OnCallReadyStart learning

Kubernetes: Architecture & Workloads: glossary

128 terms, in plain words, with where each one is taught.

--field-selector (field selector)
Filters objects on the server by a field value, e.g. --field-selector=status.phase!=Running. Lesson 17.38
--sort-by
Sorts kubectl get output by a field, e.g. --sort-by=.metadata.creationTimestamp. Lesson 17.38
$do
A shell variable for --dry-run=client -o yaml: append it to a create command to print the YAML instead of creating the object. Lesson 17.3
$now
A shell variable for --grace-period=0 --force: delete a pod immediately instead of waiting for it to stop. Lesson 17.3
admission (admission controller, admission webhook)
Checks the API server runs on a request after login and permissions and before saving: they may change the object or reject it. Lesson 17.11
adoption (adopt, release (controller))
A controller claims any matching pod that has no owner, and lets go of pods that stop matching its selector. Lesson 17.26
allocatable
How much CPU and memory a node offers to pods (its total minus what the system reserves). Lesson 17.42
API server (kube-apiserver, apiserver)
The front door of the cluster: every command and every component talks to it (over HTTPS), and only it reads and writes etcd. Lesson 17.1
apiVersion
The API group and version a manifest is written against, e.g. v1 for Pods, apps/v1 for Deployments, batch/v1 for Jobs. Lesson 17.3
authentication (authn)
Proving who you are to the API server (certificate, token); failure is HTTP 401. Lesson 17.11
authorization (authz)
Deciding whether you may do this action on this object; failure is HTTP 403 Forbidden. Lesson 17.11
Calico
The CNI plugin used in this lab: it gives pods IPs and routes between nodes, and can enforce network rules. Lesson 17.7
checksum annotation (checksum/config)
A hash of the config put on the pod template, so a config change changes the template and rolls the pods. Lesson 17.29
cloud-controller-manager
A control-plane component, only on cloud clusters, that creates the cloud's load balancers and removes nodes whose VMs are gone. Lesson 17.5
cluster-scoped (namespaced)
Object types that exist once per cluster (nodes, namespaces) rather than inside a namespace; -n is ignored for them. Lesson 17.3
CNI (Container Network Interface, CNI plugin)
The standard way a node gives each new pod its network: the runtime runs a CNI plugin (here Calico) that creates the interface and assigns an IP. Lesson 17.7
completions (parallelism, backoffLimit, activeDeadlineSeconds)
Job settings: how many successes are needed, how many pods at once, how many failures before giving up, and a time limit. Lesson 17.24
concurrencyPolicy
CronJob setting: Allow overlapping runs, Forbid a new run while one is active, or Replace the running one. Lesson 17.24
ConfigMap (ConfigMaps, cm)
An object holding plain configuration as key/value strings or whole files, handed to pods as env vars or files. Lesson 17.29
configMapKeyRef (envFrom, secretKeyRef, valueFrom)
Pod-spec ways to turn ConfigMap or Secret keys into environment variables: one key (KeyRef) or all keys (envFrom). Lesson 17.29
container runtime (runtime)
The program on each node that actually pulls images and runs containers (here containerd), driven by the kubelet. Lesson 17.7
containerPort
The port a container listens on, written in the pod spec. Informational: it opens nothing by itself. Lesson 17.3
control plane
The machines and programs that run the cluster itself: the API server, etcd, the scheduler and the controllers. Your apps run on the worker nodes. Lesson 17.1
controller
A loop that watches objects through the API server and acts to make reality match the spec, e.g. the ReplicaSet controller creating missing pods. Lesson 17.5
controller manager (kube-controller-manager)
The control-plane program that runs all the built-in controllers (deployment, replicaset, job, node, namespace...). Lesson 17.5
CrashLoopBackOff
The container keeps exiting and the kubelet waits longer and longer (up to 5 minutes) before restarting it again. Lesson 17.14
CRD (CustomResourceDefinition, custom resource)
A way to add new object types to the API server; extensions install their own kinds this way. Lesson 17.3
CreateContainerConfigError
The pod cannot start because a ConfigMap or Secret (or one of its keys) it references does not exist. Lesson 17.29
crictl
A command-line tool to talk to the node's container runtime directly - like docker ps for containerd, for debugging on a node. Lesson 17.7
CronJob (CronJobs, cj)
Creates a Job on a cron schedule (like a systemd timer, 2.16); concurrencyPolicy says what to do if the last run is still going. Lesson 17.24
custom-columns (-o custom-columns)
An output format where you name the columns and the field each one shows: -o custom-columns=POD:.metadata.name. Lesson 17.38
DaemonSet (DaemonSets, ds)
A workload that runs exactly one pod on every node (or every matching node) - log agents, node tools. Lesson 17.22
database migration
A script that changes a database's tables to the shape the new app version expects. Lesson 17.35
Deployment (Deployments)
The standard workload for stateless apps: keeps N replicas running and rolls out new versions gradually, with history and rollback. Lesson 17.16
Deployment rollout (rolling update, RollingUpdate, kubectl rollout)
