Release notes
Every release of vurulkan/kubernetes-openconsole, newest first. The image of a release is ghcr.io/vurulkan/kubernetes-openconsole:X.Y.Z; read the notes of every version you skip before upgrading (Upgrading).
v2.23.4
2026-11-17 · on GitHub
What's new
Events follow the object live. Asked for while testing 2.23.3. The Events of a pod or workload drawer used to be read again only when the object itself changed, so a repeated Warning or a growing BackOff count showed only after reopening.
- The pod and workload (Deployment, StatefulSet, DaemonSet) drawers, and the list behind a row's Events button, read the last hour of events and then follow the object's new events as they happen. They use the namespace's
/ws/eventsstream, the one Live events uses, filtered to that object; permissions are checked by the server as before. - A new event is added, briefly highlighted and announced to screen readers. A repeat updates its row in place (
×5and its time). - ● Live / Reconnecting… shows the stream state. After a dropped connection, the list is read again to fill the gap.
- The row Events dialog shows the same list as the drawers (warnings first, count, age) instead of raw JSON.
/ws/eventsframes of Kubernetes events carrycount. This is an addition only; the OpenAPI description lists the frames.
Upgrading
Replace the image: there is no change to the API's paths, the database, the manifests or deploy/elk. Rolling back to 2.23.3 works.
Tested on Zenia nonprod with the :dev-194 image and a throwaway namespace:
- a new event on a pod appeared and was announced without reopening the drawer, and its repeat became
×4in the same row; - scaling a Deployment made
ScalingReplicaSetappear in its open drawer; - the Events dialog listed the events live;
- the API checklist and the browser checks (124: light / dark, EN / TR, keyboard, axe) passed;
- this image passed a Docker smoke test.
Image
ghcr.io/vurulkan/kubernetes-openconsole:2.23.4 (= :latest, same digest sha256:80afd955…; amd64 + arm64), with signed provenance and SBOM:
gh attestation verify oci://ghcr.io/vurulkan/kubernetes-openconsole:2.23.4 --repo vurulkan/kubernetes-openconsole
🤖 Generated with Claude Code
v2.23.3
2026-11-17 · on GitHub
What's new
Two small fixes found while testing 2.23.2 on a real cluster.
- Logs and Shell have their own icons. Every Logs button (pods, deployments and jobs, in the list, the cards and the drawers) drew the same terminal icon as Shell. Logs now use a page-of-text icon.
- An object without events says so. A list answer with nothing in it was
{"items": null}, so the pod and deployment drawers kept spinning instead of saying "No events in the last hour". Empty lists are now[]in every list answer. The event reads also report a cluster refusal ascluster_forbiddeninstead of a bare500.
Upgrading
Replace the image: there is no change to the API's paths, the database, the manifests or deploy/elk. Rolling back to 2.23.2 works.
Tested on Zenia nonprod with the :dev-190 image:
- the two fixes, in the browser;
- the API checklist (43 steps);
- the browser checks (124: light / dark, EN / TR, keyboard, axe);
- a Docker smoke test of this image.
Image
ghcr.io/vurulkan/kubernetes-openconsole:2.23.3 (= :latest, same digest sha256:b68e17f3…; amd64 + arm64), with signed provenance and SBOM:
gh attestation verify oci://ghcr.io/vurulkan/kubernetes-openconsole:2.23.3 --repo vurulkan/kubernetes-openconsole
🤖 Generated with Claude Code
v2.23.2
2026-11-16 · on GitHub
What's new
Three small fixes found while testing 2.23.1 on a real cluster.
- Refused WebSockets leave no trace. With
WS_ORIGIN_CHECK=enforce, a pod log socket opened from another site was refused but still audited aspod.logssuccess, and a refused shell left an empty recording. The log socket is now audited once it is open. A cross-site shell is refused before anything starts: it is audited aspod.execdenied, reasonorigin_not_allowed, and leaves no recording. Nothing changes with the defaultwarn. - One date format. Dates used to follow the browser's region: a Turkish UI on a US-region browser showed
11/16/2026, 11:37:51 PM. Every date and time is now2026-11-16 23:37:51: ISO 8601, 24 h, in the browser's time zone, like the audit log. Admin → Backups says its schedule in words ("A backup every 24 hours" / "24 saatte bir yedek alınır"). - Wide tables keep their actions in reach. At 1280–1440 px the Dashboard tables are wider than their box. The row actions and the column picker (⋮) used to sit past the right edge, and reaching them scrolled the Name column away. The actions column now stays at the right edge.
Upgrading
Replace the image: there is no change to the API, the database, the manifests or deploy/elk. Rolling back to 2.23.1 works. Coming from 2.23.0, read Upgrading to 2.23.1 first.
Tested on Zenia nonprod: the :dev-183 image, then this one.
- The upgrade from 2.23.1 kept the encryption key in its Secret and
WS_ORIGIN_CHECK=enforce. - Cross-site log and shell sockets were refused without a trace.
- The API checklist (43 steps) and the browser checks (124: light / dark, EN / TR, keyboard, axe) passed.
- The image passed a Docker smoke test.
Image
ghcr.io/vurulkan/kubernetes-openconsole:2.23.2 (= :latest, same digest sha256:182d590f…; amd64 + arm64), with signed provenance and SBOM:
gh attestation verify oci://ghcr.io/vurulkan/kubernetes-openconsole:2.23.2 --repo vurulkan/kubernetes-openconsole
🤖 Generated with Claude Code
v2.23.1
2026-11-16 · on GitHub
What's new
2.23.1 is a patch release of fixes, not features: security hardening, supply chain, data safety, cluster correctness, frontend reliability and code health. The product is now called OpenConsole for Kubernetes; the image, repository, namespace and manifests keep their names.
Security
- The client address comes from
X-Forwarded-Foronly when the request arrives fromTRUSTED_PROXIES. - Kubeconfigs that run a program or read a file are refused.
- A local account never falls back to LDAP.
- StartTLS verifies the directory's certificate, with a CA field in Admin → LDAP.
- Inactive accounts, and accounts that must change their password, can't open shells.
- Request bodies are size-limited, and CSV exports are formula-safe.
- CORS only for
ALLOWED_ORIGINS; cross-site WebSockets are logged (WS_ORIGIN_CHECK=warn, orenforceto refuse them).
Image and supply chain
- Distroless image with a read-only root filesystem.
- Monaco editor bundled; the page no longer loads anything from a CDN.
- govulncheck,
npm auditand Trivy run in CI. - Signed provenance and SBOM are made with
actions/attest.
Data
- Database copies: daily, before an upgrade, and on demand (Admin → Backups).
- The encryption key of the stored credentials can live in a Secret (
ENCRYPTION_KEY_FILE, with rotation). - SQLite runs in WAL mode with foreign keys and migrations in one transaction.
- Graceful shutdown audits open shells.
deploy/usesstrategy: Recreate.
Clusters and streams
- Each browser tab keeps its own cluster (
X-OpenConsole-Cluster). - Informer caches are used only while synced; cluster errors keep their meaning (
cluster_forbidden,cluster_timeout, …). - Log sockets have limits and resume.
- Followed logs no longer stop after 30 s.
- The session timeout is now an inactivity timeout (12 h at most).
Frontend
- Errors are shown as sentences in English and Turkish, never raw JSON.
- Focus is trapped in dialogs and returns to the opener; the shell is keyboard-reachable (
Ctrl+]). - Smaller bundle.
- The sign-in panel is readable in the light theme.
API and audit
- Every error is RFC 9457 problem details (
application/problem+json) with a reason code inerrorand the English message indetail. - Unknown
/apipaths answer404JSON. - Every admin change is audited for every answer (success, denied, failed).
- YAML views are audited reads, and invalid cluster-write requests are audited as
failed/invalid. - A Kubernetes refusal of OpenConsole's own account on the REST log is
403 cluster_forbidden.
Upgrading
Read Upgrading to 2.23.1 before you install. In short:
- Kubeconfig clusters: clusters added with a kubeconfig that uses
exec(kubelogin),auth-provideror file paths no longer connect. Replace those kubeconfigs first. - Your own manifest: add
TRUSTED_PROXIES,strategy: Recreate,terminationGracePeriodSeconds: 30, and a/tmpemptyDir if you turn onreadOnlyRootFilesystem. - No shell in the image:
kubectl execinto the pod no longer works; usekubectl debug. - API errors: scripts that compare the API's
errortext must switch to the codes. The/api/admin/cluster*endpoints are removed; use/api/admin/clusters. - Database: the first start copies the database to
/data/backups/…-pre-upgrade.db. Backups need about 7 × the database size on the volume. - Settings to review: the session minutes are now an inactivity timeout, and
LOG_RETENTION_DAYS=0now means "keep forever". In Admin → LDAP, put one base DN per line.
Tested on dev, on a k3d cluster and on a real AKS cluster upgraded from 2.23.0 (:dev-173 … :dev-179, then this image). The upgrade kept all data, sessions and API keys. LDAP and Azure AD worked, and so did role-scoped users, scale, restart, YAML, shell, logs, the encryption key moved into a Secret, and WS_ORIGIN_CHECK=enforce. The image passed a Docker smoke test.
Image
ghcr.io/vurulkan/kubernetes-openconsole:2.23.1 (= :latest, same digest sha256:66321e1c…; amd64 + arm64), with signed provenance and SBOM:
gh attestation verify oci://ghcr.io/vurulkan/kubernetes-openconsole:2.23.1 --repo vurulkan/kubernetes-openconsole
🤖 Generated with Claude Code
v2.23.0
2026-11-14 · on GitHub
What's new
Documentation site, in English and Turkish: https://k8sopenconsole.github.io/ (Türkçe)
- The manual moved from the README to its own site: getting started, installation, user guide, administration, API and agents, operations (logging, Elastic, metrics and alerts, tracing, security model, hardening, air-gapped, troubleshooting) and release notes. Every command on it was run on a real cluster.
- Every page in Turkish, in the words of the app's Turkish UI. Switching the language keeps you on the same page and section.
- Reference pages generated from the code, so they can't drift: environment variables, application permissions, Kubernetes permissions (ClusterRole rule by rule), audit events, metrics, keyboard shortcuts, and one page per API endpoint from the OpenAPI description.
- Versioned per minor release from 2.23 on.
In the app
- The sign-in page footer reads Kubernetes OpenConsole · GitHub | Docs; a book icon in the nav rail (above GitHub) opens the docs. Both open the site in the UI's language.
- The keyboard cheat sheet (
?) lists the Saved Views keys (arrows, Enter, 1–9, Esc) andEscto clear a table selection.
For contributors
backend/cmd/docgenwrites the reference pages (English and Turkish) from the registries next to the code (config.Vars,rbac.ActionDocs,audit.*Docs,api.Metrics, the comments indeploy/rbac*.yaml,frontend/src/shortcuts.json); CI fails when they are out of date.- The README is now a short front door;
CLAUDE.mdasks every app change to update the docs in both languages.
Upgrading
- No change to the server, the API, the database or the ClusterRole: replace the image. Rolling back to 2.22.0 works.
- Bookmarks into the old README: the same sections are on the site; 2.22.0's README stays at the v2.22.0 tag.
Tested: the image on Docker (probes, first sign-in and password change, the docs links in English and Turkish) and on a k3d cluster installed with the docs' kustomize steps (?ref=v2.23.0).
Image
ghcr.io/vurulkan/kubernetes-openconsole:2.23.0 (= :latest, same digest sha256:f4877e65…; amd64 + arm64), with signed provenance and SBOM:
gh attestation verify oci://ghcr.io/vurulkan/kubernetes-openconsole:2.23.0 --repo vurulkan/kubernetes-openconsole
🤖 Generated with Claude Code
v2.22.0
2026-11-13 · on GitHub
What's new
API keys (account menu → API keys, ⌘K → Go to API keys)
- Keys (
oc_…) for Claude Code, Cursor or your own scripts, calling the API in your name. Scope: all clusters or one, all your namespaces (including later ones) or a selection, and a permission matrix with templates (Read-only (safe), Read-only (all), All my permissions, Custom). A key never does more than its owner holds at the time of the call. - Expiry 7 / 30 / 60 / 90 days, a date or none; edit, regenerate (new secret and expiry), revoke, delete. The key is shown once and stored as a SHA-256 hash.
