ananke/internal/config/config.go

package config

import (
	"fmt"
	neturl "net/url"
	"os"
	"strings"

	"gopkg.in/yaml.v3"
)

type Config struct {
	Kubeconfig          string            `yaml:"kubeconfig"`
	SSHUser             string            `yaml:"ssh_user"`
	SSHPort             int               `yaml:"ssh_port"`
	SSHConfigFile       string            `yaml:"ssh_config_file"`
	SSHIdentityFile     string            `yaml:"ssh_identity_file"`
	SSHNodeHosts        map[string]string `yaml:"ssh_node_hosts"`
	SSHNodeUsers        map[string]string `yaml:"ssh_node_users"`
	SSHManagedNodes     []string          `yaml:"ssh_managed_nodes"`
	SSHJumpHost         string            `yaml:"ssh_jump_host"`
	SSHJumpUser         string            `yaml:"ssh_jump_user"`
	IACRepoPath         string            `yaml:"iac_repo_path"`
	ExpectedFluxBranch  string            `yaml:"expected_flux_branch"`
	ExpectedFluxSource  string            `yaml:"expected_flux_source_url"`
	ControlPlanes       []string          `yaml:"control_planes"`
	Workers             []string          `yaml:"workers"`
	LocalBootstrapPaths []string          `yaml:"local_bootstrap_paths"`
	ExcludedNamespaces  []string          `yaml:"excluded_namespaces"`
	Startup             Startup           `yaml:"startup"`
	Shutdown            Shutdown          `yaml:"shutdown"`
	UPS                 UPS               `yaml:"ups"`
	Coordination        Coordination      `yaml:"coordination"`
	Metrics             Metrics           `yaml:"metrics"`
	State               State             `yaml:"state"`
}

type Startup struct {
	APIWaitSeconds                 int                          `yaml:"api_wait_seconds"`
	APIPollSeconds                 int                          `yaml:"api_poll_seconds"`
	ShutdownCooldownSeconds        int                          `yaml:"shutdown_cooldown_seconds"`
	MinimumBatteryPercent          float64                      `yaml:"minimum_battery_percent"`
	RequiredNodeLabels             map[string]map[string]string `yaml:"required_node_labels"`
	RequireTimeSync                bool                         `yaml:"require_time_sync"`
	TimeSyncWaitSeconds            int                          `yaml:"time_sync_wait_seconds"`
	TimeSyncPollSeconds            int                          `yaml:"time_sync_poll_seconds"`
	TimeSyncMode                   string                       `yaml:"time_sync_mode"`
	TimeSyncQuorum                 int                          `yaml:"time_sync_quorum"`
	ReconcileAccessOnBoot          bool                         `yaml:"reconcile_access_on_boot"`
	AutoEtcdRestoreOnAPIFailure    bool                         `yaml:"auto_etcd_restore_on_api_failure"`
	EtcdRestoreControlPlane        string                       `yaml:"etcd_restore_control_plane"`
	RequireStorageReady            bool                         `yaml:"require_storage_ready"`
	StorageReadyWaitSeconds        int                          `yaml:"storage_ready_wait_seconds"`
	StorageReadyPollSeconds        int                          `yaml:"storage_ready_poll_seconds"`
	StorageMinReadyNodes           int                          `yaml:"storage_min_ready_nodes"`
	StorageCriticalPVCs            []string                     `yaml:"storage_critical_pvcs"`
	RequirePostStartProbes         bool                         `yaml:"require_post_start_probes"`
	PostStartProbeWaitSeconds      int                          `yaml:"post_start_probe_wait_seconds"`
	PostStartProbePollSeconds      int                          `yaml:"post_start_probe_poll_seconds"`
	PostStartProbes                []string                     `yaml:"post_start_probes"`
	RequireServiceChecklist        bool                         `yaml:"require_service_checklist"`
	ServiceChecklistWaitSeconds    int                          `yaml:"service_checklist_wait_seconds"`
	ServiceChecklistPollSeconds    int                          `yaml:"service_checklist_poll_seconds"`
	ServiceChecklistStabilitySec   int                          `yaml:"service_checklist_stability_seconds"`
	ServiceChecklist               []ServiceChecklistCheck      `yaml:"service_checklist"`
	RequireIngressChecklist        bool                         `yaml:"require_ingress_checklist"`
	IngressChecklistWaitSeconds    int                          `yaml:"ingress_checklist_wait_seconds"`
	IngressChecklistPollSeconds    int                          `yaml:"ingress_checklist_poll_seconds"`
	IngressChecklistAccepted       []int                        `yaml:"ingress_checklist_accepted_statuses"`
	IngressChecklistIgnoreHosts    []string                     `yaml:"ingress_checklist_ignore_hosts"`
	IngressChecklistInsecureSkip   bool                         `yaml:"ingress_checklist_insecure_skip_tls"`
	RequireNodeSSHAuth             bool                         `yaml:"require_node_ssh_auth"`
	NodeSSHAuthWaitSeconds         int                          `yaml:"node_ssh_auth_wait_seconds"`
	NodeSSHAuthPollSeconds         int                          `yaml:"node_ssh_auth_poll_seconds"`
	RequireFluxHealth              bool                         `yaml:"require_flux_health"`
	FluxHealthWaitSeconds          int                          `yaml:"flux_health_wait_seconds"`
	FluxHealthPollSeconds          int                          `yaml:"flux_health_poll_seconds"`
	IgnoreFluxKustomizations       []string                     `yaml:"ignore_flux_kustomizations"`
	RequireWorkloadConvergence     bool                         `yaml:"require_workload_convergence"`
	WorkloadConvergenceWaitSeconds int                          `yaml:"workload_convergence_wait_seconds"`
	WorkloadConvergencePollSeconds int                          `yaml:"workload_convergence_poll_seconds"`
	IgnoreWorkloadNamespaces       []string                     `yaml:"ignore_workload_namespaces"`
	IgnoreWorkloads                []string                     `yaml:"ignore_workloads"`
	IgnoreUnavailableNodes         []string                     `yaml:"ignore_unavailable_nodes"`
	AutoRecycleStuckPods           bool                         `yaml:"auto_recycle_stuck_pods"`
	StuckPodGraceSeconds           int                          `yaml:"stuck_pod_grace_seconds"`
	VaultUnsealKeyFile             string                       `yaml:"vault_unseal_key_file"`
	VaultUnsealBreakglassCommand   string                       `yaml:"vault_unseal_breakglass_command"`
	VaultUnsealBreakglassTimeout   int                          `yaml:"vault_unseal_breakglass_timeout_seconds"`
}

