多くのアプリには、物理的な位置によってインデックスされたドキュメントがあります。たとえば、アプリで現在地付近の店舗をブラウジングできることがあります。
ソリューション: Geohash
Geohash は、(latitude, longitude)
ペアを 1 つの Base32 文字列にエンコードするシステムです。Geohash システムでは、世界は矩形グリッドに分割されます。Geohash 文字列の各文字は、プレフィックス ハッシュの 32 の下位区分の 1 つを指定します。たとえば Geohash abcd
は、32 個の 4 文字ハッシュのうちの 1 つで、より大きな Geohash abc
に完全に含まれています。
2 つのハッシュの共有プレフィックスが長いほど、これらのハッシュは相互に近くなります。たとえば、abcdef
は abcdff
よりも abcdeg
に近い値です。ただしこの逆は当てはまりません。かなり異なる Geohash を持つ 2 つの領域が、相互に非常に近いことがあります。
Cloud Firestore で位置によってドキュメントを保存およびクエリするときに Geohash を使用すると、適度に効率的に処理でき、また必要となるインデックス付きフィールドが 1 つのみです。
ヘルパー ライブラリをインストールする
Geohash の作成と解析には、複雑な数学が必要となるため、Android、Apple、ウェブで最も難しい部分を抽象化するためのヘルパー ライブラリを作成しました。
Web
// Install from NPM. If you prefer to use a static .js file visit
// https://github.com/firebase/geofire-js/releases and download
// geofire-common.min.js from the latest version
npm install --save geofire-common
Web
// Install from NPM. If you prefer to use a static .js file visit
// https://github.com/firebase/geofire-js/releases and download
// geofire-common.min.js from the latest version
npm install --save geofire-common
Swift
Kotlin+KTX
// Add this to your app/build.gradle
implementation 'com.firebase:geofire-android-common:3.2.0'
Java
// Add this to your app/build.gradle
implementation 'com.firebase:geofire-android-common:3.1.0'
Geohash を保存する
位置でインデックス登録するドキュメントごとに、Geohash フィールドを保存する必要があります。
Web
import { doc, updateDoc } from 'firebase/firestore'; // Compute the GeoHash for a lat/lng point const lat = 51.5074; const lng = 0.1278; const hash = geofire.geohashForLocation([lat, lng]); // Add the hash and the lat/lng to the document. We will use the hash // for queries and the lat/lng for distance comparisons. const londonRef = doc(db, 'cities', 'LON'); await updateDoc(londonRef, { geohash: hash, lat: lat, lng: lng });
Web
// Compute the GeoHash for a lat/lng point const lat = 51.5074; const lng = 0.1278; const hash = geofire.geohashForLocation([lat, lng]); // Add the hash and the lat/lng to the document. We will use the hash // for queries and the lat/lng for distance comparisons. const londonRef = db.collection('cities').doc('LON'); londonRef.update({ geohash: hash, lat: lat, lng: lng }).then(() => { // ... });
Swift
// Compute the GeoHash for a lat/lng point let latitude = 51.5074 let longitude = 0.12780 let location = CLLocationCoordinate2D(latitude: latitude, longitude: longitude) let hash = GFUtils.geoHash(forLocation: location) // Add the hash and the lat/lng to the document. We will use the hash // for queries and the lat/lng for distance comparisons. let documentData: [String: Any] = [ "geohash": hash, "lat": latitude, "lng": longitude ] let londonRef = db.collection("cities").document("LON") londonRef.updateData(documentData) { error in // ... }
Kotlin+KTX
// Compute the GeoHash for a lat/lng point val lat = 51.5074 val lng = 0.1278 val hash = GeoFireUtils.getGeoHashForLocation(GeoLocation(lat, lng)) // Add the hash and the lat/lng to the document. We will use the hash // for queries and the lat/lng for distance comparisons. val updates: MutableMap<String, Any> = mutableMapOf( "geohash" to hash, "lat" to lat, "lng" to lng, ) val londonRef = db.collection("cities").document("LON") londonRef.update(updates) .addOnCompleteListener { // ... }
Java
// Compute the GeoHash for a lat/lng point double lat = 51.5074; double lng = 0.1278; String hash = GeoFireUtils.getGeoHashForLocation(new GeoLocation(lat, lng)); // Add the hash and the lat/lng to the document. We will use the hash // for queries and the lat/lng for distance comparisons. Map<String, Object> updates = new HashMap<>(); updates.put("geohash", hash); updates.put("lat", lat); updates.put("lng", lng); DocumentReference londonRef = db.collection("cities").document("LON"); londonRef.update(updates) .addOnCompleteListener(new OnCompleteListener<Void>() { @Override public void onComplete(@NonNull Task<Void> task) { // ... } });
