Learn Red Hat OpenShift by playing. Built for EX280.
shifting.pro is a browser-based 3D OpenShift cluster simulator. Create Projects with real
oc commands, deploy applications, expose Routes, run S2I builds, configure SCCs, RBAC, quotas and
Operators - then investigate and fix realistic production incidents. No install, no signup, no build
step: every lab runs client-side in your browser.
Play: https://shifting.pro · Draw: https://shifting.pro/draw
Documentation: the learner-facing guides live in the
GitHub wiki - start with the
EX280 Study Guide and the
EX280 Prep mode reference. The engineering
docs (architecture map, conversion spec, declarative engine, research notes, verification report) stay in
docs/.
Target platform: Red Hat OpenShift Container Platform 4.22 (Kubernetes 1.35, CRI-O) · Exam: EX280V422 (“Red Hat Certified System Administrator in OpenShift exam” on redhat.com; the LMS objectives PDF lists the credential as “Red Hat Certified Specialist in OpenShift Administration” - same objectives, both names are shown in EX280 Prep).
octerminal - a simulated OpenShift CLI withget,describe,apply -f,create,new-app,new-project,expose,create route,start-build,set env|image|resources|volume,rollout(status --timeout|history|undo;status --timeoutblocks like realocand exits 1 witherror: timed out waiting for the conditionon a stalled rollout, so&&chains stop),logs(Pods, Deployments, Jobs),exec,rsh,cp,wait,diff -f,patch --type=strategic|merge|json,adm policy(RBAC + SCC),auth can-i,adm top,adm cordon|drain,adm taint(all real forms, including kubectl's empty-valuekey=:Effect),create quota|clusterquota(LimitRanges are declarative -oc apply -f, there is nooc create limitrangesubcommand),create token,image info,process,helm(repo/search/install/upgrade/history/rollback),htpasswd,debug node(interactivechroot /host+systemctl/journalctlkubelet troubleshooting),--dry-run=client|server,<resource>.<group>tokens, escaped jsonpath keys,custom-columns,events --field-selector, pipes, redirects and-f -stdin.oc explainprints kubectl 1.35's explain v2 plaintext (GROUP:only for non-core groups, alignedKIND:/VERSION:, aFIELD: <path> <Type>line for a field request, a wrappedDESCRIPTION:and alphabetical tab-separatedFIELDS:, including the Deployment strategy'srollingUpdate);-o jsonpathmarshals list/object fields as JSON ({.spec.template.spec.containers[0].env}→[{"name":"APP_ENV","value":"production"}]);oc get alland every list output are sorted by name like the API server;helm listUPDATED prints helm's real Go timestamp;oc create role|clusterrole --resource=securitycontextconstraints|sccresolves the real API group (security.openshift.io, so a customuseRole grants the SCC); andoc set resources job/<name>is rejected like the API server (The Job "<name>" is invalid: spec.template: … field is immutable) whilecreate job|cronjob --dry-run=client -o yamlnames the container after the object.oc describerenders like kubectl: envvalueFromentries print the<set to the key 'DB_PASS' in secret 'db'> Optional: false/<set to the key 'APP_LOG' of config map 'cfg'> Optional: falseplaceholders (fieldRef and resourceFieldRef forms too, never the raw JSON), Liveness/Readiness/Startup probes print in kubectl's order and format (Readiness: http-get http://:8080/ delay=2s timeout=1s period=10s #success=1 #failure=3,tcp-socket :PORT,exec […]),Volumes:print theType:/Name:/Optional:blocks for ConfigMap/Secret/emptyDir/PVC/hostPath/NFS/projected/ downwardAPI/CSI sources (including the injectedkube-api-access-*projected service-account volume) aligned per volume, and annotations print as sorted multilinekey: value(labels staykey=value).oc describe podcontainer states match kubectl 1.35 (a Waiting container prints onlyReason, a TerminatedLast StateprintsStarted:beforeFinished:, a Running container printsContainer ID: cri-o://<id>and an image digest),oc describe jobprintsActive Deadline Seconds: 8s,oc describe quotaprints canonical quantities (1,1Gi) andoc get identitiesfills USER UID.oc set probehonours--initial-delay-seconds,--period-seconds,--timeout-seconds,--success-thresholdand--failure-threshold(only the supplied fields change; the handler and API-server defaults are preserved),oc set probe … -- <command>stores a realexechandler (verbatim argv), and the kubelet tick actually evaluates every player-set probe: a readiness probe on a closed port/wrong path flips the Pod to0/1 Runningwith kubectl'sReadiness probe failed: Get "http://<podIP>:<port><path>": dial tcp …event, removes it from the Service endpoints (the Route answers 503) and restores it when fixed; a failing liveness/startup probe restarts the container (Container <name> failed liveness probe, will be restarted, RESTARTS climbs, then CrashLoopBackOff); exec probes run in the same simulated container filesystem asoc exec(cat /tmp/healthy→No such file or directory), andoc describe deploymentlists only the Deployment's ownScalingReplicaSetevents -Scaled up replica set <rs> to <n>per rollout andScaled down replica set <old> to 0 from <n>when one is retired, at their real rollout times (ReplicaSet-controller events stay onoc describe rs).oc adm top nodesis sorted by NAME with a realistically busy control plane and per-node worker variance. Container names follow realoc(oc create deployment --image=…names the container after the image;oc set imagenever adds one).oc get rolebinding|clusterrolebindingprints