type ServiceChecklistCheck struct {
	Name             string `yaml:"name"`
	URL              string `yaml:"url"`
	AcceptedStatuses []int  `yaml:"accepted_statuses"`
	BodyContains     string `yaml:"body_contains"`
	BodyNotContains  string `yaml:"body_not_contains"`
	TimeoutSeconds   int    `yaml:"timeout_seconds"`
	InsecureSkipTLS  bool   `yaml:"insecure_skip_tls"`
}

type Shutdown struct {
	DefaultBudgetSeconds int      `yaml:"default_budget_seconds"`
	HistoryMinSamples    int      `yaml:"history_min_samples"`
	EmergencyBudgetSec   int      `yaml:"emergency_budget_seconds"`
	EmergencyMinSamples  int      `yaml:"emergency_history_min_samples"`
	EmergencySkipEtcd    bool     `yaml:"emergency_skip_etcd_snapshot"`
	EmergencySkipDrain   bool     `yaml:"emergency_skip_drain"`
	SkipEtcdSnapshot     bool     `yaml:"skip_etcd_snapshot"`
	SkipDrain            bool     `yaml:"skip_drain"`
	DrainParallelism     int      `yaml:"drain_parallelism"`
	ScaleParallelism     int      `yaml:"scale_parallelism"`
	SSHParallelism       int      `yaml:"ssh_parallelism"`
	PoweroffEnabled      bool     `yaml:"poweroff_enabled"`
	PoweroffDelaySeconds int      `yaml:"poweroff_delay_seconds"`
	PoweroffLocalHost    bool     `yaml:"poweroff_local_host"`
	ExtraPoweroffHosts   []string `yaml:"extra_poweroff_hosts"`
}

type UPS struct {
	Enabled                 bool        `yaml:"enabled"`
	Provider                string      `yaml:"provider"`
	Target                  string      `yaml:"target"`
	Targets                 []UPSTarget `yaml:"targets"`
	PollSeconds             int         `yaml:"poll_seconds"`
	RuntimeSafetyFactor     float64     `yaml:"runtime_safety_factor"`
	DebounceCount           int         `yaml:"debounce_count"`
	TelemetryTimeoutSeconds int         `yaml:"telemetry_timeout_seconds"`
}

type UPSTarget struct {
	Name   string `yaml:"name"`
	Target string `yaml:"target"`
}

type Coordination struct {
	ForwardShutdownHost   string   `yaml:"forward_shutdown_host"`
	ForwardShutdownUser   string   `yaml:"forward_shutdown_user"`
	ForwardShutdownConfig string   `yaml:"forward_shutdown_config"`
	PeerHosts             []string `yaml:"peer_hosts"`
	FallbackLocalShutdown bool     `yaml:"fallback_local_shutdown"`
	CommandTimeoutSeconds int      `yaml:"command_timeout_seconds"`
	StartupGuardMaxAgeSec int      `yaml:"startup_guard_max_age_seconds"`
	Role                  string   `yaml:"role"`
	AllowStartupOnBattery bool     `yaml:"allow_startup_on_battery"`
}

type Metrics struct {
	Enabled  bool   `yaml:"enabled"`
	BindAddr string `yaml:"bind_addr"`
	Path     string `yaml:"path"`
}

type State struct {
	Dir            string `yaml:"dir"`
	RunHistoryPath string `yaml:"run_history_path"`
	LockPath       string `yaml:"lock_path"`
	IntentPath     string `yaml:"intent_path"`
}

