WP-12: Kubernetes overview and workload actions

kube-rs gateway (nodes, namespaces, deployments/statefulsets/daemonsets with
images, PVCs; rollout restart, scale, set image via patches) using the
microk8s kubeconfig by default, fake gateway for dev, /api/cluster routes,
Kubernetes page with node cards, workload table with admin actions and PVC
list. Also implements the streaming command runner that WP-11 relied on.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

ROADMAP.md

@@ -357,7 +357,7 @@ The server is reachable via `ssh softvisor` (as root). Findings from the inspect
 | WP-02 | M1 (shell) / M2 (rest) | done | shell 2026-09-02; encrypted settings, SMTP, job runner, scheduler 2026-09-02 |
 | WP-10 | M2 | done | 2026-09-02; snaps inventoried, no upstream check |
 | WP-11 | M2 | done | 2026-09-02; runs as root via systemd, sudoers scoping deferred to WP-41; log via polling |
-| WP-12 | M2 | todo | |
+| WP-12 | M2 | done | 2026-09-02; overview, restart, scale, set image; upstream tag check not done |
 | WP-20 | M3 | todo | |
 | WP-21 | M3 | todo | |
 | WP-30 | M4 | todo | |

backend/crates/api/src/cluster.rs

@@ -0,0 +1,96 @@
+//! /api/cluster: Kubernetes overview and workload actions.
+use axum::extract::{Path, State};
+use axum::http::StatusCode;
+use axum::routing::{get, post};
+use axum::{Json, Router};
+use domain::cluster::{ClusterOverview, WorkloadKind, WorkloadRef};
+use domain::DomainError;
+use serde::{Deserialize, Serialize};
+use utoipa::ToSchema;
+
+use crate::error::ApiError;
+use crate::extract::{AdminUser, AuthUser};
+use crate::AppState;
+
+pub fn router() -> Router<AppState> {
+    Router::new()
+        .route("/overview", get(overview))
+        .route("/workloads/{ns}/{kind}/{name}/restart", post(restart))
+        .route("/workloads/{ns}/{kind}/{name}/scale", post(scale))
+        .route("/workloads/{ns}/{kind}/{name}/image", post(set_image))
+}
+
+#[derive(Serialize, ToSchema)]
+pub struct OverviewResponse {
+    #[serde(flatten)]
+    #[schema(value_type = Object)]
+    pub overview: ClusterOverview,
+    pub images: Vec<String>,
+}
+
+#[utoipa::path(get, path = "/api/cluster/overview", tag = "cluster", security(("bearer" = [])), responses((status = 200, body = OverviewResponse), (status = 502)))]
+async fn overview(
+    State(state): State<AppState>,
+    _: AuthUser,
+) -> Result<Json<OverviewResponse>, ApiError> {
+    let overview = state.cluster.overview().await?;
+    Ok(Json(OverviewResponse {
+        images: overview.images(),
+        overview,
+    }))
+}
+
+fn workload_ref((ns, kind, name): (String, String, String)) -> Result<WorkloadRef, DomainError> {
+    let kind = WorkloadKind::parse(&kind).ok_or(DomainError::NotFound)?;
+    Ok(WorkloadRef {
+        namespace: ns,
+        kind,
+        name,
+    })
+}
+
+#[utoipa::path(post, path = "/api/cluster/workloads/{ns}/{kind}/{name}/restart", tag = "cluster", security(("bearer" = [])), responses((status = 204), (status = 404)))]
+async fn restart(
+    State(state): State<AppState>,
+    _: AdminUser,
+    Path(p): Path<(String, String, String)>,
+) -> Result<StatusCode, ApiError> {
+    state.cluster.restart(workload_ref(p)?).await?;
+    Ok(StatusCode::NO_CONTENT)
+}
+
+#[derive(Deserialize, ToSchema)]
+pub struct ScaleRequest {
+    pub replicas: i32,
+}
+
+#[utoipa::path(post, path = "/api/cluster/workloads/{ns}/{kind}/{name}/scale", tag = "cluster", security(("bearer" = [])), request_body = ScaleRequest, responses((status = 204), (status = 422)))]
+async fn scale(
+    State(state): State<AppState>,
+    _: AdminUser,
+    Path(p): Path<(String, String, String)>,
+    Json(req): Json<ScaleRequest>,
+) -> Result<StatusCode, ApiError> {
+    state.cluster.scale(workload_ref(p)?, req.replicas).await?;
+    Ok(StatusCode::NO_CONTENT)
+}
+
+#[derive(Deserialize, ToSchema)]
+pub struct ImageRequest {
+    pub image: String,
+    pub container: Option<String>,
+}
+
+#[utoipa::path(post, path = "/api/cluster/workloads/{ns}/{kind}/{name}/image", tag = "cluster", security(("bearer" = [])), request_body = ImageRequest, responses((status = 204), (status = 404), (status = 422)))]
+async fn set_image(
+    State(state): State<AppState>,
+    _: AdminUser,
+    Path(p): Path<(String, String, String)>,
+    Json(req): Json<ImageRequest>,
+) -> Result<StatusCode, ApiError> {
+    state
+        .cluster
+        .set_image(workload_ref(p)?, req.container, &req.image)
+        .await?;
+    Ok(StatusCode::NO_CONTENT)
+}

