wip: vllm example

Reviewed-on: #248

Cargo.toml

@@ -18,8 +18,7 @@ members = [
   "adr/agent_discovery/mdns",
    "brocade",
    "harmony_agent",
-  "harmony_agent/deploy", "harmony_node_readiness",
-  "examples/*",
+   "harmony_agent/deploy", "harmony_node_readiness", "harmony-k8s", "examples/vllm",
 ]
 
 [workspace.package]
@@ -38,6 +37,8 @@ tokio = { version = "1.40", features = [
   "macros",
   "rt-multi-thread",
 ] }
+tokio-retry = "0.3.0"
+tokio-util = "0.7.15"
 cidr = { features = ["serde"], version = "0.2" }
 russh = "0.45"
 russh-keys = "0.45"

brocade/src/fast_iron.rs

@@ -1,8 +1,7 @@
 use super::BrocadeClient;
 use crate::{
     BrocadeInfo, Error, ExecutionMode, InterSwitchLink, InterfaceInfo, MacAddressEntry,
-    PortChannelId, PortOperatingMode, SecurityLevel, parse_brocade_mac_address,
-    shell::BrocadeShell,
+    PortChannelId, PortOperatingMode, parse_brocade_mac_address, shell::BrocadeShell,
 };
 
 use async_trait::async_trait;

brocade/src/network_operating_system.rs

@@ -8,7 +8,7 @@ use regex::Regex;
 use crate::{
     BrocadeClient, BrocadeInfo, Error, ExecutionMode, InterSwitchLink, InterfaceInfo,
     InterfaceStatus, InterfaceType, MacAddressEntry, PortChannelId, PortOperatingMode,
-    SecurityLevel, parse_brocade_mac_address, shell::BrocadeShell,
+    parse_brocade_mac_address, shell::BrocadeShell,
 };
 
 #[derive(Debug)]

examples/cert_manager/src/main.rs

@@ -1,8 +1,8 @@
 use harmony::{
     inventory::Inventory,
     modules::cert_manager::{
-        capability::CertificateManagementConfig, score_cert_management::CertificateManagementScore,
-        score_certificate::CertificateScore, score_issuer::CertificateIssuerScore,
+        capability::CertificateManagementConfig, score_certificate::CertificateScore,
+        score_issuer::CertificateIssuerScore,
     },
     topology::K8sAnywhereTopology,
 };

examples/k8s_drain_node/Cargo.toml

@@ -10,9 +10,10 @@ publish = false
 harmony = { path = "../../harmony" }
 harmony_cli = { path = "../../harmony_cli" }
 harmony_types = { path = "../../harmony_types" }
+harmony_macros = { path = "../../harmony_macros" }
+harmony-k8s = { path = "../../harmony-k8s" }
 cidr.workspace = true
 tokio.workspace = true
-harmony_macros = { path = "../../harmony_macros" }
 log.workspace = true
 env_logger.workspace = true
 url.workspace = true

examples/k8s_drain_node/src/main.rs

@@ -1,6 +1,6 @@
 use std::time::Duration;
 
-use harmony::topology::k8s::{DrainOptions, K8sClient};
+use harmony_k8s::{DrainOptions, K8sClient};
 use log::{info, trace};
 
 #[tokio::main]

examples/k8s_write_file_on_node/Cargo.toml

@@ -10,9 +10,10 @@ publish = false
 harmony = { path = "../../harmony" }
 harmony_cli = { path = "../../harmony_cli" }
 harmony_types = { path = "../../harmony_types" }
+harmony_macros = { path = "../../harmony_macros" }
+harmony-k8s = { path = "../../harmony-k8s" }
 cidr.workspace = true
 tokio.workspace = true
-harmony_macros = { path = "../../harmony_macros" }
 log.workspace = true
 env_logger.workspace = true
 url.workspace = true

examples/k8s_write_file_on_node/src/main.rs

@@ -1,4 +1,4 @@
-use harmony::topology::k8s::{DrainOptions, K8sClient, NodeFile};
+use harmony_k8s::{K8sClient, NodeFile};
 use log::{info, trace};
 
 #[tokio::main]

examples/multisite_postgres/src/main.rs

@@ -14,7 +14,6 @@ async fn main() {
             ..Default::default() // Use harmony defaults, they are based on CNPG's default values :
                                  // "default" namespace, 1 instance, 1Gi storage
         },
-        hostname: "postgrestest.sto1.nationtech.io".to_string(),
     };
 
     harmony_cli::run(

examples/okd_installation/src/main.rs

@@ -6,7 +6,10 @@ use harmony::{
     data::{FileContent, FilePath},
     modules::{
         inventory::HarmonyDiscoveryStrategy,
-        okd::{installation::OKDInstallationPipeline, ipxe::OKDIpxeScore},
+        okd::{
+            installation::OKDInstallationPipeline, ipxe::OKDIpxeScore,
+            load_balancer::OKDLoadBalancerScore,
+        },
     },
     score::Score,
     topology::HAClusterTopology,
@@ -32,6 +35,7 @@ async fn main() {
     scores
         .append(&mut OKDInstallationPipeline::get_all_scores(HarmonyDiscoveryStrategy::MDNS).await);
 
+    scores.push(Box::new(OKDLoadBalancerScore::new(&topology)));
     harmony_cli::run(inventory, topology, scores, None)
         .await
         .unwrap();

examples/openbao/src/main.rs

@@ -5,7 +5,7 @@ use harmony::{
 #[tokio::main]
 async fn main() {
     let openbao = OpenbaoScore {
-        host: String::new(),
+        host: "openbao.sebastien.sto1.nationtech.io".to_string(),
     };
 
     harmony_cli::run(

examples/operatorhub_catalog/src/main.rs

@@ -1,5 +1,3 @@
-use std::str::FromStr;
-
 use harmony::{
     inventory::Inventory,
     modules::{k8s::apps::OperatorHubCatalogSourceScore, postgresql::CloudNativePgOperatorScore},
@@ -9,7 +7,7 @@ use harmony::{
 #[tokio::main]
 async fn main() {
     let operatorhub_catalog = OperatorHubCatalogSourceScore::default();
-    let cnpg_operator = CloudNativePgOperatorScore::default();
+    let cnpg_operator = CloudNativePgOperatorScore::default_openshift();
 
     harmony_cli::run(
         Inventory::autoload(),

examples/opnsense_node_exporter/src/main.rs

@@ -1,22 +1,13 @@
-use std::{
-    net::{IpAddr, Ipv4Addr},
-    sync::Arc,
-};
+use std::sync::Arc;
 
 use async_trait::async_trait;
-use cidr::Ipv4Cidr;
 use harmony::{
     executors::ExecutorError,
-    hardware::{HostCategory, Location, PhysicalHost, SwitchGroup},
-    infra::opnsense::OPNSenseManagementInterface,
     inventory::Inventory,
     modules::opnsense::node_exporter::NodeExporterScore,
-    topology::{
-        HAClusterTopology, LogicalHost, PreparationError, PreparationOutcome, Topology,
-        UnmanagedRouter, node_exporter::NodeExporter,
-    },
+    topology::{PreparationError, PreparationOutcome, Topology, node_exporter::NodeExporter},
 };
-use harmony_macros::{ip, ipv4, mac_address};
+use harmony_macros::ip;
 
 #[derive(Debug)]
 struct OpnSenseTopology {

examples/public_postgres/src/main.rs

@@ -1,8 +1,7 @@
 use harmony::{
     inventory::Inventory,
     modules::postgresql::{
-        K8sPostgreSQLScore, PostgreSQLConnectionScore, PublicPostgreSQLScore,
-        capability::PostgreSQLConfig,
+        PostgreSQLConnectionScore, PublicPostgreSQLScore, capability::PostgreSQLConfig,
     },
     topology::K8sAnywhereTopology,
 };
@@ -16,7 +15,6 @@ async fn main() {
             ..Default::default() // Use harmony defaults, they are based on CNPG's default values :
                                  // 1 instance, 1Gi storage
         },
-        hostname: "postgrestest.sto1.nationtech.io".to_string(),
     };
 
     let test_connection = PostgreSQLConnectionScore {

examples/rhob_application_monitoring/src/main.rs

@@ -1,4 +1,4 @@
-use std::{collections::HashMap, path::PathBuf, sync::Arc};
+use std::{path::PathBuf, sync::Arc};
 
 use harmony::{
     inventory::Inventory,

examples/rust/src/main.rs

@@ -1,4 +1,4 @@
-use std::{collections::HashMap, path::PathBuf, sync::Arc};
+use std::{path::PathBuf, sync::Arc};
 
 use harmony::{
     inventory::Inventory,

examples/tui/src/main.rs

@@ -44,6 +44,7 @@ fn build_large_score() -> LoadBalancerScore {
         ],
         listening_port: SocketAddr::V4(SocketAddrV4::new(ipv4!("192.168.0.0"), 49387)),
         health_check: Some(HealthCheck::HTTP(
+            Some(1993),
             "/some_long_ass_path_to_see_how_it_is_displayed_but_it_has_to_be_even_longer"
                 .to_string(),
             HttpMethod::GET,

examples/vllm/Cargo.toml

@@ -0,0 +1,15 @@
+[package]
+name = "vllm"
+edition = "2024"
+version.workspace = true
+readme.workspace = true
+license.workspace = true
+publish = false
+
+[dependencies]
+harmony = { path = "../../harmony" }
+harmony_cli = { path = "../../harmony_cli" }
+k8s-openapi = { workspace = true }
+tokio = { workspace = true }
+log = { workspace = true }
+env_logger = { workspace = true }

examples/vllm/src/main.rs

@@ -0,0 +1,523 @@
+//! vLLM Deployment Example for Qwen3.5-27B-FP8 on NVIDIA RTX 5090
+//!
+//! This example deploys vLLM serving Qwen3.5-27B with FP8 quantization,
+//! optimized for single RTX 5090 (32GB VRAM) with tool calling support.
+//!
+//! # Architecture & Memory Constraints
+//!
+//! **Model Details:**
+//! - Parameters: 27B (dense, not sparse/MoE)
+//! - Quantization: FP8 (8-bit weights)
+//! - Model size: ~27-28GB in memory
+//! - Native context: 262,144 tokens (will NOT fit in 32GB VRAM)
+//!
+//! **VRAM Budget for RTX 5090 (32GB):**
+//! - Model weights (FP8): ~27GB
+//! - Framework overhead: ~1-2GB
+//! - KV cache: ~2-3GB (for 16k context)
+//! - CUDA context: ~500MB
+//! - Temporary buffers: ~500MB
+//! - **Total: ~31-33GB** (tight fit, leaves minimal headroom)
+//!
+//! # OpenShift/OKD Requirements
+//!
+//! **SCC (Security Context Constraint) Setup:**
+//!
+//! The official vLLM container runs as root and writes to `/root/.cache/huggingface`.
+//! On OpenShift/OKD with the default restricted SCC, containers run as arbitrary UIDs
+//! and cannot write to `/root`. For testing, grant the `anyuid` SCC:
+//!
+//! ```bash
+//! # As cluster admin, grant anyuid SCC to the namespace's service account:
+//! oc adm policy add-scc-to-user anyuid -z default -n vllm-qwen
+//! ```
+//!
+//! This allows pods in the `vllm-qwen` namespace to run as root (UID 0).
+//! For production, consider building a custom vLLM image that runs as non-root.
+//!
+//! # Critical Configuration Notes
+//!
+//! 1. **GPU_MEMORY_UTILIZATION=1.0**: Maximum GPU memory allocation.
+//!    NEVER decrease this for dense models - CPU offloading destroys performance
+//!    (100-1000x slower) for models where every parameter is used during inference.
+//!
+//! 2. **MAX_MODEL_LEN=16384**: Conservative context length that fits in available VRAM.
+//!    Agentic workflows with long tool call histories will need careful context management.
+//!
+//! 3. **--language-model-only**: Skips loading the vision encoder, saving ~1-2GB VRAM.
+//!    Essential for fitting the model in 32GB VRAM.
+//!
+//! 4. **PVC Size**: 50Gi for HuggingFace cache. Qwen3.5-27B-FP8 is ~30GB.
+//!
+//! # Performance Expectations
+//!
+//! - Single token latency: ~50-100ms (no CPU offloading)
+//! - With CPU offloading: ~5-50 seconds per token (unusable for real-time inference)
+//! - Throughput: ~10-20 tokens/second (single stream, no batching)
+//!
+//! # Next Steps for Production
+//!
+//! To increase context length:
+//! 1. Monitor GPU memory: `kubectl exec -it deployment/qwen3-5-27b -- nvidia-smi dmon -s u`
+//! 2. If stable, increase MAX_MODEL_LEN (try 32768, then 65536)
+//! 3. If OOM: revert to lower value
+//!
+//! For full 262k context, consider:
+//! - Multi-GPU setup with tensor parallelism (--tensor-parallel-size 8)
+//! - Or use a smaller model (Qwen3.5-7B-FP8)
+
+use std::collections::BTreeMap;
+
+use harmony::{
+    inventory::Inventory,
+    modules::{
+        k8s::resource::K8sResourceScore,
+        okd::{
+            crd::route::{RoutePort, RouteSpec, RouteTargetReference, TLSConfig},
+            route::OKDRouteScore,
+        },
+    },
+    score::Score,
+    topology::{K8sAnywhereTopology, TlsRouter},
+};
+use k8s_openapi::{
+    api::{
+        apps::v1::{Deployment, DeploymentSpec, DeploymentStrategy},
+        core::v1::{
+            Container, ContainerPort, EmptyDirVolumeSource, EnvVar, EnvVarSource,
+            HTTPGetAction, PersistentVolumeClaim, PersistentVolumeClaimSpec,
+            PersistentVolumeClaimVolumeSource, PodSpec, PodTemplateSpec, Probe,
+            ResourceRequirements, Secret, SecretKeySelector, SecretVolumeSource, Service,
+            ServicePort, ServiceSpec, Volume, VolumeMount, VolumeResourceRequirements,
+        },
+    },
+    apimachinery::pkg::{
+        api::resource::Quantity,
+        apis::meta::v1::{LabelSelector, ObjectMeta},
+        util::intstr::IntOrString,
+    },
+    ByteString,
+};
+use log::info;
+
+const NAMESPACE: &str = "vllm-qwen";
+const MODEL_NAME: &str = "Qwen/Qwen3.5-27B-FP8";
+const DEPLOYMENT_NAME: &str = "qwen3-5-27b";
+const SERVICE_NAME: &str = DEPLOYMENT_NAME;
+const ROUTE_NAME: &str = DEPLOYMENT_NAME;
+const PVC_NAME: &str = "huggingface-cache";
+const SECRET_NAME: &str = "hf-token-secret";
+
+const VLLM_IMAGE: &str = "vllm/vllm-openai:latest";
+const SERVICE_PORT: u16 = 8000;
+const TARGET_PORT: u16 = 8000;
+
+/// Maximum context length for the model (in tokens).
+///
+/// **Impact on VRAM:**
+/// - Qwen3.5-27B uses per-token KV cache storage for the context window
+/// - Larger context = more KV cache memory required
+/// - Approximate KV cache per token: ~32KB for FP8 (very rough estimate)
+/// - 16k tokens ≈ 0.5-1GB KV cache
+/// - 262k tokens ≈ 8-16GB KV cache (native context length - will NOT fit in 32GB VRAM)
+///
+/// **Performance Impact:**
+/// - Context length directly impacts memory for storing conversation history
+/// - Agentic workflows with long tool call histories benefit from more context
+/// - If context > available VRAM, vLLM will OOM and fail to start
+///
+/// **Recommendations for RTX 5090 (32GB):**
+/// - Start with 16384 (conservative, should work)
+/// - If no OOM, try 32768 (better for agentic workflows)
+/// - Monitor GPU memory with `nvidia-smi` during operation
+const MAX_MODEL_LEN: i64 = 16384;
+
+/// Fraction of GPU memory to allocate for the model (0.0 to 1.0).
+///
+/// **CRITICAL WARNING: This is a dense model!**
+/// Qwen3.5-27B-FP8 is NOT a sparse/mixture-of-experts model. All 27B parameters
+/// are active during inference. CPU offloading will DESTROY performance.
+///
+/// **What this parameter controls:**
+/// - Controls how much of GPU memory vLLM pre-allocates for:
+///   1. Model weights (~27GB for FP8 quantization)
+///   2. KV cache for context window
+///   3. Activation buffers for inference
+///   4. Runtime overhead
+///
+/// **VRAM Allocation Example:**
+/// - GPU: 32GB RTX 5090
+/// - GPU_MEMORY_UTILIZATION: 0.95
+/// - vLLM will try to use: 32GB * 0.95 = 30.4GB
+/// - Model weights: ~27-28GB
+/// - Remaining for KV cache + runtime: ~2-3GB
+///
+/// **If set too LOW (e.g., 0.7):**
+/// - vLLM restricts itself to 32GB * 0.7 = 22.4GB
+/// - Model weights alone need ~27GB
+/// - vLLM will OFFLOAD model weights to CPU memory
+/// - Performance: **100-1000x slower** (single token generation can take seconds instead of milliseconds)
+/// - This is catastrophic for a dense model where every layer needs all parameters
+///
+/// **If set too HIGH (e.g., 0.99):**
+/// - vLLM tries to allocate nearly all GPU memory
+/// - Risk: CUDA OOM if any other process needs GPU memory
+/// - Risk: KV cache allocation fails during inference
+/// - System instability
+///
+/// **Current Setting: 0.95**
+/// - Leaves 5% buffer (1.6GB) for CUDA overhead, system processes
+/// - Maximum allocation for model + KV cache: ~30.4GB
+/// - Should leave enough headroom for:
+///   - CUDA context: ~500MB
+///   - Temporary buffers: ~500MB
+///   - Safety margin: ~600MB
+///
+/// **How to tune:**
+/// 1. Start with 0.95 (current setting)
+/// 2. Monitor with `nvidia-smi dmon -s u` during operation
+/// 3. If OOM during inference: reduce MAX_MODEL_LEN first
+/// 4. If stable: try increasing MAX_MODEL_LEN before increasing this
+/// 5. Only increase this if you're certain no other GPU processes run
+///
+/// **NEVER decrease this for dense models!** 
+/// If model doesn't fit, use a smaller model or quantization, not CPU offloading.
+const GPU_MEMORY_UTILIZATION : f32 = 1.0;
+
+#[tokio::main]
+async fn main() -> Result<(), Box<dyn std::error::Error>> {
+    env_logger::init();
+
+    info!("Deploying vLLM with Qwen3.5-27B-FP8 model");
+    info!("Configuration:");
+    info!("  Model: {}", MODEL_NAME);
+    info!("  Max context length: {} tokens", MAX_MODEL_LEN);
+    info!("  GPU memory utilization: {}", GPU_MEMORY_UTILIZATION);
+    info!("  Language model only: true");
+    info!("  Tool calling enabled: true");
+
+    let topology = K8sAnywhereTopology::from_env();
+    let domain = topology
+        .get_internal_domain()
+        .await
+        .ok()
+        .flatten()
+        .unwrap_or_else(|| "cluster.local".to_string());
+
+    let host = format!("{}-{}.apps.{}", SERVICE_NAME, NAMESPACE, domain);
+    info!("Creating route with host: {}", host);
+
+    let scores: Vec<Box<dyn Score<K8sAnywhereTopology>>> = vec![
+        create_namespace(),
+        create_pvc(),
+        create_secret(),
+        create_deployment(),
+        create_service(),
+        create_route(&host),
+    ];
+
+    harmony_cli::run(Inventory::autoload(), topology, scores, None)
+        .await
+        .map_err(|e| format!("Failed to deploy: {}", e))?;
+
+    info!("Successfully deployed vLLM with Qwen3.5-27B-FP8");
+    info!("Access the API at: http://{}.apps.<cluster-domain>", SERVICE_NAME);
+
+    Ok(())
+}
+
+fn create_namespace() -> Box<dyn Score<K8sAnywhereTopology>> {
+    use k8s_openapi::api::core::v1::Namespace;
+
+    let namespace = Namespace {
+        metadata: ObjectMeta {
+            name: Some(NAMESPACE.to_string()),
+            ..Default::default()
+        },
+        spec: None,
+        status: None,
+    };
+
+    Box::new(K8sResourceScore::single(namespace, None))
+}
+
+fn create_pvc() -> Box<dyn Score<K8sAnywhereTopology>> {
+    let pvc = PersistentVolumeClaim {
+        metadata: ObjectMeta {
+            name: Some(PVC_NAME.to_string()),
+            namespace: Some(NAMESPACE.to_string()),
+            ..Default::default()
+        },
+        spec: Some(PersistentVolumeClaimSpec {
+            access_modes: Some(vec!["ReadWriteOnce".to_string()]),
+            resources: Some(VolumeResourceRequirements {
+                requests: Some(BTreeMap::from([(
+                    "storage".to_string(),
+                    Quantity("50Gi".to_string()),
+                )])),
+                limits: None,
+            }),
+            ..Default::default()
+        }),
+        status: None,
+    };
+
+    Box::new(K8sResourceScore::single(
+        pvc,
+        Some(NAMESPACE.to_string()),
+    ))
+}
+
+fn create_secret() -> Box<dyn Score<K8sAnywhereTopology>> {
+    let mut data = BTreeMap::new();
+    data.insert(
+        "token".to_string(),
+        ByteString("".to_string().into_bytes()),
+    );
+
+    let secret = Secret {
+        metadata: ObjectMeta {
+            name: Some(SECRET_NAME.to_string()),
+            namespace: Some(NAMESPACE.to_string()),
+            ..Default::default()
+        },
+        data: Some(data),
+        immutable: Some(false),
+        type_: Some("Opaque".to_string()),
+        string_data: None,
+    };
+
+    Box::new(K8sResourceScore::single(
+        secret,
+        Some(NAMESPACE.to_string()),
+    ))
+}
+
+fn create_deployment() -> Box<dyn Score<K8sAnywhereTopology>> {
+    let deployment = Deployment {
+        metadata: ObjectMeta {
+            name: Some(DEPLOYMENT_NAME.to_string()),
+            namespace: Some(NAMESPACE.to_string()),
+            labels: Some(BTreeMap::from([(
+                "app".to_string(),
+                DEPLOYMENT_NAME.to_string(),
+            )])),
+            ..Default::default()
+        },
+        spec: Some(DeploymentSpec {
+            replicas: Some(1),
+            selector: LabelSelector {
+                match_labels: Some(BTreeMap::from([(
+                    "app".to_string(),
+                    DEPLOYMENT_NAME.to_string(),
+                )])),
+                ..Default::default()
+            },
+            strategy: Some(DeploymentStrategy {
+                type_: Some("Recreate".to_string()),
+                ..Default::default()
+            }),
+            template: PodTemplateSpec {
+                metadata: Some(ObjectMeta {
+                    labels: Some(BTreeMap::from([(
+                        "app".to_string(),
+                        DEPLOYMENT_NAME.to_string(),
+                    )])),
+                    ..Default::default()
+                }),
+                spec: Some(PodSpec {
+                    node_selector: Some(BTreeMap::from([(
+                        "nvidia.com/gpu.product".to_string(),
+                        "NVIDIA-GeForce-RTX-5090".to_string(),
+                    )])),
+                    volumes: Some(vec![
+                        Volume {
+                            name: "cache-volume".to_string(),
+                            persistent_volume_claim: Some(PersistentVolumeClaimVolumeSource {
+                                claim_name: PVC_NAME.to_string(),
+                                read_only: Some(false),
+                            }),
+                            ..Default::default()
+                        },
+                        Volume {
+                            name: "shm".to_string(),
+                            empty_dir: Some(EmptyDirVolumeSource {
+                                medium: Some("Memory".to_string()),
+                                size_limit: Some(Quantity("4Gi".to_string())),
+                            }),
+                            ..Default::default()
+                        },
+                        Volume {
+                            name: "hf-token".to_string(),
+                            secret: Some(SecretVolumeSource {
+                                secret_name: Some(SECRET_NAME.to_string()),
+                                optional: Some(true),
+                                ..Default::default()
+                            }),
+                            ..Default::default()
+                        },
+                    ]),
+                    containers: vec![Container {
+                        name: DEPLOYMENT_NAME.to_string(),
+                        image: Some(VLLM_IMAGE.to_string()),
+                        command: Some(vec!["/bin/sh".to_string(), "-c".to_string()]),
+                        args: Some(vec![build_vllm_command()]),
+                        env: Some(vec![
+                            EnvVar {
+                                name: "HF_TOKEN".to_string(),
+                                value_from: Some(EnvVarSource {
+                                    secret_key_ref: Some(SecretKeySelector {
+                                        key: "token".to_string(),
+                                        name: SECRET_NAME.to_string(),
+                                        optional: Some(true),
+                                    }),
+                                    ..Default::default()
+                                }),
+                                value: None,
+                            },
+                            EnvVar {
+                                name: "VLLM_WORKER_MULTIPROC_METHOD".to_string(),
+                                value: Some("spawn".to_string()),
+                                value_from: None,
+                            },
+                        ]),
+                        ports: Some(vec![ContainerPort {
+                            container_port: SERVICE_PORT as i32,
+                            protocol: Some("TCP".to_string()),
+                            ..Default::default()
+                        }]),
+                        resources: Some(ResourceRequirements {
+                            limits: Some(BTreeMap::from([
+                                ("cpu".to_string(), Quantity("10".to_string())),
+                                ("memory".to_string(), Quantity("30Gi".to_string())),
+                                ("nvidia.com/gpu".to_string(), Quantity("1".to_string())),
+                            ])),
+                            requests: Some(BTreeMap::from([
+                                ("cpu".to_string(), Quantity("2".to_string())),
+                                ("memory".to_string(), Quantity("10Gi".to_string())),
+                                ("nvidia.com/gpu".to_string(), Quantity("1".to_string())),
+                            ])),
+                            claims: None,
+                        }),
+                        volume_mounts: Some(vec![
+                            VolumeMount {
+                                name: "cache-volume".to_string(),
+                                mount_path: "/root/.cache/huggingface".to_string(),
+                                read_only: Some(false),
+                                ..Default::default()
+                            },
+                            VolumeMount {
+                                name: "shm".to_string(),
+                                mount_path: "/dev/shm".to_string(),
+                                ..Default::default()
+                            },
+                            VolumeMount {
+                                name: "hf-token".to_string(),
+                                mount_path: "/etc/secrets/hf-token".to_string(),
+                                read_only: Some(true),
+                                ..Default::default()
+                            },
+                        ]),
+                        liveness_probe: Some(Probe {
+                            http_get: Some(HTTPGetAction {
+                                path: Some("/health".to_string()),
+                                port: IntOrString::Int(SERVICE_PORT as i32),
+                                ..Default::default()
+                            }),
+                            initial_delay_seconds: Some(300),
+                            period_seconds: Some(30),
+                            ..Default::default()
+                        }),
+                        readiness_probe: Some(Probe {
+                            http_get: Some(HTTPGetAction {
+                                path: Some("/health".to_string()),
+                                port: IntOrString::Int(SERVICE_PORT as i32),
+                                ..Default::default()
+                            }),
+                            initial_delay_seconds: Some(120),
+                            period_seconds: Some(10),
+                            ..Default::default()
+                        }),
+                        ..Default::default()
+                    }],
+                    ..Default::default()
+                }),
+            },
+            ..Default::default()
+        }),
+        status: None,
+    };
+
+    Box::new(K8sResourceScore::single(
+        deployment,
+        Some(NAMESPACE.to_string()),
+    ))
+}
+
+fn build_vllm_command() -> String {
+    format!(
+        "vllm serve {} \
+         --port {} \
+         --max-model-len {} \
+         --gpu-memory-utilization {} \
+         --reasoning-parser qwen3 \
+         --enable-auto-tool-choice \
+         --tool-call-parser qwen3_coder \
+         --language-model-only",
+        MODEL_NAME, SERVICE_PORT, MAX_MODEL_LEN, GPU_MEMORY_UTILIZATION
+    )
+}
+
+fn create_service() -> Box<dyn Score<K8sAnywhereTopology>> {
+    let service = Service {
+        metadata: ObjectMeta {
+            name: Some(SERVICE_NAME.to_string()),
+            namespace: Some(NAMESPACE.to_string()),
+            ..Default::default()
+        },
+        spec: Some(ServiceSpec {
+            ports: Some(vec![ServicePort {
+                name: Some("http".to_string()),
+                port: SERVICE_PORT as i32,
+                protocol: Some("TCP".to_string()),
+                target_port: Some(IntOrString::Int(TARGET_PORT as i32)),
+                ..Default::default()
+            }]),
+            selector: Some(BTreeMap::from([(
+                "app".to_string(),
+                DEPLOYMENT_NAME.to_string(),
+            )])),
+            type_: Some("ClusterIP".to_string()),
+            ..Default::default()
+        }),
+        status: None,
+    };
+
+    Box::new(K8sResourceScore::single(
+        service,
+        Some(NAMESPACE.to_string()),
+    ))
+}
+
+fn create_route(host: &str) -> Box<dyn Score<K8sAnywhereTopology>> {
+    let route_spec = RouteSpec {
+        to: RouteTargetReference {
+            kind: "Service".to_string(),
+            name: SERVICE_NAME.to_string(),
+            weight: Some(100),
+        },
+        host: Some(host.to_string()),
+        port: Some(RoutePort {
+            target_port: SERVICE_PORT as u16,
+        }),
+        tls: Some(TLSConfig {
+            termination: "edge".to_string(),
+            insecure_edge_termination_policy: Some("Redirect".to_string()),
+            ..Default::default()
+        }),
+        wildcard_policy: None,
+        ..Default::default()
+    };
+
+    Box::new(OKDRouteScore::new(ROUTE_NAME, NAMESPACE, route_spec))
+}

