Skip to main content
Lumovi tries to say what went wrong in plain words, and what to try next. On the start screen, each cluster that can’t be opened gets a short label; hover it to read the full message. In a fleet, each cluster’s card shows the label and the full message. Inside a cluster, an error screen has a title, a hint and Try again.
Lumovi showing a cluster it can't reach, with a hint to check the API server, VPN or tunnel, and a Try again button.Lumovi showing a cluster it can't reach, with a hint to check the API server, VPN or tunnel, and a Try again button.

A cluster that doesn't answer, and what to try.

Connecting to a cluster

On the start screen, Credentials failed and Unauthorized both read Couldn’t sign in: hover it for which.
Lumovi couldn’t open a connection to the API server.
  • Check that the cluster is running, and that this computer can reach it. Many clusters are only reachable over a VPN or a tunnel.
  • If you have kubectl, kubectl --context <name> get --raw /version tells you whether the API server answers from this computer.
  • Behind a company’s proxy, Lumovi goes through the one your shell uses (HTTPS_PROXY), or your kubeconfig’s proxy-url. The proxy needs credentials (407) wants them in the proxy’s URL (http://user:password@proxy:3128), and The proxy didn’t open a tunnel (it answered 403) means it refused the address. A cluster on your own network may need to be in NO_PROXY. See Proxies and certificates.
  • Choose Try again, or Reload on the start screen, once the network is back.
The API server accepted the connection, but didn’t answer within 20 seconds. That’s usually a busy or distant API server.Try again in a moment. If it’s always slow, give it longer with LUMOVI_REQUEST_TIMEOUT_MS. On macOS, quit Lumovi, then:
Hover the label for which it is:
  • Its certificate isn’t signed by a certificate authority Lumovi trusts: the certificate authority in your kubeconfig doesn’t match the API server’s certificate. This happens when a cluster is recreated or its certificates rotate. Get a fresh kubeconfig entry from wherever the cluster came from, for example aws eks update-kubeconfig, gcloud container clusters get-credentials or az aks get-credentials, then choose Reload. Behind a proxy that inspects HTTPS, the proxy signs it with its own: ask for the cluster to be left uninspected, or put the proxy’s authority in your kubeconfig.
  • Its certificate has expired: the cluster’s, or the proxy’s.
  • Its certificate is for another name: the address in your kubeconfig isn’t one the certificate names. tls-server-name on the cluster can name the one it does.
  • TLS handshake failed, and why, for anything else.
See Proxies and certificates.
The cluster’s address starts with http://, for example through kubectl proxy. Lumovi only uses unencrypted connections when the kubeconfig says so for that cluster, the same rule as the official JavaScript client:
~/.kube/config
The context gets its credentials from a plugin, like gke-gcloud-auth-plugin, aws eks get-token or kubelogin, and the plugin failed.
  • The credential plugin “…” wasn’t found. Install the plugin, or make sure it’s on your PATH. On macOS and Linux, Lumovi reads PATH from your login shell when it starts, even when you open it from the Dock or a launcher. So the plugin must be on the PATH your shell profile sets up. Restart Lumovi after changing it.
  • Could not get credentials: … The plugin ran but failed, most often because you’re signed out of your cloud provider. Sign in again (gcloud auth login, aws sso login, az login…), then choose Try again.
The API server turned down the token or client certificate. Expired credentials end up here. Sign in again, or get a fresh kubeconfig entry, then try again.
Your account isn’t allowed to read this. If you only have access to some namespaces, pick one from the namespace menu. If your account can’t list namespaces, type the name of one you can use. See Namespaces and Permissions.
  • Your kubeconfig couldn’t be read. A file has a syntax error. Lumovi names the file and the first line of the problem. Fix the file, then choose Reload, or Choose a kubeconfig… to read another. When other files can be read, they still load, and a notice above the list names the one that couldn’t.
  • No clusters yet. Lumovi found no contexts. Choose a kubeconfig… reads the file you use with kubectl, and Add a cluster takes one you paste. See Connect your clusters.
  • The kubeconfig you chose is gone. A file you chose in Lumovi was moved or deleted. Choose it again where it is now, or go Back to KUBECONFIG and ~/.kube/config.
kubectl in your terminal uses the KUBECONFIG your shell profile exports. Lumovi opened from the Dock, Spotlight or a launcher doesn’t see that variable (it takes only PATH and the proxy from your shell), so it reads ~/.kube/config. The kubeconfig files button, at the bottom of the start screen, shows which files it read.Choose your files in Lumovi, which keeps them, start Lumovi from your terminal, or set KUBECONFIG where apps see it. See Setting environment variables.