X-OpenConsole-Cluster: <name or id>picks the cluster per request; WebSockets take a single-use ticket fromPOST /api/ws-ticket. 10 requests / s per key (burst 30), then 429.- A key can never manage keys, change the password, switch your cluster or save views (403
api_key_not_allowed). Admin endpoints need an admin's key with the Admin API box.
Admin policy
- Per role, how much of it a key may get: Off, Safe read-only (default, also for existing roles) or All of the role's permissions. With several roles each goes through its own mode, then they add up (exec in
ns1, read only inns2). - Per user override (By roles / Off / Safe read-only / All of their permissions, never more than the web), with a preview of what the user's keys may reach and Revoke all their keys.
- The policy never changes a key: a blocked cell stops working and works again when allowed (403
api_access_policy;api_access_offwhen nothing is left). - Expiry policy (Admin → Session): longest lifetime, keys without expiry allowed or not; existing keys are left alone and listed as Outside the expiry policy.
- Admin → API keys: every user's keys, filters (owner, status, includes exec, Admin API, outside policy), revoke with a reason, CSV.
API for agents
- OpenAPI 3.1 at
GET /api/openapi.json(no sign-in, nothing secret in it); a test fails the build when a route is missing from it. - API keys → Documentation: quick start, a text to hand your agent (the key stays in
OPENCONSOLE_API_KEY), every endpoint a key can call by tag with its permission, parameters, acurlexample and example responses, greyed out where your keys can't reach it (You do not have it, Policy does not allow, Switched off), and the errors. POST /api/namespaces/{ns}/pods/{name}/exec: one command (argv, never a shell), no stdin, 1 MiB per stream, 1–120 s;{stdout, stderr, exitCode, truncated, durationMs}. Samepods:exec, feature flag, shell limit and recorder as the shell; recorded with a command badge; the audit log keeps only the program (command[0]), never the command line.GET /api/namespaces/{ns}/pods/{name}/logs: a pod's log astext/plain(tailLinesup to 10 000,sinceSeconds,previous,timestamps; 5 MiB,X-OpenConsole-Truncated).
Audit and logs
- Every action and resource read made with a key carries
api_key_id/api_key_name(Audit Logs: Source = Web / API key, key name; the CSV hasapi_key_name). Successful reads with a key are keptAPI_KEY_READ_RETENTION_DAYS(7). A shell or command run with a key is always recorded, even with recording off. http.requestlog lines also carry the key (openconsole.api_key.*in ECS), refused requests and non-audited calls (/api/auth/me,/api/ws-ticket) included, soopenconsole.api_key.id : *separates API from web in Kibana.- 403 bodies name the reason:
forbidden,api_access_policy,api_access_off,api_key_not_allowed,feature_disabled,admin_only.
Also in this release (2.21.1, not released on its own)
- Any 401 to a signed-in user returns to the sign-in page with "your session ended"; an idle screen does too when the token expires. 401 now only means that: a wrong current password is 400, a user who must change the password gets 403.
Upgrading
- Database: tables and columns are added on the first start. Rolling back to 2.21 works: the old image opens the new schema (keys simply don't exist there). Back up
/data/app.dbfirst anyway. - API key policy: every existing role starts at Safe read-only. Open a role (Admin → Roles → API access) for keys to exec, edit, scale, restart or read ConfigMaps / Secrets.
- Elastic: re-run
deploy/elk/install.shonce (ten new mapped fields; the pipeline is unchanged). On Elasticsearch 8.7 followdeploy/elk/README.md→ Elasticsearch 8.7 and older (by hand, an ILM policy,PUT <ds>/_mappingfor the live index). - Scripts that read error texts: 403 bodies are now reason codes; the audit CSV has one more (last) column.
- ClusterRole: unchanged (exec and logs were already granted).
Tested on a nonprod AKS cluster (dev images): role modes and every override value, the expiry policy, admin revoke and CSV, one-command exec (output, exit codes, timeout, 1 MiB, limits, recording, no command text in the audit log, the console log or Elasticsearch), REST log limits and previous, WebSocket tickets, a key shell recorded with recording off, 401 / 403 / 429, the Documentation tab as a user and as an admin, and the key on audit events and request lines in Kibana.
Image
ghcr.io/vurulkan/kubernetes-openconsole:2.22.0 (= :latest, same digest sha256:f4d193cf…; amd64 + arm64), with signed provenance and SBOM:
gh attestation verify oci://ghcr.io/vurulkan/kubernetes-openconsole:2.22.0 --repo vurulkan/kubernetes-openconsole
🤖 Generated with Claude Code
v2.21.0
2026-10-27 · on GitHub
What's new
Column picker (Dashboard, list view)
- A ⋮ at the right end of the header row shows or hides the columns of the current tab. Name always stays; Reset to default shows them all again.
- The choice is kept per tab in the browser, so it survives a reload but doesn't follow you to another browser. Columns added in later versions show by default.
- Keyboard: Tab reaches ⋮, arrow keys move, Space toggles, Esc closes and returns focus to the button.
Pods: Containers and Controlled By
- Containers: one square per container (the pod's containers, then its init containers). Filled green = running and ready (amber ring when it has restarted), filled amber = not ready or starting, filled red = waiting on an error (CrashLoopBackOff, ImagePullBackOff, …), hollow gray = exited 0 (a finished init container), hollow red = exited with an error. Hovering shows the name (init containers and sidecars marked), state, start time and restarts. Sorting puts the pods with the most containers first, then the ones with more problem containers.
- Controlled By: the kind of the pod's direct controller (ReplicaSet, StatefulSet, DaemonSet, Node for a static pod, or a custom kind such as StrimziPodSet), its name on hover.
- Pod cards show the same squares for pods with two or more containers, and the controller kind.
Search every column: a plain term now matches the name or the text of any column on the tab, hidden columns too, so list and card view find the same rows: pod status, container names and controller; workload status and replicas (2/3); service type, cluster / external IPs and ports (80, 8080/TCP); ConfigMap / Secret key names and secret type; ingress hosts and paths; CronJob / Job schedule and status; HPA target and min / max. Ages, restarts and CPU / memory are not searched; label:key=value and label:key work as before.
Fixes
- A running pod with a native sidecar (init container with
restartPolicy: Always) showedInit:1/2; it now showsRunningand counts the sidecar in READY, likekubectl get pods. - Each Dashboard tab keeps its own sort; switching tabs no longer copies the sort to the next tab.
- Table headers are no longer upper case anywhere; Dashboard headers are translated in Turkish.
Measured on an AKS namespace with 113 pods: filtering while typing takes 10–19 ms per key (worst 35 ms).
Upgrading
Nothing to do: no backend, API, permission, ClusterRole or audit change. Column choices live in the browser's localStorage (dt:cols:dashboard:<tab>). Pods with native sidecars now show one more container in their READY count.
Image
ghcr.io/vurulkan/kubernetes-openconsole:2.21.0 (= :latest, same digest sha256:6d045d3c…; amd64 + arm64), with signed provenance and SBOM:
gh attestation verify oci://ghcr.io/vurulkan/kubernetes-openconsole:2.21.0 --repo vurulkan/kubernetes-openconsole
🤖 Generated with Claude Code
v2.20.0
2026-10-26 · on GitHub
What's new
Nodes (admins only, read only; left rail under Overview, g n, command palette)
- Every node of the selected cluster: status (Ready / NotReady / MemoryPressure / DiskPressure / PIDPressure, cordoned shown apart), roles, kubelet version, OS / architecture, age, pods against capacity.
- CPU and memory used as % of allocatable (amber from 70 %, red from 90 %), with requests and limits as % of allocatable, the way the scheduler counts them (init containers, sidecars, pod overhead).
- Sort by any column (CPU / memory % included), search by name, role or label (
key=value,key=for any value),j/k/ Enter. Refreshes every 15 s on its own. - Live Events in the header, for the whole cluster, on Warnings.
Node drawer (click a node, or open /nodes?node=<name>)
- Summary: OS image, kernel, container runtime, internal / external IP, scheduling (cordoned or not).
- Capacity, allocatable, requests, limits and usage for CPU, memory, pods and ephemeral storage.
- Usage history over 1 h / 6 h / 24 h against allocatable (7 d with Prometheus): the node's containers from Prometheus by the
nodeorinstancelabel, else its pods summed; without Prometheus, the in-memory samples. - Top 10 pods by CPU or memory, and every pod on the node (all, or only those needing attention); a pod opens in its drawer on the Dashboard.
- Conditions, taints, labels, annotations and events (Warnings first).
Ways in: a node in the Overview's node strip opens its drawer, "All nodes" opens the list; in a pod's drawer the node name is a link (admins).
Measured on a 4-node, ~700-pod AKS cluster: the list answers in under 3 ms from its 15 s cache (about 130 ms when it is rebuilt), a node's detail in under 10 ms.
Upgrading
- Re-apply
deploy/rbac.yaml(orrbac-readonly.yaml):nodesnow also haswatch, so nodes stay in the informer cache. Without it the Nodes screen still works; nodes are listed from the API server on each request instead (cached 15 s). - Admin only: there is no new permission to grant. A non-admin gets 403 and an audit event with outcome
denied. - New audit actions
node.listandnode.get(existing verbs, ECSevent.type: access); the page's own refreshes are not audited. - New endpoints:
GET /api/admin/nodes,GET /api/admin/nodes/{name},GET /api/admin/nodes/{name}/metrics/history?range=.
Image
ghcr.io/vurulkan/kubernetes-openconsole:2.20.0 (= :latest, same digest; amd64 + arm64), with signed provenance and SBOM:
gh attestation verify oci://ghcr.io/vurulkan/kubernetes-openconsole:2.20.0 --repo vurulkan/kubernetes-openconsole
🤖 Generated with Claude Code
v2.19.0
2026-10-26 · on GitHub
What's new
Admin → Overview (admins only; admins land here after sign-in, g o)
- The selected cluster at a glance: pods running, pods needing attention (red / amber), restarts in 24 h, ready Deployments, CPU and memory as % of allocatable, each with a 24 h trend.
- Cluster usage over 24 h against allocatable (Prometheus when configured, else the in-memory samples), pods needing attention (click opens the pod's drawer), workloads restarting most with their usual exit code (
5× 137 (OOMKilled)), restarts per hour, busiest namespaces with pods near their limits, Warning events of the last 15 minutes, node strip. - Opens in well under a second on a 700-pod cluster: the page doesn't wait for Prometheus, whose 24 h history is kept 5 minutes and refreshed in the background.
- Live Events opens right on the page, for the whole cluster, on Warnings.
Log viewer (pods, Deployments, Jobs)
- Status (following, reconnecting, ended, why) is shown apart from the log; a kubelet message such as "unable to retrieve container logs" no longer appears as a log line.
- Container picker, init containers included; a pod whose sidecar comes first (e.g.
[filebeat, app]) opens on the main container. - Live | Before last restart.
- A dropped connection (network, proxy timeout) or a container restart resumes from the last line's timestamp without losing or repeating lines; a restarted container is followed into its new instance.
- Keeps the last 10,000 lines with a virtual list; find (⌘F / Ctrl+F), pause, wrap, copy, download.
Shell: for a pod with several containers, a small menu under the Shell button: the main container first (Enter), containers that aren't running greyed out.
Live Events
- Now filtered by the user's grants: a user sees an object's changes and Kubernetes Events about it only where they may list that kind; node and other cluster-scoped events are for admins. Before, any signed-in user could follow any namespace.
- Repeats of a recurring event (a failing probe) are shown, not only its first occurrence.
- Opened from a button in the header (or
e) instead of the floating button that covered row actions.
Upgrading
- WebSocket clients: browsers need nothing. A script that opens
/ws/…?token=<JWT>must firstPOST /api/ws-ticket{"path": "/ws/…"}and use?ticket=<t>(single use, 30 s, that path only). The JWT no longer appears in any URL. - Re-apply
deploy/rbac.yaml(orrbac-readonly.yaml): new rulesnodesandmetrics.k8s.ionodesget / list, used by the Overview (node count, % of allocatable, node strip). Without them the Overview shows absolute CPU / memory and no node strip; nothing breaks. - Log streams per user: 10 at once, then one per second (was 2, then one every 5 s).
- New audit verb
view_overview(cluster.view_overview, ECSevent.type: access).