examples/zitadel/Cargo.toml

@@ -0,0 +1,14 @@
+[package]
+name = "example-zitadel"
+edition = "2024"
+version.workspace = true
+readme.workspace = true
+license.workspace = true
+
+[dependencies]
+harmony = { path = "../../harmony" }
+harmony_cli = { path = "../../harmony_cli" }
+harmony_macros = { path = "../../harmony_macros" }
+harmony_types = { path = "../../harmony_types" }
+tokio.workspace = true
+url.workspace = true

examples/zitadel/src/main.rs

@@ -0,0 +1,20 @@
+use harmony::{
+    inventory::Inventory, modules::zitadel::ZitadelScore, topology::K8sAnywhereTopology,
+};
+
+#[tokio::main]
+async fn main() {
+    let zitadel = ZitadelScore {
+        host: "sso.sto1.nationtech.io".to_string(),
+        zitadel_version: "v4.12.1".to_string(),
+    };
+
+    harmony_cli::run(
+        Inventory::autoload(),
+        K8sAnywhereTopology::from_env(),
+        vec![Box::new(zitadel)],
+        None,
+    )
+    .await
+    .unwrap();
+}

harmony-k8s/Cargo.toml

@@ -0,0 +1,23 @@
+[package]
+name = "harmony-k8s"
+edition = "2024"
+version.workspace = true
+readme.workspace = true
+license.workspace = true
+
+[dependencies]
+kube.workspace = true
+k8s-openapi.workspace = true
+tokio.workspace = true
+tokio-retry.workspace = true
+serde.workspace = true
+serde_json.workspace = true
+serde_yaml.workspace = true
+log.workspace = true
+similar.workspace = true
+reqwest.workspace = true
+url.workspace = true
+inquire.workspace = true
+
+[dev-dependencies]
+pretty_assertions.workspace = true

harmony-k8s/src/apply.rs

@@ -0,0 +1,593 @@
+use kube::{
+    Client, Error, Resource,
+    api::{
+        Api, ApiResource, DynamicObject, GroupVersionKind, Patch, PatchParams, PostParams,
+        ResourceExt,
+    },
+    core::ErrorResponse,
+    discovery::Scope,
+    error::DiscoveryError,
+};
+use log::{debug, error, trace, warn};
+use serde::{Serialize, de::DeserializeOwned};
+use serde_json::Value;
+use similar::TextDiff;
+use url::Url;
+
+use crate::client::K8sClient;
+use crate::helper;
+use crate::types::WriteMode;
+
+/// The field-manager token sent with every server-side apply request.
+pub const FIELD_MANAGER: &str = "harmony-k8s";
+
+// ── Private helpers ──────────────────────────────────────────────────────────
+
+/// Serialise any `Serialize` payload to a [`DynamicObject`] via JSON.
+fn to_dynamic<T: Serialize>(payload: &T) -> Result<DynamicObject, Error> {
+    serde_json::from_value(serde_json::to_value(payload).map_err(Error::SerdeError)?)
+        .map_err(Error::SerdeError)
+}
+
+/// Fetch the current resource, display a unified diff against `payload`, and
+/// return `()`.  All output goes to stdout (same behaviour as before).
+///
+/// A 404 is treated as "resource would be created" — not an error.
+async fn show_dry_run<T: Serialize>(
+    api: &Api<DynamicObject>,
+    name: &str,
+    payload: &T,
+) -> Result<(), Error> {
+    let new_yaml = serde_yaml::to_string(payload)
+        .unwrap_or_else(|_| "Failed to serialize new resource".to_string());
+
+    match api.get(name).await {
+        Ok(current) => {
+            println!("\nDry-run for resource: '{name}'");
+            let mut current_val = serde_yaml::to_value(&current).unwrap_or(serde_yaml::Value::Null);
+            if let Some(map) = current_val.as_mapping_mut() {
+                map.remove(&serde_yaml::Value::String("status".to_string()));
+            }
+            let current_yaml = serde_yaml::to_string(&current_val)
+                .unwrap_or_else(|_| "Failed to serialize current resource".to_string());
+
+            if current_yaml == new_yaml {
+                println!("No changes detected.");
+            } else {
+                println!("Changes detected:");
+                let diff = TextDiff::from_lines(&current_yaml, &new_yaml);
+                for change in diff.iter_all_changes() {
+                    let sign = match change.tag() {
+                        similar::ChangeTag::Delete => "-",
+                        similar::ChangeTag::Insert => "+",
+                        similar::ChangeTag::Equal => " ",
+                    };
+                    print!("{sign}{change}");
+                }
+            }
+            Ok(())
+        }
+        Err(Error::Api(ErrorResponse { code: 404, .. })) => {
+            println!("\nDry-run for new resource: '{name}'");
+            println!("Resource does not exist. Would be created:");
+            for line in new_yaml.lines() {
+                println!("+{line}");
+            }
+            Ok(())
+        }
+        Err(e) => {
+            error!("Failed to fetch resource '{name}' for dry-run: {e}");
+            Err(e)
+        }
+    }
+}
+
+/// Execute the real (non-dry-run) apply, respecting [`WriteMode`].
+async fn do_apply<T: Serialize + std::fmt::Debug>(
+    api: &Api<DynamicObject>,
+    name: &str,
+    payload: &T,
+    patch_params: &PatchParams,
+    write_mode: &WriteMode,
+) -> Result<DynamicObject, Error> {
+    match write_mode {
+        WriteMode::CreateOrUpdate => {
+            // TODO refactor this arm to perform self.update and if fail with 404 self.create
+            // This will avoid the repetition of the api.patch and api.create calls within this
+            // function body. This makes the code more maintainable
+            match api.patch(name, patch_params, &Patch::Apply(payload)).await {
+                Ok(obj) => Ok(obj),
+                Err(Error::Api(ErrorResponse { code: 404, .. })) => {
+                    debug!("Resource '{name}' not found via SSA, falling back to POST");
+                    let dyn_obj = to_dynamic(payload)?;
+                    api.create(&PostParams::default(), &dyn_obj)
+                        .await
+                        .map_err(|e| {
+                            error!("Failed to create '{name}': {e}");
+                            e
+                        })
+                }
+                Err(e) => {
+                    error!("Failed to apply '{name}': {e}");
+                    Err(e)
+                }
+            }
+        }
+        WriteMode::Create => {
+            let dyn_obj = to_dynamic(payload)?;
+            api.create(&PostParams::default(), &dyn_obj)
+                .await
+                .map_err(|e| {
+                    error!("Failed to create '{name}': {e}");
+                    e
+                })
+        }
+        WriteMode::Update => match api.patch(name, patch_params, &Patch::Apply(payload)).await {
+            Ok(obj) => Ok(obj),
+            Err(Error::Api(ErrorResponse { code: 404, .. })) => Err(Error::Api(ErrorResponse {
+                code: 404,
+                message: format!("Resource '{name}' not found and WriteMode is UpdateOnly"),
+                reason: "NotFound".to_string(),
+                status: "Failure".to_string(),
+            })),
+            Err(e) => {
+                error!("Failed to update '{name}': {e}");
+                Err(e)
+            }
+        },
+    }
+}
+
+// ── Public API ───────────────────────────────────────────────────────────────
+
+impl K8sClient {
+    /// Server-side apply: create if absent, update if present.
+    /// Equivalent to `kubectl apply`.
+    pub async fn apply<K>(&self, resource: &K, namespace: Option<&str>) -> Result<K, Error>
+    where
+        K: Resource + Clone + std::fmt::Debug + DeserializeOwned + Serialize,
+        <K as Resource>::DynamicType: Default,
+    {
+        self.apply_with_strategy(resource, namespace, WriteMode::CreateOrUpdate)
+            .await
+    }
+
+    /// POST only — returns an error if the resource already exists.
+    pub async fn create<K>(&self, resource: &K, namespace: Option<&str>) -> Result<K, Error>
+    where
+        K: Resource + Clone + std::fmt::Debug + DeserializeOwned + Serialize,
+        <K as Resource>::DynamicType: Default,
+    {
+        self.apply_with_strategy(resource, namespace, WriteMode::Create)
+            .await
+    }
+
+    /// Server-side apply only — returns an error if the resource does not exist.
+    pub async fn update<K>(&self, resource: &K, namespace: Option<&str>) -> Result<K, Error>
+    where
+        K: Resource + Clone + std::fmt::Debug + DeserializeOwned + Serialize,
+        <K as Resource>::DynamicType: Default,
+    {
+        self.apply_with_strategy(resource, namespace, WriteMode::Update)
+            .await
+    }
+
+    pub async fn apply_with_strategy<K>(
+        &self,
+        resource: &K,
+        namespace: Option<&str>,
+        write_mode: WriteMode,
+    ) -> Result<K, Error>
+    where
+        K: Resource + Clone + std::fmt::Debug + DeserializeOwned + Serialize,
+        <K as Resource>::DynamicType: Default,
+    {
+        debug!(
+            "apply_with_strategy: {:?} ns={:?}",
+            resource.meta().name,
+            namespace
+        );
+        trace!("{:#}", serde_json::to_value(resource).unwrap_or_default());
+
+        let dyntype = K::DynamicType::default();
+        let gvk = GroupVersionKind {
+            group: K::group(&dyntype).to_string(),
+            version: K::version(&dyntype).to_string(),
+            kind: K::kind(&dyntype).to_string(),
+        };
+
+        let discovery = self.discovery().await?;
+        let (ar, caps) = discovery.resolve_gvk(&gvk).ok_or_else(|| {
+            Error::Discovery(DiscoveryError::MissingResource(format!(
+                "Cannot resolve GVK: {gvk:?}"
+            )))
+        })?;
+
+        let effective_ns = if caps.scope == Scope::Cluster {
+            None
+        } else {
+            namespace.or_else(|| resource.meta().namespace.as_deref())
+        };
+
+        let api: Api<DynamicObject> =
+            get_dynamic_api(ar, caps, self.client.clone(), effective_ns, false);
+
+        let name = resource
+            .meta()
+            .name
+            .as_deref()
+            .expect("Kubernetes resource must have a name");
+
+        if self.dry_run {
+            show_dry_run(&api, name, resource).await?;
+            return Ok(resource.clone());
+        }
+
+        let patch_params = PatchParams::apply(FIELD_MANAGER);
+        do_apply(&api, name, resource, &patch_params, &write_mode)
+            .await
+            .and_then(helper::dyn_to_typed)
+    }
+
+    /// Applies resources in order, one at a time
+    pub async fn apply_many<K>(&self, resources: &[K], ns: Option<&str>) -> Result<Vec<K>, Error>
+    where
+        K: Resource + Clone + std::fmt::Debug + DeserializeOwned + Serialize,
+        <K as Resource>::DynamicType: Default,
+    {
+        let mut result = Vec::new();
+        for r in resources.iter() {
+            let res = self.apply(r, ns).await;
+            if res.is_err() {
+                // NOTE: this may log sensitive data; downgrade to debug if needed.
+                warn!(
+                    "Failed to apply k8s resource: {}",
+                    serde_json::to_string_pretty(r).map_err(Error::SerdeError)?
+                );
+            }
+            result.push(res?);
+        }
+        Ok(result)
+    }
+
+    /// Apply a [`DynamicObject`] resource using server-side apply.
+    pub async fn apply_dynamic(
+        &self,
+        resource: &DynamicObject,
+        namespace: Option<&str>,
+        force_conflicts: bool,
+    ) -> Result<DynamicObject, Error> {
+        trace!("apply_dynamic {resource:#?} ns={namespace:?} force={force_conflicts}");
+
+        let discovery = self.discovery().await?;
+        let type_meta = resource.types.as_ref().ok_or_else(|| {
+            Error::BuildRequest(kube::core::request::Error::Validation(
+                "DynamicObject must have types (apiVersion and kind)".to_string(),
+            ))
+        })?;
+
+        let gvk = GroupVersionKind::try_from(type_meta).map_err(|_| {
+            Error::BuildRequest(kube::core::request::Error::Validation(format!(
+                "Invalid GVK in DynamicObject: {type_meta:?}"
+            )))
+        })?;
+
+        let (ar, caps) = discovery.resolve_gvk(&gvk).ok_or_else(|| {
+            Error::Discovery(DiscoveryError::MissingResource(format!(
+                "Cannot resolve GVK: {gvk:?}"
+            )))
+        })?;
+
+        let effective_ns = if caps.scope == Scope::Cluster {
+            None
+        } else {
+            namespace.or_else(|| resource.metadata.namespace.as_deref())
+        };
+
+        let api = get_dynamic_api(ar, caps, self.client.clone(), effective_ns, false);
+        let name = resource.metadata.name.as_deref().ok_or_else(|| {
+            Error::BuildRequest(kube::core::request::Error::Validation(
+                "DynamicObject must have metadata.name".to_string(),
+            ))
+        })?;
+
+        debug!(
+            "apply_dynamic kind={:?} name='{name}' ns={effective_ns:?}",
+            resource.types.as_ref().map(|t| &t.kind),
+        );
+
+        // NOTE would be nice to improve cohesion between the dynamic and typed apis and avoid copy
+        // pasting the dry_run and some more logic
+        if self.dry_run {
+            show_dry_run(&api, name, resource).await?;
+            return Ok(resource.clone());
+        }
+
+        let mut patch_params = PatchParams::apply(FIELD_MANAGER);
+        patch_params.force = force_conflicts;
+
+        do_apply(
+            &api,
+            name,
+            resource,
+            &patch_params,
+            &WriteMode::CreateOrUpdate,
+        )
+        .await
+    }
+
+    pub async fn apply_dynamic_many(
+        &self,
+        resources: &[DynamicObject],
+        namespace: Option<&str>,
+        force_conflicts: bool,
+    ) -> Result<Vec<DynamicObject>, Error> {
+        let mut result = Vec::new();
+        for r in resources.iter() {
+            result.push(self.apply_dynamic(r, namespace, force_conflicts).await?);
+        }
+        Ok(result)
+    }
+
+    pub async fn apply_yaml_many(
+        &self,
+        #[allow(clippy::ptr_arg)] yaml: &Vec<serde_yaml::Value>,
+        ns: Option<&str>,
+    ) -> Result<(), Error> {
+        for y in yaml.iter() {
+            self.apply_yaml(y, ns).await?;
+        }
+        Ok(())
+    }
+
+    pub async fn apply_yaml(
+        &self,
+        yaml: &serde_yaml::Value,
+        ns: Option<&str>,
+    ) -> Result<(), Error> {
+        // NOTE wouldn't it be possible to parse this into a DynamicObject and simply call
+        // apply_dynamic instead of reimplementing api interactions?
+        let obj: DynamicObject =
+            serde_yaml::from_value(yaml.clone()).expect("YAML must deserialise to DynamicObject");
+        let name = obj.metadata.name.as_ref().expect("YAML must have a name");
+
+        let api_version = yaml["apiVersion"].as_str().expect("missing apiVersion");
+        let kind = yaml["kind"].as_str().expect("missing kind");
+
+        let mut it = api_version.splitn(2, '/');
+        let first = it.next().unwrap();
+        let (g, v) = match it.next() {
+            Some(second) => (first, second),
+            None => ("", first),
+        };
+
+        let api_resource = ApiResource::from_gvk(&GroupVersionKind::gvk(g, v, kind));
+        let namespace = ns.unwrap_or_else(|| {
+            obj.metadata
+                .namespace
+                .as_deref()
+                .expect("YAML must have a namespace when ns is not provided")
+        });
+
+        let api: Api<DynamicObject> =
+            Api::namespaced_with(self.client.clone(), namespace, &api_resource);
+
+        println!("Applying '{name}' in namespace '{namespace}'...");
+        let patch_params = PatchParams::apply(FIELD_MANAGER);
+        let result = api.patch(name, &patch_params, &Patch::Apply(&obj)).await?;
+        println!("Successfully applied '{}'.", result.name_any());
+        Ok(())
+    }
+
+    /// Equivalent to `kubectl apply -f <url>`.
+    pub async fn apply_url(&self, url: Url, ns: Option<&str>) -> Result<(), Error> {
+        let patch_params = PatchParams::apply(FIELD_MANAGER);
+        let discovery = self.discovery().await?;
+
+        let yaml = reqwest::get(url)
+            .await
+            .expect("Could not fetch URL")
+            .text()
+            .await
+            .expect("Could not read response body");
+
+        for doc in multidoc_deserialize(&yaml).expect("Failed to parse YAML from URL") {
+            let obj: DynamicObject =
+                serde_yaml::from_value(doc).expect("YAML document is not a valid object");
+            let namespace = obj.metadata.namespace.as_deref().or(ns);
+            let type_meta = obj.types.as_ref().expect("Object is missing TypeMeta");
+            let gvk =
+                GroupVersionKind::try_from(type_meta).expect("Object has invalid GroupVersionKind");
+            let name = obj.name_any();
+
+            if let Some((ar, caps)) = discovery.resolve_gvk(&gvk) {
+                let api = get_dynamic_api(ar, caps, self.client.clone(), namespace, false);
+                trace!(
+                    "Applying {}:\n{}",
+                    gvk.kind,
+                    serde_yaml::to_string(&obj).unwrap_or_default()
+                );
+                let data: Value = serde_json::to_value(&obj).expect("serialisation failed");
+                let _r = api.patch(&name, &patch_params, &Patch::Apply(data)).await?;
+                debug!("Applied {} '{name}'", gvk.kind);
+            } else {
+                warn!("Skipping document with unknown GVK: {gvk:?}");
+            }
+        }
+        Ok(())
+    }
+
+    /// Build a dynamic API client from a [`DynamicObject`]'s type metadata.
+    pub(crate) fn get_api_for_dynamic_object(
+        &self,
+        object: &DynamicObject,
+        ns: Option<&str>,
+    ) -> Result<Api<DynamicObject>, Error> {
+        let ar = object
+            .types
+            .as_ref()
+            .and_then(|t| {
+                let parts: Vec<&str> = t.api_version.split('/').collect();
+                match parts.as_slice() {
+                    [version] => Some(ApiResource::from_gvk(&GroupVersionKind::gvk(
+                        "", version, &t.kind,
+                    ))),
+                    [group, version] => Some(ApiResource::from_gvk(&GroupVersionKind::gvk(
+                        group, version, &t.kind,
+                    ))),
+                    _ => None,
+                }
+            })
+            .ok_or_else(|| {
+                Error::BuildRequest(kube::core::request::Error::Validation(format!(
+                    "Invalid apiVersion in DynamicObject: {object:#?}"
+                )))
+            })?;
+
+        Ok(match ns {
+            Some(ns) => Api::namespaced_with(self.client.clone(), ns, &ar),
+            None => Api::default_namespaced_with(self.client.clone(), &ar),
+        })
+    }
+}
+
+// ── Free functions ───────────────────────────────────────────────────────────
+
+pub(crate) fn get_dynamic_api(
+    resource: kube::api::ApiResource,
+    capabilities: kube::discovery::ApiCapabilities,
+    client: Client,
+    ns: Option<&str>,
+    all: bool,
+) -> Api<DynamicObject> {
+    if capabilities.scope == Scope::Cluster || all {
+        Api::all_with(client, &resource)
+    } else if let Some(namespace) = ns {
+        Api::namespaced_with(client, namespace, &resource)
+    } else {
+        Api::default_namespaced_with(client, &resource)
+    }
+}
+
+pub(crate) fn multidoc_deserialize(
+    data: &str,
+) -> Result<Vec<serde_yaml::Value>, serde_yaml::Error> {
+    use serde::Deserialize;
+    let mut docs = vec![];
+    for de in serde_yaml::Deserializer::from_str(data) {
+        docs.push(serde_yaml::Value::deserialize(de)?);
+    }
+    Ok(docs)
+}
+
+// ── Tests ────────────────────────────────────────────────────────────────────
+
+#[cfg(test)]
+mod apply_tests {
+    use std::collections::BTreeMap;
+    use std::time::{SystemTime, UNIX_EPOCH};
+
+    use k8s_openapi::api::core::v1::ConfigMap;
+    use k8s_openapi::apimachinery::pkg::apis::meta::v1::ObjectMeta;
+    use kube::api::{DeleteParams, TypeMeta};
+
+    use super::*;
+
+    #[tokio::test]
+    #[ignore = "requires kubernetes cluster"]
+    async fn apply_creates_new_configmap() {
+        let client = K8sClient::try_default().await.unwrap();
+        let ns = "default";
+        let name = format!(
+            "test-cm-{}",
+            SystemTime::now()
+                .duration_since(UNIX_EPOCH)
+                .unwrap()
+                .as_millis()
+        );
+
+        let cm = ConfigMap {
+            metadata: ObjectMeta {
+                name: Some(name.clone()),
+                namespace: Some(ns.to_string()),
+                ..Default::default()
+            },
+            data: Some(BTreeMap::from([("key1".to_string(), "value1".to_string())])),
+            ..Default::default()
+        };
+
+        assert!(client.apply(&cm, Some(ns)).await.is_ok());
+
+        let api: Api<ConfigMap> = Api::namespaced(client.client.clone(), ns);
+        let _ = api.delete(&name, &DeleteParams::default()).await;
+    }
+
+    #[tokio::test]
+    #[ignore = "requires kubernetes cluster"]
+    async fn apply_is_idempotent() {
+        let client = K8sClient::try_default().await.unwrap();
+        let ns = "default";
+        let name = format!(
+            "test-idem-{}",
+            SystemTime::now()
+                .duration_since(UNIX_EPOCH)
+                .unwrap()
+                .as_millis()
+        );
+
+        let cm = ConfigMap {
+            metadata: ObjectMeta {
+                name: Some(name.clone()),
+                namespace: Some(ns.to_string()),
+                ..Default::default()
+            },
+            data: Some(BTreeMap::from([("key".to_string(), "value".to_string())])),
+            ..Default::default()
+        };
+
+        assert!(
+            client.apply(&cm, Some(ns)).await.is_ok(),
+            "first apply failed"
+        );
+        assert!(
+            client.apply(&cm, Some(ns)).await.is_ok(),
+            "second apply failed (not idempotent)"
+        );
+
+        let api: Api<ConfigMap> = Api::namespaced(client.client.clone(), ns);
+        let _ = api.delete(&name, &DeleteParams::default()).await;
+    }
+
+    #[tokio::test]
+    #[ignore = "requires kubernetes cluster"]
+    async fn apply_dynamic_creates_new_resource() {
+        let client = K8sClient::try_default().await.unwrap();
+        let ns = "default";
+        let name = format!(
+            "test-dyn-{}",
+            SystemTime::now()
+                .duration_since(UNIX_EPOCH)
+                .unwrap()
+                .as_millis()
+        );
+
+        let obj = DynamicObject {
+            types: Some(TypeMeta {
+                api_version: "v1".to_string(),
+                kind: "ConfigMap".to_string(),
+            }),
+            metadata: ObjectMeta {
+                name: Some(name.clone()),
+                namespace: Some(ns.to_string()),
+                ..Default::default()
+            },
+            data: serde_json::json!({}),
+        };
+
+        let result = client.apply_dynamic(&obj, Some(ns), false).await;
+        assert!(result.is_ok(), "apply_dynamic failed: {:?}", result.err());
+
+        let api: Api<ConfigMap> = Api::namespaced(client.client.clone(), ns);
+        let _ = api.delete(&name, &DeleteParams::default()).await;
+    }
+}

