Many apps have documents that are indexed by physical locations. For example, your app might allow users to browse stores near their current location.
Cloud Firestore only allows a single range clause per compound query, which means we can't perform geo queries by simply storing latitude and longitude as separate fields and querying a bounding box.
Solution: Geohashes
Geohash is a system for encoding a (latitude, longitude) pair into a single
Base32 string. In the Geohash system the world is divided into a rectangular grid.
Each character of a Geohash string specifies one of 32 subdivisions of the
prefix hash. For example the Geohash abcd is one of 32 four-character hashes
fully contained within the larger Geohash abc.
The longer the shared prefix between two hashes, the closer they are to
each other. For example abcdef is closer to abcdeg than abcdff. However
the converse is not true! Two areas may be very close to each other while
having very different Geohashes:
We can use Geohashes to store and query documents by position in Cloud Firestore with reasonable efficiency while only requiring a single indexed field.
Install helper library
Creating and parsing Geohashes involves some tricky math, so we created helper libraries to abstract the most difficult parts on Android, iOS, and Web:
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
// Add this to your Podfile
pod 'GeoFire/Utils'
Java
// Add this to your app/build.gradle
implementation 'com.firebase:geofire-android-common:3.1.0'
Store Geohashes
For each document you want to index by location, you will need to store a Geohash field:
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
// ...
}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) {
// ...
}
});Query Geohashes
Geohashes allow us to approximate area queries by joining a set of queries on the Geohash field and then filtering out some false positives:
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])
}
var matchingDocs = [QueryDocumentSnapshot]()
// Collect all the query results together into a single list
func getDocumentsCompletion(snapshot: QuerySnapshot?, error: Error?) -> () {
guard let documents = snapshot?.documents else {
print("Unable to fetch snapshot data. \(String(describing: error))")
return
}
for document in documents {
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)
if distance <= radiusInM {
matchingDocs.append(document)
}
}
}
// After all callbacks have executed, matchingDocs contains the result. Note that this
// sample does not demonstrate how to wait on all callbacks to complete.
for query in queries {
query.getDocuments(completion: getDocumentsCompletion)
}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
// ...
}
});Limitations
Using Geohashes for querying locations gives us new capabilities, but comes with its own set of limitations:
- False Positives - querying by Geohash is not exact, and you have to filter out false-positive results on the client side. These extra reads add cost and latency to your app.
- Edge Cases - this query method relies on estimating the distance between lines of longitude/latitude. The accuracy of this estimate decreases as points get closer to the North or South Pole which means Geohash queries have more false positives at extreme latitudes.