backend/crates/api/src/config.rs

@@ -9,6 +9,8 @@ pub struct Config {
     pub master_key: String,
     /// Use fake host adapters (dev machines without apt/kubectl).
     pub fake_host: bool,
+    /// Path to a kubeconfig; None infers. Defaults to the microk8s client config if present.
+    pub kubeconfig: Option<String>,
     pub bind: SocketAddr,
     pub bootstrap_admin: Option<(String, String)>,
     pub cookie_secure: bool,
@@ -32,6 +34,12 @@ impl Config {
             jwt_secret,
             master_key,
             fake_host: env("FAKE_HOST").is_some_and(|v| v == "true" || v == "1"),
+            kubeconfig: env("KUBECONFIG").or_else(|| {
+                let microk8s = "/var/snap/microk8s/current/credentials/client.config";
+                std::path::Path::new(microk8s)
+                    .exists()
+                    .then(|| microk8s.to_string())
+            }),
             bind: env("BIND")
                 .unwrap_or_else(|| "127.0.0.1:8080".into())
                 .parse()?,

backend/crates/api/src/lib.rs

@@ -1,5 +1,6 @@
 //! HTTP API layer (axum). `build_app` is used by both the binary and the integration tests.
 pub mod auth;
+pub mod cluster;
 pub mod config;
 pub mod error;
 pub mod extract;
@@ -15,17 +16,18 @@ use std::sync::Arc;
 
 use application::scheduler::Scheduler;
 use application::{
-    AuthService, InventoryService, JobRunner, PackageRefreshJob, PackageUpgradeJob,
+    AuthService, ClusterService, InventoryService, JobRunner, PackageRefreshJob, PackageUpgradeJob,
     SettingsService, UserService,
 };
 use axum::{routing::get, Json, Router};
 use domain::jobs::JobKind;
 use domain::ports::Mailer;
-use domain::ports::{HostInspector, HostUpdater};
+use domain::ports::{ClusterGateway, HostInspector, HostUpdater};
 use infrastructure::{
-    AesGcmCipher, Argon2Hasher, DbPool, DebianInspector, DebianUpdater, FakeHostInspector,
+    AesGcmCipher, Argon2Hasher, DbPool, DebianInspector, DebianUpdater, FakeClusterGateway,
-    FakeHostUpdater, JwtIssuer, LettreMailer, SqliteAuditLog, SqliteInventory, SqliteJobRuns,
+    FakeHostInspector, FakeHostUpdater, JwtIssuer, KubeGateway, LettreMailer, SqliteAuditLog,
-    SqliteRefreshTokens, SqliteSettings, SqliteUsers, SystemCommandRunner,
+    SqliteInventory, SqliteJobRuns, SqliteRefreshTokens, SqliteSettings, SqliteUsers,
+    SystemCommandRunner,
 };
 use tower_http::services::{ServeDir, ServeFile};
 use tower_http::trace::TraceLayer;
@@ -40,6 +42,7 @@ pub struct AppState {
     pub settings: Arc<SettingsService>,
     pub jobs: Arc<JobRunner>,
     pub inventory: Arc<InventoryService>,
+    pub cluster: Arc<ClusterService>,
     pub login_limiter: Arc<rate_limit::RateLimiter>,
 }
 
@@ -47,16 +50,32 @@ impl AppState {
     /// Wire the services on top of a connected database.
     pub fn new(cfg: Config, pool: DbPool) -> anyhow::Result<Self> {
         let runner = Arc::new(SystemCommandRunner);
-        let (inspector, updater): (Arc<dyn HostInspector>, Arc<dyn HostUpdater>) = if cfg.fake_host
+        let (inspector, updater, cluster): (
-        {
+            Arc<dyn HostInspector>,
-            (Arc::new(FakeHostInspector), Arc::new(FakeHostUpdater))
+            Arc<dyn HostUpdater>,
+            Arc<dyn ClusterGateway>,
+        ) = if cfg.fake_host {
+            (
+                Arc::new(FakeHostInspector),
+                Arc::new(FakeHostUpdater),
+                Arc::new(FakeClusterGateway),
+            )
         } else {
             (
                 Arc::new(DebianInspector::new(runner.clone())),
                 Arc::new(DebianUpdater::new(runner)),
+                Arc::new(KubeGateway::new(cfg.kubeconfig.clone())),
             )
         };
-        Self::with_adapters(cfg, pool, Arc::new(LettreMailer), inspector, updater, |r| r)
+        Self::with_adapters(
+            cfg,
+            pool,
+            Arc::new(LettreMailer),
+            inspector,
+            updater,
+            cluster,
+            |r| r,
+        )
     }
 
