Files
paliad/cmd/server/main.go
m d2790a0461 feat(paliadin): reconcile late responses via janitor + chat polling
When Claude writes the response file after the 60 s pollForResponse
window expires (e.g. the tmux pane was busy mid-turn when the message
arrived), the SSE stream has already closed with an error and the
file sits unread on disk forever. The chat shows a permanent timeout
even though the answer exists.

Backend:
- LocalPaliadinService.StartJanitor: scans responseDir every 2 s and
  patches rows whose response is still NULL when the file lands.
  completeTurnLate stamps error_code='late' so the FE can render a
  marker. Guarded with WHERE response IS NULL to never overwrite a
  real response if RunTurn races.
- Paliadin.GetTurn(callerID, turnID) on the shared paliadinDB. Same
  visibility predicate as ListRecentTurns.
- GET /api/paliadin/turns/{id} — owner-gated; lets the chat UI
  discover late-arrived responses without a refresh.

Frontend:
- paliadin-late-poll.ts: shared 3 s / 10 min poller.
- paliadin.ts + paliadin-widget.ts: on SSE error, show
  "wartet auf späte Antwort", kick off the poller, swap bubble in
  place when response arrives + retroactively persist to history.
- i18n: paliadin.late.waiting + paliadin.late.marker (DE/EN).
- CSS: --late-pending opacity tweak, --late neutral background,
  italic-grey "verspätet" tag.
2026-05-08 20:56:53 +02:00

376 lines
15 KiB
Go

package main
import (
"context"
"encoding/base64"
"fmt"
"log"
"net/http"
"os"
"os/exec"
"os/signal"
"strconv"
"strings"
"syscall"
// Embed Go's IANA tz database into the binary so time.LoadLocation works
// without OS tzdata. The runtime image (alpine) doesn't ship /usr/share/
// zoneinfo — without this import, every reminder timezone lookup fails
// silently and the hourly reminder slot fires in UTC instead of the
// user's chosen tz (t-paliad-064 root cause). Adds ~450KB to the binary.
_ "time/tzdata"
"mgit.msbls.de/m/paliad/internal/auth"
"mgit.msbls.de/m/paliad/internal/branding"
"mgit.msbls.de/m/paliad/internal/db"
"mgit.msbls.de/m/paliad/internal/handlers"
"mgit.msbls.de/m/paliad/internal/services"
)
func main() {
port := os.Getenv("PORT")
if port == "" {
port = "8080"
}
// Surface the firm name in the boot log so a deployer can confirm
// FIRM_NAME took effect without curl-ing a rendered page.
log.Printf("branding: firm=%q (override with FIRM_NAME)", branding.Name)
supabaseURL := os.Getenv("SUPABASE_URL")
supabaseAnonKey := os.Getenv("SUPABASE_ANON_KEY")
if supabaseURL == "" || supabaseAnonKey == "" {
log.Fatal("SUPABASE_URL and SUPABASE_ANON_KEY must be set")
}
jwtSecret := os.Getenv("SUPABASE_JWT_SECRET")
if jwtSecret == "" {
log.Fatal("SUPABASE_JWT_SECRET must be set — session cookies cannot be trusted without signature verification")
}
client := auth.NewClient(supabaseURL, supabaseAnonKey, []byte(jwtSecret))
giteaToken := os.Getenv("GITEA_TOKEN")
if giteaToken == "" {
log.Println("GITEA_TOKEN not set — file proxy will not be able to access private repos")
}
// MailService is wired regardless of DB availability — it no-ops when
// SMTP env vars are unset, so the server stays runnable for knowledge-
// platform-only deployments. Template-parse errors at boot are fatal.
mailSvc, err := services.NewMailService()
if err != nil {
log.Fatalf("mail service init: %v", err)
}
// Shared context for background goroutines (CalDAV sync + reminder job).
bgCtx, stop := signal.NotifyContext(context.Background(), syscall.SIGINT, syscall.SIGTERM)
defer stop()
// DATABASE_URL is optional during the Phase A → Phase D transition. The
// existing knowledge-platform features (Kostenrechner, Glossar, etc.) work
// without a DB. matter-management endpoints return 503 until DATABASE_URL is set.
dbURL := os.Getenv("DATABASE_URL")
var svcBundle *handlers.Services
var caldavSvc *services.CalDAVService
if dbURL != "" {
log.Println("applying database migrations…")
if err := db.ApplyMigrations(dbURL); err != nil {
log.Fatalf("migration failed: %v", err)
}
log.Println("database migrations applied")
pool, err := db.OpenPool(dbURL)
if err != nil {
log.Fatalf("open db pool: %v", err)
}
// Refresh paliad.deadline_search whenever migrations run, so
// search reflects any newly-seeded rule / concept / trigger.
