│→ enemies=aliens
│→ lives=3 ┌───────────────────────────────────
│→ enemies.cheat=true │ºSTEP 3:ºUse in Pod container*
│→ enemies.cheat.level=noGoodRotten │ apiVersion: v1
│→ secret.code.passphrase=UUDDLRLRBABAS │ kind: Pod
│→ secret.code.allowed=true │ metadata:
│→ secret.code.lives=30 │ name: dapi-test-pod
│→ ºui.properties:º| │ spec:
│→ color.good=purple │ containers:
│→ color.bad=yellow │ - name: test-container
│→ allow.textmode=true │ ...
│→ how.nice.to.look=fairlyNice │ env:
│ º- name: ENEMIES_CHEAT º
│ º valueFrom: º
│ º configMapKeyRef: º
│ º name: game-config º
│ º key: enemies.cheat º
ConfigMapºfrom env-fileº:
│ºSTEP 0:º Input to config map │ºSTEP 1:º create ConfigMap object
│ ...configure─pod─container/ │ $ kubectl create configmap \
│ └─ configmap/ │ game-config-env-file \
│ └─ºgame-env-file.propertiesº │ --from-env-file=game-env-file.properties
│ ^^^^^^^^^^^^^^^^^^^^^^^^
│ game-env-file.properties
│ enemies=aliens │ ºSTEP 2:º Check STEP 1
│ lives=3 │ $ kubectl get configmap game-config-env-file -o yaml
│ allowed="true" │ → apiVersion: v1
│ → kind: ConfigMap
│ → metadata:
│ → creationTimestamp: 2017-12-27T18:36:28Z
│ → name: game-config-env-file
│ → namespace: default
│ → resourceVersion: "809965"
│ → selfLink: /api/v1/namespaces/default/configmaps/game-config-env-file
│ → uid: d9d1ca5b-eb34-11e7-887b-42010a8002b8
│ → data:
│ → allowed: '"true"'
│ → enemies: aliens
│ → lives: "3"
NOTE: kubectl create configmap ... --from-file=Bº'my-key-name'º='path-to-file'
will create the data under:
→ ....
→ data:
→ Bºmy-key-name:º
→ key1: value1
→ ...
ºConfigMaps from literal valuesº
º STEP 1º | º STEP 2º
$ kubectl create configmap special-config \ | → ...
--from-literal=special.how=very \ | → data:
--from-literal=special.type=charm | → special.how: very
| → special.type: charm
restartPolicy: Never
Containers
Init Containers
@[https://kubernetes.io/docs/concepts/workloads/pods/init-containers/]
- one or more specialized Containers that run before app Containers
-ºcan contain utilities or setup scripts not present in an app imageº
- exactly equal to regular Containers, except:
- They always run to completion.
k8s will restart it repeatedly until succeeds
(unless restartPolicy == Never)
- Each one must complete successfully before the next one is started.
- status is returned in .status.initContainerStatuses
^^^^^^^^^^^^^^^^^^^^^
vs .status.containerStatuses
- readiness probes do not apply
Note: Init Containers can also be given access to Secrets that app Containers
are not able to access.
- Ex: Wait for 'myService' and then for 'mydb'.
│ apiVersion: v1 │apiVersion: v1 │apiVersion: v1
│ kind:ºPodº │kind:ºServiceº │kind:ºServiceº
│ metadata: │metadata: │metadata:
│ name: myapp─pod │ name:Gºmyserviceº │ name:Bºmydbº
│ labels: │spec: │spec:
│ app: myapp │ ports: │ ports:
│ spec: │ ─ protocol: TCP │ ─ protocol: TCP
│ containers: │ port: 80 │ port: 80
│ ─ name: myapp─container │ targetPort: 9376 │ targetPort: 9377
│ image: busybox:1.28 └──────────────────── └─────────────────────
│ command: ['sh', '─c', 'echo The app is running! ⅋⅋ sleep 3600']
│ ºinitContainers:º
│ ─ name:Oºinit─myserviceº
│ image: busybox:1.28
│ command: ['sh', '─c', 'until nslookup Gºmyserviceº; sleep 2; done;']
│ ─ name:Oºinit─mydbº
│ image: busybox:1.28
│ command: ['sh', '─c', 'until nslookup Bºmydbº ; sleep 2; done;']
Inspect init Containers like:
$ kubectl logs myapp-pod -c Oºinit-myserviceº
$ kubectl logs myapp-pod -c Oºinit-mydb º
RºWARN:º
- Use activeDeadlineSeconds on the Pod and livenessProbe on the Container
to prevent Init Containers from failing forever.
- Debugging Init Containers:
@[https://kubernetes.io/docs/tasks/debug-application-cluster/debug-init-containers/]
Containers Limits
@[https://kubernetes.io/docs/concepts/configuration/manage-compute-resources-container/]
- See also design proposal:
@[https://github.com/kubernetes/community/blob/master/contributors/design-proposals/node/resource-qos.md]
A Container:
- ºcan exceedº its resourceºrequestºif the Node has memory available,
- isºnot allowedºto use more than its resource ºlimitº.
- spec.containers[].resources.limits.cpu ← can NOT exceed resource limit → it will be scheduled
- spec.containers[].resources.limits.memory for termination
- spec.containers[].resources.requests.cpu ← can exceed resource request
- spec.containers[].resources.requests.memory
- spec.containers[].resources.limits.ephemeral-storage k8s 1.8+
- spec.containers[].resources.requests.ephemeral-storage k8s 1.8+
Note: resource quota feature can be configured to limit the total amount of resources
that can be consumed. In conjunction namespaces, it can prevent one team from
hogging all the resources.
ContaineRºComputeºResource types:
- CPU : units of cores
- total amount of CPU time that a container can use every 100ms.
(minimum resolution can be setup to 1ms)
- 1 cpu is equivalent to: 1 (AWS vCPU|GCP Core|Azure vCore,IBM vCPU,Hyperthread bare-metal )
- 0.1 cpu == 100m
- memory : units of bytes:
- Ex: 128974848, 129e6, 129M, 123Mi
- Local ephemeral storage(k8s 1.8+):
- No QoS can be applied.
- Ex: 128974848, 129e6, 129M, 123Mi
When using Docker:
greater of this number or 2 is used as the value of the --cpu-shares flag in the docker run command.
ºExtended resourcesº
- fully-qualified resource names outside the kubernetes.io domain.
Ussage:
- STEP 1: cluster operator must advertise an Extended Resource.
- STEP 2: users must request the Extended Resource in Pods.
(See official doc for more info)
It is planned to add new resource types, including a node disk space resource, and a framework for adding custom resource types.
Pods
Life Cicle
DEFINITION END
defined → scheduled in node → run → alt1: run until their container(s) exit
alt2: pod removed (for any other reason)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
depending on policy and exit code,
may be removed after exiting, or
may be retained in order to enable
access containers logs.
@[https://kubernetes.io/docs/concepts/workloads/pods/pod-lifecycle/]
Pod.PodStatus.phase := Pending|Running|Succeeded|Failed
Pod Preset
@[https://kubernetes.io/docs/concepts/workloads/pods/podpreset/]
- objects used to inject information into pods at creation time
like secrets, volumes, volume mounts, and environment variables.
- label selectors are used to specify Pods affected.
ºDisable Pod Preset for a Specific Podº
Add next annotation to the Pod Spec:
- podpreset.admission.kubernetes.io/exclude: "true".
Pod QoS
@[https://kubernetes.io/docs/tasks/configure-pod-container/quality-service-pod/]
Pod.spec.qosClass :=
Guaranteed Assigned to pod when:
- All Pod-Containers have a defined cpu/mem. limit.
- All Pod-Containers have same cpu/mem.limit and same cpu/mem.request
Burstable Assigned to pod when:
- criteria for QoS class Guaranteed isºNOTºmet.
- 1+ Pod-Containers have memory or CPU request.
BestEffort Assigned to pod when:
- Pod-Containers do NOT have any memory/CPU request/limit
Ex. Pod YAML Definition:
apiVersion: v1
kind: Pod
metadata:
name: test-pd
spec:
containers:
- image: gcr.io/google_containers/test-webserver
name: test-container
volumeMounts:
- mountPath: /cache ← where to mount
name:Bºcache-volumeº ← volume (name) to mount
- mountPath: /test-pd
name:Gºtest-volumeº
volumes:
- name:Bºcache-volumeº
emptyDir: {}
- name:Gºtest-volumeº
hostPath:
# directory location on host
path: /data
# this field is optional
type: Directory
Managing running pods
$ kubectl get pod shell-demo # ← ensure pod is running
$ kubectlºlogsºmy-pod (-c my-container) (-f) # ← Get logs
^^
"tail"
$ kubectlºattachºmy-pod -i
$ kubectlºport-forwardºmy-pod 5000:6000
^ ^
local pod
machine port
$ kubectlºrunº-i \ # ← Exec. shell for image
--tty busybox \ # (launch new temporal pod)
--image=busybox \
-- sh
$ kubectlºexecºmy-pod \ # ← Get a shell inside already-running Pod
(-c my-container) # ← Required if Pod has 2+ containers
-it # ← Allocate terminal.
-- /bin/bash
$ kubectlºexecºmy-pod \ # ← Execute non-interatively inside already-running Pod
(-c my-container) # ← Required if Pod has 2+ containers
env # ← env execution will dump all ENV.VARs inside container
$ kubectlºtopºpod POD_NAME --containers
SecurityContext
@[https://kubernetes.io/docs/tasks/configure-pod-container/security-context/]
@[https://github.com/kubernetes/community/blob/master/contributors/design-proposals/auth/security_context.md]
Pod.spec.securityContext Ex:
Ex:
apiVersion: v1 Exec a terminal into a running container adnd execute:
kind: Pod $ id
metadata: ºuid=1000 gid=3000 groups=2000º
name: security-context-demo ^^^^ ^^^^
spec: uid/gid 0/0 if not specified
ºsecurityContext: º
º runAsUser: 1000 º
º runAsGroup: 3000º
º fsGroup: 2000º ← owned/writable by this GID when supported by volume
volumes:
- name: sec-ctx-vol ← Volumes will be relabeled with provided seLinuxOptions values
emptyDir: {}
containers:
- name: sec-ctx-demo
...
volumeMounts:
- name: sec-ctx-vol
mountPath: /data/demo
ºsecurityContext:º
º allowPrivilegeEscalation: falseº
º capabilities:º
º add: ["NET_ADMIN", "SYS_TIME"]º ← Provides a subset of 'root' capabilities:
@[https://github.com/torvalds/linux/blob/master/include/uapi/linux/capability.h]
º seLinuxOptions:º
º level: "s0:c123,c456"º
SecurityContext holds security configuration that will be applied to a container.
Some fields are present in both SecurityContext and PodSecurityContext. When both
are set, the values in SecurityContext take precedence.
typeºSecurityContextºstruct {
// The capabilities to add/drop when running containers.
// Defaults to the default set of capabilities granted by the container runtime.
// +optional
Capabilities *Capabilities
// Run container in privileged mode.
// Processes in privileged containers are essentially equivalent to root on the host.
// Defaults to false.
// +optional
Privileged *bool
// The SELinux context to be applied to the container.
// If unspecified, the container runtime will allocate a random SELinux context for each
// container. May also be set in PodSecurityContext. If set in both SecurityContext and
// PodSecurityContext, the value specified in SecurityContext takes precedence.
// +optional
SELinuxOptions *SELinuxOptions
// Windows security options.
// +optional
WindowsOptions *WindowsSecurityContextOptions
// The UID to run the entrypoint of the container process.
// Defaults to user specified in image metadata if unspecified.
// May also be set in PodSecurityContext. If set in both SecurityContext and
// PodSecurityContext, the value specified in SecurityContext takes precedence.
// +optional
RunAsUser *int64
// The GID to run the entrypoint of the container process.
// Uses runtime default if unset.
// May also be set in PodSecurityContext. If set in both SecurityContext and
// PodSecurityContext, the value specified in SecurityContext takes precedence.
// +optional
RunAsGroup *int64
// Indicates that the container must run as a non-root user.
// If true, the Kubelet will validate the image at runtime to ensure that it
// does not run as UID 0 (root) and fail to start the container if it does.
// If unset or false, no such validation will be performed.
// May also be set in PodSecurityContext. If set in both SecurityContext and
// PodSecurityContext, the value specified in SecurityContext takes precedence.
// +optional
RunAsNonRoot *bool
// The read-only root filesystem allows you to restrict the locations that an application can write
// files to, ensuring the persistent data can only be written to mounts.
// +optional
ReadOnlyRootFilesystem *bool
// AllowPrivilegeEscalation controls whether a process can gain more
// privileges than its parent process. This bool directly controls if
// the no_new_privs flag will be set on the container process.
// +optional
AllowPrivilegeEscalation *bool
// ProcMount denotes the type of proc mount to use for the containers.
// The default is DefaultProcMount which uses the container runtime defaults for
// readonly paths and masked paths.