Image
ghcr.io/vurulkan/kubernetes-openconsole:2.19.0 (= :latest, same digest; amd64 + arm64), with signed provenance and SBOM:
gh attestation verify oci://ghcr.io/vurulkan/kubernetes-openconsole:2.19.0 --repo vurulkan/kubernetes-openconsole
🤖 Generated with Claude Code
v2.18.0
2026-10-26 · on GitHub
What's new
Why did it restart?
- Pod status as
kubectl get podsprints it (CrashLoopBackOff, OOMKilled, ImagePullBackOff, Init:Error, Unschedulable …), red / amber / green / gray. - The restart count shows the last exit code on hover, e.g.
137 (OOMKilled), with a plain-language explanation. - Logs: Before the last restart reads the previous container instance (
kubectl logs --previous), audited withprevious: true. - Restart history kept for 30 days (
RESTART_HISTORY_RETENTION_DAYS), also for pods that no longer exist; a crash loop is one row.
Detail drawers for pods, Deployments, StatefulSets and DaemonSets: containers, requests / limits, probes, last exit, rollout, the workload's pods, Deployment revisions (read only, no rollback), conditions, events. Workload lists flag workloads whose pods are failing.
CPU / memory
- Usage against the limit in lists, cards and drawers; CPU / Memory columns sort by share of the limit.
- 1h / 6h / 24h charts from a 30 s in-memory sampler (needs metrics-server), or up to 7 days from the cluster's Prometheus (Admin → Clusters → Prometheus): in the cluster through the API server proxy, or any Prometheus-compatible URL with basic / bearer / Azure AD / mTLS, extra headers and a multi-cluster label. Test connection explains what is wrong, step by step.
Also
- Logs: switching streams no longer shows the old stream's close / error lines; multi-container pods open on the default container.
- The dashboard opens in list view; every cluster is connected at startup.
- Accessibility: 0 axe violations on the new screens in a real browser, light and dark.
Upgrading
- Re-apply
deploy/rbac.yaml(orrbac-readonly.yaml): new rules forreplicasets(revisions) andmetrics.k8s.iopods(usage). Without them those parts are hidden, nothing breaks. - Optional: for Prometheus "in the cluster" mode, uncomment the
services/proxyrule and name your Prometheus Service. - New env var
RESTART_HISTORY_RETENTION_DAYS(default 30).
Image
ghcr.io/vurulkan/kubernetes-openconsole:2.18.0 (= :latest, same digest; amd64 + arm64), with signed provenance and SBOM:
gh attestation verify oci://ghcr.io/vurulkan/kubernetes-openconsole:2.18.0 --repo vurulkan/kubernetes-openconsole
🤖 Generated with Claude Code
v2.17.0
2026-10-15 · on GitHub
Structured audit events
Every audited action is now one event with its own fields, so you can filter on what was done instead of parsing the message:
mustafa: deployment.list success boh-cy/* on Zenia-NonProd
mustafa: deployment.restart success boh-cy/boh on Zenia-NonProd
event.action : "deployment.restart" and event.outcome : "failure"
openconsole.verb : "exec" and openconsole.cluster : "Zenia-NonProd"
openconsole.replicas_to >= 10
Audit events
actionis<resource type>.<verb>(deployment.list,deployment.restart,pod.exec,secret.reveal,user.login,permission.create). The result is a separateoutcome(success/denied/failed/rate_limited/locked) with areason(forbidden,conflict,k8s_forbidden,bad_password, …).- Details are typed fields instead of a
(k=v)suffix on the resource name:restarted_at,replicas_from/replicas_to,exit_code/bytes_in/bytes_out/duration_ms,key(never the value),dry_run/yaml_from/yaml_to,login_method,grant,changed,error, … - Every event records who and from where: source IP, user agent, account source (
local/ldap/azure), request id. - The cluster is the one the action targeted: empty for sign-ins, users, groups, roles, settings and sessions (it used to be the admin's selected cluster); a permission grant records the grant's cluster; plus
cluster_id. - Reads refused for lack of permission are audited as
denied. Lists the UI fetches on its own (the dashboard's 10-second refresh, the namespace list behind the command palette and the role editor) are not audited. - One stdout line per action. The extra
deployment.action,workload.action,secret.action,yaml.apply,recording.action,pod.exec.start/end,*.logs.start,user.password_reset,user.auto_provisionlines and thelevel=errorlines repeating a failed action are gone.level=erroris left for faults in OpenConsole itself.
ECS
event.actionwithout the outcome,event.outcome,event.typefrom the verb,event.category,event.reason;openconsole.verb,openconsole.outcome,openconsole.auth_source,openconsole.cluster_idand the details asopenconsole.<key>.
Admin → Audit Logs
- Outcome badge, Verb / Resource type / Cluster filters, expandable details per row; the CSV export has a column per field.
deploy/elk
- Mappings for every detail with its type (dates, longs, booleans), and
event.reason. - The pipeline gives 2.16 audit lines the 2.17 shape, so one query finds old and new events.
- Verified on Elasticsearch 8.15.3 and 9.1.0.
Upgrade notes
- Audit rows already in the database are converted on the first start (action, verb, outcome, singular resource type). Their resource names keep the old
(k=v)suffix until retention removes them. - Re-run
deploy/elk/install.sh. The new mapping applies from the next rollover (POST logs-openconsole.audit-default/_rolloverto apply it now). - Breaking for Kibana: saved searches, visualizations and alerts on
event.action : *.success/*.denied/login.*must move toevent.action+event.outcome.?action=deployment.restart.successstill works in the audit API.
See Audit events in the README.
v2.16.2
2026-10-13 · on GitHub
ECS logs aligned with other ECS-logging services
LOG_FORMAT=ecs now produces the same shape as other ECS-logging services on the same Elastic stack, so one Kibana query works across them:
- Nested JSON objects ("http":{"response":{"status_code":401}}) instead of dotted keys.
- HTTP request records:
- message "GET /api/namespaces 401" (was "http.request");
- event.kind / category [web] / type [access] / outcome (failure for 4xx and 5xx);
- event.duration measured in nanoseconds (was milliseconds × 10⁶, so sub-millisecond requests showed 0);
- http.response.body.bytes;
- the client address is source.ip (was client.ip).
- Audit records:
- message "admin: deployment.restart.success deployment kube-system/coredns on prod";
- event.kind and event.outcome; event.category for sign-ins (authentication) and pod shells (process).
- Every record carries event.dataset (openconsole.audit or openconsole.app) and log.logger (http or audit).
- Successful /healthz, /livez, /readyz and /metrics requests are logged at debug level only; failing ones still show up. This applies to every log format.
- deploy/deployment.yaml sets APP_ENV=prod (service.environment).
- LOG_FORMAT=json is unchanged.
deploy/elk
- The pipeline recognises both the nested 2.16.2 lines and the dotted 2.16.1 lines, so an upgrade with both versions running briefly needs no extra step. It fills event.dataset / event.outcome only where a line lacks them.
- Mappings add source.ip (ip), http.response.body.bytes (long), event.category / event.type and log.logger (keyword).
- Verified on Elasticsearch 8.15.3 and 9.1.0 with real lines from both versions.
Upgrade notes
- Re-run deploy/elk/install.sh to update the pipeline and mappings before or right after rolling out 2.16.2.
- Kibana searches on client.ip need source.ip; searches on message:"http.request" need log.logger:"http" or event.category:"web".
- Set APP_ENV to your cluster's environment if it isn't prod.
Image: ghcr.io/vurulkan/kubernetes-openconsole:2.16.2 = :latest (same digest), linux/amd64 + linux/arm64, with GitHub-attested provenance and SBOM.
🤖 Generated with Claude Code
v2.16.1
2026-10-13 · on GitHub
Logging for Elastic
- New LOG_FORMAT=ecs: JSON with Elastic Common Schema field names (@timestamp in UTC, log.level, message, service., user.name, user_agent.original, client.ip, http.request.method, url.path, http.response.status_code, http.request.id, event.action, event.duration in ns, error.message, trace.id). Everything else goes under openconsole.. So the namespace / pod / container an action ran against never collides with the shipper's kubernetes.* / container.* metadata, and nothing clashes with the ECS index templates.
- deploy/deployment.yaml now sets LOG_FORMAT=ecs. LOG_FORMAT=json keeps the previous field names for Loki / Vector or existing queries.
- Fix: client-go's own messages (klog: watch errors, throttling, reflector failures) were written as plain text, even with LOG_FORMAT=json. They now go through the same logger, so every line is one JSON object.
- deploy/elk: an Elasticsearch ingest pipeline, mappings and an
idempotent install.sh for Elastic Agent's Kubernetes integration. The
pipeline:
- decodes only OpenConsole's lines and keeps Agent's kubernetes.* / container.* / log.file.path;
- sets event.outcome on audit events;
- routes records to logs-openconsole.audit-* and logs-openconsole.app-* with typed fields (client.ip as ip, durations and status codes as long). Tested on Elasticsearch 8.15.3 with lines from this image, sent the way the Kubernetes integration sends them.
- README "Shipping logs to Elastic" has ready-to-use configurations:
- Elastic Agent with Fleet: run deploy/elk/install.sh;
- Elastic Agent standalone / Filebeat: a filestream input with the container and ndjson parsers;
- Logstash: the json filter.
Upgrade note
- If you apply deploy/deployment.yaml, the field names change to ECS. Kibana searches or dashboards built on the old names (user, msg, status, …) need the new ones (user.name, message, http.response.status_code, …). Keep LOG_FORMAT=json if you prefer the old names.
Image: ghcr.io/vurulkan/kubernetes-openconsole:2.16.1 = :latest (same digest), linux/amd64 + linux/arm64, with GitHub-attested provenance and SBOM.
🤖 Generated with Claude Code
v2.16.0
2026-10-13 · on GitHub
LDAP: accounts on first sign-in (opt-in)
- New setting in Admin → LDAP: "Create an account on a user's first sign-in". It is off by default.
- When it is on, a directory user without an account who signs in with valid credentials gets an LDAP account on the spot. The account uses the name the directory uses (typing "John" signs in to "john", no duplicate) and is added to the groups you pick, so it can see something right away.
- New accounts are never admins and are audited as user.auto_provision.success.
- Who may sign in is decided by the LDAP user filter. Add a memberOf clause to allow only one directory group.
- A disabled account stays disabled. A deleted one is created again on the next sign-in, so disable accounts to lock people out.
Fixes
- Multi-select fields: the remove button on a selected chip had no accessible name, and the chips' buttons were nested inside the field's button. The field is now a box holding separate buttons; the search box and the options are labelled and translated.
- The LDAP "Test connection" button is translated.
Upgrade notes
- Two automatic columns on ldap_config (auto_provision, auto_provision_groups). Nothing changes until you turn the setting on.
Image: ghcr.io/vurulkan/kubernetes-openconsole:2.16.0 = :latest (same digest), linux/amd64 + linux/arm64, with GitHub-attested provenance and SBOM.
🤖 Generated with Claude Code
v2.15.0
2026-10-13 · on GitHub
Security
- Password sign-in now follows the account's source. LDAP accounts are checked against the directory only, and Azure AD accounts never accept a password. Before, a local password left on such an account (for example from a pre-2.13 reset) still worked and bypassed the directory. Local accounts still fall back to LDAP, which tags users imported before 2.13 on their next sign-in.
- Successful password sign-ins are audited as login.success with the method (local / ldap). Every failed sign-in gets the same 401, so the login form reveals nothing about the account; login.failed records the reason (bad_password, unknown_user, azure_account).
Users
- Admin → Users → New user is a split button. The main part opens the
local-user form; the arrow opens a menu (when LDAP or Azure AD is
enabled):
- Import from LDAP…: the same panel as Admin → LDAP, now fully translated. Accounts that already exist are listed separately and can't be picked again. Imported users can go straight into groups, and the panel reports imported and skipped users.
- Azure AD user: pre-creates the Microsoft account by UPN or e-mail, with no password and its groups set up front.
- New users can be put into groups when they are created. Users just created or imported are listed first and highlighted.
- The admin.import audit entry names the imported users.
- README: SQL to set the source of accounts created before 2.13, which show as Local.
