// Get the index keys for elements that are GeoJSON.
void S2AccessMethod::getGeoKeys(const BSONElementSet& elements, BSONObjSet* out) const {
for (BSONElementSet::iterator i = elements.begin(); i != elements.end(); ++i) {
uassert(16754, "Can't parse geometry from element: " + i->toString(),
i->isABSONObj());
const BSONObj &obj = i->Obj();
vector<string> cells;
bool succeeded = S2SearchUtil::getKeysForObject(obj, _params, &cells);
uassert(16755, "Can't extract geo keys from object, malformed geometry?:"
+ obj.toString(), succeeded);
uassert(16756, "Unable to generate keys for (likely malformed) geometry: "
+ obj.toString(),
cells.size() > 0);
for (vector<string>::const_iterator it = cells.begin(); it != cells.end(); ++it) {
BSONObjBuilder b;
b.append("", *it);
out->insert(b.obj());
}
}
if (0 == out->size()) {
BSONObjBuilder b;
b.appendNull("");
out->insert(b.obj());
}
}
// Get the index keys for elements that are GeoJSON.
void S2AccessMethod::getGeoKeys(const BSONElementSet& elements, BSONObjSet* out) const {
S2RegionCoverer coverer;
_params.configureCoverer(&coverer);
// See here for GeoJSON format: geojson.org/geojson-spec.html
for (BSONElementSet::iterator i = elements.begin(); i != elements.end(); ++i) {
uassert(16754, "Can't parse geometry from element: " + i->toString(),
i->isABSONObj());
const BSONObj &obj = i->Obj();
vector<string> cells;
S2Polyline line;
S2Cell point;
// We only support GeoJSON polygons. Why?:
// 1. we don't automagically do WGS84/flat -> WGS84, and
// 2. the old polygon format must die.
if (GeoParser::isGeoJSONPolygon(obj)) {
S2Polygon polygon;
GeoParser::parseGeoJSONPolygon(obj, &polygon);
keysFromRegion(&coverer, polygon, &cells);
} else if (GeoParser::parseLineString(obj, &line)) {
keysFromRegion(&coverer, line, &cells);
} else if (GeoParser::parsePoint(obj, &point)) {
S2CellId parent(point.id().parent(_params.finestIndexedLevel));
cells.push_back(parent.toString());
} else {
uasserted(16755, "Can't extract geo keys from object, malformed geometry?:"
+ obj.toString());
}
uassert(16756, "Unable to generate keys for (likely malformed) geometry: "
+ obj.toString(),
cells.size() > 0);
for (vector<string>::const_iterator it = cells.begin(); it != cells.end(); ++it) {
BSONObjBuilder b;
b.append("", *it);
out->insert(b.obj());
}
}
if (0 == out->size()) {
BSONObjBuilder b;
b.appendNull("");
out->insert(b.obj());
}
}
// This is the actual search.
bool S2Cursor::advance() {
if (_numToReturn <= 0) { return false; }
for (; _btreeCursor->ok(); _btreeCursor->advance()) {
++_nscanned;
if (_seen.end() != _seen.find(_btreeCursor->currLoc())) { continue; }
_seen.insert(_btreeCursor->currLoc());
++_matchTested;
MatchDetails details;
bool matched = _matcher->matchesCurrent(_btreeCursor.get(), &details);
if (!matched) { continue; }
const BSONObj &indexedObj = _btreeCursor->currLoc().obj();
++_geoTested;
size_t geoFieldsMatched = 0;
// OK, cool, non-geo match satisfied. See if the object actually overlaps w/the geo
// query fields.
for (size_t i = 0; i < _fields.size(); ++i) {
BSONElementSet geoFieldElements;
indexedObj.getFieldsDotted(_fields[i].getField(), geoFieldElements, false);
if (geoFieldElements.empty()) { continue; }
bool match = false;
for (BSONElementSet::iterator oi = geoFieldElements.begin();
!match && (oi != geoFieldElements.end()); ++oi) {
if (!oi->isABSONObj()) { continue; }
const BSONObj &geoObj = oi->Obj();
GeometryContainer geoContainer;
uassert(16698, "malformed geometry: " + geoObj.toString(),
geoContainer.parseFrom(geoObj));
match = _fields[i].satisfiesPredicate(geoContainer);
}
if (match) { ++geoFieldsMatched; }
}
if (geoFieldsMatched == _fields.size()) {
// We have a winner! And we point at it.
