OnCallReadyStart learning

Kubernetes: Cluster Operations & Troubleshooting: glossary

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

admin.conf (super-admin.conf)
The cluster-admin kubeconfig kubeadm writes on the control plane. super-admin.conf is the system:masters version that bypasses RBAC. Lesson 20.1
apt-mark hold (apt-mark unhold, held packages)
Freezes a package so apt upgrade never changes it; unhold releases it. Kubernetes packages are held so upgrades happen only on purpose. Lesson 20.6
blackbox probe (blackbox check)
A check that tests a service from outside, like a client would (connect, read the certificate), without looking inside it. Lesson 20.11
bootstrap token (join token, kubeadm token)
A short-lived password (24 hours by default) a new node uses to join the cluster. kubeadm token create --print-join-command makes a new one. Lesson 20.6
cgroup driver (cgroupDriver, SystemdCgroup)
Who creates and manages the cgroups for containers. The kubelet and containerd must agree; kubeadm uses systemd for both. Lesson 20.1
CRI (Container Runtime Interface)
Container Runtime Interface: the API the kubelet uses to ask a runtime such as containerd to start and stop containers. Lesson 20.1
crictl
A command-line client that talks straight to the node's container runtime (containerd). Shows what really runs on the node, even when the apiserver is down. Lesson 20.1
crictl rmi --prune
Deletes every image on the node that no container uses - the safe way to free image space.
data-dir (--data-dir)
The directory where etcd keeps its database. In kubeadm the flag is a path inside the container; the hostPath volume decides which node directory that is. Lesson 20.22
discovery-token-ca-cert-hash (CA cert hash)
The hash in a join command that lets the new node check it is talking to the real cluster CA, not an impostor. Lesson 20.6
etcd snapshot (snapshot save, etcdctl snapshot save)
A single-file copy of the whole etcd database, taken with etcdctl snapshot save; the backup of the entire cluster state. Lesson 20.20
etcdutl (snapshot restore, snapshot status)
etcd's offline tool that works on snapshot files: snapshot status checks one, snapshot restore turns it into a data directory. Lesson 20.20
finalizer (finalizers, metadata.finalizers)
A key on an object that makes the apiserver wait with deletion until a controller has cleaned up and removed it. A finalizer nobody removes keeps the object Terminating forever. Lesson 20.28
front-proxy (front-proxy-ca, aggregated API)
The certificate pair the apiserver uses when it forwards requests to extra APIs served by add-ons (aggregated APIs), such as metrics-server. Lesson 20.11
ghost pod (Terminating ghost)
A pod on a dead node that shows Running or Terminating because nobody can confirm its end; it disappears when the node returns or the Node object is deleted. Lesson 20.25
imagePullSecrets (docker-registry Secret)
The pod field that names Secrets holding registry credentials, so the kubelet can pull from a private registry. Lesson 20.28
JWS (JWS signature)
JSON Web Signature: a signed piece of data. kubeadm signs its public cluster-info with each join token; no matching signature means that token is gone or expired.
kernel module (modprobe, lsmod, br_netfilter)
A piece of the kernel loaded on demand. modprobe loads one, lsmod lists them, /etc/modules-load.d lists those to load at boot. Lesson 20.6
klog
The logging library of every Kubernetes component. Its lines start with a severity letter (I, W, E, F), the date and time, and the source file. Lesson 20.1
kubeadm
The official tool that turns plain Linux machines into a Kubernetes cluster: kubeadm init builds the control plane, kubeadm join adds nodes, kubeadm upgrade moves versions. Lesson 20.1
kubeadm certs check-expiration (check-expiration, RESIDUAL TIME)
Lists every kubeadm-managed certificate with its expiry date and time left (RESIDUAL TIME); <invalid> means expired. Lesson 20.11
kubeadm certs renew
Re-issues kubeadm's leaf certificates with the same CA and names. The components must then be restarted to load them. Lesson 20.11
kubeadm reset
Undoes kubeadm init or join on a machine. It does not clean CNI config, iptables rules or kubeconfig files - it says so. Lesson 20.6
kubeadm upgrade (kubeadm upgrade plan, kubeadm upgrade apply, kubeadm upgrade node)
plan shows what an upgrade would do, apply upgrades the first control-plane node, node upgrades every other node. Kubelets are upgraded separately with the packages. Lesson 20.16
kubectl debug node (kubectl debug)
Starts a debugging pod on a node (or next to a pod), for when you cannot ssh to the machine.
Kustomize
A YAML-customising tool built into kubectl (kubectl apply -k); its version appears in kubectl version. Lesson 20.15
Lease (kube-node-lease, node lease)
A tiny heartbeat object each kubelet renews every few seconds in the kube-node-lease namespace; the node controller watches it. Lesson 20.25
minor version (patch version, minor)
In 1.37.1, 37 is the minor version (a new one about every four months) and 1 the patch (bug fixes). Upgrades move one minor at a time. Lesson 20.15
mirror pod (mirror pods)
The read-only copy of a static pod the kubelet puts in the apiserver so kubectl can see it. Deleting it does not stop the real pod. Lesson 20.3
network partition
Part of the network is cut off from the rest, so machines that are running cannot reach each other. Lesson 20.25
node-monitor-grace-period
How long the node controller waits without a heartbeat before marking a node Unknown (50s by default). Lesson 20.25
object storage
A cloud file store reached over HTTP, used for backups and large files (like a network drive without folders). Lesson 20.20
pkgs.k8s.io
The official Kubernetes package repositories, one per minor version, so a node pointed at v1.37 can only install 1.37.x. Lesson 20.6
PKI (public key infrastructure)
Public key infrastructure: the CAs and all the certificates they signed. kubeadm keeps the cluster's in /etc/kubernetes/pki. Lesson 20.6
PLEG (Pod Lifecycle Event Generator)
The kubelet loop that regularly asks the runtime what changed. "PLEG is not healthy" means the runtime is not answering. Lesson 20.1
preflight (preflight checks)
The checks kubeadm runs before doing anything (swap, ip_forward, ports, leftover files). Each failure is printed as an [ERROR ...] line. Lesson 20.6
quorum
The majority of etcd members that must agree for a write to count. 3 members survive 1 failure, 5 survive 2. Lesson 20.20
Raft (member, etcd member)
The agreement algorithm etcd uses to keep its copies (members) identical; a write is committed once a majority has it. Lesson 20.20
rotateCertificates
A kubelet setting that makes it renew its own client certificate before expiry, so kubelets never need manual renewal. Lesson 20.11
static pod (static pods)
A pod the kubelet runs from a file in its staticPodPath, with no scheduler or apiserver involved. The control plane runs this way. Lesson 20.3
staticPodPath (/etc/kubernetes/manifests)
The directory the kubelet watches for pod manifests to run by itself; on kubeadm nodes /etc/kubernetes/manifests. Lesson 20.1
sysctl (sysctl.d, ip_forward, net.ipv4.ip_forward)
The tool that reads and sets kernel settings. Files in /etc/sysctl.d are applied at every boot; sysctl --system applies them now. Lesson 20.6
TLS bootstrap
How a joining kubelet gets its own client certificate: it sends a CSR with the bootstrap token, the controller-manager approves it automatically. Lesson 20.6
Velero
A backup tool that saves Kubernetes objects and volume snapshots per namespace, so one namespace can be restored without touching etcd. Lesson 20.22
version skew policy (version skew)
The official rules for which component versions may run together: nothing newer than the apiserver, kubelets up to three minors older, kubectl within one. Lesson 20.15