Azure AD sign-in
- A pre-created account that nobody has signed in to yet is also found by the token's e-mail or upn claim, or by preferred_username in another letter case. So a pre-created jane.doe@corp.com works when the UPN is jdoe@corp.onmicrosoft.com; before, a second account without groups was created.
- Local, LDAP and already used accounts are never claimed this way. The login.azure audit entry notes a claim.
- Every successful sign-in is recorded in users.last_login_at, a new column backfilled for existing accounts.
Audit
- User, Action and Namespace filters suggest the values that occur in the log as you type (GET /api/admin/audit-logs/facets). Typed text matches anywhere; a picked action matches exactly. A family row such as pod.exec.* selects the whole family.
- New Outcome filter: success / denied / failed / rate_limited.
- The CSV export uses the same filters.
- % and _ in typed filters are literal, no longer LIKE wildcards.
Version and project link
- Images carry their version, taken from the release tag. It is used in logs and traces (APP_VERSION still overrides it) and served to signed-in users at GET /api/version.
- Login page: a faint "Kubernetes OpenConsole · GitHub" line under the card. The version is not shown before login.
- Signed-in users: a GitHub icon at the bottom of the nav rail, with the version in its tooltip.
Upgrade notes
- One automatic migration: users.last_login_at (existing accounts count as having signed in).
- Accounts tagged LDAP or Azure AD can no longer sign in with a local password. Check Admin → Users → Source before upgrading if you have ever given such accounts a local password on purpose.
Image: ghcr.io/vurulkan/kubernetes-openconsole:2.15.0 = :latest (same digest), linux/amd64 + linux/arm64, with GitHub-attested provenance and SBOM.
🤖 Generated with Claude Code
v2.14.2
2026-10-10 · on GitHub
Release images only, two tags, no registry clutter. No application changes.
- Only release tags publish images. One build pushes both :X.Y.Z and :latest, so :latest is always the newest release with the same digest. (Before, main and the release tag were built separately: :latest and :2.14.1 had the same code but different digests.)
- Pushes to main are still built and tested, but no longer publish
:latest or :
. - Signed build provenance and an SPDX SBOM are now GitHub artifact
attestations (Sigstore, keyless) instead of cosign .sig tags and
in-registry attestation manifests, so the package holds only the image
tags. Verify with
gh attestation verify(README "Verifying an image").
Upgrade note: deploy/deployment.yaml uses :latest, which now means "newest release" instead of "newest main commit".
Image: ghcr.io/vurulkan/kubernetes-openconsole:2.14.2 = :latest (same digest), linux/amd64 + linux/arm64, with GitHub-attested provenance and SBOM.
🤖 Generated with Claude Code
v2.14.1
2026-10-09 · on GitHub
Fix
- /metrics: openconsole_pod_exec_sessions_total, openconsole_pod_log_streams_total and openconsole_deployment_actions_total were declared in 2.0.0 but never incremented, so they always read 0. They now count pod shells opened, log streams opened (pod, deployment and job logs) and deployment restart / scale attempts. A test guards them.
No other changes; no migration.
See v2.14.0 for everything else in 2.14.
Image: ghcr.io/vurulkan/kubernetes-openconsole:2.14.1 (published by CI from this tag), linux/amd64 + linux/arm64, cosign-signed with SBOM and provenance.
🤖 Generated with Claude Code
v2.14.0
2026-10-09 · on GitHub
2.14.0 closes the backlog: tracing, feature flags, shareable saved views, table keyboard navigation, CSV exports, a password policy, shell limits, per-cluster metrics, arm64 and signed images, an accessibility pass and a much larger test suite. Only the two on-hold security items (encryption key outside the DB, CORS allow-list) remain open.
Observability
- OpenTelemetry tracing, opt-in: set OTEL_EXPORTER_OTLP_ENDPOINT (OTLP/HTTP) and every API request becomes a span named after its route (GET /api/namespaces/{namespace}/pods) with the user, cluster and request id, and the Kubernetes API calls it made become child spans. Incoming traceparent headers are honoured. Probes, /metrics and background informer traffic are not traced. Request log lines carry trace_id. Standard OTEL_* variables (sampler, headers, service name) apply.
- New metrics: openconsole_cluster_requests_total{cluster}, openconsole_cluster_connected{cluster}, openconsole_pod_exec_active.
Operator controls
- Feature flags: FEATURE_POD_EXEC, FEATURE_YAML_EDIT, FEATURE_WORKLOAD_ACTIONS, FEATURE_SECRETS, FEATURE_SECRET_REVEAL, FEATURE_SECRET_EDIT = false switch a capability off for everyone, admins included. The buttons disappear and the endpoints answer 403 with a denied audit entry. GET /api/features shows the state.
- Password policy: PASSWORD_MIN_LENGTH (default 8) now applies to user creation and self-service change, not only to admin reset. Passwords are capped at 72 bytes (bcrypt), must not be blank, and a change must actually change the password.
- Pod shell: at most MAX_EXEC_SESSIONS_PER_USER (default 3) open shells per user, the next is refused and audited as pod.exec.rate_limited. EXEC_IDLE_TIMEOUT (default 5m) sets the idle timeout, and the shell banner shows the configured value. pod.exec.end records exit code, bytes in / out and duration; a shell ending with a non-zero status is a normal end, -1 means the session was cut.
- AZURE_AD_AUTHORITY for the US Government and China clouds.
UI
- Saved views live on the server (with their cluster), follow you to any browser and can be shared with everyone. Opening a view saved on another cluster switches to it. Views stored in the browser by older versions move to the server the first time the menu opens.
- Admin tables (Users, Groups, Roles, Sessions, Recordings): j / k to move, Enter to open, d to delete (asks first), n / p to page, Esc to clear.
- Users and Groups: Export CSV (audited as admin.export).
- Accessibility: a real-browser axe scan of every screen in light and dark theme now reports 0 issues. Muted text meets WCAG AA contrast in both themes, scrolling dialogs are keyboard-focusable, and the role matrix toggles report their state.
Fixes
- Audit log CSV export failed: the download sent no auth header.
- Azure AD: an expired ID token was not rejected.
- Restoring a saved view that changed tab or namespace lost its search.
- When the shell limit was reached, the browser only showed "connection error"; it now says why.
- Untranslated strings in the dashboard (namespace panel, search, view toggles) and the role list.
Images and supply chain
- Images are multi-arch: linux/amd64 and linux/arm64.
- Every published image is signed with cosign (keyless, GitHub OIDC) and carries an SPDX SBOM and SLSA provenance. README "Verifying an image" shows the cosign verify command.
Tests
- Store / migration tests (pre-2.x schema upgrade, idempotent migrations, encryption at rest), LDAP end to end against an in-process LDAP server, Azure AD end to end against a fake Entra endpoint, tracing, feature flags, views, exports, shell limits.
- Frontend suite (npm test, run in CI): Vitest + Testing Library + axe, EN / TR key parity and a check that every t('…') key exists.
Upgrade notes
- The saved_views table is created automatically.
- The default limit of 3 open shells per user is new; set MAX_EXEC_SESSIONS_PER_USER=0 for the old unlimited behaviour.
- Existing passwords keep working; the length rule applies to new ones.
- Nothing to change in deploy/; new env vars are listed, commented out, in deploy/deployment.yaml.
Image: ghcr.io/vurulkan/kubernetes-openconsole:2.14.0 (published by CI from this tag), linux/amd64 + linux/arm64, cosign-signed with SBOM and provenance.
Known issue: the counters openconsole_pod_exec_sessions_total, openconsole_pod_log_streams_total and openconsole_deployment_actions_total have stayed at 0 since 2.0.0. Fixed in 2.14.1.
🤖 Generated with Claude Code
v2.13.0
2026-10-09 · on GitHub
Per-user cluster
- Every user picks their own cluster in the header switcher (or with
c); other users are unaffected. The choice is stored per user and survives sessions and devices. - Users who never picked one work on the default cluster. Admins set it with Admin → Clusters → "Set as default" (formerly "Activate"; "Unset default" no longer disconnects anyone).
- The switcher lists only clusters the user's permissions cover (admins: all). A selection that is deleted or no longer allowed falls back to the default.
- One connection + informer cache per cluster, opened on first use.
- Audit entries record the cluster: new column in Admin → Audit Logs, the CSV export and the console mirror.
- /readyz checks the DB only, so one unreachable cluster no longer takes the console (and the admin screens to fix it) out of service.
Users and passwords
- Admin → Users shows each account's source: Local, LDAP or Azure AD.
- "Reset password" for local users: minimum 8 characters, forces a change at next login by default, signs the user out everywhere, and is audited as user.password_reset.{success,denied,failed}. LDAP and Azure AD users are refused, because a local password would bypass the directory.
- Local users get a "Change password" key icon in the header.
Fixes
- Cluster credentials are really validated before saving (a namespace list call). Before, only a client object was built, so a wrong server, token or CA was accepted.
- LDAP login: with a wildcard user filter such as (sAMAccountName=%s*), the login lookup could resolve a username to another user's DN (for example "jo" → john). Login now also requires an exact match on the username attribute. Exact filters are unchanged.
Docs
- README "Using OpenConsole": first login, admin checklist, users, LDAP / Active Directory (every field, test, search + import, sign-in), Azure AD (app registration, auto-created accounts, username claim), groups / roles / permissions, choosing your cluster, daily Dashboard use, admin tools.
Upgrade notes
- DB migrations run automatically: users.active_cluster_id, users.auth_source, audit_logs.cluster.
- Existing LDAP / Azure AD accounts show as "Local" until their next sign-in, which tags them. LDAP users imported from now on are tagged at import.
- The previously "active" cluster becomes the default, so nothing changes for users until they pick another cluster.
- The legacy single-cluster write endpoint (POST/PUT /api/admin/cluster) now returns 410; use /api/admin/clusters.
Image: ghcr.io/vurulkan/kubernetes-openconsole:2.13.0 (published by CI from this tag), linux/amd64.
🤖 Generated with Claude Code
v2.12.0
2026-10-09 · on GitHub
Secrets
- New Secrets tab with four permissions you can grant to any role, per
namespace and cluster (admins have all four):
- secrets:list: name, type, key count.
- secrets:get: Keys modal with type, labels, annotations, key names and sizes. Values are never sent on list or get, and kubectl's last-applied-configuration annotation is stripped.
- secrets:reveal: "Show" returns one key's value per request (POST, no-store), with copy / hide. Binary values come back base64-encoded and labelled. Audited as secret.reveal.{success,denied,failed}.
- secrets:edit: the same YAML editor as other resources (view → Edit → dry-run → diff → Apply). Text values appear as plain-text stringData, binary values stay base64 in data, and removing a key deletes it. Opening the YAML is audited as secret.yaml.view., applying as secrets.apply..
- Secrets are read from the API server on demand; they are never put in the informer cache.
Recordings
- Download now asks for confirmation first: the file can contain secrets, leaves OpenConsole's access control once downloaded, and must not be shared in tickets, chat or email.
Tests
- An RBAC engine test and an API authorization suite, run on the real router and SQLite store against a fake clientset. They cover tokens (missing / forged / revoked / deactivated), login lockout, every admin endpoint, namespace / resource / cluster-scoped grants, write denials with their audit entries, exec, and secrets (no value leaks; reveal and edit enforced and audited).
Deploy and docs
- deploy/rbac.yaml: ServiceAccount, long-lived token Secret, ClusterRole (now with secrets get / list / update) and binding.
- deploy/rbac-readonly.yaml: a view-only ClusterRole as a drop-in replacement. Writes, shell and secrets are refused by Kubernetes.
- deploy/kustomization.yaml: the whole install is
kubectl apply -k deploy/. - deployment.yaml:
- The image is pinned instead of :latest.
- The pod mounts no ServiceAccount token (serviceAccountName is commented out, automountServiceAccountToken: false). OpenConsole only uses the credentials added in Admin → Clusters.
- Dockerfile: DATA_PATH=/data/app.db by default, with /data writable by
the app user. Before,
docker runwithout DATA_PATH crashed. - README:
- New "Connecting clusters" guide: token method field by field, kubeconfig method, rotation, troubleshooting.
- New sections: "Read-only mode", "Application permissions", "Secrets", "Tests".
- Removed the wrong claim that a cluster is detected automatically when running in-cluster.