While you work

Can’t reach … Checking again every 15 seconds. The connection dropped. The last data stays on screen, and Lumovi reconnects by itself when it can. Choose Retry now to check at once, or All clusters to open another.Couldn’t refresh — showing the last data. A refresh failed, but what you see is still the latest Lumovi has. Choose Retry.
Lists load in chunks of 500 and stop at 5,000 objects, so huge clusters stay fast. A note says Showing the first … of …. Choose Filter by label to narrow the list on the server, pick a namespace, or raise the limit with LUMOVI_MAX_LIST_ITEMS.On the Workloads page and plugin pages, the note says Some lists are too long to load whole, and a kind that can’t be listed at all says Couldn’t list … above the rest.
Lumovi asks the cluster what you’re allowed to do before offering an action, and says why it can’t:
  • Your account can’t … in …. Your RBAC doesn’t allow it. See Permissions.
  • Changes are turned off for this cluster. You, LUMOVI_READ_ONLY, or your organization’s policy made it read-only. On a server, it’s read-only for everyone, and says who made it so: “Changes are turned off for this cluster: jane@example.com made it read-only for everyone.” Ask them, or Lumovi’s admins. See Read-only mode.
As the desktop app started, it couldn’t use something it was given, and says what, and why:
  • Your organization’s policy, which then locks the most it could: see below.
  • A certificate authority’s file the policy names: The certificate authorities in … can’t be read: …. Lumovi trusts the others, and the system’s.
  • A proxy that isn’t a URL, in HTTPS_PROXY or HTTP_PROXY: HTTPS_PROXY isn’t a proxy’s URL: give it as http://host:port (or https://). Lumovi connects without it. Check your shell’s profile, or the environment variable.
See Proxies and certificates.
On a server, a cluster’s read-only switch, where its metrics come from, and where its node shells run are the same for everyone on it, and only Lumovi’s admins change them: “Only Lumovi’s admins change what’s set for everyone on this server.” Ask one of them. A server without admins lets anyone signed in change them. See Read-only mode.
Your organization’s policy can’t be used, so Lumovi locks the most it could until it’s put right: every cluster read-only, AI assistants off, no kubectl got to match a cluster (terminals use the one installed), and no kubeconfig files read but those in KUBECONFIG, or ~/.kube/config. As it starts, a notice says so, titled “Lumovi started without something it was given”, and the Read-only badge next to an object’s actions says why: “Lumovi can’t use your organization’s policy (…), so it changes no cluster until it’s put right.” Send that to your IT: the policy isn’t an administrator’s alone, isn’t JSON, has a key Lumovi doesn’t know, or, on Windows, couldn’t be read from the registry (where people may run neither reg.exe nor PowerShell, let them run reg.exe). See When it can’t be used.
Someone else changed the object while you were editing its YAML. Lumovi doesn’t overwrite their change. Choose Start over from the latest, and make your edit again. See Edit YAML.
Images built without a shell (distroless ones, say) can’t run one. Choose Debug to add a debug container with tools to the pod instead. See Shells and debug containers.Shells are off means the cluster is read-only in Lumovi: shells count as changes. Terminals on your computer aren’t covered.
Lumovi starts a privileged pod on the node for it, and says why when it can’t:
  • … doesn’t allow privileged pods: its Pod Security level is stricter than privileged. Choose Settings in the node’s Shell tab, and a namespace that allows privileged pods, like kube-system.
  • … couldn’t start … (ErrImagePull): …. The node can’t pull the image. Choose one it can pull in Settings, like a copy in your own registry.
  • … didn’t start the node shell’s pod in … seconds. The node’s kubelet may not be running, or the image takes long to pull.
  • … has no shell of its own. Talos and Bottlerocket nodes have none. Choose Shell in the pod.
  • … has no nsenter. A shell on the node itself needs the image’s nsenter. Choose an image that has it, like alpine, in Settings, or a shell in the pod.
  • No node shell access. Your account can’t create pods, read them, or open shells in them, in the node shells’ namespace. Ask for them, or choose another namespace in Settings (on a server, Lumovi’s admins do, for everyone).
  • Node shells need a Linux node. Windows nodes can’t have one.
  • Node shells are off. The Lumovi server you use turned them off.
See Shells on nodes.
