dyzur-bot/service.go

package dyzurbot

import (
	"context"
	"errors"
	"fmt"
	"log/slog"
	"os"
	"strconv"
	"strings"
	"time"

	"github.com/joho/godotenv"
)

// Config holds the cache/sync settings, sourced from the environment/.env.
type Config struct {
	SpreadsheetID string
	SheetRange    string
	CachePath     string
	SyncInterval  time.Duration
	// Reminder is nil when GROUP_CHAT_ID is unset (reminders disabled).
	Reminder *ReminderConfig
}

// reminderTimezone is the fixed zone the reminder schedule is expressed in, so
// REMINDER_HOUR means Polish local time regardless of the host/container TZ.
const reminderTimezone = "Europe/Warsaw"

// LoadConfig reads configuration from the environment (and .env if present),
// applying defaults for everything except the required spreadsheet ID.
func LoadConfig() (Config, error) {
	_ = godotenv.Load()

	// SheetRange is left empty when unset so the fetcher can resolve the
	// current-year tab at fetch time (see resolveRange). Freezing the default
	// here would pin a long-running process to its start year and silently miss
	// the next year's tab after the New Year rollover.
	cfg := Config{
		SpreadsheetID: os.Getenv("GOOGLE_SHEETS_ID"),
		SheetRange:    os.Getenv("SHEET_RANGE"),
		CachePath:     getenvDefault("ROTA_CACHE_PATH", "rota.json"),
	}
	if cfg.SpreadsheetID == "" {
		return Config{}, fmt.Errorf("missing GOOGLE_SHEETS_ID in environment/.env")
	}

	// Fail fast at startup if the service-account key is missing or unreadable,
	// rather than on the first sheet fetch minutes later. ReadSheet consumes the
	// path directly from the environment.
	credPath := os.Getenv("GOOGLE_CREDENTIALS_FILE")
	if credPath == "" {
		return Config{}, fmt.Errorf("missing GOOGLE_CREDENTIALS_FILE in environment/.env")
	}
	if _, err := os.Stat(credPath); err != nil { // #nosec G703 -- credPath is operator-controlled config (env/.env)
		return Config{}, fmt.Errorf("GOOGLE_CREDENTIALS_FILE %q: %w", credPath, err)
	}

	raw := getenvDefault("SYNC_INTERVAL", "30m")
	interval, err := time.ParseDuration(raw)
	if err != nil {
		return Config{}, fmt.Errorf("invalid SYNC_INTERVAL %q: %w", raw, err)
	}
	if interval <= 0 {
		return Config{}, fmt.Errorf("SYNC_INTERVAL must be positive, got %v", interval)
	}
	cfg.SyncInterval = interval

	reminder, err := loadReminderConfig()
	if err != nil {
		return Config{}, err
	}
	cfg.Reminder = reminder

	return cfg, nil
}

// loadReminderConfig parses the reminder schedule from the environment. It
// returns nil (reminders disabled) when GROUP_CHAT_ID is unset, and an error
// for any malformed value when it is set.
func loadReminderConfig() (*ReminderConfig, error) {
	raw := os.Getenv("GROUP_CHAT_ID")
	if raw == "" {
		return nil, nil
	}
	chatID, err := strconv.ParseInt(raw, 10, 64)
	if err != nil {
		return nil, fmt.Errorf("invalid GROUP_CHAT_ID %q: %w", raw, err)
	}

	weekday, err := parseWeekday(getenvDefault("REMINDER_WEEKDAY", "Monday"))
	if err != nil {
		return nil, fmt.Errorf("invalid REMINDER_WEEKDAY: %w", err)
	}

	hourRaw := getenvDefault("REMINDER_HOUR", "9")
	hour, err := strconv.Atoi(hourRaw)
	if err != nil {
		return nil, fmt.Errorf("invalid REMINDER_HOUR %q: %w", hourRaw, err)
	}
	if hour < 0 || hour > 23 {
		return nil, fmt.Errorf("REMINDER_HOUR must be 0–23, got %d", hour)
	}

	loc, err := time.LoadLocation(reminderTimezone)
	if err != nil {
		return nil, fmt.Errorf("load timezone %q: %w", reminderTimezone, err)
	}

	return &ReminderConfig{
		ChatID:  chatID,
		Weekday: weekday,
		Hour:    hour,
		Loc:     loc,
	}, nil
}

func getenvDefault(key, fallback string) string {
	if v := os.Getenv(key); v != "" {
		return v
	}
	return fallback
}

