oikos/internal/httpapi/sse.go

package httpapi

import (
	"container/list"
	"context"
	"encoding/json"
	"fmt"
	"io"
	"log/slog"
	"net/http"
	"strconv"
	"sync"
	"time"

	"github.com/dtoro/oikos/internal/db/sqlcgen"
	"github.com/dtoro/oikos/internal/httpapi/gen"
	"github.com/jackc/pgx/v5"
	"github.com/jackc/pgx/v5/pgxpool"
)

// SSE broker is a keep-last-event-id in-memory buffer used for subscriber
// fan-out. The database NOTIFY is the primary delivery mechanism; this buffer
// just supports the Last-Event-ID replay on connect.
type sseBroker struct {
	mu     sync.Mutex
	buf    *list.List             // list of sqlcgen.Event
	cache  map[int64]*list.Element // id → list element for O(1) lookup
	cap    int
	lastID int64
}

func newSSEBroker(capacity int) *sseBroker {
	return &sseBroker{
		buf:   list.New(),
		cache: make(map[int64]*list.Element),
		cap:   capacity,
	}
}

func (b *sseBroker) push(ev sqlcgen.Event) {
	b.mu.Lock()
	defer b.mu.Unlock()

	// Evict oldest if at capacity
	for b.buf.Len() >= b.cap && b.buf.Len() > 0 {
		front := b.buf.Front()
		b.cache[front.Value.(sqlcgen.Event).ID] = nil // don't delete, just nil
		b.buf.Remove(front)
	}

	elem := b.buf.PushBack(ev)
	b.cache[ev.ID] = elem
	if ev.ID > b.lastID {
		b.lastID = ev.ID
	}
}

func (b *sseBroker) after(id int64) []sqlcgen.Event {
	b.mu.Lock()
	defer b.mu.Unlock()

	if id >= b.lastID {
		return nil
	}

	// Walk from the front to find the first element after id
	var events []sqlcgen.Event
	for e := b.buf.Front(); e != nil; e = e.Next() {
		ev := e.Value.(sqlcgen.Event)
		if ev.ID > id {
			events = append(events, ev)
		}
	}
	return events
}

func (b *sseBroker) latestID() int64 {
	b.mu.Lock()
	defer b.mu.Unlock()
	return b.lastID
}

// notifyPayload is the JSON payload from the pg_notify trigger (migration 008).
type notifyPayload struct {
	ID            int64   `json:"id"`
	Ts            string  `json:"ts"`
	Type          string  `json:"type"`
	EntityID      *string `json:"entity_id"`
	Severity      string  `json:"severity"`
	Source        string  `json:"source"`
	CorrelationID *string `json:"correlation_id"`
}

// sseSubscriber holds the channels and cancel func for one SSE client.
type sseSubscriber struct {
	ch     chan sqlcgen.Event
	done   chan struct{}
	cancel context.CancelFunc
}

// sseListener runs in a background goroutine: it opens a dedicated pgx
// connection, LISTENs on oikos_events, and fans out each notification to
// all live subscribers. Runs until ctx is cancelled.
func (s *Server) sseListener(ctx context.Context) {
	poolConn, err := s.pool.Acquire(ctx)
	if err != nil {
		slog.Error("sse listener acquire failed", "error", err)
		return
	}
	defer poolConn.Release()

	conn := poolConn.Conn()
	if _, err := conn.Exec(ctx, "LISTEN oikos_events"); err != nil {
		slog.Error("sse listener listen failed", "error", err)
		return
	}

	slog.Info("sse listener started on oikos_events")
	defer slog.Info("sse listener stopped")

	for {
		nt, err := conn.WaitForNotification(ctx)
		if err != nil {
			if ctx.Err() != nil {
				return // normal shutdown
			}
			slog.Error("sse listener notification error", "error", err)
			// Reconnect on error after a brief delay
			select {
			case <-time.After(5 * time.Second):
			case <-ctx.Done():
				return
			}
			// Re-acquire connection
			poolConn.Release()
			var reconnErr error
			poolConn, reconnErr = s.pool.Acquire(ctx)
			if reconnErr != nil {
				slog.Error("sse listener reconnect failed", "error", reconnErr)
				return
			}
			conn = poolConn.Conn()
			if _, err := conn.Exec(ctx, "LISTEN oikos_events"); err != nil {
				slog.Error("sse listener re-listen failed", "error", err)
				return
			}
			continue
		}

		var p notifyPayload
		if err := json.Unmarshal([]byte(nt.Payload), &p); err != nil {
			slog.Error("sse listener unmarshal failed", "error", err)
			continue
		}

		// Fetch full event from DB
		q := sqlcgen.New(s.pool)
		events, err := q.ListEventsAfter(ctx, sqlcgen.ListEventsAfterParams{
			ID:    p.ID - 1,
			Limit: 1,
		})
		if err != nil || len(events) == 0 {
			slog.Warn("sse listener event fetch failed", "id", p.ID, "error", err)
			continue
		}
		ev := events[0]

		// Push to broker
		s.sseBroker.push(ev)

		// Fan out to subscribers (non-blocking send)
		s.sseMu.Lock()
		for sub := range s.sseSubs {
			select {
			case sub.ch <- ev:
			default:
				// Subscriber too slow — drop event for them
				// (they'll reconnect via Last-Event-ID)
			}
		}
		s.sseMu.Unlock()
	}
}