harmony-k8s/src/bundle.rs

@@ -25,9 +25,9 @@
 //!
 //! ## Example
 //!
-//! ```rust,no_run
-//! use harmony::topology::k8s::{K8sClient, helper};
-//! use harmony::topology::KubernetesDistribution;
+//! ```
+//! use harmony_k8s::{K8sClient, helper};
+//! use harmony_k8s::KubernetesDistribution;
 //!
 //! async fn write_network_config(client: &K8sClient, node: &str) {
 //!     // Create a bundle with platform-specific RBAC
@@ -56,7 +56,7 @@ use kube::{Error, Resource, ResourceExt, api::DynamicObject};
 use serde::Serialize;
 use serde_json;
 
-use crate::domain::topology::k8s::K8sClient;
+use crate::K8sClient;
 
 /// A ResourceBundle represents a logical unit of work consisting of multiple
 /// Kubernetes resources that should be applied or deleted together.

harmony-k8s/src/client.rs

@@ -0,0 +1,99 @@
+use std::sync::Arc;
+
+use kube::config::{KubeConfigOptions, Kubeconfig};
+use kube::{Client, Config, Discovery, Error};
+use log::error;
+use serde::Serialize;
+use tokio::sync::OnceCell;
+
+use crate::types::KubernetesDistribution;
+
+// TODO not cool, should use a proper configuration mechanism
+// cli arg, env var, config file
+fn read_dry_run_from_env() -> bool {
+    std::env::var("DRY_RUN")
+        .map(|v| v == "true" || v == "1")
+        .unwrap_or(false)
+}
+
+#[derive(Clone)]
+pub struct K8sClient {
+    pub(crate) client: Client,
+    /// When `true` no mutation is sent to the API server; diffs are printed
+    /// to stdout instead. Initialised from the `DRY_RUN` environment variable.
+    pub(crate) dry_run: bool,
+    pub(crate) k8s_distribution: Arc<OnceCell<KubernetesDistribution>>,
+    pub(crate) discovery: Arc<OnceCell<Discovery>>,
+}
+
+impl Serialize for K8sClient {
+    fn serialize<S>(&self, _serializer: S) -> Result<S::Ok, S::Error>
+    where
+        S: serde::Serializer,
+    {
+        todo!("K8sClient serialization is not meaningful; remove this impl if unused")
+    }
+}
+
+impl std::fmt::Debug for K8sClient {
+    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
+        f.write_fmt(format_args!(
+            "K8sClient {{ namespace: {}, dry_run: {} }}",
+            self.client.default_namespace(),
+            self.dry_run,
+        ))
+    }
+}
+
+impl K8sClient {
+    /// Create a client, reading `DRY_RUN` from the environment.
+    pub fn new(client: Client) -> Self {
+        Self {
+            dry_run: read_dry_run_from_env(),
+            client,
+            k8s_distribution: Arc::new(OnceCell::new()),
+            discovery: Arc::new(OnceCell::new()),
+        }
+    }
+
+    /// Create a client that always operates in dry-run mode, regardless of
+    /// the environment variable.
+    pub fn new_dry_run(client: Client) -> Self {
+        Self {
+            dry_run: true,
+            ..Self::new(client)
+        }
+    }
+
+    /// Returns `true` if this client is operating in dry-run mode.
+    pub fn is_dry_run(&self) -> bool {
+        self.dry_run
+    }
+
+    pub async fn try_default() -> Result<Self, Error> {
+        Ok(Self::new(Client::try_default().await?))
+    }
+
+    pub async fn from_kubeconfig(path: &str) -> Option<Self> {
+        Self::from_kubeconfig_with_opts(path, &KubeConfigOptions::default()).await
+    }
+
+    pub async fn from_kubeconfig_with_context(path: &str, context: Option<String>) -> Option<Self> {
+        let mut opts = KubeConfigOptions::default();
+        opts.context = context;
+        Self::from_kubeconfig_with_opts(path, &opts).await
+    }
+
+    pub async fn from_kubeconfig_with_opts(path: &str, opts: &KubeConfigOptions) -> Option<Self> {
+        let k = match Kubeconfig::read_from(path) {
+            Ok(k) => k,
+            Err(e) => {
+                error!("Failed to load kubeconfig from {path}: {e}");
+                return None;
+            }
+        };
+        Some(Self::new(
+            Client::try_from(Config::from_custom_kubeconfig(k, opts).await.unwrap()).unwrap(),
+        ))
+    }
+}

harmony-k8s/src/discovery.rs

@@ -0,0 +1,83 @@
+use std::time::Duration;
+
+use kube::{Discovery, Error};
+use log::{debug, error, info, trace, warn};
+use tokio::sync::Mutex;
+use tokio_retry::{Retry, strategy::ExponentialBackoff};
+
+use crate::client::K8sClient;
+use crate::types::KubernetesDistribution;
+
+impl K8sClient {
+    pub async fn get_apiserver_version(
+        &self,
+    ) -> Result<k8s_openapi::apimachinery::pkg::version::Info, Error> {
+        self.client.clone().apiserver_version().await
+    }
+
+    /// Runs (and caches) Kubernetes API discovery with exponential-backoff retries.
+    pub async fn discovery(&self) -> Result<&Discovery, Error> {
+        let retry_strategy = ExponentialBackoff::from_millis(1000)
+            .max_delay(Duration::from_secs(32))
+            .take(6);
+
+        let attempt = Mutex::new(0u32);
+        Retry::spawn(retry_strategy, || async {
+            let mut n = attempt.lock().await;
+            *n += 1;
+            match self
+                .discovery
+                .get_or_try_init(async || {
+                    debug!("Running Kubernetes API discovery (attempt {})", *n);
+                    let d = Discovery::new(self.client.clone()).run().await?;
+                    debug!("Kubernetes API discovery completed");
+                    Ok(d)
+                })
+                .await
+            {
+                Ok(d) => Ok(d),
+                Err(e) => {
+                    warn!("Kubernetes API discovery failed (attempt {}): {}", *n, e);
+                    Err(e)
+                }
+            }
+        })
+        .await
+        .map_err(|e| {
+            error!("Kubernetes API discovery failed after all retries: {}", e);
+            e
+        })
+    }
+
+    /// Detect which Kubernetes distribution is running. Result is cached for
+    /// the lifetime of the client.
+    pub async fn get_k8s_distribution(&self) -> Result<KubernetesDistribution, Error> {
+        self.k8s_distribution
+            .get_or_try_init(async || {
+                debug!("Detecting Kubernetes distribution");
+                let api_groups = self.client.list_api_groups().await?;
+                trace!("list_api_groups: {:?}", api_groups);
+
+                let version = self.get_apiserver_version().await?;
+
+                if api_groups
+                    .groups
+                    .iter()
+                    .any(|g| g.name == "project.openshift.io")
+                {
+                    info!("Detected distribution: OpenshiftFamily");
+                    return Ok(KubernetesDistribution::OpenshiftFamily);
+                }
+
+                if version.git_version.contains("k3s") {
+                    info!("Detected distribution: K3sFamily");
+                    return Ok(KubernetesDistribution::K3sFamily);
+                }
+
+                info!("Distribution not identified, using Default");
+                Ok(KubernetesDistribution::Default)
+            })
+            .await
+            .cloned()
+    }
+}

harmony-k8s/src/helper.rs

@@ -1,7 +1,7 @@
 use std::collections::BTreeMap;
 use std::time::Duration;
 
-use crate::topology::KubernetesDistribution;
+use crate::KubernetesDistribution;
 
 use super::bundle::ResourceBundle;
 use super::config::PRIVILEGED_POD_IMAGE;
@@ -133,9 +133,9 @@ pub fn host_root_volume() -> (Volume, VolumeMount) {
 ///
 /// # Example
 ///
-/// ```rust,no_run
-/// # use harmony::topology::k8s::helper::{build_privileged_bundle, PrivilegedPodConfig};
-/// # use harmony::topology::KubernetesDistribution;
+/// ```
+/// use harmony_k8s::helper::{build_privileged_bundle, PrivilegedPodConfig};
+/// use harmony_k8s::KubernetesDistribution;
 /// let bundle = build_privileged_bundle(
 ///     PrivilegedPodConfig {
 ///         name: "network-setup".to_string(),

harmony-k8s/src/lib.rs

@@ -0,0 +1,13 @@
+pub mod apply;
+pub mod bundle;
+pub mod client;
+pub mod config;
+pub mod discovery;
+pub mod helper;
+pub mod node;
+pub mod pod;
+pub mod resources;
+pub mod types;
+
+pub use client::K8sClient;
+pub use types::{DrainOptions, KubernetesDistribution, NodeFile, ScopeResolver, WriteMode};