Replacing a Deployment's pods with the new version step by step; kubectl rollout status/history/undo/restart manage it. Lesson 17.16
desired state (observed state)
What the spec says should exist, versus what actually exists; controllers work to close the gap. Lesson 17.9
Downward API (fieldRef, resourceFieldRef)
Lets a container read facts about its own pod (name, namespace, node, IP, limits) as env vars or files. Lesson 17.33
emptyDir
A scratch volume created empty when the pod starts and deleted with it; the pod's containers share it. Lesson 17.35
etcd
The small replicated database that holds every object of the cluster. Only the API server talks to it; back it up. Lesson 17.5
event (events, kubectl get events)
A short record a component writes about what it did to an object (Scheduled, Pulled, FailedMount). Shown at the bottom of describe; kept about an hour. Lesson 17.11
FailedMount
The event when the kubelet cannot set up a volume for a pod (e.g. the ConfigMap does not exist); the pod stays ContainerCreating. Lesson 17.29
FailedScheduling
The event when the scheduler found no node for a pod; its message counts why each node was rejected. Lesson 17.42
finalizer (finalizers)
A note on an object that blocks its deletion until some controller has finished cleaning up. Lesson 17.42
GA
Generally available: a Kubernetes feature that is stable and on by default (after alpha and beta). Lesson 17.35
generation (observedGeneration)
A counter that goes up whenever an object's spec changes; observedGeneration says which one the controller has acted on. Lesson 17.9
generator (kubectl create deployment, kubectl run)
A kubectl subcommand that builds an object from flags (k create deployment web --image=nginx), the quick start for a manifest. Lesson 17.3
headless Service (clusterIP: None)
A Service without a shared address, used to give each StatefulSet pod its own DNS name (db-0.db). Services are Ch 18. Lesson 17.19
ImagePullBackOff (ErrImagePull)
The node could not pull the image (wrong name or tag, no access to the registry) and is waiting before trying again. Lesson 17.14
imagePullSecrets (docker-registry secret, dockerconfigjson)
A pod field naming a Secret with registry credentials, so the node can pull private images. Lesson 17.33
immutable ConfigMap (immutable: true, immutable field)
A ConfigMap or Secret marked immutable cannot be edited, only replaced; some object fields (a pod's command) are immutable too. Lesson 17.29
imperative command
A kubectl command that says what to do (create, scale, set image) instead of describing the end state. Lesson 17.40
init container (initContainers, init containers, Init:N/M)
A container that runs to completion before the app containers start, one after another; Init:1/2 = one of two done. Lesson 17.35
Job (Jobs)
A workload that runs pods until a task succeeds a given number of times, then stops. Lesson 17.24
jsonpath (-o jsonpath, range)
A small query language for -o jsonpath='{...}' that prints chosen fields of objects, like a mini jq. Lesson 17.38
k alias (alias k=kubectl)
The shortcut alias k=kubectl (6.1) the course uses everywhere, with shell completion for it. Lesson 17.3
kind
The type of a Kubernetes object, written in its manifest: kind: Deployment, kind: Pod. Lesson 17.3
KMS
Key management service: a separate (usually cloud) store that holds encryption keys so they never sit next to the data. Lesson 17.33
kubeadm
The standard tool that installs a Kubernetes cluster on plain machines; this lab was built with it (Ch 20). Lesson 17.5
kubeconfig context (context, current-context, use-context)
A named pair of cluster + user (plus a default namespace) in your kubeconfig; kubectl config use-context NAME switches between them. Lesson 17.1
kubectl
The command-line client for Kubernetes. It sends your requests to the API server; it never touches the nodes directly. Lesson 17.1
kubectl api-resources (api-resources, short name)
Lists every object type the cluster serves, with its short name (po, deploy, cm), API version, kind and whether it is namespaced. Lesson 17.3
kubectl apply (apply -f)
Makes the cluster match a file: creates what is missing, updates what changed, says unchanged otherwise. Lesson 17.40
kubectl debug (ephemeral container)
Adds a temporary extra container with tools to a running pod (or a node), for images without a shell. Lesson 17.42
kubectl describe (describe)
Prints a readable summary of an object plus its recent Events - the first command when something is wrong. Lesson 17.11
kubectl diff
Shows what apply would change, as a diff; exit code 1 means there are differences - Kubernetes' terraform plan. Lesson 17.40
kubectl dry run (--dry-run, --dry-run=client, --dry-run=server)
Run a command without saving anything: client builds the object locally, server sends it through all checks and stops before storing. Lesson 17.3
kubectl edit
Opens the live object in an editor and saves your change straight to the cluster - quick but unreviewed. Lesson 17.40
kubectl exec (exec -it)
Runs a command inside a running container, like docker exec; -it for an interactive shell. Lesson 17.14
kubectl explain
Prints the documentation of any field: k explain deployment.spec.strategy (add --recursive for the whole tree). Lesson 17.3
kubectl expose (k expose)
Creates a Service for a Deployment or pod in one command (Services are Ch 18).