// +optional
ProcMount *ProcMountType
}
─────────────────────────────────────────────────────────────────────────────
// PodSecurityContext holds pod-level security attributes and common container settings
// Some fields are also present in container.securityContext. Field values of
// container.securityContext take precedence over field values of PodSecurityContext.
typeºPodSecurityContextºstruct {
// Use the host's network namespace. If this option is set, the ports that will be
// used must be specified.
// Optional: Default to false
// +k8s:conversion-gen=false
// +optional
HostNetwork bool
// Use the host's pid namespace.
// Optional: Default to false.
// +k8s:conversion-gen=false
// +optional
HostPID bool
// Use the host's ipc namespace.
// Optional: Default to false.
// +k8s:conversion-gen=false
// +optional
HostIPC bool
// Share a single process namespace between all of the containers in a pod.
// When this is set containers will be able to view and signal processes from other conts
// in the same pod, and the first process in each container will not be assigned PID 1.
// HostPID and ShareProcessNamespace cannot both be set.
// Optional: Default to false.
// This field is beta-level and may be disabled with the PodShareProcessNamespace feature.
// +k8s:conversion-gen=false
// +optional
ShareProcessNamespace *bool
// The SELinux context to be applied to all containers.
// If unspecified, the container runtime will allocate a random SELinux context for each
// container. May also be set in SecurityContext. If set in
// both SecurityContext and PodSecurityContext, the value specified in SecurityContext
// takes precedence for that container.
// +optional
SELinuxOptions *SELinuxOptions
// Windows security options.
// +optional
WindowsOptions *WindowsSecurityContextOptions
// The UID to run the entrypoint of the container process.
// Defaults to user specified in image metadata if unspecified.
// May also be set in SecurityContext. If set in both SecurityContext and
// PodSecurityContext, the value specified in SecurityContext takes precedence
// for that container.
// +optional
RunAsUser *int64
// The GID to run the entrypoint of the container process.
// Uses runtime default if unset.
// May also be set in SecurityContext. If set in both SecurityContext and
// PodSecurityContext, the value specified in SecurityContext takes precedence
// for that container.
// +optional
RunAsGroup *int64
// Indicates that the container must run as a non-root user.
// If true, the Kubelet will validate the image at runtime to ensure that it
// does not run as UID 0 (root) and fail to start the container if it does.
// If unset or false, no such validation will be performed.
// May also be set in SecurityContext. If set in both SecurityContext and
// PodSecurityContext, the value specified in SecurityContext takes precedence
// for that container.
// +optional
RunAsNonRoot *bool
// A list of groups applied to the first process run in each container, in addition
// to the container's primary GID. If unspecified, no groups will be added to
// any container.
// +optional
SupplementalGroups []int64
// A special supplemental group that applies to all containers in a pod.
// Some volume types allow the Kubelet to change the ownership of that volume
// to be owned by the pod:
//
// 1. The owning GID will be the FSGroup
// 2. The setgid bit is set (new files created in the volume will be owned by FSGroup)
// 3. The permission bits are OR'd with rw-rw----
//
// If unset, the Kubelet will not modify the ownership and permissions of any volume.
// +optional
FSGroup *int64
// Sysctls hold a list of namespaced sysctls used for the pod. Pods with unsupported
// sysctls (by the container runtime) might fail to launch.
// +optional
Sysctls []Sysctl
}
ServiceAccount
@[https://kubernetes.io/docs/reference/access-authn-authz/service-accounts-admin/]
@[https://kubernetes.io/docs/tasks/configure-pod-container/configure-service-account/]
- Processes in containers inside pods can also contact the apiserver.
When they do, they are authenticated as a particular Service Account
- provides an identity for processes that run in a Pod.
- to access the API from inside a pod a mounted service account with
a 1-hour-expiration-token is provided.
(can be disabled with automountServiceAccountToken)
- User accounts are for humans. Service accounts are for processes running on pods.
- Service accounts are namespaced.
- cluster users can create service accounts for specific tasks (i.e. principle of least privilege).
- Allows to split auditing between humans and services.
- Three separate components cooperate to implement the automation around service accounts:
- AºService account admission controllerº, part of the apiserver, thatsynchronously modify
pods as they are created or updated:
- sets the ServiceAccount to default when not specified by pod.
- abort if ServiceAccount in the pod spec does NOT exists.
- ImagePullSecrets of ServiceAccount are added to the pod if none is specified by pod.
- Adds a volume to the pod which contains a token for API access.
(service account token expires after 1 hour or on pod deletion)
- Adds a volumeSource to each container of the pod mounted at
*/var/run/secrets/kubernetes.io/serviceaccount*
- AºToken controllerº, running as as part of controller-manager. It acts asynchronously and
- observes serviceAccount creation and creates a corresponding Secret to allow API access.
- observes serviceAccount deletion and deletes all corresponding ServiceAccountToken Secrets.
- observes secret addition, and ensures the referenced ServiceAccount exists, and adds a
token to the secret if needed.
- observes secret deletion and removes a reference from the corresponding ServiceAccount if needed.
NOTE:
controller-manageRº'--service-account-private-key-file'º indicates the priv.key signing tokens
- kube-apiserver º'--service-account-key-file'º indicates the matching pub.key verifying signatures
- A controller loop ensures a secret with an API token exists for each service account.
To create additional API tokens for a service account:
- create a new secret of type ServiceAccountToken like:
(controller will update it with a generated token)
{
"kind":º"Secret"º,
"type":º"kubernetes.io/service-account-token"º
"apiVersion": "v1",
"metadata": {
"name": "mysecretname",
"annotations": {
"kubernetes.io/service-account.name" :
"myserviceaccount" ← reference to service account
}
},
}
- AºService account controller:º
- manages ServiceAccount inside namespaces,
- ensures a ServiceAccount named “default” exists in every active namespace.
$ kubectl get serviceAccounts
NAME SECRETS AGE
default 1 1d ← Exists for each namespace
...
- Additional ServiceAccount's can be created like:
-
- $ kubectl apply -f - ˂˂EOF
- apiVersion: v1
- kind: ServiceAccount
- metadata:
- name: build-robot
- EOF
Liveness/Readiness Probes
@[https://kubernetes.io/docs/tasks/configure-pod-container/configure-liveness-readiness-probes/]
@[https://kubernetes.io/docs/reference/generated/kubernetes-api/v1.14/#probe-v1-core]
.
- liveness probes : used by kubelet to know when to restart a Container.
- readiness probes : used by kubelet to know whether a Container can accept new requests.
- A Pod is considered ready when all of its Containers are ready.
- The Pod is not restarted in this case but no new requests are
forwarded.
- Readiness probes continue to execute for the lifetime of the Pod
(not only at startup)
ºliveness COMMANDº │ ºliveness HTTP requestº │ ºTCP liveness probeº
│ │
apiVersion: v1 │ apiVersion: v1 │ apiVersion: v1
kind:ºPodº │ kind:ºPodº │ kind:ºPodº
metadata: │ metadata: │ metadata:
... │ ... │ ...
spec: │ spec: │ spec:
ºcontainers:º │ ºcontainers:º │ ºcontainers:º
- name: ... │ - name: ... │ - name: ...
... │ ... │ ºreadinessProbe: º
ºlivenessProbe: º │ ºlivenessProbe: º │ º tcpSocket: º
º exec: º │ º httpGet: º │ º port: 8080 º
º command: º │ º path: /healthz º │ º initialDelaySeconds: 5 º
º - cat º │ º port: 8080 º │ º periodSeconds: 10 º
º - /tmp/healthy º │ º httpHeaders: º │ ºlivenessProbe: º
º initialDelaySeconds: 5º │ º - name: Custom-HeadeRº │ º tcpSocket: º
º periodSeconds: 5 º │ º value: Awesome º │ º port: 8080 º
│ º initialDelaySeconds: 3 º │ º initialDelaySeconds: 15º
│ º periodSeconds: 3 º │ º periodSeconds: 20 º
HTTP proxy ENV.VARS is ignored
in liveness probes (k8s v1.13+)
↑ ↑
│ │
ºnamedºContainerPort can also ──────┴────────────────────────────────┘
be used for HTTP and TCP probes
like:
│ ports:
│ - name: liveness-port
│ containerPort: 8080
│ hostPort: 8080
│
│ livenessProbe:
│ httpGet:
│ path: /healthz
│ port: liveness-port
ºMonitor liveness test resultº
$ kubectl describe pod liveness-exec
Other optional parameters:
- timeoutSeconds : secs after which the probe times out. (1 sec by default)
- successThreshold: Minimum consecutive successes for the probe to be
considered successful after having failed. (1 by default)
- failureThreshold: number of failures before "giving up": Pod will be marked Unready.(Def 3)
HTTP optional parameters:
- host : for example 127.0.0.1 (defaults to pod IP)
- port : Name or number(1-65535)
- scheme: HTTP or HTTPS(skiping cert.validation)
- path :
- httpHeaders: Custom headers
PodDisruptionBudget
@[https://kubernetes.io/docs/concepts/workloads/pods/disruptions/#how-disruption-budgets-work]
@[https://kubernetes.io/docs/tasks/run-application/configure-pdb/]
A PodDisruptionBudget object (PDB) can be defined for each deployment(application),
limiting the number of pods of a replicated application that
are down simultaneously from ºvoluntaryº disruptions.
Ex: A Deployment has:
- .spec.replicas: 5 (5 pods desired at any given time)
- The PDB is defined as:
apiVersion: policy/v1beta1
kind: PodDisruptionBudget
metadata:
name: zookeeper-pdb
spec:
maxUnavailable: 1 ←-- Eviction API will allow voluntary disruption of one,
selector: but not two pods, at a time trying to have 4 running
matchLabels: at all times
app: zookeeper
Eviction API compatible tools/commands
like 'kubectl drain' must be used
(vs directly deleting pods/deployments)
Services
Service
Problem: Pods can be rescheduled to a different node in k8s pool,
"at random", changing their internal cluster IP.
How Pods can deterministically access some other Pod's
services (TCP/UDP/Web service, ...)?
Solution: Use k8s Services
- A Services logically groups a set of Pods through a Label-selection,
and for such logical group, it also established a ºnetwork policyº.
to access them.
- Kubernetes-native applications: K8s offers a simple Endpoints API
- non-native applications: K8s offers virtual-IP-based-2-Service bridge
-OºService Types:º
- OºClusterIPº : (default) internal-to-cluster virtual IP.
No IP exposed outside the cluster.
- OºExternalNameº: Maps apps (Pods services) to an externally
visible DNS entry (ex: foo.example.com)
─ OºNodePortº : Exposes an external NodeIP:NodePort to ←───────┐ k8s ºIngressº can also be used
Pods providing the service. ├─ to expose HTTP/S services to
│ external clients.
─ OºLoadBalancerº: Exposes Service externally using a cloud provider's ←─┘ NodePort/LoadBalancer Services
load balancer. NodePort and ClusterIP services, allows for TCP,UDP,PROXY and SCTP(k8s 1.2+)
to which the external load balancer routes, are - Advanced load balancing
automatically created. (persistent sessions, dynamic weights)
are RºNOTº yet supported by Ingress rules
- On the other side ingress allows
to expose services based on
HTTP host (virtual hosting), paths/regex,...
not available to NodePort/LoadBalancer
Services
kube-proxy, installed on each node of the k8s pool, will get in charge of
forwarding correctly to the correct target.
Note: kube-proxy is more reliable than DNS due to problems with TTLs,
DNS caching on clients,...
OºClusterIPº │ OºNodePortº │ OºExternalNameº
Oº=========º │ Oº========º │ Oº============º
│ │
┌───┐ │ :3100┌────┐ :80┌───┐ │ ┌────┐:80───┐
┌─→:80│Pod│ │ ┌───→│node├───┐ ┌──→│Pod│ │ ┌─→│node├─→│Pod│
│ └───┘ │ │ └────┘ │ │ └───┘ │ │ └────┘ └───┘
:80┌───────────┐ │ Incomming ┌──v───────┐ │ :80 ┌──────────┐
Internal──→│ºClusterIPº│ │ traffic │ºNodePortº│ │ ┌─→:│ ºLoadº │ ┌───┐
Traffic └───────────┘ │ │ └──^───────┘ │ │ │ºBalancerº│ │Pod│
│ ┌───┐ │ │ ┌────┐ │ │ ┌───┐ │ │ └──────────┘ └───┘
└─→:80│Pod│ │ └───→│node├───┘ └──→│Pod│ │ Incomming │ ┌────┐ ↑:80
└───┘ │ :3100└────┘ :80└───┘ │ traffic └──→│node├──┘
│ ^ ^ │ └────┘
│ Creates mapping between │ configures an external
│ node port(3100) and │ IP address in balancer.