harmony-k8s/src/main.rs

@@ -0,0 +1,3 @@
+fn main() {
+    println!("Hello, world!");
+}

harmony-k8s/src/node.rs

@@ -0,0 +1,722 @@
+use std::collections::BTreeMap;
+use std::time::{Duration, SystemTime, UNIX_EPOCH};
+
+use k8s_openapi::api::core::v1::{
+    ConfigMap, ConfigMapVolumeSource, Node, Pod, Volume, VolumeMount,
+};
+use k8s_openapi::apimachinery::pkg::apis::meta::v1::ObjectMeta;
+use kube::{
+    Error,
+    api::{Api, DeleteParams, EvictParams, ListParams, PostParams},
+    core::ErrorResponse,
+    error::DiscoveryError,
+};
+use log::{debug, error, info, warn};
+use tokio::time::sleep;
+
+use crate::client::K8sClient;
+use crate::helper::{self, PrivilegedPodConfig};
+use crate::types::{DrainOptions, NodeFile};
+
+impl K8sClient {
+    pub async fn cordon_node(&self, node_name: &str) -> Result<(), Error> {
+        Api::<Node>::all(self.client.clone())
+            .cordon(node_name)
+            .await?;
+        Ok(())
+    }
+
+    pub async fn uncordon_node(&self, node_name: &str) -> Result<(), Error> {
+        Api::<Node>::all(self.client.clone())
+            .uncordon(node_name)
+            .await?;
+        Ok(())
+    }
+
+    pub async fn wait_for_node_ready(&self, node_name: &str) -> Result<(), Error> {
+        self.wait_for_node_ready_with_timeout(node_name, Duration::from_secs(600))
+            .await
+    }
+
+    async fn wait_for_node_ready_with_timeout(
+        &self,
+        node_name: &str,
+        timeout: Duration,
+    ) -> Result<(), Error> {
+        let api: Api<Node> = Api::all(self.client.clone());
+        let start = tokio::time::Instant::now();
+        let poll = Duration::from_secs(5);
+        loop {
+            if start.elapsed() > timeout {
+                return Err(Error::Discovery(DiscoveryError::MissingResource(format!(
+                    "Node '{node_name}' did not become Ready within {timeout:?}"
+                ))));
+            }
+            match api.get(node_name).await {
+                Ok(node) => {
+                    if node
+                        .status
+                        .as_ref()
+                        .and_then(|s| s.conditions.as_ref())
+                        .map(|conds| {
+                            conds
+                                .iter()
+                                .any(|c| c.type_ == "Ready" && c.status == "True")
+                        })
+                        .unwrap_or(false)
+                    {
+                        debug!("Node '{node_name}' is Ready");
+                        return Ok(());
+                    }
+                }
+                Err(e) => debug!("Error polling node '{node_name}': {e}"),
+            }
+            sleep(poll).await;
+        }
+    }
+
+    async fn wait_for_node_not_ready(
+        &self,
+        node_name: &str,
+        timeout: Duration,
+    ) -> Result<(), Error> {
+        let api: Api<Node> = Api::all(self.client.clone());
+        let start = tokio::time::Instant::now();
+        let poll = Duration::from_secs(5);
+        loop {
+            if start.elapsed() > timeout {
+                return Err(Error::Discovery(DiscoveryError::MissingResource(format!(
+                    "Node '{node_name}' did not become NotReady within {timeout:?}"
+                ))));
+            }
+            match api.get(node_name).await {
+                Ok(node) => {
+                    let is_ready = node
+                        .status
+                        .as_ref()
+                        .and_then(|s| s.conditions.as_ref())
+                        .map(|conds| {
+                            conds
+                                .iter()
+                                .any(|c| c.type_ == "Ready" && c.status == "True")
+                        })
+                        .unwrap_or(false);
+                    if !is_ready {
+                        debug!("Node '{node_name}' is NotReady");
+                        return Ok(());
+                    }
+                }
+                Err(e) => debug!("Error polling node '{node_name}': {e}"),
+            }
+            sleep(poll).await;
+        }
+    }
+
+    async fn list_pods_on_node(&self, node_name: &str) -> Result<Vec<Pod>, Error> {
+        let api: Api<Pod> = Api::all(self.client.clone());
+        Ok(api
+            .list(&ListParams::default().fields(&format!("spec.nodeName={node_name}")))
+            .await?
+            .items)
+    }
+
+    fn is_mirror_pod(pod: &Pod) -> bool {
+        pod.metadata
+            .annotations
+            .as_ref()
+            .map(|a| a.contains_key("kubernetes.io/config.mirror"))
+            .unwrap_or(false)
+    }
+
+    fn is_daemonset_pod(pod: &Pod) -> bool {
+        pod.metadata
+            .owner_references
+            .as_ref()
+            .map(|refs| refs.iter().any(|r| r.kind == "DaemonSet"))
+            .unwrap_or(false)
+    }
+
+    fn has_emptydir_volume(pod: &Pod) -> bool {
+        pod.spec
+            .as_ref()
+            .and_then(|s| s.volumes.as_ref())
+            .map(|vols| vols.iter().any(|v| v.empty_dir.is_some()))
+            .unwrap_or(false)
+    }
+
+    fn is_completed_pod(pod: &Pod) -> bool {
+        pod.status
+            .as_ref()
+            .and_then(|s| s.phase.as_deref())
+            .map(|phase| phase == "Succeeded" || phase == "Failed")
+            .unwrap_or(false)
+    }
+
+    fn classify_pods_for_drain(
+        pods: &[Pod],
+        options: &DrainOptions,
+    ) -> Result<(Vec<Pod>, Vec<String>), String> {
+        let mut evictable = Vec::new();
+        let mut skipped = Vec::new();
+        let mut blocking = Vec::new();
+
+        for pod in pods {
+            let name = pod.metadata.name.as_deref().unwrap_or("<unknown>");
+            let ns = pod.metadata.namespace.as_deref().unwrap_or("<unknown>");
+            let qualified = format!("{ns}/{name}");
+
+            if Self::is_mirror_pod(pod) {
+                skipped.push(format!("{qualified} (mirror pod)"));
+                continue;
+            }
+            if Self::is_completed_pod(pod) {
+                skipped.push(format!("{qualified} (completed)"));
+                continue;
+            }
+            if Self::is_daemonset_pod(pod) {
+                if options.ignore_daemonsets {
+                    skipped.push(format!("{qualified} (DaemonSet-managed)"));
+                } else {
+                    blocking.push(format!(
+                        "{qualified} is managed by a DaemonSet (set ignore_daemonsets to skip)"
+                    ));
+                }
+                continue;
+            }
+            if Self::has_emptydir_volume(pod) && !options.delete_emptydir_data {
+                blocking.push(format!(
+                    "{qualified} uses emptyDir volumes (set delete_emptydir_data to allow eviction)"
+                ));
+                continue;
+            }
+            evictable.push(pod.clone());
+        }
+
+        if !blocking.is_empty() {
+            return Err(format!(
+                "Cannot drain node — the following pods block eviction:\n  - {}",
+                blocking.join("\n  - ")
+            ));
+        }
+        Ok((evictable, skipped))
+    }
+
+    async fn evict_pod(&self, pod: &Pod) -> Result<(), Error> {
+        let name = pod.metadata.name.as_deref().unwrap_or_default();
+        let ns = pod.metadata.namespace.as_deref().unwrap_or_default();
+        debug!("Evicting pod {ns}/{name}");
+        Api::<Pod>::namespaced(self.client.clone(), ns)
+            .evict(name, &EvictParams::default())
+            .await
+            .map(|_| ())
+    }
+
+    /// Drains a node: cordon → classify → evict & wait.
+    pub async fn drain_node(&self, node_name: &str, options: &DrainOptions) -> Result<(), Error> {
+        debug!("Cordoning '{node_name}'");
+        self.cordon_node(node_name).await?;
+
+        let pods = self.list_pods_on_node(node_name).await?;
+        debug!("Found {} pod(s) on '{node_name}'", pods.len());
+
+        let (evictable, skipped) =
+            Self::classify_pods_for_drain(&pods, options).map_err(|msg| {
+                error!("{msg}");
+                Error::Discovery(DiscoveryError::MissingResource(msg))
+            })?;
+
+        for s in &skipped {
+            info!("Skipping pod: {s}");
+        }
+        if evictable.is_empty() {
+            info!("No pods to evict on '{node_name}'");
+            return Ok(());
+        }
+        info!("Evicting {} pod(s) from '{node_name}'", evictable.len());
+
+        let mut start = tokio::time::Instant::now();
+        let poll = Duration::from_secs(5);
+        let mut pending = evictable;
+
+        loop {
+            for pod in &pending {
+                match self.evict_pod(pod).await {
+                    Ok(()) => {}
+                    Err(Error::Api(ErrorResponse { code: 404, .. })) => {}
+                    Err(Error::Api(ErrorResponse { code: 429, .. })) => {
+                        warn!(
+                            "PDB blocked eviction of {}/{}; will retry",
+                            pod.metadata.namespace.as_deref().unwrap_or(""),
+                            pod.metadata.name.as_deref().unwrap_or("")
+                        );
+                    }
+                    Err(e) => {
+                        error!(
+                            "Failed to evict {}/{}: {e}",
+                            pod.metadata.namespace.as_deref().unwrap_or(""),
+                            pod.metadata.name.as_deref().unwrap_or("")
+                        );
+                        return Err(e);
+                    }
+                }
+            }
+
+            sleep(poll).await;
+
+            let mut still_present = Vec::new();
+            for pod in pending {
+                let ns = pod.metadata.namespace.as_deref().unwrap_or_default();
+                let name = pod.metadata.name.as_deref().unwrap_or_default();
+                match self.get_pod(name, Some(ns)).await? {
+                    Some(_) => still_present.push(pod),
+                    None => debug!("Pod {ns}/{name} evicted"),
+                }
+            }
+            pending = still_present;
+
+            if pending.is_empty() {
+                break;
+            }
+
+            if start.elapsed() > options.timeout {
+                match helper::prompt_drain_timeout_action(
+                    node_name,
+                    pending.len(),
+                    options.timeout,
+                )? {
+                    helper::DrainTimeoutAction::Accept => break,
+                    helper::DrainTimeoutAction::Retry => {
+                        start = tokio::time::Instant::now();
+                        continue;
+                    }
+                    helper::DrainTimeoutAction::Abort => {
+                        return Err(Error::Discovery(DiscoveryError::MissingResource(format!(
+                            "Drain aborted. {} pod(s) remaining on '{node_name}'",
+                            pending.len()
+                        ))));
+                    }
+                }
+            }
+            debug!("Waiting for {} pod(s) on '{node_name}'", pending.len());
+        }
+
+        debug!("'{node_name}' drained successfully");
+        Ok(())
+    }
+
+    /// Safely reboots a node: drain → reboot → wait for Ready → uncordon.
+    pub async fn reboot_node(
+        &self,
+        node_name: &str,
+        drain_options: &DrainOptions,
+        timeout: Duration,
+    ) -> Result<(), Error> {
+        info!("Starting reboot for '{node_name}'");
+        let node_api: Api<Node> = Api::all(self.client.clone());
+
+        let boot_id_before = node_api
+            .get(node_name)
+            .await?
+            .status
+            .as_ref()
+            .and_then(|s| s.node_info.as_ref())
+            .map(|ni| ni.boot_id.clone())
+            .ok_or_else(|| {
+                Error::Discovery(DiscoveryError::MissingResource(format!(
+                    "Node '{node_name}' has no boot_id in status"
+                )))
+            })?;
+
+        info!("Draining '{node_name}'");
+        self.drain_node(node_name, drain_options).await?;
+
+        let start = tokio::time::Instant::now();
+
+        info!("Scheduling reboot for '{node_name}'");
+        let reboot_cmd =
+            "echo rebooting ; nohup bash -c 'sleep 5 && nsenter -t 1 -m -- systemctl reboot'";
+        match self
+            .run_privileged_command_on_node(node_name, reboot_cmd)
+            .await
+        {
+            Ok(_) => debug!("Reboot command dispatched"),
+            Err(e) => debug!("Reboot command error (expected if node began shutdown): {e}"),
+        }
+
+        info!("Waiting for '{node_name}' to begin shutdown");
+        self.wait_for_node_not_ready(node_name, timeout.saturating_sub(start.elapsed()))
+            .await?;
+
+        if start.elapsed() > timeout {
+            return Err(Error::Discovery(DiscoveryError::MissingResource(format!(
+                "Timeout during reboot of '{node_name}' (shutdown phase)"
+            ))));
+        }
+
+        info!("Waiting for '{node_name}' to come back online");
+        self.wait_for_node_ready_with_timeout(node_name, timeout.saturating_sub(start.elapsed()))
+            .await?;
+
+        if start.elapsed() > timeout {
+            return Err(Error::Discovery(DiscoveryError::MissingResource(format!(
+                "Timeout during reboot of '{node_name}' (ready phase)"
+            ))));
+        }
+
+        let boot_id_after = node_api
+            .get(node_name)
+            .await?
+            .status
+            .as_ref()
+            .and_then(|s| s.node_info.as_ref())
+            .map(|ni| ni.boot_id.clone())
+            .ok_or_else(|| {
+                Error::Discovery(DiscoveryError::MissingResource(format!(
+                    "Node '{node_name}' has no boot_id after reboot"
+                )))
+            })?;
+
+        if boot_id_before == boot_id_after {
+            return Err(Error::Discovery(DiscoveryError::MissingResource(format!(
+                "Node '{node_name}' did not actually reboot (boot_id unchanged: {boot_id_before})"
+            ))));
+        }
+
+        info!("'{node_name}' rebooted ({boot_id_before} → {boot_id_after})");
+        self.uncordon_node(node_name).await?;
+        info!("'{node_name}' reboot complete ({:?})", start.elapsed());
+        Ok(())
+    }
+
+    /// Write a set of files to a node's filesystem via a privileged ephemeral pod.
+    pub async fn write_files_to_node(
+        &self,
+        node_name: &str,
+        files: &[NodeFile],
+    ) -> Result<String, Error> {
+        let ns = self.client.default_namespace();
+        let suffix = SystemTime::now()
+            .duration_since(UNIX_EPOCH)
+            .unwrap()
+            .as_millis();
+        let name = format!("harmony-k8s-writer-{suffix}");
+
+        debug!("Writing {} file(s) to '{node_name}'", files.len());
+
+        let mut data = BTreeMap::new();
+        let mut script = String::from("set -e\n");
+        for (i, file) in files.iter().enumerate() {
+            let key = format!("f{i}");
+            data.insert(key.clone(), file.content.clone());
+            script.push_str(&format!("mkdir -p \"$(dirname \"/host{}\")\"\n", file.path));
+            script.push_str(&format!("cp \"/payload/{key}\" \"/host{}\"\n", file.path));
+            script.push_str(&format!("chmod {:o} \"/host{}\"\n", file.mode, file.path));
+        }
+
+        let cm = ConfigMap {
+            metadata: ObjectMeta {
+                name: Some(name.clone()),
+                namespace: Some(ns.to_string()),
+                ..Default::default()
+            },
+            data: Some(data),
+            ..Default::default()
+        };
+
+        let cm_api: Api<ConfigMap> = Api::namespaced(self.client.clone(), ns);
+        cm_api.create(&PostParams::default(), &cm).await?;
+        debug!("Created ConfigMap '{name}'");
+
+        let (host_vol, host_mount) = helper::host_root_volume();
+        let payload_vol = Volume {
+            name: "payload".to_string(),
+            config_map: Some(ConfigMapVolumeSource {
+                name: name.clone(),
+                ..Default::default()
+            }),
+            ..Default::default()
+        };
+        let payload_mount = VolumeMount {
+            name: "payload".to_string(),
+            mount_path: "/payload".to_string(),
+            ..Default::default()
+        };
+
+        let bundle = helper::build_privileged_bundle(
+            PrivilegedPodConfig {
+                name: name.clone(),
+                namespace: ns.to_string(),
+                node_name: node_name.to_string(),
+                container_name: "writer".to_string(),
+                command: vec!["/bin/bash".to_string(), "-c".to_string(), script],
+                volumes: vec![payload_vol, host_vol],
+                volume_mounts: vec![payload_mount, host_mount],
+                host_pid: false,
+                host_network: false,
+            },
+            &self.get_k8s_distribution().await?,
+        );
+
+        bundle.apply(self).await?;
+        debug!("Created privileged pod bundle '{name}'");
+
+        let result = self.wait_for_pod_completion(&name, ns).await;
+
+        debug!("Cleaning up '{name}'");
+        let _ = bundle.delete(self).await;
+        let _ = cm_api.delete(&name, &DeleteParams::default()).await;
+
+        result
+    }
+
+    /// Run a privileged command on a node via an ephemeral pod.
+    pub async fn run_privileged_command_on_node(
+        &self,
+        node_name: &str,
+        command: &str,
+    ) -> Result<String, Error> {
+        let namespace = self.client.default_namespace();
+        let suffix = SystemTime::now()
+            .duration_since(UNIX_EPOCH)
+            .unwrap()
+            .as_millis();
+        let name = format!("harmony-k8s-cmd-{suffix}");
+
+        debug!("Running privileged command on '{node_name}': {command}");
+
+        let (host_vol, host_mount) = helper::host_root_volume();
+        let bundle = helper::build_privileged_bundle(
+            PrivilegedPodConfig {
+                name: name.clone(),
+                namespace: namespace.to_string(),
+                node_name: node_name.to_string(),
+                container_name: "runner".to_string(),
+                command: vec![
+                    "/bin/bash".to_string(),
+                    "-c".to_string(),
+                    command.to_string(),
+                ],
+                volumes: vec![host_vol],
+                volume_mounts: vec![host_mount],
+                host_pid: true,
+                host_network: true,
+            },
+            &self.get_k8s_distribution().await?,
+        );
+
+        bundle.apply(self).await?;
+        debug!("Privileged pod '{name}' created");
+
+        let result = self.wait_for_pod_completion(&name, namespace).await;
+
+        debug!("Cleaning up '{name}'");
+        let _ = bundle.delete(self).await;
+
+        result
+    }
+}
+
+// ── Tests ────────────────────────────────────────────────────────────────────
+
+#[cfg(test)]
+mod tests {
+    use k8s_openapi::api::core::v1::{EmptyDirVolumeSource, PodSpec, PodStatus, Volume};
+    use k8s_openapi::apimachinery::pkg::apis::meta::v1::{ObjectMeta, OwnerReference};
+
+    use super::*;
+
+    fn base_pod(name: &str, ns: &str) -> Pod {
+        Pod {
+            metadata: ObjectMeta {
+                name: Some(name.to_string()),
+                namespace: Some(ns.to_string()),
+                ..Default::default()
+            },
+            spec: Some(PodSpec::default()),
+            status: Some(PodStatus {
+                phase: Some("Running".to_string()),
+                ..Default::default()
+            }),
+        }
+    }
+
+    fn mirror_pod(name: &str, ns: &str) -> Pod {
+        let mut pod = base_pod(name, ns);
+        pod.metadata.annotations = Some(std::collections::BTreeMap::from([(
+            "kubernetes.io/config.mirror".to_string(),
+            "abc123".to_string(),
+        )]));
+        pod
+    }
+
+    fn daemonset_pod(name: &str, ns: &str) -> Pod {
+        let mut pod = base_pod(name, ns);
+        pod.metadata.owner_references = Some(vec![OwnerReference {
+            api_version: "apps/v1".to_string(),
+            kind: "DaemonSet".to_string(),
+            name: "some-ds".to_string(),
+            uid: "uid-ds".to_string(),
+            ..Default::default()
+        }]);
+        pod
+    }
+
+    fn emptydir_pod(name: &str, ns: &str) -> Pod {
+        let mut pod = base_pod(name, ns);
+        pod.spec = Some(PodSpec {
+            volumes: Some(vec![Volume {
+                name: "scratch".to_string(),
+                empty_dir: Some(EmptyDirVolumeSource::default()),
+                ..Default::default()
+            }]),
+            ..Default::default()
+        });
+        pod
+    }
+
+    fn completed_pod(name: &str, ns: &str, phase: &str) -> Pod {
+        let mut pod = base_pod(name, ns);
+        pod.status = Some(PodStatus {
+            phase: Some(phase.to_string()),
+            ..Default::default()
+        });
+        pod
+    }
+
+    fn default_opts() -> DrainOptions {
+        DrainOptions::default()
+    }
+
+    // All test bodies are identical to the original — only the module path changed.
+
+    #[test]
+    fn empty_pod_list_returns_empty_vecs() {
+        let (e, s) = K8sClient::classify_pods_for_drain(&[], &default_opts()).unwrap();
+        assert!(e.is_empty());
+        assert!(s.is_empty());
+    }
+
+    #[test]
+    fn normal_pod_is_evictable() {
+        let pods = vec![base_pod("web", "default")];
+        let (e, s) = K8sClient::classify_pods_for_drain(&pods, &default_opts()).unwrap();
+        assert_eq!(e.len(), 1);
+        assert!(s.is_empty());
+    }
+
+    #[test]
+    fn mirror_pod_is_skipped() {
+        let pods = vec![mirror_pod("kube-apiserver", "kube-system")];
+        let (e, s) = K8sClient::classify_pods_for_drain(&pods, &default_opts()).unwrap();
+        assert!(e.is_empty());
+        assert!(s[0].contains("mirror pod"));
+    }
+
+    #[test]
+    fn completed_pods_are_skipped() {
+        for phase in ["Succeeded", "Failed"] {
+            let pods = vec![completed_pod("job", "batch", phase)];
+            let (e, s) = K8sClient::classify_pods_for_drain(&pods, &default_opts()).unwrap();
+            assert!(e.is_empty());
+            assert!(s[0].contains("completed"));
+        }
+    }
+
+    #[test]
+    fn daemonset_skipped_when_ignored() {
+        let pods = vec![daemonset_pod("fluentd", "logging")];
+        let opts = DrainOptions {
+            ignore_daemonsets: true,
+            ..default_opts()
+        };
+        let (e, s) = K8sClient::classify_pods_for_drain(&pods, &opts).unwrap();
+        assert!(e.is_empty());
+        assert!(s[0].contains("DaemonSet-managed"));
+    }
+
+    #[test]
+    fn daemonset_blocks_when_not_ignored() {
+        let pods = vec![daemonset_pod("fluentd", "logging")];
+        let opts = DrainOptions {
+            ignore_daemonsets: false,
+            ..default_opts()
+        };
+        let err = K8sClient::classify_pods_for_drain(&pods, &opts).unwrap_err();
+        assert!(err.contains("DaemonSet") && err.contains("logging/fluentd"));
+    }
+
+    #[test]
+    fn emptydir_blocks_without_flag() {
+        let pods = vec![emptydir_pod("cache", "default")];
+        let opts = DrainOptions {
+            delete_emptydir_data: false,
+            ..default_opts()
+        };
+        let err = K8sClient::classify_pods_for_drain(&pods, &opts).unwrap_err();
+        assert!(err.contains("emptyDir") && err.contains("default/cache"));
+    }
+
+    #[test]
+    fn emptydir_evictable_with_flag() {
+        let pods = vec![emptydir_pod("cache", "default")];
+        let opts = DrainOptions {
+            delete_emptydir_data: true,
+            ..default_opts()
+        };
+        let (e, s) = K8sClient::classify_pods_for_drain(&pods, &opts).unwrap();
+        assert_eq!(e.len(), 1);
+        assert!(s.is_empty());
+    }
+
+    #[test]
+    fn multiple_blocking_all_reported() {
+        let pods = vec![daemonset_pod("ds", "ns1"), emptydir_pod("ed", "ns2")];
+        let opts = DrainOptions {
+            ignore_daemonsets: false,
+            delete_emptydir_data: false,
+            ..default_opts()
+        };
+        let err = K8sClient::classify_pods_for_drain(&pods, &opts).unwrap_err();
+        assert!(err.contains("ns1/ds") && err.contains("ns2/ed"));
+    }
+
+    #[test]
+    fn mixed_pods_classified_correctly() {
+        let pods = vec![
+            base_pod("web", "default"),
+            mirror_pod("kube-apiserver", "kube-system"),
+            daemonset_pod("fluentd", "logging"),
+            completed_pod("job", "batch", "Succeeded"),
+            base_pod("api", "default"),
+        ];
+        let (e, s) = K8sClient::classify_pods_for_drain(&pods, &default_opts()).unwrap();
+        let names: Vec<&str> = e
+            .iter()
+            .map(|p| p.metadata.name.as_deref().unwrap())
+            .collect();
+        assert_eq!(names, vec!["web", "api"]);
+        assert_eq!(s.len(), 3);
+    }
+
+    #[test]
+    fn mirror_checked_before_completed() {
+        let mut pod = mirror_pod("static-etcd", "kube-system");
+        pod.status = Some(PodStatus {
+            phase: Some("Succeeded".to_string()),
+            ..Default::default()
+        });
+        let (_, s) = K8sClient::classify_pods_for_drain(&[pod], &default_opts()).unwrap();
+        assert!(s[0].contains("mirror pod"), "got: {}", s[0]);
+    }
+
+    #[test]
+    fn completed_checked_before_daemonset() {
+        let mut pod = daemonset_pod("collector", "monitoring");
+        pod.status = Some(PodStatus {
+            phase: Some("Failed".to_string()),
+            ..Default::default()
+        });
+        let (_, s) = K8sClient::classify_pods_for_drain(&[pod], &default_opts()).unwrap();
+        assert!(s[0].contains("completed"), "got: {}", s[0]);
+    }
+}

harmony-k8s/src/pod.rs

@@ -0,0 +1,193 @@
+use std::time::Duration;
+
+use k8s_openapi::api::core::v1::Pod;
+use kube::{
+    Error,
+    api::{Api, AttachParams, ListParams},
+    error::DiscoveryError,
+    runtime::reflector::Lookup,
+};
+use log::debug;
+use tokio::io::AsyncReadExt;
+use tokio::time::sleep;
+
+use crate::client::K8sClient;
+
+impl K8sClient {
+    pub async fn get_pod(&self, name: &str, namespace: Option<&str>) -> Result<Option<Pod>, Error> {
+        let api: Api<Pod> = match namespace {
+            Some(ns) => Api::namespaced(self.client.clone(), ns),
+            None => Api::default_namespaced(self.client.clone()),
+        };
+        api.get_opt(name).await
+    }
+
+    pub async fn wait_for_pod_ready(
+        &self,
+        pod_name: &str,
+        namespace: Option<&str>,
+    ) -> Result<(), Error> {
+        let mut elapsed = 0u64;
+        let interval = 5u64;
+        let timeout_secs = 120u64;
+        loop {
+            if let Some(p) = self.get_pod(pod_name, namespace).await? {
+                if let Some(phase) = p.status.and_then(|s| s.phase) {
+                    if phase.to_lowercase() == "running" {
+                        return Ok(());
+                    }
+                }
+            }
+            if elapsed >= timeout_secs {
+                return Err(Error::Discovery(DiscoveryError::MissingResource(format!(
+                    "Pod '{}' in '{}' did not become ready within {timeout_secs}s",
+                    pod_name,
+                    namespace.unwrap_or("<default>"),
+                ))));
+            }
+            sleep(Duration::from_secs(interval)).await;
+            elapsed += interval;
+        }
+    }
+
+    /// Polls a pod until it reaches `Succeeded` or `Failed`, then returns its
+    /// logs.  Used internally by node operations.
+    pub(crate) async fn wait_for_pod_completion(
+        &self,
+        name: &str,
+        namespace: &str,
+    ) -> Result<String, Error> {
+        let api: Api<Pod> = Api::namespaced(self.client.clone(), namespace);
+        let poll_interval = Duration::from_secs(2);
+        for _ in 0..60 {
+            sleep(poll_interval).await;
+            let p = api.get(name).await?;
+            match p.status.and_then(|s| s.phase).as_deref() {
+                Some("Succeeded") => {
+                    let logs = api
+                        .logs(name, &Default::default())
+                        .await
+                        .unwrap_or_default();
+                    debug!("Pod {namespace}/{name} succeeded. Logs: {logs}");
+                    return Ok(logs);
+                }
+                Some("Failed") => {
+                    let logs = api
+                        .logs(name, &Default::default())
+                        .await
+                        .unwrap_or_default();
+                    debug!("Pod {namespace}/{name} failed. Logs: {logs}");
+                    return Err(Error::Discovery(DiscoveryError::MissingResource(format!(
+                        "Pod '{name}' failed.\n{logs}"
+                    ))));
+                }
+                _ => {}
+            }
+        }
+        Err(Error::Discovery(DiscoveryError::MissingResource(format!(
+            "Timed out waiting for pod '{name}'"
+        ))))
+    }
+
+    /// Execute a command in the first pod matching `{label}={name}`.
+    pub async fn exec_app_capture_output(
+        &self,
+        name: String,
+        label: String,
+        namespace: Option<&str>,
+        command: Vec<&str>,
+    ) -> Result<String, String> {
+        let api: Api<Pod> = match namespace {
+            Some(ns) => Api::namespaced(self.client.clone(), ns),
+            None => Api::default_namespaced(self.client.clone()),
+        };
+        let pod_list = api
+            .list(&ListParams::default().labels(&format!("{label}={name}")))
+            .await
+            .expect("Failed to list pods");
+
+        let pod_name = pod_list
+            .items
+            .first()
+            .expect("No matching pod")
+            .name()
+            .expect("Pod has no name")
+            .into_owned();
+
+        match api
+            .exec(
+                &pod_name,
+                command,
+                &AttachParams::default().stdout(true).stderr(true),
+            )
+            .await
+        {
+            Err(e) => Err(e.to_string()),
+            Ok(mut process) => {
+                let status = process
+                    .take_status()
+                    .expect("No status handle")
+                    .await
+                    .expect("Status channel closed");
+
+                if let Some(s) = status.status {
+                    let mut buf = String::new();
+                    if let Some(mut stdout) = process.stdout() {
+                        stdout
+                            .read_to_string(&mut buf)
+                            .await
+                            .map_err(|e| format!("Failed to read stdout: {e}"))?;
+                    }
+                    debug!("exec status: {} - {:?}", s, status.details);
+                    if s == "Success" { Ok(buf) } else { Err(s) }
+                } else {
+                    Err("No inner status from pod exec".to_string())
+                }
+            }
+        }
+    }
+
+    /// Execute a command in the first pod matching
+    /// `app.kubernetes.io/name={name}`.
+    pub async fn exec_app(
+        &self,
+        name: String,
+        namespace: Option<&str>,
+        command: Vec<&str>,
+    ) -> Result<(), String> {
+        let api: Api<Pod> = match namespace {
+            Some(ns) => Api::namespaced(self.client.clone(), ns),
+            None => Api::default_namespaced(self.client.clone()),
+        };
+        let pod_list = api
+            .list(&ListParams::default().labels(&format!("app.kubernetes.io/name={name}")))
+            .await
+            .expect("Failed to list pods");
+
+        let pod_name = pod_list
+            .items
+            .first()
+            .expect("No matching pod")
+            .name()
+            .expect("Pod has no name")
+            .into_owned();
+
+        match api.exec(&pod_name, command, &AttachParams::default()).await {
+            Err(e) => Err(e.to_string()),
+            Ok(mut process) => {
+                let status = process
+                    .take_status()
+                    .expect("No status handle")
+                    .await
+                    .expect("Status channel closed");
+
+                if let Some(s) = status.status {
+                    debug!("exec status: {} - {:?}", s, status.details);
+                    if s == "Success" { Ok(()) } else { Err(s) }
+                } else {
+                    Err("No inner status from pod exec".to_string())
+                }
+            }
+        }
+    }
+}