the ROLE as<roleRef.kind>/<roleRef.name>(ClusterRole/admin,Role/edit), so Role vs ClusterRole is visible like realoc, and-o wideprints ServiceAccount subjects as<namespace>/<name>(qa8/deployer) likerbac.SubjectsStrings.oc exec/oc rshreturn the remote command's exit status: an unmodelled binary fails with the real OCIexecutable file not found in $PATHerror andcommand terminated with exit code 127, coreutils behave like the image's own binaries (trueexits 0 silently,falseexits 1), and a real command's failure is surfaced with its real status (lson a missing path exits 2,cat/grepfailures exit 1) followed by the realcommand terminated with exit code <N>line, so&&chains stop (||still runs). Cluster DNS is NetworkPolicy-aware:nslookup,digandgetent hostsquery the sameopenshift-dnsdns-default Pods on 5353 thatcurl's name resolution uses, print the real tool formats (nslookupapplies the Pod's search list, so barewebprintsName: web.<ns>.svc.cluster.local;dig +shortprints just the Service ClusterIP anddighas its own header/ANSWER/SERVER format;getent hostsprints the glibcIP canonicalline) and fail with;; connection timed out; no servers could be reachedplus the real exit status when an egress policy blocks DNS - the lab 13 diagnostic flow. VerbosecurlprintsTrying/Connectedonly in the real order (never before a name resolves, neverConnectedon a failed connection). RBAC is genuinely enforced: a plain user gets the realError from server (Forbidden): …for reads and writes they are not allowed to perform.ocis the taught command; a leadingkubectlis accepted as an alias with an educational hint.- A real shell around
oc-vim/vi(opens the manifest editor),cat,echowith>/>>redirects, heredocs, command substitution,;/&&/||sequencing, session variables (export NS=default,$NS/${NS}expansion), bash's$?/${?}exit-status parameter (oc exec … -- ls /nope; echo rc=$?→rc=2,false; echo $?→1) and pipes throughgrep,sed,tail,head,cut,wc,sort,base64 -d,openssl(including realreq -x509/x509PEM handling),htpasswd,skopeo- so exam workflows like writing YAML in the terminal, decoding a Secret andoc apply -f -work as on a real workstation.$(...)also works at the end of a word (curl -s http://$(oc get route webapp -o jsonpath={.spec.host})),/tmpis one real virtual directory (oc get … > /tmp/webapp.yaml,cat /tmp/webapp.yaml,ls /tmpandoc apply -f /tmp/…agree),echoprints exactly one trailing newline, andoc create secret tls x --cert=missing.crt --key=missing.keyfails with the realerror: open missing.crt: no such file or directory(theopenssl req -x509flow is the taught path). Unsupported shell grammar (loops, conditionals) is rejected with a clear simulator message instead of silently running part of the line. A workstation-sidecurlresolves Route hosts through the router model (edge/passthrough/reencrypt, http redirect and NetworkPolicy reachability), with real flags and-w/--write-out(%{http_code}etc.), andoc exec/oc rshshow the Pod's real volumes, ConfigMaps, Secrets, service-account token and env.oc exec … -- sh -c '<script>'runs the script in the container's own shell, so single-quoted$DB_PASS/$APP_MODEexpand from the container environment (s3cr3t prod) and$HOSTNAMEprints the pod name; shell builtins behave like bash (sh -c "exit 3"; echo rc=$?→command terminated with exit code 3+rc=3),id -u/-g/-unprint bare IDs like coreutils,;/&&/||/pipes follow the real exit-status rules, and a double-quoted$APP_MODEis still expanded by the workstation shell first (oc exec -- exit 3stays the authentic missing-binary 127 -exitis only a shell builtin).cat /etc/resolv.confinside a Pod prints the cluster DNS configuration (search <project>.svc.cluster.local svc.cluster.local cluster.local,nameserver 172.30.0.10,options ndots:5). - A simulated RHEL 9 workstation (
js/engine/LinuxHost.js+js/ui/LinuxCommands.js) - the terminal is[student@workstation ~]$onworkstation.lab.example.com(RHEL 9.6, SELinux enforcing,studentinwheelwith lab sudo). It models a filesystem with modes, owners, SELinux contexts and symlinks (home and/tmpfiles stay the same files the Files panel andoc apply -fuse), users and groups, processes, systemd units with a journal (a unit can fail to start until its config is fixed; httpd and sshd validate their real config files), dnf/rpm packages, the network (ip,ss,ping,/etc/hostsbefore DNS,dig/hostfor the lab cluster names), firewalld runtime vs--permanentand rootless podman. About 110 commands with real RHEL output and exit codes:cd/pwdwith a cwd every command honours,ls -l(with the SELinux.),cp -r,mv,rm -rf, globbing and brace expansion,find,tee,tar,chmod/chown/sudo/id,getenforce/restorecon/ls -Z,ps/kill/top -b,systemctl/journalctl,firewall-cmd,dnf/rpm,podman, scripts with a shebang run by name - and permission errors where RHEL gives them (bash: /etc/hosts: Permission denied). - First-class OpenShift resources - Project, Pod, Deployment, ReplicaSet, StatefulSet, DaemonSet,
Job, CronJob, Service, Route, Ingress, NetworkPolicy, ConfigMap, Secret, PVC, PV, StorageClass,
HPA, PDB, ResourceQuota, LimitRange, ClusterResourceQuota, ServiceAccount, Role, RoleBinding,
ClusterRole, ClusterRoleBinding, SCC, BuildConfig, Build, ImageStream, Template, Subscription,
OperatorGroup, InstallPlan, ClusterServiceVersion, PackageManifest, MetalLB, User, Group,
ClusterOperator, ClusterVersion, IPAddressPool, L2Advertisement, PrometheusRule, Alertmanager.