func Load(path string) (Config, error) {
	cfg := defaults()

	b, err := os.ReadFile(path)
	if err != nil {
		return Config{}, fmt.Errorf("read config %s: %w", path, err)
	}
	if err := yaml.Unmarshal(b, &cfg); err != nil {
		return Config{}, fmt.Errorf("decode config %s: %w", path, err)
	}

	cfg.applyDefaults()
	if err := cfg.Validate(); err != nil {
		return Config{}, err
	}
	return cfg, nil
}

func (c Config) Validate() error {
	if len(c.ControlPlanes) == 0 {
		return fmt.Errorf("config.control_planes must not be empty")
	}
	if c.ExpectedFluxBranch == "" {
		return fmt.Errorf("config.expected_flux_branch must not be empty")
	}
	if c.ExpectedFluxSource == "" {
		return fmt.Errorf("config.expected_flux_source_url must not be empty")
	}
	if c.IACRepoPath == "" {
		return fmt.Errorf("config.iac_repo_path must not be empty")
	}
	if c.Shutdown.DefaultBudgetSeconds <= 0 {
		return fmt.Errorf("config.shutdown.default_budget_seconds must be > 0")
	}
	if c.Shutdown.HistoryMinSamples <= 0 {
		return fmt.Errorf("config.shutdown.history_min_samples must be > 0")
	}
	if c.Shutdown.EmergencyBudgetSec <= 0 {
		return fmt.Errorf("config.shutdown.emergency_budget_seconds must be > 0")
	}
	if c.Shutdown.EmergencyMinSamples <= 0 {
		return fmt.Errorf("config.shutdown.emergency_history_min_samples must be > 0")
	}
	if c.Shutdown.DrainParallelism <= 0 {
		return fmt.Errorf("config.shutdown.drain_parallelism must be > 0")
	}
	if c.Shutdown.ScaleParallelism <= 0 {
		return fmt.Errorf("config.shutdown.scale_parallelism must be > 0")
	}
	if c.Shutdown.SSHParallelism <= 0 {
		return fmt.Errorf("config.shutdown.ssh_parallelism must be > 0")
	}
	if c.Startup.APIWaitSeconds <= 0 {
		return fmt.Errorf("config.startup.api_wait_seconds must be > 0")
	}
	if c.Startup.APIPollSeconds <= 0 {
		return fmt.Errorf("config.startup.api_poll_seconds must be > 0")
	}
	if c.Startup.ShutdownCooldownSeconds <= 0 {
		return fmt.Errorf("config.startup.shutdown_cooldown_seconds must be > 0")
	}
	if c.Startup.MinimumBatteryPercent < 0 || c.Startup.MinimumBatteryPercent > 100 {
		return fmt.Errorf("config.startup.minimum_battery_percent must be between 0 and 100")
	}
	for node, labels := range c.Startup.RequiredNodeLabels {
		if strings.TrimSpace(node) == "" {
			return fmt.Errorf("config.startup.required_node_labels keys must not be empty")
		}
		if len(labels) == 0 {
			return fmt.Errorf("config.startup.required_node_labels[%q] must include at least one label", node)
		}
		for key, value := range labels {
			if strings.TrimSpace(key) == "" {
				return fmt.Errorf("config.startup.required_node_labels[%q] contains empty label key", node)
			}
			if strings.TrimSpace(value) == "" {
				return fmt.Errorf("config.startup.required_node_labels[%q][%q] must not be empty", node, key)
			}
		}
	}
	if c.Startup.TimeSyncWaitSeconds <= 0 {
		return fmt.Errorf("config.startup.time_sync_wait_seconds must be > 0")
	}
	if c.Startup.TimeSyncPollSeconds <= 0 {
		return fmt.Errorf("config.startup.time_sync_poll_seconds must be > 0")
	}
	if c.Startup.TimeSyncMode != "strict" && c.Startup.TimeSyncMode != "quorum" {
		return fmt.Errorf("config.startup.time_sync_mode must be strict or quorum")
	}
	if c.Startup.TimeSyncMode == "quorum" && c.Startup.TimeSyncQuorum <= 0 {
		return fmt.Errorf("config.startup.time_sync_quorum must be > 0 when time_sync_mode=quorum")
	}
	if c.Startup.EtcdRestoreControlPlane != "" {
		found := false
		for _, cp := range c.ControlPlanes {
			if cp == c.Startup.EtcdRestoreControlPlane {
				found = true
				break
			}
		}
		if !found {