Your shell, … isn’t there: set SHELL to one that is. SHELL names a shell that isn’t on your computer. Set it to one that is, and start Lumovi again. Without SHELL, terminals start zsh on macOS and bash on Linux. On Windows, they start Windows PowerShell.PowerShell is still starting. On Windows, a terminal says so when PowerShell has drawn nothing for 5 seconds; the line goes once it draws. Two things usually hold it up. Your profile: Measure-Command { . $PROFILE }, in PowerShell, says how long it takes, and moving slow work out of it helps every terminal. A virus scanner that checks PowerShell as it starts: your IT can tell you whether it does, and whether that can be eased.kubectl uses another cluster. Your shell’s startup files probably set KUBECONFIG, which replaces the terminal’s own. Have them add to it instead, so the terminal’s file stays first. See Terminals.
Lumovi couldn’t give the terminal a kubectl that matches its cluster, and its first line says why, after “It’s the one on your PATH:”:
  • Lumovi couldn’t ask the cluster its version (it didn’t answer within 3 seconds). The cluster isn’t reachable now, and Lumovi has never learned its version: connect its VPN, say, and open another terminal.
  • the cluster says it’s Kubernetes 1.13, and Lumovi gets kubectl for 1.25 or later. Lumovi gets no kubectl older than 1.25. Use a kubectl of your own that suits that cluster.
  • Lumovi gets kubectl only over HTTPS, and … isn’t (LUMOVI_KUBECTL_MIRROR). LUMOVI_KUBECTL_MIRROR names a mirror over plain http, or isn’t a URL. Give an https one, or unset it.
  • kubectl v1.34.9 is still downloading, for the terminals opened once it’s done. Open another terminal once it’s done.
  • dl.k8s.io didn’t answer within 10 seconds, or Lumovi couldn’t get kubectl 1.34 to match the cluster (…): dl.k8s.io can’t be reached from this computer, and Lumovi kept none of that version. Behind a company’s proxy, see Proxies and certificates. Lumovi tries again 10 minutes later. Your organization can name a mirror in its policy.
  • Lumovi couldn’t get kubectl 1.34 to match the cluster (dl.k8s.io’s kubectl v1.34.9 couldn’t be verified as Kubernetes’ own, and may have been changed: …), or (… published no signature for kubectl v1.34.9): the download isn’t signed by Kubernetes’ release, or its signature doesn’t hold, so it isn’t used. A proxy that changes what it passes on can do that. If it’s a mirror’s, check what it keeps. See A kubectl that matches the cluster.
  • (… couldn’t be verified as Kubernetes’ own, so it may have been changed, or Kubernetes has changed how it signs and Lumovi needs an update: …), or the same with Sigstore has changed its certificate authority or Sigstore has changed its logs: it’s signed, but not as this Lumovi knows Kubernetes signs, or not under Sigstore’s authority and logs as it knows them. If Kubernetes or Sigstore has changed them, every kubectl signed since is refused until you update Lumovi, once a release knows the new ones; meanwhile, terminals use yours, or the newest one kept. If neither has, it may have been changed, as above.
  • (artifacts.corp.example.com published no signature for kubectl v1.34.9, and your organization’s policy requires one): your organization’s policy requires Kubernetes’ signature from its mirror, and the mirror keeps none for that kubectl. Whoever runs the mirror can add Kubernetes’ .sig and .cert beside it.
A terminal with no such line at all has View → Match kubectl to Each Cluster turned off, by you or your organization. One that says … has no Kubernetes signature for it, so Lumovi checked it against its SHA-256 only. got its kubectl from a mirror that keeps no signatures.Just after a cluster’s upgrade, the first terminal you open can still get the kubectl for its old version: Lumovi remembers each cluster’s version as it last said it, and asks again for the next. Open another terminal.
Port … on this computer can’t be used (EADDRINUSE). Something else on your computer listens on that port. Pick another local port. See Port forwarding.
The object refers to something that isn’t there: a ConfigMap or Secret it mounts, the service account it runs as, an ingress’s backend service, or the node a pod is on. The card is dashed, and can’t be opened. Create what’s missing, or fix the reference.A card without a status is different: Lumovi couldn’t list that kind, usually because your account can’t, so it doesn’t know whether the object exists. A route’s gateway or service in another namespace shows the same way, since the map lists only the object’s own namespace: click it to open it. See Map.
Lumovi didn’t expect this. Choose Try again first. When a page itself fails, its screen says Lumovi ran into an unexpected problem. Your clusters were not changed., with Reload, Back to clusters, Copy details, and Report issue, which opens a GitHub issue with the error filled in. A detail panel that fails says This object couldn’t be displayed (or This release couldn’t be displayed) the same way, and the list next to it keeps working.
An address Lumovi doesn’t know, like an old link, says This page doesn’t exist, with Go to the overview, or Back to clusters outside a cluster. For a custom kind, a page saying the cluster doesn’t serve it means the cluster doesn’t have that kind: Its CustomResourceDefinition may have been removed, or never installed in this cluster.