A single central registry (
js/data/kindRegistry.js) drives the palette, 3D renderer, inspector, command bar, minimap and YAML export. - S2I builds & ImageStreams -
oc new-app, BuildConfig → Build → ImageStream lifecycle, image change triggers and rollouts, internal registry image references. In the paused pre-start sandbox theoc new-appScalingReplicaSet event already prints the full deterministic<name>-<10-char pod-template-hash>ReplicaSet name, and starting the simulation creates exactly that ReplicaSet (theoc get rs/oc describe rsoutput matches the event). A build's digest image (…/<project>/<name>@sha256:<digest>) resolves through the ImageStream tag history, so the new Deployment's Pods start the code the build produced and the rollout completes; a pre-build:latestPod isImagePullBackOffuntil the first successful push, then recovers.oc new-appnames Service ports8080-tcp/3306-tcp(newapp naming) and the Route PORT column follows, andoc create route passthrough <name> --service=<svc>setsspec.port.targetPortto the Service port name/number like realoc. - Routes & networking -
oc expose svc/..., edge/passthrough/reencrypt TLS Routes, Route admission state, Services, NetworkPolicies, MetalLBLoadBalancerservices with EXTERNAL-IP. Route admission follows the real 4.22 router rules: host/path ownership (HostAlreadyClaimed), external-certificate validation and target checks - a Service with no ready endpoints does not reject the Route; the router answers 503 andoc describe routeshowsEndpoints: <none>, so the player debugs the selector/endpoints instead of chasing a fake admission failure. The service-ca controller issues Service serving certificates (service.beta.openshift.io/serving-cert-secret-name) and injects CA bundles (inject-cabundle) for reencrypt Routes; a default-deny policy really blocks a project's Route until an ingress NetworkPolicy admits the router namespace (the Network Policies level teaches the fix and the capstone requires it). The platform namespaces are populated by real controllers like a real install:openshift-ingressrunsdeployment.apps/router-default(2/2 Ready) with its ReplicaSet and two shared-hash router pods, andopenshift-dnsruns thedaemonset.apps/dns-default+daemonset.apps/node-resolverDaemonSets with one owned Pod per Node (thedns-defaultPods report READY2/2). DNS pods run on the pod network (10.128.x-10.131.x- each Pod IP is allocated inside its Node'sspec.podCIDR/23,hostNetwork: false) whilenode-resolverand the router stay hostNetwork, both DNS DaemonSets show the realkubernetes.io/os=linuxNODE SELECTOR inoc get ds -n openshift-dns, and the router pods carryingresscontroller.operator.openshift.io/deployment-ingresscontroller=default, so the documented label-selectedoc delete pod -n openshift-ingress -l …flow finds and deletes them. Those pods ARE theoc debug/NetworkPolicy labs' DNS egress target, and deleting one makes the owning controller heal it. The real platform Services are seeded in every mode, including the paused pre-start sandbox:openshift-dns/dns-default(ClusterIP172.30.0.10, ports53/UDP,53/TCP,9154/TCP) andopenshift-ingress/router-internal-default(80/TCP,443/TCP,1936/TCP), sooc get svc,endpoints -n openshift-dnsis the real DNS-troubleshooting check (lab 13) andoc exec <pod> -- cat /etc/resolv.confshows the cluster search domains,nameserver 172.30.0.10andoptions ndots:5. Service objects carry the API server's realspec.clusterIP(unique172.30.0.0/16;Nonewhen headless) plus the defaultedclusterIPs/ipFamilies/ipFamilyPolicy/sessionAffinityand per-portprotocol/targetPort, and that one field drivesoc get svc,-o yaml/json,-o jsonpath={.spec.clusterIP},oc describe svc, cluster DNSnslookupandoc exec … curl http://<ClusterIP>:<port>alike. Status is the real{loadBalancer: {}}subresource (no invented phase/conditions/endpoint counters),oc describe svcprints the kubectl 1.35 / OCP 4.22 shape (IP Family Policy: SingleStack,IP:/IPs:, onePort:/TargetPort:/Endpoints:block per port,Session Affinity: None,Internal Traffic Policy: Cluster, and - when the operator sets it -Traffic Distribution:at the shared tabwriter column),oc get endpointstruncates the ENDPOINTS cell like kubectl (ip:port,ip:port,ip:port + N more...), andnslookupanswers a Service name with its own ClusterIP (headless → ready pod IPs, ExternalName → its target) and an unknown name with the realNXDOMAINline. Thedns-defaultService carries the real 4.22 named targetPorts (TargetPort: dns/UDP,dns-tcp/TCP,metrics/TCP) while its ready endpoints stay 5353/5353/9154, and itsspec.trafficDistribution: PreferSameNode(the cluster-dns-operator's 4.22 asset) shows up inoc describeand inoc get svc dns-default -n openshift-dns -o jsonpath={.spec.trafficDistribution};oc describealigns itsEvents: <none>placeholder like kubectl's single tabwriter (column 27 for ClusterIP Services), and each Pod's/etc/hostsis the kubelet-managed file (# Kubernetes-managed hosts file., the loopback/localnet lines and<podIP>\t<podName>- no Service entry), sooc exec … cat /etc/hostsmatches a real Pod. The control-plane Node carries the realnode-role.kubernetes.io/master:NoScheduletaint, so application workloads schedule onto the workers while the tolerated DaemonSet pods keep running on the master - the taints/tolerations and "why is my Pod on the wrong node / Pending" troubleshooting flows behave like 4.22: an unschedulable Pod reports0/1 Pending(ContainerCreatingonly after it is bound),oc describe podprintsNode: <none>and no fabricated kubelet detail (noContainer ID,Image ID,State,ReadyorRestart Countlines until the Pod is bound - only Image/Port/Host Port/Environment/Mounts/Conditions), and FailedScheduling retries follow the real scheduler backoff (1 s doubling to 10 s max) sooc describe/oc get eventsshow realistic small counts like(x4 over 7s)instead of a zero-second burst. hello-openshift answers plain textHello OpenShift!; a passthrough Route to a plain-HTTP backend fails the TLS handshake (curl: (35) error:0A00010B:SSL routines::wrong version number, exit status 35) while lab 12's TLS fixture serves 200; and MetalLB LoadBalancer Services emit the controller'sNormal IPAllocated Assigned IP ["<ip>"]event (no cloud-providerEnsuredLoadBalancertext). - Declarative workflows - a Files panel with editable YAML manifests,