			return fmt.Errorf("config.startup.etcd_restore_control_plane must be one of config.control_planes when set")
		}
	}
	if c.Startup.StorageReadyWaitSeconds <= 0 {
		return fmt.Errorf("config.startup.storage_ready_wait_seconds must be > 0")
	}
	if c.Startup.StorageReadyPollSeconds <= 0 {
		return fmt.Errorf("config.startup.storage_ready_poll_seconds must be > 0")
	}
	if c.Startup.StorageMinReadyNodes <= 0 {
		return fmt.Errorf("config.startup.storage_min_ready_nodes must be > 0")
	}
	for _, pvc := range c.Startup.StorageCriticalPVCs {
		if strings.Count(strings.TrimSpace(pvc), "/") != 1 {
			return fmt.Errorf("config.startup.storage_critical_pvcs entries must be namespace/name, got %q", pvc)
		}
	}
	if c.Startup.PostStartProbeWaitSeconds <= 0 {
		return fmt.Errorf("config.startup.post_start_probe_wait_seconds must be > 0")
	}
	if c.Startup.PostStartProbePollSeconds <= 0 {
		return fmt.Errorf("config.startup.post_start_probe_poll_seconds must be > 0")
	}
	if c.Startup.RequirePostStartProbes && len(c.Startup.PostStartProbes) == 0 {
		return fmt.Errorf("config.startup.post_start_probes must not be empty when require_post_start_probes is true")
	}
	if c.Startup.ServiceChecklistWaitSeconds <= 0 {
		return fmt.Errorf("config.startup.service_checklist_wait_seconds must be > 0")
	}
	if c.Startup.ServiceChecklistPollSeconds <= 0 {
		return fmt.Errorf("config.startup.service_checklist_poll_seconds must be > 0")
	}
	if c.Startup.ServiceChecklistStabilitySec < 0 {
		return fmt.Errorf("config.startup.service_checklist_stability_seconds must be >= 0")
	}
	if c.Startup.RequireServiceChecklist && len(c.Startup.ServiceChecklist) == 0 {
		return fmt.Errorf("config.startup.service_checklist must not be empty when require_service_checklist is true")
	}
	for i, check := range c.Startup.ServiceChecklist {
		if strings.TrimSpace(check.Name) == "" {
			return fmt.Errorf("config.startup.service_checklist[%d].name must not be empty", i)
		}
		rawURL := strings.TrimSpace(check.URL)
		if rawURL == "" {
			return fmt.Errorf("config.startup.service_checklist[%d].url must not be empty", i)
		}
		parsed, err := neturl.Parse(rawURL)
		if err != nil || parsed.Scheme == "" || parsed.Host == "" {
			return fmt.Errorf("config.startup.service_checklist[%d].url is invalid: %q", i, rawURL)
		}
		if check.TimeoutSeconds <= 0 {
			return fmt.Errorf("config.startup.service_checklist[%d].timeout_seconds must be > 0", i)
		}
		for _, code := range check.AcceptedStatuses {
			if code < 100 || code > 599 {
				return fmt.Errorf("config.startup.service_checklist[%d].accepted_statuses contains invalid HTTP code %d", i, code)
			}
		}
	}
	if c.Startup.IngressChecklistWaitSeconds <= 0 {
		return fmt.Errorf("config.startup.ingress_checklist_wait_seconds must be > 0")
	}
	if c.Startup.IngressChecklistPollSeconds <= 0 {
		return fmt.Errorf("config.startup.ingress_checklist_poll_seconds must be > 0")
	}
	for _, code := range c.Startup.IngressChecklistAccepted {
		if code < 100 || code > 599 {
			return fmt.Errorf("config.startup.ingress_checklist_accepted_statuses contains invalid HTTP code %d", code)
		}
	}
	for _, host := range c.Startup.IngressChecklistIgnoreHosts {
		if strings.TrimSpace(host) == "" {
			return fmt.Errorf("config.startup.ingress_checklist_ignore_hosts entries must not be empty")
		}
	}
	if c.Startup.NodeSSHAuthWaitSeconds <= 0 {
		return fmt.Errorf("config.startup.node_ssh_auth_wait_seconds must be > 0")
	}
	if c.Startup.NodeSSHAuthPollSeconds <= 0 {
		return fmt.Errorf("config.startup.node_ssh_auth_poll_seconds must be > 0")
	}
	if c.Startup.FluxHealthWaitSeconds <= 0 {