Geohash に対してクエリを実行する
Geohash を使用すると、Geohash フィールドに対する一連のクエリを結合し、偽陽性を除外することで、領域クエリを近似できます。
Web
import { collection, query, orderBy, startAt, endAt, getDocs } from 'firebase/firestore'; // Find cities within 50km of London const center = [51.5074, 0.1278]; const radiusInM = 50 * 1000; // Each item in 'bounds' represents a startAt/endAt pair. We have to issue // a separate query for each pair. There can be up to 9 pairs of bounds // depending on overlap, but in most cases there are 4. const bounds = geofire.geohashQueryBounds(center, radiusInM); const promises = []; for (const b of bounds) { const q = query( collection(db, 'cities'), orderBy('geohash'), startAt(b[0]), endAt(b[1])); promises.push(getDocs(q)); } // Collect all the query results together into a single list const snapshots = await Promise.all(promises); const matchingDocs = []; for (const snap of snapshots) { for (const doc of snap.docs) { const lat = doc.get('lat'); const lng = doc.get('lng'); // We have to filter out a few false positives due to GeoHash // accuracy, but most will match const distanceInKm = geofire.distanceBetween([lat, lng], center); const distanceInM = distanceInKm * 1000; if (distanceInM <= radiusInM) { matchingDocs.push(doc); } } }
Web
// Find cities within 50km of London const center = [51.5074, 0.1278]; const radiusInM = 50 * 1000; // Each item in 'bounds' represents a startAt/endAt pair. We have to issue // a separate query for each pair. There can be up to 9 pairs of bounds // depending on overlap, but in most cases there are 4. const bounds = geofire.geohashQueryBounds(center, radiusInM); const promises = []; for (const b of bounds) { const q = db.collection('cities') .orderBy('geohash') .startAt(b[0]) .endAt(b[1]); promises.push(q.get()); } // Collect all the query results together into a single list Promise.all(promises).then((snapshots) => { const matchingDocs = []; for (const snap of snapshots) { for (const doc of snap.docs) { const lat = doc.get('lat'); const lng = doc.get('lng'); // We have to filter out a few false positives due to GeoHash // accuracy, but most will match const distanceInKm = geofire.distanceBetween([lat, lng], center); const distanceInM = distanceInKm * 1000; if (distanceInM <= radiusInM) { matchingDocs.push(doc); } } } return matchingDocs; }).then((matchingDocs) => { // Process the matching documents // ... });
Swift
// Find cities within 50km of London let center = CLLocationCoordinate2D(latitude: 51.5074, longitude: 0.1278) let radiusInM: Double = 50 * 1000 // Each item in 'bounds' represents a startAt/endAt pair. We have to issue // a separate query for each pair. There can be up to 9 pairs of bounds // depending on overlap, but in most cases there are 4. let queryBounds = GFUtils.queryBounds(forLocation: center, withRadius: radiusInM) let queries = queryBounds.map { bound -> Query in return db.collection("cities") .order(by: "geohash") .start(at: [bound.startValue]) .end(at: [bound.endValue]) } @Sendable func fetchMatchingDocs(from query: Query, center: CLLocationCoordinate2D, radiusInMeters: Double) async throws -> [QueryDocumentSnapshot] { let snapshot = try await query.getDocuments() // Collect all the query results together into a single list return snapshot.documents.filter { document in let lat = document.data()["lat"] as? Double ?? 0 let lng = document.data()["lng"] as? Double ?? 