harmony-k8s/src/resources.rs

@@ -0,0 +1,316 @@
+use std::collections::HashMap;
+
+use k8s_openapi::api::{
+    apps::v1::Deployment,
+    core::v1::{Node, ServiceAccount},
+};
+use k8s_openapi::apiextensions_apiserver::pkg::apis::apiextensions::v1::CustomResourceDefinition;
+use kube::api::ApiResource;
+use kube::{
+    Error, Resource,
+    api::{Api, DynamicObject, GroupVersionKind, ListParams, ObjectList},
+    runtime::conditions,
+    runtime::wait::await_condition,
+};
+use log::debug;
+use serde::de::DeserializeOwned;
+use serde_json::Value;
+use std::time::Duration;
+
+use crate::client::K8sClient;
+use crate::types::ScopeResolver;
+
+impl K8sClient {
+    pub async fn has_healthy_deployment_with_label(
+        &self,
+        namespace: &str,
+        label_selector: &str,
+    ) -> Result<bool, Error> {
+        let api: Api<Deployment> = Api::namespaced(self.client.clone(), namespace);
+        let list = api
+            .list(&ListParams::default().labels(label_selector))
+            .await?;
+        for d in list.items {
+            let available = d
+                .status
+                .as_ref()
+                .and_then(|s| s.available_replicas)
+                .unwrap_or(0);
+            if available > 0 {
+                return Ok(true);
+            }
+            if let Some(conds) = d.status.as_ref().and_then(|s| s.conditions.as_ref()) {
+                if conds
+                    .iter()
+                    .any(|c| c.type_ == "Available" && c.status == "True")
+                {
+                    return Ok(true);
+                }
+            }
+        }
+        Ok(false)
+    }
+
+    pub async fn list_namespaces_with_healthy_deployments(
+        &self,
+        label_selector: &str,
+    ) -> Result<Vec<String>, Error> {
+        let api: Api<Deployment> = Api::all(self.client.clone());
+        let list = api
+            .list(&ListParams::default().labels(label_selector))
+            .await?;
+
+        let mut healthy_ns: HashMap<String, bool> = HashMap::new();
+        for d in list.items {
+            let ns = match d.metadata.namespace.clone() {
+                Some(n) => n,
+                None => continue,
+            };
+            let available = d
+                .status
+                .as_ref()
+                .and_then(|s| s.available_replicas)
+                .unwrap_or(0);
+            let is_healthy = if available > 0 {
+                true
+            } else {
+                d.status
+                    .as_ref()
+                    .and_then(|s| s.conditions.as_ref())
+                    .map(|c| {
+                        c.iter()
+                            .any(|c| c.type_ == "Available" && c.status == "True")
+                    })
+                    .unwrap_or(false)
+            };
+            if is_healthy {
+                healthy_ns.insert(ns, true);
+            }
+        }
+        Ok(healthy_ns.into_keys().collect())
+    }
+
+    pub async fn get_controller_service_account_name(
+        &self,
+        ns: &str,
+    ) -> Result<Option<String>, Error> {
+        let api: Api<Deployment> = Api::namespaced(self.client.clone(), ns);
+        let list = api
+            .list(&ListParams::default().labels("app.kubernetes.io/component=controller"))
+            .await?;
+        if let Some(dep) = list.items.first() {
+            if let Some(sa) = dep
+                .spec
+                .as_ref()
+                .and_then(|s| s.template.spec.as_ref())
+                .and_then(|s| s.service_account_name.clone())
+            {
+                return Ok(Some(sa));
+            }
+        }
+        Ok(None)
+    }
+
+    pub async fn list_clusterrolebindings_json(&self) -> Result<Vec<Value>, Error> {
+        let gvk = GroupVersionKind::gvk("rbac.authorization.k8s.io", "v1", "ClusterRoleBinding");
+        let ar = ApiResource::from_gvk(&gvk);
+        let api: Api<DynamicObject> = Api::all_with(self.client.clone(), &ar);
+        let list = api.list(&ListParams::default()).await?;
+        Ok(list
+            .items
+            .into_iter()
+            .map(|o| serde_json::to_value(&o).unwrap_or(Value::Null))
+            .collect())
+    }
+
+    pub async fn is_service_account_cluster_wide(&self, sa: &str, ns: &str) -> Result<bool, Error> {
+        let sa_user = format!("system:serviceaccount:{ns}:{sa}");
+        for crb in self.list_clusterrolebindings_json().await? {
+            if let Some(subjects) = crb.get("subjects").and_then(|s| s.as_array()) {
+                for subj in subjects {
+                    let kind = subj.get("kind").and_then(|v| v.as_str()).unwrap_or("");
+                    let name = subj.get("name").and_then(|v| v.as_str()).unwrap_or("");
+                    let subj_ns = subj.get("namespace").and_then(|v| v.as_str()).unwrap_or("");
+                    if (kind == "ServiceAccount" && name == sa && subj_ns == ns)
+                        || (kind == "User" && name == sa_user)
+                    {
+                        return Ok(true);
+                    }
+                }
+            }
+        }
+        Ok(false)
+    }
+
+    pub async fn has_crd(&self, name: &str) -> Result<bool, Error> {
+        let api: Api<CustomResourceDefinition> = Api::all(self.client.clone());
+        let crds = api
+            .list(&ListParams::default().fields(&format!("metadata.name={name}")))
+            .await?;
+        Ok(!crds.items.is_empty())
+    }
+
+    pub async fn service_account_api(&self, namespace: &str) -> Api<ServiceAccount> {
+        Api::namespaced(self.client.clone(), namespace)
+    }
+
+    pub async fn get_resource_json_value(
+        &self,
+        name: &str,
+        namespace: Option<&str>,
+        gvk: &GroupVersionKind,
+    ) -> Result<DynamicObject, Error> {
+        let ar = ApiResource::from_gvk(gvk);
+        let api: Api<DynamicObject> = match namespace {
+            Some(ns) => Api::namespaced_with(self.client.clone(), ns, &ar),
+            None => Api::default_namespaced_with(self.client.clone(), &ar),
+        };
+        api.get(name).await
+    }
+
+    pub async fn get_secret_json_value(
+        &self,
+        name: &str,
+        namespace: Option<&str>,
+    ) -> Result<DynamicObject, Error> {
+        self.get_resource_json_value(
+            name,
+            namespace,
+            &GroupVersionKind {
+                group: String::new(),
+                version: "v1".to_string(),
+                kind: "Secret".to_string(),
+            },
+        )
+        .await
+    }
+
+    pub async fn get_deployment(
+        &self,
+        name: &str,
+        namespace: Option<&str>,
+    ) -> Result<Option<Deployment>, Error> {
+        let api: Api<Deployment> = match namespace {
+            Some(ns) => {
+                debug!("Getting namespaced deployment '{name}' in '{ns}'");
+                Api::namespaced(self.client.clone(), ns)
+            }
+            None => {
+                debug!("Getting deployment '{name}' in default namespace");
+                Api::default_namespaced(self.client.clone())
+            }
+        };
+        api.get_opt(name).await
+    }
+
+    pub async fn scale_deployment(
+        &self,
+        name: &str,
+        namespace: Option<&str>,
+        replicas: u32,
+    ) -> Result<(), Error> {
+        let api: Api<Deployment> = match namespace {
+            Some(ns) => Api::namespaced(self.client.clone(), ns),
+            None => Api::default_namespaced(self.client.clone()),
+        };
+        use kube::api::{Patch, PatchParams};
+        use serde_json::json;
+        let patch = json!({ "spec": { "replicas": replicas } });
+        api.patch_scale(name, &PatchParams::default(), &Patch::Merge(&patch))
+            .await?;
+        Ok(())
+    }
+
+    pub async fn delete_deployment(
+        &self,
+        name: &str,
+        namespace: Option<&str>,
+    ) -> Result<(), Error> {
+        let api: Api<Deployment> = match namespace {
+            Some(ns) => Api::namespaced(self.client.clone(), ns),
+            None => Api::default_namespaced(self.client.clone()),
+        };
+        api.delete(name, &kube::api::DeleteParams::default())
+            .await?;
+        Ok(())
+    }
+
+    pub async fn wait_until_deployment_ready(
+        &self,
+        name: &str,
+        namespace: Option<&str>,
+        timeout: Option<Duration>,
+    ) -> Result<(), String> {
+        let api: Api<Deployment> = match namespace {
+            Some(ns) => Api::namespaced(self.client.clone(), ns),
+            None => Api::default_namespaced(self.client.clone()),
+        };
+        let timeout = timeout.unwrap_or(Duration::from_secs(120));
+        let establish = await_condition(api, name, conditions::is_deployment_completed());
+        tokio::time::timeout(timeout, establish)
+            .await
+            .map(|_| ())
+            .map_err(|_| "Timed out waiting for deployment".to_string())
+    }
+
+    /// Gets a single named resource, using the correct API scope for `K`.
+    pub async fn get_resource<K>(
+        &self,
+        name: &str,
+        namespace: Option<&str>,
+    ) -> Result<Option<K>, Error>
+    where
+        K: Resource + Clone + std::fmt::Debug + DeserializeOwned,
+        <K as Resource>::Scope: ScopeResolver<K>,
+        <K as Resource>::DynamicType: Default,
+    {
+        let api: Api<K> =
+            <<K as Resource>::Scope as ScopeResolver<K>>::get_api(&self.client, namespace);
+        api.get_opt(name).await
+    }
+
+    pub async fn list_resources<K>(
+        &self,
+        namespace: Option<&str>,
+        list_params: Option<ListParams>,
+    ) -> Result<ObjectList<K>, Error>
+    where
+        K: Resource + Clone + std::fmt::Debug + DeserializeOwned,
+        <K as Resource>::Scope: ScopeResolver<K>,
+        <K as Resource>::DynamicType: Default,
+    {
+        let api: Api<K> =
+            <<K as Resource>::Scope as ScopeResolver<K>>::get_api(&self.client, namespace);
+        api.list(&list_params.unwrap_or_default()).await
+    }
+
+    pub async fn list_all_resources_with_labels<K>(&self, labels: &str) -> Result<Vec<K>, Error>
+    where
+        K: Resource + Clone + std::fmt::Debug + DeserializeOwned,
+        <K as Resource>::DynamicType: Default,
+    {
+        Api::<K>::all(self.client.clone())
+            .list(&ListParams::default().labels(labels))
+            .await
+            .map(|l| l.items)
+    }
+
+    pub async fn get_all_resource_in_all_namespace<K>(&self) -> Result<Vec<K>, Error>
+    where
+        K: Resource + Clone + std::fmt::Debug + DeserializeOwned,
+        <K as Resource>::Scope: ScopeResolver<K>,
+        <K as Resource>::DynamicType: Default,
+    {
+        Api::<K>::all(self.client.clone())
+            .list(&Default::default())
+            .await
+            .map(|l| l.items)
+    }
+
+    pub async fn get_nodes(
+        &self,
+        list_params: Option<ListParams>,
+    ) -> Result<ObjectList<Node>, Error> {
+        self.list_resources(None, list_params).await
+    }
+}

harmony-k8s/src/types.rs

@@ -0,0 +1,100 @@
+use std::time::Duration;
+
+use k8s_openapi::{ClusterResourceScope, NamespaceResourceScope};
+use kube::{Api, Client, Resource};
+use serde::Serialize;
+
+/// Which Kubernetes distribution is running. Detected once at runtime via
+/// [`crate::discovery::K8sClient::get_k8s_distribution`].
+#[derive(Debug, Clone, PartialEq, Eq, Serialize)]
+pub enum KubernetesDistribution {
+    Default,
+    OpenshiftFamily,
+    K3sFamily,
+}
+
+/// A file to be written to a node's filesystem.
+#[derive(Debug, Clone)]
+pub struct NodeFile {
+    /// Absolute path on the host where the file should be written.
+    pub path: String,
+    /// Content of the file.
+    pub content: String,
+    /// UNIX permissions (e.g. `0o600`).
+    pub mode: u32,
+}
+
+/// Options controlling the behaviour of a [`crate::K8sClient::drain_node`] operation.
+#[derive(Debug, Clone)]
+pub struct DrainOptions {
+    /// Evict pods that use `emptyDir` volumes (ephemeral data is lost).
+    /// Equivalent to `kubectl drain --delete-emptydir-data`.
+    pub delete_emptydir_data: bool,
+    /// Silently skip DaemonSet-managed pods instead of blocking the drain.
+    /// Equivalent to `kubectl drain --ignore-daemonsets`.
+    pub ignore_daemonsets: bool,
+    /// Maximum wall-clock time to wait for all evictions to complete.
+    pub timeout: Duration,
+}
+
+impl Default for DrainOptions {
+    fn default() -> Self {
+        Self {
+            delete_emptydir_data: false,
+            ignore_daemonsets: true,
+            timeout: Duration::from_secs(1),
+        }
+    }
+}
+
+impl DrainOptions {
+    pub fn default_ignore_daemonset_delete_emptydir_data() -> Self {
+        Self {
+            delete_emptydir_data: true,
+            ignore_daemonsets: true,
+            ..Self::default()
+        }
+    }
+}
+
+/// Controls how [`crate::K8sClient::apply_with_strategy`] behaves when the
+/// resource already exists (or does not).
+pub enum WriteMode {
+    /// Server-side apply; create if absent, update if present (default).
+    CreateOrUpdate,
+    /// POST only; return an error if the resource already exists.
+    Create,
+    /// Server-side apply only; return an error if the resource does not exist.
+    Update,
+}
+
+// ── Scope resolution trait ───────────────────────────────────────────────────
+
+/// Resolves the correct [`kube::Api`] for a resource type based on its scope
+/// (cluster-wide vs. namespace-scoped).
+pub trait ScopeResolver<K: Resource> {
+    fn get_api(client: &Client, ns: Option<&str>) -> Api<K>;
+}
+
+impl<K> ScopeResolver<K> for ClusterResourceScope
+where
+    K: Resource<Scope = ClusterResourceScope>,
+    <K as Resource>::DynamicType: Default,
+{
+    fn get_api(client: &Client, _ns: Option<&str>) -> Api<K> {
+        Api::all(client.clone())
+    }
+}
+
+impl<K> ScopeResolver<K> for NamespaceResourceScope
+where
+    K: Resource<Scope = NamespaceResourceScope>,
+    <K as Resource>::DynamicType: Default,
+{
+    fn get_api(client: &Client, ns: Option<&str>) -> Api<K> {
+        match ns {
+            Some(ns) => Api::namespaced(client.clone(), ns),
+            None => Api::default_namespaced(client.clone()),
+        }
+    }
+}

harmony/Cargo.toml

@@ -21,6 +21,8 @@ semver = "1.0.23"
 serde.workspace = true
 serde_json.workspace = true
 tokio.workspace = true
+tokio-retry.workspace = true
+tokio-util.workspace = true
 derive-new.workspace = true
 log.workspace = true
 env_logger.workspace = true
@@ -31,6 +33,7 @@ opnsense-config-xml = { path = "../opnsense-config-xml" }
 harmony_macros = { path = "../harmony_macros" }
 harmony_types = { path = "../harmony_types" }
 harmony_execution = { path = "../harmony_execution" }
+harmony-k8s = { path = "../harmony-k8s" }
 uuid.workspace = true
 url.workspace = true
 kube = { workspace = true, features = ["derive"] }
@@ -60,7 +63,6 @@ temp-dir = "0.1.14"
 dyn-clone = "1.0.19"
 similar.workspace = true
 futures-util = "0.3.31"
-tokio-util = "0.7.15"
 strum = { version = "0.27.1", features = ["derive"] }
 tempfile.workspace = true
 serde_with = "3.14.0"
@@ -80,7 +82,7 @@ sqlx.workspace = true
 inquire.workspace = true
 brocade = { path = "../brocade" }
 option-ext = "0.2.0"
-tokio-retry = "0.3.0"
+rand.workspace = true
 
 [dev-dependencies]
 pretty_assertions.workspace = true

harmony/src/domain/interpret/mod.rs

@@ -4,8 +4,6 @@ use std::error::Error;
 use async_trait::async_trait;
 use derive_new::new;
 
-use crate::inventory::HostRole;
-
 use super::{
     data::Version, executors::ExecutorError, inventory::Inventory, topology::PreparationError,
 };

harmony/src/domain/topology/ha_cluster.rs

@@ -1,4 +1,5 @@
 use async_trait::async_trait;
+use harmony_k8s::K8sClient;
 use harmony_macros::ip;
 use harmony_types::{
     id::Id,
@@ -16,7 +17,7 @@ use super::{
     DHCPStaticEntry, DhcpServer, DnsRecord, DnsRecordType, DnsServer, Firewall, HostNetworkConfig,
     HttpServer, IpAddress, K8sclient, LoadBalancer, LoadBalancerService, LogicalHost, NetworkError,
     NetworkManager, PreparationError, PreparationOutcome, Router, Switch, SwitchClient,
-    SwitchError, TftpServer, Topology, k8s::K8sClient,
+    SwitchError, TftpServer, Topology,
 };
 use std::{
     process::Command,

harmony/src/domain/topology/k8s_anywhere/k8s_anywhere.rs

@@ -2,6 +2,7 @@ use std::{collections::BTreeMap, process::Command, sync::Arc, time::Duration};
 
 use async_trait::async_trait;
 use base64::{Engine, engine::general_purpose};
+use harmony_k8s::{K8sClient, KubernetesDistribution};
 use harmony_types::rfc1123::Rfc1123Name;
 use k8s_openapi::api::{
     core::v1::{Pod, Secret},
@@ -58,7 +59,6 @@ use crate::{
 use super::super::{
     DeploymentTarget, HelmCommand, K8sclient, MultiTargetTopology, PreparationError,
     PreparationOutcome, Topology,
-    k8s::K8sClient,
     oberservability::monitoring::AlertReceiver,
     tenant::{
         TenantConfig, TenantManager,
@@ -76,13 +76,6 @@ struct K8sState {
     message: String,
 }
 
-#[derive(Debug, Clone, Serialize)]
-pub enum KubernetesDistribution {
-    OpenshiftFamily,
-    K3sFamily,
-    Default,
-}
-
 #[derive(Debug, Clone)]
 enum K8sSource {
     LocalK3d,
@@ -116,6 +109,13 @@ impl K8sclient for K8sAnywhereTopology {
 
 #[async_trait]
 impl TlsRouter for K8sAnywhereTopology {
+    async fn get_public_domain(&self) -> Result<String, String> {
+        match &self.config.public_domain {
+            Some(public_domain) => Ok(public_domain.to_string()),
+            None => Err("Public domain not available".to_string()),
+        }
+    }
+
     async fn get_internal_domain(&self) -> Result<Option<String>, String> {
         match self.get_k8s_distribution().await.map_err(|e| {
             format!(
@@ -1131,6 +1131,7 @@ pub struct K8sAnywhereConfig {
     ///
     /// If the context name is not found, it will fail to initialize.
     pub k8s_context: Option<String>,
+    public_domain: Option<String>,
 }
 
 impl K8sAnywhereConfig {
@@ -1158,6 +1159,7 @@ impl K8sAnywhereConfig {
 
         let mut kubeconfig: Option<String> = None;
         let mut k8s_context: Option<String> = None;
+        let mut public_domain: Option<String> = None;
 
         for part in env_var_value.split(',') {
             let kv: Vec<&str> = part.splitn(2, '=').collect();
@@ -1165,6 +1167,7 @@ impl K8sAnywhereConfig {
                 match kv[0].trim() {
                     "kubeconfig" => kubeconfig = Some(kv[1].trim().to_string()),
                     "context" => k8s_context = Some(kv[1].trim().to_string()),
+                    "public_domain" => public_domain = Some(kv[1].trim().to_string()),
                     _ => {}
                 }
             }
@@ -1182,6 +1185,7 @@ impl K8sAnywhereConfig {
         K8sAnywhereConfig {
             kubeconfig,
             k8s_context,
+            public_domain,
             use_system_kubeconfig,
             autoinstall: false,
             use_local_k3d: false,
@@ -1224,6 +1228,7 @@ impl K8sAnywhereConfig {
             use_local_k3d: std::env::var("HARMONY_USE_LOCAL_K3D")
                 .map_or_else(|_| true, |v| v.parse().ok().unwrap_or(true)),
             k8s_context: std::env::var("HARMONY_K8S_CONTEXT").ok(),
+            public_domain: std::env::var("HARMONY_PUBLIC_DOMAIN").ok(),
         }
     }
 }

harmony/src/domain/topology/k8s_anywhere/postgres.rs

@@ -1,7 +1,6 @@
 use async_trait::async_trait;
 
 use crate::{
-    interpret::Outcome,
     inventory::Inventory,
     modules::postgresql::{
         K8sPostgreSQLScore,

harmony/src/domain/topology/load_balancer.rs

@@ -106,6 +106,7 @@ pub enum SSL {
 
 #[derive(Debug, Clone, PartialEq, Serialize)]
 pub enum HealthCheck {
-    HTTP(String, HttpMethod, HttpStatusCode, SSL),
+    /// HTTP(None, "/healthz/ready", HttpMethod::GET, HttpStatusCode::Success2xx, SSL::Disabled)
+    HTTP(Option<u16>, String, HttpMethod, HttpStatusCode, SSL),
     TCP(Option<u16>),
 }

harmony/src/domain/topology/mod.rs

@@ -16,7 +16,6 @@ pub mod tenant;
 use derive_new::new;
 pub use k8s_anywhere::*;
 pub use localhost::*;
-pub mod k8s;
 mod load_balancer;
 pub mod router;
 mod tftp;

harmony/src/domain/topology/network.rs

@@ -9,6 +9,7 @@ use std::{
 use async_trait::async_trait;
 use brocade::PortOperatingMode;
 use derive_new::new;
+use harmony_k8s::K8sClient;
 use harmony_types::{
     id::Id,
     net::{IpAddress, MacAddress},
@@ -18,7 +19,7 @@ use serde::Serialize;
 
 use crate::executors::ExecutorError;
 
-use super::{LogicalHost, k8s::K8sClient};
+use super::LogicalHost;
 
 #[derive(Debug)]
 pub struct DHCPStaticEntry {

harmony/src/domain/topology/router.rs

@@ -122,4 +122,6 @@ pub trait TlsRouter: Send + Sync {
 
     /// Returns the port that this router exposes externally.
     async fn get_router_port(&self) -> u16;
+
+    async fn get_public_domain(&self) -> Result<String, String>;
 }

harmony/src/domain/topology/tenant/k8s.rs

@@ -1,10 +1,8 @@
 use std::sync::Arc;
 
-use crate::{
-    executors::ExecutorError,
-    topology::k8s::{ApplyStrategy, K8sClient},
-};
+use crate::executors::ExecutorError;
 use async_trait::async_trait;
+use harmony_k8s::K8sClient;
 use k8s_openapi::{
     api::{
         core::v1::{LimitRange, Namespace, ResourceQuota},
@@ -14,7 +12,7 @@ use k8s_openapi::{
     },
     apimachinery::pkg::util::intstr::IntOrString,
 };
-use kube::{Resource, api::DynamicObject};
+use kube::Resource;
 use log::debug;
 use serde::de::DeserializeOwned;
 use serde_json::json;
@@ -59,7 +57,6 @@ impl K8sTenantManager {
     ) -> Result<K, ExecutorError>
     where
         <K as kube::Resource>::DynamicType: Default,
-        <K as kube::Resource>::Scope: ApplyStrategy<K>,
     {
         self.apply_labels(&mut resource, config);
         self.k8s_client

harmony/src/infra/network_manager.rs

@@ -5,6 +5,7 @@ use std::{
 
 use askama::Template;
 use async_trait::async_trait;
+use harmony_k8s::{DrainOptions, K8sClient, NodeFile};
 use harmony_types::id::Id;
 use k8s_openapi::api::core::v1::Node;
 use kube::{
@@ -15,10 +16,7 @@ use log::{debug, info, warn};
 
 use crate::{
     modules::okd::crd::nmstate,
-    topology::{
-        HostNetworkConfig, NetworkError, NetworkManager,
-        k8s::{DrainOptions, K8sClient, NodeFile},
-    },
+    topology::{HostNetworkConfig, NetworkError, NetworkManager},
 };
 