		return fmt.Errorf("config.startup.flux_health_wait_seconds must be > 0")
	}
	if c.Startup.FluxHealthPollSeconds <= 0 {
		return fmt.Errorf("config.startup.flux_health_poll_seconds must be > 0")
	}
	if c.Startup.WorkloadConvergenceWaitSeconds <= 0 {
		return fmt.Errorf("config.startup.workload_convergence_wait_seconds must be > 0")
	}
	if c.Startup.WorkloadConvergencePollSeconds <= 0 {
		return fmt.Errorf("config.startup.workload_convergence_poll_seconds must be > 0")
	}
	if c.Startup.StuckPodGraceSeconds <= 0 {
		return fmt.Errorf("config.startup.stuck_pod_grace_seconds must be > 0")
	}
	for _, probe := range c.Startup.PostStartProbes {
		if strings.TrimSpace(probe) == "" {
			return fmt.Errorf("config.startup.post_start_probes entries must not be empty")
		}
	}
	for _, item := range c.Startup.IgnoreFluxKustomizations {
		item = strings.TrimSpace(item)
		if item == "" {
			return fmt.Errorf("config.startup.ignore_flux_kustomizations entries must not be empty")
		}
		if strings.Count(item, "/") != 1 {
			return fmt.Errorf("config.startup.ignore_flux_kustomizations entries must be namespace/name, got %q", item)
		}
	}
	for _, item := range c.Startup.IgnoreWorkloads {
		item = strings.TrimSpace(item)
		if item == "" {
			return fmt.Errorf("config.startup.ignore_workloads entries must not be empty")
		}
		parts := strings.Split(item, "/")
		if len(parts) != 2 && len(parts) != 3 {
			return fmt.Errorf("config.startup.ignore_workloads entries must be namespace/name or namespace/kind/name, got %q", item)
		}
	}
	for _, ns := range c.Startup.IgnoreWorkloadNamespaces {
		if strings.TrimSpace(ns) == "" {
			return fmt.Errorf("config.startup.ignore_workload_namespaces entries must not be empty")
		}
	}
	for _, node := range c.Startup.IgnoreUnavailableNodes {
		if strings.TrimSpace(node) == "" {
			return fmt.Errorf("config.startup.ignore_unavailable_nodes entries must not be empty")
		}
	}
	if strings.TrimSpace(c.Startup.VaultUnsealKeyFile) == "" {
		return fmt.Errorf("config.startup.vault_unseal_key_file must not be empty")
	}
	if c.SSHPort <= 0 || c.SSHPort > 65535 {
		return fmt.Errorf("config.ssh_port must be in range 1-65535")
	}
	if c.UPS.Enabled {
		if c.UPS.Provider == "" {
			return fmt.Errorf("config.ups.provider must not be empty when ups is enabled")
		}
		if c.UPS.Target == "" && len(c.UPS.Targets) == 0 {
			return fmt.Errorf("config.ups.target or config.ups.targets must be set when ups is enabled")
		}
		for _, t := range c.UPS.Targets {
			if t.Target == "" {
				return fmt.Errorf("config.ups.targets[].target must not be empty")
			}
		}
	}
	if c.Coordination.ForwardShutdownHost != "" {
		if c.Coordination.ForwardShutdownConfig == "" {
			return fmt.Errorf("config.coordination.forward_shutdown_config must not be empty when forward_shutdown_host is set")
		}
	}
	for _, peer := range c.Coordination.PeerHosts {
		if strings.TrimSpace(peer) == "" {
			return fmt.Errorf("config.coordination.peer_hosts entries must not be empty")
		}
	}
	if c.Coordination.StartupGuardMaxAgeSec <= 0 {
		return fmt.Errorf("config.coordination.startup_guard_max_age_seconds must be > 0")
	}
	if c.Coordination.Role != "coordinator" && c.Coordination.Role != "peer" {
		return fmt.Errorf("config.coordination.role must be coordinator or peer")
	}
	if c.State.RunHistoryPath == "" || c.State.LockPath == "" {
		return fmt.Errorf("config.state.run_history_path and config.state.lock_path must not be empty")
	}
	if c.State.IntentPath == "" {
		return fmt.Errorf("config.state.intent_path must not be empty")
	}
	return nil
}