Fixes
- Restart and scale returned a generic 502 when Kubernetes refused them. They now return the real status (for example 403 under the read-only ClusterRole) together with the API server's message.
- Confirm dialogs showed English "Cancel" / "Confirm" in the Turkish UI.
Upgrade notes
- To use the Secrets tab, re-apply deploy/rbac.yaml (or add secrets get / list, plus update for edit, to your own ClusterRole). Without it the tab shows a clear 403 hint, and everything else keeps working.
- No other changes are required. If your deployment still sets serviceAccountName, it is harmless; you can drop it.
Image: ghcr.io/vurulkan/kubernetes-openconsole:2.12.0 (published by CI from this tag), linux/amd64.
🤖 Generated with Claude Code
v2.11.1
2026-10-09 · on GitHub
Fixes
- Pod shell could crash the whole server. The idle watchdog wrote its "session idle timeout" frame without the WebSocket write lock while stdout was streaming. gorilla/websocket panics on overlapping writes, and because the panic happens in a goroutine, it kills the process for every user. Trigger: a shell that keeps printing (for example tail -f) with no keystrokes for 5 minutes. All exec writes are now serialized, with a 15 s write deadline. A regression test reproduces the panic without the fix.
- Classic role-permission form: after adding permissions, the reset dropped the DaemonSet / StatefulSet / HPA rows and every "edit" toggle. It now resets to the full default.
- TypeScript: tsc --noEmit goes from 39 errors to 0 (icon prop types).
i18n (TR)
- Pod shell modal: title, Connected / Connecting… / Reconnect, connection error, session closed, auto-shell option.
- Admin: subtitle; Clusters, Add Cluster, Sessions, LDAP import and Login Logo cards; cluster form (Edit / Delete, kubeconfig upload, Save Cluster, placeholders); group modal; the "Press ⌘K" header hint.
CI / release process
- Backend tests run with -race and fail the build. Before this, a
|| echoswallowed test failures. - The frontend is type-checked (tsc --noEmit) before the vite build.
- CI is now the only publisher of images: a push to main publishes
:latest and :
; a vX.Y.Z tag publishes :X.Y.Z (linux/amd64). Manual pushes no longer race CI over :latest.
Image: ghcr.io/vurulkan/kubernetes-openconsole:2.11.1 (published by CI from this tag), linux/amd64.
🤖 Generated with Claude Code
v2.11.0
2026-10-08 · on GitHub
Every pod shell (exec) session is now recorded as an asciicast v2 file and can be replayed by admins in the browser.
Recording
- Terminal output (stdout + stderr) and resizes are streamed to
/<session_id>.cast. Typed input is not recorded as input events; whatever the shell echoes back is part of the output. - The shell shows a red "This session is being recorded" banner that warns that secrets printed to the screen are stored too. The banner appears only when the server confirms the session is recorded.
- Fail-open: a disabled, full or broken recorder never blocks the shell; the session just isn't recorded and the reason is audited. The 5-minute idle timeout is unchanged.
Admin → Recordings
- Settings, editable without a restart: on/off, retention days, per-session cap, total quota, free-space floor, disk policy. The panel shows disk usage and warns when the limits are hit.
- List: filters for user, namespace, pod and date range, server-side paging, in-progress / truncated badges, Play / Download / Delete.
- Player: asciinema-player bundled from npm and lazy-loaded (no CDN). Play/pause, seek, 1x/2x/4x, skip idle time.
Disk protection
- Recordings share the /data PVC by default. A total quota (2 GB) and a free-space floor (512 MB) are checked before every session and every 30 s while sessions run, so the volume never fills up under app.db.
- Disk policy evict_oldest (default) deletes the oldest recordings to make room, audited as recording.evicted. Policy stop keeps them and opens new shells unrecorded.
- A daily retention purge also removes orphan .cast files. After a crash, the next start closes the recordings that were left open.
Audit
- pod.exec.session.recorded (session_id, size, dur_ms, truncation) carries the same request_id as pod.exec.start / end.
- Also: pod.exec.session.record_skipped / record_failed, recording.view / download, recording.delete., recording.settings.update., recording.evicted, recording.purge.
- recording.delete.* records whose recording was deleted (owner, start time, size, cluster, container). The row and the file are gone afterwards, so the audit entry is the only remaining trace.
- Every recording action (view / download / delete) also writes its own "recording.action" console log line with request_id. This line does not depend on LOG_INCLUDE_AUDIT.
API (admin only)
- GET /api/admin/recordings, GET /api/admin/recordings/{id}, GET /api/admin/recordings/{id}/cast (?download=1), DELETE /api/admin/recordings/{id}, GET / PUT /api/admin/recordings/settings
Deploy
- New env vars SESSION_RECORDING_ENABLED / DIR / RETENTION_DAYS / MAX_SIZE_MB / MAX_TOTAL_MB / MIN_FREE_MB / DISK_POLICY. All except DIR only seed the settings on first boot; the admin UI owns them after that.
- deploy/deployment.yaml documents them (commented). deploy/pvc.yaml and deployment.yaml include a commented, optional dedicated recordings PVC.
- CSP script-src adds 'wasm-unsafe-eval' for the player's WebAssembly terminal. It allows WebAssembly compilation only, not JS eval.
- No ClusterRole change is needed.
Fixes
- The active cluster is now reconnected after a restart. Before this fix, a cluster activated through the multi-cluster API was not restored on boot, and the dashboard showed "Service Unavailable" until an admin re-activated it.
- The Admin page title is translated (it was always the English tab label).
Image: ghcr.io/vurulkan/kubernetes-openconsole:2.11.0 (also :latest), linux/amd64.
🤖 Generated with Claude Code
v2.10.2
2026-10-08 · on GitHub
Last English stragglers across the UI are now translated:
- ResourceListView status badges in both card and list views (Running / Available / Progressing / Scaled to 0 / Ready N/M / Available N/M / Progressing N/M / No nodes / Suspended / Active / Succeeded / Failed · N / Running · N / Pending).
- AdminPage status alerts (LDAP + Azure test success, cluster add/update/validate/needs-fields/syncing, logo select/uploaded/removed).
- "Administration" header, Kubeconfig / ServiceAccount Token option labels, "All clusters" option + badge, Active badge in clusters table.
- Delete-user / Delete-group / Delete-role / Delete-cluster confirm dialog titles and the deactivate-cluster message.
- PodExecModal status (Error / Closed / Idle).
- Role Permissions table headers (Resource / Actions).
v2.10.1
2026-10-08 · on GitHub
Follow-up to 2.10.0.
Symptom: ConfigMap values that contained trailing whitespace on any
line (common for Spring Boot application.yml snippets stored in a
data: entry) still rendered as the ugly "spring:\n kafka:\n…"
single-liner even though 2.10.0 wrapped the YAML in the prettifier.
Cause: yaml.v3 encoder silently falls back from LiteralStyle to
double-quoted when any line of the content ends with whitespace or the
string as a whole has trailing spaces — the YAML spec forbids trailing
whitespace inside a block scalar.
Fix: strip trailing whitespace from each line (and the string as a
whole) before setting Style = LiteralStyle. The encoder now emits a
real | block.
Trade-off: a handful of trailing spaces in the original value are lost in
the UI's YAML view. Byte-for-byte fidelity is still available via
kubectl get -o yaml.
v2.10.0
2026-10-08 · on GitHub
v2.10.0 — i18n phase 4 + YAML prettifier + live modal titles
ConfigMap YAML readability
Multi-line string values (ConfigMap data entries that hold scripts, JSON
blobs, config files) used to render as a single double-quoted line with every
newline shown as a literal \n. The backend now runs every YAML through a
yaml.v3 node-tree walk that converts such scalars into block scalars (|- /
|). Every workload YAML (Pod, Deployment, DaemonSet, StatefulSet, HPA,
Service, ConfigMap, Ingress, CronJob, Job) benefits.
Before (what sigs.k8s.io/yaml emitted):
data:
app.conf: "listen: 0.0.0.0\nport: 8080\ngraceful: 10s\n"
After (what the UI now shows):
data:
app.conf: |
listen: 0.0.0.0
port: 8080
graceful: 10s
Guards: control characters other than LF/TAB fall back to the original scalar (block scalars can't carry them legally); leading whitespace on the first line also disqualifies.
Live locale switch for Dashboard modal titles
The Dashboard modal title (Pod Logs / Deployment Logs / YAML view / Events /
ConfigMap Data) used to live in React state as an English kind-marker. The
state didn't react to language changes, so flipping EN↔TR while the modal
was open left the title in English. The Modal title prop now resolves the
marker through t() on every render — the title updates live without a
refresh.
i18n phase 4 — admin forms
Translated in this release:
- Role Permissions Add / Edit / Copy modals: template buttons
(Viewer / Developer / SRE / Admin), cluster + namespaces labels, filter
placeholder,
+ Select N matching+Clear selectionbuttons, chip tooltip wording, Save counter, cancel / copy / copying copy, "all clusters (wildcard)" option. - AdminPage body: Session Lifetime label, LDAP bind-password toggle + field, user base DN / filter / username attribute, LDAP search + import flow, Azure AD client secret toggle, cluster create / edit drawer labels (Name / Description / Method / API Server / Token / CA Cert / Replace credentials), user create and edit drawer labels (Username / Password / Grant admin / Active / Admin / Groups / Assign groups), Activate / Deactivate buttons and their confirm dialog + status messages.
Shared common.* keys now cover Name / Description / Method / Session
Lifetime / Activate / Deactivate etc. in one entry per locale rather than
repeating across every form.
Upgrade
docker pull ghcr.io/vurulkan/kubernetes-openconsole:2.10.0
docker rm -f openconsole && \
docker run -d --name openconsole -p 8080:8080 \
-e DATA_PATH=/tmp/app.db -e LOG_FORMAT=json \
ghcr.io/vurulkan/kubernetes-openconsole:2.10.0
No schema change.
v2.9.0
2026-10-05 · on GitHub
v2.9.0 — i18n phase 3 + audit log request_id correlation
Logs — ES-ready, correlation complete
- Audit mirror now carries
request_id. Every write op the console records (deployment.scale, pod.exec, yaml.apply, session.revoke, cluster.activate, …) is mirrored to stdout with the samerequest_idthe HTTP request log already carries. In Elasticsearch / Kibana you can filter byrequest_idto see the full chain: request landed → audit entry → outcome. - Threading goes through a background context (goroutine outlives
r.Context()), so the HTTP cancellation won't abort the DB insert and the slog line still gets the id.
Env knobs (reminder)
LOG_FORMAT=json # recommended for ES; default text
LOG_LEVEL=info # debug|info|warn|error
LOG_OUTPUT=stdout
LOG_ADD_SOURCE=true # file:line in each line
LOG_INCLUDE_AUDIT=true # mirror audit records to stdout (default on)
An audit line now looks like (with LOG_FORMAT=json):
{
"time": "2026-10-05T08:14:05Z",
"level": "INFO",
"msg": "audit",
"event": "deployment.scale.success",
"user": "mustafav",
"namespace": "payments",
"resource_type": "deployment",
"resource_name": "nginx (from=3 to=5)",
"request_id": "qXpB6yXONDiuXKX2",
"service": "openconsole"
}
i18n phase 3 — Turkish now covers almost everything visible
Translated in this release:
- Section tables: Users / Groups / Roles / Sessions (column headers, empty states, search placeholders, "new …" buttons, status badges, revoke confirm dialogs).
- ClusterSwitcher: "no cluster" label, dropdown title, switching indicator, admin-only hint, activation failure dialog, OK / Dismiss.
- YamlEditModal: every button (Edit / Dry-run / Apply / Reset / Reload / Close), state badges (Read-only / Editing / Edited / Loading), dry-run and applied notices, read-only hint, apply-confirm dialog.
- ChangePasswordPage: title, subtitle, labels, submit button copy, failure message, success alert.
- Admin audit page: filter labels (User / Action / Namespace / dates), pagination (Previous / Next), Export CSV.
- LDAP form: Enabled, Host, Port, Use SSL, StartTLS, Skip Verify, Timeout, Bind DN; Save button.
- Azure AD form: Tenant ID, Client ID, Redirect URL; Save button.
Still English (phase 4 backlog)
Deeper remaining strings in the LDAP search / import flow, Azure AD client secret rotation, Customization logo upload hint, individual action-modal copy — the infra is now in those files, so expansion is a per-string change in later patches.