     /// Wiring with replaceable adapters (used by tests and the fake-host mode).
@@ -66,6 +85,7 @@ impl AppState {
         mailer: Arc<dyn Mailer>,
         inspector: Arc<dyn HostInspector>,
         updater: Arc<dyn HostUpdater>,
+        cluster: Arc<dyn ClusterGateway>,
         register_jobs: impl FnOnce(JobRunner) -> JobRunner,
     ) -> anyhow::Result<Self> {
         let users = Arc::new(SqliteUsers(pool.clone()));
@@ -100,6 +120,7 @@ impl AppState {
                 }),
             );
         let jobs = Arc::new(register_jobs(runner));
+        let cluster = Arc::new(ClusterService::new(cluster));
         Ok(Self {
             cfg,
             auth: Arc::new(auth),
@@ -107,6 +128,7 @@ impl AppState {
             settings,
             jobs,
             inventory,
+            cluster,
             login_limiter: Arc::new(rate_limit::RateLimiter::new(
                 10,
                 std::time::Duration::from_secs(60),
@@ -140,6 +162,7 @@ pub fn build_app(state: AppState) -> Router {
         .nest("/api/settings", settings::router())
         .nest("/api/jobs", jobs::router())
         .nest("/api/system", system::router())
+        .nest("/api/cluster", cluster::router())
         .fallback_service(spa)
         .layer(TraceLayer::new_for_http())
         .with_state(state)

backend/crates/api/src/openapi.rs

@@ -27,6 +27,7 @@ impl Modify for BearerAuth {
         crate::settings::get_smtp, crate::settings::put_smtp, crate::settings::test_smtp, crate::settings::list_schedules, crate::settings::put_schedule,
         crate::jobs::list, crate::jobs::kinds, crate::jobs::get_one, crate::jobs::run,
         crate::system::inventory, crate::system::upgrade,
+        crate::cluster::overview, crate::cluster::restart, crate::cluster::scale, crate::cluster::set_image,
     ),
     modifiers(&BearerAuth)
 )]

backend/crates/api/src/test_support.rs

@@ -16,6 +16,7 @@ pub fn test_config() -> Config {
         jwt_secret: "test-secret-test-secret-test-secret-1234".into(),
         master_key: "000102030405060708090a0b0c0d0e0f101112131415161718191a1b1c1d1e1f".into(),
         fake_host: true,
+        kubeconfig: None,
         bind: "127.0.0.1:0".parse().unwrap(),
         bootstrap_admin: None,
         cookie_secure: false,
@@ -85,6 +86,7 @@ async fn build_test_app_with(cfg: Config) -> Router {
         Arc::new(RecordingMailer),
         Arc::new(infrastructure::FakeHostInspector),
         Arc::new(infrastructure::FakeHostUpdater),
+        Arc::new(infrastructure::FakeClusterGateway),
         register_test_jobs,
     )
     .expect("state");

backend/crates/application/src/cluster_service.rs

@@ -1,6 +1,8 @@
 use std::sync::Arc;
 
-use domain::cluster::{ClusterOverview, WorkloadRef};
+use domain::cluster::{
+    validate_image, validate_k8s_name, ClusterOverview, WorkloadRef, MAX_REPLICAS,
+};
 use domain::ports::ClusterGateway;
 use domain::DomainError;
 
@@ -10,28 +12,55 @@ pub struct ClusterService {
 
 impl ClusterService {
     pub fn new(gateway: Arc<dyn ClusterGateway>) -> Self {
-        let _ = &gateway;
         Self { gateway }
     }
 
     pub async fn overview(&self) -> Result<ClusterOverview, DomainError> {
-        todo!()
+        self.gateway.overview().await
     }
 
-    pub async fn restart(&self, _w: WorkloadRef) -> Result<(), DomainError> {
+    pub async fn restart(&self, w: WorkloadRef) -> Result<(), DomainError> {
-        todo!()
+        validate_ref(&w)?;
+        self.gateway.restart(&w).await
     }
 
-    pub async fn scale(&self, _w: WorkloadRef, _replicas: i32) -> Result<(), DomainError> {
+    pub async fn scale(&self, w: WorkloadRef, replicas: i32) -> Result<(), DomainError> {
-        todo!()
+        validate_ref(&w)?;
+        if !(0..=MAX_REPLICAS).contains(&replicas) {
+            return Err(DomainError::Validation(format!(
+                "replicas must be between 0 and {MAX_REPLICAS}"
+            )));
+        }
+        self.gateway.scale(&w, replicas).await
     }
 
+    /// Change the image of `container` (default: the workload's first container).
     pub async fn set_image(
         &self,
-        _w: WorkloadRef,
+        w: WorkloadRef,
-        _container: Option<String>,
+        container: Option<String>,
-        _image: &str,
+        image: &str,
     ) -> Result<(), DomainError> {
-        todo!()
+        validate_ref(&w)?;
+        validate_image(image)?;
+        let container = match container {
+            Some(c) => c,
+            None => {
+                let overview = self.gateway.overview().await?;
+                overview
+                    .workloads
+                    .iter()
+                    .find(|x| x.namespace == w.namespace && x.kind == w.kind && x.name == w.name)
+                    .and_then(|x| x.containers.first())
+                    .map(|c| c.name.clone())
+                    .ok_or(DomainError::NotFound)?
+            }
+        };
+        self.gateway.set_image(&w, &container, image).await
     }
 }
+
+fn validate_ref(w: &WorkloadRef) -> Result<(), DomainError> {
+    validate_k8s_name(&w.namespace)?;
+    validate_k8s_name(&w.name)
+}