--_numToReturn;
return true;
}
}
return false;
}
// Get the index keys for elements that are GeoJSON.
void getGeoKeys(const BSONElementSet &elements, BSONObjSet *out) const {
S2RegionCoverer coverer;
_params.configureCoverer(&coverer);
// See here for GeoJSON format: geojson.org/geojson-spec.html
for (BSONElementSet::iterator i = elements.begin(); i != elements.end(); ++i) {
if (!i->isABSONObj()) { continue; } // error?
const BSONObj &obj = i->Obj();
vector<string> cells;
S2Polygon polygon;
S2Polyline line;
S2Cell point;
if (GeoParser::parsePolygon(obj, &polygon)) {
keysFromRegion(&coverer, polygon, &cells);
} else if (GeoParser::parseLineString(obj, &line)) {
keysFromRegion(&coverer, line, &cells);
} else if (GeoParser::parsePoint(obj, &point)) {
keysFromRegion(&coverer, point, &cells);
} else {
uasserted(16572, "Can't extract geo keys from object, malformed geometry?:"
+ obj.toString());
}
for (vector<string>::const_iterator it = cells.begin(); it != cells.end(); ++it) {
BSONObjBuilder b;
b.append("", *it);
out->insert(b.obj());
}
}
if (0 == out->size()) {
BSONObjBuilder b;
b.appendNull("");
out->insert(b.obj());
}
}
PlanStage::StageState S2NearStage::addResultToQueue(WorkingSetID* out) {
PlanStage::StageState state = _child->work(out);
// All done reading from _child.
if (PlanStage::IS_EOF == state) {
_child.reset();
_keyGeoFilter.reset();
// Adjust the annulus size depending on how many results we got.
if (_results.empty()) {
_radiusIncrement *= 2;
} else if (_results.size() < 300) {
_radiusIncrement *= 2;
} else if (_results.size() > 600) {
_radiusIncrement /= 2;
}
// Make a new ixscan next time.
return PlanStage::NEED_TIME;
}
// Nothing to do unless we advance.
if (PlanStage::ADVANCED != state) { return state; }
WorkingSetMember* member = _ws->get(*out);
// Must have an object in order to get geometry out of it.
verify(member->hasObj());
// The scans we use don't dedup so we must dedup them ourselves. We only put locs into here
// if we know for sure whether or not we'll return them in this annulus.
if (member->hasLoc()) {
if (_seenInScan.end() != _seenInScan.find(member->loc)) {
return PlanStage::NEED_TIME;
}
}
// Get all the fields with that name from the document.
BSONElementSet geom;
member->obj.getFieldsDotted(_params.nearQuery.field, geom, false);
if (geom.empty()) {
return PlanStage::NEED_TIME;
}
// Some value that any distance we can calculate will be less than.
double minDistance = numeric_limits<double>::max();
BSONObj minDistanceObj;
for (BSONElementSet::iterator git = geom.begin(); git != geom.end(); ++git) {
if (!git->isABSONObj()) {
mongoutils::str::stream ss;
ss << "s2near stage read invalid geometry element " << *git << " from child";
Status status(ErrorCodes::InternalError, ss);
*out = WorkingSetCommon::allocateStatusMember( _ws, status);
return PlanStage::FAILURE;
}
BSONObj obj = git->Obj();
double distToObj;
if (S2SearchUtil::distanceBetween(_params.nearQuery.centroid.point, obj, &distToObj)) {
if (distToObj < minDistance) {
minDistance = distToObj;
minDistanceObj = obj;
}
}
else {
warning() << "unknown geometry: " << obj.toString();
}
}
// If we're here we'll either include the doc in this annulus or reject it. It's safe to
// ignore it if it pops up again in this annulus.
if (member->hasLoc()) {
_seenInScan.insert(member->loc);
}
// If the distance to the doc satisfies our distance criteria, add it to our buffered
// results.
if (minDistance >= _innerRadius &&
(_outerRadiusInclusive ? minDistance <= _outerRadius : minDistance < _outerRadius)) {
_results.push(Result(*out, minDistance));
if (_params.addDistMeta) {
// FLAT implies the output distances are in radians. Convert to meters.