Upgrade
docker pull ghcr.io/vurulkan/kubernetes-openconsole:2.9.0
docker rm -f openconsole && \
docker run -d --name openconsole -p 8080:8080 -e DATA_PATH=/tmp/app.db \
-e LOG_FORMAT=json \
ghcr.io/vurulkan/kubernetes-openconsole:2.9.0
No schema change.
v2.8.0
2026-10-05 · on GitHub
v2.8.0 — i18n phase 2 + Views v shortcut + Login locale switcher
Three polish wins. All frontend, no schema change.
Views keyboard navigation (v)
- Press
vanywhere to open the Saved Views dropdown. - Inside the menu: 1-9 picks a view by index, ↑/↓ moves focus,
Enter restores the focused view, Esc closes. Each row shows its
1..9prefix as akbdso the shortcut is self-explanatory. Shortcut is suspended while the "name this view" input is open so typing a digit in a name doesn't accidentally restore a view.
LoginPage locale switcher
LocaleSwitcherpinned to the top-right of the login screen so a user on a TR browser lands on TR on an EN-defaulted deploy without logging in to flip the toggle.- Default locale behavior is unchanged: localStorage pick wins, otherwise
navigator.languageis used, otherwise EN.
i18n phase 2 — TR covers more of the surface
Dictionary expansion with full EN + TR translations for:
- LoginPage (headline, subheadline, stats, username/password placeholders, Microsoft button, footer, failure message).
- ChangePasswordPage (headers, labels, validation copy).
- Dashboard status badges: Running / Available / Progressing / Scaled to 0 / Ready N/M / Succeeded / Failed · N / Pending / No nodes / Suspended / Active.
- Dashboard action buttons: Logs / Events / YAML / Scale / Restart / Shell / Data.
- Dashboard empty states, auto-refresh pill, "no namespace" fallback, scale modal, restart confirm.
- LiveEventsPanel: title, status (streaming / disconnected), filter (All / K8s events / Warnings), pause / resume / clear tooltips.
- SavedViewsMenu: button label, dropdown title, empty state, save prompt input placeholder.
- AdminPage: SectionCard titles for every sub-tab (Role Permissions, LDAP Configuration, Azure AD Configuration, Session Settings, Audit Logs, Sessions), "New user" drawer title, Classic/New view toggle links.
- RolePermissionsPanel: panel description, role sidebar search + "No roles", "No description" placeholder, "Copy from role" / "Add permissions" buttons, GrantCard ⋯ menu (Edit permissions… / Duplicate to another namespace… / Clear card), "all clusters" badge, empty state.
- Keyboard cheat sheet: all group titles AND row descriptions localised.
Deferred to phase 3
The deeper admin forms (LDAP body, Azure AD body, Customization, table column headers, individual action-modal copy) remain in English for this release. The i18n infrastructure now lives inside those files, so further translation is a per-string change rather than a restructure.
Upgrade
docker pull ghcr.io/vurulkan/kubernetes-openconsole:2.8.0
docker rm -f openconsole && \
docker run -d --name openconsole -p 8080:8080 -e DATA_PATH=/tmp/app.db \
ghcr.io/vurulkan/kubernetes-openconsole:2.8.0
No schema change, no backend-contract change.
v2.7.0
2026-10-02 · on GitHub
v2.7.0 — Polish pack
Five small-to-medium UX wins in one release. All frontend, no schema change.
Role Permissions bulk selection
The Add Permissions modal's namespace picker gets "+ Select N matching" and "Clear selection" buttons next to the filter input. Type a prefix, bulk-pick; type the next prefix, bulk-pick. No regex, no new syntax — composable with the filter you already know.
Dashboard: label filters + saved views
- Label tokens in the search box. Space-separated.
foomatches a resource name;label:app=apimatches an exact label;label:tiermatches any resource that carries the key. Multiple tokens AND. Example:web label:tier=frontkeeps rows whose name containswebAND havetier=front. - "Views" dropdown in the resource card header saves the current
(namespace, tab, search, viewMode)under a name inlocalStorage. One click restores. The active view is marked. Delete on row hover.
i18n foundation — EN + TR (phase 1)
- Compact EN / TR locale switcher next to the theme toggle. The choice persists; browser language is the first-run fallback.
- Phase 1 translated surface: top-level nav (Dashboard / Admin / Sign out / Shortcuts), Dashboard overview heading + subtitle + refresh, resource tab labels (Pods / Deployments / …), Admin tab labels, Saved Views copy.
- Everything else continues to render in English until later releases widen coverage; the infra (react-i18next) is in place so deeper translation is a pure string-level change from here on.
a11y + dark mode polish
- Explicit
aria-labelon the Card / List view-mode toggles so screen readers announcecard view, pressedinstead of silence. - Two Admin cluster-form labels were missing a dark variant and read as
low-contrast slate-on-slate; both got
dark:text-slate-400.
Upgrade
docker pull ghcr.io/vurulkan/kubernetes-openconsole:2.7.0
docker rm -f openconsole && \
docker run -d --name openconsole -p 8080:8080 -e DATA_PATH=/tmp/app.db \
ghcr.io/vurulkan/kubernetes-openconsole:2.7.0
No schema change, no backend-contract change.
v2.6.1
2026-10-02 · on GitHub
v2.6.1 — YAML apply returned 502 (fix)
Two bugs combined to turn every apply attempt into a Cloudflare 502 for the deployed instance. This patch fixes both.
Fixes
- Served YAML was missing
apiVersionandkind. Typed client-go objects come back with an emptyTypeMeta;sigs.k8s.io/yaml.Marshalhonors the jsonomitemptytag and drops both fields. When the operator posted that same YAML to apply, the dynamic client built an unstructured object with no GVK and the API server rejected the body with "object has no kind". Each YAML GET handler (Pod, Deployment, DaemonSet, StatefulSet, HPA, Service, ConfigMap, Ingress, CronJob, Job) now setsTypeMetaexplicitly before marshaling. - Apply handler returned
502 Bad Gatewayfor every error. 502 at the origin triggers Cloudflare's branded error page, which hid the real message. The handler now maps the underlying k8s API error to its proper HTTP status:- NotFound →
404 - Conflict (stale
resourceVersion) →409 - Forbidden →
403 - Invalid / BadRequest / RequestEntityTooLarge →
400 - Timeout / ServerTimeout →
504 - ServiceUnavailable →
503 - InternalError →
500 - Client-side validation we raised (name/namespace mismatch, missing
resourceVersion, YAML parse) →
422 Unprocessable Entity
- NotFound →
Only a true upstream outage reaches 5xx now, which Cloudflare passes
through unchanged.
Belt-and-braces
The Apply path also injects apiVersion/kind on the submitted object when
GVK is empty, so a YAML that was fetched by a browser tab on 2.6.0 (before
TypeMeta was populated) keeps working after this upgrade.
Upgrade
docker pull ghcr.io/vurulkan/kubernetes-openconsole:2.6.1
docker rm -f openconsole && \
docker run -d --name openconsole -p 8080:8080 -e DATA_PATH=/tmp/app.db \
ghcr.io/vurulkan/kubernetes-openconsole:2.6.1
No schema change. Any browser tab that was holding YAML from 2.6.0 should Reload before editing (header button) to pick up the full YAML including apiVersion/kind.
v2.6.0
2026-10-02 · on GitHub
v2.6.0 — YAML view/edit with Monaco + server dry-run
Every resource that already had a YAML button now opens a view/edit modal powered by Monaco. Operators can audit the YAML, flip to Edit, run a server-side dry-run, and apply with a confirmation — all without leaving the console.
Flow
- View (default) — the YAML button opens a read-only Monaco pane with syntax highlighting. Non-editors stop here.
- Edit — a visible toggle (only when you have the new
{resource}:editpermission) flips the body to a side-by-side DiffEditor: left is the server copy, right is your editable buffer. - Dry-run — sends the draft to the API server with
DryRun=All. The canonicalized object (defaulted fields, timestamps) replaces the left pane so you see what the cluster would actually store. - Apply — gates behind a confirmation and persists. The modal swaps in
the fresh
resourceVersionso you can iterate without reopening.
Permissions
- New per-resource action
edit. The admin wildcard includes it on every resource; non-admins get it via a role grant. - Role Permissions "Add" matrix shows
editas a destructive-tinted chip (rose) so operators can grant it deliberately. - SRE template now also grants
editondeployments,statefulsets,configmapsso bumping an env var or scaling via YAML doesn't need an Admin role.
Backend
POST /api/namespaces/{ns}/{resource}/{name}/applywith{yaml, dryRun}. Supported resources: pods, deployments, daemonsets, statefulsets, hpas, services, configmaps, ingresses, cronjobs, jobs.- Guardrails:
metadata.nameandmetadata.namespacemust match the URL;metadata.resourceVersionis required (optimistic concurrency — a stale edit fails with 409 instead of clobbering). - Audit entries:
{resource}.apply.{success,denied,rate_limited,failed}withbefore/aftersha256 prefixes stored in the row'sResourceName— enough to distinguish writes without logging full YAML. - Per-user rate limiter (shared with scale/restart buckets).
CDN / air-gapped note
Monaco loads from https://cdn.jsdelivr.net by default. CSP widens to allow
that CDN and worker-src blob: so Monaco can spawn its web workers.
Air-gapped deployments can self-host the vs folder and point
@monaco-editor/react's loader.config({ paths: { vs: '/vs' } }) at a local
path, then tighten CSP back to self.
Upgrade
docker pull ghcr.io/vurulkan/kubernetes-openconsole:2.6.0
docker rm -f openconsole && \
docker run -d --name openconsole -p 8080:8080 -e DATA_PATH=/tmp/app.db \
ghcr.io/vurulkan/kubernetes-openconsole:2.6.0
No schema change. Monaco is lazy-loaded, so first page open is unchanged; opening the YAML modal pulls Monaco from the CDN on first use.
Rollback
docker pull ghcr.io/vurulkan/kubernetes-openconsole:2.5.2
v2.5.2
2026-10-02 · on GitHub
v2.5.2 — Modal portal + single-line namespace strip
Two fixes on top of 2.5.1.
Fixes
- Modal is now viewport-centered everywhere. The dialog renders via
createPortal(..., document.body)so a parent element'sbackdrop-filter/transform/filtercan no longer trapposition: fixedinside its own containing block. Add Permissions, Copy from Role and every other modal now open in the middle of the screen, not anchored to the card that triggered them. - Grant card header stays on one line. When a card merges many
namespaces, the strip no longer wraps to a second row. The first 5
namespaces show inline; the rest collapse into a
+N morebadge that opens a hover / focus popover listing every namespace. "Same grants" marker stays on the right.
Upgrade
docker pull ghcr.io/vurulkan/kubernetes-openconsole:2.5.2
docker rm -f openconsole && \
docker run -d --name openconsole -p 8080:8080 -e DATA_PATH=/tmp/app.db \
ghcr.io/vurulkan/kubernetes-openconsole:2.5.2
No schema change, backend untouched.
v2.5.1
2026-10-02 · on GitHub
v2.5.1 — Role Permissions polish
Four fixes on top of 2.5.0.
Role Permissions
- Identical namespaces merge into one card. If a role grants the same
(resource, action)set onns=aandns=bof the same cluster, the panel renders a single card with both namespaces listed in the header, not two identical cards. The counterN namespaces, same grantsappears next to the chips when more than one is merged in. - Edit permissions. The row
⋯menu gains Edit permissions…. It opens the Add modal in edit mode seeded with the card's current state. Save computes a diff — unticked chips are deleted, newly-ticked chips are added, everything else is untouched — scoped only to this card so unrelated grants stay put. - Chip remove on merged cards removes the
(resource, action)from every namespace listed on the card. The tooltip spells that out so there is no surprise. - Terminology follows the merge: the destructive menu entry is now Clear card (not "Clear row"); Duplicate to another namespace… opens Add with no namespace preselected so the operator picks the targets.
Modal + Checkbox
- Modal locks body scroll while open. Previously, wheeling over the backdrop moved content behind the dialog, which made the modal feel like it "opened too far below". Scroll is now frozen until the modal closes.