Usage and metrics

Live usage comes from metrics-server, as for kubectl top. Without it, the overview says Live usage needs metrics-server, and Lumovi shows requests and limits against capacity instead. Install metrics-server in the cluster to see live usage. See Live usage.
History comes from a Prometheus or VictoriaMetrics in the cluster.
  • No Prometheus found. None of the cluster’s services looks like Prometheus or VictoriaMetrics. If yours runs under another name, choose Choose a service to pick its namespace, service, port and path (/select/0/prometheus for vmselect). Look again searches again, after you install one, say.
  • Can’t read usage history. There’s a source, or something that looked like one, but reading it failed. The message under it says why:
    • Your account can’t list services, so Lumovi can’t look for Prometheus. Choose its service yourself, or ask for list on services.
    • Your account can’t reach … through the API server (it needs get on services/proxy). Lumovi reaches the source through the API server with your credentials, so you need get on services/proxy in its namespace.
    • … didn’t answer PromQL or It answered, but not like Prometheus. The service, port or path isn’t a PromQL endpoint. Choose the right one.
  • Usage history is off. It was turned off for this cluster. Choose Change to turn it back on.
See Usage history.
Right-sizing needs usage history, and enough of it. A workload’s note says what’s missing:
  • Too new: it has less than a day of history. Recommendations need a day, and grow more confident up to a week.
  • No usage: Prometheus has no CPU and memory samples for its containers in the last week, or it had no pods.
  • Autoscaled: a VerticalPodAutoscaler sets its requests, so Lumovi leaves it to it.
If Prometheus couldn’t answer for some namespaces, an alert names up to three, and how many more, with Try again: the other namespaces still get their recommendations. A namespace that keeps failing on its own usually asks more of Prometheus than its query limits or timeout allow.If VerticalPodAutoscalers couldn’t be listed, an alert says so: workloads they manage get recommendations too, rather than Autoscaled. See Without nodes or VerticalPodAutoscalers.If Apply… is grayed out, your account can’t change the workload, or the cluster is read-only. If the dialog says the API server refused the change, a quota, a LimitRange or an admission webhook turned it down: untick the change it refuses, or fix the cause first. See Right-sizing.

Helm

Lumovi ships with helm, so this happens only when LUMOVI_HELM names a helm that isn’t there, or can’t run, or when you run it from source (npm run dev) without helm on your login shell’s PATH. Set LUMOVI_HELM to a helm that runs, or unset it to use the one Lumovi ships with.
The error shown is what helm said. A failed upgrade is often a chart or values problem, which the dry run usually catches first.If the release is managed by Flux, Lumovi says so and links to its HelmRelease. Flux puts back changes made any other way, so make the change in Flux instead.
Behind a company’s proxy, the message says what the proxy or the certificate made of it:
  • The proxy needs credentials (407): give them in its URL, as http://user:password@host:port.
  • The proxy didn’t open a tunnel (it answered 403).: the proxy doesn’t allow the address.
  • Its certificate isn’t signed by a certificate authority Lumovi trusts (…): a proxy that inspects HTTPS signs with an authority your computer doesn’t trust yet. IT installs it in your system’s certificates, or names it in your organization’s policy. On a server, give it with extraCA.
See Proxies and certificates, or for a server, Proxies and certificate authorities.