│ Pod port (80) │ Then connects requested
│ │ pods to balancer
apiVersion: v1
kind:ºServiceº
metadata:
name: Redis_Master ← Must be a valid DNS label name
spec:
┌→ · selector: ← Targeted Pods defined in selector
| · app: MyApp ← Matching label.
| · clusterIP: 1.1.1.1 ← Optional. set it to match any existing DNS entry or
| · legacy systems with hardcoded IPs difficult to reconfigure
| · In k8s apps, clusterIP can be auto-assigned by k8s,
| · App will use Environment Variables (injected into the Pod)
| · and/or DNS.
| · REDIS_MASTER_SERVICE_HOST=10.0.0.11
| · REDIS_MASTER_SERVICE_PORT=6379
| · REDIS_MASTER_PORT=tcp://10.0.0.11:6379
| · REDIS_MASTER_PORT_6379_TCP=tcp://10.0.0.11:6379
| · REDIS_MASTER_PORT_6379_TCP_PROTO=tcp
| · REDIS_MASTER_PORT_6379_TCP_PORT=6379
| · REDIS_MASTER_PORT_6379_TCP_ADDR=10.0.0.11
| · RºWARNº: Service must be setup before Pod is started.
|
|
| · The DNS entry (if CoreDNS or other k8s-aware DNS server is setup)
| · will be like:
| · redis_master.myNameSpace
| · ^^^^^^^^^^^^
| · Needed only for Pods in another Namespace.
| · ports:
| · - name: http
| · protocol: TCP ← TCP*, UDP, HTTP, PROXY, SCTP(k8s 1.2+) º*1º
| · port: 80 ← Request to port 80 will be forwarded to Port 9376
| · targetPort: 9376 on "any" pod matching the selector.
| · - name: https
| · protocol: TCP
| · port: 443
| · targetPort: 9377
| · sessionAffinity: "ClientIP" ← Optionally sessionAffinityConfig.clientIP.timeoutSeconds
| can be set (defaults to 10800)
|
└ The selector can be empty if Service doesn't target internal pod.
This is the case when we want the Service to point to:
┌ - External tcp service (DDBB, ERP, ...)
| - Different Namespace or differente K8s cluster.
| - k8s migration is in progress.
| In this case the Endpoint object is not automatically created. Admins can point manually
| to an external end-point like:
|
| apiVersion: v1
| kind:ºEndpointsº
| metadata:
| name: my-service ← Must also be a valid DNS subdomain name.
└→ subsets:
- addresses:
- ip: 10.0.0.10 ← External PostgreSQL @10.0.0.10:5432
- ports:
- port: 5432
ºUser-space Proxy Mode:º
(kube-proxy on each k8s node of k8s node-pool keeps listening for events from Master.ApiServer)
master → kube-proxy: Event +Service.EndPoint
kube-proxy → kube-proxy: - Open Random Port 01
- Setup round-robin from random-port 01 to *all existing*
Backend Pods matching the Service selector.
kube-proxy → iptables : Install iptables rule in kernel to capture traffice to
(virtual) clusterIP:port to Random Port 01
...
app → kernel : request to clusterIP:port
kernel → iptables : request to clusterIP:port
iptables → kube-proxy: request to random port
kube-proxy → kube-proxy: Round-robin amongst all availables Pods.
(The kube-proxy keeps a list of live-Pods matching the Service.Selector)
(Using also SessionAffinity for better balancing)
kube-proxy → Pod N : request
ºiptables Proxy Mode:º
(kube-proxy on each k8s node of k8s node-pool keeps listening for events from Master.ApiServer)
master → kube-proxy: Event +Service.EndPoint
kube-proxy → iptables : Install iptables rule in kernel to capture traffice to
(virtual) clusterIP:port to some Pod i (and just one) from
all Pods matching the Service.selecotr
...
app → kernel : request to clusterIP:port
kernel → iptables : request to clusterIP:port
iptables → Pod i : request
^^^^^
- If Pod is down request fails.
- readiness probes can be set to avoid forwarding to a faulty Pod.
ºIPVS proxy modeº
PRE-SETUP: IPVS module must be available on the node.
master → kube-proxy: Event +Service.EndPoint
kube-proxy → netlink : Create IPVS rules (sync status periodically)
to all Pods in Selector (allowing for balancing)
app → kernel : request to clusterIP:port
kernel → netlink : request to clusterIP:port
netlink → Pod i : request
Note: netlink allows next balancing options:
· rr: round-robin
· lc: least connection (smallest number of open connections)
· dh: destination hashing
· sh: source hashing
· sed: shortest expected delay
· nq: never queue
º*1º:@[https://kubernetes.io/docs/concepts/services-networking/service/#protocol-support]
ºLoad Balancerºworks at ºlayer 4º.
unaware of the actual apps.
(Layer 7) ┌───┐ ┌───┐ ☜ ☞ Use OºIngress controllersº(Oºlayer 7º)
┌─────────────────────┐ │Pod│ │Pod│ for advanced rules based on inbound
│OºIngress controllerº│ └─^─┘ └─^─┘ URL, ...
│ ┌─────────────────┐ │ ┌─┴────────┴───┐ - Another common feature of Ingress
│ │ myapp.com/blog │ │───→ │Blog service │ isºSSL/TLS terminationºremoving
│ └─────────────────┘ │ └──────────────┘ TLS cert complexity from Apps
│ ┌─────────────────┐ │ ┌──────────────┐
│ │ myapp.com/store │ │────→│Store service │
│ └─────────────────┘ │ └─┬────────┬───┘
└───^─────────────────┘ ┌─v─┐ ┌─v─┐
| │Pod│ │Pod│
Incomming └───┘ └───┘
Traffic
-ºService.spec.clusterIP can be used to force a given IPº
┌─────────────────┐
│SERVICE DISCOVERY│
├─────────────────┴─────────────────────────────────────────────────────────────────────────────────────────
│ ALT.1, using ENV.VARS │ ALT.2, using the k8s DNS add─on
│ │ (strongly recommended).
├─────────────────────────────────────────────────┼──────────────────────────────────────────────────────────
│Ex: given service "redis─master" with │ Updates dynamically the set of DNS records for services.
│ OºClusterIPº = 10.0.0.11:6379 │ The entry will be similar to: "my─service.my─namespace"
│ next ENV.VARs are created: │
│ REDIS_MASTER_SERVICE_HOST=10.0.0.11 │
│ REDIS_MASTER_SERVICE_PORT=6379 │
│ REDIS_MASTER_PORT=tcp://10.0.0.11:6379 │
│ REDIS_MASTER_PORT_6379_TCP=tcp://10.0.0.11:6379│
│ REDIS_MASTER_PORT_6379_TCP_PROTO=tcp │
│ REDIS_MASTER_PORT_6379_TCP_PORT=6379 │
│ REDIS_MASTER_PORT_6379_TCP_ADDR=10.0.0.11 │
└─────────────────────────────────────────────────┴──────────────────────────────────────────────────────────
ºQuick App(Deployment) service creationº
REF: @[https://kubernetes.io/docs/concepts/services-networking/connect-applications-service/]
$ kubectl expose deployment/my-nginx
is equivalent to:
$ kubectl apply -f service.yaml
^^^^^^^^^^^^
apiVersion: v1
kind: Service
metadata:
name: my-nginx
labels:
run: my-nginx
spec:
ports:
- port: 80
protocol: TCP
selector:
run: my-nginx
@[https://kubernetes.io/docs/concepts/services-networking/add-entries-to-pod-etc-hosts-with-host-aliases/]
Debug Services
@[https://kubernetes.io/docs/tasks/debug-application-cluster/debug-service/]
☞ (Much more) detailed info available at:
( about kube-proxy + iptables +... advanced config settings)
@[https://kubernetes.io/docs/concepts/services-networking/service/]
Ingress Rules
@[https://kubernetes.io/docs/concepts/services-networking/ingress/]
- Ingress objects (and its rules) allows to exposes HTTP/S based k8s services to
the internal cluster.
Ussually a WebApp (HTTP/s) can be summarised as:
Creating Pods → Creating Services → Creating Ingress Rules
running on to access Pods from to expose HTTP/S Services
k8s pool-nodes other Ports. externally allowing to
listening for Services listen for split requests based on host name,
HTTP/s requests Pod creation/destruction HTTP path/subpaths, ...
events to adjust
the correct node/port
target
PRE-SETUP:
Prerequisites
- an ingress controller must be in place.
ingress-nginx, ingress-istio, etc...
RºWARNº: Different Ingress controllers operate slightly differently to expected specification.
BºIngress Resource Example:º
apiVersion: networking.k8s.io/v1beta1
ºkind: Ingressº
metadata:
· name: test-ingress
· annotations:
· nginx.ingress.kubernetes.io/rewrite-target: / ← Annotation used before IngressClass was added
· to define the underlying controller implementation.
· .ingressClassName can replace it now.
spec:
tls: ← Optional but recomended. Only 443 port supported.
· - hosts: ← values must apiVersion: v1
· - secure.foo.com match cert CNs kind: Secret
· metadata:
· secretName: secretTLSPrivateKey ←·····→ name: secretTLSPrivateKey
· namespace: default
· data:
· tls.crt: base64 encoded cert
· tls.key: base64 encoded key
· type: kubernetes.io/tls
rules: ← A backend with no rules could be use to expose a single service
- host: reddis.bar.com ← Optional. filter-out HTTP requests not addressing this host
http:
paths: ← For each path in paths list a
If more than on path in list matches the request, longest matchs "wins".
- path: /reddis
pathType: Prefix ← ImplementationSpecific: delegate matching to IngressClass (nginx, istio, ...)
Exact : Matches URL path exactly (RºWARNº:case sensitive).
Prefix : Matching based on request URL prefix (RºWARNº:case sensitive)
backend:
serviceName: reddissrv ← targeted k8s service
servicePort: 80 ← Need if more than a Listening port is defined for Service
- host: monit.bar.com ← Optional. filter-out HTTP requests not addressing this host
http:
- path: /grafana
pathType: Prefix
backend:
serviceName: grafanasrv ← targeted k8s service
- path: /prometheus
pathType: Prefix
backend:
serviceName: prometheussrv ← targeted k8s service
- http: ← No host specified. All traffice not matching reddis.bar.com/monit.bar.com
- path: / will be forwarded here.
pathType: Prefix
backend:
serviceName: httpdsrv ← targeted k8s service
Labels
Labels
@[https://kubernetes.io/docs/concepts/overview/working-with-objects/labels/#label-selectors/]
- Labels = key/value pairs attached to objects.
- labelsºdo not provide uniquenessº
- we expect many objects to carry the same label(s)
- used to specify ºidentifying attributes of objectsº
ºmeaningful and relevant to usersº, (vs k8s core system)
- Normally used to organize and to select subsets of objects.
- Label format:
("prefix"/)name
name : [a-z0-9A-Z\-_.]{1,63}
prefix : must be a DNS subdomain no longer than 253 chars
- Example labels:
"release" : "stable" # "canary" ...
"environment": "dev" # "qa", "pre", "production"
"tier" : "frontend" # "backend" "cache"
"partition" : "customerA", "partition" : "customerB"
"track" : "daily" # "weekly" "monthly" ...
ºRecommended Labelsº
@[https://kubernetes.io/docs/concepts/overview/working-with-objects/common-labels/]
Key Description Example Type
app.kubernetes.io/name The name of the application mysql string
app.kubernetes.io/instance A unique name identifying wordpress-abcxzy string
the instance of an application
app.kubernetes.io/version current version of the app 5.7.21 string
app.kubernetes.io/component component within the arch database string
app.kubernetes.io/part-of name of higher level app wordpress string
app.kubernetes.io/managed-by tool used to manage the helm string
operation of an application
Ex.1 use in an StatefulSet object:
apiVersion: apps/v1 |apiVersion: apps/v1 |apiVersion: v1
kind: StatefulSet |kind: Deployment |kind: Service
metadata: |metadata: |metadata:
labels: | labels: | labels:
app.kubernetes.io/name : mysql | .../name : wordpress | .../name : wordpress
app.kubernetes.io/instance : wordpress-abc | .../instance : wordpress-abc | .../instance : wordpress-abcxzy
app.kubernetes.io/version : "5.7.21" | .../version : "4.9.4" | .../version : "4.9.4"
app.kubernetes.io/component : database | .../component : server | .../component : server
ºapp.kubernetes.io/part-of :ºwordpressº | .../part-of :ºwordpressº | .../part-of : wordpressº
app.kubernetes.io/managed-by: helm | .../managed-by: helm | .../managed-by: helm
|... |...
-ºWell-Known Labels, Annotations and Taintsº
@[https://kubernetes.io/docs/reference/kubernetes-api/labels-annotations-taints/]
kubernetes.io/arch
kubernetes.io/os
beta.kubernetes.io/arch (deprecated)
beta.kubernetes.io/os (deprecated)
kubernetes.io/hostname
beta.kubernetes.io/instance-type
failure-domain.beta.kubernetes.io/region
failure-domain.beta.kubernetes.io/zone
Label selectors
- core (object) grouping primitive.