oc apply -f, Kustomize base/overlays (oc apply -k), OpenShift templates (oc process -f … | oc apply -f -), Helm charts (helm install|upgrade|uninstall),--dry-run=client -o yamlmanifest skeletons, terminal-authored YAML (heredocs/vim) and anhtpasswdfile for identity labs.oc kustomize/oc apply -krender resources in Kustomize's legacy sort order (ConfigMap → Service → Deployment). - Security & identity - HTPasswd identity provider +
oc login(validated against thehtpass-secretSecret, like a real cluster), users/groups/identities, RBAC evaluation (oc auth can-i,oc adm policy …,oc policy …) enforced on every command, OCP 4.x SCC grants (RBAC bindings tosystem:openshift:scc:<name>, with the SCC users/groups lists left untouched like 4.x) and spec-based SCC admission: a Pod whose manifest explicitly requestsrunAsUser: 0,privileged: trueorhostNetworkis rejected as a ReplicaSetFailedCreateand no Pod is created. The classic exam trap is modelled the way real OCP behaves: a barenginx:latest(nosecurityContext) is admitted underrestricted-v2and then crash-loops withmkdir() "/var/cache/nginx/client_temp" failed (13: Permission denied)inoc logs- theServiceAccount+anyuidgrant +oc set serviceaccountflow, or an explicit non-rootrunAsUser, is the escape hatch.htpasswd -Bwrites real-shaped bcrypt hashes and login validates bcrypt and legacy{SHA}entries. The installer-side kubeadmin password is discoverable the real way:cat ~/auth/kubeadmin-password, andoc login -u kubeadminwithout-pprompts for it like realoc. As in a real install, thekube-system/kubeadminSecret stores only a bcrypt hash of that password (oc extract secret/kubeadmin -n kube-system --to=-prints$2a$10$…) and never the plain text; kubeadmin is a virtual user (absent fromoc get users), and the exam removal step isoc delete secret kubeadmin -n kube-systemonce another cluster-admin exists. The full exam identity lifecycle is practised: rotate the HTPasswd password through theopenshift-config/htpass-secret(oc extract→htpasswd→oc set data) and wait for the OAuth rollout, then delete a departing user and its Identity (oc delete userleaves the Identity behind).oc loginalso prints loginoptions.go's project text (You have access to the following projects and can switch between them with 'oc project <projectname>':with*marking the current project, orYou have one project on this server: "<name>"). A Deployment whoseserviceAccountNamedoes not exist gets no Pod - the ReplicaSet recordsError creating: … error looking up service account <project>/<sa>: serviceaccount "<sa>" not found- and the SCCFailedCreatemessage appends the per-provider reasons (provider "anyuid": Forbidden: not usable by user or serviceaccount, spec.containers[0].securityContext.privileged: Invalid value: true: Privileged containers are not allowed, …). - Quotas & self-service - Namespace/Project quotas, LimitRanges, ClusterResourceQuota, project
templates (
oc adm create-bootstrap-project-template+project.config.openshift.io/cluster) and the realself-provisionersClusterRoleBinding (remove it andoc new-projectreturns Forbidden). ChangingprojectRequestTemplaterolls out the simulated openshift-apiserver for 8 game seconds (Progressing=True reason=Working) before the nextoc new-projectapplies the template. - Operators (OLM) - a software catalog of PackageManifests, OperatorGroups, Subscriptions that
move through InstallPlan and ClusterServiceVersion phases (
Installing → Succeeded), real operator effects (e.g. the MetalLB operator creates its controller/speaker pods and serves LoadBalancer IPs only once the MetalLB CR, IPAddressPool and L2Advertisement exist), and a correct two-step uninstall: deleting the Subscription keeps the CSV; deleting the CSV removes the operands. Operator install variants are modelled: AllNamespaces installs through the global OperatorGroup intoopenshift-operators, ManualinstallPlanApprovalwaits for theoc patch installplan … approved:truestep, andstartingCSVpins the installed version - after a pinned install OLM immediately resolves the channel head into a second, unapproved Manual InstallPlan (InstallPlanPending True RequiresApproval,currentCSVhead /installedCSVpin), so the real exam decision "Manual and why" is playable end to end. The catalog package names follow OCP 4.x (metallb-operator); OLM resolves per namespace, so a Subscription for a package that is not in the catalog makes every Subscription in that namespace reportResolutionFailed(with no fake CSV, InstallPlan or install-plan conditions), while other namespaces install normally. Themetallb-operatorbundle follows redhat-operators' OCP-minor versioning (metallb-operator.v4.22.0-202610281200) instead of upstreamv0.14.9. - Monitoring & alerting -
oc get prometheusrules -A,oc describe prometheusrule, liveAlertmanagerstate (oc get|describe alertmanager -n openshift-monitoring): the Watchdog alert always fires and cluster state (crash-looping or unready Pods) surfaces the matching alert rules. - 22-level EX280 campaign (5 chapters) + an optional Linux & RHEL Foundations prelude - the
prelude (levels 0.1-0.7) runs on a simulated RHEL 9 workstation; then Manage OpenShift →