AI assistants

Lumovi can’t use port 47830 (listen EADDRINUSE: …): choose another one. Another program on your computer listens on that port. Type another one, from 1024 to 65535, in Port, on the Connect tab of the AI assistants page, and choose Use. Then set up again each assistant that connects over HTTP: the address it has names the old port. Claude Desktop, through Lumovi’s bridge, finds the new one by itself. See The port.
The assistant usually shows what Lumovi answered:
  • Lumovi needs its token: set this assistant up again from Lumovi. It sent no token, or an old one, after New token…, say. Claude Desktop is told something else: see Claude Desktop doesn’t use Lumovi, below. Set it up again from the AI assistants page. See Connect an assistant.
  • That session has ended: start a new one. Lumovi was restarted, turned off and on, or let the assistant go after six hours without a word from it. Most assistants start a new session by themselves. If yours doesn’t, reconnect it, or restart it.
  • Web pages may not connect to Lumovi. The request came with an Origin header, as a browser’s requests do. Tools that run in a browser can’t connect.
  • Only this computer may connect. The request was addressed to a name other than 127.0.0.1 or localhost. Use the address the dialog shows.
  • That’s more than Lumovi takes: 4 MB at most. A message, like a manifest to apply, was larger than that.
When it can’t connect at all, AI assistants are off, Lumovi isn’t running, or the assistant has an old port. The AI assistants page shows where Lumovi listens: Listening at …
Add to Claude Desktop says why it couldn’t:
  • Claude Desktop isn’t installed on this computer. Lumovi found no folder of Claude Desktop’s settings. Install it, open it once, and try again.
  • Claude Desktop’s settings (…) aren’t valid JSON, so Lumovi left them alone. Fix the file it names, and choose Add to Claude Desktop again.
Once it’s added, quit Claude Desktop and open it again. Then, Claude Desktop shows what Lumovi answered:
  • AI assistants are turned off in Lumovi: open Lumovi, and turn them on in its AI assistants settings.
  • Lumovi didn’t start: open it, and check that AI assistants are on. Lumovi wasn’t running, and didn’t listen within 30 seconds of being started. Open it yourself.
  • Lumovi isn’t answering: open it, and try again.
  • What listens on Lumovi’s port (47830) can’t show it’s Lumovi, so Lumovi’s token isn’t sent to it: open Lumovi, and set this assistant up again. Another program listens on Lumovi’s port while Lumovi isn’t running, or the token in Lumovi’s settings isn’t the one Lumovi uses. Lumovi’s token isn’t sent there, and Lumovi isn’t started over it. Open Lumovi, and if it can’t use its port, choose another; then choose Add to Claude Desktop again.
  • Lumovi isn’t running: open it, and try again. Lumovi isn’t open, and can’t be started for Claude Desktop yet: Lumovi notes how to start it each time you open it. Open it yourself.
  • Lumovi’s settings name no port it listens on: open Lumovi, and choose AI assistants’ port again. The port in Lumovi’s settings isn’t one Lumovi could listen on, so Claude Desktop’s messages, and the token with them, aren’t sent anywhere. Choose the port again on the AI assistants page.
Moved Lumovi? Add it to Claude Desktop again: Claude Desktop’s settings name where Lumovi’s app was when you added it, and choosing Add to Claude Desktop again updates them. Claude Desktop isn’t made for Linux, so there’s no Add to Claude Desktop there.
Some changes are refused before you’re asked. Only the assistant is told why:
  • … wouldn’t work: …. The cluster refused the dry run: your account can’t make the change, the cluster is read-only, or the object isn’t valid. The rest of the message is the cluster’s.
  • Lumovi doesn’t let AI assistants change …: the person’s AI permissions say so (…). Your permissions say Never there, by the rule named, or the defaults. Check a namespace, on the Permissions tab, says which. When it says “this server’s administrator says so”, it’s a limit only they can change.
  • Lumovi doesn’t show AI assistants the Secrets in … or Lumovi doesn’t let AI assistants read logs in …: the same, for Secrets and logs. See What assistants see.
Lumovi tells you about the others too:
  • Not approved in time. Nobody answered within five minutes.
  • … withdrew a change. The assistant gave up on a call that was waiting for the change: it was stopped, or it ended its session, or you turned assistants off. On a server, also when you disconnected it, or your session ended. Ask it again.
  • … wasn’t made: … changed after it was shown (…). You approved it, but it would no longer do what the diff showed: the object changed meanwhile, or, for a deletion, it’s another object under the same name. Lumovi tells the assistant to ask again, so you see it as it is now.
  • … failed: …. You approved it, but the cluster refused it then: the object is gone, say. The rest of the message is the cluster’s.
Still waiting for the person’s answer in Lumovi isn’t an error: the call stopped waiting after about 50 seconds, and the change still waits for you. The assistant calls wait_for_change to hear how it went. That says No change with id … is waiting when the change was answered a while ago, or never asked for.See Approving changes.
On a Lumovi server, an assistant signs in through Lumovi, in your browser. Its first try is answered Sign in to Lumovi.: that’s how it learns where to sign in, and it opens Lumovi for you to allow it.If the page for allowing it says “This request to allow an assistant can’t be answered:”, the rest says why:
  • It doesn’t say which assistant it is. or It goes back somewhere the assistant didn’t name. The request didn’t come from the assistant as it registered. Start again from the assistant.
  • It’s for …, not Lumovi’s …. The assistant was set up with another address. Set it up again with the one the AI assistants page shows.
  • Allow assistants from Lumovi’s own page. Lumovi doesn’t recognize the address the page was opened at as its own: the administrator should set url.
  • This server’s administrator has turned AI assistants off. So does the AI assistants page.
An assistant that can’t register at all hears redirect_uris must go back to this computer (http://localhost), to an app (its own scheme), or to a site this server allows (LUMOVI_ASSISTANT_REDIRECT_HOSTS). It wants to be sent back to a site the server doesn’t name. Assistants that run on your computer, or open an app, don’t: for one on a site of its own, ask the administrator.If the assistant can’t find where to sign in at all, the server is below a path whose addresses for signing in don’t reach Lumovi, or a proxy in front of it asks the assistant to sign in to it. See AI assistants for your team.An assistant you allowed has to sign in again when you sign out, or your session ends, and when Lumovi let it go because a token it had used already came back. A restart lets go of none.
Lumovi restarted since alice@example.com last used it: open Lumovi in your browser, and Claude Code picks up where it left off. You sign in with a token, or your team’s single sign-on passes your own token on, and Lumovi keeps it sealed with a key only your browser has. Open Lumovi, and the assistant carries on. See When Lumovi restarts.