backend/crates/infrastructure/src/host/command.rs

@@ -43,11 +43,52 @@ impl CommandRunner for SystemCommandRunner {
 
     async fn run_streaming(
         &self,
-        _program: &str,
+        program: &str,
-        _args: &[&str],
+        args: &[&str],
-        _out: &dyn LineSink,
+        out: &dyn LineSink,
     ) -> Result<bool, DomainError> {
-        todo!()
+        use tokio::io::{AsyncBufReadExt, BufReader};
+        let mut child = tokio::process::Command::new(program)
+            .args(args)
+            .env("DEBIAN_FRONTEND", "noninteractive")
+            .env("LC_ALL", "C")
+            .stdin(std::process::Stdio::null())
+            .stdout(std::process::Stdio::piped())
+            .stderr(std::process::Stdio::piped())
+            .spawn()
+            .map_err(|e| DomainError::Unavailable(format!("{program}: {e}")))?;
+        let (tx, mut rx) = tokio::sync::mpsc::unbounded_channel::<String>();
+        let mut readers = Vec::new();
+        if let Some(stdout) = child.stdout.take() {
+            let tx = tx.clone();
+            readers.push(tokio::spawn(async move {
+                let mut lines = BufReader::new(stdout).lines();
+                while let Ok(Some(l)) = lines.next_line().await {
+                    let _ = tx.send(l);
+                }
+            }));
+        }
+        if let Some(stderr) = child.stderr.take() {
+            let tx = tx.clone();
+            readers.push(tokio::spawn(async move {
+                let mut lines = BufReader::new(stderr).lines();
+                while let Ok(Some(l)) = lines.next_line().await {
+                    let _ = tx.send(l);
+                }
+            }));
+        }
+        drop(tx);
+        while let Some(line) = rx.recv().await {
+            out.line(&line);
+        }
+        for r in readers {
+            let _ = r.await;
+        }
+        let status = child
+            .wait()
+            .await
+            .map_err(|e| DomainError::Unavailable(format!("{program}: {e}")))?;
+        Ok(status.success())
     }
 