if (FLAT == _params.nearQuery.centroid.crs) {
member->addComputed(new GeoDistanceComputedData(minDistance
/ kRadiusOfEarthInMeters));
}
else {
member->addComputed(new GeoDistanceComputedData(minDistance));
}
}
if (_params.addPointMeta) {
member->addComputed(new GeoNearPointComputedData(minDistanceObj));
}
if (member->hasLoc()) {
_invalidationMap[member->loc] = *out;
}
}
return PlanStage::NEED_TIME;
}
static bool run2DSphereGeoNear(NamespaceDetails* nsDetails, int idxNo, BSONObj& cmdObj,
const GeoNearArguments &parsedArgs, string& errmsg,
BSONObjBuilder& result) {
auto_ptr<IndexDescriptor> descriptor(CatalogHack::getDescriptor(nsDetails, idxNo));
auto_ptr<S2AccessMethod> sam(new S2AccessMethod(descriptor.get()));
const S2IndexingParams& params = sam->getParams();
auto_ptr<S2NearIndexCursor> nic(new S2NearIndexCursor(descriptor.get(), params));
vector<string> geoFieldNames;
BSONObjIterator i(descriptor->keyPattern());
while (i.more()) {
BSONElement e = i.next();
if (e.type() == String && IndexNames::GEO_2DSPHERE == e.valuestr()) {
geoFieldNames.push_back(e.fieldName());
}
}
// NOTE(hk): If we add a new argument to geoNear, we could have a
// 2dsphere index with multiple indexed geo fields, and the geoNear
// could pick the one to run over. Right now, we just require one.
uassert(16552, "geoNear requires exactly one indexed geo field", 1 == geoFieldNames.size());
NearQuery nearQuery(geoFieldNames[0]);
uassert(16679, "Invalid geometry given as arguments to geoNear: " + cmdObj.toString(),
nearQuery.parseFromGeoNear(cmdObj, params.radius));
uassert(16683, "geoNear on 2dsphere index requires spherical",
parsedArgs.isSpherical);
// NOTE(hk): For a speedup, we could look through the query to see if
// we've geo-indexed any of the fields in it.
vector<GeoQuery> regions;
nic->seek(parsedArgs.query, nearQuery, regions);
// We do pass in the query above, but it's just so we can possibly use it in our index
// scan. We have to do our own matching.
auto_ptr<Matcher> matcher(new Matcher(parsedArgs.query));
double totalDistance = 0;
BSONObjBuilder resultBuilder(result.subarrayStart("results"));
double farthestDist = 0;
int results;
for (results = 0; results < parsedArgs.numWanted && !nic->isEOF(); ++results) {
BSONObj currObj = nic->getValue().obj();
if (!matcher->matches(currObj)) {
--results;
nic->next();
continue;
}
double dist = nic->currentDistance();
// If we got the distance in radians, output it in radians too.
if (nearQuery.fromRadians) { dist /= params.radius; }
dist *= parsedArgs.distanceMultiplier;
totalDistance += dist;
if (dist > farthestDist) { farthestDist = dist; }
BSONObjBuilder oneResultBuilder(
resultBuilder.subobjStart(BSONObjBuilder::numStr(results)));
oneResultBuilder.append("dis", dist);
if (parsedArgs.includeLocs) {
BSONElementSet geoFieldElements;
currObj.getFieldsDotted(geoFieldNames[0], geoFieldElements, false);
for (BSONElementSet::iterator oi = geoFieldElements.begin();
oi != geoFieldElements.end(); ++oi) {
if (oi->isABSONObj()) {
oneResultBuilder.appendAs(*oi, "loc");
}
}
}
oneResultBuilder.append("obj", currObj);
oneResultBuilder.done();
nic->next();
}
resultBuilder.done();
BSONObjBuilder stats(result.subobjStart("stats"));
stats.appendNumber("nscanned", nic->nscanned());
stats.append("avgDistance", totalDistance / results);
stats.append("maxDistance", farthestDist);
stats.append("time", cc().curop()->elapsedMillis());
stats.done();
return true;
}
// Fill _results with all of the results in the annulus defined by _innerRadius and
// _outerRadius. If no results are found, grow the annulus and repeat until success (or
// until the edge of the world).
void S2NearIndexCursor::fillResults() {
verify(_results.empty());
if (_innerRadius >= _outerRadius) {
return;
}
if (_innerRadius > _maxDistance) {
return;
}
// We iterate until 1. our search radius is too big or 2. we find results.
do {
// Some of these arguments are opaque, look at the definitions of the involved classes.