The audit log

On a server, the audit log shows you what you did, and what your assistants did as you. Its auditors see everyone’s, and only they have Who and Check integrity. The page says which you are. The administrator names auditors: see Auditors.
It lists each place an event doesn’t follow from the one before it, and why. Two can happen with nothing changed: a line left unfinished as the computer or pod stopped, and an event the history couldn’t keep, which an Audit events lost event explains. See Check it hasn’t changed.
A search that has looked through many events without filling the list stops, so a filter that matches few doesn’t keep the page waiting. Look further back goes on from there.Every line read counts. An auditor reads how many it looked through; anyone else, “Lumovi looked as far back as one search goes.” Choosing how far back makes each search shorter.
An export holds what the search and filters find, in the time they cover, up to 50,000 events: the most recent, when there are more. Its notification says so. Export a shorter time at once, and the rest after.
On a server, its auditors see it at the top of the Audit log page, and the server’s log says so too. See When events are lost.What was let go, and why, is in the server’s log and in an Audit events lost event, every minute and as Lumovi stops.

Installing

The installer is signed, by Open Source Developer Péter Kóta, but its certificate is new, and Windows warns about a publisher until enough people have installed what it signs. Choose More info, check that the publisher is Open Source Developer Péter Kóta, then Run anyway. You can check the download, its checksum and signature, first.
Make the file executable first: chmod +x Lumovi-*.AppImage. Then run it from a terminal to see any error it prints.