func defaults() Config {
	c := Config{
		IACRepoPath:        "/opt/titan-iac",
		ExpectedFluxBranch: "main",
		ExpectedFluxSource: "ssh://git@scm.bstein.dev:2242/bstein/titan-iac.git",
		SSHPort:            2277,
		ControlPlanes:      []string{"titan-0a", "titan-0b", "titan-0c"},
		LocalBootstrapPaths: []string{
			"infrastructure/core",
			"clusters/atlas/flux-system",
			"infrastructure/sources/helm",
			"infrastructure/metallb",
			"infrastructure/traefik",
			"infrastructure/cert-manager",
			"infrastructure/vault-csi",
			"infrastructure/vault-injector",
			"services/vault",
			"infrastructure/postgres",
			"services/gitea",
			"services/keycloak",
			"services/oauth2-proxy",
		},
		ExcludedNamespaces: []string{
			"kube-system",
			"kube-public",
			"kube-node-lease",
			"flux-system",
			"traefik",
			"metallb-system",
			"cert-manager",
			"longhorn-system",
			"vault",
			"postgres",
			"maintenance",
		},
		Startup: Startup{
			APIWaitSeconds:              1200,
			APIPollSeconds:              2,
			ShutdownCooldownSeconds:     45,
			RequireTimeSync:             true,
			TimeSyncWaitSeconds:         240,
			TimeSyncPollSeconds:         5,
			TimeSyncMode:                "quorum",
			TimeSyncQuorum:              2,
			ReconcileAccessOnBoot:       true,
			AutoEtcdRestoreOnAPIFailure: true,
			EtcdRestoreControlPlane:     "titan-0a",
			RequireStorageReady:         true,
			StorageReadyWaitSeconds:     420,
			StorageReadyPollSeconds:     5,
			StorageMinReadyNodes:        2,
			StorageCriticalPVCs: []string{
				"vault/data-vault-0",
				"postgres/postgres-data-postgres-0",
				"gitea/gitea-data",
				"sso/keycloak-data",
			},
			MinimumBatteryPercent: 20,
			RequiredNodeLabels: map[string]map[string]string{
				"titan-09": {
					"ananke.bstein.dev/harbor-bootstrap": "true",
				},
			},
			RequirePostStartProbes:    true,
			PostStartProbeWaitSeconds: 240,
			PostStartProbePollSeconds: 5,
			PostStartProbes: []string{
				"https://sso.bstein.dev/realms/atlas/.well-known/openid-configuration",
				"https://scm.bstein.dev/api/healthz",
				"https://metrics.bstein.dev/api/health",
			},
			RequireServiceChecklist:      true,
			ServiceChecklistWaitSeconds:  420,
			ServiceChecklistPollSeconds:  5,
			ServiceChecklistStabilitySec: 120,
			ServiceChecklist: []ServiceChecklistCheck{
				{
					Name:             "gitea-api",
					URL:              "https://scm.bstein.dev/api/healthz",
					AcceptedStatuses: []int{200},
					BodyContains:     "pass",
					TimeoutSeconds:   12,
				},
				{
					Name:             "grafana-api",
					URL:              "https://metrics.bstein.dev/api/health",
					AcceptedStatuses: []int{200},
					BodyContains:     "\"database\":\"ok\"",
					TimeoutSeconds:   12,
				},
				{
					Name:             "keycloak-oidc",
					URL:              "https://sso.bstein.dev/realms/atlas/.well-known/openid-configuration",
					AcceptedStatuses: []int{200},
					BodyContains:     "\"issuer\":\"https://sso.bstein.dev/realms/atlas\"",
					TimeoutSeconds:   12,
				},
				{
					Name:             "harbor-registry",
					URL:              "https://registry.bstein.dev/v2/",
					AcceptedStatuses: []int{401},
					BodyContains:     "unauthorized",
					TimeoutSeconds:   12,
				},
				{
					Name:             "longhorn-auth",
					URL:              "https://longhorn.bstein.dev/",
					AcceptedStatuses: []int{200, 302},
					TimeoutSeconds:   12,
				},
			},
			RequireIngressChecklist:        true,
			IngressChecklistWaitSeconds:    420,
			IngressChecklistPollSeconds:    5,
			IngressChecklistAccepted:       []int{200, 301, 302, 307, 308, 401, 403, 404},
			IngressChecklistIgnoreHosts:    []string{},
			RequireNodeSSHAuth:             true,
			NodeSSHAuthWaitSeconds:         240,
			NodeSSHAuthPollSeconds:         5,
			RequireFluxHealth:              true,
			FluxHealthWaitSeconds:          900,
			FluxHealthPollSeconds:          5,
			IgnoreFluxKustomizations:       []string{},
			RequireWorkloadConvergence:     true,
			WorkloadConvergenceWaitSeconds: 900,
			WorkloadConvergencePollSeconds: 5,
			IgnoreWorkloadNamespaces:       []string{},
			IgnoreWorkloads:                []string{},
			IgnoreUnavailableNodes:         []string{},
			AutoRecycleStuckPods:           true,
			StuckPodGraceSeconds:           180,
			VaultUnsealKeyFile:             "/var/lib/ananke/vault-unseal.key",
			VaultUnsealBreakglassTimeout:   15,
		},
		Shutdown: Shutdown{
			DefaultBudgetSeconds: 1380,
			HistoryMinSamples:    3,
			EmergencyBudgetSec:   420,
			EmergencyMinSamples:  3,
			EmergencySkipEtcd:    true,
			EmergencySkipDrain:   true,
			DrainParallelism:     6,
			ScaleParallelism:     8,
			SSHParallelism:       8,
			PoweroffEnabled:      false,
			PoweroffDelaySeconds: 25,
			PoweroffLocalHost:    false,
		},
		UPS: UPS{
			Enabled:                 true,
			Provider:                "nut",
			PollSeconds:             5,
			RuntimeSafetyFactor:     1.25,
			DebounceCount:           3,
			TelemetryTimeoutSeconds: 90,
		},
		Coordination: Coordination{
			ForwardShutdownConfig: "/etc/ananke/ananke.yaml",
			PeerHosts:             []string{},
			FallbackLocalShutdown: true,
			CommandTimeoutSeconds: 25,
			StartupGuardMaxAgeSec: 900,
			Role:                  "coordinator",
			AllowStartupOnBattery: false,
		},
		Metrics: Metrics{
			Enabled:  true,
			BindAddr: "0.0.0.0:9560",
			Path:     "/metrics",
		},
		State: State{
			Dir:            "/var/lib/ananke",
			RunHistoryPath: "/var/lib/ananke/runs.json",
			LockPath:       "/var/lib/ananke/ananke.lock",
			IntentPath:     "/var/lib/ananke/intent.json",
		},
	}
	c.applyDefaults()
	return c
}

