oc-discovery/daemons/node/indexer/nameindex.go

package indexer

import (
	"context"
	"encoding/json"
	"strings"
	"sync"
	"time"

	"oc-discovery/daemons/node/common"

	oclib "cloud.o-forge.io/core/oc-lib"
	pubsub "github.com/libp2p/go-libp2p-pubsub"
	pp "github.com/libp2p/go-libp2p/core/peer"
)

// TopicNameIndex is the GossipSub topic shared by regular indexers to exchange
// add/delete events for the distributed name→peerID mapping.
const TopicNameIndex = "oc-name-index"

// nameIndexDedupWindow suppresses re-emission of the same (action, name, peerID)
// tuple within this window, reducing duplicate events when a node is registered
// with multiple indexers simultaneously.
const nameIndexDedupWindow = 30 * time.Second

// NameIndexAction indicates whether a name mapping is being added or removed.
type NameIndexAction string

const (
	NameIndexAdd    NameIndexAction = "add"
	NameIndexDelete NameIndexAction = "delete"
)

// NameIndexEvent is published on TopicNameIndex by each indexer when a node
// registers (add) or is evicted by the GC (delete).
type NameIndexEvent struct {
	Action NameIndexAction `json:"action"`
	Name   string          `json:"name"`
	PeerID string          `json:"peer_id"`
	DID    string          `json:"did"`
}

// nameIndexState holds the local in-memory name index and the sender-side
// deduplication tracker.
//
// Search strategy: trigram inverted index.
//   - byName:  lowercased name → peerID → DID  (for delete and exact resolution)
//   - byPeer:  peerID → lowercased name         (to recompute trigrams on delete)
//   - trigrams: 3-char substring → set of peerIDs (for O(1) substring lookup)
//
// For needles shorter than 3 chars the trigram index cannot help; a linear
// scan of byName is used as fallback (rare and fast enough at small N).
type nameIndexState struct {
	byName   map[string]map[string]string  // name → peerID → DID
	byPeer   map[string]string             // peerID → name
	trigrams map[string]map[string]struct{} // trigram → peerID set
	indexMu  sync.RWMutex

	// emitted deduplicates GossipSub emissions within nameIndexDedupWindow.
	// Purged periodically to prevent unbounded growth.
	emitted   map[string]time.Time
	emittedMu sync.Mutex
}

// trigramsOf returns all overlapping 3-char substrings of s (already lowercased).
// If s is shorter than 3 chars the string itself is returned as the sole token.
func trigramsOf(s string) []string {
	if len(s) < 3 {
		return []string{s}
	}
	out := make([]string, 0, len(s)-2)
	for i := 0; i <= len(s)-3; i++ {
		out = append(out, s[i:i+3])
	}
	return out
}

// addTrigrams inserts peerID into every trigram bucket for name.
func (s *nameIndexState) addTrigrams(name, peerID string) {
	for _, tg := range trigramsOf(name) {
		if s.trigrams[tg] == nil {
			s.trigrams[tg] = map[string]struct{}{}
		}
		s.trigrams[tg][peerID] = struct{}{}
	}
}

// removeTrigrams deletes peerID from every trigram bucket for name,
// cleaning up empty buckets to keep memory tight.
func (s *nameIndexState) removeTrigrams(name, peerID string) {
	for _, tg := range trigramsOf(name) {
		if m := s.trigrams[tg]; m != nil {
			delete(m, peerID)
			if len(m) == 0 {
				delete(s.trigrams, tg)
			}
		}
	}
}

// shouldEmit returns true if the (action, name, peerID) tuple has not been
// emitted within nameIndexDedupWindow, updating the tracker if so.
//
// On DELETE: the ADD entry for the same peer is immediately removed — the peer
// is gone, keeping it would cause the map to grow with departed peers forever.
// The DELETE entry itself is kept for the dedup window to absorb duplicate
// delete events, then cleaned by the purgeEmitted ticker.
func (s *nameIndexState) shouldEmit(action NameIndexAction, name, peerID string) bool {
	key := string(action) + ":" + name + ":" + peerID
	s.emittedMu.Lock()
	defer s.emittedMu.Unlock()
	if t, ok := s.emitted[key]; ok && time.Since(t) < nameIndexDedupWindow {
		return false
	}
	s.emitted[key] = time.Now()
	if action == NameIndexDelete {
		// Peer is leaving: drop its ADD entry — no longer needed.
		delete(s.emitted, string(NameIndexAdd)+":"+name+":"+peerID)
	}
	return true
}

// purgeEmitted removes stale DELETE entries from the emitted dedup map.
// ADD entries are cleaned eagerly on DELETE, so only short-lived DELETE
// entries remain here; the ticker just trims those stragglers.
func (s *nameIndexState) purgeEmitted() {
	now := time.Now()
	s.emittedMu.Lock()
	defer s.emittedMu.Unlock()
	for k, t := range s.emitted {
		if now.Sub(t) >= nameIndexDedupWindow {
			delete(s.emitted, k)
		}
	}
}