Declarative & Kustomize → Deploy & Expose → Identity, RBAC & Self-service → Security & Operations,
which teaches NetworkPolicies (default-deny, the router allow, DNS egress) and node troubleshooting
(
oc adm top,oc describe node, cordon/drain/uncordon,oc debug node→chroot /host→systemctl/journalctl) before the production platform build (Level 22) whose capstone only completes once its scheduled incidents are resolved. Nothing is used before it is taught: Level 12 configures a MetalLB operator the platform team already installed (Level 18 teaches OLM installs), Level 4 teaches--dry-run=client -o yaml > fileskeletons, and SCC admission details wait for Level 17. Level ids are stable save keys - campaign order comes from the chapter table and the numbers players see are positional - so inserting levels never moves anyone's progress (the capstone keeps id 20 and is displayed as Level 22). The capstone's node failure is fixed the real way:oc debug node/<n>→chroot /host→systemctl restart kubelet(the incidents drawer action remains as a fallback). Grading follows state: L13's OAuth-referenced htpasswd Secret, the rotated password hash plus a successful login, L17's admitted privileged Pod and equivalent RBAC/SCC grants are credited (lab 7/8/11 too), PendingScheduling patches the Deployment template (the new ReplicaSet Pod stays Pending with a real FailedScheduling event while old Pods serve; rollout undo, patch/edit oroc label nodefix it), L3 crashes from the first start, L14's denial shows inoc logs deploy/web(no invented events), L16 proves in-cluster API access with a PodList/403, and L7/L11 objectives start unticked. - Guided missions - every campaign level says where to work and how to do each step. The
mission card's project row shows the lab project the level is graded in and, when you are
somewhere else (every level starts in
default), the one command that fixes it; Next step has an opt-in Show me how with the exact commands (plus what they do and a check command), each with Copy and Put in terminal; revealed hints render commands as clickable chips in one scrolling card (never a cramped nested scroller); and an unticked objective explains itself- the right object in the wrong project, a Template that was only applied (not processed) or only
rendered, a password key in a ConfigMap. Help (
H) mid-level is a "where am I" view (done objectives, the current step, its how-to), and the Files panel says what Apply -f will not do for a Template or a Helm chart. All of it is authored data (objective.how,workProject) that the browser suite replays (qa-guidance-replay: every how-to command runs in a real browser and the level must complete), so the guidance cannot silently rot.
- the right object in the wrong project, a Template that was only applied (not processed) or only
rendered, a password key in a ConfigMap. Help (
- 13 timed EX280 labs - speed drills: three-tier app, CrashLoopBackOff, edge TLS Routes, Kustomize prod overlay, source-to-image, quota guardrail, RBAC onboarding, SCC, network isolation, operator install, HTPasswd onboarding, production network policies and an EX280 speed run. Labs 7/8/11 grade state-equivalent RBAC/password/SCC fixes, lab 12 seeds a real TLS fixture (its passthrough serves 200, while a plain-HTTP backend fails the handshake) and teaches the immutable-Job delete/recreate (or LimitRange) flow, and the lab 1/3/8 hints match the real Route admission, certificate and ServiceAccount behavior.
- EX280 Prep mode - objective tracker for the official EX280V422 objective groups (with Red Hat
docs links), quick command labs, a searchable
occheat sheet, exam facts and an evidence-based study-coverage verdict (mastery requires the bullet’s own commands inside the linked drill run). - Chaos mode - endless incident survival; Sandbox - free build scored 0-100 by the Architecture Advisor (HA, security, networking, storage, scaling, RBAC, quotas, probes, config, cost).
- A start page worth opening - a dark, premium landing screen with the live 3D cluster in the
hero: a banded mode launcher, an interactive EX280 objective→drill picker built from the real
objective data, real product captures, and a light colophon that credits K8s Games by Rohit
(Apache-2.0), the project this one is a derivative of. Native page scrolling, 390px responsive,
prefers-reduced-motionsupport and a progress-aware primary action. - 3D cluster view - Three.js r152 diorama with redesigned models for all 47 kinds, studio lighting, floor tier zones, camera fit-to-view, a screen-space label layer with collision culling, ownership/network/traffic connections, project floor planes, minimap, inspector, live metrics and an importable/exportable YAML view, and a visible paused state (HUD chip + pause glyph) whenever an overlay freezes the run. The renderer boots as the start-page hero and expands to full screen when a mode launches.
- OpenShift Draw (
/draw) - a 3D architecture whiteboard: drag OpenShift resources, draw connections, edit properties, auto-layout, export YAML/PNG and share via URL.
- Pick a mode on the start page - Campaign, Timed Labs, Sandbox, Chaos - or open EX280 Prep for the objective tracker. The start page also maps each EX280 objective group to the campaign level and timed lab that drills it. New to the Linux command line? The campaign starts with the optional Linux & RHEL Foundations prelude (0.1-0.7); if you already live in a RHEL shell, Level 1 is open from the start.
- Create resources from the left palette or with the
octerminal (/). The name prompt lets you match the objective name (e.g.frontend). - Click a resource to inspect it (Overview / YAML / Describe / Events); right-click for actions.