0 let coordinates = CLLocation(latitude: lat, longitude: lng) let centerPoint = CLLocation(latitude: center.latitude, longitude: center.longitude) // We have to filter out a few false positives due to GeoHash accuracy, but // most will match let distance = GFUtils.distance(from: centerPoint, to: coordinates) return distance <= radiusInM } } // After all callbacks have executed, matchingDocs contains the result. Note that this code // executes all queries serially, which may not be optimal for performance. do { let matchingDocs = try await withThrowingTaskGroup(of: [QueryDocumentSnapshot].self) { group -> [QueryDocumentSnapshot] in for query in queries { group.addTask { try await fetchMatchingDocs(from: query, center: center, radiusInMeters: radiusInM) } } var matchingDocs = [QueryDocumentSnapshot]() for try await documents in group { matchingDocs.append(contentsOf: documents) } return matchingDocs } print("Docs matching geoquery: \(matchingDocs)") } catch { print("Unable to fetch snapshot data. \(error)") }
Kotlin+KTX
// Find cities within 50km of London val center = GeoLocation(51.5074, 0.1278) val radiusInM = 50.0 * 1000.0 // Each item in 'bounds' represents a startAt/endAt pair. We have to issue // a separate query for each pair. There can be up to 9 pairs of bounds // depending on overlap, but in most cases there are 4. val bounds = GeoFireUtils.getGeoHashQueryBounds(center, radiusInM) val tasks: MutableList<Task<QuerySnapshot>> = ArrayList() for (b in bounds) { val q = db.collection("cities") .orderBy("geohash") .startAt(b.startHash) .endAt(b.endHash) tasks.add(q.get()) } // Collect all the query results together into a single list Tasks.whenAllComplete(tasks) .addOnCompleteListener { val matchingDocs: MutableList<DocumentSnapshot> = ArrayList() for (task in tasks) { val snap = task.result for (doc in snap!!.documents) { val lat = doc.getDouble("lat")!! val lng = doc.getDouble("lng")!! // We have to filter out a few false positives due to GeoHash // accuracy, but most will match val docLocation = GeoLocation(lat, lng) val distanceInM = GeoFireUtils.getDistanceBetween(docLocation, center) if (distanceInM <= radiusInM) { matchingDocs.add(doc) } } } // matchingDocs contains the results // ... }
Java
// Find cities within 50km of London final GeoLocation center = new GeoLocation(51.5074, 0.1278); final double radiusInM = 50 * 1000; // Each item in 'bounds' represents a startAt/endAt pair. We have to issue // a separate query for each pair. There can be up to 9 pairs of bounds // depending on overlap, but in most cases there are 4. List<GeoQueryBounds> bounds = GeoFireUtils.getGeoHashQueryBounds(center, radiusInM); final List<Task<QuerySnapshot>> tasks = new ArrayList<>(); for (GeoQueryBounds b : bounds) { Query q = db.collection("cities") .orderBy("geohash") .startAt(b.startHash) .endAt(b.endHash); tasks.add(q.get()); } // Collect all the query results together into a single list Tasks.whenAllComplete(tasks) .addOnCompleteListener(new OnCompleteListener<List<Task<?>>>() { @Override public void onComplete(@NonNull Task<List<Task<?>>> t) { List<DocumentSnapshot> matchingDocs = new ArrayList<>(); for (Task<QuerySnapshot> task : tasks) { QuerySnapshot snap = task.getResult(); for (DocumentSnapshot doc : snap.getDocuments()) { double lat = doc.getDouble("lat"); double lng = doc.getDouble("lng"); // We have to filter out a few false positives due to GeoHash // accuracy, but most will match GeoLocation docLocation = new GeoLocation(lat, lng); double distanceInM = GeoFireUtils.getDistanceBetween(docLocation, center); if (distanceInM <= radiusInM) { matchingDocs.add(doc); } } } // matchingDocs contains the results // ... } });
制限事項
位置のクエリに Geohash を使用することで新たな可能性がもたらされますが、それに伴い特有の制限事項もあります。
- 偽陽性 - Geohash によるクエリは正確でないため、クライアントサイドで偽陽性の結果を除外する必要があります。このような追加の読み込みのために、アプリのコストとレイテンシが増加します。
- エッジケース - このクエリ手法は、経線 / 緯線間の距離の推定を使用します。この推定の精度は、位置点が北極または南極に近づくにつれて下がります。つまり、超高緯度では Geohash クエリの偽陽性率が高くなります。