FieldRangeSet frs(_descriptor->parentNS().c_str(), makeFRSObject(), false, false);
shared_ptr<FieldRangeVector> frv(new FieldRangeVector(frs, _specForFRV, 1));
scoped_ptr<BtreeCursor> cursor(BtreeCursor::make(nsdetails(_descriptor->parentNS()),
_descriptor->getOnDisk(), frv, 0, 1));
// The cursor may return the same obj more than once for a given
// FRS, so we make sure to only consider it once in any given annulus.
//
// We don't want this outside of the 'do' loop because the covering
// for an annulus may return an object whose distance to the query
// point is actually contained in a subsequent annulus. If we
// didn't consider every object in a given annulus we might miss
// the point.
//
// We don't use a global 'seen' because we get that by requiring
// the distance from the query point to the indexed geo to be
// within our 'current' annulus, and I want to dodge all yield
// issues if possible.
unordered_set<DiskLoc, DiskLoc::Hasher> seen;
LOG(1) << "looking at annulus from " << _innerRadius << " to " << _outerRadius << endl;
LOG(1) << "Total # returned: " << _stats._numReturned << endl;
// Do the actual search through this annulus.
for (; cursor->ok(); cursor->advance()) {
// Don't bother to look at anything we've returned.
if (_returned.end() != _returned.find(cursor->currLoc())) {
++_stats._returnSkip;
continue;
}
++_stats._nscanned;
if (seen.end() != seen.find(cursor->currLoc())) {
++_stats._btreeDups;
continue;
}
// Get distance interval from our query point to the cell.
// If it doesn't overlap with our current shell, toss.
BSONObj currKey(cursor->currKey());
BSONObjIterator it(currKey);
BSONElement geoKey;
for (int i = 0; i <= _nearFieldIndex; ++i) {
geoKey = it.next();
}
S2Cell keyCell = S2Cell(S2CellId::FromString(geoKey.String()));
if (!_annulus.MayIntersect(keyCell)) {
++_stats._keyGeoSkip;
continue;
}
// We have to add this document to seen *AFTER* the key intersection test.
// A geometry may have several keys, one of which may be in our search shell and one
// of which may be outside of it. We don't want to ignore a document just because
// one of its covers isn't inside this annulus.
seen.insert(cursor->currLoc());
// At this point forward, we will not examine the document again in this annulus.
const BSONObj& indexedObj = cursor->currLoc().obj();
// Match against indexed geo fields.
++_stats._geoMatchTested;
size_t geoFieldsMatched = 0;
// See if the object actually overlaps w/the geo query fields.
for (size_t i = 0; i < _indexedGeoFields.size(); ++i) {
BSONElementSet geoFieldElements;
indexedObj.getFieldsDotted(_indexedGeoFields[i].getField(), geoFieldElements,
false);
if (geoFieldElements.empty()) {
continue;
}
bool match = false;
for (BSONElementSet::iterator oi = geoFieldElements.begin();
!match && (oi != geoFieldElements.end()); ++oi) {
if (!oi->isABSONObj()) {
continue;
}
const BSONObj &geoObj = oi->Obj();
GeometryContainer geoContainer;
uassert(16762, "ill-formed geometry: " + geoObj.toString(),
geoContainer.parseFrom(geoObj));
match = _indexedGeoFields[i].satisfiesPredicate(geoContainer);
}
if (match) {
++geoFieldsMatched;
}
}
if (geoFieldsMatched != _indexedGeoFields.size()) {
continue;
}
// Get all the fields with that name from the document.
BSONElementSet geoFieldElements;
indexedObj.getFieldsDotted(_nearQuery.field, geoFieldElements, false);
if (geoFieldElements.empty()) {
continue;
}
++_stats._inAnnulusTested;
double minDistance = 1e20;
// Look at each field in the document and take the min. distance.
for (BSONElementSet::iterator oi = geoFieldElements.begin();
oi != geoFieldElements.end(); ++oi) {
if (!oi->isABSONObj()) {
continue;
}
double dist = distanceTo(oi->Obj());
minDistance = min(dist, minDistance);
}
// We could be in an annulus, yield, add new points closer to
// query point than the last point we returned, then unyield.