- two types of selectors:
- equality-based
Ex. environment=production,tier!=frontend
^
"AND" represented with ','
- set-based
- filter keys according to a set of values.
(in, notin and exists )
Ex.'ºkeyºequal to environment andºvalueºequal to production or qa'
Ex.'environment in (production, qa)'
Ex.'tier notin (frontend, backend)'
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
all resources with
(key == "tier" AND
values != frontend or backend ) AND
all resources with
(key != "tier")
Ex:'partition' !partition
^^^^^^^^^ ^^^^^^^^^
all resources including all resources without
a label with key 'partition' a label with key 'partition'
- LIST and WATCH operations may specify label selectors to filter the sets
of objects returned using a query parameter.
Ex.:
$ kubectl get pods -l environment=production,tier=frontend # equality-based
$ kubectl get pods -l 'environment in (production),tier in (frontend)' # set-based
$ kubectl get pods -l 'environment in (production, qa)' # "OR" only set-based
$ kubectl get pods -l 'environment,environment notin (frontend)' # "NOTIN"
Storage
Volume
@[https://github.com/kubernetes/community/blob/master/contributors/design-proposals/storage/volume-ownership-management.md]
REF: https://www.youtube.com/watch?v=OulmwTYTauI
once underlying "hardware" resource applying
storage has been to the whole cluster
assigned to a pod (ºlifespanº independent of pods)
┌──────────────────┐ PV is bound to PVC ┌─↓───────────┐
│ Pod │ one-to-one relation┌···→│ Persistent ·····plugin to
│ │ ┌───────────────────┐· │ Volume(ºPVº)│ one of
│┌───────────────┐ │ ┌··→│ Persistent *1 │· └─────────────┘ · Underlying
││ºVolume Mountsº│ │ · │ Volume │· · Storage
││ /foo │ │ · │ Claim (ºPVCº) │· ┌───↓────────────────┐
│└───────────────┘ │ · │ │· ┌────────────┐ │-Local HD │
│ ┌──────────────┐· │- 100Gi │· ┌·→│ Storage │ │-NFS: path,serverDNS│
│ │ Volumes: │· │- Selector ·········┘ · │ Class(ºSCº)│ │-Amazon:EBS,EFS,... │
│ │ -PVC ·········┘ │- StorageClassName····┘ └────────────┘ │-Azure:... │
│ │ ─claimName ←┐ └───────────────────┘ │-... │
│ └──────────────┘└─ Volumeºlifespamºis └────────────────────┘
└──────────────────┘ that of the Pod.
├────────Created by users (Dev)───────────────────┤ ├──────Created by Admin (Ops)────────────┤
- many types supported, and a Pod can use any number of them simultaneously.
- a pod specifies what volumes to provide for the pod (spec.volumes) and where
to mount those into each container(spec.containers.volumeMounts).
- Sometimes, it is useful to share one volume for multiple uses in a single
pod. The volumeMounts.subPath property can be used to specify a sub-path inside
the referenced volume instead of its root.
*1:Similar to how Pods can request specific levels of resources
(CPU and Memory), Claims can request specific size and access modes
(e.g., can be mounted once read/write or many times read-only)
☞ BºPV contains max size, PVC contains min sizeº
wih PV max.size ˃ PVC min.size must be ˂ PV max.size
Networking Model
@[https://kubernetes.io/docs/concepts/cluster-administration/networking/]
Storage
Local Persistent Vol (1.12+)
REF: @[https://kubernetes.io/blog/2019/04/04/kubernetes-1.14-local-persistent-volumes-ga/]
Authors: Michelle Au (Google), Matt Schallert (Uber), Celina Ward (Uber)
- local persistent volume (k8s GA 1.14+) leverage local disks and it enables to use
them with persistent volume claims (pvc).
- Local PV provide for consistent high performance guarantees of local directly-attached
storage.
-Bºthis type of storage is suitable for applications which handles the dataº
Bºreplication themself. software defined storage or replicated databases º
Bºlike kafka, Ethereum, "p2p", cassandra, ...)º.
Apps that do not will better opt for standard (remote) PVs that transparently
takes care of data replication.
- to use this as a persistent volume, we have some manual steps to do:
- Pre-partition, format and mount disks to nodes
- Create Persistent Volumes:
- Alt 1: Manually
- Alt 2: Let a DaemonSet handle the creation.
- Create Storage Class
(can be skipeed in newer k8s releases)
NOTE:
Q: k8s "hostPath" already allows to use local disk as a storage for Pods.
why SHOULD we use Local Persistent Volume instead?
A: With Local PV, the Scheduler is aware of the Local PV, so on Pod
re-start, execution will be assigned to the same worker node.
With hostPath, k8s can re-schedule in a different node, loosing all
previously stored data.
Note: Most other k8s block and file storage plugins provide for remote storage
where data is persisted independently of the node excuting the Pod.
- raw block device support is also provided.
- K8s is also able to format the block device on demand, avoiding manual formating.
- dynamic volume provisioning RºNOTº supported.
(@[https://github.com/kubernetes-sigs/sig-storage-local-static-provisioner]
can be used to help manage the Local PV lifecycle - create,clean up,reuse -
for individual disks),
BºUssage Steps:º
Bº└RE-SETUP Plannification:º
- how many local disks would each node cluster have?
- How would they be partitioned?
(The local static provisioner provides guidance to help answer these questions).
Hint: It’s best to be able to dedicate a full disk to each local volume (for IO isolation)
and a full partition per-volume (for capacity isolation).
Bº└1.- Create StorageClass:º
Qº kind: StorageClass º
Qº apiVersion: storage.k8s.io/v1 º
Qº metadata: º
Qº name: local-storage º
Qº provisioner: kubernetes.io/no-provisionerº
Qº volumeBindingMode: WaitForFirstConsumer º ← *1: enable Bºvolume topology-aware schedulingº
└────── *1 ────────┘ Consumer == "Pod requesting the volume"
Bº└2.-ºthe external static provisioner can be configured and run to (TODO)
create PVs for all the local disks on your nodes.
$º$ kubectl get pv º
NAME CAPA ACC. RECLAIM STATUS CLAIM STORAGECLASS REASON AGE
MODE POLICY
local-pv-27c0f084 368Gi RWO Delete Available local-storage 8s
local-pv-3796b049 368Gi RWO Delete Available local-storage 7s
local-pv-3ddecaea 368Gi RWO Delete Available local-storage 7s
Bº└3.- Start using PVs in workloads by:º
- Alt 1: creating a PVC and Pod
- Alt 2: creating a StatefulSet with volumeClaimTemplates
Ex:
QºapiVersion: apps/v1 º
Qºkind: StatefulSet º
Qºmetadata: º
Qº name: local-test º
Qºspec: º
Qº serviceName: "local-service" º
Qº replicas: 3 º
Qº selector: º
Qº matchLabels: º
Qº app: local-test º
Qº template: º
Qº metadata: º
Qº labels: º
Qº app: local-test º
Qº spec: º
Qº containers: º
Qº - name: test-container º
Qº image: k8s.gcr.io/busybox º
Qº command: º
Qº - "/bin/sh" º
Qº args: º
Qº - "-c" º
Qº - "sleep 100000" º
Qº volumeMounts: º
Qº - name: local-vol º
Qº mountPath: /usr/test-pod º
Qº volumeClaimTemplates: º
Qº - metadata: º
Qº name: local-vol º
Qº spec: º
Qº accessModes: [ "ReadWriteOnce" ] º
Qº storageClassName: "local-storage"º
Qº resources: º
Qº requests: º
Qº storage: 368Gi º
Once the StatefulSet is up and running,
the PVCs will be bound:
$º$ kubectl get pvc º
NAME STATUS VOLUME CAPACITY ACC.. STORAGECLASS AGE
local-vol-local-test-0 Bound local-pv-27c0f084 368Gi RWO local-storage 3m45s
local-vol-local-test-1 Bound local-pv-3ddecaea 368Gi RWO local-storage 3m40s
local-vol-local-test-2 Bound local-pv-3796b049 368Gi RWO local-storage 3m36s
Bº└4.- Automatic Clean Up Ex:º
$º$ kubectl patch sts local-test \ º
$º -p '{"spec":{"replicas":2}}' º
└────────────┘
Reduce Pod replicates 3→2
Associated local-pv-... is not needed anymore.
The external static provisioner will clean up
the disk and make the PV available for use again.
$ºstatefulset.apps/local-test patchedº
$º$ kubectl delete pvc local-vol-local-test-2º
persistentvolumeclaim "local-vol-local-test-2" deleted
$ kubectl get pv
$º$ kubectl get pv º
NAME CAPA ACC. RECLAIM STATUS CLAIM STORAGECLASS REASON AGE
MODE POLICY
local-pv-27c0f084 368Gi RWO Delete Bound ...-0 local-storage 11m
local-pv-3796b049 368Gi RWO Delete Available local-storage 7s
local-pv-3ddecaea 368Gi RWO Delete Bound ...-1 local-storage 19m
└─┬─┘
default/local-vol-local-test-0
default/local-vol-local-test-1
BºLIMITATIONS AND CAVEATSº:
- It ties apps to a specific node, making it harder to schedule.
Those apps Bºshould specify a high priorityº so that lower priority pods,
can be preempted if necessary.
Also if the tied-to node|local volume become inaccessible, then the pod
also becomes inaccessible requiring manual intervention, external controllers
or operators.
If a node becomes unavailable (removed from the cluster or drained), pods using
local volumes on that node are stuck in "Unknown" or "Pending" state depending
on whether or not the node was removed gracefully.
Recovering pods from these interim states means having to delete the PVC binding the pod
to its local volume and then delete the pod in order for it to be rescheduled
(or wait until the node and disk are available again).
"... We took this into account when building our operator for M3DB, which makes changes
to the cluster topology when a pod is rescheduled such that the new one gracefully
streams data from the remaining two peers..."
Because of these constraints, Rºit’s best to exclude nodes with º
Rºlocal volumes from automatic upgrades or repairsº, and in fact some
cloud providers explicitly mention this as a best practice.
(Basically most of the benefits of k8s are lost for apps managing
storage at App level. This is the case with most DDBBs and stream
architectures).
Running storage services
(GlusterFS,iSCSI,...)
@[https://opensource.com/article/17/12/storage-services-kubernetes]
Rook.io
@[https://rook.io/]
Rook turns distributed storage systems into self-managing,
self-scaling, self-healing storage services. It automates the tasks
of a storage administrator: deployment, bootstrapping, configuration,
provisioning, scaling, upgrading, migration, disaster recovery,
monitoring, and resource management.
Rook uses the power of the Kubernetes platform to deliver its
services via a Kubernetes Operator for each storage provider.
https://www.infoq.com/news/2019/08/rook-v1-release/
Rook, a storage orchestrator for Kubernetes, has released version 1.0
for production-ready workloads that use file, block, and object
storage in containers. Highlights of Rook 1.0 include support for
storage providers through operators like Ceph Nautilus, EdgeFS, and
NFS. For instance, when a pod requests an NFS file system, Rook can
provision it without any manual intervention.
Rook was the first storage project accepted into the Cloud Native
Computing Foundation (CNCF), and it helps storage administrators to
automate everyday tasks like provisioning, configuration, disaster
recovery, deployment, and upgrading storage providers. Rook turns a
distributed file system into storage services that scale and heal
automatically by leveraging the Kubernetes features with the operator
pattern. When administrators use Rook with a storage provider like
Ceph, they only have to worry about declaring the desired state of
the cluster and the operator will be responsible for setting up and
configuring the storage layer in the cluster.
Ceph
@[https://ceph.io/]
Noobaa
https://www.noobaa.io/try
NooBaa can collapse multiple storage silos into a single, scalable
storage fabric, by its ability to virtualize any local storage,
whether shared or dedicated, physical or virtual and include both
private and public cloud storage, using the same S3 API and
management tools. NooBaa also gives you full control over data
placement, letting you place data based on security, strategy and
cost considerations, in the granularity of an application.
- Easily scales locally on top of PVs or Ceph external clusters
- Workload Portability
Easily mirror data to other cluster or native cloud storage
Networking
Networking Explained
@[https://www.slideshare.net/CJCullen/kubernetes-networking-55835829]
SDNs
SDN intro:
@[https://www.redhat.com/sysadmin/getting-started-sdn]
- provide sdns: allows admin teams to control network traffic in
complex networking topologies through a centralized panel,
rather than handling each network device, such as routers
and switches, manually ("hierarchical" topology)
ºcontainer network interface (cni):º
- library definition and a set of tools to configure network
interfaces in linux containers through many supported plugins.
- multiple plugins can run at the same time in a container
that participates in a network driven by different plugins.
- networks use json configuration files and instantiated as
new namespaces when the cni plugin is invoked.
- common cni plugins include:
-ºcalicoº:
- high scalability.
@[https://docs.projectcalico.org/v3.7/getting-started/kubernetes/]
@[https://kubernetes.io/docs/tasks/administer-cluster/network-policy-provider/calico-network-policy/]
-ºciliumº:
- provides network connectivity and load balancing
between application workloads, such as application
containers and processes, and ensures transparent security.