- New size
lg(64 rem, viewport-capped). The Add Permissions modal uses this instead offull(which was 95 vh and dominated the viewport). The form reads as a focused dialog now. - Resource-row checkbox renders native indeterminate. When some but not all actions for a resource are ticked, the parent box shows the browser's mixed state instead of silently flipping back to unticked. Ticking an indeterminate box fills in the remaining actions.
Upgrade
docker pull ghcr.io/vurulkan/kubernetes-openconsole:2.5.1
docker rm -f openconsole && \
docker run -d --name openconsole -p 8080:8080 -e DATA_PATH=/tmp/app.db \
ghcr.io/vurulkan/kubernetes-openconsole:2.5.1
No schema change, backend untouched.
Rollback
docker pull ghcr.io/vurulkan/kubernetes-openconsole:2.5.0
## or :2.4.0, :2.3.0, :2.1.4 — all safe
v2.5.0
2026-10-02 · on GitHub
v2.5.0 — Cluster switcher shortcut + Role Permissions redesign
Two usability wins, both frontend-only, both behind a safety net.
Cluster switcher
- Global
copens the header cluster dropdown (admins only; non-admin switcher stays read-only). - Inside the menu: 1..9 pick a cluster by index, ↑/↓ navigate, Enter
activates the focused row, Esc closes. Each row shows its
1..9prefix as akbdso the shortcut is self-explanatory.
Keyboard cheat sheet (?) lists the new keys under a new Cluster switcher
section.
Role Permissions redesign (new default · classic reachable)
The old form (one role → one cluster/ns → checkbox grid → flat pill list) was noisy and tedious for anything beyond a single grant. The new layout:
- Role sidebar with search and a
N perms · M ns · K clusterssummary per role (parallel-loaded on mount). - Grouped grant cards — one card per
(cluster, namespace)pair, one row per resource. Action chips are tinted by family:read= slate,write= amber,destructive (exec)= rose. One glance tells you the risk shape of a grant. - Chip X / shift-click removes a single
(resource, action)grant. - Row ⋯ menu: Duplicate to another namespace… (seeds the Add modal with the current grants pre-selected) and Clear row.
- Add permissions modal — cluster + multi-namespace picker + four
templates (Viewer / Developer / SRE / Admin). Chips are struck-through
when every selected namespace already has that
(resource, action), and a live counter at the bottom saysWill create N new permissions · M will be skipped. - Copy from another role modal — pick a source role, bulk-copy grants. Overlaps are skipped automatically.
Rollback
If the new UI isn't for you, click "Switch to classic view" in the Role
Permissions header. localStorage.rolePermissionsLayout persists the choice.
The classic UI stays fully intact and offers a reverse link. Nothing changes
on the backend; a future release can delete the classic code once we've lived
with the new one for a while.
Upgrade
docker pull ghcr.io/vurulkan/kubernetes-openconsole:2.5.0
docker rm -f openconsole && \
docker run -d --name openconsole -p 8080:8080 -e DATA_PATH=/tmp/app.db \
ghcr.io/vurulkan/kubernetes-openconsole:2.5.0
No schema change, no backend-contract change.
Rollback (if the whole release goes sideways)
docker pull ghcr.io/vurulkan/kubernetes-openconsole:2.4.0
docker rm -f openconsole && docker run -d --name openconsole -p 8080:8080 \
-e DATA_PATH=/tmp/app.db ghcr.io/vurulkan/kubernetes-openconsole:2.4.0
v2.4.0
2026-10-02 · on GitHub
v2.4.0 — DaemonSet / StatefulSet / HPA tabs + Running-badge polish
Workload coverage lands. The Dashboard gets three new tabs between Deployments and Services, each with the card and list views already available for the other workload kinds.
New tabs
- DaemonSets — status badge
Ready N/M, Nodes column (current/desired), YAML. - StatefulSets — status badge
Available N/M/Progressing N/M/Scaled to 0, Replicas column, YAML, and Scale (same modal as Deployment scale, same rate limit, sameMAX_REPLICAScap). - HPAs — Target (
Kind/name), Min/Max, Current (current/desired), YAML.
Fix
- Pod card: the amber ring is gone from the dot inside the "Running" badge.
The ring stays only on the outer status dot (left of the pod name) as a
quiet "recovered from a crash" tell. The badge itself is always a plain
green
Runningregardless of restart count.
Internals
kube.Manager's informer factory now watchesDaemonSet,StatefulSet(apps/v1) andHorizontalPodAutoscaler(autoscaling/v2). The informer-event WS feed also carries add/update/delete frames for these kinds.- New endpoints:
GET /api/namespaces/{ns}/daemonsets[/{name}[/yaml]]GET /api/namespaces/{ns}/statefulsets[/{name}[/yaml]]POST /api/namespaces/{ns}/statefulsets/{name}/scale— writesstatefulset.scale.{success,denied,rate_limited,failed}audit outcomes.GET /api/namespaces/{ns}/hpas[/{name}[/yaml]]
- Admin wildcard permissions include the three new resource types, and the
Admin → Roles → Permissions matrix now lets you grant
list/get/scale(StatefulSet only) per cluster/namespace.
Upgrade
docker pull ghcr.io/vurulkan/kubernetes-openconsole:2.4.0
docker rm -f openconsole && \
docker run -d --name openconsole -p 8080:8080 -e DATA_PATH=/tmp/app.db \
ghcr.io/vurulkan/kubernetes-openconsole:2.4.0
No schema change. First page load may take a few seconds while the three additional informers warm up.
Rollback
docker pull ghcr.io/vurulkan/kubernetes-openconsole:2.3.0
docker rm -f openconsole && docker run -d --name openconsole -p 8080:8080 \
-e DATA_PATH=/tmp/app.db ghcr.io/vurulkan/kubernetes-openconsole:2.3.0
No data migration — safe to roll back.
v2.3.0
2026-10-02 · on GitHub
v2.3.0 — Session rotation + Admin Sessions tab (M4-B3 phase 1)
Admins can now see every token this backend has issued and revoke any one of them. Change-password automatically kills every live session for the affected user — so a stolen token can't survive a password reset.
Highlights
- Admin → Sessions: new tab lists every session (user, status, IP, UA, issued / last-used / expires times) with an "Only active" toggle. Per-row Revoke or Revoke all for user, both with confirmation. Your own session is tagged "this is you".
- Password rotation kills tokens: change-password now revokes every session belonging to the user, including the browser that initiated the change. The UI falls back to the login screen on the next request.
- Browser kept open longer than it should be? Click revoke in Admin, and the next request from that browser gets a 401.
Internals
- JWT remains the primary auth artifact. A new
session_tokenstable pins each issued token by itsjtiand holds the revocation flag. AuthMiddleware consults the table on every request and fails open on DB errors — a storage blip can only degrade admin control, never log everyone out. last_used_atwrites are throttled in-memory (30s perjti), so AuthMiddleware never serializes an UPDATE behind SQLite's single writer.- Pre-tracking JWTs (issued before this upgrade and still in flight) are honoured until their natural expiry — rolling upgrades don't kick anyone out of a live tab.
- New endpoints (admin only):
GET /api/admin/sessions[?activeOnly=1&userId=…],DELETE /api/admin/sessions/{id},POST /api/admin/users/{id}/revoke-sessions. Audit actions:session.revoke,session.revoke_all.
Upgrade
docker pull ghcr.io/vurulkan/kubernetes-openconsole:2.3.0
docker rm -f openconsole && \
docker run -d --name openconsole -p 8080:8080 -e DATA_PATH=/tmp/app.db \
ghcr.io/vurulkan/kubernetes-openconsole:2.3.0
The migration adds session_tokens and two indexes. It is idempotent and
touches no existing rows. Already-issued JWTs stay valid until they expire.
Rollback
docker pull ghcr.io/vurulkan/kubernetes-openconsole:2.1.4
docker rm -f openconsole && docker run -d --name openconsole -p 8080:8080 \
-e DATA_PATH=/tmp/app.db ghcr.io/vurulkan/kubernetes-openconsole:2.1.4
The session_tokens table stays in the DB after rollback (unused). No cleanup
required.
v2.1.4
2026-10-02 · on GitHub
v2.1.4 — Pod badge + namespace filter fixes
Small patch on top of 2.1.3.
Fixes
- Running badge text. The pod badge now reads just
Runningregardless of restart count. The green dot in the badge grows an amber ring when the pod has restarted at least once (and the card's outer status dot keeps the same ring, as before 2.1.2) — operators keep the "recovered from a crash" tell without the badge flipping to amber for a long-ago restart. The exact restart count still lives in the pod body / list column. - Namespace filter unreadable in dark mode. The left-panel namespace
search input had
dark:text-slate-500, which happens to be the same color as its own placeholder — so what you typed looked identical to the hint. Fixed toslate-900/slate-100.
Upgrade
docker pull ghcr.io/vurulkan/kubernetes-openconsole:2.1.4
docker rm -f openconsole && \
docker run -d --name openconsole -p 8080:8080 -e DATA_PATH=/tmp/app.db \
ghcr.io/vurulkan/kubernetes-openconsole:2.1.4
No schema change.
v2.1.3
2026-10-02 · on GitHub
v2.1.3 — Keyboard shortcut fixes
Small patch on top of 2.1.2.
Changes
nnow focuses the namespace filter in the left panel instead of opening the command palette. ⌘K still owns the palette, and the namespace filter is what people actually reach for.- Cheat sheet reads correctly for alternatives. Previously the physical-
key fallbacks for TR layout (
ğ/ü/.) showed as[ then (ğ)— as if it were a chord sequence. They're now joined withor, and the explanation at the top of the sheet distinguishesthen(chord) fromor(alternative).
Upgrade
docker pull ghcr.io/vurulkan/kubernetes-openconsole:2.1.3
docker rm -f openconsole && \
docker run -d --name openconsole -p 8080:8080 -e DATA_PATH=/tmp/app.db \
ghcr.io/vurulkan/kubernetes-openconsole:2.1.3
No schema change.
v2.1.2
2026-10-02 · on GitHub
v2.1.2 — Pod badge polish
Patch on top of 2.1.1.
Changed
- The Pods list and Pods cards now show Running in green regardless of restart count. The restart counter in the pod body (and the dedicated "Restarts" column in list view) still carries the number, so the badge no longer flips to amber for a long-ago crash recovery that has nothing to tell the operator about the pod's current state.
Upgrade
docker pull ghcr.io/vurulkan/kubernetes-openconsole:2.1.2
docker rm -f openconsole && \
docker run -d --name openconsole -p 8080:8080 -e DATA_PATH=/tmp/app.db \
ghcr.io/vurulkan/kubernetes-openconsole:2.1.2
No schema change.
v2.1.1
2026-10-02 · on GitHub
v2.1.1 — Live events polish + stable list sort
Follow-up patch to 2.1.0.
Fixes
- Live Events panel closes on
Escwhen it is open. Esc still goes to the focused input first, so filter boxes elsewhere keep working. - New shortcut
etoggles the Live Events panel from anywhere on the Dashboard (while not typing in an input). - The keyboard cheat sheet (
?) lists both. - Alphabetical list sort restored. 2.1.0 served lists directly from the informer indexer, which is backed by a Go map and iterates in random order — so the namespace list (and every other list) shuffled between refreshes. ResourceClient now sorts lister results by Name, matching the pre-2.1.0 ordering the API server used to return.
Upgrade
docker pull ghcr.io/vurulkan/kubernetes-openconsole:2.1.1
docker rm -f openconsole && \
docker run -d --name openconsole -p 8080:8080 -e DATA_PATH=/tmp/app.db \
ghcr.io/vurulkan/kubernetes-openconsole:2.1.1
No schema change. Safe to roll back to 2.1.0 or 2.0.0 the same way.
v2.1.0
2026-10-02 · on GitHub
v2.1.0 — Live cluster
This release flips the dashboard from a 10-second polling loop to a client-go informer watch, and adds a live events panel so you see what the cluster is doing in real time.
Highlights
- Informer cache replaces polling for Pods, Deployments, Services, ConfigMaps, Ingresses, CronJobs, Jobs, Namespaces. Lists are served from the local lister once the initial sync completes; the old list endpoints fall back to a live API call during warmup or when no cluster is configured, so the UI never shows an empty page because the cache is cold.