func (c *Config) applyDefaults() {
	if c.ExpectedFluxBranch == "" {
		c.ExpectedFluxBranch = "main"
	}
	if c.IACRepoPath == "" {
		c.IACRepoPath = "/opt/titan-iac"
	}
	if c.ExpectedFluxSource == "" {
		c.ExpectedFluxSource = "ssh://git@scm.bstein.dev:2242/bstein/titan-iac.git"
	}
	if c.Startup.APIWaitSeconds <= 0 {
		c.Startup.APIWaitSeconds = 1200
	}
	if c.Startup.APIPollSeconds <= 0 {
		c.Startup.APIPollSeconds = 2
	}
	if c.Startup.ShutdownCooldownSeconds <= 0 {
		c.Startup.ShutdownCooldownSeconds = 45
	}
	if c.Startup.MinimumBatteryPercent <= 0 {
		c.Startup.MinimumBatteryPercent = 20
	}
	if c.Startup.RequiredNodeLabels == nil {
		c.Startup.RequiredNodeLabels = map[string]map[string]string{
			"titan-09": {
				"ananke.bstein.dev/harbor-bootstrap": "true",
			},
		}
	}
	if c.Startup.TimeSyncWaitSeconds <= 0 {
		c.Startup.TimeSyncWaitSeconds = 240
	}
	if c.Startup.TimeSyncPollSeconds <= 0 {
		c.Startup.TimeSyncPollSeconds = 5
	}
	if c.Startup.TimeSyncMode == "" {
		c.Startup.TimeSyncMode = "quorum"
	}
	if c.Startup.TimeSyncQuorum <= 0 {
		c.Startup.TimeSyncQuorum = 2
	}
	if c.Startup.TimeSyncQuorum > len(c.ControlPlanes) && len(c.ControlPlanes) > 0 {
		c.Startup.TimeSyncQuorum = len(c.ControlPlanes)
	}
	if c.Startup.EtcdRestoreControlPlane == "" && len(c.ControlPlanes) > 0 {
		c.Startup.EtcdRestoreControlPlane = c.ControlPlanes[0]
	}
	if c.Startup.StorageReadyWaitSeconds <= 0 {
		c.Startup.StorageReadyWaitSeconds = 420
	}
	if c.Startup.StorageReadyPollSeconds <= 0 {
		c.Startup.StorageReadyPollSeconds = 5
	}
	if c.Startup.StorageMinReadyNodes <= 0 {
		c.Startup.StorageMinReadyNodes = 2
	}
	if len(c.Startup.StorageCriticalPVCs) == 0 {
		c.Startup.StorageCriticalPVCs = []string{
			"vault/data-vault-0",
			"postgres/postgres-data-postgres-0",
			"gitea/gitea-data",
			"sso/keycloak-data",
		}
	}
	if c.Startup.PostStartProbeWaitSeconds <= 0 {
		c.Startup.PostStartProbeWaitSeconds = 240
	}
	if c.Startup.PostStartProbePollSeconds <= 0 {
		c.Startup.PostStartProbePollSeconds = 5
	}
	if len(c.Startup.PostStartProbes) == 0 {
		c.Startup.PostStartProbes = []string{
			"https://sso.bstein.dev/realms/atlas/.well-known/openid-configuration",
			"https://scm.bstein.dev/api/healthz",
			"https://metrics.bstein.dev/api/health",
		}
	}
	if c.Startup.ServiceChecklistWaitSeconds <= 0 {
		c.Startup.ServiceChecklistWaitSeconds = 420
	}
	if c.Startup.ServiceChecklistPollSeconds <= 0 {
		c.Startup.ServiceChecklistPollSeconds = 5
	}
	if c.Startup.ServiceChecklistStabilitySec < 0 {
		c.Startup.ServiceChecklistStabilitySec = 0
	}
	if len(c.Startup.ServiceChecklist) == 0 {
		c.Startup.ServiceChecklist = []ServiceChecklistCheck{
			{
				Name:             "gitea-api",
				URL:              "https://scm.bstein.dev/api/healthz",
				AcceptedStatuses: []int{200},
				BodyContains:     "pass",
				TimeoutSeconds:   12,
			},
			{
				Name:             "grafana-api",
				URL:              "https://metrics.bstein.dev/api/health",
				AcceptedStatuses: []int{200},
				BodyContains:     "\"database\":\"ok\"",
				TimeoutSeconds:   12,
			},
		}
	}
	for i := range c.Startup.ServiceChecklist {
		if c.Startup.ServiceChecklist[i].TimeoutSeconds <= 0 {
			c.Startup.ServiceChecklist[i].TimeoutSeconds = 12
		}
	}
	if c.Startup.IngressChecklistWaitSeconds <= 0 {
		c.Startup.IngressChecklistWaitSeconds = 420
	}
	if c.Startup.IngressChecklistPollSeconds <= 0 {