// Migration 047 created the matview already-populated; this
// is only a no-op for the boot that introduced it. CONCURRENTLY
// keeps reads online and stays well under 100 ms at < 1k rows.
if err := services.RefreshSearchView(bgCtx, pool); err != nil {
log.Printf("refresh deadline_search: %v", err)
}
holidays := services.NewHolidayService(pool)
courts := services.NewCourtService(pool)
users := services.NewUserService(pool)
projectSvc := services.NewProjectService(pool, users)
teamSvc := services.NewTeamService(pool, projectSvc)
partnerUnitSvc := services.NewPartnerUnitService(pool, users)
rules := services.NewDeadlineRuleService(pool)
// Phase F: optional CalDAV cipher. If CALDAV_ENCRYPTION_KEY is unset
// the service exists but Enabled() reports false; handlers return 501.
// If the env var is malformed, fail fast — silently skipping would
// leave plaintext-credential bugs hidden.
cipher, err := services.LoadCalDAVCipher()
if err != nil {
log.Fatalf("CALDAV_ENCRYPTION_KEY: %v", err)
}
if cipher == nil {
log.Println("CALDAV_ENCRYPTION_KEY not set — CalDAV endpoints will return 501")
} else {
log.Println("CalDAV encryption configured (AES-256-GCM)")
}
appointmentSvc := services.NewAppointmentService(pool, projectSvc)
caldavSvc = services.NewCalDAVService(pool, cipher, appointmentSvc)
// Wire the push hook so user-driven mutations sync to the external
// calendar without waiting for the next 60-second tick.
appointmentSvc.SetCalDAVPusher(caldavSvc)
baseURL := os.Getenv("PALIAD_BASE_URL")
inviteSvc := services.NewInviteService(pool, mailSvc, handlers.AllowedEmailDomains, baseURL)
reminderSvc := services.NewReminderService(pool, mailSvc, users, baseURL)
// Wire EmailTemplateService onto the MailService so DB-backed admin
// edits propagate without a process restart. The constructor is split
// from MailService creation because the DB pool isn't available yet
// at the point we build mailSvc above.
emailTemplateSvc := services.NewEmailTemplateService(pool)
mailSvc.SetTemplateService(emailTemplateSvc)
eventTypeSvc := services.NewEventTypeService(pool, users)
deadlineSvc := services.NewDeadlineService(pool, projectSvc, eventTypeSvc)
svcBundle = &handlers.Services{
Project: projectSvc,
Team: teamSvc,
PartnerUnit: partnerUnitSvc,
Party: services.NewPartyService(pool, projectSvc),
Deadline: deadlineSvc,
Appointment: appointmentSvc,
CalDAV: caldavSvc,
Rules: rules,
Calculator: services.NewDeadlineCalculator(holidays),
Users: users,
Fristenrechner: services.NewFristenrechnerService(rules, holidays, courts),
EventDeadline: services.NewEventDeadlineService(pool, services.NewDeadlineCalculator(holidays), holidays, courts),
Courts: courts,
DeadlineSearch: services.NewDeadlineSearchService(pool),
EventCategory: nil, // wired below; cross-link order matters
EventType: eventTypeSvc,
Dashboard: services.NewDashboardService(pool, users),
Note: services.NewNoteService(pool, projectSvc, appointmentSvc),
ChecklistInst: services.NewChecklistInstanceService(pool, projectSvc),
Mail: mailSvc,
Invite: inviteSvc,
Agenda: services.NewAgendaService(pool, users, eventTypeSvc),
Audit: services.NewAuditService(pool),
EmailTemplate: emailTemplateSvc,
Link: services.NewLinkService(pool),
Event: services.NewEventService(pool, deadlineSvc, appointmentSvc),
Approval: services.NewApprovalService(pool, users),
Derivation: services.NewDerivationService(pool, projectSvc, partnerUnitSvc),
UserView: services.NewUserViewService(pool),
Broadcast: services.NewBroadcastService(pool, mailSvc, users, teamSvc, emailTemplateSvc),
Pin: services.NewPinService(pool, projectSvc),
CardLayout: services.NewCardLayoutService(pool),
}
// Paliadin backend selection (t-paliad-146 + t-paliad-151):
// PALIADIN_REMOTE_HOST set → RemotePaliadinService (ssh to mRiver)
// else: local tmux available → LocalPaliadinService (PoC path)
// else: DisabledPaliadinService (handlers still 404 for non-owners
// via the gate; for m, RunTurn returns ErrPaliadinDisabled
// which surfaces as a friendly error).