// FetchFunc retrieves and parses the current rota from upstream (the Sheet).
type FetchFunc func(ctx context.Context) (Rota, error)

// errEmptyRota signals that a fetch succeeded but returned no duty weeks, so the
// cache was deliberately left untouched rather than clobbered with empty data.
var errEmptyRota = errors.New("fetched rota has no duty weeks; keeping cached data")

// defaultStaleAfter is how long the cache may go without a successful sync
// before Sync escalates from a per-failure warning to a stale-cache error. At
// the default 30m sync interval this is ~12 consecutive failures.
const defaultStaleAfter = 6 * time.Hour

// Syncer refreshes a Store from an upstream fetch.
type Syncer struct {
	store      *Store
	fetch      FetchFunc
	now        func() time.Time
	staleAfter time.Duration
}

// NewSyncer wires a Store to a fetch function.
func NewSyncer(store *Store, fetch FetchFunc) *Syncer {
	return &Syncer{store: store, fetch: fetch, now: time.Now, staleAfter: defaultStaleAfter}
}

// Sync fetches once. On success with a non-empty rota it replaces the cache.
// When the fetch fails — or succeeds but returns zero weeks (e.g. a freshly
// created, not-yet-filled year tab) — it leaves the last-good cache in place,
// warns, and returns an error so a caller (e.g. a manual /sync) can report it.
// A guard against the empty case matters because the range targets the current
// year's tab, which around New Year may exist but be empty; replacing the cache
// with it would silently wipe the schedule.
func (sy *Syncer) Sync(ctx context.Context) error {
	rota, err := sy.fetch(ctx)
	if err != nil {
		slog.Warn("sync fetch failed, keeping cached data", "err", err)
		sy.warnIfStale("fetch error")
		return err
	}
	if len(rota.Weeks) == 0 {
		slog.Warn("sync returned no duty weeks, keeping cached data")
		sy.warnIfStale("empty rota")
		return errEmptyRota
	}
	return sy.store.Replace(rota, sy.now())
}

// warnIfStale emits a loud error when the cache has gone too long without a
// successful sync, turning a silently-stale cache (which is not a crash, so
// process-level supervision won't catch it) into a visible log signal.
func (sy *Syncer) warnIfStale(reason string) {
	last := sy.store.LastSynced()
	if age, stale := staleAge(last, sy.now(), sy.staleAfter); stale {
		slog.Error("rota cache is stale; sync has not succeeded recently",
			"age", age.Round(time.Minute), "last_synced", last.Format(time.RFC3339), "reason", reason)
	}
}

// staleAge reports the cache's age and whether it exceeds threshold. A
// never-synced cache (zero last) is not "stale" — there is no good data to be
// behind — so it reports false.
func staleAge(last, now time.Time, threshold time.Duration) (time.Duration, bool) {
	if last.IsZero() {
		return 0, false
	}
	age := now.Sub(last)
	return age, age >= threshold
}

// Run polls Sync on the given interval until ctx is canceled.
func (sy *Syncer) Run(ctx context.Context, interval time.Duration) {
	ticker := time.NewTicker(interval)
	defer ticker.Stop()
	for {
		select {
		case <-ctx.Done():
			return
		case <-ticker.C:
			_ = sy.Sync(ctx)
		}
	}
}

// fetchTimeout bounds a single Sheets fetch so a hung HTTP call can't block a
// caller (e.g. a /sync command) indefinitely — it would otherwise wait until
// the parent context is canceled at shutdown.
const fetchTimeout = 30 * time.Second