kubectl logs --previous (--previous, -p)
Shows the log of the container instance that died, not the one that just restarted. Lesson 17.14
kubectl patch (strategic merge patch, merge patch, JSON patch)
Changes only some fields; strategic (default) merges lists by name, merge replaces lists, JSON patch edits exact paths. Lesson 17.40
kubectl replace --force (replace --force, kubectl replace)
Deletes and recreates an object from a file - the way to change fields that cannot be updated in place. Lesson 17.40
Kubernetes (k8s)
A system that runs containers across a group of machines for you: you describe what should run, and it keeps that true - restarting, replacing and spreading containers. Lesson 17.1
Kubernetes annotation (annotations, kubectl annotate)
Key/value metadata for tools and humans; unlike labels it cannot be selected on and can be large. Lesson 17.26
Kubernetes garbage collector (cascading delete, --cascade)
The controller that deletes owned objects when their owner is deleted (--cascade=orphan keeps them). Lesson 17.9
Kubernetes label
A key/value pair in an object's metadata.labels (app: web), used by selectors to group objects. Lesson 17.26
Kubernetes namespace (ns, kube-system, default namespace)
A folder for objects inside a cluster: names are unique per namespace; deleting one deletes everything in it. Not a network boundary. Lesson 17.26
Kubernetes object (resource (Kubernetes))
One record stored in the API server - a pod, a Deployment, a ConfigMap - with metadata, a spec (what you want) and a status (what is). Lesson 17.1
Kubernetes Secret (Opaque, kubectl create secret)
Like a ConfigMap but for passwords and keys: values stored base64-encoded (not encrypted), volumes kept in RAM. Lesson 17.33
label selector (selector, selectors, matchLabels)
A query over labels ("app=web", "tier in (frontend)") that workloads and other objects use to find their pods. Lesson 17.26
last-applied-configuration (3-way merge, three-way merge)
The annotation where apply remembers what you applied last, so it can tell what you removed from the file. Lesson 17.40
leader election
How several copies of a control-plane component agree that only one acts at a time. Lesson 17.5
maxSurge (maxUnavailable)
Rollout limits: maxSurge = how many extra pods above replicas may exist; maxUnavailable = how many may be missing. Lesson 17.16
metadata
An object's name, namespace, labels, annotations and bookkeeping fields (uid, creationTimestamp, ownerReferences). Lesson 17.3
mirror pod
The read-only copy of a static pod that the kubelet shows in the API, so kubectl get pods can see it. Lesson 17.5
node lease (kube-node-lease)
A small object each kubelet renews every few seconds as a heartbeat; when it stops, the node becomes NotReady. Lesson 17.7
nodeSelector
A pod field listing node labels the pod requires: it only runs on nodes that have them. Lesson 17.22
NotReady (Ready (node))
A node's Ready condition is False or Unknown: the kubelet is not reporting healthy, so no new pods go there. Lesson 17.7
ordinal
The number in a StatefulSet pod's name (db-0, db-1); it is the pod's stable identity. Lesson 17.19
ownerReference (ownerReferences)
A field in an object's metadata naming the controller object that owns it (pod -> ReplicaSet -> Deployment). Lesson 17.9
pause container
A tiny container that holds a pod's network namespace; the pod's real containers join it. Lesson 17.7
PersistentVolumeClaim (PVC, PVCs)
A pod's request for a piece of storage that outlives the pod (Ch 18 in detail). Lesson 17.19
pod grace period (terminationGracePeriodSeconds, --grace-period)
How long Kubernetes waits after SIGTERM before SIGKILL when a pod is deleted (default 30 s) - the pod-deletion flow of 3.18. Lesson 17.14
pod phase (phase, Succeeded, Failed)
A pod's coarse state in status.phase: Pending, Running, Succeeded, Failed or Unknown. Lesson 17.14
pod template
The pod description inside a workload (spec.template); every pod the workload creates is stamped from it. Lesson 17.16
pod-template-hash
A label a Deployment adds to each ReplicaSet and its pods, so versions never adopt each other's pods. Lesson 17.16
PodInitializing
The STATUS shown after all init containers finished and while the app containers start. Lesson 17.35
ProgressDeadlineExceeded (progressDeadlineSeconds)