-ºcontivº:
- integrates containers, virtualization, and physical servers
based on the container network using a single networking fabric.
-ºcontrailº:
- provides overlay networking for multi-cloud and
hybrid cloud through network policy enforcement.
-ºflannelº:
- makes it easier for developers to configure a layer 3
network fabric for kubernetes.
-ºmultusº:
- supports multiple network interfaces in a single pod on
kubernetes for sriov, sriov-dpdk, ovs-dpdk, and vpp workloads.
-ºopen vswitch (ovs)º:
- production-grade cni platform with a standard management
interface on openshift and openstack.
-ºovn-kubernetesº:
- enables virtual networks for multiple containers on different
hosts using an overlay function.
-ºromanaº:
- makes cloud network functions less expensive to build,
easier to operate, and better performing than traditional
cloud networks.
- in addition to network namespaces, an sdn should increase security by
offering isolation between multiple namespaces with the multi-tenant plugin:
packets from one namespace, by default, will not be visible to
other namespaces, so containers from different namespaces cannot
send packets to or receive packets from pods and services of a
different namespace.
network policy providers
use cilium for networkpolicy
use kube-router for networkpolicy
romana for networkpolicy
weave net for networkpolicy
access clusters using the kubernetes api
access services running on clusters
advertise extended resources for a node
autoscale the dns service in a cluster
change the reclaim policy of a persistentvolume
change the default storageclass
cluster management
configure multiple schedulers
configure out of resource handling
configure quotas for api objects
control cpu management policies on the node
customizing dns service
debugging dns resolution
declare network policy
developing cloud controller manager
encrypting secret data at rest
guaranteed scheduling for critical add-on pods
ip masquerade agent user guide
kubernetes cloud controller manager
limit storage consumption
operating etcd clusters for kubernetes
reconfigure a node's kubelet in a live cluster
reserve compute resources for system daemons
safely drain a node while respecting application slos
securing a cluster
set kubelet parameters via a config file
set up high-availability kubernetes masters
static pods
storage object in use protection
using coredns for service discovery
using a kms provider for data encryption
using sysctls in a kubernetes cluster
service mesh
istio.io
linkerd
Controllers
ReplicaSet
@[https://kubernetes.io/docs/concepts/workloads/controllers/replicaset/]
- Target: ensure "N" pod replicas are running simultaneously.
- Most of the times used indirectly by "Deployments" to orchestrate pod creation/deletion/updates.
BºNOTE:º Job controller prefered for pods that terminate on their own.
BºNOTE:º DaemonSet controller prefered for pods providing a machine-level function.
(monitoring,logging, pods that need to be running before others pods starts).
OºDaemonSet pods lifetime == machine lifetimeº
|controllers/frontend.yaml
|
|apiVersion: apps/v1
|kind:ºReplicaSetº
|metadata:
| name: frontend
| labels:
| app: guestbook
| tier: frontend
|spec:
| # modify replicas according to your case
| ºreplicas: 3º ← Default to 1
|Oºselector:º
|Oº matchLabels:º
|Oº tier: frontendº
|Oº matchExpressions:º
|Oº - {key: tier, operator: In, values: [frontend]}º
| template: ← ················· Pod template (nested pod schema, removing
| metadata: apiVersion/kind properties -)
| labels: ←····· Needed in pod-template (vs isolated pod)
| app: guestbook .spec.template.metadata.labels must match
| Oºtier: frontendº .spec.selector
| spec:
| ºrestartPolicy: Alwaysº←··· (default/only allowed value)
| Qºcontainers:º
| - name: php-redis
| image: gcr.io/google_samples/gb-frontend:v3
| resources: ←············ "Best Pattern". Indicate resources.
| requests:
| cpu: 100m
| memory: 100Mi
| env:
| - name: GET_HOSTS_FROM
| value: dns ←·········· replace 'dns' by 'env' if DNS is not configured
| ports:
| - containerPort: 80
$º$ kubectl create -f http://.../frontend.yaml º
→ replicaset.apps/frontend created
$º$ kubectl describe rs/frontend º
→ Name: frontend
· Namespace: default
· Selector: tier=frontend,tier in (frontend)
· Labels: app=guestbook
· tier=frontend
· Annotations: ˂none˃
· Replicas: Bº3 current / 3 desiredº
· Pods Status: 3 Running / 0 Waiting / 0 Succeeded / 0 Failed
· Pod Template:
· OºLabels: app=guestbookº
· Oº tier=frontendº
· Containers:
· Qºphp-redis:º
· Image: gcr.io/google_samples/gb-frontend:v3
· Port: 80/TCP
· Requests:
· cpu: 100m
· memory: 100Mi
· Environment:
· GET_HOSTS_FROM: dns
· Mounts: ˂none˃
· Volumes: ˂none˃
· Events:
· FirstSeen LastSeen Count From ··· Type Reason Message
· --------- -------- ----- ---- -------- ------ -------
· 1m 1m 1 {replicaset-cont } Normal SuccessfulCreate Created pod: frontend-qhloh
· 1m 1m 1 {replicaset-cont } Normal SuccessfulCreate Created pod: frontend-dnjpy
· 1m 1m 1 {replicaset-cont } Normal SuccessfulCreate Created pod: frontend-9si5l
$º$ kubectl get pods º
→ NAME READY STATUS RESTARTS AGE
· frontend-9si5l 1/1 Running 0 1m
· frontend-dnjpy 1/1 Running 0 1m
· frontend-qhloh 1/1 Running 0 1m
Deployment
("Application")
@[https://kubernetes.io/docs/concepts/workloads/controllers/deployment/]
- Built "on top" of ReplicaSets adding lifecycle management (creation, updates,...).
For example, if a Pod definition is changed, Deployments takes care of gradually
moving from a running (old) ReplicaSet to a new (automatically created) ReplicaSet.
Deployments also rollback to (old) ReplicaSet if new one is not stable.
Old ReplicaSets are automatically removed once the updated one is detected to be
stable.
- Ussually a running applications maps to a Deployment (plus some network
ingress rules, to make pod exposed services visibles "outside" k8s)
- Ex. Deployment:
| # for versions before 1.7.0 use apps/v1beta1
| apiVersion: apps/v1beta2
| kind: Deployment
| metadata:
| name: nginx-deployment
| labels:
| app: nginx
| # namespace: production
| spec:
| replicas: 3 ← 3 replicated Pods
| strategy:
| - type : Recreate ← Recreate | RollingUpdate*
| # Alt. strategy example
| # strategy:
| # rollingUpdate:
| # maxSurge: 2
| # maxUnavailable: 0
| # type: RollingUpdate
| selector:
| matchLabels:
| app: nginx
| template: ← pod template
| metadata:
| labels:
| app: nginx
| spec: ← template pod spec
| containers: change triggers new rollout
| - name: nginx
| image: nginx:1.7.9
| ports:
| - containerPort: 80
| livenessProbe:
| httpGet:
| path: /heartbeat
| port: 80
| scheme: HTTP
$º$ kubectl create -f nginx-deployment.yaml º
$º$ kubectl get deployments º
→ NAME DESIRED CURRENT ...
· nginx-deployment 3 0
$º$ kubectl rollout status deployment/nginx-deployment º
→ Waiting for rollout to finish: 2 out of 3 new replicas
· have been updated...
· deployment "nginx-deployment" successfully rolled out
$º$ kubectl get deployments º
→ NAME DESIRED CURRENT ...
· nginx-deployment 3 3
ºTo see the ReplicaSet (rs) created by the deployment:º
$ kubectl get rs
NAME DESIRED ...
nginx-deployment-...4211 3
└─────────┬────────────┘
Format: [deployment-name]-[pod-template-hash-value]
$º$ kubectl get pods --show-labels º ← Display ALL labels automatically
→ NAME ... LABELS generated for ALL pods
· nginx-..7ci7o ... app=nginx,...,
· nginx-..kzszj ... app=nginx,...,
· nginx-..qqcnn ... app=nginx,...,
- Ex: Update nginx Pods from nginx:1.7.9 to nginx:1.9.1:
$º$ kubectl set image deployment/nginx-deployment \ º
$º nginx=nginx:1.9.1 º
ºCheck the revisions of deployment:º
$ kubectl rollout history deployment/nginx-deployment
deployments "nginx-deployment"
R CHANGE-CAUSE
1 kubectl create -f nginx-deployment.yaml ---record
2 kubectl set image deployment/nginx-deployment \
nginx=nginx:1.9.1
3 kubectl set image deployment/nginx-deployment \
nginx=nginx:1.91
$ kubectl rollout undo deployment/nginx-deployment \
--to-revision=2
ºScale Deployment:º
$ kubectl scale deployment \
nginx-deployment --replicas=10
$ kubectl autoscale deployment nginx-deployment \
--min=10 --max=15 --cpu-percent=80
StatefulSet (v:1.9+)
@[https://kubernetes.io/docs/concepts/workloads/controllers/statefulset/]
- naming convention, network names, and storage persist as replicas are
rescheduled.
- underlying persistent storage remains even when the StatefulSet is deleted.
- Pods in StatefulSet are scheduled and run across any available node in an
AKS cluster (vs DaemonSet pods, attached to a given node).
- Manages stateful apps:
- Useful for apps requiring one+ of:
{
- Stable, unique network identifiers.
- persistent storage across Pod (re)scheduling
- Ordered, graceful deployment and scaling.
- Ordered, graceful deletion and termination.
- Ordered, automated rolling updates.
}
- Manages the deploy+scaling of Pods providing
guarantees about ordering and uniqueness
- Unlike Deployments, a StatefulSet maintains a sticky identity for each
of their Pods. These pods are created from the same spec, but are not
interchangeable: each has a persistent identifier that it maintains
across any rescheduling
- Pod Identity: StatefulSet Pods have a unique identity that is comprised
of [ordinal, stable network identity, stable storage] that sticks even
if Pods is rescheduled on another node.
- Ordinal:Each Pod will be assigned an unique integer ordinal,
from 0 up through N-1, where N = number of replicas
- Stable Network Pod host-name = $(statefulset name)-$(ordinal)
Ex. full DNS using Stateless service:
Pod full DNS (web == StatefullSet.name)
Oº← pod-host →º Bº← service ns →º Qº←clusterDoma→º
Oºweb-{0..N-1}º.Bºnginx.default.svcº.Qºcluster.localº
Oºweb-{0..N-1}º.Bºnginx.foo .svcº.Qºcluster.localº
Oºweb-{0..N-1}º.Bºnginx.foo .svcº.Qºkube.local º
*1: Cluster Domain defaults to cluster.local
- Pod Name Label: When the controller creates a Pod,
it adds a label statefulset.kubernetes.io/"pod-name" set to
the name of the pod, allowing to attach a Service to an unique Pod
ºLimitationsº
- The storage for a given Pod must either be provisioned by a
PersistentVolume Provisioner based on the requested storage class, or
pre-provisioned by an admin.
- Deleting and/or scaling a StatefulSet down will not delete the
volumes associated with the StatefulSet in order to ensure data safety,
which is generally more valuable than an automatic purge of all related
StatefulSet resources.
- StatefulSets currently require a Headless Service to be
responsible for the network identity of the Pods. You are responsible for
creating this Service.
ºExample StatefulSetº
The Bºheadless Service (named nginx)º, is used to control the network domain
The OºStatefulSet(named web)º, has a Spec that indicates
that 3 replicas of the nginx container will be launched in unique Pods.
The GºvolumeClaimTemplatesº will provide stable storage
using PersistentVolumes provisioned by a PersistentVolume Provisioner.
apiVersion: v1
kind: Service
metadata:
name: nginx
labels:
app: nginx
spec:
ports:
- port: 80
name: web
clusterIP: None
selector:
app: nginx
---
apiVersion: apps/v1
kind: StatefulSet
metadata:
name: web
spec:
replicas: 3 # by default is 1
selector:
matchLabels:
Qºapp: nginx # has to match .spec.template.metadata.labelsº
serviceName: "nginx"
template:
metadata:
labels:
app: nginx # has to match .spec.selector.matchLabels
spec:
terminationGracePeriodSeconds: 10
containers:
- name: nginx
image: k8s.gcr.io/nginx-slim:0.8
ports:
- containerPort: 80
name: web
volumeMounts:
- name: www
mountPath: /usr/share/nginx/html
ºvolumeClaimTemplatesº:
# Kubernetes creates one PersistentVolume for each VolumeClaimTemplate
- metadata:
name: www
spec:
accessModes: [ "ReadWriteOnce" ]
# each Pod will receive a single PersistentVolume
# When a Pod is (re)scheduled onto a node, its volumeMounts mount
# the PersistentVolumes associated with its PersistentVolume Claims.
storageClassName: "my-storage-class"
resources:
requests:
storage: 1Gi
DONT'S
- The StatefulSet should NOT specify a pod.Spec.TerminationGracePeriodSeconds of 0.
