felhom-controller/controller/internal/selfupdate/updater.go

package selfupdate

import (
	"bytes"
	"encoding/json"
	"fmt"
	"log"
	"net/http"
	"os"
	"os/exec"
	"regexp"
	"strings"
	"sync"
	"time"

	"gitea.dooplex.hu/admin/felhom-controller/internal/config"
)

// CheckResult holds the result of a version check.
type CheckResult struct {
	CurrentVersion  string `json:"current_version"`
	LatestVersion   string `json:"latest_version"`
	UpdateAvailable bool   `json:"update_available"`
	Error           string `json:"error,omitempty"`
	CheckedAt       string `json:"checked_at"`
}

// UpdateStatus is the complete status returned by the API.
type UpdateStatus struct {
	Running   bool         `json:"running"`
	LastCheck *CheckResult `json:"last_check,omitempty"`
	LastState *UpdateState `json:"last_state,omitempty"`
}

// Updater manages controller self-updates.
type Updater struct {
	cfg         *config.SelfUpdateConfig
	gitCfg      *config.GitConfig
	currentVer  string
	dataDir     string
	composePath string // e.g., "/opt/docker/felhom-controller/docker-compose.yml"
	logger      *log.Logger
	debug       bool

	mu            sync.Mutex
	latestVersion string
	lastCheck     *CheckResult
	updateRunning bool
	backupRunning func() bool
}

// NewUpdater creates a new Updater instance.
func NewUpdater(cfg *config.SelfUpdateConfig, gitCfg *config.GitConfig, currentVersion, dataDir, composePath string, logger *log.Logger, debug bool) *Updater {
	return &Updater{
		cfg:         cfg,
		gitCfg:      gitCfg,
		currentVer:  currentVersion,
		dataDir:     dataDir,
		composePath: composePath,
		logger:      logger,
		debug:       debug,
	}
}

func (u *Updater) dbg(format string, args ...interface{}) {
	if u.debug {
		u.logger.Printf("[DEBUG] [selfupdate] "+format, args...)
	}
}

// SetBackupRunningCheck sets the callback to check if a backup is in progress.
func (u *Updater) SetBackupRunningCheck(fn func() bool) {
	u.mu.Lock()
	defer u.mu.Unlock()
	u.backupRunning = fn
}

// IsUpdateRunning returns true if an update is currently in progress.
func (u *Updater) IsUpdateRunning() bool {
	u.mu.Lock()
	defer u.mu.Unlock()
	return u.updateRunning
}

// GetStatus returns the current update status for API/UI.
func (u *Updater) GetStatus() UpdateStatus {
	u.mu.Lock()
	lastCheck := u.lastCheck
	running := u.updateRunning
	u.mu.Unlock()

	state, err := LoadState(u.dataDir)
	if err != nil {
		u.logger.Printf("[WARN] Failed to load update state: %v", err)
	}

	return UpdateStatus{
		Running:   running,
		LastCheck: lastCheck,
		LastState: state,
	}
}

// CheckForUpdate queries the Gitea registry for the latest version tag.
// Caches the result. Thread-safe.
func (u *Updater) CheckForUpdate() CheckResult {
	result := CheckResult{
		CurrentVersion: u.currentVer,
		CheckedAt:      time.Now().UTC().Format(time.RFC3339),
	}

	// Dev version can't check for updates
	currentVer, err := ParseVersion(u.currentVer)
	if err != nil {
		result.Error = "Dev verzió nem ellenőrizhető"
		u.mu.Lock()
		u.lastCheck = &result
		u.mu.Unlock()
		return result
	}

	// Query registry
	latestStr, err := u.queryRegistry()
	if err != nil {
		result.Error = fmt.Sprintf("Registry lekérdezés sikertelen: %v", err)
		u.logger.Printf("[WARN] Registry check failed: %v", err)
		u.mu.Lock()
		u.lastCheck = &result
		u.mu.Unlock()
		return result
	}

	result.LatestVersion = latestStr

	latestVer, err := ParseVersion(latestStr)
	if err != nil {
		result.Error = fmt.Sprintf("Érvénytelen verzió a registry-ben: %s", latestStr)
		u.mu.Lock()
		u.lastCheck = &result
		u.mu.Unlock()
		return result
	}

	cmp := latestVer.Compare(currentVer)
	if cmp > 0 {
		result.UpdateAvailable = true
	}

	u.dbg("version comparison: current=%s (%d.%d.%d), latest=%s (%d.%d.%d), cmp=%d, updateAvailable=%v",
		u.currentVer, currentVer.Major, currentVer.Minor, currentVer.Patch,
		latestStr, latestVer.Major, latestVer.Minor, latestVer.Patch,
		cmp, result.UpdateAvailable)

	u.mu.Lock()
	u.latestVersion = latestStr
	u.lastCheck = &result
	u.mu.Unlock()

	return result
}

// queryRegistry queries the Gitea Docker Registry V2 API for available tags.
// Returns the highest valid semver tag found.
func (u *Updater) queryRegistry() (string, error) {
	if u.gitCfg.Username == "" || u.gitCfg.Token == "" {
		return "", fmt.Errorf("registry hitelesítő adatok hiányoznak")
	}