 /// NetworkManager bond configuration template

harmony/src/infra/opnsense/load_balancer.rs

@@ -216,7 +216,15 @@ pub(crate) fn get_health_check_for_backend(
                     SSL::Other(other.to_string())
                 }
             };
-            Some(HealthCheck::HTTP(path, method, status_code, ssl))
+
+            let port = haproxy_health_check
+                .checkport
+                .content_string()
+                .parse::<u16>()
+                .ok();
+            debug!("Found haproxy healthcheck port {port:?}");
+
+            Some(HealthCheck::HTTP(port, path, method, status_code, ssl))
         }
         _ => panic!("Received unsupported health check type {}", uppercase),
     }
@@ -251,7 +259,7 @@ pub(crate) fn harmony_load_balancer_service_to_haproxy_xml(
     // frontend points to backend
     let healthcheck = if let Some(health_check) = &service.health_check {
         match health_check {
-            HealthCheck::HTTP(path, http_method, _http_status_code, ssl) => {
+            HealthCheck::HTTP(port, path, http_method, _http_status_code, ssl) => {
                 let ssl: MaybeString = match ssl {
                     SSL::SSL => "ssl".into(),
                     SSL::SNI => "sslni".into(),
@@ -259,14 +267,21 @@ pub(crate) fn harmony_load_balancer_service_to_haproxy_xml(
                     SSL::Default => "".into(),
                     SSL::Other(other) => other.as_str().into(),
                 };
+                let path_without_query = path.split_once('?').map_or(path.as_str(), |(p, _)| p);
+                let (port, port_name) = match port {
+                    Some(port) => (Some(port.to_string()), port.to_string()),
+                    None => (None, "serverport".to_string()),
+                };
+
                 let haproxy_check = HAProxyHealthCheck {
-                    name: format!("HTTP_{http_method}_{path}"),
+                    name: format!("HTTP_{http_method}_{path_without_query}_{port_name}"),
                     uuid: Uuid::new_v4().to_string(),
-                    http_method: http_method.to_string().into(),
+                    http_method: http_method.to_string().to_lowercase().into(),
                     health_check_type: "http".to_string(),
                     http_uri: path.clone().into(),
                     interval: "2s".to_string(),
                     ssl,
+                    checkport: MaybeString::from(port.map(|p| p.to_string())),
                     ..Default::default()
                 };
 
@@ -305,7 +320,10 @@ pub(crate) fn harmony_load_balancer_service_to_haproxy_xml(
     let mut backend = HAProxyBackend {
         uuid: Uuid::new_v4().to_string(),
         enabled: 1,
-        name: format!("backend_{}", service.listening_port),
+        name: format!(
+            "backend_{}",
+            service.listening_port.to_string().replace(':', "_")
+        ),
         algorithm: "roundrobin".to_string(),
         random_draws: Some(2),
         stickiness_expire: "30m".to_string(),
@@ -337,10 +355,22 @@ pub(crate) fn harmony_load_balancer_service_to_haproxy_xml(
     let frontend = Frontend {
         uuid: uuid::Uuid::new_v4().to_string(),
         enabled: 1,
-        name: format!("frontend_{}", service.listening_port),
+        name: format!(
+            "frontend_{}",
+            service.listening_port.to_string().replace(':', "_")
+        ),
         bind: service.listening_port.to_string(),
         mode: "tcp".to_string(), // TODO do not depend on health check here
         default_backend: Some(backend.uuid.clone()),
+        stickiness_expire: "30m".to_string().into(),
+        stickiness_size: "50k".to_string().into(),
+        stickiness_conn_rate_period: "10s".to_string().into(),
+        stickiness_sess_rate_period: "10s".to_string().into(),
+        stickiness_http_req_rate_period: "10s".to_string().into(),
+        stickiness_http_err_rate_period: "10s".to_string().into(),
+        stickiness_bytes_in_rate_period: "1m".to_string().into(),
+        stickiness_bytes_out_rate_period: "1m".to_string().into(),
+        ssl_hsts_max_age: 15768000,
         ..Default::default()
     };
     info!("HAPRoxy frontend and backend mode currently hardcoded to tcp");

harmony/src/modules/application/backend_app.rs

@@ -1,5 +1,5 @@
 use async_trait::async_trait;
-use log::{debug, info, trace};
+use log::{debug, info};
 use serde::Serialize;
 use std::path::PathBuf;
 

harmony/src/modules/application/features/helm_argocd_score.rs

@@ -1,4 +1,5 @@
 use async_trait::async_trait;
+use harmony_k8s::K8sClient;
 use harmony_macros::hurl;
 use log::{debug, info, trace, warn};
 use non_blank_string_rs::NonBlankString;
@@ -14,7 +15,7 @@ use crate::{
         helm::chart::{HelmChartScore, HelmRepository},
     },
     score::Score,
-    topology::{HelmCommand, K8sclient, Topology, ingress::Ingress, k8s::K8sClient},
+    topology::{HelmCommand, K8sclient, Topology, ingress::Ingress},
 };
 use harmony_types::id::Id;
 

harmony/src/modules/application/rust.rs

@@ -1,6 +1,5 @@
 use std::fs::{self};
-use std::path::{Path, PathBuf};
-use std::process;
+use std::path::PathBuf;
 use std::sync::Arc;
 
 use async_trait::async_trait;
@@ -65,6 +64,7 @@ pub struct RustWebapp {
     ///
     /// This is the place to put the public host name if this is a public facing webapp.
     pub dns: String,
+    pub version: String,
 }
 
 impl Application for RustWebapp {
@@ -465,6 +465,7 @@ impl RustWebapp {
 
         let app_name = &self.name;
         let service_port = self.service_port;
+        let version = &self.version;
         // Create Chart.yaml
         let chart_yaml = format!(
             r#"
@@ -472,7 +473,7 @@ apiVersion: v2
 name: {chart_name}
 description: A Helm chart for the {app_name} web application.
 type: application
-version: 0.2.1
+version: {version}
 appVersion: "{image_tag}"
 "#,
         );

harmony/src/modules/argocd/mod.rs

@@ -1,8 +1,9 @@
 use std::sync::Arc;
 
+use harmony_k8s::K8sClient;
 use log::{debug, info};
 
-use crate::{interpret::InterpretError, topology::k8s::K8sClient};
+use crate::interpret::InterpretError;
 
 #[derive(Clone, Debug, PartialEq, Eq)]
 pub enum ArgoScope {

harmony/src/modules/cert_manager/cluster_issuer.rs

@@ -1,3 +1,4 @@
+use harmony_k8s::K8sClient;
 use std::sync::Arc;
 
 use async_trait::async_trait;
@@ -11,7 +12,7 @@ use crate::{
     interpret::{Interpret, InterpretError, InterpretName, InterpretStatus, Outcome},
     inventory::Inventory,
     score::Score,
-    topology::{K8sclient, Topology, k8s::K8sClient},
+    topology::{K8sclient, Topology},
 };
 
 #[derive(Clone, Debug, Serialize)]

harmony/src/modules/k8s/failover.rs

@@ -3,7 +3,8 @@ use std::sync::Arc;
 use async_trait::async_trait;
 use log::warn;
 
-use crate::topology::{FailoverTopology, K8sclient, k8s::K8sClient};
+use crate::topology::{FailoverTopology, K8sclient};
+use harmony_k8s::K8sClient;
 
 #[async_trait]
 impl<T: K8sclient> K8sclient for FailoverTopology<T> {

harmony/src/modules/k8s/resource.rs

@@ -1,5 +1,5 @@
 use async_trait::async_trait;
-use k8s_openapi::{NamespaceResourceScope, ResourceScope};
+use k8s_openapi::ResourceScope;
 use kube::Resource;
 use log::info;
 use serde::{Serialize, de::DeserializeOwned};
@@ -109,7 +109,7 @@ where
         topology
             .k8s_client()
             .await
-            .expect("Environment should provide enough information to instanciate a client")
+            .map_err(|e| InterpretError::new(format!("Failed to get k8s client : {e}")))?
             .apply_many(&self.score.resource, self.score.namespace.as_deref())
             .await?;
 

harmony/src/modules/monitoring/kube_prometheus/crd/crd_alertmanager_config.rs

@@ -6,7 +6,7 @@ use schemars::JsonSchema;
 use serde::{Deserialize, Serialize};
 
 use crate::{
-    interpret::{InterpretError, Outcome},
+    interpret::InterpretError,
     inventory::Inventory,
     modules::{
         monitoring::{
@@ -17,10 +17,10 @@ use crate::{
     topology::{
         K8sclient, Topology,
         installable::Installable,
-        k8s::K8sClient,
         oberservability::monitoring::{AlertReceiver, AlertSender, ScrapeTarget},
     },
 };
+use harmony_k8s::K8sClient;
 
 #[derive(CustomResource, Serialize, Deserialize, Debug, Clone, JsonSchema)]
 #[kube(

harmony/src/modules/monitoring/kube_prometheus/crd/rhob_alertmanager_config.rs

@@ -4,10 +4,8 @@ use kube::CustomResource;
 use schemars::JsonSchema;
 use serde::{Deserialize, Serialize};
 
-use crate::topology::{
-    k8s::K8sClient,
-    oberservability::monitoring::{AlertReceiver, AlertSender},
-};
+use crate::topology::oberservability::monitoring::{AlertReceiver, AlertSender};
+use harmony_k8s::K8sClient;
 
 #[derive(CustomResource, Serialize, Deserialize, Debug, Clone, JsonSchema)]
 #[kube(

harmony/src/modules/monitoring/ntfy/ntfy.rs

@@ -11,8 +11,9 @@ use crate::{
     inventory::Inventory,
     modules::monitoring::ntfy::helm::ntfy_helm_chart::ntfy_helm_chart_score,
     score::Score,
-    topology::{HelmCommand, K8sclient, MultiTargetTopology, Topology, k8s::K8sClient},
+    topology::{HelmCommand, K8sclient, MultiTargetTopology, Topology},
 };
+use harmony_k8s::K8sClient;
 use harmony_types::id::Id;
 
 #[derive(Debug, Clone, Serialize)]

harmony/src/modules/monitoring/okd/config.rs

@@ -1,9 +1,7 @@
 use std::{collections::BTreeMap, sync::Arc};
 
-use crate::{
-    interpret::{InterpretError, Outcome},
-    topology::k8s::K8sClient,
-};
+use crate::interpret::{InterpretError, Outcome};
+use harmony_k8s::K8sClient;
 use k8s_openapi::api::core::v1::ConfigMap;
 use kube::api::ObjectMeta;
 

harmony/src/modules/nats/pki.rs

@@ -1,3 +1,7 @@
+use std::time::Duration;
+
+use tokio_retry::{Retry, strategy::ExponentialBackoff};
+
 use crate::modules::{
     cert_manager::{
         capability::{CertificateManagement, CertificateManagementConfig},
@@ -69,9 +73,28 @@ where
             .await
             .map_err(|e| e.to_string())?;
 
-        self.topology
+        let strategy = ExponentialBackoff::from_millis(250)
+            .factor(2)
+            .max_delay(Duration::from_millis(1000))
+            .take(10);
+
+        Retry::spawn(strategy, || async {
+            log::debug!("Attempting CA cert fetch");
+
+            let res = self
+                .topology
                 .get_ca_certificate(root_ca_cert_name.into(), &root_ca_config)
+                .await;
+
+            match res {
+                Ok(cert) => Ok(cert),
+                Err(e) => {
+                    log::warn!("Retryable error: {:?}", e);
+                    Err(e)
+                }
+            }
+        })
         .await
-            .map_err(|e| e.to_string())
+        .map_err(|e| format!("Retries exhausted: {:?}", e))
     }
 }

harmony/src/modules/nats/score_nats_k8s.rs

@@ -1,6 +1,7 @@
 use std::{collections::BTreeMap, str::FromStr};
 
 use async_trait::async_trait;
+use harmony_k8s::KubernetesDistribution;
 use harmony_macros::hurl;
 use harmony_secret::{Secret, SecretManager};
 use harmony_types::id::Id;
@@ -25,7 +26,7 @@ use crate::{
         },
     },
     score::Score,
-    topology::{HelmCommand, K8sclient, KubernetesDistribution, TlsRouter, Topology},
+    topology::{HelmCommand, K8sclient, TlsRouter, Topology},
 };
 
 #[derive(Debug, Clone, Serialize)]

harmony/src/modules/network/failover.rs

@@ -4,7 +4,20 @@ use log::warn;
 use crate::topology::{FailoverTopology, TlsRoute, TlsRouter};
 
 #[async_trait]
-impl<T: TlsRouter> TlsRouter for FailoverTopology<T> {
+impl<T: TlsRouter + Send> TlsRouter for FailoverTopology<T> {
+    async fn get_public_domain(&self) -> Result<String, String> {
+        /*
+        let primary_domain = self
+            .primary
+            .get_public_domain()
+            .await
+            .map_err(|e| e.to_string())?;
+
+        Ok(primary_domain)
+        */
+        todo!()
+    }
+
     async fn get_internal_domain(&self) -> Result<Option<String>, String> {
         todo!()
     }

harmony/src/modules/node_health/mod.rs

@@ -1,5 +1,6 @@
 use async_trait::async_trait;
 use harmony_types::id::Id;
+use k8s_openapi::api::scheduling::v1::PriorityClass;
 use k8s_openapi::api::{
     apps::v1::{DaemonSet, DaemonSetSpec},
     core::v1::{
@@ -144,6 +145,19 @@ impl<T: Topology + K8sclient> Interpret<T> for NodeHealthInterpret {
             },
         };
 
+        // PriorityClass
+        let priority_class_name = "node-healthcheck-critical".to_string();
+        let priority_class = PriorityClass {
+            metadata: ObjectMeta {
+                name: Some(priority_class_name.clone()),
+                ..ObjectMeta::default()
+            },
+            value: 1000000000,
+            global_default: Some(false),
+            preemption_policy: Some("PreemptLowerPriority".to_string()),
+            description: Some("Highest priority for node health check daemonset - can preempt lower priority pods".to_string()),
+        };
+
         // DaemonSet
         let mut daemonset_labels = BTreeMap::new();
         daemonset_labels.insert("app".to_string(), "node-healthcheck".to_string());
@@ -168,6 +182,7 @@ impl<T: Topology + K8sclient> Interpret<T> for NodeHealthInterpret {
                     spec: Some(PodSpec {
                         service_account_name: Some(service_account_name.clone()),
                         host_network: Some(true),
+                        priority_class_name: Some(priority_class_name),
                         tolerations: Some(vec![Toleration {
                             operator: Some("Exists".to_string()),
                             ..Toleration::default()
@@ -182,6 +197,7 @@ impl<T: Topology + K8sclient> Interpret<T> for NodeHealthInterpret {
                                 name: "NODE_NAME".to_string(),
                                 value_from: Some(EnvVarSource {
                                     field_ref: Some(ObjectFieldSelector {
+                                        api_version: Some("v1".to_string()),
                                         field_path: "spec.nodeName".to_string(),
                                         ..ObjectFieldSelector::default()
                                     }),
@@ -233,6 +249,9 @@ impl<T: Topology + K8sclient> Interpret<T> for NodeHealthInterpret {
         K8sResourceScore::single(cluster_role_binding, None)
             .interpret(inventory, topology)
             .await?;
+        K8sResourceScore::single(priority_class, None)
+            .interpret(inventory, topology)
+            .await?;
         K8sResourceScore::single(daemon_set, Some(namespace_name.clone()))
             .interpret(inventory, topology)
             .await?;

harmony/src/modules/okd/bootstrap_load_balancer.rs

@@ -41,6 +41,7 @@ impl OKDBootstrapLoadBalancerScore {
                 backend_servers: Self::topology_to_backend_server(topology, 6443),
                 listening_port: SocketAddr::new(private_ip, 6443),
                 health_check: Some(HealthCheck::HTTP(
+                    None,
                     "/readyz".to_string(),
                     HttpMethod::GET,
                     HttpStatusCode::Success2xx,

harmony/src/modules/okd/load_balancer.rs

@@ -8,7 +8,7 @@ use crate::{
     score::Score,
     topology::{
         BackendServer, HAClusterTopology, HealthCheck, HttpMethod, HttpStatusCode, LoadBalancer,
-        LoadBalancerService, SSL, Topology,
+        LoadBalancerService, LogicalHost, Router, SSL, Topology,
     },
 };
 
@@ -23,32 +23,72 @@ pub struct OKDLoadBalancerScore {
     load_balancer_score: LoadBalancerScore,
 }
 
+/// OKD Load Balancer Score configuration
+///
+/// This module configures the load balancer for OKD (OpenShift Kubernetes Distribution)
+/// bare metal installations.
+///
+/// # Backend Server Configuration
+///
+/// For ports 80 and 443 (ingress traffic), the load balancer includes both control plane
+/// and worker nodes in the backend pool. This is consistent with OKD's requirement that
+/// ingress traffic should be load balanced across all nodes that may run ingress router pods.
+///
+/// For ports 22623 (Ignition API) and 6443 (Kubernetes API), only control plane nodes
+/// are included as backends, as these services are control plane specific.
+///
+/// # References
+///
+/// - [OKD Bare Metal Installation - External Load Balancer Configuration]
+///   (<https://docs.okd.io/latest/installing/installing_bare_metal/ipi/ipi-install-installation-workflow.html#nw-osp-configuring-external-load-balancer_ipi-install-installation-workflow>)
+///
+/// # Example
+///
+/// ```ignore
+/// use harmony::topology::HAClusterTopology;
+/// use harmony::modules::okd::OKDLoadBalancerScore;
+///
+/// let topology: HAClusterTopology = /* get topology from your infrastructure */;
+/// let score = OKDLoadBalancerScore::new(&topology);
+/// ```
 impl OKDLoadBalancerScore {
     pub fn new(topology: &HAClusterTopology) -> Self {
         let public_ip = topology.router.get_gateway();
         let public_services = vec![
             LoadBalancerService {
-                backend_servers: Self::control_plane_to_backend_server(topology, 80),
+                backend_servers: Self::nodes_to_backend_server(topology, 80),
                 listening_port: SocketAddr::new(public_ip, 80),
-                health_check: Some(HealthCheck::TCP(None)),
+                health_check: None,
             },
             LoadBalancerService {
-                backend_servers: Self::control_plane_to_backend_server(topology, 443),
+                backend_servers: Self::nodes_to_backend_server(topology, 443),
                 listening_port: SocketAddr::new(public_ip, 443),
-                health_check: Some(HealthCheck::TCP(None)),
+                health_check: None,
             },
         ];
 
         let private_services = vec![
             LoadBalancerService {
-                backend_servers: Self::control_plane_to_backend_server(topology, 80),
+                backend_servers: Self::nodes_to_backend_server(topology, 80),
                 listening_port: SocketAddr::new(public_ip, 80),
-                health_check: Some(HealthCheck::TCP(None)),
+                health_check: Some(HealthCheck::HTTP(
+                    Some(25001),
+                    "/health?check=okd_router_1936,node_ready".to_string(),
+                    HttpMethod::GET,
+                    HttpStatusCode::Success2xx,
+                    SSL::Default,
+                )),
             },
             LoadBalancerService {
-                backend_servers: Self::control_plane_to_backend_server(topology, 443),
+                backend_servers: Self::nodes_to_backend_server(topology, 443),
                 listening_port: SocketAddr::new(public_ip, 443),
-                health_check: Some(HealthCheck::TCP(None)),
+                health_check: Some(HealthCheck::HTTP(
+                    Some(25001),
+                    "/health?check=okd_router_1936,node_ready".to_string(),
+                    HttpMethod::GET,
+                    HttpStatusCode::Success2xx,
+                    SSL::Default,
+                )),
             },
             LoadBalancerService {
                 backend_servers: Self::control_plane_to_backend_server(topology, 22623),
@@ -59,6 +99,7 @@ impl OKDLoadBalancerScore {
                 backend_servers: Self::control_plane_to_backend_server(topology, 6443),
                 listening_port: SocketAddr::new(public_ip, 6443),
                 health_check: Some(HealthCheck::HTTP(
+                    None,
                     "/readyz".to_string(),
                     HttpMethod::GET,
                     HttpStatusCode::Success2xx,
@@ -74,6 +115,11 @@ impl OKDLoadBalancerScore {
         }
     }
 
+    /// Creates backend servers list for control plane nodes only
+    ///
+    /// Use this for control plane-specific services like:
+    /// - Port 22623: Ignition API (machine configuration during bootstrap)
+    /// - Port 6443: Kubernetes API server
     fn control_plane_to_backend_server(
         topology: &HAClusterTopology,
         port: u16,
@@ -87,6 +133,194 @@ impl OKDLoadBalancerScore {
             })
             .collect()
     }
+
+    /// Creates backend servers list for all nodes (control plane + workers)
+    ///
+    /// Use this for ingress traffic that should be distributed across all nodes:
+    /// - Port 80: HTTP ingress traffic
+    /// - Port 443: HTTPS ingress traffic
+    ///
+    /// In OKD, ingress router pods can run on any node, so both control plane
+    /// and worker nodes should be included in the load balancer backend pool.
+    fn nodes_to_backend_server(topology: &HAClusterTopology, port: u16) -> Vec<BackendServer> {
+        let mut nodes = Vec::new();
+        for cp in &topology.control_plane {
+            nodes.push(BackendServer {
+                address: cp.ip.to_string(),
+                port,
+            });
+        }
+        for worker in &topology.workers {
+            nodes.push(BackendServer {
+                address: worker.ip.to_string(),
+                port,
+            });
+        }
+        nodes
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use std::sync::{Arc, OnceLock};
+
+    use super::*;
+    use crate::topology::DummyInfra;
+    use harmony_macros::ip;
+    use harmony_types::net::IpAddress;
+
+    fn create_test_topology() -> HAClusterTopology {
+        let router = Arc::new(DummyRouter {
+            gateway: ip!("192.168.1.1"),
+        });
+
+        HAClusterTopology {
+            domain_name: "test.example.com".to_string(),
+            router,
+            load_balancer: Arc::new(DummyInfra),
+            firewall: Arc::new(DummyInfra),
+            dhcp_server: Arc::new(DummyInfra),
+            tftp_server: Arc::new(DummyInfra),
+            http_server: Arc::new(DummyInfra),
+            dns_server: Arc::new(DummyInfra),
+            node_exporter: Arc::new(DummyInfra),
+            switch_client: Arc::new(DummyInfra),
+            bootstrap_host: LogicalHost {
+                ip: ip!("192.168.1.100"),
+                name: "bootstrap".to_string(),
+            },
+            control_plane: vec![
+                LogicalHost {
+                    ip: ip!("192.168.1.10"),
+                    name: "control-plane-0".to_string(),
+                },
+                LogicalHost {
+                    ip: ip!("192.168.1.11"),
+                    name: "control-plane-1".to_string(),
+                },
+                LogicalHost {
+                    ip: ip!("192.168.1.12"),
+                    name: "control-plane-2".to_string(),
+                },
+            ],
+            workers: vec![
+                LogicalHost {
+                    ip: ip!("192.168.1.20"),
+                    name: "worker-0".to_string(),
+                },
+                LogicalHost {
+                    ip: ip!("192.168.1.21"),
+                    name: "worker-1".to_string(),
+                },
+            ],
+            kubeconfig: None,
+            network_manager: OnceLock::new(),
+        }
+    }
+
+    struct DummyRouter {
+        gateway: IpAddress,
+    }
+
+    impl Router for DummyRouter {
+        fn get_gateway(&self) -> IpAddress {
+            self.gateway
+        }
+        fn get_cidr(&self) -> cidr::Ipv4Cidr {
+            let ipv4 = match self.gateway {
+                IpAddress::V4(ip) => ip,
+                IpAddress::V6(_) => panic!("IPv6 not supported"),
+            };
+            cidr::Ipv4Cidr::new(ipv4, 24).unwrap()
+        }
+        fn get_host(&self) -> LogicalHost {
+            LogicalHost {
+                ip: self.gateway,
+                name: "router".to_string(),
+            }
+        }
+    }
+
+    #[test]
+    fn test_nodes_to_backend_server_includes_control_plane_and_workers() {
+        let topology = create_test_topology();
+
+        let backend_servers = OKDLoadBalancerScore::nodes_to_backend_server(&topology, 80);
+
+        assert_eq!(backend_servers.len(), 5);
+
+        let addresses: Vec<&str> = backend_servers.iter().map(|s| s.address.as_str()).collect();
+        assert!(addresses.contains(&"192.168.1.10"));
+        assert!(addresses.contains(&"192.168.1.11"));
+        assert!(addresses.contains(&"192.168.1.12"));
+        assert!(addresses.contains(&"192.168.1.20"));
+        assert!(addresses.contains(&"192.168.1.21"));
+    }
+
+    #[test]
+    fn test_control_plane_to_backend_server_only_includes_control_plane() {
+        let topology = create_test_topology();
+
+        let backend_servers = OKDLoadBalancerScore::control_plane_to_backend_server(&topology, 80);
+
+        assert_eq!(backend_servers.len(), 3);
+
+        let addresses: Vec<&str> = backend_servers.iter().map(|s| s.address.as_str()).collect();
+        assert!(addresses.contains(&"192.168.1.10"));
+        assert!(addresses.contains(&"192.168.1.11"));
+        assert!(addresses.contains(&"192.168.1.12"));
+        assert!(!addresses.contains(&"192.168.1.20"));
+        assert!(!addresses.contains(&"192.168.1.21"));
+    }
+
+    #[test]
+    fn test_public_services_include_all_nodes_on_port_80_and_443() {
+        let topology = create_test_topology();
+        let score = OKDLoadBalancerScore::new(&topology);
+
+        let public_service_80 = score
+            .load_balancer_score
+            .public_services
+            .iter()
+            .find(|s| s.listening_port.port() == 80)
+            .expect("Public service on port 80 not found");
+
+        let public_service_443 = score
+            .load_balancer_score
+            .public_services
+            .iter()
+            .find(|s| s.listening_port.port() == 443)
+            .expect("Public service on port 443 not found");
+
+        assert_eq!(public_service_80.backend_servers.len(), 5);
+        assert_eq!(public_service_443.backend_servers.len(), 5);
+    }
+
+    #[test]
+    fn test_private_service_port_22623_only_control_plane() {
+        let topology = create_test_topology();
+        let score = OKDLoadBalancerScore::new(&topology);
+
+        let private_service_22623 = score
+            .load_balancer_score
+            .private_services
+            .iter()
+            .find(|s| s.listening_port.port() == 22623)
+            .expect("Private service on port 22623 not found");
+
+        assert_eq!(private_service_22623.backend_servers.len(), 3);
+    }
+
+    #[test]
+    fn test_all_backend_servers_have_correct_port() {
+        let topology = create_test_topology();
+
+        let backend_servers = OKDLoadBalancerScore::nodes_to_backend_server(&topology, 443);
+
+        for server in backend_servers {
+            assert_eq!(server.port, 443);
+        }
+    }
 }
 