     async fn read_file(&self, path: &str) -> Result<Option<String>, DomainError> {

backend/crates/infrastructure/src/k8s.rs

@@ -0,0 +1,383 @@
+//! Kubernetes gateway on top of kube-rs, plus a fake for development.
+use async_trait::async_trait;
+use chrono::Utc;
+use domain::cluster::{
+    ClusterOverview, Container, NodeInfo, VolumeClaim, Workload, WorkloadKind, WorkloadRef,
+};
+use domain::ports::ClusterGateway;
+use domain::DomainError;
+use k8s_openapi::api::apps::v1::{DaemonSet, Deployment, StatefulSet};
+use k8s_openapi::api::core::v1::{Namespace, Node, PersistentVolumeClaim, PodTemplateSpec};
+use kube::api::{Patch, PatchParams};
+use kube::config::{KubeConfigOptions, Kubeconfig};
+use kube::{Api, Client, Config};
+use serde_json::json;
+use tokio::sync::OnceCell;
+
+pub struct KubeGateway {
+    kubeconfig: Option<String>,
+    client: OnceCell<Client>,
+}
+
+impl KubeGateway {
+    /// `kubeconfig`: path to a kubeconfig file; `None` infers (in-cluster or ~/.kube/config).
+    pub fn new(kubeconfig: Option<String>) -> Self {
+        Self {
+            kubeconfig,
+            client: OnceCell::new(),
+        }
+    }
+
+    async fn client(&self) -> Result<Client, DomainError> {
+        let unavailable = |e: String| DomainError::Unavailable(format!("kubernetes: {e}"));
+        self.client
+            .get_or_try_init(|| async {
+                let config = match &self.kubeconfig {
+                    Some(path) => {
+                        let kc =
+                            Kubeconfig::read_from(path).map_err(|e| unavailable(e.to_string()))?;
+                        Config::from_custom_kubeconfig(kc, &KubeConfigOptions::default())
+                            .await
+                            .map_err(|e| unavailable(e.to_string()))?
+                    }
+                    None => Config::infer()
+                        .await
+                        .map_err(|e| unavailable(e.to_string()))?,
+                };
+                Client::try_from(config).map_err(|e| unavailable(e.to_string()))
+            })
+            .await
+            .cloned()
+    }
+}
+
+fn map_err(e: kube::Error) -> DomainError {
+    match e {
+        kube::Error::Api(ref r) if r.code == 404 => DomainError::NotFound,
+        kube::Error::Api(ref r) if r.code == 422 => DomainError::Validation(r.message.clone()),
+        e => DomainError::Unavailable(format!("kubernetes: {e}")),
+    }
+}
+
+fn containers(t: &Option<PodTemplateSpec>) -> Vec<Container> {
+    t.as_ref()
+        .and_then(|t| t.spec.as_ref())
+        .map(|s| {
+            s.containers
+                .iter()
+                .map(|c| Container {
+                    name: c.name.clone(),
+                    image: c.image.clone().unwrap_or_default(),
+                })
+                .collect()
+        })
+        .unwrap_or_default()
+}
+
+fn restart_patch() -> Patch<serde_json::Value> {
+    Patch::Merge(
+        json!({"spec": {"template": {"metadata": {"annotations": {"kubectl.kubernetes.io/restartedAt": Utc::now().to_rfc3339()}}}}}),
+    )
+}
+
+fn image_patch(container: &str, image: &str) -> Patch<serde_json::Value> {
+    Patch::Strategic(
+        json!({"spec": {"template": {"spec": {"containers": [{"name": container, "image": image}]}}}}),
+    )
+}
+
+macro_rules! patch_kind {
+    ($client:expr, $w:expr, $patch:expr) => {{
+        let pp = PatchParams::apply("softvisor-monitoring").force();
+        let pp = PatchParams {
+            field_manager: pp.field_manager,
+            ..PatchParams::default()
+        };
+        match $w.kind {
+            WorkloadKind::Deployment => Api::<Deployment>::namespaced($client, &$w.namespace)
+                .patch(&$w.name, &pp, &$patch)
+                .await
+                .map(|_| ()),
+            WorkloadKind::StatefulSet => Api::<StatefulSet>::namespaced($client, &$w.namespace)
+                .patch(&$w.name, &pp, &$patch)
+                .await
+                .map(|_| ()),
+            WorkloadKind::DaemonSet => Api::<DaemonSet>::namespaced($client, &$w.namespace)
+                .patch(&$w.name, &pp, &$patch)
+                .await
+                .map(|_| ()),
+        }
+        .map_err(map_err)
+    }};
+}
+
+#[async_trait]
+impl ClusterGateway for KubeGateway {
+    async fn overview(&self) -> Result<ClusterOverview, DomainError> {
+        let client = self.client().await?;
+        let lp = Default::default();
+        let nodes = Api::<Node>::all(client.clone())
+            .list(&lp)
+            .await
+            .map_err(map_err)?;
+        let namespaces = Api::<Namespace>::all(client.clone())
+            .list(&lp)
+            .await
+            .map_err(map_err)?;
+        let deployments = Api::<Deployment>::all(client.clone())
+            .list(&lp)
+            .await
+            .map_err(map_err)?;
+        let statefulsets = Api::<StatefulSet>::all(client.clone())
+            .list(&lp)
+            .await
+            .map_err(map_err)?;
+        let daemonsets = Api::<DaemonSet>::all(client.clone())
+            .list(&lp)
+            .await
+            .map_err(map_err)?;
+        let pvcs = Api::<PersistentVolumeClaim>::all(client)
+            .list(&lp)
+            .await
+            .map_err(map_err)?;
+
+        let mut workloads = Vec::new();
+        for d in deployments {
+            workloads.push(Workload {
+                namespace: d.metadata.namespace.clone().unwrap_or_default(),
+                kind: WorkloadKind::Deployment,
+                name: d.metadata.name.clone().unwrap_or_default(),
+                ready: d
+                    .status
+                    .as_ref()
+                    .and_then(|s| s.ready_replicas)
+                    .unwrap_or(0),
+                desired: d.spec.as_ref().and_then(|s| s.replicas).unwrap_or(0),
+                containers: containers(&d.spec.as_ref().map(|s| s.template.clone())),
+            });
+        }
+        for s in statefulsets {
+            workloads.push(Workload {
+                namespace: s.metadata.namespace.clone().unwrap_or_default(),
+                kind: WorkloadKind::StatefulSet,
+                name: s.metadata.name.clone().unwrap_or_default(),
+                ready: s
+                    .status
+                    .as_ref()
+                    .and_then(|s| s.ready_replicas)
+                    .unwrap_or(0),
+                desired: s.spec.as_ref().and_then(|s| s.replicas).unwrap_or(0),
+                containers: containers(&s.spec.as_ref().map(|s| s.template.clone())),
+            });
+        }
+        for d in daemonsets {
+            workloads.push(Workload {
+                namespace: d.metadata.namespace.clone().unwrap_or_default(),
+                kind: WorkloadKind::DaemonSet,
+                name: d.metadata.name.clone().unwrap_or_default(),
+                ready: d.status.as_ref().map(|s| s.number_ready).unwrap_or(0),
+                desired: d
+                    .status
+                    .as_ref()
+                    .map(|s| s.desired_number_scheduled)
+                    .unwrap_or(0),
+                containers: containers(&d.spec.as_ref().map(|s| s.template.clone())),
+            });
+        }