In your cluster

Each page keeps a WebSocket open to Lumovi, and Reconnecting to Lumovi… shows while it’s down. If it drops every minute or so, something in front of Lumovi closes idle connections. Give your ingress a long timeout, for example with ingress-nginx:
values.yaml
Lumovi also checks each connection every 30 seconds (LUMOVI_HEARTBEAT_SECONDS). Keep that shorter than the idle timeouts of the proxies in between. See Address and ingress.
LUMOVI_CA_FILE, which the chart sets from extraCA, names a file Lumovi can’t read, one without certificates in it, or one with something that isn’t a certificate:
  • The certificate authorities in /etc/lumovi/ca/ca.crt can’t be read: ENOENT …: the ConfigMap or Secret has no key extraCA.key (ca.crt unless set). Check the key it has with kubectl describe configmap.
  • /etc/lumovi/ca/ca.crt has no certificates in it (PEM, as -----BEGIN CERTIFICATE-----).: the file isn’t PEM. Convert a DER certificate with openssl x509 -inform der -in corp-root.cer -out corp-root.pem.
  • … can’t be read: its certificate 2 isn’t one (…): the second block between -----BEGIN CERTIFICATE----- and -----END CERTIFICATE----- is damaged. Check it with openssl x509 -noout -text.
See Proxies and certificate authorities.
With proxy.secret, the pod needs that Secret, in Lumovi’s namespace, with the key HTTPS_PROXY. Without either, Kubernetes doesn’t start the container, and kubectl describe pod says which is missing. Create it, or fix its key’s name: see Proxies and certificate authorities.
HTTPS_PROXY or HTTP_PROXY, which the chart sets from proxy, can’t be read as a URL: HTTPS_PROXY isn’t a proxy’s URL: give it as http://host:port (or https://). Lumovi doesn’t say what it was, as it can hold credentials. Check it, and encode characters like @ in a password as %40. See Proxies and certificate authorities.
More than one replica (replicaCount) needs auth.keepSessions: without it, each replica keeps who's signed in, and what's set for everyone (the clusters made read-only, say), in its own memory, and they don't agree. Leave auth.keepSessions on, or run one replica. More than one also needs assistants.enabled: false, audit.persistence.enabled: false, and sticky sessions. See replicaCount.
Its settings were deleted, or an older copy of them put back, where the server keeps them (the Secret <release>-state, or state.json), by whoever can write there, not through Lumovi. Lumovi kept the stricter: read-only if either copy was, and its metrics source and node shells as it last set them. The banner says what it did: “production’s read-only was turned off outside Lumovi, and Lumovi made it read-only again.”, “production was made read-only outside Lumovi.”, or “What’s set for production was changed outside Lumovi, and Lumovi put it back.” The audit log has it, by Lumovi itself, with outside: true.
  • Choose Keep it read-only or Allow changes (Keep it changeable or Make it read-only, where it’s changeable). Either is your own change, recorded as yours, and the banner goes.
  • Find who could write there: whoever may patch that Secret, or write the file, and controllers that sync Secrets into Lumovi’s namespace, or put a backup back.
See Changed outside Lumovi.
Behind a proxy that inspects HTTPS, Lumovi gets a certificate signed by the proxy’s own authority. The sign-in page says Signing in didn’t work, and Lumovi’s log, or the Audit log page for the webhook, says Its certificate isn’t signed by a certificate authority Lumovi trusts (…). Give Lumovi that authority with extraCA, or leave the address out of the proxy with proxy.noProxy. Its certificate has expired and Its certificate is for another name are said the same way. See Proxies and certificate authorities.
The chart keeps the audit log on a volume, from the cluster’s default storage class. Without a default, the volume’s claim waits, and so does the pod. Set audit.persistence.storageClass, or audit.persistence.enabled: false. See Where it’s kept.
Two Lumovis were given one audit folder: the second waits for the first to let go, about 40 seconds at most, then doesn’t start. Run one, or give each a folder of its own. See Where it’s kept.
Lumovi’s admins don’t let you do that there: their access rules say less than your RBAC. See your access says what you may do, and which grant or limit decides. Disabled actions and tabs like Shell and Logs say the same, with who the admins are. Ask one of them for more.
LUMOVI_ACCESS (the chart’s access.policy), or what admins set and Lumovi kept, doesn’t make sense, and the log says where: LUMOVI_ACCESS: grants[0].who[0] names a group there isn’t: “nobody”. Lumovi stops rather than let anyone do more. See Access for your team.
A restart, as an upgrade does, signs nobody out, unless Lumovi had nowhere to keep sessions, couldn’t, or can’t open what it kept:
  • auth.keepSessions: false, or without the chart, neither LUMOVI_STATE_SECRET nor LUMOVI_DATA_DIR. The log says, as Lumovi starts: Who’s signed in, the AI assistants they allowed, and people’s pins and saved views are kept in memory: they’re lost when Lumovi stops., then, in a cluster, The Helm chart’s auth.keepSessions keeps them (LUMOVI_STATE_SECRET)., and Run a single replica so: …
  • The key changed, or was lost: 12 of what Lumovi kept doesn’t open with its key (it changed, they were, or another replica uses another): left as they are, unused, until they’ve been so for 15 minutes, then let go. With Argo CD, or anything else that renders the chart with helm template, each render makes a new key: give your own, in auth.stateKeySecret. See Sessions across restarts.
  • Writing it failed, before the restart: Lumovi can’t keep who’s signed in (…): if it restarts now, what changed since is lost., then Lumovi keeps who’s signed in again. once it can.
  • It didn’t all fit in a megabyte: What Lumovi keeps would be more than the 1000 kB it may hold: the 40 sessions that end soonest aren’t kept (they sign in again after a restart).
  • What’s kept can’t be read, edited by hand, say: What Secret lumovi/lumovi-state kept can’t be read (…): everyone signs in again., with why in the brackets
  • Lumovi was uninstalled, which deletes the Secret and its key.
  • More than one replica, without sticky sessions: each knows the sessions started on it.
Sessions also end after 12 hours, unless you set it otherwise. See Across restarts.