- Press
/to open theocterminal,Ffor the Files panel,?for EX280 Prep,Hfor help. In the campaign, read the mission card: its project row says where to work, Next step shows what is next, and Show me how gives the exact commands (click one to put it in the terminal, then press Enter). - Handle incidents as they appear - diagnose with
oc describe,oc logs,oc get events- and resolve them by fixing the resource state.
| Input | Action |
|---|---|
/ |
oc command bar |
F |
Files / manifests panel (apply -f, apply -k, templates, Helm; New/Import/Delete files) |
? |
EX280 Prep (objective tracker, quick labs, cheat sheet) |
H |
Help / How to play |
Space |
Pause / resume the simulation |
M |
Metrics dashboard |
Esc |
Close the topmost panel/overlay; with nothing open, open the pause menu (Resume, Restart level, Main menu) |
1-9 |
Quick-place a resource from the palette (opens the name prompt) |
| Left-click | Select resource |
| Left-drag | Move resource or rotate camera |
| Right-click | Context menu (Expose, Start Build, Scale, Rollout, Delete, …) |
| Right-drag | Pan camera |
| Scroll | Zoom |
Bottom toolbar: Auto-align | Reset view | Terminal | Files | YAML | EX280 | Help.
The game is organised around the official EX280V422 objective groups. Campaign level data carries
examObjectives and docsLinks; challenges and quick labs carry docs links; the in-app EX280 Prep
tracker maps every bullet to its campaign levels, timed labs and quick labs. The table below is
derived from that same data (9 objective groups, 55 bullets; timed labs are the 13 challenge drills):
| EX280V422 objective group | Where it is practiced | Levels | Timed labs | Quick lab |
|---|---|---|---|---|
Manage OpenShift Container Platform (projects, oc CLI, querying, health) |
Campaign ch. 1, Sandbox, oc terminal |
1, 2, 3, 4, 9, 10, 15, 21, 22 | 1, 2, 4, 5, 9, 12 | Projects & the oc CLI |
| Declarative Resource Management (manifests, Kustomize, secrets/configmaps) | Files panel, oc apply -f/-k, oc process |
4, 5, 6, 7, 10, 11 | 1, 3, 4, 10, 11, 12 | Kustomize prod overlay |
| Deploy applications (templates, Helm, builds, rollouts, Services, Routes) | Builds/ImageStream controllers, oc new-app, oc start-build, oc rollout |
2, 8, 10, 11, 12, 20 | 1, 3, 5, 9, 10, 12 | Route with edge TLS |
| Manage authentication and authorization (HTPasswd, users, groups, RBAC) | oc login, users/groups, oc adm policy, oc auth can-i |
13, 14, 16 | 7, 11, 12 | HTPasswd users & login |
| Configure network security (Services, Routes, NetworkPolicy, TLS) | Route admission, TLS edge/passthrough/reencrypt, service serving certs, default-deny + router-ingress NetworkPolicy | 11, 12, 20, 22 | 3, 8, 9, 10, 13 | Service serving cert → reencrypt Route |
| Expose non-HTTP/SNI Applications | passthrough Routes, LoadBalancer services | 12 | 10 | - |
| Enable developer self-service (quotas, LimitRanges, project templates) | ResourceQuota/LimitRange/ClusterResourceQuota; project-request template flow (generate → edit → create in openshift-config → point project.config.openshift.io/cluster at it) |
15, 22 | 6, 12 | ResourceQuota + LimitRange |
| Manage OpenShift operators (OLM) | Subscription → CSV phases, operator effects, uninstall | 18 | 10 | Install an operator |
| Configure application security (SCCs, service accounts, jobs/cronjobs) | SCC grants/denials, oc set serviceaccount, oc debug, Jobs/CronJobs |
7, 16, 17, 19, 22 | 3, 7, 8, 12 | Service account + SCC grant; Job + CronJob |
Level numbers are the in-game display numbers (Level 20 is Network Policies, Level 21 Nodes &
Cluster Troubleshooting, Level 22 the capstone). Cross-cutting drills (not a separate V422 objective
group): troubleshooting/operations content - logs, events, oc adm top, oc adm cordon|drain,
oc debug node, Jobs/CronJobs - appears in levels 3, 19, 21 and 22 and timed labs 2, 6, 12, plus
Chaos mode. The Linux & RHEL Foundations prelude (0.1-0.7) drills the workstation skills these
objectives assume.
Official sources (verify against these before your exam):
- EX280 exam page: https://www.redhat.com/en/services/training/red-hat-certified-openshift-administrator-exam
- OpenShift Container Platform 4.22 documentation: https://docs.redhat.com/en/documentation/openshift_container_platform/4.22
- OpenShift 4.22 release notes: https://docs.redhat.com/en/documentation/openshift_container_platform/4.22/html/release_notes
shifting.pro is a study aid, not an exam dump. It simplifies cluster behaviour for teaching, and it is not affiliated with or endorsed by Red Hat.
There is no build step and no runtime dependencies to install. The app is plain ES modules
served statically (Three.js and Tailwind are loaded from CDNs; an internet connection is required
for those, and a WebGL-capable browser is required for the 3D view). The only optional npm
install is the dev-only formatting/linting toolchain - see
Development tooling.
git clone https://github.com/G-man3207/openshiftgames.git
cd openshiftgames
node scripts/dev-server.mjs
# open http://localhost:8080 for the game and http://localhost:8080/draw for the whiteboardnpm run dev is the same zero-dependency server (Node built-ins only; port 8080, or pass
--port 9000). It emulates the production Vercel rewrites - the /draw page and the versioned
/v/<version>/js/* module prefix - with Cache-Control: no-store, so the ES-module graph loads
exactly as deployed and local edits are never stale.