	// Gitea registry V2: GET /v2/<owner>/<repo>/tags/list
	url := fmt.Sprintf("https://gitea.dooplex.hu/v2/%s/tags/list", registryImagePath(u.cfg.Image))

	u.dbg("queryRegistry: url=%s user=%s", url, u.gitCfg.Username)

	req, err := http.NewRequest("GET", url, nil)
	if err != nil {
		return "", fmt.Errorf("creating request: %w", err)
	}
	req.SetBasicAuth(u.gitCfg.Username, u.gitCfg.Token)

	client := &http.Client{Timeout: 15 * time.Second}
	resp, err := client.Do(req)
	if err != nil {
		u.dbg("queryRegistry: HTTP request failed: %v", err)
		return "", fmt.Errorf("HTTP request failed: %w", err)
	}
	defer resp.Body.Close()
	u.dbg("queryRegistry: HTTP %d", resp.StatusCode)

	if resp.StatusCode == 401 {
		return "", fmt.Errorf("authentication failed (401)")
	}
	if resp.StatusCode != 200 {
		return "", fmt.Errorf("unexpected status: %d", resp.StatusCode)
	}

	var tagsResp struct {
		Name string   `json:"name"`
		Tags []string `json:"tags"`
	}
	if err := json.NewDecoder(resp.Body).Decode(&tagsResp); err != nil {
		return "", fmt.Errorf("decoding response: %w", err)
	}

	u.dbg("queryRegistry: %d tags returned: %v", len(tagsResp.Tags), tagsResp.Tags)

	// Find highest semver tag
	var highest *Version
	for _, tag := range tagsResp.Tags {
		v, err := ParseVersion(tag)
		if err != nil {
			continue // skip non-semver tags ("latest", "dev", etc.)
		}
		if highest == nil || v.Compare(*highest) > 0 {
			highest = &v
		}
	}

	if highest == nil {
		return "", fmt.Errorf("no valid semver tags found")
	}

	return highest.String(), nil
}

// registryImagePath extracts the "owner/repo" from a full image reference.
// e.g., "gitea.dooplex.hu/admin/felhom-controller" → "admin/felhom-controller"
func registryImagePath(image string) string {
	// Remove registry host
	parts := strings.SplitN(image, "/", 2)
	if len(parts) == 2 {
		return parts[1]
	}
	return image
}

// DryRunResult holds the result of a self-update dry run.
type DryRunResult struct {
	CurrentVersion  string `json:"current_version"`
	LatestVersion   string `json:"latest_version"`
	UpdateAvailable bool   `json:"update_available"`
	ComposeWritable bool   `json:"compose_writable"`
	CurrentImageLine string `json:"current_image_line"`
	NewImageLine     string `json:"new_image_line"`
	BackupRunning   bool   `json:"backup_running"`
	Error           string `json:"error,omitempty"`
}

// DryRun checks for updates and reports what would happen without performing any changes.
func (u *Updater) DryRun() *DryRunResult {
	result := &DryRunResult{
		CurrentVersion: u.currentVer,
	}

	// Check for update
	check := u.CheckForUpdate()
	result.LatestVersion = check.LatestVersion
	result.UpdateAvailable = check.UpdateAvailable
	if check.Error != "" {
		result.Error = check.Error
		return result
	}

	// Check compose file
	data, err := os.ReadFile(u.composePath)
	if err != nil {
		result.Error = fmt.Sprintf("Compose fájl nem olvasható: %v", err)
		return result
	}

	// Find current image line
	re := regexp.MustCompile(`(image:\s*)gitea\.dooplex\.hu/admin/felhom-controller:\S+`)
	match := re.Find(data)
	if match != nil {
		result.CurrentImageLine = string(match)
	}

	// Build new image line
	if check.UpdateAvailable {
		result.NewImageLine = fmt.Sprintf("image: %s:%s", u.cfg.Image, check.LatestVersion)
	}

	// Check writability
	f, err := os.OpenFile(u.composePath, os.O_WRONLY, 0)
	if err == nil {
		f.Close()
		result.ComposeWritable = true
	}