		c.Startup.IngressChecklistPollSeconds = 5
	}
	if len(c.Startup.IngressChecklistAccepted) == 0 {
		c.Startup.IngressChecklistAccepted = []int{200, 301, 302, 307, 308, 401, 403, 404}
	}
	if c.Startup.IngressChecklistIgnoreHosts == nil {
		c.Startup.IngressChecklistIgnoreHosts = []string{}
	}
	if c.Startup.NodeSSHAuthWaitSeconds <= 0 {
		c.Startup.NodeSSHAuthWaitSeconds = 240
	}
	if c.Startup.NodeSSHAuthPollSeconds <= 0 {
		c.Startup.NodeSSHAuthPollSeconds = 5
	}
	if c.Startup.FluxHealthWaitSeconds <= 0 {
		c.Startup.FluxHealthWaitSeconds = 900
	}
	if c.Startup.FluxHealthPollSeconds <= 0 {
		c.Startup.FluxHealthPollSeconds = 5
	}
	if c.Startup.IgnoreFluxKustomizations == nil {
		c.Startup.IgnoreFluxKustomizations = []string{}
	}
	if c.Startup.WorkloadConvergenceWaitSeconds <= 0 {
		c.Startup.WorkloadConvergenceWaitSeconds = 900
	}
	if c.Startup.WorkloadConvergencePollSeconds <= 0 {
		c.Startup.WorkloadConvergencePollSeconds = 5
	}
	if c.Startup.IgnoreWorkloadNamespaces == nil {
		c.Startup.IgnoreWorkloadNamespaces = []string{}
	}
	if c.Startup.IgnoreWorkloads == nil {
		c.Startup.IgnoreWorkloads = []string{}
	}
	if c.Startup.IgnoreUnavailableNodes == nil {
		c.Startup.IgnoreUnavailableNodes = []string{}
	}
	if c.Startup.StuckPodGraceSeconds <= 0 {
		c.Startup.StuckPodGraceSeconds = 180
	}
	if strings.TrimSpace(c.Startup.VaultUnsealKeyFile) == "" {
		c.Startup.VaultUnsealKeyFile = "/var/lib/ananke/vault-unseal.key"
	}
	if c.Startup.VaultUnsealBreakglassTimeout <= 0 {
		c.Startup.VaultUnsealBreakglassTimeout = 15
	}
	if c.SSHPort <= 0 {
		c.SSHPort = 2277
	}
	if c.Shutdown.DefaultBudgetSeconds <= 0 {
		c.Shutdown.DefaultBudgetSeconds = 1380
	}
	if c.Shutdown.HistoryMinSamples <= 0 {
		c.Shutdown.HistoryMinSamples = 3
	}
	if c.Shutdown.EmergencyBudgetSec <= 0 {
		c.Shutdown.EmergencyBudgetSec = 420
	}
	if c.Shutdown.EmergencyMinSamples <= 0 {
		c.Shutdown.EmergencyMinSamples = 3
	}
	if c.Shutdown.DrainParallelism <= 0 {
		c.Shutdown.DrainParallelism = 6
	}
	if c.Shutdown.ScaleParallelism <= 0 {
		c.Shutdown.ScaleParallelism = 8
	}
	if c.Shutdown.SSHParallelism <= 0 {
		c.Shutdown.SSHParallelism = 8
	}
	if c.Shutdown.PoweroffDelaySeconds <= 0 {
		c.Shutdown.PoweroffDelaySeconds = 25
	}
	if c.UPS.PollSeconds <= 0 {
		c.UPS.PollSeconds = 5
	}
	if c.UPS.RuntimeSafetyFactor <= 0 {
		c.UPS.RuntimeSafetyFactor = 1.25
	}
	if c.UPS.DebounceCount <= 0 {
		c.UPS.DebounceCount = 3
	}
	if c.UPS.TelemetryTimeoutSeconds <= 0 {
		c.UPS.TelemetryTimeoutSeconds = 90
	}
	if c.Coordination.ForwardShutdownConfig == "" {
		c.Coordination.ForwardShutdownConfig = "/etc/ananke/ananke.yaml"
	}
	if c.Coordination.PeerHosts == nil {
		c.Coordination.PeerHosts = []string{}
	}
	if c.Coordination.CommandTimeoutSeconds <= 0 {
		c.Coordination.CommandTimeoutSeconds = 25
	}
	if c.Coordination.StartupGuardMaxAgeSec <= 0 {
		c.Coordination.StartupGuardMaxAgeSec = 900
	}
	if c.Coordination.Role == "" {
		c.Coordination.Role = "coordinator"
	}
	if c.Metrics.BindAddr == "" {
		c.Metrics.BindAddr = "0.0.0.0:9560"
	}
	if c.Metrics.Path == "" {
		c.Metrics.Path = "/metrics"
	}
	if c.State.Dir == "" {
		c.State.Dir = "/var/lib/ananke"
	}
	if c.State.RunHistoryPath == "" {
		c.State.RunHistoryPath = "/var/lib/ananke/runs.json"
	}
	if c.State.LockPath == "" {
		c.State.LockPath = "/var/lib/ananke/ananke.lock"
	}
	if c.State.IntentPath == "" {
		c.State.IntentPath = "/var/lib/ananke/intent.json"
	}
}