// SheetFetcher builds a FetchFunc that reads the rota from Google Sheets. When
// readRange is empty the range is resolved per fetch (see resolveRange), so a
// long-running process follows the current-year tab across a New Year rollover
// instead of pinning the year it started in.
func SheetFetcher(spreadsheetID, readRange string) FetchFunc {
	return func(ctx context.Context) (Rota, error) {
		ctx, cancel := context.WithTimeout(ctx, fetchTimeout)
		defer cancel()
		rng := resolveRange(readRange, time.Now())
		rows, err := ReadSheet(ctx, spreadsheetID, rng)
		if err != nil {
			return Rota{}, err
		}
		return stampTabs(ParseRota(rows), rng), nil
	}
}

// stampTabs records on each week the tab its rows were fetched from, so an
// assignment written later (possibly after the clock-resolved tab has rolled to
// the next year while the cache still holds this fetch) targets the tab the
// week's Row is valid on.
func stampTabs(rota Rota, resolvedRange string) Rota {
	tab, _, _ := strings.Cut(resolvedRange, "!")
	for i := range rota.Weeks {
		rota.Weeks[i].Tab = tab
	}
	return rota
}

// SheetAssigner returns an AssignFunc that writes a week's two names into its
// B:C cells on the same tab the rota is read from. sheetRange mirrors the
// fetcher's range so the tab (year or explicit) matches.
func SheetAssigner(spreadsheetID, sheetRange string) AssignFunc {
	return func(ctx context.Context, week DutyWeek, p1, p2 string) error {
		ctx, cancel := context.WithTimeout(ctx, fetchTimeout)
		defer cancel()
		rng := assignRangeFor(week, sheetRange, time.Now())
		return WriteSheet(ctx, spreadsheetID, rng, [][]interface{}{{p1, p2}})
	}
}

// assignRangeFor builds the A1 range for a week's assignment cells (Osoba 1/2
// are columns B and C). An explicit sheetRange keeps its tab verbatim, matching
// reads. Otherwise the week's fetch-time tab is preferred over clock resolution:
// around New Year the cache can hold the old tab's rows (the empty-rota guard
// keeps them while the next year's tab is unfilled) after resolveRange has moved
// to the new tab, and writing an old-tab Row onto the new tab would corrupt it.
// A week without a recorded tab (cache from an older version) falls back to the
// clock, preserving prior behavior.
func assignRangeFor(week DutyWeek, sheetRange string, now time.Time) string {
	tab := week.Tab
	if sheetRange != "" || tab == "" {
		tab, _, _ = strings.Cut(resolveRange(sheetRange, now), "!")
	}
	return fmt.Sprintf("%s!B%d:C%d", tab, week.Row, week.Row)
}

// resolveRange returns the explicit range when set, otherwise the default that
// targets the current year's tab (sheets are named per year, e.g. "2026").
// Resolving against now (not a startup-captured value) is what lets the range
// advance to the next year's tab automatically. The 200-row cap bounds the
// fetch; rows past it are silently dropped, so it must stay comfortably above
// one year of weekly rows (~53) plus roster entries.
func resolveRange(explicit string, now time.Time) string {
	if explicit != "" {
		return explicit
	}
	return fmt.Sprintf("%d!A1:G200", now.Year())
}

// Service ties the Store and Syncer together for the bot to consume.
type Service struct {
	Store    *Store
	Syncer   *Syncer
	interval time.Duration
}

// NewService builds the cache + syncer from config.
func NewService(cfg Config) *Service {
	store := NewStore(cfg.CachePath)
	syncer := NewSyncer(store, SheetFetcher(cfg.SpreadsheetID, cfg.SheetRange))
	return &Service{Store: store, Syncer: syncer, interval: cfg.SyncInterval}
}

// Start loads the cached rota, runs one best-effort sync in the background
// (a failure leaves the cached/empty data in place — offline-tolerant), then
// polls in the background until ctx is canceled.
func (s *Service) Start(ctx context.Context) error {
	if err := s.Store.Load(); err != nil {
		return err
	}
	go func() { _ = s.Syncer.Sync(ctx) }()
	go s.Syncer.Run(ctx, s.interval)
	return nil
}