Serve over HTTP (not
file://) - ES modules, import maps and the optional Files/EX280 Prep panels rely on HTTP semantics. A plainpython3 -m http.servercannot resolve the import-map URLs:/v/<version>/js/*is a production rewrite, not a directory on disk, so the module requests would 404.
The game itself stays no-build and dependency-free; the repo carries an optional, dev-only
toolchain (Prettier 3, Knip 5, husky 9) that keeps the tree clean. It is not needed to play, serve
or deploy the app, and node_modules/ is gitignored.
npm install # dev tooling only - no runtime dependencies
npm run dev # zero-dependency local server: /draw + versioned /v/ module rewrite (no-store)
npm run format # Prettier: rewrite all text files
npm run format:check # Prettier: strict check (same as the pre-commit gate)
npm run knip # Knip: dead files, unused/unlisted deps, unresolved imports, binaries
npm run lint # both gates-
Release bump rule - a release bumps
package.jsonand the import-map module prefixes inindex.html+draw.htmltogether ("./js/": "./v/<pkg.version>/js/").scripts/verify.mjsfails the fast gate when the three drift, and the browser sections prove the versioned URLs load. -
Prettier is strict and repo-wide (
.prettierrc: single quotes, semicolons, print width 100, ES5 trailing commas)..prettierignoreskipsnode_modules/, binary assets,docs/screenshots/, vendored skill assets (.taste-skill/,.agents/) and.husky/. -
Knip runs in "real garbage only" mode (
knip.jsonc): dead files, unused/unlisted dependencies, unresolved imports and binaries are errors; unused-export, duplicate-export, type and enum-member rules are off by design. This is a no-build vanilla-ESM app: much of its API is consumed from the inline<script type="module">blocks inindex.html/draw.html, from thewindow.gameruntime and from browser-context dynamic imports, which static analysis cannot trace - enforcing unused-export strictness would mean deleting legitimate, used code, so that cleanup is deliberately deferred.scripts/knip-entrypoints.mjsmirrors every import of the inline HTML modules so dead-file detection still works, andscripts/verify.mjsfails if the mirror drifts.three(browser import map) andpuppeteer-core(installed out-of-repo for the QA harnesses) are explicitly ignored, as is/js/*for the browser-context import inscripts/browser-test.mjs. -
husky pre-commit gate (
.husky/pre-commit) runsprettier --check .andnpx knipand blocks the commit on any issue; ifnode_modules/is missing it tells you to runnpm install. Keep the toolchain toprettier,knipandhusky:scripts/verify.mjsenforces that the onlydevDependenciesare those three, with no runtimedependenciesand no bundler/build config.
The repo is a static site and deploys to any static host. The canonical demo is https://shifting.pro (hosted on Vercel; point the domain at the Vercel project):
- Canonical URL:
https://shifting.pro(used inog:url/og:image/ canonical tags; updateindex.html,draw.html,og-image.htmland README if the domain changes). vercel.jsonkeeps the/draw→/draw.htmlrewrite, the/v/:version/js/:path*→/js/:path*rewrite, cache headers and bot policy. No server code and no environment variables are required. Because the site has no build step, release delivery of the module graph uses versioned URLs: every release bumps the import-map module prefix inindex.htmlanddraw.html("./js/": "./v/<package.json version>/js/") together withpackage.json, and the whole ES-module graph then loads from/v/<version>/js/...servedCache-Control: public, max-age=31536000, immutable- a returning visitor's old cache entries are never requested by the new entry document.index.html, the import map, CSS,/and/drawstill revalidate (public, max-age=0, must-revalidate), so a direct unversioned request always sees the current release; only images/fonts keep the long-livedmax-age=86400, s-maxage=604800, stale-while-revalidate=2592000cache.scripts/verify.mjsfails whenpackage.jsonand either import map drift. Any deploy that changes a file underjs/or the entry documents must bump the version (and both import maps) in the same commit - an unchanged/v/<version>/prefix can keep serving a whole year of old cached modules.- Deploy with the Vercel dashboard (import the repo), add
shifting.proas the project domain, or usevercel deploy(Vercel CLI, optional).