// This would return points out of order.
if (minDistance < _returnedDistance) {
continue;
}
// If the min. distance satisfies our distance criteria
if (minDistance >= _innerRadius && minDistance < _outerRadius) {
// The result is valid. We have to de-dup ourselves here.
if (_returned.end() == _returned.find(cursor->currLoc())) {
_results.push(Result(cursor->currLoc(), cursor->currKey(),
minDistance));
}
}
}
if (_results.empty()) {
LOG(1) << "results empty!\n";
_radiusIncrement *= 2;
nextAnnulus();
} else if (_results.size() < 300) {
_radiusIncrement *= 2;
} else if (_results.size() > 600) {
_radiusIncrement /= 2;
}
} while (_results.empty()
&& _innerRadius < _maxDistance
&& _innerRadius < _outerRadius);
LOG(1) << "Filled shell with " << _results.size() << " results" << endl;
}
PlanStage::StageState S2NearStage::addResultToQueue(WorkingSetID* out) {
PlanStage::StageState state = _child->work(out);
// All done reading from _child.
if (PlanStage::IS_EOF == state) {
_child.reset();
// Adjust the annulus size depending on how many results we got.
if (_results.empty()) {
_radiusIncrement *= 2;
} else if (_results.size() < 300) {
_radiusIncrement *= 2;
} else if (_results.size() > 600) {
_radiusIncrement /= 2;
}
// Make a new ixscan next time.
return PlanStage::NEED_TIME;
}
// Nothing to do unless we advance.
if (PlanStage::ADVANCED != state) { return state; }
// TODO Speed improvements:
//
// 0. Modify fetch to preserve key data and test for intersection w/annulus.
//
// 1. keep track of what we've seen in this scan and possibly ignore it.
//
// 2. keep track of results we've returned before and ignore them.
WorkingSetMember* member = _ws->get(*out);
// Must have an object in order to get geometry out of it.
verify(member->hasObj());
// Get all the fields with that name from the document.
BSONElementSet geom;
member->obj.getFieldsDotted(_params.nearQuery.field, geom, false);
if (geom.empty()) {return PlanStage::NEED_TIME; }
// Some value that any distance we can calculate will be less than.
double minDistance = numeric_limits<double>::max();
BSONObj minDistanceObj;
for (BSONElementSet::iterator git = geom.begin(); git != geom.end(); ++git) {
if (!git->isABSONObj()) {
mongoutils::str::stream ss;
ss << "s2near stage read invalid geometry element " << *git << " from child";
Status status(ErrorCodes::InternalError, ss);
*out = WorkingSetCommon::allocateStatusMember( _ws, status);
return PlanStage::FAILURE;
}
BSONObj obj = git->Obj();
double distToObj;
if (S2SearchUtil::distanceBetween(_params.nearQuery.centroid.point, obj, &distToObj)) {
if (distToObj < minDistance) {
minDistance = distToObj;
minDistanceObj = obj;
}
}
else {
warning() << "unknown geometry: " << obj.toString();
}
}
// If the distance to the doc satisfies our distance criteria, add it to our buffered
// results.
if (minDistance >= _innerRadius &&
(_outerRadiusInclusive ? minDistance <= _outerRadius : minDistance < _outerRadius)) {
_results.push(Result(*out, minDistance));
if (_params.addDistMeta) {
member->addComputed(new GeoDistanceComputedData(minDistance));
}
if (_params.addPointMeta) {
member->addComputed(new GeoNearPointComputedData(minDistanceObj));
}
if (member->hasLoc()) {
_invalidationMap[member->loc] = *out;
}
}
return PlanStage::NEED_TIME;
}
// Fill _results with the next shell of results. We may have to search several times to do
// this. If _results.empty() after calling fillResults, there are no more possible results.
void S2NearCursor::fillResults() {
verify(_results.empty());
if (_innerRadius >= _outerRadius) { return; }
if (_innerRadius > _maxDistance) { return; }
// We iterate until 1. our search radius is too big or 2. we find results.
do {
// Some of these arguments are opaque, look at the definitions of the involved classes.