Unsafe and ºstrongly discouragedº
Deployment and Scaling Guarantees
- Pods are deployed sequentially in order from {0..N-1}.
- When Pods are deleted they are terminated in reverse order, from {N-1..0}
- Before a scaling operation is applied to a Pod, all of its predecessors must be Running and Ready
- Before a Pod is terminated, all of its successors must be completely shutdown
-ºOrdering Policies guarantees can be relaxed via .spec.podManagementPolicy(K8s 1.7+)º
- OrderedReady: Defaults, implements the behavior described above
- Parallel Pod Management: launch/terminate all Pods in parallel, not
waiting for Pods to become Running and Ready or completely terminated
-
-º.spec.updateStrategyº allows to configure and disable
automated rolling updates for containers, labels, resource request/limits,
and annotations for the Pods in a StatefulSet.
- "OnDelete" implements the legacy (1.6 and prior) behavior.
StatefulSet controller will not automatically update the Pods in a StatefulSet.
Users must manually delete Pods to cause the controller to create new Pods
that reflect modifications made to a StatefulSet’s .spec.template.
- "RollingUpdate" (default 1.7+) implements automated, rolling update for Pods.
The StatefulSet controller will delete and recreate each Pod proceeding
in the same order as Pod termination (largest to smallest ordinal), updating
each Pod one at a time waiting until an updated Pod is Running and Ready
prior to updating its predecessor.
Partitions
{
- RollingUpdate strategy can be partitioned, by specifying a
.spec.updateStrategy.rollingUpdate.partition
- If specified all Pods with an ordinal greater than or equal to
the partition will be updated when the StatefulSet’s .spec.template
is updated. All Pods with an ordinal that is less than the partition
will not be updated, and, even if they are deleted, they will be
recreated at the previous version.
- If it is greater than its .spec.replicas, updates to its
.spec.template will not be propagated to its Pods.
- In most cases you will not need to use a partition, but they
are useful if you want to stage an update, roll out a canary,
or perform a phased roll out.
}
}
Debug
@[https://kubernetes.io/docs/tasks/debug-application-cluster/debug-stateful-set/]
DaemonSet
- ensures "N" Nodes run a Pod instance
- typical uses: cluster storage, log collection or monitoring
- As nodes are added to the cluster, Pods are added to them. As nodes are
removed from the cluster, those Pods are garbage collected. Deleting a
DaemonSet will clean up the Pods it created.
- Ex (simple case): one DaemonSet, covering all nodes, would be used
for each type of daemon. A more complex setup might use multiple DaemonSets
for a single type of daemon, but with different flags and/or different memory
and cpu requests for different hardware types.
Ex. DaemonSet for Bºfluentd-elasticsearchº:
$ cat daemonset.yaml
apiVersion: apps/v1
kind: ºDaemonSetº
metadata:
name: fluentd-elasticsearch
namespace: kube-system
labels:
k8s-app: fluentd-logging
spec:
selector:
matchLabels:
name: fluentd-elasticsearch
template: # Pod template
# Pod Template must have RestartPolicy equal to Always (default if un-specified)
metadata:
labels:
name: fluentd-elasticsearch
spec:
tolerations:
- key: node-role.kubernetes.io/master
effect: NoSchedule
containers:
- name: fluentd-elasticsearch
image: Bºk8s.gcr.io/fluentd-elasticsearch:1.20º
resources:
limits:
memory: 200Mi
requests:
cpu: 100m
memory: 200Mi
volumeMounts:
- name: varlog
mountPath: /var/log
- name: varlibdockercontainers
mountPath: /var/lib/docker/containers
readOnly: true
terminationGracePeriodSeconds: 30
volumes:
- name: varlog
hostPath:
path: /var/log
- name: varlibdockercontainers
hostPath:
path: /var/lib/docker/containers
Garbage Collection
@[https://kubernetes.io/docs/concepts/workloads/controllers/garbage-collection/]
Jobs
@[https://kubernetes.io/docs/concepts/workloads/controllers/jobs-run-to-completion/]
Running Automated Tasks with a CronJob
Parallel Processing using Expansions
Coarse Parallel Processing Using a Work Queue
Fine Parallel Processing Using a Work Queue
- One example of Job pattern would be a Job which starts a Pod which
runs a script that in turn starts a Spark master controller (see
spark example), runs a spark driver, and then cleans up.
- reliably run 1+ Pod/s to "N" completions
- creates one+ pods and ensures that a specified number of them successfully terminate
- Jobs are complementary to Deployment Controllers. A Deployment Controller
manages pods which are not expected to terminate (e.g. web servers), and
a Job manages pods that are expected to terminate (e.g. batch jobs).
- As pods successfully complete, the job tracks the successful completions. When a specified number of successful completions is reached, the job itself is complete. Deleting a Job will cleanup the pods it created.
- Pod Backoff failure policy: ifyou want to fail a Job after N retries set
.spec.backoffLimit (defaults to 6).
- Pods are not deleted on completion in order to allow view logs/output/errors for completed pods. They will show up with kubectl get pods º-aº.
Neither the job object in order to allow viewing its status.
- Another way to terminate a Job is by setting an active deadline
in .spec.activeDeadlineSeconds or .specs.template.specs.activeDeadlineSeconds
}
example. Compute 2000 digits of "pi"
$ cat job.yaml
apiVersion: batch/v1
kind: ºJobº
metadata:
name: pi
spec:
template: # Required (== Pod template - apiVersion - kind)
spec:
containers:
- name: pi
ºimage: perlº
ºcommand: ["perl", "-Mbignum=bpi", "-wle", "print bpi(2000)"]º
OºrestartPolicy: Neverº # Only Never/OnFailure allowed
backoffLimit: 4
# ºRun job using:º
# $ kubectl create -f ./job.yaml
# ºCheck job current status like:º
# $ kubectl describe jobs/pi
# output will be similar to:
# Name: pi
# Namespace: default
# Selector: controller-uid=b1db589a-2c8d-11e6-b324-0209dc45a495
# Labels: controller-uid=b1db589a-2c8d-11e6-b324-0209dc45a495
# job-name=pi
# Annotations: ˂none˃
# Parallelism: 1
# Completions: 1
# Start Time: Tue, 07 Jun 2016 10:56:16 +0200
# Pods Statuses: 0 Running / 1 Succeeded / 0 Failed
# Pod Template:
# Labels: controller-uid=b1db589a-2c8d-11e6-b324-0209dc45a495
# job-name=pi
# Containers:
# pi:
# Image: perl
# Port:
# Command:
# perl
# -Mbignum=bpi
# -wle
# print bpi(2000)
# Environment: ˂none˃
# Mounts: ˂none˃
# Volumes: ˂none˃
# Events:
# FirstSeen LastSeen Count From SubobjectPath Type Reason Message
# --------- -------- ----- ---- ------------- ------- ------ -------
# 1m 1m 1 {job-controller} Normal SuccessfulCreate Created pod: pi-dtn4q
#
# ºTo view completed pods of a job, use º
# $ kubectl get pods
# ºTo list all pods belonging to job in machine-readable-formº:
#
# $ pods=$(kubectl get pods --selector=ºjob-name=piº --output=ºjsonpath={.items..metadata.name}º)
# $ echo $pods
pi-aiw0a
# ºView the standard output of one of the pods:º
# $ kubectl logs $pods
# 3.1415926535897a....9
Parallel Jobs
- Parallel Jobs with a fixed completion count
(.spec.completions greater than zero).
the job is complete when there is one successful pod for
each value in the range 1 to .spec.completions.
- Parallel Jobs with a work queue: do not specify .spec.completions:
pods must coordinate with themselves or external service to determine
what each should work on.
each pod is independently capable of determining whether or not all its peers
are done, thus the entire Job is done.
- For Non-parallel job, leave both .spec.completions and
.spec.parallelism unset.
- Actual parallelism (number of pods running at any instant) may be more or less than requested parallelism, for a variety of reasons
- (read official K8s docs for Job Patterns ussages)
}
Cron Jobs (1.8+)
- written in Cron format (question mark (?) has the same meaning as an asterisk *)
- Concurrency Policy
- Allow (default): allows concurrently running jobs
- Forbid: forbids concurrent runs, skipping next
if previous still running
- Replace: cancels currently running job and replaces with new one
CronJob
- @[https://kubernetes.io/docs/concepts/workloads/controllers/cron-jobs/]
- Ex. cronjob:
$ cat cronjob.yaml
apiVersion: batch/v1beta1
kind: CronJob
metadata:
name: hello
spec:
schedule: Oº"*/1 º º º º"º
jobTemplate:
spec:
template:
spec:
containers:
Oº- name: hello º
Oº image: busybox º
Oº args: º
Oº - /bin/sh º
Oº - -c º
Oº - date; echo 'Hi from K8s'º
restartPolicy: OnFailure
# Alternatively:
$ kubectl run hello \
--schedule="*/1 º º º º" \
--restart=OnFailure \
--image=busybox \
-- /bin/sh -c "date; echo Hello from the Kubernetes cluster"
# get status:
$ kubectl get cronjob hello
NAME SCHEDULE SUSPEND ACTIVE LAST-SCHEDULE
hello */1 º º º º False 0
# Watch for the job to be created:
$ kubectl get jobs --watch
NAME DESIRED SUCCESSFUL AGE
hello-4111706356 1 1 2s
Monitoring
Node Health
@[https://kubernetes.io/docs/tasks/debug-application-cluster/monitor-node-health/]
crictl: Debug node
@[https://kubernetes.io/docs/tasks/debug-application-cluster/crictl/]
Monitoring the cluster
Auditing
@[https://kubernetes.io/docs/tasks/debug-application-cluster/audit/]
Events in Stackdriver
@[href=https://kubernetes.io/docs/tasks/debug-application-cluster/events-stackdriver/]
- Kubernetes events are objects that provide insight into what is
happening inside a cluster, such as what decisions were made by
scheduler or why some pods were evicted from the node.
- Since events are API objects, they are stored in the apiserver on
master. To avoid filling up master's disk, a retention policy is
enforced: events are removed one hour after the last occurrence. To
provide longer history and aggregation capabilities, a third party
solution should be installed to capture events.
- This article describes a solution that exports Kubernetes events to
Stackdriver Logging, where they can be processed and analyzed.
Metrics API+Pipeline
- Resource usage metrics, such as container CPU and memory usage, are
available in Kubernetes through the Metrics API. These metrics can be
either accessed directly by user, for example by using kubectl top
command, or used by a controller in the cluster, e.g. Horizontal Pod
Autoscaler, to make decisions.
@[https://kubernetes.io/docs/tasks/debug-application-cluster/resource-metrics-pipeline/]
@[https://github.com/kubernetes-incubator/metrics-server]
Extracted from @[https://kubernetes.io/docs/tasks/configure-pod-container/assign-memory-resource/]
"""...If running on Minikube, run the following command to enable the metrics-server:
$º$ minikube addons enable metrics-server º
... to see whether the metrics-server is running, or another provider of the resource metrics
API (metrics.k8s.io), run the following command:
$º$ kubectl get apiservices º
output must include a reference to metrics.k8s.io.
→ ...
→ v1beta1.metrics.k8s.io
"""
Auditing
Auditing
@[https://kubernetes.io/docs/tasks/debug-application-cluster/audit/]
- Kubernetes auditing provides a security-relevant chronological set of
records documenting the sequence of activities that have affected
system by individual users, administrators or other components of the
system. It allows cluster administrator to answer the following
questions:
- what happened?
- when did it happen?
- who initiated it?
- on what did it happen?
- where was it observed?
- from where was it initiated?
- to where was it going?
K8s Extension Tools
HELM Charts
@[https://helm.sh/docs/intro/quickstart/]
ºPRE-SETUPº
└ "kubectl" installed locally.
ºINITIAL-SETUPº
└ Download from @[https://github.com/helm/helm/releases], use homebrew (MacOSX),...
and add "helm" command to path.
└ Initialize aºHelm Chart Repositoryº
$º$ helm repo add stable \ º
$º https://kubernetes-charts.storage.googleapis.com/º ← Official repo.
ºDAILY USEº
$º$ helm get -h º
$º$ helm search repo stable º ← list charts available for installation.
$ºNAME CHART VERSION APP VERSION DESCRIPTION º
$ºstable/acs-engine-autoscaler 2.2.2 2.1.1 DEPRECATED Scales worker nodes within agent pools º
$ºstable/aerospike 0.2.8 v4.5.0.5 A Helm chart for Aerospike in Kubernetes º
$ºstable/airflow 4.1.0 1.10.4 Airflow is a platform to programmatically autho...º
$º... º
$º$ helm repo update º ← Make sure we get the latest list of charts
┌→$º$ helm install stable/mysql --generate-nameº ← Install MySQL
│ ^^^^^^
│ Helm has several ways to find and install a chart,
│ official stable charts is the easiests.