  • Lumovi can’t read Secret lumovi/lumovi-state, where it keeps who’s signed in, the AI assistants they allowed, and people’s pins and saved views: …. The Secret isn’t there, or Lumovi may not get it: with rbac.create: false, give its service account get and patch on <release>-state.
  • LUMOVI_STATE_SECRET needs LUMOVI_STATE_KEY: the key what’s kept is sealed with, from a Secret of its own (the chart makes one). Without the chart, set both.
  • LUMOVI_STATE_KEY must be at least 32 characters, random ones. The key in auth.stateKeySecret’s Secret is too short: make one with openssl rand -base64 48.
Kubernetes doesn’t start the pod at all when the key’s Secret, or its key, isn’t there (CreateContainerConfigError). See LUMOVI_STATE_SECRET.
The token is wrong, expired or revoked. Create a new one, for example kubectl create token NAME --namespace NAMESPACE. Lumovi asks the cluster who a token belongs to with a SelfSubjectReview, which needs Kubernetes 1.28 or later. See Tokens.
The token you signed in with expired, or the cluster stopped accepting it. Sign in again to carry on where you were. With single sign-on, Lumovi renews tokens by itself when the provider gives it a refresh token.In a fleet, one cluster turning your token down doesn’t end your session: its card says Unauthorized, and the other clusters keep working.
Lumovi checks a fleet’s settings when it starts, and stops with a message that names the one to fix:
  • A fleet needs single sign-on or a proxy (LUMOVI_AUTH=oidc or proxy): a pasted token only says who someone is to one cluster.
  • LUMOVI_FLEET_SECRETS reads Secrets of the cluster Lumovi runs in, and it isn’t running in one, or LUMOVI_FLEET_LOCAL is true, but Lumovi isn’t running in a cluster: outside Kubernetes, use a kubeconfig or agents instead.
  • LUMOVI_FLEET_KUBECONFIG isn’t a kubeconfig: …
  • LUMOVI_FLEET_AGENTS[0] (private-eu) needs a token of at least 32 characters, or its tokenSha256.
See Configuration.
The agent’s log says why it stopped or what it’s waiting for:
  • The hub at … refused this agent: it doesn’t know this name and token. Lumovi’s log says An agent was refused, with the name. The token’s SHA-256 was taken with a trailing newline, the agent’s name (clusterName, in-cluster unless you set it) isn’t the one Lumovi lists, Lumovi wasn’t restarted after its list of agents changed, or a proxy in front of Lumovi answers api/agent itself.
  • … refused this agent: another agent connected as it (do two clusters use the same name and token?). Give each agent its own name and token.
  • Can’t reach the hub at …, then Connecting again in … s: the agent can’t reach Lumovi’s address. It keeps trying. Behind a proxy, give it proxy in the chart: The proxy … needs credentials to … (407) wants them in its URL, and … didn’t open a tunnel to … refused the address. See Proxies and certificate authorities.
  • Nothing from the hub in 61 s: something between them, often a proxy’s idle timeout, drops a quiet connection. Lumovi pings every 30 seconds by default, so let idle connections live longer than that.
  • LUMOVI_HUB_URL must be https, not http://…: the agent’s token and its cluster’s would cross the network as they are. The agent stops: give Lumovi’s https address. http reaches only a Lumovi on the same machine (localhost, 127.0.0.1 or [::1]). The chart refuses agent.hubUrl over http the same way, but on the same machine.
Until it’s connected, and Lumovi trusts its cluster, the agent’s pod isn’t ready. Until it’s connected, its card says Its agent isn’t connected. See Agents.
The agent is connected, but sent a certificate authority other than the one Lumovi trusts for its cluster, and its log says The hub doesn’t trust this cluster:, and why. Lumovi’s audit log has it as agent.refused.
  • Its cluster’s certificate authority changed, rotated or rebuilt: an admin (or anyone, where Lumovi has none) trusts the new one with Trust it… on the page of every cluster, pasting its SHA-256 from the cluster itself: kubectl get configmap kube-root-ca.crt -n default -o jsonpath='{.data.ca\.crt}' | openssl x509 -noout -fingerprint -sha256, run with your own access to that cluster, not through Lumovi. Lumovi trusts it only if the agent sends that one. Where Lumovi has no admins, anyone signed in can. Or set its caSha256 in LUMOVI_FLEET_AGENTS, and restart Lumovi.
  • It didn’t change: someone else has the agent’s token, and connected with it. Don’t trust it. Make a new token, and give its SHA-256 to Lumovi.
  • The card says Certificate error, with “other than the one LUMOVI_FLEET_AGENTS names for it”: its caSha256 is set, and isn’t this one. Change it there, and restart Lumovi.
See Its cluster’s certificate authority.
Lumovi reads a fleet’s kubeconfig files and Secrets again every 30 seconds, and a Secret mounted into a pod can take a minute or two to change there. If it still isn’t there, Lumovi’s log says why: … can’t be read: … for a kubeconfig file, … can’t be listed: … for Secrets it may not list, or Two clusters are called …: the one from … is left out (the one from … stays). See A fleet of clusters.
A fleet shows each person only the clusters shared with their groups. One that isn’t shared with you isn’t on the page of every cluster, and a link to it says This server has no cluster called “staging” that you can see. If none is shared with you, the page says No clusters for you here. Ask whoever runs Lumovi to share it with one of your groups. See Clusters in a fleet.
Its card has a label and the reason, like Unreachable with Its agent isn’t connected., or Credentials failed with It takes each person’s own token, and this server doesn’t pass tokens on: set LUMOVI_OIDC_FORWARD_TOKEN. Opened, it shows the error with Choose another cluster, and a banner with All clusters. The other clusters keep working. See A fleet of clusters, and Agents for a cluster behind one.

Reporting a problem

If none of this helps, please open an issue. In the desktop app, Help → Report an Issue… opens the same page. It helps to include:
  • your Lumovi version, shown at the bottom of the sidebar,
  • your operating system, and how the cluster is run (EKS, GKE, kind…),
  • what you did, what you expected, and what happened instead,
  • the details from the error screen, which Copy details copies.
For security problems, please don’t open an issue: see Reporting a vulnerability.