- Live Events panel (collapsible, right side of the Dashboard) streams
added/updated/deletedframes plus nativecorev1.Eventobjects with theirReason/Message/Type. Filter: All / K8s events / Warnings. Pause without dropping the socket; 200-event ring buffer; remembers open/closed across refreshes. - New endpoint
GET /ws/events?namespace=<ns>— authenticated via the usual Bearer token (query-stringtoken=also accepted for browser WS).
Internals
kube.Managernow owns theSharedInformerFactoryand anEventBus; credential swaps tear down the previous factory before starting a new one, so deactivating a cluster shuts the informer goroutines cleanly.- The event bus drops frames on a slow consumer rather than stalling the informer workers.
- Initial cache sync is bounded to 20s so a slow cluster can't delay readiness forever; after the window the listers go into service even if a resource type is still catching up (serves what it has).
How to upgrade
docker pull ghcr.io/vurulkan/kubernetes-openconsole:2.1.0
docker rm -f openconsole && \
docker run -d --name openconsole -p 8080:8080 -e DATA_PATH=/tmp/app.db \
ghcr.io/vurulkan/kubernetes-openconsole:2.1.0
No schema change. First page load after upgrade may take a few seconds while the informer cache warms up for each namespace you have access to; subsequent loads are instant.
Rollback
docker pull ghcr.io/vurulkan/kubernetes-openconsole:2.0.0
docker rm -f openconsole && docker run -d --name openconsole -p 8080:8080 \
-e DATA_PATH=/tmp/app.db ghcr.io/vurulkan/kubernetes-openconsole:2.0.0
No data migration — the DB is unchanged between 2.0.0 and 2.1.0.
v2.0.0
2026-10-01 · on GitHub
v2.0.0 — Multi-cluster, write actions, dark mode, shortcuts
OpenConsole 2.0.0 turns the dashboard from a single-cluster read-only viewer into a full multi-cluster operator console with write actions, interactive shells, structured observability, and a modern UI. Backend data is migrated in place; everything else is drop-in.
Highlights
Multi-cluster
- Save many clusters, name them, switch via a dropdown in the header.
- Admin → Clusters: add / edit / activate / deactivate / delete each cluster with validation against the live API server before saving.
- Per-cluster RBAC: every permission row carries a
cluster_id(NULL = "all clusters" wildcard). Rows written on 1.0.x automatically become wildcards, so nothing breaks on upgrade. - Audit, logs and metrics all gain a cluster dimension.
Write actions (opt-in)
- Deployment restart (
deployments:restart) — rolling restart viakubectl.kubernetes.io/restartedAtpatch. - Deployment scale (
deployments:scale) — Scale subresource, bounded byMAX_REPLICAS(default 100), per-user rate limited. - Pod shell (
pods:exec) — interactive xterm.js terminal over SPDY, with container picker, bash → ash → sh fallback, PS1 injection for images that ship without a default prompt, 5-minute idle timeout, start/end audit entries. - Deployment logs fan-out — one WebSocket, every pod×container
streamed with
[pod/container]prefix, matchingkubectl logs -f -l … --all-containers --prefix. - Each write action is off by default on every role; admins opt in per role + per namespace + per cluster.
Observability & security
- Structured logging via
log/slog.LOG_FORMAT=jsonfor Elastic / Loki / Vector.X-Request-Idon every response, correlated with the structured request log and the DB audit entry. - Prometheus
/metrics(zero-dep exposition) — HTTP request counters by status bucket, duration sum, audit events, exec sessions, log streams, deployment-action totals. /livez+/readyzsplit: liveness is process-only, readiness pings SQLite and confirms the configured cluster is Ready.- Login brute-force gate: 5 fails per (IP, username) in 5 minutes
→ 429 +
Retry-After, logged aslogin.locked. - Strict security headers: Content-Security-Policy, X-Frame-Options, Referrer-Policy, Permissions-Policy on every response.
- Admin guards: cannot delete your own account, cannot demote or deactivate the last active administrator.
UI / UX
- Dark mode (Light / Dark / System). Persisted; follows OS when on System.
- Nav rail + section panel + content layout replaces the single sidebar. Admin sub-sections live in the panel.
- Card / List view toggle on every resource tab, remembered across sessions via localStorage. Card view shows the hero chip + metrics; List view is a dense sortable table with per-resource columns (Pods: Ready / Restarts / Age; Deployments: Replicas; Services: Type / Cluster IP / Ports; Jobs: Completions, etc.). Default is Card.
- Sortable list columns — click any header to toggle asc → desc →
off. Sort choice persists per-tab via localStorage, so operators keep
their preferred ordering (
Age descon Pods, say) across sessions. - Jobs tab (next to CronJobs) with its own status chips (Succeeded / Running / Failed / Pending) and a Logs button that fans out across every pod of the Job. The Pods tab no longer lists job-owned pods, so CronJob runs and migration helpers stop cluttering the view.
- Compact age column everywhere (
77d,3h,18m) alongside the full timestamp on cards; list view shows the short form only. - Users / Groups / Roles: table + icon actions (✎ edit, 🗑 delete)
- Create/Edit modal with Enter-to-submit. Single Save replaces Save / Save Groups / Save Roles trio.
- Role Permissions: namespaces picker restricted to real values
(plus explicit
* (all namespaces)wildcard) — no more typos saving as "permissions" that silently never apply. - In-app confirm dialogs: every destructive action uses the themed Modal instead of the browser's confirm/alert boxes.
- Deployment restart uses a themed confirm modal; Scale opens a compact dialog with ±1 stepper, Shift+↑/↓ for ±10, and offline warning at 0 replicas.
- Modals: ESC closes everything. Non-full modals clamp to viewport so long content scrolls inside. All dropdowns portal to body so they never fall behind cards or sidebars.
- Pod cards: name truncates while the status chip stays on one line
(no more two-line headers). Healthy pods that have restarted show a
thin amber halo and the chip reads
Running · N restart(s). - Not Ready pods show an amber triangle inside the chip; hover reveals the pod's latest Warning events (reason + message + lastSeen), parsed from the Events API so you don't have to open a separate modal to diagnose ImagePullBackOff / CrashLoopBackOff / etc.
- Azure AD sign-in works under the new strict CSP — the callback bootstrap script carries a per-response nonce.
- Namespace tab preservation: switching namespace keeps the current resource tab (Deployments, Pods, …) when the new namespace also grants it.
- Soft background: grid pattern replaced by a wide, low-opacity gradient glow — eye-friendly for long sessions.
Keyboard shortcuts
⌘K/Ctrl+Kcommand palette (nav + namespaces + theme)?opens an in-app shortcut cheat sheet (also reachable via the Shortcuts chip in the header)- Chord nav:
g d→ Dashboard,g a→ Admin - Theme chord:
t l / t d / t s - Dashboard:
[/]cycle tabs,rrefresh,/focus search,nopen palette - Log viewer:
ppause,wword wrap - Scale modal: Shift+Arrow for ±10
- Admin:
[/]cycle sub-sections [/]//are matched by physical key position, so Turkish Q users pressğ/ü/.(same spot on the keyboard) and get the same action — no need to reach for AltGr.
DX / release
- GitHub Actions CI: frontend build + backend vet/build/test + Docker
image build; publishes to GHCR on
main.
Upgrading from 1.0.x
The data migration is automatic. Pull the new image, roll the
Deployment, done. SQLite file stays where it is — the schema grows
(new clusters table + cluster_id column on namespace_permissions)
and the legacy kube_credentials row is copied into a cluster named
default with is_active=1 on the first boot so nothing goes offline.
## Optional: snapshot the SQLite file beforehand
kubectl -n kubernetes-openconsole exec deploy/kubernetes-openconsole -- \
cp /data/app.db /data/app.db.pre-2.0.0.bak
## Roll the image
kubectl -n kubernetes-openconsole set image deploy/kubernetes-openconsole \
openconsole=ghcr.io/vurulkan/kubernetes-openconsole:2.0.0
kubectl -n kubernetes-openconsole rollout status deploy/kubernetes-openconsole
ClusterRole changes (only if you want write actions)
The new application permissions deployments:restart, deployments:scale,
and pods:exec are off by default. If you want to grant them to any role
through the Admin UI, the ServiceAccount's ClusterRole must also carry the
matching Kubernetes verbs:
## apps.deployments: patch → required for Restart
## apps.deployments/scale: get,update → required for Scale
## pods/exec: create → required for Pod Shell
rules:
- apiGroups: [""]
resources: [namespaces, pods, services, configmaps, events]
verbs: [get, list, watch]
- apiGroups: [""]
resources: [pods/log]
verbs: [get]
- apiGroups: [""]
resources: [pods/exec]
verbs: [create] # Pod Shell (optional)
- apiGroups: ["apps"]
resources: [deployments]
verbs: [get, list, watch, patch] # patch enables Restart
- apiGroups: ["apps"]
resources: [deployments/scale]
verbs: [get, update] # Scale
- apiGroups: ["networking.k8s.io"]
resources: [ingresses]
verbs: [get, list, watch]
- apiGroups: ["batch"]
resources: [cronjobs]
verbs: [get, list, watch]
If you don't grant these verbs, 2.0.0 behaves exactly like 1.0.5 for your users — read-only with the new UI. The write buttons remain hidden until admins hand out the matching application permission.
New env variables (all optional, safe defaults)
| Variable | Default | Purpose |
|---|---|---|
LOG_LEVEL | info | debug / info / warn / error |
LOG_FORMAT | text | json for log shippers |
LOG_OUTPUT | stdout | stdout / stderr |
LOG_INCLUDE_AUDIT | true | Mirror audit DB rows to console |
LOG_ADD_SOURCE | false | Add file:line to every record |
APP_ENV, APP_VERSION | — | Static labels in every log record |
MAX_REPLICAS | 100 | Upper bound for the Scale action |
Rollback
Schema additions are backward compatible to a restore, not to the running binary. If 2.0.0 misbehaves:
kubectl -n kubernetes-openconsole exec deploy/kubernetes-openconsole -- \
cp /data/app.db.pre-2.0.0.bak /data/app.db
kubectl -n kubernetes-openconsole set image deploy/kubernetes-openconsole \
openconsole=ghcr.io/vurulkan/kubernetes-openconsole:1.0.5
Image
ghcr.io/vurulkan/kubernetes-openconsole:2.0.0ghcr.io/vurulkan/kubernetes-openconsole:latest
1.0.5
2026-05-04 · on GitHub
New UI improvements
1.0.4
2026-03-17 · on GitHub
Kubernetes OpenConsole - Update Release Notes
This release adds major usability and authentication improvements across Dashboard, Admin, and Login experiences.
Highlights
-
New resources in Dashboard
- Added Ingress and CronJob support.
- CronJob cards now show:
- schedule
- last schedule time
- enabled/disabled state with color indicator (green/orange)
- Added YAML view for both Ingress and CronJob.
-
Role permissions UX improvements
- Permission management is now shown in a cleaner, responsive table grouped by namespace.
- Permissions are displayed as chips.
- Individual permissions can be removed quickly (hover delete icon).
- Entire namespace permission sets can be removed in one action.
- Role selection no longer auto-selects on refresh/re-entry (starts empty).
-
Logs modal improvements
- Fixed auto-scroll behavior so disabling auto-scroll no longer jumps to top while live logs continue.
- Initial log modal still opens at the bottom (latest logs).
- Added keyboard handling so
Ctrl/Cmd + Aselects modal content directly.
-
YAML output cleanup
- YAML responses are cleaner by removing noisy metadata fields (such as managed fields) for better readability closer to
kubectl -o yaml.
- YAML responses are cleaner by removing noisy metadata fields (such as managed fields) for better readability closer to
-
Namespace navigation search
- Added live namespace search in the left namespace panel.
- Namespace list filters as you type.
-
Login branding customization
- Added logo upload + remove in Admin > Customization.
- Logo is persisted in persistent storage and displayed on login.
- Login card now constrains large logos to prevent layout breakage.
-
Azure AD login (Phase 1)
- Added Azure AD single-tenant login alongside existing LDAP/local login.
- Users can sign in with Microsoft.
- On first successful Azure login, user is auto-created locally (JIT provisioning).
- Existing local RBAC model remains unchanged (assign roles/groups in Admin).
- LDAP remains fully supported and can run in parallel.
- Logout remains app-local only.