│ $º$ helm show chart stable/mysql º ← get an idea of the chart features
│ $º$ helm show all stable/mysql º ← get all information about chart
│
└ Whenever an install is done, a new local release is created
cluster, allowing to install multiple times into the same cluster.
Each local release can be independently managed and upgraded.
To list local released:
$º$ helm ls º
$ºNAME VERSION UPDATED STATUS CHART º
$ºsmiling-penguin 1 Wed Sep 28 12:59:46 2016 DEPLOYED mysql-0.1.0 º
To uninstall a release:
$º$ helm uninstall \ º
$º$ smiling-penguin --keep-history º ←·························┐
$ºRemoved smiling-penguin º |
$º$ helm status smiling-penguin º ← works only in using the --keep-history
$ºStatus: UNINSTALLED º flag during uninstall.
$... º It also allows to roolback a deleted
release using $º$ helm rollback ...º
TODO: @[https://helm.sh/docs/intro/using_helm/]
NOTE:@[https://www.infoworld.com/article/3541608/kubernetes-helm-gets-full-cncf-approval.html]
Kubeapps
- Application Dashboard for Kubernetes
@[https://kubeapps.com/]
- Deploy Apps Internally:
Browse Helm charts from public or your own private chart repositories
and deploy them into your cluster.
- Manage Apps:
Upgrade, manage and delete the applications that are deployed in your
Kubernetes cluster.
- Service Catalog:
Browse and provision external services from the Service Catalog and
available Service Brokers.
Teresa
@[https://github.com/luizalabs/teresa]
- extremely simple platform as a service that runs on top
of Kubernetes. It uses a client-server model: the client sends high
level commands (create application, deploy, etc.) to the server,
which translates them to the Kubernetes API.
Gravity Portable clusters
"take a cluster as is and deploy it somewhere else"
- Gravity takes snapshots of Kubernetes clusters, their container
registries, and their running applications, called "application
bundles." The bundle, which is just a .tar file, can replicate the
cluster anywhere Kubernetes runs.
- Gravity also ensures that the target infrastructure can support the
same behavioral requirements as the source, and that the Kubernetes
runtime on the target is up to snuff. The enterprise version of
Gravity adds security features including role-based access controls
and the ability to synchronize security configurations across
multiple cluster deployments.
- latest major version, Gravity 7, can deploy a Gravity image into
an existing Kubernetes cluster, versus spinning up an all-new cluster
using the image. Gravity 7 can also deploy into clusters that
aren’t already running a Gravity-defined image. Plus, Gravity now
supports SELinux and integrates natively with the Teleport SSH
gateway.
Kaniko, Container Build
- Kaniko performs container builds inside a container environment,
Bºwithout relying on a container daemon (Docker Daemon)º.
- Kaniko extracts the file system from the base image,
executes the build commands in user space atop the extracted
file system, taking a snapshot of the file system after each command.
-
KubeDB: Production DBs
- operators exists for MySQL/PostgreSQL/Redis/... Rºbut there are plenty of gapsº.
BºKubeDB allows to create custom Kubernetes operators for managing databasesº
- Running backups, cloning, monitoring, snapshotting, and declaratively
creating databases are all part of the mix.
Rºsupported features vary among databasesº. Ex:
clustering is available for PostgreSQL but not MySQL.
Kube-monkey: Chaos testing
- stress test, breaking stuff at random.
- It works by randomly killing pods in a cluster
that you specifically designate, and can be fine-tuned
to operate within specific time windows.
Ingress Controller for AWS
- Allows to reuse AWS load balancing functionality.
- It uses AWS CloudFormation to ensure that cluster state
remains consistent.
Governance
Gatekeeper policy controls
The Open Policy Agent project (OPA) provides a way to create policies
across cloud-native application stacks, from ingress to service-mesh
components to Kubernetes. Gatekeeper provides a Kubernetes-native way
to enforce OPA policies on a cluster automatically, and to audit for
any events or resources violating policy. All this is handled by a
relatively new mechanism in Kubernetes, admission controller
Webhooks, that fire on changes to resources. With Gatekeeper, OPA
policies can be maintained as just another part of your Kubernetes
cluster’s defined state, without needing constant babysitting.
Teleport
- implement industry-best practices for SSH and
Kubernetes access, meet compliance requirements, and have complete
visibility into access and behavior.
· Security best practices out-of-the-box.
· Isolate critical infra and enforce 2FA with SSH and Kubernetes.
· Provide role-based access controls (RBAC) using short-lived
certificates and your existing identity management service.
· Log events and record session activity for full auditability.
Kubecost: Cost metrics running k8s
- monitor "the dollars" cost of running Kubernetes?
- Kubecost uses real-time Kubernetes metrics, and real-world cost
information derived from running clusters on the major cloud
providers, to provide a dashboard view of the monthly cost of each
cluster deployment. Costs for memory, CPU, GPU, and storage are all
broken out by Kubernetes component (container, pod, service,
deployment, etc.).
- Kubecost can also track the costs of “out of cluster” resources,
such as Amazon S3 buckets, although this is currently limited to AWS.
- Kubecost is free to use if you only need to keep 15 days of logs. For
more advanced features, pricing starts at $199 per month for
monitoring 50 nodes.
cluster Administration
Bootstrap/Manage k8s clusters
Kubespray: Cluster Bootstrap
@[https://github.com/kubernetes-sigs/kubespray/blob/master/docs/comparisons.md]
- Kubespray is formed by a set of Ansible playbooks (optionally Vagrant) allowing
to create a new production-ready k8s cluster (mono/multi-master,
single/distributed etcd, Flannel|Calico|Weave|... network...).
Extra playbooks allows to add/remove new nodes to a running k8s cluster.
- It targets "mostly" any environment, from bare metal (RedHat, CoreOS, Ubuntu,...)
to public clouds.
- Compared to "Kops", "Kops" is more tightly integrated and profits better from
the unique features of the supported clouds (AWS,GCE, ...). "Kops" doesn't
allow to deploy on standard linux distros running on bare-metal/VMs.
KOPS: Deploy on Cloud
@[https://github.com/kubernetes/kops/blob/master/README.md]
@[https://github.com/kubernetes/kops/blob/master/README.md]
- Kops cli-tool allows the automation of production-grade k8s cluster
(creation/destruction, upgrade, maintenance).
- AWS is currently officially supported, with GCE in beta support,
and VMware vSphere in alpha, and other platforms planned.
- It does NOT support standard linux distros on bare-metal/VMs.
Kubespray is prefered in this case.
Kubeadm
Overview of kubeadm
kubeadm init
kubeadm join
kubeadm upgrade
kubeadm config
kubeadm reset
kubeadm token
kubeadm version
kubeadm alpha
Implementation details
Note: Kubespray offers a "simpler" front-end based on Ansible playbooks to Kubeadm.
Example Tasks:
- Upgrading kubeadm HA clusters from 1.9.x to 1.9.y
@[https://kubernetes.io/docs/tasks/administer-cluster/kubeadm/kubeadm-upgrade-ha/]
- Upgrading/downgrading kubeadm clusters between v1.8 to v1.9:
@[https://kubernetes.io/docs/tasks/administer-cluster/kubeadm/kubeadm-upgrade-1-9/]
- Upgrading kubeadm clusters from 1.7 to 1.8
@[https://kubernetes.io/docs/tasks/administer-cluster/kubeadm/kubeadm-upgrade-1-8/]
- Upgrading kubeadm clusters from v1.10 to v1.11:
@[https://kubernetes.io/docs/tasks/administer-cluster/kubeadm/kubeadm-upgrade-1-11/]
WKSctl GitOps install
- Tool for Kubernetes Cluster Management Using BºGitOpsº.
WKSctl is an open-source project to install, bootstrap, and manage Kubernetes
clusters, including add-ons, through SSH. WKS is a provider of the Cluster API
(CAPI) using the GitOps approach. Kubernetes cluster configuration is defined
in YAML, and WKSctl applies the updates after every push in Git, allowing users
to have repeatable clusters on-demand.
- CAPI (@[https://cluster-api.sigs.k8s.io/] k8s sub-project is focused on providing
declarative APIs and tooling to simplify provisioning, upgrading, and operating
multiple Kubernetes clusters.
Other Admin Tasks
TLS Cert Mng:
- Managing TLS Certs:
@[https://kubernetes.io/docs/tasks/tls/managing-tls-in-a-cluster/]
@[https://kubernetes.io/docs/tasks/tls/certificate-rotation/]
- kubelet TLS setup:
@[https://kubernetes.io/docs/reference/command-line-tools-reference/kubelet-authentication-authorization/]
@[https://kubernetes.io/docs/reference/command-line-tools-reference/kubelet-tls-bootstrapping/]
private Reg.
- using a private (image) registry:
@[https://kubernetes.io/docs/concepts/containers/images/#using-a-private-registry]
- Pull Image from a Private Registry:
@[https://kubernetes.io/docs/tasks/configure-pod-container/pull-image-private-registry/]
- imagePullSecrets:
@[https://kubernetes.io/docs/concepts/configuration/secret/]
- method to pass a secret that contains a Docker image registry password
to the Kubelet so it can pull a private image on behalf of your Pod.
Mng.Cluster DaemonSets
- Perform a Rollback on a DaemonSet:
@[https://kubernetes.io/docs/tasks/manage-daemon/rollback-daemon-set/]
- Perform a Rolling Update on a DaemonSet:
@[https://kubernetes.io/docs/tasks/manage-daemon/update-daemon-set/]
Install Service Catalog
- Install Service Catalog using Helm:
@[https://kubernetes.io/docs/tasks/service-catalog/install-service-catalog-using-helm/]
- Install Service Catalog using SC:
@[https://kubernetes.io/docs/tasks/service-catalog/install-service-catalog-using-sc/]
Max Node Lattency
@[https://stackoverflow.com/questions/46891273/kubernetes-what-is-the-maximum-distance-latency-supported-between-kubernetes-no]
"""....There is no latency limitation between nodes in kubernetes cluster. They are configurable parameters.
└ For kubelet on worker node:
--node-status-update-frequency duration Specifies how often kubelet posts node status to master.
Note: be cautious when changing the constant, it must work
with nodeMonitorGracePeriod in nodecontroller. (default 10s)
└ For controller-manager on master node:
--node-monitor-grace-period duration Amount of time which we allow running Node to be unresponsive
before marking it unhealthy. Must be N times more than kubelet's
nodeStatusUpdateFrequency, where N means number of retries allowed
for kubelet to post node status. (default 40s)
--node-monitor-period duration The period for syncing NodeStatus in NodeController. (default 5s)
--node-startup-grace-period duration Amount of time which we allow starting Node to be unresponsive before
marking it unhealthy. (default 1m0s)
"""
Cluster Admin (Cont)
Admin Tasks
Config Ref()
@[https://kubernetes.io/docs/concepts/cluster-administration/cluster-administration-overview/]
@[https://kubernetes.io/docs/concepts/cluster-administration/certificates/]
@[https://kubernetes.io/docs/concepts/cluster-administration/cloud-providers/]
@[https://kubernetes.io/docs/concepts/cluster-administration/manage-deployment/]
@[https://kubernetes.io/docs/concepts/cluster-administration/networking/]
@[https://kubernetes.io/docs/concepts/cluster-administration/logging/]
@[https://kubernetes.io/docs/concepts/cluster-administration/kubelet-garbage-collection/]
@[https://kubernetes.io/docs/concepts/cluster-administration/proxies/]
@[https://kubernetes.io/docs/concepts/cluster-administration/controller-metrics/]
@[https://kubernetes.io/docs/concepts/cluster-administration/addons/]
Config. files
(documented in the Reference section of the online documentation, under each binary:)
kubelet
kube-apiserver
kube-controller-manager
kube-scheduler.
Compute, Storage, and Networking Extensions
Network Plugins
Device Plugins
Service Catalog
kubecfg
k8s as code
@[https://github.com/ksonnet/kubecfg]
@[https://www.youtube.com/watch?v=zpgp3yCmXok] Writing less yaml
A tool for managing Kubernetes resources as code.
kubecfg allows you to express the patterns across your infrastructure and reuse
these powerful "templates" across many services, and then manage those templates
as files in version control. The more complex your infrastructure is, the more
you will gain from using kubecfg.
The idea is to describe as much as possible about your configuration as files
in version control (eg: git).
Understanding K8S Code
K8s Implementation Summary
- Julia Evans "A few things I've learned about Kubernetes"
- """... you can run the kubelet by itself! And if you have a kubelet, you
can add the API server and just run those two things by themselves! Okay,
awesome, now let’s add the scheduler!"""
- the “kubelet” is in charge of running containers on nodes
- If you tell the API server to run a container on a node, it will tell the kubelet to get it done (indirectly)
- The scheduler translates "run a container" to "run a container on node X"
ºetcd is Kubernetes’ brainº
- Every component in Kubernetes (API server, scheduler, kubelets, controller manager, ...) is stateless.
All of the state is stored in the (key-value store) etcd database.
