metis/pkg/service/app.go

package service

import (
	"fmt"
	"os"
	"path/filepath"
	"sort"
	"strings"
	"sync"
	"time"

	"metis/pkg/facts"
	"metis/pkg/inventory"
	"metis/pkg/sentinel"
)

// JobStatus identifies the current lifecycle state of a queued job because
// the UI and metrics need a stable shared vocabulary for progress updates.
type JobStatus string

const (
	JobQueued  JobStatus = "queued"
	JobRunning JobStatus = "running"
	JobDone    JobStatus = "done"
	JobError   JobStatus = "error"
)

// Device describes a flashable block device.
type Device struct {
	Name      string `json:"name"`
	Path      string `json:"path"`
	Model     string `json:"model,omitempty"`
	Transport string `json:"transport,omitempty"`
	Type      string `json:"type,omitempty"`
	Note      string `json:"note,omitempty"`
	Removable bool   `json:"removable"`
	Hotplug   bool   `json:"hotplug"`
	SizeBytes int64  `json:"size_bytes"`
}

// Job is a long-running Metis action visible in the UI.
type Job struct {
	ID             string    `json:"id"`
	Kind           string    `json:"kind"`
	Node           string    `json:"node,omitempty"`
	Host           string    `json:"host,omitempty"`
	Builder        string    `json:"builder,omitempty"`
	Device         string    `json:"device,omitempty"`
	Status         JobStatus `json:"status"`
	Stage          string    `json:"stage,omitempty"`
	StageStartedAt time.Time `json:"stage_started_at,omitempty"`
	Message        string    `json:"message,omitempty"`
	Artifact       string    `json:"artifact,omitempty"`
	ProgressPct    float64   `json:"progress_pct"`
	Written        int64     `json:"written_bytes,omitempty"`
	Total          int64     `json:"total_bytes,omitempty"`
	Error          string    `json:"error,omitempty"`
	StartedAt      time.Time `json:"started_at"`
	UpdatedAt      time.Time `json:"updated_at,omitempty"`
	FinishedAt     time.Time `json:"finished_at,omitempty"`
}

// Event is a user-facing activity item for recent changes and runs.
type Event struct {
	Time    time.Time      `json:"time"`
	Kind    string         `json:"kind"`
	Summary string         `json:"summary"`
	Details map[string]any `json:"details,omitempty"`
}

// SnapshotRecord stores the last fact snapshot pushed by a node sentinel.
type SnapshotRecord struct {
	Node        string            `json:"node"`
	CollectedAt time.Time         `json:"collected_at"`
	Snapshot    sentinel.Snapshot `json:"snapshot"`
}

// PageState is the UI/API view model.
type PageState struct {
	LocalHost        string                     `json:"local_host"`
	DefaultFlashHost string                     `json:"default_flash_host"`
	SelectedHost     string                     `json:"selected_host"`
	FlashHosts       []string                   `json:"flash_hosts"`
	Nodes            []inventory.NodeSpec       `json:"nodes"`
	Jobs             []*Job                     `json:"jobs"`
	Devices          []Device                   `json:"devices"`
	PreferredDevice  string                     `json:"preferred_device,omitempty"`
	DeviceError      string                     `json:"device_error,omitempty"`
	Events           []Event                    `json:"events"`
	Snapshots        []SnapshotRecord           `json:"snapshots"`
	Targets          map[string]facts.Targets   `json:"targets"`
	Artifacts        map[string]ArtifactSummary `json:"artifacts"`
}

// ArtifactSummary describes the latest built image for a node.
type ArtifactSummary struct {
	Node        string    `json:"node,omitempty"`
	Ref         string    `json:"ref,omitempty"`
	BuildTag    string    `json:"build_tag,omitempty"`
	LocalPath   string    `json:"local_path,omitempty"`
	HostPath    string    `json:"host_path,omitempty"`
	BuilderHost string    `json:"builder_host,omitempty"`
	Compressed  bool      `json:"compressed,omitempty"`
	UpdatedAt   time.Time `json:"updated_at"`
	SizeBytes   int64     `json:"size_bytes"`
}

type deviceSnapshot struct {
	Devices   []Device
	Err       string
	CheckedAt time.Time
}

// App coordinates builds, flashes, sentinel snapshots, and the web UI state.
type App struct {
	settings  Settings
	inventory *inventory.Inventory
	metrics   *Metrics

	mu            sync.RWMutex
	jobs          map[string]*Job
	snapshots     map[string]SnapshotRecord
	targets       map[string]facts.Targets
	artifactStore map[string]ArtifactSummary
	deviceStore   map[string]deviceSnapshot
}

// NewApp creates a Metis service app instance.
func NewApp(settings Settings) (*App, error) {
	if err := os.MkdirAll(settings.CacheDir, 0o755); err != nil {
		return nil, err
	}
	if err := os.MkdirAll(settings.ArtifactDir, 0o755); err != nil {
		return nil, err
	}
	if err := os.MkdirAll(filepath.Dir(settings.HistoryPath), 0o755); err != nil {
		return nil, err
	}
	inv, err := inventory.Load(settings.InventoryPath)
	if err != nil {
		return nil, err
	}
	app := &App{
		settings:      settings,
		inventory:     inv,
		metrics:       NewMetrics(),
		jobs:          map[string]*Job{},
		snapshots:     map[string]SnapshotRecord{},
		targets:       map[string]facts.Targets{},
		artifactStore: map[string]ArtifactSummary{},
		deviceStore:   map[string]deviceSnapshot{},
	}
	_ = app.loadSnapshots()
	_ = app.loadTargets()
	_ = app.loadArtifacts()
	return app, nil
}