	// Check backup running
	if u.backupRunning != nil {
		result.BackupRunning = u.backupRunning()
	}

	return result
}

// TriggerUpdate starts the self-update process. Returns error immediately if
// preconditions fail. The actual update runs in a goroutine.
func (u *Updater) TriggerUpdate(initiatedBy string) error {
	u.dbg("TriggerUpdate: initiatedBy=%s currentVer=%s", initiatedBy, u.currentVer)
	u.mu.Lock()
	if u.updateRunning {
		u.mu.Unlock()
		u.dbg("TriggerUpdate: rejected — update already running")
		return fmt.Errorf("Frissítés már folyamatban")
	}

	// Dev version check
	if _, err := ParseVersion(u.currentVer); err != nil {
		u.mu.Unlock()
		return fmt.Errorf("Dev verzió nem frissíthető")
	}

	// Backup running check
	if u.backupRunning != nil && u.backupRunning() {
		u.mu.Unlock()
		return fmt.Errorf("Mentés fut, próbálja később")
	}

	// Compose file accessible check
	if _, err := os.Stat(u.composePath); err != nil {
		u.mu.Unlock()
		return fmt.Errorf("docker-compose.yml nem elérhető: %w", err)
	}

	u.updateRunning = true
	u.mu.Unlock()

	// Check for update (or use cached)
	result := u.CheckForUpdate()
	if !result.UpdateAvailable {
		u.mu.Lock()
		u.updateRunning = false
		u.mu.Unlock()
		return fmt.Errorf("Nincs elérhető frissítés")
	}

	targetVersion := result.LatestVersion
	targetImage := fmt.Sprintf("%s:%s", u.cfg.Image, targetVersion)
	previousImage := fmt.Sprintf("%s:%s", u.cfg.Image, u.currentVer)

	u.logger.Printf("[INFO] Starting self-update: %s → %s (initiated by: %s)", u.currentVer, targetVersion, initiatedBy)
	u.dbg("TriggerUpdate: target=%s image=%s previousImage=%s", targetVersion, targetImage, previousImage)

	go u.performUpdate(targetVersion, targetImage, previousImage, initiatedBy)

	return nil
}

// performUpdate runs the actual update steps in a goroutine.
func (u *Updater) performUpdate(targetVersion, targetImage, previousImage, initiatedBy string) {
	defer func() {
		u.mu.Lock()
		u.updateRunning = false
		u.mu.Unlock()
	}()

	u.dbg("performUpdate: starting — target=%s image=%s", targetVersion, targetImage)
	// 1. Write pending state
	state := &UpdateState{
		Status:          "pending",
		PreviousVersion: u.currentVer,
		PreviousImage:   previousImage,
		TargetVersion:   targetVersion,
		TargetImage:     targetImage,
		InitiatedAt:     time.Now().UTC().Format(time.RFC3339),
		InitiatedBy:     initiatedBy,
	}
	if err := SaveState(u.dataDir, state); err != nil {
		u.logger.Printf("[ERROR] Failed to save update state: %v", err)
		return
	}

	// 2. Docker pull
	u.dbg("performUpdate: step 2 — docker pull %s", targetImage)
	u.logger.Printf("[INFO] Pulling image: %s", targetImage)
	pullStart := time.Now()
	pullOut, pullErr := runCommand("docker", "pull", targetImage)
	if pullErr != nil {
		state.Status = "failed"
		state.Error = fmt.Sprintf("docker pull failed: %v — %s", pullErr, pullOut)
		state.CompletedAt = time.Now().UTC().Format(time.RFC3339)
		SaveState(u.dataDir, state)
		u.logger.Printf("[ERROR] Docker pull failed: %v — %s", pullErr, pullOut)
		return
	}
	u.logger.Printf("[INFO] Image pulled successfully: %s", targetImage)
	u.dbg("performUpdate: docker pull completed in %s", time.Since(pullStart).Round(time.Millisecond))

	// 3. Update compose file (replace image tag)
	u.dbg("performUpdate: step 3 — updating compose file %s", u.composePath)
	if err := u.updateComposeFile(targetImage); err != nil {
		state.Status = "failed"
		state.Error = fmt.Sprintf("compose update failed: %v", err)
		state.CompletedAt = time.Now().UTC().Format(time.RFC3339)
		SaveState(u.dataDir, state)
		u.logger.Printf("[ERROR] Compose file update failed: %v", err)
		return
	}
	u.logger.Printf("[INFO] Compose file updated with new image: %s", targetImage)