+        workloads.sort_by(|a, b| (&a.namespace, &a.name).cmp(&(&b.namespace, &b.name)));
+
+        Ok(ClusterOverview {
+            nodes: nodes
+                .iter()
+                .map(|n| {
+                    let info = n.status.as_ref().and_then(|s| s.node_info.as_ref());
+                    NodeInfo {
+                        name: n.metadata.name.clone().unwrap_or_default(),
+                        version: info.map(|i| i.kubelet_version.clone()).unwrap_or_default(),
+                        ready: n
+                            .status
+                            .as_ref()
+                            .and_then(|s| s.conditions.as_ref())
+                            .is_some_and(|c| {
+                                c.iter().any(|c| c.type_ == "Ready" && c.status == "True")
+                            }),
+                        os_image: info.map(|i| i.os_image.clone()).unwrap_or_default(),
+                        kernel: info.map(|i| i.kernel_version.clone()).unwrap_or_default(),
+                        container_runtime: info
+                            .map(|i| i.container_runtime_version.clone())
+                            .unwrap_or_default(),
+                    }
+                })
+                .collect(),
+            namespaces: namespaces
+                .iter()
+                .filter_map(|n| n.metadata.name.clone())
+                .collect(),
+            workloads,
+            volume_claims: pvcs
+                .iter()
+                .map(|p| VolumeClaim {
+                    namespace: p.metadata.namespace.clone().unwrap_or_default(),
+                    name: p.metadata.name.clone().unwrap_or_default(),
+                    capacity: p
+                        .status
+                        .as_ref()
+                        .and_then(|s| s.capacity.as_ref())
+                        .and_then(|c| c.get("storage"))
+                        .map(|q| q.0.clone())
+                        .unwrap_or_default(),
+                    storage_class: p
+                        .spec
+                        .as_ref()
+                        .and_then(|s| s.storage_class_name.clone())
+                        .unwrap_or_default(),
+                    status: p
+                        .status
+                        .as_ref()
+                        .and_then(|s| s.phase.clone())
+                        .unwrap_or_default(),
+                })
+                .collect(),
+            fetched_at: Utc::now(),
+        })
+    }
+
+    async fn restart(&self, w: &WorkloadRef) -> Result<(), DomainError> {
+        let client = self.client().await?;
+        patch_kind!(client.clone(), w, restart_patch())
+    }
+
+    async fn scale(&self, w: &WorkloadRef, replicas: i32) -> Result<(), DomainError> {
+        if w.kind == WorkloadKind::DaemonSet {
+            return Err(DomainError::Validation(
+                "daemonsets cannot be scaled".into(),
+            ));
+        }
+        let client = self.client().await?;
+        patch_kind!(
+            client.clone(),
+            w,
+            Patch::<serde_json::Value>::Merge(json!({"spec": {"replicas": replicas}}))
+        )
+    }
+
+    async fn set_image(
+        &self,
+        w: &WorkloadRef,
+        container: &str,
+        image: &str,
+    ) -> Result<(), DomainError> {
+        let client = self.client().await?;
+        patch_kind!(client.clone(), w, image_patch(container, image))
+    }
+}
+
+/// Sample cluster for development machines (FAKE_HOST=true).
+pub struct FakeClusterGateway;
+
+#[async_trait]
+impl ClusterGateway for FakeClusterGateway {
+    async fn overview(&self) -> Result<ClusterOverview, DomainError> {
+        let c = |name: &str, image: &str| {
+            vec![Container {
+                name: name.into(),
+                image: image.into(),
+            }]
+        };
+        let w = |ns: &str, kind, name: &str, ready, desired, containers| Workload {
+            namespace: ns.into(),
+            kind,
+            name: name.into(),
+            ready,
+            desired,
+            containers,
+        };
+        Ok(ClusterOverview {
+            nodes: vec![NodeInfo {
+                name: "fake-node".into(),
+                version: "v1.32.13".into(),
+                ready: true,
+                os_image: "Debian GNU/Linux 12 (bookworm)".into(),
+                kernel: "6.1.0-42-amd64".into(),
+                container_runtime: "containerd://1.6.36".into(),
+            }],
+            namespaces: vec![
+                "default".into(),
+                "gitea".into(),
+                "cert-manager".into(),
+                "kube-system".into(),
+            ],
+            workloads: vec![
+                w(
+                    "cert-manager",
+                    WorkloadKind::Deployment,
+                    "cert-manager",
+                    1,
+                    1,
+                    c(
+                        "cert-manager",
+                        "quay.io/jetstack/cert-manager-controller:v1.16.1",
+                    ),
+                ),
+                w(
+                    "gitea",
+                    WorkloadKind::Deployment,
+                    "gitea",
+                    1,
+                    1,
+                    c("gitea", "gitea/gitea:1.22.3"),
+                ),
+                w(
+                    "gitea",
+                    WorkloadKind::StatefulSet,
+                    "gitea-postgresql",
+                    1,
+                    1,
+                    c("postgresql", "bitnami/postgresql:16.4.0"),
+                ),
+                w(
+                    "gitea",
+                    WorkloadKind::StatefulSet,
+                    "gitea-valkey-primary",
+                    1,
+                    1,
+                    c("valkey", "bitnami/valkey:8.0.1"),
+                ),
+                w(
+                    "kube-system",
+                    WorkloadKind::DaemonSet,
+                    "calico-node",
+                    1,
+                    1,
+                    c("calico-node", "docker.io/calico/node:v3.28.1"),
+                ),
+            ],
+            volume_claims: vec![
+                VolumeClaim {
+                    namespace: "gitea".into(),
+                    name: "gitea-shared-storage".into(),
+                    capacity: "10Gi".into(),
+                    storage_class: "microk8s-hostpath".into(),
+                    status: "Bound".into(),
+                },
+                VolumeClaim {
+                    namespace: "gitea".into(),
+                    name: "data-gitea-postgresql-0".into(),
+                    capacity: "10Gi".into(),
+                    storage_class: "microk8s-hostpath".into(),
+                    status: "Bound".into(),
+                },
+            ],
+            fetched_at: Utc::now(),
+        })
+    }
+    async fn restart(&self, _: &WorkloadRef) -> Result<(), DomainError> {
+        Ok(())
+    }
+    async fn scale(&self, _: &WorkloadRef, _: i32) -> Result<(), DomainError> {
+        Ok(())
+    }
+    async fn set_image(&self, _: &WorkloadRef, _: &str, _: &str) -> Result<(), DomainError> {
+        Ok(())
+    }
+}