//
// All three implement services.Paliadin; the per-request handler
// gate (requirePaliadinOwner) is unchanged and applies to every
// backend.
if remoteHost := os.Getenv("PALIADIN_REMOTE_HOST"); remoteHost != "" {
cfg, err := buildPaliadinRemoteConfig(remoteHost)
if err != nil {
log.Fatalf("paliadin: remote config: %v", err)
}
svcBundle.Paliadin = services.NewRemotePaliadinService(pool, users, cfg)
log.Printf("paliadin: remote mode → ssh %s@%s:%d (owner=%s)",
cfg.SSHUser, cfg.SSHHost, cfg.SSHPort, services.PaliadinOwnerEmail)
} else if _, err := exec.LookPath("tmux"); err == nil {
sessionPrefix := os.Getenv("PALIADIN_SESSION_PREFIX")
responseDir := os.Getenv("PALIADIN_RESPONSE_DIR")
local := services.NewLocalPaliadinService(pool, users, sessionPrefix, responseDir)
// Late-response janitor — patches rows when Claude writes the
// response file after the 60 s pollForResponse window expires.
// Runs for the process lifetime; cleaned up when bgCtx
// cancels on SIGTERM.
local.StartJanitor(bgCtx)
svcBundle.Paliadin = local
log.Printf("paliadin: local tmux mode (owner=%s, janitor=on)", services.PaliadinOwnerEmail)
} else {
svcBundle.Paliadin = services.NewDisabledPaliadinService(pool, users)
log.Printf("paliadin: disabled (no PALIADIN_REMOTE_HOST, no local tmux; owner=%s)",
services.PaliadinOwnerEmail)
}
// Wire ApprovalService into the entity services so Create / Update /
// Complete / Delete consult paliad.approval_policies (t-paliad-138).
// Without this wiring, the policies and request tables exist but no
// mutation path consults them — paliad behaves as before.
deadlineSvc.SetApprovalService(svcBundle.Approval)
appointmentSvc.SetApprovalService(svcBundle.Approval)
// v3 (t-paliad-133): wire EventCategoryService and cross-link
// it into DeadlineSearchService so ?event_category_slug= can
// resolve to a concept-id allow-list during search.
eventCategorySvc := services.NewEventCategoryService(pool)
svcBundle.EventCategory = eventCategorySvc
svcBundle.DeadlineSearch.SetEventCategoryService(eventCategorySvc)
log.Println("Phase B services initialised")
// Spawn background goroutines: CalDAV sync (one per enabled user)
// and the hourly reminder scanner. Both live for the process
// lifetime; the signal-scoped context cleans them up on SIGTERM.
if err := caldavSvc.Start(bgCtx); err != nil {
log.Printf("CalDAV start: %v", err)
}
reminderSvc.Start(bgCtx)
go func() {
<-bgCtx.Done()
log.Println("background services: shutdown signal received")
caldavSvc.Stop()
}()
} else {
log.Println("DATABASE_URL not set — matter-management endpoints will return 503")
}
mux := http.NewServeMux()
handlers.Register(mux, client, giteaToken, svcBundle)
log.Printf("paliad server starting on :%s", port)
if err := http.ListenAndServe(":"+port, mux); err != nil {
log.Fatal(err)
}
}
// buildPaliadinRemoteConfig assembles a RemotePaliadinConfig from
// environment variables, materialising the SSH private key and
// known_hosts blobs into chmod-600/644 tmpfiles for OpenSSH to read.
//
// The blobs travel as Dokploy secrets (multi-line env vars). We never
// persist them to disk — tmpfiles live for the process lifetime in
// /tmp and disappear on container restart. Re-creating them every boot
// is fine; the keys themselves rotate independently via Dokploy
// secret updates.
//
// Required: PALIADIN_REMOTE_HOST, PALIADIN_SSH_PRIVATE_KEY, PALIADIN_KNOWN_HOSTS.