- Communication between components (often) happens via etcd.
-Oºbasically everything in Kubernetes works by watching etcd for stuff it has to do,º
Oºdoing it, and then writing the new state back into etcd º
Ex 1: Run a container on Machine "X":
Wrong way: ask kubelet@Machine"X" to run the container.
Right way: kubectl*1 →(API Server)→ etcd: "This pod should run on Machine X"
kubelet@Machine"X" → etcd: check work to do
kubelet@Machine"X" ← etcd: "This pod should run on Machine X"
kubelet@Machine"X" ← kubelet@Machine"X": Run pod
Ex 2: Run a container anywhere on the k8s cluster
kubectl*1 → (API Server) → etcd: "This pod should run somewhere"
scheduler → etcd: Something to run?
scheduler ← etcd: "This pod should run somewhere"
scheduler → kuberlet@Machine"Y": Run pod
*1 The kubectl is used from the command line.
In the sequence diagram it can be replaced for any
of the existing controllers (ReplicaSet, Deployment, DaemonSet, Job,...)
ºAPI server roles in cluster:º
API Server is responsible for:
1.- putting stuff into etcd
kubectl → API Server : put "stuff" in etcd
API Server → API Server : check "stuff"
alt 1:
kubectl ← API Server : error: "stuff" is wrong
alt 2:
API Server → etcd : set/update "stuff"
2.- Managing authentication:
("who is allowed to put what stuff into etcd")
The normal way is through X509 client certs.
ºcontroller manager does a bunch of stuffº
Responsible for:
- Inspect etcd for pending to schedule pods.
- daemon-set-controllers will inspect etcd for
pending daemonsets and will call the scheduler
to run them on every machine with the given
pod configuration.
- The "replica set controller" will inspect etcd for
pending replicasets and will create 5 pods that
the scheduler will then schedule.
- "deployment controller" ...
ºTroubleshooting:º
something isn’t working? figure out which controller is
responsible and look at its logs
ºCore K8s components run inside of k8sº
- Only 5 things needs to be running before k8s starts up:
- the scheduler
- the API server
- etcd
- kubelets on every node (to actually execute containers)
- the controller manager (because to set up daemonsets you
need the controller manager)
Any other core system (DNS, overlay network,... ) can
be scheduled by k8s inside k8s
API Conventions
@[https://github.com/kubernetes/community/blob/master/contributors/devel/sig-architecture/api-conventions.md]
Source Code Layout
Note: Main k8s are placed in kubernetes/pkg/
(API, kubectl, kubelet, controller, ...)
- REF: A Tour of the Kubernetes Source Code Part One: From kubectl to API Server
ºExamining kubectl sourceº
Locating the implementation of kubectl commands in the Kubernetes source code
- kubectl entry point for all commands
- Inside there is a name of a go directory that matches the kubectl command:
Example:
kubectl create/create.go
ºK8s loves the Cobra Command Frameworkº
- k8s commands are implemented using the Cobra command framework.
- Cobra provides lot of features for building command line interfaces
amongst them, Cobra puts the command usage message and command
descriptions adjacent to the code that runs the command.
Ex.:
| // NewCmdCreate returns new initialized instance of create sub command
| func NewCmdCreate(f cmdutil.Factory, ioStreams genericclioptions.IOStreams) *cobra.Command {
| o := NewCreateOptions(ioStreams)
|
| cmd := ⅋cobra.Command{
| Use: "create -f FILENAME",
| DisableFlagsInUseLine: true,
| Short: i18n.T("Create a resource from a file or from stdin."),
| Long: createLong,
| Example: createExample,
| Run: func(cmd *cobra.Command, args []string) {
| if cmdutil.IsFilenameSliceEmpty(o.FilenameOptions.Filenames) {
| defaultRunFunc := cmdutil.DefaultSubCommandRun(ioStreams.ErrOut)
| defaultRunFunc(cmd, args)
| return
| }
| cmdutil.CheckErr(o.Complete(f, cmd))
| cmdutil.CheckErr(o.ValidateArgs(cmd, args))
| cmdutil.CheckErr(o.RunCreate(f, cmd))
| },
| }
|
| // bind flag structs
| o.RecordFlags.AddFlags(cmd)
|
| usage := "to use to create the resource"
| cmdutil.AddFilenameOptionFlags(cmd, ⅋o.FilenameOptions, usage)
| ...
| o.PrintFlags.AddFlags(cmd)
|
| // create subcommands
| cmd.AddCommand(NewCmdCreateNamespace(f, ioStreams))
| ...
| return cmd
| }
ºBuilders and Visitors Abound in Kubernetesº
Ex. code:
| r := f.NewBuilder().
| Unstructured().
| Schema(schema).
| ContinueOnError().
| NamespaceParam(cmdNamespace).DefaultNamespace().
| FilenameParam(enforceNamespace, ⅋o.FilenameOptions).
| LabelSelectorParam(o.Selector).
| Flatten().
| Do()
The functions Unstructured, Schema, ContinueOnError,... Flatten
all take in a pointer to a Builder struct, perform some form of
modification on the Builder struct, and then return the pointer to
the Builder struct for the next method in the chain to use when it
performs its modifications defined at:
https://github.com/kubernetes/cli-runtime/blob/master/pkg/genericclioptions/resource/builder.go:
| ...
| func (b ºBuilder) Schema(schema validation.Schema) ºBuilder {
| b.schema = schema
| return b
| }
| ...
| func (b ºBuilder) ContinueOnError() ºBuilder {
| b.continueOnError = true
| return b
| }
The Do function finally returns a Result object that will be used to drive
the creation of our resource. It also creates a Visitor object that can be
used to traverse the list of resources that were associated with this
invocation of resource.NewBuilder. The Do function implementation is shown below.
a new DecoratedVisitor is created and stored as part of the Result object
that is returned by the Builder Do function. The DecoratedVisitor has a Visit
function that will call the Visitor function that is passed into it.
|// Visit implements Visitor
|func (v DecoratedVisitor) Visit(fn VisitorFunc) error {
| return v.visitor.Visit(func(info *Info, err error) error {
| if err != nil {
| return err
| }
| for i := range v.decorators {
| if err := v.decorators[i](info, nil); err != nil {
| return err
| }
| }
| return fn(info, nil)
| })
|}
Create eventually will call the anonymous function that contains the
createAndRefresh function that will lead to the code making a REST call
to the API server.
The createAndRefresh function invokes the Resource NewHelper
function found in ...helper.go returning a new Helper object:
| func NewHelper(client RESTClient, mapping ºmeta.RESTMapping) ºHelper {
| return ⅋Helper{
| Resource: mapping.Resource,
| RESTClient: client,
| Versioner: mapping.MetadataAccessor,
| NamespaceScoped: mapping.Scope.Name() == meta.RESTScopeNameNamespace,
| }
| }
Finally the Create function iwill invoke a createResource function of the
Helper Create function.
The Helper createResource function, will performs the actual REST call to
the API server to create the resource we defined in our YAML file.
ºCompiling and Running Kubernetesº
- we are going to use a special option that informs the Kubernetes build process
$ make WHAT='cmd/kubectl' # ← compile only kubectl
Test it like:
On terminal 1 boot up local test hack cluster:
$ PATH=$PATH KUBERNETES_PROVIDER=local hack/local-up-cluster.sh
On terminal 2 execute the compiled kubectl:
$ cluster/kubectl.sh create -f nginx_replica_pod.yaml
ºCode Learning Toolsº
Tools and techniques that can really help accelerate your ability to learn the k8s src:
- Chrome Sourcegraph Plugin:
provides several advanced IDE features that make it dramatically
easier to understand Kubernetes Go code when browsing GitHub repositories.
Ussage:
- start by looking at an absolutely depressing snippet of code,
with ton of functions.
- Hover over each code function with Chrome browser + Sourcegraph extension
installed:
It will popup a description of the function, what is passed into it
and what it returns.
- It also provides advanced view with the ability to peek into the function
being invoked.
- Properly formatted print statements:
fmt.Println("\n createAndRefresh Info = %#v", info)
- Use of a go panic to get desperately needed stack traces:
| func createAndRefresh(info *resource.Info) error {
| fmt.Println("\n createAndRefresh Info = %#v", info)
| ºpanic("Want Stack Trace")º
| obj, err := resource.NewHelper(info.Client, info.Mapping).Create(info.Namespace, true, info.Object)
| if err != nil {
| return err
| }
| info.Refresh(obj, true)
| return nil
| }
- GitHub Blame to travel back in time:
"What was the person thinking when they committed those lines of code?"
- GitHub browser interface has a blame option available as a button on the user interface:
It returns a view of the code that has the commits responsible for each line of code
in the source file. This allows you to go back in time and look at the commit that added
a particular line of code and determine what the developer was trying to accomplish when
that line of code was added.
Extending k8s API
- Custom Resources:
@[https://kubernetes.io/docs/concepts/extend-kubernetes/api-extension/custom-resources/]
- Extend API with CustomResourceDefinitions:
@[https://kubernetes.io/docs/tasks/access-kubernetes-api/custom-resources/custom-resource-definitions/]
- Versions of CustomResourceDefinitions:
@[https://kubernetes.io/docs/tasks/access-kubernetes-api/custom-resources/custom-resource-definition-versioning/]
- Migrate a ThirdPartyResource to CustomResourceDefinition:
@[https://kubernetes.io/docs/tasks/access-kubernetes-api/custom-resources/migrate-third-party-resource/]
- Aggregation Layer:
@[https://kubernetes.io/docs/concepts/extend-kubernetes/api-extension/apiserver-aggregation/]
- Configure the Aggregation Layer:
@[https://kubernetes.io/docs/tasks/access-kubernetes-api/configure-aggregation-layer/]
- Setup an Extension API Server:
@[https://kubernetes.io/docs/tasks/access-kubernetes-api/setup-extension-api-server/">
- Use HTTP Proxy to Access the k8s API:
@[https://kubernetes.io/docs/tasks/access-kubernetes-api/http-proxy-access-api/]
Unordered notes
GPU and Kubernetes
@[https://www.infoq.com/news/2018/01/gpu-workflows-kubernetes]
@[https://kubernetes.io/docs/tasks/manage-gpus/scheduling-gpus/]
Skaffold
@[https://www.infoq.com/news/2018/03/skaffold-kubernetes/]
- Tool to facilitate Continuous Development by allowing Local Kubernetes Development.
optimized “Source to Kubernetes” - Skaffold detects changes in
your source code and handles the pipeline to build, push, and deploy
your application automatically with policy-based image tagging and
highly optimized, fast local workflows.
- Skaffold handles the workflow for building, pushing and deploying the application,
allowing to focus on what matters most: writing code.
-ºLightweightº: client-side only.
-ºWorks Everywhereº: share your project with 'git clone' → 'skaffold run'.
- Support forºprofiles, local user config, environment variables, and flagsº
to easily incorporate differences across environments.
-ºFeature Richº: policy-based image tagging, resource port-forwarding and
logging, file syncing, and much more.
-ºOptimized Developmentº: Skaffold's inner loop is tight and highly optimized,
providing instant feedback while developing.
JAVA apli Client
- Useful for example to get k8s service config inside
running container.
Summary extracted from:
https://github.com/hyperledger/besu/blob/master/nat/src/main/java/org/hyperledger/besu/nat/kubernetes/KubernetesNatManager.java
import io.kubernetes.client.ApiClient;
import io.kubernetes.client.Configuration;
import io.kubernetes.client.apis.CoreV1Api;
import io.kubernetes.client.models.V1Service;
import io.kubernetes.client.util.ClientBuilder;
import io.kubernetes.client.util.KubeConfig;
import io.kubernetes.client.util.authenticators.GCPAuthenticator;
...
try {
KubeConfig.registerAuthenticator(new GCPAuthenticator());
final ApiClient client = ClientBuilder.cluster().build();
Configuration.setDefaultApiClient(client);
final CoreV1Api api = new CoreV1Api();
// invokes the CoreV1Api client
final V1Service service =
api.listServiceForAllNamespaces(null, null, null, null, null, null, null, null, null)
.getItems().stream()
.filter(
v1Service -˃ v1Service.getMetadata().getName().contains(besuServiceNameFilter))
.findFirst()
.orElseThrow(() -˃ new NatInitializationException("Service not found"));
updateUsingBesuService(service);
internalAdvertisedHost =
getIpDetector(service)
.detectAdvertisedIp()
.orElseThrow(
() -˃ new NatInitializationException("Unable to retrieve IP from service"));
final String internalHost = queryLocalIPAddress().get(TIMEOUT_SECONDS, TimeUnit.SECONDS);
service.getSpec().getPorts().forEach( v1ServicePort -˃ { ... } );
final String serviceType = service.getSpec().getType();
final Optional˂String˃ clusterIP = v1Service.getSpec().getClusterIP();
} catch (Exception e) {
throw new RuntimeException(
"Failed update information using pod metadata : " + e.getMessage(), e);
}