backend/crates/infrastructure/src/lib.rs

@@ -2,6 +2,7 @@
 pub mod cipher;
 pub mod db;
 pub mod host;
+pub mod k8s;
 pub mod mail;
 pub mod password;
 pub mod sqlite;
@@ -12,6 +13,7 @@ pub use db::{connect, DbPool};
 pub use host::{
     DebianInspector, DebianUpdater, FakeHostInspector, FakeHostUpdater, SystemCommandRunner,
 };
+pub use k8s::{FakeClusterGateway, KubeGateway};
 pub use mail::LettreMailer;
 pub use password::Argon2Hasher;
 pub use sqlite::SqliteInventory;

frontend/e2e/cluster.spec.ts

@@ -10,7 +10,7 @@ test('cluster page shows node, workloads and lets an admin restart a workload',
   await page.getByRole('link', { name: 'Kubernetes' }).click()
 
   await expect(page.getByTestId('node-version')).toContainText('v1.32')
-  const row = page.getByRole('row', { name: /gitea-postgresql/ })
+  const row = page.getByRole('row', { name: /^gitea gitea-postgresql statefulset/ })
   await expect(row).toContainText('statefulset')
   page.once('dialog', (d) => d.accept())
   await page

frontend/src/components/AppShell.vue

@@ -7,6 +7,7 @@ const router = useRouter()
 const nav = [
   { to: '/', label: 'Dashboard' },
   { to: '/updates', label: 'Updates' },
+  { to: '/cluster', label: 'Kubernetes' },
   { to: '/vulnerabilities', label: 'Vulnerabilities' },
   { to: '/backups', label: 'Backups' },
   { to: '/jobs', label: 'Jobs' },

frontend/src/components/WorkloadTable.vue

@@ -0,0 +1,61 @@
+<script setup lang="ts">
+import type { Workload } from '../api/types'
+
+defineProps<{ workloads: Workload[]; canAct: boolean }>()
+defineEmits<{ restart: [w: Workload]; scale: [w: Workload]; image: [w: Workload] }>()
+</script>
+
+<template>
+  <table class="w-full text-left text-sm">
+    <thead class="border-b border-gray-200 text-gray-500">
+      <tr>
+        <th class="py-2">Namespace</th>
+        <th>Name</th>
+        <th>Kind</th>
+        <th>Ready</th>
+        <th>Images</th>
+        <th></th>
+      </tr>
+    </thead>
+    <tbody>
+      <tr
+        v-for="w in workloads"
+        :key="`${w.namespace}/${w.kind}/${w.name}`"
+        class="border-b border-gray-100"
+        :class="w.ready < w.desired ? 'bg-red-50' : ''"
+      >
+        <td class="py-1.5">{{ w.namespace }}</td>
+        <td class="font-mono">{{ w.name }}</td>
+        <td>{{ w.kind }}</td>
+        <td :class="w.ready < w.desired ? 'font-medium text-red-700' : ''">
+          {{ w.ready }}/{{ w.desired }}
+        </td>
+        <td class="font-mono text-xs text-gray-600">
+          <div v-for="c in w.containers" :key="c.name">{{ c.image }}</div>
+        </td>
+        <td class="space-x-3 whitespace-nowrap text-right">
+          <template v-if="canAct">
+            <button
+              name="restart"
+              class="text-blue-600 hover:underline"
+              @click="$emit('restart', w)"
+            >
+              Restart
+            </button>
+            <button
+              v-if="w.kind !== 'daemonset'"
+              name="scale"
+              class="text-blue-600 hover:underline"
+              @click="$emit('scale', w)"
+            >
+              Scale
+            </button>
+            <button name="image" class="text-blue-600 hover:underline" @click="$emit('image', w)">
+              Set image
+            </button>
+          </template>
+        </td>
+      </tr>
+    </tbody>
+  </table>
+</template>