// State returns the current UI/API snapshot.
func (a *App) State(deviceHost string) PageState {
	if strings.TrimSpace(deviceHost) == "" {
		deviceHost = a.settings.DefaultFlashHost
	}
	a.mu.RLock()
	jobs := make([]*Job, 0, len(a.jobs))
	for _, job := range a.jobs {
		copyJob := *job
		jobs = append(jobs, &copyJob)
	}
	sort.Slice(jobs, func(i, j int) bool {
		return jobs[i].StartedAt.After(jobs[j].StartedAt)
	})

	snaps := make([]SnapshotRecord, 0, len(a.snapshots))
	for _, snap := range a.snapshots {
		snaps = append(snaps, snap)
	}
	aTargets := map[string]facts.Targets{}
	for key, value := range a.targets {
		aTargets[key] = value
	}
	a.mu.RUnlock()

	sort.Slice(snaps, func(i, j int) bool {
		return snaps[i].Node < snaps[j].Node
	})

	flashHosts := a.flashHosts()
	devices, deviceErr := a.cachedDevices(deviceHost)
	preferredDevice := preferredDevice(devices)
	return PageState{
		LocalHost:        a.settings.LocalHost,
		DefaultFlashHost: a.settings.DefaultFlashHost,
		SelectedHost:     deviceHost,
		FlashHosts:       flashHosts,
		Nodes:            a.replacementNodes(),
		Jobs:             jobs,
		Devices:          devices,
		PreferredDevice:  preferredDevice,
		DeviceError:      errorString(deviceErr),
		Events:           a.recentEvents(40),
		Snapshots:        snaps,
		Targets:          aTargets,
		Artifacts:        a.artifacts(),
	}
}

// Build starts a background image build for a node.
func (a *App) Build(node string) (*Job, error) {
	if err := a.ensureReplacementReady(node); err != nil {
		return nil, err
	}
	job := a.newJob("build", node, "", "")
	go a.runBuild(job, false)
	return job, nil
}

// Replace starts a background build+flash workflow for a node.
func (a *App) Replace(node, host, device string) (*Job, error) {
	if host == "" {
		host = a.settings.DefaultFlashHost
	}
	if err := a.ensureReplacementReady(node); err != nil {
		return nil, err
	}
	if _, err := a.ensureDevice(host, device); err != nil {
		return nil, err
	}
	job := a.newJob("replace", node, host, device)
	go a.runBuild(job, true)
	return job, nil
}

// StoreSnapshot records a pushed sentinel snapshot.
func (a *App) StoreSnapshot(record SnapshotRecord) error {
	if record.Node == "" {
		record.Node = record.Snapshot.Hostname
	}
	if record.CollectedAt.IsZero() {
		record.CollectedAt = time.Now().UTC()
	}
	if strings.TrimSpace(record.Node) == "" {
		return fmt.Errorf("snapshot node required")
	}
	a.mu.Lock()
	a.snapshots[record.Node] = record
	a.mu.Unlock()
	if err := a.persistSnapshots(); err != nil {
		return err
	}
	a.metrics.RecordSnapshot(record.Node, "ok", record.CollectedAt)
	a.appendEvent(Event{
		Time:    record.CollectedAt,
		Kind:    "sentinel.snapshot",
		Summary: fmt.Sprintf("Captured sentinel snapshot for %s", record.Node),
		Details: map[string]any{
			"node":        record.Node,
			"kernel":      record.Snapshot.Kernel,
			"k3s_version": record.Snapshot.K3sVersion,
		},
	})
	return nil
}

// WatchSentinel recomputes class targets and logs meaningful drift.
func (a *App) WatchSentinel() (*Event, error) {
	a.mu.RLock()
	snaps := make([]facts.Snapshot, 0, len(a.snapshots))
	for _, snap := range a.snapshots {
		snaps = append(snaps, facts.Snapshot{
			Hostname:      snap.Node,
			Kernel:        snap.Snapshot.Kernel,
			OSImage:       snap.Snapshot.OSImage,
			K3sVersion:    firstLine(snap.Snapshot.K3sVersion),
			Containerd:    firstLine(snap.Snapshot.Containerd),
			PackageSample: snap.Snapshot.PackageSample,
			DropInsSample: snap.Snapshot.DropInsSample,
		})
	}
	prevTargets := map[string]facts.Targets{}
	for key, value := range a.targets {
		prevTargets[key] = value
	}
	a.mu.RUnlock()

	nextTargets := facts.RecommendTargets(a.inventory, snaps)
	changes := diffTargets(prevTargets, nextTargets)

	a.mu.Lock()
	a.targets = nextTargets
	a.mu.Unlock()
	if err := a.persistTargets(); err != nil {
		return nil, err
	}

	event := &Event{
		Time:    time.Now().UTC(),
		Kind:    "sentinel.watch",
		Summary: "Metis sentinel watch completed with no template changes",
		Details: map[string]any{
			"classes": len(nextTargets),
			"changes": 0,
		},
	}
	if len(changes) > 0 {
		event.Summary = fmt.Sprintf("Metis sentinel watch detected %d template change(s)", len(changes))
		event.Details["changes"] = changes
	}
	a.appendEvent(*event)
	a.metrics.RecordWatch("ok")
	a.metrics.SetDriftTargets(nextTargets, len(changes))
	return event, nil
}