- No build step - vanilla ES modules + import maps. No bundler, no framework, no npm runtime
dependencies. Each entry page's import map resolves
./js/to a versioned/v/<package.json version>/js/prefix thatvercel.jsonrewrites to the real/js/tree and servesimmutable, so a release changes every module URL without a build step (and the revalidated entry document can never boot a module graph cached from an older release). Tailwind is loaded from the CDN for utility classes; custom styling lives instyle.css(in-game shell) andstart.css(start page). Type is set in Schibsted Grotesk with JetBrains Mono reserved for real terminal/YAML code. An optional Prettier + Knip + husky dev toolchain guards repo hygiene - see Development tooling. - Three.js r152 (
https://unpkg.com/three@0.152.0/..., pinned in the import map) powers the 3D cluster view and the whiteboard; WebGL is required. - Central kind registry -
js/data/kindRegistry.jsis the single source of truth for kinds, apiVersions, scope, aliases,oc getcolumns, palette metadata, default specs, owner chains, explain text, minimap sizes and the shared manifest serializer. Adding a kind means adding one registry entry + a resource class + a mesh creator (+ optional controller/data changes). - Simulation -
js/engine/SimulationTick.jsruns fixed-step OpenShift controllers: scheduler (real retry backoff + unschedulable-PodPendingdisplay), Pod lifecycle, Deployments/ReplicaSets, StatefulSets, DaemonSets, Jobs/CronJobs, HPA, PVC binding, Pod-network IP allocation inside each Node'sspec.podCIDR(OpenShift's10.128.0.0/14clusterNetwork,/23per node), Services and endpoints (API-server ClusterIP defaulting, named/numeric targetPort resolution, one canonical ready-endpoint resolver, the seeded platform DNS/router Services), NetworkPolicy evaluation shared by the traffic simulation and the cluster-DNS diagnostics, Routes (admission), BuildConfig→Build→ImageStream, Subscriptions/operators, MetalLB LoadBalancer IPs, SCC enforcement, LimitRange/ResourceQuota/ClusterResourceQuota, OAuth/HTPasswd, ClusterOperator health. - Command layer -
js/ui/CommandBar.jsparsesoccommands (with pipes/redirects and tab completion), runs them through the RBAC authorization gate and drives the same engine APIs as the UI;js/ui/ShellTools.jsis the simulated shell (heredocs, redirects, cat/echo/grep/tail/head/wc/ vim/openssl/skopeo/htpasswd/…) that shares the ManifestEngine virtual filesystem. Cluster-DNS diagnostics (nslookup/dig/getent hosts) and verbosecurluse the engine's NetworkPolicy evaluation, andoc exec/oc rshpropagate remote exit statuses (unmodelled binaries terminate with exit code 127). - RHEL workstation -
js/engine/LinuxHost.jsis the workstation model (filesystem metadata overlay on the ManifestEngine store, users, processes, systemd + journal, packages, network, firewalld, SELinux, rootless podman) withreset(seed)for a level'slinuxHostseed;js/ui/LinuxCommands.jsmixes its ~110 commands into ShellTools. One cwd-aware path resolver serves the shell andocfile arguments, and the Linux objective types (file,service,process,package,firewall,container,selinux,linuxUser,cwd,env,ran) read host state injs/engine/ObjectiveEvaluator.js. - Campaign structure -
js/data/CampaignLevels.jsowns campaign order (CHAPTERS), unlocking (isLevelUnlocked), the Continue rule (continueLevelId) and display numbers (levelLabel). Level ids are stable save keys: never renumber a level; add a new id and place it inCHAPTERS. - Declarative engine -
js/engine/YamlParser.js+js/engine/ManifestEngine.jsprovide a virtual file system, YAML manifests, Kustomize overlays, OpenShift templates, Helm charts and HTPasswd file operations;js/ui/ManifestPanel.jsis the Files panel. - UI - StartMenu (the start page: live diorama stage, mode launcher, EX280 objective picker,
scroll reveals), HUD, Inspector, ContextMenu, Metrics, Settings, Incidents, Minimap, docked
octerminal and pause overlay;index.htmlwires the shell and the palette (generated from the registry);draw.htmlis the whiteboard. - Data -
js/data/CampaignLevels.js(EX280 levels + campaign structure),js/data/LinuxLevels.js(the Linux & RHEL Foundations prelude),js/data/IncidentDefs.js,js/data/Achievements.js,js/data/EX280Objectives.js,js/data/ManifestFiles.js.
Repo layout:
index.html start page markup + game shell (inline ES module)
draw.html OpenShift Draw whiteboard
style.css in-game design system and shell styling
start.css start page styling (dark editorial, one accent)
img/start/ product captures used by the start page
js/data/ kind registry, campaign levels (+ Linux prelude), incidents, achievements,
EX280 objectives, manifest files
js/engine/ GameEngine, ClusterState, SimulationTick, IncidentEngine, ScoringEngine,
YamlParser, ManifestEngine, ObjectiveEvaluator, RBAC helpers, LinuxHost
js/resources/ ResourceBase + per-kind classes (OpenShift + legacy)
js/rendering/ Three.js renderer, meshes, connection lines, particles, label layer
js/ui/ CommandBar, ShellTools, LinuxCommands, HUD, Inspector, ContextMenu, Metrics,
IncidentPanel, Minimap, Settings, ManifestPanel, StartMenu
js/modes/ CampaignMode, ChallengeMode, ChaosMode, SandboxMode, ExamPrepMode
scripts/ verify / browser-test / visual-qa harnesses, local dev server, Knip entry mirror
vercel.json static hosting: /draw + versioned /v/ module rewrites, cache/bot headers
og-image.html OpenGraph card source (1200×630) - PNG regenerated by the verifier
package.json dev-only toolchain (Prettier/Knip/husky) + dev server script; no runtime deps
knip.jsonc Knip "real garbage only" configuration (documented rules)
.husky/pre-commit pre-commit gate: prettier --check + knip
This is an education project. Useful contributions:
- Fix an inaccurate
occommand/flag (cite the OpenShift 4.22 docs page in the PR). - Add a campaign level, timed lab or incident aligned to an EX280 objective.
- Improve an EX280 Prep objective mapping or quick lab.
- Report a bug with browser, mode, level and the exact
occommand/output.
Please keep the no-build, dependency-free architecture (scripts/verify.mjs checks syntax, imports,
registry consistency and forbidden legacy strings). The app must not gain runtime dependencies or a
bundler; dev tooling is limited to Prettier, Knip and husky behind the pre-commit gate (see
Development tooling).
shifting.pro is a derivative work of K8s Games by Rohit (rohitg00), used under the
Apache License 2.0.
- Original project: https://github.com/rohitg00/k8sgames
- Original copyright: K8s Games, Copyright (c) Rohit G / rohitg00, Apache-2.0.
- See the NOTICE file for the derivative-work statement and full attribution.
This project is licensed under the Apache License 2.0 - see LICENSE. The original license and copyright notices are preserved. It is an independent, unofficial study project: it is not affiliated with, sponsored or endorsed by Red Hat, Inc. OpenShift, Red Hat and the Red Hat logo are trademarks of Red Hat, Inc. Kubernetes is a trademark of The Linux Foundation.