frontend/src/pages/ClusterPage.vue

@@ -0,0 +1,132 @@
+<script setup lang="ts">
+import { onMounted, ref } from 'vue'
+import { api, ApiError } from '../api/client'
+import type { ClusterOverview, Workload } from '../api/types'
+import { useAuthStore } from '../stores/auth'
+import { useToastStore } from '../stores/toast'
+import WorkloadTable from '../components/WorkloadTable.vue'
+
+const auth = useAuthStore()
+const toast = useToastStore()
+const overview = ref<ClusterOverview | null>(null)
+const error = ref('')
+
+async function load() {
+  error.value = ''
+  try {
+    overview.value = await api.get<ClusterOverview>('/api/cluster/overview')
+  } catch (e) {
+    error.value = e instanceof ApiError ? e.message : 'Request failed'
+  }
+}
+onMounted(load)
+
+const path = (w: Workload) => `/api/cluster/workloads/${w.namespace}/${w.kind}/${w.name}`
+
+async function act(label: string, fn: () => Promise<unknown>) {
+  try {
+    await fn()
+    toast.success(label)
+    await load()
+  } catch (e) {
+    toast.error(e instanceof ApiError ? e.message : 'Request failed')
+  }
+}
+
+function restart(w: Workload) {
+  if (!confirm(`Rolling restart of ${w.kind} ${w.namespace}/${w.name}?`)) return
+  act('Restart requested', () => api.post(path(w) + '/restart', {}))
+}
+
+function scale(w: Workload) {
+  const v = prompt(`Replicas for ${w.namespace}/${w.name}`, String(w.desired))
+  if (v === null) return
+  act(`Scaled to ${v}`, () => api.post(path(w) + '/scale', { replicas: Number(v) }))
+}
+
+function setImage(w: Workload) {
+  const current = w.containers[0]?.image ?? ''
+  const image = prompt(`New image for ${w.namespace}/${w.name} (${w.containers[0]?.name})`, current)
+  if (!image || image === current) return
+  act('Image updated, rollout started', () => api.post(path(w) + '/image', { image }))
+}
+</script>
+
+<template>
+  <div class="flex items-center justify-between">
+    <h1 class="text-2xl font-semibold">Kubernetes</h1>
+    <button class="rounded-md border border-gray-300 px-4 py-2 text-sm" @click="load">
+      Reload
+    </button>
+  </div>
+
+  <p v-if="error" role="alert" class="mt-6 rounded-md bg-red-50 p-4 text-sm text-red-800">
+    Cluster unreachable: {{ error }}
+  </p>
+
+  <template v-if="overview">
+    <div class="mt-6 grid gap-4 md:grid-cols-3">
+      <div
+        v-for="n in overview.nodes"
+        :key="n.name"
+        class="rounded-lg border border-gray-200 bg-white p-4"
+      >
+        <div class="text-xs uppercase text-gray-500">Node</div>
+        <div class="mt-1 text-lg font-medium">{{ n.name }}</div>
+        <div class="text-sm text-gray-600">
+          <span data-testid="node-version">{{ n.version }}</span> · {{ n.container_runtime }}
+        </div>
+        <div class="text-sm text-gray-600">{{ n.os_image }} · {{ n.kernel }}</div>
+        <span
+          class="mt-2 inline-block rounded-full px-2 py-0.5 text-xs font-medium"
+          :class="n.ready ? 'bg-green-100 text-green-800' : 'bg-red-100 text-red-800'"
+        >
+          {{ n.ready ? 'Ready' : 'NotReady' }}
+        </span>
+      </div>
+      <div class="rounded-lg border border-gray-200 bg-white p-4">
+        <div class="text-xs uppercase text-gray-500">Workloads</div>
+        <div class="mt-1 text-2xl font-semibold">{{ overview.workloads.length }}</div>
+        <div class="text-sm text-gray-600">
+          {{ overview.images.length }} distinct images · {{ overview.namespaces.length }} namespaces
+        </div>
+      </div>
+    </div>
+
+    <h2 class="mt-8 text-lg font-medium">Workloads</h2>
+    <WorkloadTable
+      class="mt-3"
+      :workloads="overview.workloads"
+      :can-act="auth.isAdmin"
+      @restart="restart"
+      @scale="scale"
+      @image="setImage"
+    />
+
+    <h2 class="mt-8 text-lg font-medium">Persistent volume claims</h2>
+    <table class="mt-3 w-full text-left text-sm">
+      <thead class="border-b border-gray-200 text-gray-500">
+        <tr>
+          <th class="py-2">Namespace</th>
+          <th>Name</th>
+          <th>Capacity</th>
+          <th>Storage class</th>
+          <th>Status</th>
+        </tr>
+      </thead>
+      <tbody>
+        <tr
+          v-for="p in overview.volume_claims"
+          :key="`${p.namespace}/${p.name}`"
+          class="border-b border-gray-100"
+        >
+          <td class="py-1.5">{{ p.namespace }}</td>
+          <td class="font-mono">{{ p.name }}</td>
+          <td>{{ p.capacity }}</td>
+          <td>{{ p.storage_class }}</td>
+          <td>{{ p.status }}</td>
+        </tr>
+      </tbody>
+    </table>
+  </template>
+</template>

frontend/src/router.ts

@@ -9,6 +9,7 @@ import PlaceholderPage from './pages/PlaceholderPage.vue'
 import SettingsPage from './pages/SettingsPage.vue'
 import JobsPage from './pages/JobsPage.vue'
 import UpdatesPage from './pages/UpdatesPage.vue'
+import ClusterPage from './pages/ClusterPage.vue'
 
 export const router = createRouter({
   history: createWebHistory(),
@@ -20,6 +21,7 @@ export const router = createRouter({
       children: [
         { path: '', name: 'dashboard', component: DashboardPage },
         { path: 'updates', component: UpdatesPage },
+        { path: 'cluster', component: ClusterPage },
         {
           path: 'vulnerabilities',
           component: PlaceholderPage,