 impl<T: Topology + LoadBalancer> Score<T> for OKDLoadBalancerScore {

harmony/src/modules/postgresql/capability.rs

@@ -37,6 +37,7 @@ pub struct PostgreSQLConfig {
     /// settings incompatible with the default CNPG behavior.
     pub namespace: String,
 }
+
 impl PostgreSQLConfig {
     pub fn with_namespace(&self, namespace: &str) -> PostgreSQLConfig {
         let mut new = self.clone();

harmony/src/modules/postgresql/cnpg/crd.rs

@@ -1,3 +1,6 @@
+use std::collections::BTreeMap;
+use std::collections::HashMap;
+
 use kube::{CustomResource, api::ObjectMeta};
 use serde::{Deserialize, Serialize};
 
@@ -13,9 +16,18 @@ use serde::{Deserialize, Serialize};
 #[serde(rename_all = "camelCase")]
 pub struct ClusterSpec {
     pub instances: u32,
+    #[serde(skip_serializing_if = "Option::is_none")]
     pub image_name: Option<String>,
     pub storage: Storage,
     pub bootstrap: Bootstrap,
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub external_clusters: Option<Vec<ExternalCluster>>,
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub replica: Option<ReplicaSpec>,
+    /// This must be set to None if you want cnpg to generate a superuser secret
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub superuser_secret: Option<BTreeMap<String, String>>,
+    pub enable_superuser_access: bool,
 }
 
 impl Default for Cluster {
@@ -34,6 +46,10 @@ impl Default for ClusterSpec {
             image_name: None,
             storage: Storage::default(),
             bootstrap: Bootstrap::default(),
+            external_clusters: None,
+            replica: None,
+            superuser_secret: None,
+            enable_superuser_access: false,
         }
     }
 }
@@ -47,7 +63,13 @@ pub struct Storage {
 #[derive(Deserialize, Serialize, Clone, Debug, Default)]
 #[serde(rename_all = "camelCase")]
 pub struct Bootstrap {
-    pub initdb: Initdb,
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub initdb: Option<Initdb>,
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub recovery: Option<Recovery>,
+    #[serde(rename = "pg_basebackup")]
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub pg_basebackup: Option<PgBaseBackup>,
 }
 
 #[derive(Deserialize, Serialize, Clone, Debug, Default)]
@@ -56,3 +78,50 @@ pub struct Initdb {
     pub database: String,
     pub owner: String,
 }
+
+#[derive(Deserialize, Serialize, Clone, Debug)]
+#[serde(rename_all = "camelCase")]
+pub struct Recovery {
+    pub source: String,
+}
+
+#[derive(Deserialize, Serialize, Clone, Debug)]
+pub struct PgBaseBackup {
+    pub source: String,
+}
+
+#[derive(Deserialize, Serialize, Clone, Debug)]
+#[serde(rename_all = "camelCase")]
+pub struct ExternalCluster {
+    pub name: String,
+    pub connection_parameters: HashMap<String, String>,
+    pub ssl_key: Option<SecretKeySelector>,
+    pub ssl_cert: Option<SecretKeySelector>,
+    pub ssl_root_cert: Option<SecretKeySelector>,
+}
+
+#[derive(Deserialize, Serialize, Clone, Debug)]
+#[serde(rename_all = "camelCase")]
+pub struct ConnectionParameters {
+    pub host: String,
+    pub user: String,
+    pub dbname: String,
+    pub sslmode: String,
+    pub sslnegotiation: String,
+}
+
+#[derive(Deserialize, Serialize, Clone, Debug)]
+#[serde(rename_all = "camelCase")]
+pub struct ReplicaSpec {
+    pub enabled: bool,
+    pub source: String,
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub primary: Option<String>,
+}
+
+#[derive(Deserialize, Serialize, Clone, Debug)]
+#[serde(rename_all = "camelCase")]
+pub struct SecretKeySelector {
+    pub name: String,
+    pub key: String,
+}

harmony/src/modules/postgresql/failover.rs

@@ -3,6 +3,8 @@ use log::debug;
 use log::info;
 use std::collections::HashMap;
 
+use crate::interpret::InterpretError;
+use crate::topology::TlsRoute;
 use crate::topology::TlsRouter;
 use crate::{
     modules::postgresql::capability::{
@@ -49,8 +51,18 @@ impl<T: PostgreSQL + TlsRouter> PostgreSQL for FailoverTopology<T> {
         // TODO we should be getting the public endpoint for a service by calling a method on
         // TlsRouter capability.
         // Something along the lines of `TlsRouter::get_hostname_for_service(...).await?;`
+        let host = format!(
+            "{}.{}.{}",
+            config.cluster_name,
+            config.namespace,
+            self.primary
+                .get_public_domain()
+                .await
+                .expect("failed to retrieve public domain")
+                .to_string()
+        );
         let endpoint = PostgreSQLEndpoint {
-            host: "postgrestest.sto1.nationtech.io".to_string(),
+            host,
             port: self.primary.get_router_port().await,
         };
 
@@ -59,6 +71,46 @@ impl<T: PostgreSQL + TlsRouter> PostgreSQL for FailoverTopology<T> {
             endpoint.host, endpoint.port
         );
 
+        info!("installing primary postgres route");
+        let prim_hostname = format!(
+            "{}.{}.{}",
+            config.cluster_name,
+            config.namespace,
+            self.primary.get_public_domain().await?
+        );
+        let rw_backend = format!("{}-rw", config.cluster_name);
+        let tls_route = TlsRoute {
+            hostname: prim_hostname,
+            backend: rw_backend,
+            target_port: 5432,
+            namespace: config.namespace.clone(),
+        };
+        // Expose RW publicly via TLS passthrough
+        self.primary
+            .install_route(tls_route.clone())
+            .await
+            .map_err(|e| InterpretError::new(e))?;
+
+        info!("installing replica postgres route");
+        let rep_hostname = format!(
+            "{}.{}.{}",
+            config.cluster_name,
+            config.namespace,
+            self.replica.get_public_domain().await?
+        );
+        let rw_backend = format!("{}-rw", config.cluster_name);
+        let tls_route = TlsRoute {
+            hostname: rep_hostname,
+            backend: rw_backend,
+            target_port: 5432,
+            namespace: config.namespace.clone(),
+        };
+
+        // Expose RW publicly via TLS passthrough
+        self.replica
+            .install_route(tls_route.clone())
+            .await
+            .map_err(|e| InterpretError::new(e))?;
         info!("Configuring replica connection parameters and bootstrap");
 
         let mut connection_parameters = HashMap::new();

harmony/src/modules/postgresql/operator.rs

@@ -20,7 +20,7 @@ use crate::topology::{K8sclient, Topology};
 /// # Usage
 /// ```
 /// use harmony::modules::postgresql::CloudNativePgOperatorScore;
-/// let score = CloudNativePgOperatorScore::default();
+/// let score = CloudNativePgOperatorScore::default_openshift();
 /// ```
 ///
 /// Or, you can take control of most relevant fiedls this way :
@@ -52,8 +52,8 @@ pub struct CloudNativePgOperatorScore {
     pub source_namespace: String,
 }
 
-impl Default for CloudNativePgOperatorScore {
-    fn default() -> Self {
+impl CloudNativePgOperatorScore {
+    pub fn default_openshift() -> Self {
         Self {
             namespace: "openshift-operators".to_string(),
             channel: "stable-v1".to_string(),
@@ -68,7 +68,7 @@ impl CloudNativePgOperatorScore {
     pub fn new(namespace: &str) -> Self {
         Self {
             namespace: namespace.to_string(),
-            ..Default::default()
+            ..Self::default_openshift()
         }
     }
 }

harmony/src/modules/postgresql/score_k8s.rs

@@ -1,12 +1,21 @@
-use serde::Serialize;
+use crate::data::Version;
+use crate::interpret::{Interpret, InterpretError, InterpretName, InterpretStatus, Outcome};
+use crate::inventory::Inventory;
 
-use crate::interpret::Interpret;
 use crate::modules::k8s::resource::K8sResourceScore;
 use crate::modules::postgresql::capability::PostgreSQLConfig;
-use crate::modules::postgresql::cnpg::{Bootstrap, Cluster, ClusterSpec, Initdb, Storage};
+use crate::modules::postgresql::cnpg::{
+    Bootstrap, Cluster, ClusterSpec, ExternalCluster, Initdb, PgBaseBackup, ReplicaSpec,
+    SecretKeySelector, Storage,
+};
 use crate::score::Score;
 use crate::topology::{K8sclient, Topology};
+use async_trait::async_trait;
+use harmony_types::id::Id;
+use k8s_openapi::ByteString;
+use k8s_openapi::api::core::v1::Secret;
 use k8s_openapi::apimachinery::pkg::apis::meta::v1::ObjectMeta;
+use serde::Serialize;
 
 /// Deploys an opinionated, highly available PostgreSQL cluster managed by CNPG.
 ///
@@ -49,6 +58,30 @@ impl K8sPostgreSQLScore {
 
 impl<T: Topology + K8sclient> Score<T> for K8sPostgreSQLScore {
     fn create_interpret(&self) -> Box<dyn Interpret<T>> {
+        Box::new(K8sPostgreSQLInterpret {
+            config: self.config.clone(),
+        })
+    }
+
+    fn name(&self) -> String {
+        format!("PostgreSQLScore({})", self.config.namespace)
+    }
+}
+
+#[derive(Debug)]
+pub struct K8sPostgreSQLInterpret {
+    config: PostgreSQLConfig,
+}
+
+#[async_trait]
+impl<T: Topology + K8sclient> Interpret<T> for K8sPostgreSQLInterpret {
+    async fn execute(
+        &self,
+        inventory: &Inventory,
+        topology: &T,
+    ) -> Result<Outcome, InterpretError> {
+        match &self.config.role {
+            super::capability::PostgreSQLClusterRole::Primary => {
                 let metadata = ObjectMeta {
                     name: Some(self.config.cluster_name.clone()),
                     namespace: Some(self.config.namespace.clone()),
@@ -61,20 +94,148 @@ impl<T: Topology + K8sclient> Score<T> for K8sPostgreSQLScore {
                         size: self.config.storage_size.to_string(),
                     },
                     bootstrap: Bootstrap {
-                initdb: Initdb {
+                        initdb: Some(Initdb {
                             database: "app".to_string(),
                             owner: "app".to_string(),
+                        }),
+                        recovery: None,
+                        pg_basebackup: None,
                     },
+                    enable_superuser_access: true,
+                    ..ClusterSpec::default()
+                };
+                let cluster = Cluster { metadata, spec };
+
+                Ok(
+                    K8sResourceScore::single(cluster, Some(self.config.namespace.clone()))
+                        .create_interpret()
+                        .execute(inventory, topology)
+                        .await?,
+                )
+            }
+            super::capability::PostgreSQLClusterRole::Replica(replica_config) => {
+                let metadata = ObjectMeta {
+                    name: Some("streaming-replica-certs".to_string()),
+                    namespace: Some(self.config.namespace.clone()),
+                    ..ObjectMeta::default()
+                };
+
+                // The data must be base64-encoded. If you already have PEM strings in your config, encode them:
+                let mut data = std::collections::BTreeMap::new();
+                data.insert(
+                    "tls.key".to_string(),
+                    ByteString(
+                        replica_config
+                            .replication_certs
+                            .streaming_replica_key_pem
+                            .as_bytes()
+                            .to_vec(),
+                    ),
+                );
+                data.insert(
+                    "tls.crt".to_string(),
+                    ByteString(
+                        replica_config
+                            .replication_certs
+                            .streaming_replica_cert_pem
+                            .as_bytes()
+                            .to_vec(),
+                    ),
+                );
+                data.insert(
+                    "ca.crt".to_string(),
+                    ByteString(
+                        replica_config
+                            .replication_certs
+                            .ca_cert_pem
+                            .as_bytes()
+                            .to_vec(),
+                    ),
+                );
+
+                let secret = Secret {
+                    metadata,
+                    data: Some(data),
+                    string_data: None, // You could use string_data if you prefer raw strings
+                    type_: Some("Opaque".to_string()),
+                    ..Secret::default()
+                };
+
+                K8sResourceScore::single(secret, Some(self.config.namespace.clone()))
+                    .create_interpret()
+                    .execute(inventory, topology)
+                    .await?;
+
+                let metadata = ObjectMeta {
+                    name: Some(self.config.cluster_name.clone()),
+                    namespace: Some(self.config.namespace.clone()),
+                    ..ObjectMeta::default()
+                };
+
+                let spec = ClusterSpec {
+                    instances: self.config.instances,
+                    storage: Storage {
+                        size: self.config.storage_size.to_string(),
                     },
+                    bootstrap: Bootstrap {
+                        initdb: None,
+                        recovery: None,
+                        pg_basebackup: Some(PgBaseBackup {
+                            source: replica_config.primary_cluster_name.clone(),
+                        }),
+                    },
+                    external_clusters: Some(vec![ExternalCluster {
+                        name: replica_config.primary_cluster_name.clone(),
+                        connection_parameters: replica_config
+                            .external_cluster
+                            .connection_parameters
+                            .clone(),
+                        ssl_key: Some(SecretKeySelector {
+                            name: "streaming-replica-certs".to_string(),
+                            key: "tls.key".to_string(),
+                        }),
+                        ssl_cert: Some(SecretKeySelector {
+                            name: "streaming-replica-certs".to_string(),
+                            key: "tls.crt".to_string(),
+                        }),
+                        ssl_root_cert: Some(SecretKeySelector {
+                            name: "streaming-replica-certs".to_string(),
+                            key: "ca.crt".to_string(),
+                        }),
+                    }]),
+                    replica: Some(ReplicaSpec {
+                        enabled: true,
+                        source: replica_config.primary_cluster_name.clone(),
+                        primary: None,
+                    }),
                     ..ClusterSpec::default()
                 };
 
                 let cluster = Cluster { metadata, spec };
 
-        K8sResourceScore::single(cluster, Some(self.config.namespace.clone())).create_interpret()
+                Ok(
+                    K8sResourceScore::single(cluster, Some(self.config.namespace.clone()))
+                        .create_interpret()
+                        .execute(inventory, topology)
+                        .await?,
+                )
+            }
+        }
     }
 
-    fn name(&self) -> String {
-        format!("PostgreSQLScore({})", self.config.namespace)
+    fn get_name(&self) -> InterpretName {
+        InterpretName::Custom("K8sPostgreSQLInterpret")
+    }
+
+    fn get_version(&self) -> Version {
+        todo!()
+    }
+
+    fn get_status(&self) -> InterpretStatus {
+        todo!()
+    }
+
+    fn get_children(&self) -> Vec<Id> {
+        todo!()
     }
 }

harmony/src/modules/postgresql/score_public.rs

@@ -18,46 +18,31 @@ use crate::topology::Topology;
 /// # Usage
 /// ```
 /// use harmony::modules::postgresql::PublicPostgreSQLScore;
-/// let score = PublicPostgreSQLScore::new("harmony", "pg-rw.example.com");
+/// let score = PublicPostgreSQLScore::new("harmony");
 /// ```
 #[derive(Debug, Clone, Serialize)]
 pub struct PublicPostgreSQLScore {
     /// Inner non-public Postgres cluster config.
     pub config: PostgreSQLConfig,
-    /// Public hostname for RW TLS passthrough (port 443 → cluster-rw:5432).
-    pub hostname: String,
 }
 
 impl PublicPostgreSQLScore {
-    pub fn new(namespace: &str, hostname: &str) -> Self {
+    pub fn new(namespace: &str) -> Self {
         Self {
             config: PostgreSQLConfig::default().with_namespace(namespace),
-            hostname: hostname.to_string(),
         }
     }
 }
 
 impl<T: Topology + PostgreSQL + TlsRouter> Score<T> for PublicPostgreSQLScore {
     fn create_interpret(&self) -> Box<dyn Interpret<T>> {
-        let rw_backend = format!("{}-rw", self.config.cluster_name);
-        let tls_route = TlsRoute {
-            namespace: self.config.namespace.clone(),
-            hostname: self.hostname.clone(),
-            backend: rw_backend,
-            target_port: 5432,
-        };
-
         Box::new(PublicPostgreSQLInterpret {
             config: self.config.clone(),
-            tls_route,
         })
     }
 
     fn name(&self) -> String {
-        format!(
-            "PublicPostgreSQLScore({}:{})",
-            self.config.namespace, self.hostname
-        )
+        format!("PublicPostgreSQLScore({})", self.config.namespace)
     }
 }
 
@@ -65,7 +50,6 @@ impl<T: Topology + PostgreSQL + TlsRouter> Score<T> for PublicPostgreSQLScore {
 #[derive(Debug, Clone)]
 struct PublicPostgreSQLInterpret {
     config: PostgreSQLConfig,
-    tls_route: TlsRoute,
 }
 
 #[async_trait]
@@ -76,15 +60,28 @@ impl<T: Topology + PostgreSQL + TlsRouter> Interpret<T> for PublicPostgreSQLInte
             .await
             .map_err(|e| InterpretError::new(e))?;
 
+        let hostname = format!(
+            "{}.{}.{}",
+            self.config.cluster_name,
+            self.config.namespace,
+            topo.get_public_domain().await?
+        );
+        let rw_backend = format!("{}-rw", self.config.cluster_name);
+        let tls_route = TlsRoute {
+            hostname,
+            backend: rw_backend,
+            target_port: 5432,
+            namespace: self.config.namespace.clone(),
+        };
         // Expose RW publicly via TLS passthrough
-        topo.install_route(self.tls_route.clone())
+        topo.install_route(tls_route.clone())
             .await
             .map_err(|e| InterpretError::new(e))?;
 
         Ok(Outcome::success(format!(
             "Public CNPG cluster '{}' deployed with TLS passthrough route '{}'",
             self.config.cluster_name.clone(),
-            self.tls_route.hostname
+            tls_route.hostname
         )))
     }
 

harmony/src/modules/prometheus/k8s_prometheus_alerting_score.rs

@@ -12,8 +12,7 @@ use crate::modules::monitoring::kube_prometheus::crd::crd_alertmanager_config::C
 use crate::modules::monitoring::kube_prometheus::crd::crd_default_rules::build_default_application_rules;
 use crate::modules::monitoring::kube_prometheus::crd::crd_grafana::{
     Grafana, GrafanaDashboard, GrafanaDashboardSpec, GrafanaDatasource, GrafanaDatasourceConfig,
-    GrafanaDatasourceJsonData, GrafanaDatasourceSpec, GrafanaSecretKeyRef, GrafanaSpec,
-    GrafanaValueFrom, GrafanaValueSource,
+    GrafanaDatasourceJsonData, GrafanaDatasourceSpec, GrafanaSpec,
 };
 use crate::modules::monitoring::kube_prometheus::crd::crd_prometheus_rules::{
     PrometheusRule, PrometheusRuleSpec, RuleGroup,
@@ -23,7 +22,7 @@ use crate::modules::monitoring::kube_prometheus::crd::service_monitor::{
     ServiceMonitor, ServiceMonitorSpec,
 };
 use crate::topology::oberservability::monitoring::AlertReceiver;
-use crate::topology::{K8sclient, Topology, k8s::K8sClient};
+use crate::topology::{K8sclient, Topology};
 use crate::{
     data::Version,
     interpret::{Interpret, InterpretError, InterpretName, InterpretStatus, Outcome},
@@ -38,6 +37,7 @@ use crate::{
     },
     score::Score,
 };
+use harmony_k8s::K8sClient;
 use harmony_types::id::Id;
 
 use super::prometheus::PrometheusMonitoring;

harmony/src/modules/prometheus/rhob_alerting_score.rs

@@ -30,12 +30,13 @@ use crate::modules::monitoring::kube_prometheus::crd::rhob_service_monitor::{
 use crate::score::Score;
 use crate::topology::ingress::Ingress;
 use crate::topology::oberservability::monitoring::AlertReceiver;
-use crate::topology::{K8sclient, Topology, k8s::K8sClient};
+use crate::topology::{K8sclient, Topology};
 use crate::{
     data::Version,
     interpret::{Interpret, InterpretError, InterpretName, InterpretStatus, Outcome},
     inventory::Inventory,
 };
+use harmony_k8s::K8sClient;
 use harmony_types::id::Id;
 
 use super::prometheus::PrometheusMonitoring;

harmony/src/modules/storage/ceph/ceph_remove_osd_score.rs

@@ -4,6 +4,7 @@ use std::{
 };
 
 use async_trait::async_trait;
+use harmony_k8s::K8sClient;
 use log::{debug, warn};
 use serde::{Deserialize, Serialize};
 use tokio::time::sleep;
@@ -13,7 +14,7 @@ use crate::{
     interpret::{Interpret, InterpretError, InterpretName, InterpretStatus, Outcome},
     inventory::Inventory,
     score::Score,
-    topology::{K8sclient, Topology, k8s::K8sClient},
+    topology::{K8sclient, Topology},
 };
 use harmony_types::id::Id;
 

harmony/src/modules/storage/ceph/ceph_validate_health_score.rs

@@ -9,8 +9,9 @@ use crate::{
     interpret::{Interpret, InterpretError, InterpretName, InterpretStatus, Outcome},
     inventory::Inventory,
     score::Score,
-    topology::{K8sclient, Topology, k8s::K8sClient},
+    topology::{K8sclient, Topology},
 };
+use harmony_k8s::K8sClient;
 use harmony_types::id::Id;
 