FieldRangeSet frs(_details->parentNS().c_str(), makeFRSObject(), false, false);
shared_ptr<FieldRangeVector> frv(new FieldRangeVector(frs, _specForFRV, 1));
scoped_ptr<BtreeCursor> cursor(BtreeCursor::make(nsdetails(_details->parentNS().c_str()),
*_details, frv, 0, 1));
// Do the actual search through this annulus.
size_t considered = 0;
for (; cursor->ok(); cursor->advance()) {
++_nscanned;
++considered;
MatchDetails details;
bool matched = _matcher->matchesCurrent(cursor.get(), &details);
if (!matched) { continue; }
const BSONObj& indexedObj = cursor->currLoc().obj();
size_t geoFieldsInRange = 0;
double minMatchingDistance = 1e20;
// Calculate the distance from our query point(s) to the geo field(s).
for (size_t i = 0; i < _fields.size(); ++i) {
const GeoQueryField& field = _fields[i];
BSONElementSet geoFieldElements;
indexedObj.getFieldsDotted(field.field, geoFieldElements);
if (geoFieldElements.empty()) { continue; }
S2Point us = field.getCentroid();
for (BSONElementSet::iterator oi = geoFieldElements.begin();
oi != geoFieldElements.end(); ++oi) {
const BSONObj &geoObj = oi->Obj();
double dist = -1;
S2Point them;
if (GeoJSONParser::isPolygon(geoObj)) {
S2Polygon shape;
GeoJSONParser::parsePolygon(geoObj, &shape);
them = shape.Project(us);
} else if (GeoJSONParser::isLineString(geoObj)) {
S2Polyline shape;
GeoJSONParser::parseLineString(geoObj, &shape);
int tmp;
them = shape.Project(us, &tmp);
} else if (GeoJSONParser::isPoint(geoObj)) {
S2Cell point;
GeoJSONParser::parsePoint(geoObj, &point);
them = point.GetCenter();
}
S1Angle angle(us, them);
dist = angle.radians() * _params.radius;
if (dist >= _innerRadius && dist <= _outerRadius) {
++geoFieldsInRange;
minMatchingDistance = min(dist, minMatchingDistance);
}
}
}
if (_fields.size() == geoFieldsInRange) {
if (_returned.end() == _returned.find(cursor->currLoc())) {
_results.push(Result(cursor->currLoc(), cursor->currKey(), minMatchingDistance));
}
}
}
if (_results.empty()) {
_radiusIncrement *= 2;
nextAnnulus();
}
} while (_results.empty() && _innerRadius < M_PI * _params.radius);
// TODO: consider shrinking _radiusIncrement if _results.size() meets some criteria.
}
// Fill _results with the next shell of results. We may have to search several times to do
// this. If _results.empty() after calling fillResults, there are no more possible results.
void S2NearCursor::fillResults() {
verify(_results.empty());
if (_innerRadius >= _outerRadius) { return; }
if (_innerRadius > _maxDistance) { return; }
// We iterate until 1. our search radius is too big or 2. we find results.
do {
// Some of these arguments are opaque, look at the definitions of the involved classes.
FieldRangeSet frs(_details->parentNS().c_str(), makeFRSObject(), false, false);
shared_ptr<FieldRangeVector> frv(new FieldRangeVector(frs, _specForFRV, 1));
scoped_ptr<BtreeCursor> cursor(BtreeCursor::make(nsdetails(_details->parentNS().c_str()),
*_details, frv, 0, 1));
// Do the actual search through this annulus.
size_t considered = 0;
for (; cursor->ok(); cursor->advance()) {
++considered;
MatchDetails details;
bool matched = _matcher->matchesCurrent(cursor.get(), &details);
if (!matched) { continue; }
const BSONObj& indexedObj = cursor->currLoc().obj();
size_t geoFieldsInRange = 0;
double minMatchingDistance = 1e20;
// Calculate the distance from our query point(s) to the geo field(s).
// For each geo field in the query...
for (size_t i = 0; i < _fields.size(); ++i) {
const QueryGeometry& field = _fields[i];
// Get all the fields with that name from the document.
BSONElementSet geoFieldElements;
indexedObj.getFieldsDotted(field.field, geoFieldElements, false);
if (geoFieldElements.empty()) { continue; }
// For each field with that name in the document...
for (BSONElementSet::iterator oi = geoFieldElements.begin();
oi != geoFieldElements.end(); ++oi) {
if (!oi->isABSONObj()) { continue; }
double dist = distanceBetween(field, oi->Obj());
// If it satisfies our distance criteria...
if (dist >= _innerRadius && dist <= _outerRadius) {
// Success! For this field.