// Optional: PALIADIN_REMOTE_USER (default "m"), PALIADIN_REMOTE_PORT
// (default 22022 — bypasses Tailscale SSH on :22, see design §4.5).
func buildPaliadinRemoteConfig(host string) (services.RemotePaliadinConfig, error) {
cfg := services.RemotePaliadinConfig{
SSHHost: host,
SSHUser: cmpOr(os.Getenv("PALIADIN_REMOTE_USER"), "m"),
SSHPort: 22022,
SessionPrefix: os.Getenv("PALIADIN_SESSION_PREFIX"),
}
if p := os.Getenv("PALIADIN_REMOTE_PORT"); p != "" {
n, err := strconv.Atoi(p)
if err != nil || n <= 0 || n > 65535 {
return cfg, fmt.Errorf("PALIADIN_REMOTE_PORT %q: not a valid port", p)
}
cfg.SSHPort = n
}
// Dokploy stores compose env vars in a single-line .env file: multi-line
// PEM bodies get truncated to the first line. Base64-encode the
// private key in the secret to survive that round-trip; here we
// detect base64 vs raw PEM and decode either way.
keyBlob, err := decodePaliadinPrivateKey(os.Getenv("PALIADIN_SSH_PRIVATE_KEY"))
if err != nil {
return cfg, fmt.Errorf("PALIADIN_SSH_PRIVATE_KEY: %w", err)
}
keyPath, err := writeSecretFile("paliadin-id_ed25519-", keyBlob, 0o600)
if err != nil {
return cfg, fmt.Errorf("PALIADIN_SSH_PRIVATE_KEY: %w", err)
}
if keyPath == "" {
return cfg, fmt.Errorf("PALIADIN_REMOTE_HOST set but PALIADIN_SSH_PRIVATE_KEY empty")
}
cfg.SSHKeyPath = keyPath
knownHostsPath, err := writeSecretFile("paliadin-known_hosts-", os.Getenv("PALIADIN_KNOWN_HOSTS"), 0o644)
if err != nil {
return cfg, fmt.Errorf("PALIADIN_KNOWN_HOSTS: %w", err)
}
if knownHostsPath == "" {
return cfg, fmt.Errorf("PALIADIN_REMOTE_HOST set but PALIADIN_KNOWN_HOSTS empty")
}
cfg.KnownHostsPath = knownHostsPath
return cfg, nil
}
// decodePaliadinPrivateKey accepts either a raw PEM (multi-line) or a
// base64-encoded PEM. Returns the raw PEM bytes ready to write to a
// keyfile. Empty input → ("", nil) so the caller can distinguish
// "secret not set" from "decode failed".
//
// Why base64: Dokploy stores compose env vars in a one-line-per-key
// .env file, which silently truncates multi-line values to their first
// line. Empirically, a multi-line `-----BEGIN OPENSSH PRIVATE KEY-----`
// arrived inside the container as just the BEGIN header (36 bytes).
// Base64-encoding the key in the Dokploy secret survives that
// round-trip. We still accept raw PEM for local-dev convenience.
func decodePaliadinPrivateKey(blob string) (string, error) {
blob = strings.TrimSpace(blob)
if blob == "" {
return "", nil
}
// Raw PEM: starts with ----- and contains a newline. Use as-is.
if strings.HasPrefix(blob, "-----") && strings.Contains(blob, "\n") {
return blob + "\n", nil
}
// Otherwise treat as base64. Strip any whitespace OpenSSH keygen
// helpers might insert (line breaks every 64 chars in some tools).
clean := strings.Map(func(r rune) rune {
if r == ' ' || r == '\n' || r == '\r' || r == '\t' {
return -1
}
return r
}, blob)
decoded, err := base64.StdEncoding.DecodeString(clean)
if err != nil {
return "", fmt.Errorf("not raw PEM (no newline) and base64 decode failed: %w", err)
}
out := string(decoded)
if !strings.HasPrefix(out, "-----BEGIN") {
return "", fmt.Errorf("decoded body does not look like a PEM key (no -----BEGIN prefix)")
}
if !strings.HasSuffix(out, "\n") {
out += "\n"
}
return out, nil
}
// writeSecretFile writes blob to a tmpfile with the given mode and
// returns its path. Returns ("", nil) when blob is empty so callers
// can distinguish "not set" from real I/O errors.
func writeSecretFile(prefix, blob string, mode os.FileMode) (string, error) {
if blob == "" {
return "", nil
}
f, err := os.CreateTemp("", prefix+"*")
if err != nil {
return "", err
}
if _, err := f.WriteString(blob); err != nil {
_ = f.Close()
_ = os.Remove(f.Name())
return "", err
}
if err := f.Close(); err != nil {
return "", err
}
if err := os.Chmod(f.Name(), mode); err != nil {
return "", err
}
return f.Name(), nil
}
func cmpOr(s, fallback string) string {
if s != "" {
return s
}
return fallback
}