 #[derive(Clone, Debug, Serialize)]

harmony/src/modules/zitadel/mod.rs

@@ -1,43 +1,476 @@
+use k8s_openapi::api::core::v1::Namespace;
+use k8s_openapi::apimachinery::pkg::apis::meta::v1::ObjectMeta;
+use k8s_openapi::{ByteString, api::core::v1::Secret};
+use kube::{Error as KubeError, core::ErrorResponse};
+use rand::distr::Distribution;
+use rand::{Rng, rng, seq::SliceRandom};
+use std::collections::BTreeMap;
 use std::str::FromStr;
 
+use async_trait::async_trait;
 use harmony_macros::hurl;
+use harmony_types::id::Id;
+use harmony_types::storage::StorageSize;
+use log::{debug, error, info, trace, warn};
 use non_blank_string_rs::NonBlankString;
 use serde::Serialize;
 
 use crate::{
-    interpret::Interpret,
+    data::Version,
+    interpret::{Interpret, InterpretError, InterpretName, InterpretStatus, Outcome},
+    inventory::Inventory,
     modules::helm::chart::{HelmChartScore, HelmRepository},
+    modules::k8s::resource::K8sResourceScore,
+    modules::postgresql::capability::{PostgreSQL, PostgreSQLClusterRole, PostgreSQLConfig},
     score::Score,
     topology::{HelmCommand, K8sclient, Topology},
 };
 
+const NAMESPACE: &str = "zitadel";
+const PG_CLUSTER_NAME: &str = "zitadel-pg";
+const MASTERKEY_SECRET_NAME: &str = "zitadel-masterkey";
+
+/// Opinionated Zitadel deployment score.
+///
+/// Deploys a PostgreSQL cluster (via the [`PostgreSQL`] trait) and the Zitadel
+/// Helm chart into the same namespace. Intended as a central multi-tenant IdP
+/// with SSO for OKD/OpenShift, OpenBao, Harbor, Grafana, Nextcloud, Ente
+/// Photos, and others.
+///
+/// # Ingress annotations
+/// No controller-specific ingress annotations are set by default. On
+/// OKD/OpenShift, the ingress should request TLS so the generated Route is
+/// edge-terminated instead of HTTP-only. Optional cert-manager annotations are
+/// included for clusters that have cert-manager installed; clusters without
+/// cert-manager will ignore them.
+/// Add or adjust annotations via `values_overrides` depending on your
+/// distribution:
+/// - NGINX: `nginx.ingress.kubernetes.io/backend-protocol: GRPC`
+/// - OpenShift HAProxy: `route.openshift.io/termination: edge`
+/// - AWS ALB: set `ingress.controller: aws`
+
+///
+/// # Database credentials
+/// CNPG creates a `<cluster>-superuser` secret with key `password`. Because
+/// `envVarsSecret` injects secret keys verbatim as env var names and the CNPG
+/// key (`password`) does not match ZITADEL's expected name
+/// (`ZITADEL_DATABASE_POSTGRES_USER_PASSWORD`), individual `env` entries with
+/// `valueFrom.secretKeyRef` are used instead. For environments with an
+/// External Secrets Operator or similar, create a dedicated secret with the
+/// correct ZITADEL env var names and switch to `envVarsSecret`.
 #[derive(Debug, Serialize, Clone)]
 pub struct ZitadelScore {
-    /// Host used for external access (ingress)
+    /// External domain (e.g. `"auth.example.com"`).
     pub host: String,
+    pub zitadel_version: String,
 }
 
-impl<T: Topology + K8sclient + HelmCommand> Score<T> for ZitadelScore {
+impl<T: Topology + K8sclient + HelmCommand + PostgreSQL> Score<T> for ZitadelScore {
     fn name(&self) -> String {
         "ZitadelScore".to_string()
     }
 
     #[doc(hidden)]
     fn create_interpret(&self) -> Box<dyn Interpret<T>> {
-        // TODO exec pod commands to initialize secret store if not already done
+        Box::new(ZitadelInterpret {
+            host: self.host.clone(),
+            zitadel_version: self.zitadel_version.clone(),
+        })
+    }
+}
+
+// ---------------------------------------------------------------------------
+
+#[derive(Debug, Clone)]
+struct ZitadelInterpret {
+    host: String,
+    zitadel_version: String,
+}
+
+#[async_trait]
+impl<T: Topology + K8sclient + HelmCommand + PostgreSQL> Interpret<T> for ZitadelInterpret {
+    async fn execute(
+        &self,
+        inventory: &Inventory,
+        topology: &T,
+    ) -> Result<Outcome, InterpretError> {
+        info!(
+            "[Zitadel] Starting full deployment — namespace: '{NAMESPACE}', host: '{}'",
+            self.host
+        );
+
+        info!("Creating namespace {NAMESPACE} if it does not exist");
+        K8sResourceScore::single(
+            Namespace {
+                metadata: ObjectMeta {
+                    name: Some(NAMESPACE.to_string()),
+                    ..Default::default()
+                },
+                ..Default::default()
+            },
+            None,
+        )
+        .interpret(inventory, topology)
+        .await?;
+
+        // --- Step 1: PostgreSQL -------------------------------------------
+
+        let pg_config = PostgreSQLConfig {
+            cluster_name: PG_CLUSTER_NAME.to_string(),
+            instances: 2,
+            storage_size: StorageSize::gi(10),
+            role: PostgreSQLClusterRole::Primary,
+            namespace: NAMESPACE.to_string(),
+        };
+
+        debug!(
+            "[Zitadel] Deploying PostgreSQL cluster '{}' — instances: {}, storage: 10Gi, namespace: '{}'",
+            pg_config.cluster_name, pg_config.instances, pg_config.namespace
+        );
+
+        topology.deploy(&pg_config).await.map_err(|e| {
+            let msg = format!(
+                "[Zitadel] PostgreSQL deployment failed for '{}': {e}",
+                pg_config.cluster_name
+            );
+            error!("{msg}");
+            InterpretError::new(msg)
+        })?;
+
+        info!(
+            "[Zitadel] PostgreSQL cluster '{}' deployed",
+            pg_config.cluster_name
+        );
+
+        // --- Step 2: Resolve internal DB endpoint -------------------------
+
+        debug!(
+            "[Zitadel] Resolving internal endpoint for cluster '{}'",
+            pg_config.cluster_name
+        );
+
+        let endpoint = topology.get_endpoint(&pg_config).await.map_err(|e| {
+            let msg = format!(
+                "[Zitadel] Failed to resolve endpoint for cluster '{}': {e}",
+                pg_config.cluster_name
+            );
+            error!("{msg}");
+            InterpretError::new(msg)
+        })?;
+
+        info!(
+            "[Zitadel] DB endpoint resolved — host: '{}', port: {}",
+            endpoint.host, endpoint.port
+        );
+
+        // The CNPG-managed superuser secret contains 'password', 'username',
+        // 'host', 'port', 'dbname', 'uri'. We reference 'password' directly
+        // via env.valueFrom.secretKeyRef because CNPG's key names do not
+        // match ZITADEL's required env var names.
+        let pg_user_secret = format!("{PG_CLUSTER_NAME}-app");
+        let pg_superuser_secret = format!("{PG_CLUSTER_NAME}-superuser");
+        let db_host = &endpoint.host;
+        let db_port = endpoint.port;
         let host = &self.host;
 
-        let values_yaml = Some(format!(r#""#));
+        debug!("[Zitadel] DB credentials source — secret: '{pg_user_secret}', key: 'password'");
+        debug!(
+            "[Zitadel] DB credentials source — superuser secret: '{pg_superuser_secret}', key: 'password'"
+        );
 
-        todo!("This is not complete yet");
+        // Zitadel requires one symbol, one number and more. So let's force it.
+        fn generate_secure_password(length: usize) -> String {
+            const ALPHA_UPPER: &[u8] = b"ABCDEFGHIJKLMNOPQRSTUVWXYZ";
+            const ALPHA_LOWER: &[u8] = b"abcdefghijklmnopqrstuvwxyz";
+            const DIGITS: &[u8] = b"0123456789";
+            const SYMBOLS: &[u8] = b"!@#$%^&*()_+-=[]{}|;:',.<>?/";
 
-        HelmChartScore {
-            namespace: Some(NonBlankString::from_str("zitadel").unwrap()),
+            let mut rng = rand::rng();
+            let uniform_alpha_upper = rand::distr::Uniform::new(0, ALPHA_UPPER.len())
+                .expect("Failed to create distribution");
+            let uniform_alpha_lower = rand::distr::Uniform::new(0, ALPHA_LOWER.len())
+                .expect("Failed to create distribution");
+            let uniform_digits =
+                rand::distr::Uniform::new(0, DIGITS.len()).expect("Failed to create distribution");
+            let uniform_symbols =
+                rand::distr::Uniform::new(0, SYMBOLS.len()).expect("Failed to create distribution");
+
+            let mut chars: Vec<char> = Vec::with_capacity(length);
+
+            // Ensure at least one of each: upper, lower, digit, symbol
+            chars.push(ALPHA_UPPER[uniform_alpha_upper.sample(&mut rng)] as char);
+            chars.push(ALPHA_LOWER[uniform_alpha_lower.sample(&mut rng)] as char);
+            chars.push(DIGITS[uniform_digits.sample(&mut rng)] as char);
+            chars.push(SYMBOLS[uniform_symbols.sample(&mut rng)] as char);
+
+            // Fill remaining with random from all categories
+            let all_chars: Vec<u8> = [ALPHA_UPPER, ALPHA_LOWER, DIGITS, SYMBOLS].concat();
+
+            let uniform_all = rand::distr::Uniform::new(0, all_chars.len())
+                .expect("Failed to create distribution");
+
+            for _ in 0..(length - 4) {
+                chars.push(all_chars[uniform_all.sample(&mut rng)] as char);
+            }
+
+            // Shuffle
+            let mut shuffled = chars;
+            shuffled.shuffle(&mut rng);
+
+            return shuffled.iter().collect();
+        }
+
+        let admin_password = generate_secure_password(16);
+
+        // --- Step 3: Create masterkey secret ------------------------------------
+
+        debug!(
+            "[Zitadel] Creating masterkey secret '{}' in namespace '{}'",
+            MASTERKEY_SECRET_NAME, NAMESPACE
+        );
+
+        // Masterkey for symmetric encryption — must be exactly 32 ASCII bytes (alphanumeric only).
+        let masterkey = rng()
+            .sample_iter(&rand::distr::Alphanumeric)
+            .take(32)
+            .map(char::from)
+            .collect::<String>();
+
+        debug!(
+            "[Zitadel] Created masterkey secret '{}' in namespace '{}'",
+            MASTERKEY_SECRET_NAME, NAMESPACE
+        );
+
+        let mut masterkey_data: BTreeMap<String, ByteString> = BTreeMap::new();
+        masterkey_data.insert("masterkey".to_string(), ByteString(masterkey.into()));
+
+        let masterkey_secret = Secret {
+            metadata: ObjectMeta {
+                name: Some(MASTERKEY_SECRET_NAME.to_string()),
+                namespace: Some(NAMESPACE.to_string()),
+                ..ObjectMeta::default()
+            },
+            data: Some(masterkey_data),
+            ..Secret::default()
+        };
+
+        match topology
+            .k8s_client()
+            .await
+            .map_err(|e| InterpretError::new(format!("Failed to get k8s client : {e}")))?
+            .create(&masterkey_secret, Some(NAMESPACE))
+            .await
+        {
+            Ok(_) => {
+                info!(
+                    "[Zitadel] Masterkey secret '{}' created",
+                    MASTERKEY_SECRET_NAME
+                );
+            }
+            Err(KubeError::Api(ErrorResponse { code: 409, .. })) => {
+                info!(
+                    "[Zitadel] Masterkey secret '{}' already exists, leaving it untouched",
+                    MASTERKEY_SECRET_NAME
+                );
+            }
+            Err(other) => {
+                let msg = format!(
+                    "[Zitadel] Failed to create masterkey secret '{}': {other}",
+                    MASTERKEY_SECRET_NAME
+                );
+                error!("{msg}");
+                return Err(InterpretError::new(msg));
+            }
+        };
+
+        debug!(
+            "[Zitadel] Masterkey secret '{}' created successfully",
+            MASTERKEY_SECRET_NAME
+        );
+
+        // --- Step 4: Build Helm values ------------------------------------
+
+        warn!(
+            "[Zitadel] Applying TLS-enabled ingress defaults for OKD/OpenShift. \
+             cert-manager annotations are included as optional hints and are \
+             ignored on clusters without cert-manager."
+        );
+
+        let values_yaml = format!(
+            r#"image:
+  tag: {zitadel_version}
+zitadel:
+  masterkeySecretName: "{MASTERKEY_SECRET_NAME}"
+  configmapConfig:
+    ExternalDomain: "{host}"
+    ExternalSecure: true
+    FirstInstance:
+      Org:
+        Human:
+          UserName: "admin"
+          Password: "{admin_password}"
+          FirstName: "Zitadel"
+          LastName: "Admin"
+          Email: "admin@zitadel.example.com"
+          PasswordChangeRequired: true
+    TLS:
+      Enabled: false
+    Database:
+      Postgres:
+        Host: "{db_host}"
+        Port: {db_port}
+        Database: zitadel
+        MaxOpenConns: 20
+        MaxIdleConns: 10
+        User:
+          Username: postgres
+          SSL:
+            Mode: require
+        Admin:
+          Username: postgres
+          SSL:
+            Mode: require
+# Directly import credentials from the postgres secret
+# TODO : use a less privileged postgres user
+env:
+  - name: ZITADEL_DATABASE_POSTGRES_USER_USERNAME
+    valueFrom:
+      secretKeyRef:
+        name: "{pg_superuser_secret}"
+        key: user
+  - name: ZITADEL_DATABASE_POSTGRES_USER_PASSWORD
+    valueFrom:
+      secretKeyRef:
+        name: "{pg_superuser_secret}"
+        key: password
+  - name: ZITADEL_DATABASE_POSTGRES_ADMIN_USERNAME
+    valueFrom:
+      secretKeyRef:
+        name: "{pg_superuser_secret}"
+        key: user
+  - name: ZITADEL_DATABASE_POSTGRES_ADMIN_PASSWORD
+    valueFrom:
+      secretKeyRef:
+        name: "{pg_superuser_secret}"
+        key: password
+# Security context for OpenShift restricted PSA compliance
+podSecurityContext:
+  runAsNonRoot: true
+  runAsUser: null
+  fsGroup: null
+  seccompProfile:
+    type: RuntimeDefault
+securityContext:
+  allowPrivilegeEscalation: false
+  capabilities:
+    drop:
+      - ALL
+  runAsNonRoot: true
+  runAsUser: null
+  fsGroup: null
+  seccompProfile:
+    type: RuntimeDefault
+# Init job security context (runs before main deployment)
+initJob:
+  podSecurityContext:
+    runAsNonRoot: true
+    runAsUser: null
+    fsGroup: null
+    seccompProfile:
+      type: RuntimeDefault
+  securityContext:
+    allowPrivilegeEscalation: false
+    capabilities:
+      drop:
+        - ALL
+    runAsNonRoot: true
+    runAsUser: null
+    fsGroup: null
+    seccompProfile:
+      type: RuntimeDefault
+# Setup job security context
+setupJob:
+  podSecurityContext:
+    runAsNonRoot: true
+    runAsUser: null
+    fsGroup: null
+    seccompProfile:
+      type: RuntimeDefault
+  securityContext:
+    allowPrivilegeEscalation: false
+    capabilities:
+      drop:
+        - ALL
+    runAsNonRoot: true
+    runAsUser: null
+    fsGroup: null
+    seccompProfile:
+      type: RuntimeDefault
+ingress:
+  enabled: true
+  annotations:
+    cert-manager.io/cluster-issuer: letsencrypt-prod
+    route.openshift.io/termination: edge
+  hosts:
+    - host: "{host}"
+      paths:
+        - path: /
+          pathType: Prefix
+  tls:
+    - hosts:
+        - "{host}"
+      secretName: "{host}-tls"
+
+login:
+  enabled: true
+  podSecurityContext:
+    runAsNonRoot: true
+    runAsUser: null
+    fsGroup: null
+    seccompProfile:
+      type: RuntimeDefault
+  securityContext:
+    allowPrivilegeEscalation: false
+    capabilities:
+      drop:
+        - ALL
+    runAsNonRoot: true
+    runAsUser: null
+    fsGroup: null
+    seccompProfile:
+      type: RuntimeDefault
+  ingress:
+    enabled: true
+    annotations:
+      cert-manager.io/cluster-issuer: letsencrypt-prod
+      route.openshift.io/termination: edge
+    hosts:
+      - host: "{host}"
+        paths:
+          - path: /ui/v2/login
+            pathType: Prefix
+    tls:
+      - hosts:
+          - "{host}"
+        secretName: "{host}-tls""#,
+            zitadel_version = self.zitadel_version
+        );
+
+        trace!("[Zitadel] Helm values YAML:\n{values_yaml}");
+
+        // --- Step 5: Deploy Helm chart ------------------------------------
+
+        info!(
+            "[Zitadel] Deploying Helm chart 'zitadel/zitadel' as release 'zitadel' in namespace '{NAMESPACE}'"
+        );
+
+        let result = HelmChartScore {
+            namespace: Some(NonBlankString::from_str(NAMESPACE).unwrap()),
             release_name: NonBlankString::from_str("zitadel").unwrap(),
             chart_name: NonBlankString::from_str("zitadel/zitadel").unwrap(),
             chart_version: None,
             values_overrides: None,
-            values_yaml,
+            values_yaml: Some(values_yaml),
             create_namespace: true,
             install_only: false,
             repository: Some(HelmRepository::new(
@@ -46,6 +479,40 @@ impl<T: Topology + K8sclient + HelmCommand> Score<T> for ZitadelScore {
                 true,
             )),
         }
-        .create_interpret()
+        .interpret(inventory, topology)
+        .await;
+
+        match &result {
+            Ok(_) => info!(
+                "[Zitadel] Helm chart deployed successfully\n\n\
+                ===== ZITADEL DEPLOYMENT COMPLETE =====\n\
+                Login URL: https://{host}\n\
+                Username: admin@zitadel.{host}\n\
+                Password: {admin_password}\n\n\
+                IMPORTANT: The password is saved in ConfigMap 'zitadel-config-yaml'\n\
+                and must be changed on first login. Save the credentials in a\n\
+                secure location after changing them.\n\
+                ========================================="
+            ),
+            Err(e) => error!("[Zitadel] Helm chart deployment failed: {e}"),
+        }
+
+        result
+    }
+
+    fn get_name(&self) -> InterpretName {
+        InterpretName::Custom("Zitadel")
+    }
+
+    fn get_version(&self) -> Version {
+        todo!()
+    }
+
+    fn get_status(&self) -> InterpretStatus {
+        todo!()
+    }
+
+    fn get_children(&self) -> Vec<Id> {
+        vec![]
     }
 }

opnsense-config-xml/src/data/opnsense.rs

@@ -344,7 +344,7 @@ pub struct StaticMap {
     pub mac: String,
     pub ipaddr: String,
     pub cid: Option<MaybeString>,
-    pub hostname: String,
+    pub hostname: Option<String>,
     pub descr: Option<MaybeString>,
     pub winsserver: MaybeString,
     pub dnsserver: MaybeString,
@@ -383,24 +383,24 @@ pub struct Outbound {
 
 #[derive(Default, PartialEq, Debug, YaSerialize, YaDeserialize)]
 pub struct NatRule {
-    pub protocol: String,
-    pub interface: String,
-    pub category: MaybeString,
-    pub ipprotocol: String,
-    pub descr: MaybeString,
-    pub tag: MaybeString,
+    pub protocol: Option<String>,
+    pub interface: Option<String>,
+    pub category: Option<MaybeString>,
+    pub ipprotocol: Option<String>,
+    pub descr: Option<MaybeString>,
+    pub tag: Option<MaybeString>,
     pub tagged: Option<MaybeString>,
-    pub poolopts: PoolOpts,
+    pub poolopts: Option<PoolOpts>,
     #[yaserde(rename = "associated-rule-id")]
     pub associated_rule_id: Option<MaybeString>,
     pub disabled: Option<u8>,
-    pub target: String,
+    pub target: Option<String>,
     #[yaserde(rename = "local-port")]
-    pub local_port: i32,
-    pub source: Source,
-    pub destination: Destination,
-    pub updated: Updated,
-    pub created: Created,
+    pub local_port: Option<i32>,
+    pub source: Option<Source>,
+    pub destination: Option<Destination>,
+    pub updated: Option<Updated>,
+    pub created: Option<Created>,
 }
 
 #[derive(Default, PartialEq, Debug, YaSerialize, YaDeserialize)]
@@ -1540,12 +1540,12 @@ pub struct Dyndns {
 pub struct Vlans {
     #[yaserde(attribute = true)]
     pub version: String,
-    pub vlan: MaybeString,
+    pub vlan: RawXml,
 }
 
 #[derive(Default, PartialEq, Debug, YaSerialize, YaDeserialize)]
 pub struct Bridges {
-    pub bridged: Option<MaybeString>,
+    pub bridged: Option<RawXml>,
 }
 
 #[derive(Default, PartialEq, Debug, YaSerialize, YaDeserialize)]

opnsense-config/src/modules/dhcp_legacy.rs

@@ -48,7 +48,7 @@ impl<'a> DhcpConfigLegacyISC<'a> {
         hostname: &str,
     ) -> Result<(), DhcpError> {
         let mac = mac.to_string();
-        let hostname = hostname.to_string();
+        let hostname = Some(hostname.to_string());
         let lan_dhcpd = self.get_lan_dhcpd();
         let existing_mappings: &mut Vec<StaticMap> = &mut lan_dhcpd.staticmaps;
 
@@ -121,7 +121,7 @@ impl<'a> DhcpConfigLegacyISC<'a> {
             .map(|entry| StaticMap {
                 mac: entry["mac"].as_str().unwrap_or_default().to_string(),
                 ipaddr: entry["ipaddr"].as_str().unwrap_or_default().to_string(),
-                hostname: entry["hostname"].as_str().unwrap_or_default().to_string(),
+                hostname: Some(entry["hostname"].as_str().unwrap_or_default().to_string()),
                 descr: entry["descr"].as_str().map(MaybeString::from),
                 ..Default::default()
             })

opnsense-config/src/modules/dnsmasq.rs

@@ -213,7 +213,7 @@ impl<'a> DhcpConfigDnsMasq<'a> {
             .map(|entry| StaticMap {
                 mac: entry["mac"].as_str().unwrap_or_default().to_string(),
                 ipaddr: entry["ipaddr"].as_str().unwrap_or_default().to_string(),
-                hostname: entry["hostname"].as_str().unwrap_or_default().to_string(),
+                hostname: Some(entry["hostname"].as_str().unwrap_or_default().to_string()),
                 descr: entry["descr"].as_str().map(MaybeString::from),
                 ..Default::default()
             })