	// 4. Docker compose up -d (this kills the current container)
	u.dbg("performUpdate: step 4 — docker compose up -d")
	u.logger.Printf("[INFO] Running docker compose up -d — container will restart")
	composeDir := strings.TrimSuffix(u.composePath, "/docker-compose.yml")
	upOut, upErr := runCommand("docker", "compose", "-f", u.composePath, "-p", "felhom-controller", "up", "-d")
	if upErr != nil {
		state.Status = "failed"
		state.Error = fmt.Sprintf("docker compose up -d failed: %v — %s", upErr, upOut)
		state.CompletedAt = time.Now().UTC().Format(time.RFC3339)
		SaveState(u.dataDir, state)
		u.logger.Printf("[ERROR] docker compose up -d failed: %v — %s (dir: %s)", upErr, upOut, composeDir)
		return
	}

	// If we're still alive after compose up -d, log it.
	// Normally this process should be killed when Docker replaces the container.
	u.logger.Printf("[WARN] Still running after docker compose up -d — expected to be replaced")
	time.Sleep(30 * time.Second)
	u.logger.Printf("[WARN] Still alive 30s after docker compose up -d")
}

// updateComposeFile reads the compose file, replaces the image tag, and writes it back atomically.
func (u *Updater) updateComposeFile(newImage string) error {
	data, err := os.ReadFile(u.composePath)
	if err != nil {
		return fmt.Errorf("reading compose file: %w", err)
	}

	// Replace image line: "image: gitea.dooplex.hu/admin/felhom-controller:..." → new image
	re := regexp.MustCompile(`(image:\s*)gitea\.dooplex\.hu/admin/felhom-controller:\S+`)

	// Log old image line for debugging
	oldMatch := re.Find(data)
	if oldMatch != nil {
		u.dbg("updateComposeFile: %q → %q", string(oldMatch), "image: "+newImage)
	} else {
		u.dbg("updateComposeFile: no matching image line found in %s", u.composePath)
	}

	newData := re.ReplaceAll(data, []byte("${1}"+newImage))

	if bytes.Equal(data, newData) {
		return fmt.Errorf("no image line found to replace in compose file")
	}

	// Atomic write: write to .tmp, then rename
	tmpPath := u.composePath + ".tmp"
	if err := os.WriteFile(tmpPath, newData, 0644); err != nil {
		return fmt.Errorf("writing temp compose file: %w", err)
	}
	if err := os.Rename(tmpPath, u.composePath); err != nil {
		return fmt.Errorf("renaming compose file: %w", err)
	}

	return nil
}

// VerifyStartup checks the update state file on startup.
// Called once from main.go before the scheduler starts.
// Returns the state if a pending update was detected, nil otherwise.
func (u *Updater) VerifyStartup() *UpdateState {
	u.dbg("VerifyStartup: checking update state in %s", u.dataDir)
	state, err := LoadState(u.dataDir)
	if err != nil {
		u.logger.Printf("[WARN] Failed to load update state on startup: %v — clearing", err)
		ClearState(u.dataDir, u.logger)
		return nil
	}
	if state == nil || state.Status != "pending" {
		u.dbg("VerifyStartup: no pending update (state=%v)", state)
		return nil
	}
	u.dbg("VerifyStartup: pending update found — target=%s previous=%s", state.TargetVersion, state.PreviousVersion)

	// Compare current version with target
	currentVer, curErr := ParseVersion(u.currentVer)
	targetVer, tgtErr := ParseVersion(state.TargetVersion)

	if curErr != nil || tgtErr != nil {
		state.Status = "failed"
		state.Error = "Version parse error on startup verification"
		state.CompletedAt = time.Now().UTC().Format(time.RFC3339)
		SaveState(u.dataDir, state)
		u.logger.Printf("[WARN] Post-update startup: version parse error (current=%s, target=%s)", u.currentVer, state.TargetVersion)
		return state
	}

	if currentVer.Compare(targetVer) == 0 {
		// Success — we're running the target version
		state.Status = "success"
		state.CompletedAt = time.Now().UTC().Format(time.RFC3339)
		SaveState(u.dataDir, state)
		u.logger.Printf("[INFO] Post-update startup: update successful (%s → %s)", state.PreviousVersion, state.TargetVersion)
	} else {
		// Version mismatch — update may have failed
		state.Status = "failed"
		state.Error = fmt.Sprintf("Version mismatch: expected %s, running %s", state.TargetVersion, u.currentVer)
		state.CompletedAt = time.Now().UTC().Format(time.RFC3339)
		SaveState(u.dataDir, state)
		u.logger.Printf("[WARN] Post-update startup: version mismatch (expected %s, running %s)", state.TargetVersion, u.currentVer)
	}

	return state
}

// runCommand executes a command and returns combined stdout+stderr and error.
func runCommand(name string, args ...string) (string, error) {
	cmd := exec.Command(name, args...)
	var out bytes.Buffer
	cmd.Stdout = &out
	cmd.Stderr = &out
	err := cmd.Run()
	return out.String(), err
}