// onEvent applies a received NameIndexEvent to the local index.
// "add" inserts/updates the mapping; "delete" removes it.
// Operations are idempotent — duplicate events from multiple indexers are harmless.
func (s *nameIndexState) onEvent(evt NameIndexEvent) {
	if evt.Name == "" || evt.PeerID == "" {
		return
	}
	nameLow := strings.ToLower(evt.Name)
	s.indexMu.Lock()
	defer s.indexMu.Unlock()
	switch evt.Action {
	case NameIndexAdd:
		// If the peer previously had a different name, clean up old trigrams.
		if old, ok := s.byPeer[evt.PeerID]; ok && old != nameLow {
			s.removeTrigrams(old, evt.PeerID)
			if s.byName[old] != nil {
				delete(s.byName[old], evt.PeerID)
				if len(s.byName[old]) == 0 {
					delete(s.byName, old)
				}
			}
		}
		if s.byName[nameLow] == nil {
			s.byName[nameLow] = map[string]string{}
		}
		s.byName[nameLow][evt.PeerID] = evt.DID
		s.byPeer[evt.PeerID] = nameLow
		s.addTrigrams(nameLow, evt.PeerID)

	case NameIndexDelete:
		// Use stored name so trigrams match exactly what was indexed.
		name := nameLow
		if stored, ok := s.byPeer[evt.PeerID]; ok {
			name = stored
		}
		s.removeTrigrams(name, evt.PeerID)
		delete(s.byPeer, evt.PeerID)
		if s.byName[name] != nil {
			delete(s.byName[name], evt.PeerID)
			if len(s.byName[name]) == 0 {
				delete(s.byName, name)
			}
		}
	}
}

// initNameIndex joins TopicNameIndex and starts consuming events.
// Must be called after ix.PS is ready.
func (ix *IndexerService) initNameIndex(ps *pubsub.PubSub) {
	logger := oclib.GetLogger()
	state := &nameIndexState{
		byName:   map[string]map[string]string{},
		byPeer:   map[string]string{},
		trigrams: map[string]map[string]struct{}{},
		emitted:  map[string]time.Time{},
	}
	ix.nameIndex = state

	// Periodically purge the emitted dedup map so it doesn't grow forever.
	go func() {
		t := time.NewTicker(nameIndexDedupWindow)
		defer t.Stop()
		for range t.C {
			state.purgeEmitted()
		}
	}()

	ps.RegisterTopicValidator(TopicNameIndex, func(_ context.Context, _ pp.ID, _ *pubsub.Message) bool {
		return true
	})
	topic, err := ps.Join(TopicNameIndex)
	if err != nil {
		logger.Err(err).Msg("name index: failed to join topic")
		return
	}
	ix.LongLivedStreamRecordedService.LongLivedPubSubService.PubsubMu.Lock()
	ix.LongLivedStreamRecordedService.LongLivedPubSubService.LongLivedPubSubs[TopicNameIndex] = topic
	ix.LongLivedStreamRecordedService.LongLivedPubSubService.PubsubMu.Unlock()

	common.SubscribeEvents(
		ix.LongLivedStreamRecordedService.LongLivedPubSubService,
		context.Background(),
		TopicNameIndex,
		-1,
		func(_ context.Context, evt NameIndexEvent, _ string) {
			ix.nameIndex.onEvent(evt)
		},
	)
}

// publishNameEvent emits a NameIndexEvent on TopicNameIndex, subject to the
// sender-side deduplication window.
func (ix *IndexerService) publishNameEvent(action NameIndexAction, name, peerID, did string) {
	if ix.nameIndex == nil || name == "" || peerID == "" {
		return
	}
	if !ix.nameIndex.shouldEmit(action, name, peerID) {
		return
	}
	ix.LongLivedStreamRecordedService.LongLivedPubSubService.PubsubMu.RLock()
	topic := ix.LongLivedStreamRecordedService.LongLivedPubSubService.LongLivedPubSubs[TopicNameIndex]
	ix.LongLivedStreamRecordedService.LongLivedPubSubService.PubsubMu.RUnlock()
	if topic == nil {
		return
	}
	evt := NameIndexEvent{Action: action, Name: name, PeerID: peerID, DID: did}
	b, err := json.Marshal(evt)
	if err != nil {
		return
	}
	_ = topic.Publish(context.Background(), b)
}

// LookupNameIndex searches the distributed name index for peers whose name
// contains needle (case-insensitive). Returns peerID → DID for matched peers.
// Returns nil if the name index is not initialised.
//
// Algorithm:
//   - needle ≥ 3 chars: trigram intersection → O(|candidates|) verify pass.
//     The trigram index immediately narrows the candidate set; false positives
//     are eliminated by the full-string contains check.
//   - needle < 3 chars: linear scan of byName (rare, still fast at small N).
func (ix *IndexerService) LookupNameIndex(needle string) map[string]string {
	if ix.nameIndex == nil {
		return nil
	}
	needleLow := strings.ToLower(needle)
	result := map[string]string{}

	ix.nameIndex.indexMu.RLock()
	defer ix.nameIndex.indexMu.RUnlock()

	if len(needleLow) < 3 {
		// Short needle: linear scan fallback.
		for name, peers := range ix.nameIndex.byName {
			if strings.Contains(name, needleLow) {
				for peerID, did := range peers {
					result[peerID] = did
				}
			}
		}
		return result
	}

	// Trigram intersection: start with the first trigram's set, then
	// progressively intersect with each subsequent trigram's set.
	tgs := trigramsOf(needleLow)
	var candidates map[string]struct{}
	for _, tg := range tgs {
		set := ix.nameIndex.trigrams[tg]
		if len(set) == 0 {
			return result // any empty trigram set → no possible match
		}
		if candidates == nil {
			candidates = make(map[string]struct{}, len(set))
			for pid := range set {
				candidates[pid] = struct{}{}
			}
		} else {
			for pid := range candidates {
				if _, ok := set[pid]; !ok {
					delete(candidates, pid)
				}
			}
		}
		if len(candidates) == 0 {
			return result
		}
	}

	// Full-string verification pass: trigrams admit false positives
	// (e.g. "abc" and "bca" share the trigram "bc_" with a rotated name).
	for peerID := range candidates {
		name := ix.nameIndex.byPeer[peerID]
		if strings.Contains(name, needleLow) {
			did := ""
			if m := ix.nameIndex.byName[name]; m != nil {
				did = m[peerID]
			}
			result[peerID] = did
		}
	}
	return result
}