++geoFieldsInRange;
minMatchingDistance = min(dist, minMatchingDistance);
}
}
}
// If all the geo query fields had something in range
if (_fields.size() == geoFieldsInRange) {
// The result is valid. We have to de-dup ourselves here.
if (_returned.end() == _returned.find(cursor->currLoc())) {
_results.push(Result(cursor->currLoc(), cursor->currKey(),
minMatchingDistance));
}
}
}
if (_results.empty()) {
_radiusIncrement *= 2;
nextAnnulus();
}
} while (_results.empty()
&& _innerRadius < _maxDistance
&& _innerRadius < _outerRadius
&& _innerRadius < M_PI * _params.radius);
// TODO: consider shrinking _radiusIncrement if _results.size() meets some criteria.
}
bool run2DSphereGeoNear(const IndexDetails &id, BSONObj& cmdObj, string& errmsg,
BSONObjBuilder& result) {
S2IndexType *idxType = static_cast<S2IndexType*>(id.getSpec().getType());
verify(&id == idxType->getDetails());
// We support both "num" and "limit" options to control limit
int numWanted = 100;
const char* limitName = cmdObj["num"].isNumber() ? "num" : "limit";
if (cmdObj[limitName].isNumber()) {
numWanted = cmdObj[limitName].numberInt();
verify(numWanted >= 0);
}
// Don't count any docs twice. Isn't this default behavior? Or will yields screw this up?
//bool uniqueDocs = false;
//if (!cmdObj["uniqueDocs"].eoo()) uniqueDocs = cmdObj["uniqueDocs"].trueValue();
// Add the location information to each result as a field with name 'loc'.
bool includeLocs = false;
if (!cmdObj["includeLocs"].eoo()) includeLocs = cmdObj["includeLocs"].trueValue();
// The actual query point
uassert(16551, "'near' param missing/invalid", !cmdObj["near"].eoo());
BSONObj nearObj = cmdObj["near"].embeddedObject();
// nearObj must be a point.
uassert(16571, "near must be called with a point, called with " + nearObj.toString(),
GeoParser::isPoint(nearObj));
// The non-near query part.
BSONObj query;
if (cmdObj["query"].isABSONObj())
query = cmdObj["query"].embeddedObject();
// The farthest away we're willing to look.
double maxDistance = numeric_limits<double>::max();
if (cmdObj["maxDistance"].isNumber())
maxDistance = cmdObj["maxDistance"].number();
vector<string> geoFieldNames;
idxType->getGeoFieldNames(&geoFieldNames);
uassert(16552, "geoNear called but no indexed geo fields?", 1 == geoFieldNames.size());
QueryGeometry queryGeo(geoFieldNames[0]);
uassert(16553, "geoNear couldn't parse geo: " + nearObj.toString(), queryGeo.parseFrom(nearObj));
vector<QueryGeometry> regions;
regions.push_back(queryGeo);
scoped_ptr<S2NearCursor> cursor(new S2NearCursor(idxType->keyPattern(), idxType->getDetails(),
query, regions, idxType->getParams(),
numWanted, maxDistance));
double totalDistance = 0;
int results = 0;
BSONObjBuilder resultBuilder(result.subarrayStart("results"));
double farthestDist = 0;
while (cursor->ok()) {
double dist = cursor->currentDistance();
totalDistance += dist;
if (dist > farthestDist) { farthestDist = dist; }
BSONObjBuilder oneResultBuilder(resultBuilder.subobjStart(BSONObjBuilder::numStr(results)));
oneResultBuilder.append("dis", dist);
if (includeLocs) {
BSONElementSet geoFieldElements;
cursor->current().getFieldsDotted(geoFieldNames[0], geoFieldElements, false);
for (BSONElementSet::iterator oi = geoFieldElements.begin();
oi != geoFieldElements.end(); ++oi) {
if (oi->isABSONObj()) {
oneResultBuilder.appendAs(*oi, "loc");
}
}
}
oneResultBuilder.append("obj", cursor->current());
oneResultBuilder.done();
++results;
cursor->advance();
}
resultBuilder.done();
BSONObjBuilder stats(result.subobjStart("stats"));
stats.append("time", cc().curop()->elapsedMillis());
stats.appendNumber("nscanned", cursor->nscanned());
stats.append("avgDistance", totalDistance / results);
stats.append("maxDistance", farthestDist);
stats.done();
return true;
}
// Fill _results with the next shell of results. We may have to search several times to do
// this. If _results.empty() after calling fillResults, there are no more possible results.
void S2NearCursor::fillResults() {
verify(_results.empty());
if (_innerRadius >= _outerRadius) { return; }
if (_innerRadius > _maxDistance) { return; }
if (0 == _numToReturn) { return; }
// We iterate until 1. our search radius is too big or 2. we find results.
do {
// Some of these arguments are opaque, look at the definitions of the involved classes.