// sqlcEventToGen converts a DB event row to the canonical wire shape so the
// SSE `data:` payload matches GET /events (snake_case keys, decoded data
// object) rather than leaking Go field names and base64-encoded JSONB.
func sqlcEventToGen(ev sqlcgen.Event) gen.Event {
	out := gen.Event{
		Id:            int(ev.ID),
		Ts:            ev.Ts,
		Type:          ev.Type,
		Severity:      gen.EventSeverity(ev.Severity),
		Source:        ev.Source,
		CorrelationId: ev.CorrelationID,
	}
	if ev.EntityID != nil {
		s := ev.EntityID.String()
		out.EntityId = &s
	}
	if len(ev.Data) > 0 {
		var data map[string]any
		if json.Unmarshal(ev.Data, &data) == nil && len(data) > 0 {
			out.Data = &data
		}
	}
	return out
}

// writeSSE writes a single Event as an SSE message. Returns false if the
// write failed (client disconnected). flusher may be nil (io.Pipe path,
// which has no separate flush step).
func writeSSE(w ioWriter, flusher http.Flusher, ev sqlcgen.Event) bool {
	data, err := json.Marshal(sqlcEventToGen(ev))
	if err != nil {
		return true // skip un-serializable events
	}
	_, err = fmt.Fprintf(w, "id: %d\nevent: %s\ndata: %s\n\n", ev.ID, ev.Type, data)
	if err != nil {
		return false
	}
	if flusher != nil {
		flusher.Flush()
	}
	return true
}

// ioWriter is an interface satisfied by both http.ResponseWriter and
// io.StringWriter, letting writeSSE work with the writer directly.
type ioWriter interface {
	Write([]byte) (int, error)
}

// StreamEvents implements the OpenAPI interface (SSE event stream).
//
// The strict-server model hands us only an io.Reader Body (copied to the
// client via io.Copy in the generated Visit method), not the raw
// ResponseWriter/Flusher — so we bridge with an io.Pipe. ctx is the request
// context (the generated wrapper passes r.Context()); when the client
// disconnects it cancels, the SSE loop returns, and pw.Close() ends the
// io.Copy. Without threading ctx through, each disconnected client would
// leak a goroutine holding a pipe.
//
// Note: io.Copy does not flush per-write, so delivery is chunk-buffered
// rather than strictly real-time. Adequate for the event feed; a raw
// flushing handler is a possible future refinement.
func (s *Server) StreamEvents(ctx context.Context, req gen.StreamEventsRequestObject) (gen.StreamEventsResponseObject, error) {
	pr, pw := io.Pipe()
	go func() {
		s.serveSSEWriter(ctx, pw, req.Params)
		pw.Close()
	}()
	return gen.StreamEvents200TexteventStreamResponse{
		Body:          pr,
		ContentLength: 0, // unknown/streaming — omit the Content-Length header
	}, nil
}

// serveSSEWriter runs the SSE loop writing to an io.Writer until the request
// context is cancelled (client disconnect / server shutdown).
func (s *Server) serveSSEWriter(parent context.Context, w io.Writer, params gen.StreamEventsParams) {
	ctx, cancel := context.WithCancel(parent)
	defer cancel()

	sub := &sseSubscriber{
		ch:     make(chan sqlcgen.Event, 64),
		done:   make(chan struct{}),
		cancel: cancel,
	}
	s.sseMu.Lock()
	s.sseSubs[sub] = struct{}{}
	s.sseMu.Unlock()
	defer func() {
		s.sseMu.Lock()
		delete(s.sseSubs, sub)
		s.sseMu.Unlock()
		close(sub.done)
	}()

	// Replay on Last-Event-ID
	if params.LastEventID != nil && *params.LastEventID != "" {
		id, err := strconv.ParseInt(*params.LastEventID, 10, 64)
		if err == nil {
			events := s.sseBroker.after(id)
			if len(events) > 0 {
				for _, ev := range events {
					if !writeSSE(w, nil, ev) {
						return
					}
				}
			}
			// If broker didn't have them all, fetch from DB
			if len(events) == 0 || events[len(events)-1].ID != s.sseBroker.latestID() {
				q := sqlcgen.New(s.pool)
				dbEvents, err := q.ListEventsAfter(ctx, sqlcgen.ListEventsAfterParams{ID: id, Limit: 5000})
				if err != nil {
					slog.Error("sse db replay failed", "error", err)
				} else {
					for _, ev := range dbEvents {
						if !writeSSE(w, nil, ev) {
							return
						}
					}
				}
			}
		}
	}

	// Subscribe and forward
	heartbeat := time.NewTicker(15 * time.Second)
	defer heartbeat.Stop()

	for {
		select {
		case <-ctx.Done():
			return
		case ev, ok := <-sub.ch:
			if !ok {
				return
			}
			if !writeSSE(w, nil, ev) {
				return
			}
		case <-heartbeat.C:
			_, err := fmt.Fprintf(w, ": heartbeat\n\n")
			if err != nil {
				return
			}
		}
	}
}

// Ensure pgxpool is imported — used via Acquire.
var _ = &pgxpool.Pool{}
var _ = pgx.ErrNoRows