The Deployment condition set when a rollout made no progress for 10 minutes (default); rollout status then fails. Lesson 17.16
readiness probe (readinessProbe)
A health check the kubelet keeps running against a container; while it fails the pod is not Ready (and gets no traffic), but it is not restarted. Ch 19 covers probes. Lesson 17.14
READY column (READY)
In kubectl get pods: ready containers / total containers, e.g. 1/1 or 0/2. Lesson 17.14
reconciliation loop (reconcile, reconciliation)
Watch, compare desired with actual, act, repeat - forever. Delete a pod of a Deployment and the loop makes a new one. Lesson 17.9
Recreate strategy (Recreate)
A Deployment strategy that stops all old pods before starting new ones - downtime by design. Lesson 17.16
replicas (replica)
How many copies of the pod a workload should keep running. Lesson 17.16
ReplicaSet (ReplicaSets, rs)
Keeps N identical pods matching its selector running. Deployments create one per version; you rarely make one yourself. Lesson 17.16
resource request (resources.requests)
How much CPU and memory a container asks the scheduler to reserve for it; pods only land where requests fit (Ch 19). Lesson 17.42
resourceVersion
A version number on every object, changed on each write; used to detect conflicting updates and to resume watches. Lesson 17.9
restartPolicy
What the kubelet does when a pod's container exits: Always (default), OnFailure or Never - like systemd Restart= (2.10). Lesson 17.14
RESTARTS column (RESTARTS)
How many times the pod's containers were restarted, with how long ago the last one was. Lesson 17.14
revision (rollout history)
A numbered version of a Deployment's pod template; rollout undo --to-revision=N goes back to one. Lesson 17.16
rollout restart (kubectl rollout restart)
Replaces all pods of a workload under its rollout strategy, e.g. so they re-read changed env vars. Lesson 17.29
scheduler (kube-scheduler)
The control-plane component that picks a node for each new pod (filter, score, bind) by setting spec.nodeName. It does not start anything. Lesson 17.5
Secret encryption at rest (EncryptionConfiguration)
Configuring the API server so Secrets are stored encrypted in etcd, not just base64. Lesson 17.33
set-based selector (in, notin)
A selector using in, notin, exists or does-not-exist instead of just key=value. Lesson 17.26
sidecar (sidecars, native sidecar)
A helper container running beside the app for the pod's whole life (log shipper, proxy); natively an init container with restartPolicy: Always. Lesson 17.35
spec
The part of an object that says what you want (desired state). You write it. Lesson 17.1
StatefulSet (StatefulSets, sts)
A workload for apps that need a stable identity: pods named web-0, web-1... started in order, each keeping its own storage. Lesson 17.19
StatefulSet partition (rollingUpdate.partition)
A StatefulSet setting: only pods with an ordinal >= partition get the new version - a built-in canary. Lesson 17.19
static pod
A pod the kubelet starts straight from a file in /etc/kubernetes/manifests, without the API server - how kubeadm runs the control plane. Lesson 17.5
status (.status)
The part of an object that says what is true right now (observed state). Controllers and the kubelet write it. Lesson 17.1
stringData
A write-only field of a Secret that takes plain text and stores it base64-encoded under data. Lesson 17.33
subPath
Mount one file from a volume instead of the whole directory. subPath mounts are never updated when the source changes. Lesson 17.29
Terminating
The STATUS of a pod (or namespace) that is being deleted and has not finished yet. Lesson 17.14
toleration (tolerations)
A pod setting that lets it run on a node carrying a matching taint ("keep off" mark), like the control plane's. Lesson 17.22
volumeClaimTemplates
A StatefulSet field that gives each pod its own storage claim (data-db-0, data-db-1). Storage is Ch 18. Lesson 17.19
watch (-w, --watch)
A long-lived request asking the API server to push every change; kubectl get -w shows changes as they happen. Lesson 17.5
worker node (worker, workers)
A machine in the cluster that runs your pods (as opposed to the control-plane node). Lesson 17.1
workload (workload resource)
An object that runs pods for you: Deployment, StatefulSet, DaemonSet, Job, CronJob. Lesson 17.16