FieldRangeSet frs(_details->parentNS().c_str(), makeFRSObject(), false, false);
shared_ptr<FieldRangeVector> frv(new FieldRangeVector(frs, _specForFRV, 1));
scoped_ptr<BtreeCursor> cursor(BtreeCursor::make(nsdetails(_details->parentNS()),
*_details, frv, 0, 1));
// The cursor may return the same obj more than once for a given
// FRS, so we make sure to only consider it once in any given annulus.
//
// We don't want this outside of the 'do' loop because the covering
// for an annulus may return an object whose distance to the query
// point is actually contained in a subsequent annulus. If we
// didn't consider every object in a given annulus we might miss
// the point.
//
// We don't use a global 'seen' because we get that by requiring
// the distance from the query point to the indexed geo to be
// within our 'current' annulus, and I want to dodge all yield
// issues if possible.
set<DiskLoc> seen;
LOG(1) << "looking at annulus from " << _innerRadius << " to " << _outerRadius << endl;
// Do the actual search through this annulus.
for (; cursor->ok(); cursor->advance()) {
++_nscanned;
if (seen.end() != seen.find(cursor->currLoc())) { continue; }
seen.insert(cursor->currLoc());
// Match against non-indexed fields.
++_matchTested;
MatchDetails details;
bool matched = _matcher->matchesCurrent(cursor.get(), &details);
if (!matched) { continue; }
const BSONObj& indexedObj = cursor->currLoc().obj();
++_geoTested;
// Match against indexed geo fields.
size_t geoFieldsMatched = 0;
// OK, cool, non-geo match satisfied. See if the object actually overlaps w/the geo
// query fields.
for (size_t i = 0; i < _indexedGeoFields.size(); ++i) {
BSONElementSet geoFieldElements;
indexedObj.getFieldsDotted(_indexedGeoFields[i].getField(), geoFieldElements,
false);
if (geoFieldElements.empty()) { continue; }
bool match = false;
for (BSONElementSet::iterator oi = geoFieldElements.begin();
!match && (oi != geoFieldElements.end()); ++oi) {
if (!oi->isABSONObj()) { continue; }
const BSONObj &geoObj = oi->Obj();
GeometryContainer geoContainer;
uassert(16699, "ill-formed geometry: " + geoObj.toString(),
geoContainer.parseFrom(geoObj));
match = _indexedGeoFields[i].satisfiesPredicate(geoContainer);
}
if (match) { ++geoFieldsMatched; }
}
if (geoFieldsMatched != _indexedGeoFields.size()) { continue; }
// Finally, see if the item is in our search annulus.
size_t geoFieldsInRange = 0;
double minMatchingDistance = 1e20;
// Get all the fields with that name from the document.
BSONElementSet geoFieldElements;
indexedObj.getFieldsDotted(_nearQuery.field, geoFieldElements, false);
if (geoFieldElements.empty()) { continue; }
// For each field with that name in the document...
for (BSONElementSet::iterator oi = geoFieldElements.begin();
oi != geoFieldElements.end(); ++oi) {
if (!oi->isABSONObj()) { continue; }
double dist = distanceTo(oi->Obj());
// If it satisfies our distance criteria...
if (dist >= _innerRadius && dist <= _outerRadius) {
// Success! For this field.
++geoFieldsInRange;
minMatchingDistance = min(dist, minMatchingDistance);
}
}
// If all the geo query fields had something in range
if (geoFieldsInRange > 0) {
// The result is valid. We have to de-dup ourselves here.
if (_returned.end() == _returned.find(cursor->currLoc())) {
_results.push(Result(cursor->currLoc(), cursor->currKey(),
minMatchingDistance));
}
}
}
if (_results.empty()) {
_radiusIncrement *= 2;
nextAnnulus();
}
} while (_results.empty()
&& _innerRadius < _maxDistance
&& _innerRadius < _outerRadius);
}
