From 41a2d01e419cabd2011ce4ce7262eb8b7181d9e7 Mon Sep 17 00:00:00 2001 From: Hari Khalsa Date: Mon, 29 Apr 2013 12:17:05 -0400 Subject: SERVER-8791 SERVER-9212 retire indexplugin/indextype --- src/mongo/db/cloner.h | 1 - src/mongo/db/db.cpp | 4 +- src/mongo/db/fts/SConscript | 1 - src/mongo/db/fts/fts_index.cpp | 96 ------- src/mongo/db/fts/fts_index.h | 64 ----- src/mongo/db/fts/fts_search.h | 1 - src/mongo/db/geo/2d.cpp | 458 ------------------------------ src/mongo/db/geo/2d.h | 23 -- src/mongo/db/geo/haystack.cpp | 269 +----------------- src/mongo/db/geo/s2index.cpp | 278 ------------------ src/mongo/db/hashindex.cpp | 112 -------- src/mongo/db/hashindex.h | 105 ------- src/mongo/db/index.cpp | 39 +-- src/mongo/db/index/2d_index_cursor.cpp | 10 +- src/mongo/db/index/catalog_hack.h | 54 ++++ src/mongo/db/index/hash_access_method.cpp | 19 +- src/mongo/db/index/hash_access_method.h | 7 +- src/mongo/db/index/hash_index_cursor.cpp | 4 +- src/mongo/db/index/s2_access_method.cpp | 2 +- src/mongo/db/index/s2_near_cursor.cpp | 3 +- src/mongo/db/index/s2_near_cursor.h | 2 +- src/mongo/db/index/s2_simple_cursor.cpp | 4 +- src/mongo/db/index_names.h | 22 ++ src/mongo/db/index_selection.cpp | 121 ++++---- src/mongo/db/indexkey.cpp | 391 +------------------------ src/mongo/db/indexkey.h | 152 +--------- src/mongo/db/pdfile.cpp | 12 +- src/mongo/db/query_optimizer_internal.cpp | 13 +- src/mongo/db/query_plan.cpp | 26 +- src/mongo/db/query_plan.h | 5 +- src/mongo/db/query_runner.cpp | 1 + src/mongo/db/queryutil.cpp | 42 ++- src/mongo/db/queryutil.h | 13 +- src/mongo/db/scanandorder.cpp | 2 +- src/mongo/db/scanandorder.h | 7 +- 35 files changed, 280 insertions(+), 2083 deletions(-) delete mode 100644 src/mongo/db/fts/fts_index.cpp delete mode 100644 src/mongo/db/fts/fts_index.h delete mode 100644 src/mongo/db/geo/2d.cpp delete mode 100644 src/mongo/db/geo/2d.h delete mode 100644 src/mongo/db/geo/s2index.cpp delete mode 100644 src/mongo/db/hashindex.cpp delete mode 100644 src/mongo/db/hashindex.h (limited to 'src/mongo/db') diff --git a/src/mongo/db/cloner.h b/src/mongo/db/cloner.h index 813725e38bb..400d8413753 100644 --- a/src/mongo/db/cloner.h +++ b/src/mongo/db/cloner.h @@ -23,7 +23,6 @@ namespace mongo { struct CloneOptions; - class IndexSpec; class DBClientBase; class DBClientCursor; class Query; diff --git a/src/mongo/db/db.cpp b/src/mongo/db/db.cpp index 2c42fec7dea..1513796145b 100644 --- a/src/mongo/db/db.cpp +++ b/src/mongo/db/db.cpp @@ -368,8 +368,8 @@ namespace mongo { for ( ; cursor && cursor->ok(); cursor->advance()) { const BSONObj index = cursor->current(); const BSONObj key = index.getObjectField("key"); - const string plugin = IndexPlugin::findPluginName(key); - if (IndexPlugin::existedBefore24(plugin)) + const string plugin = KeyPattern::findPluginName(key); + if (IndexNames::existedBefore24(plugin)) continue; log() << "Index " << index << " claims to be of type '" << plugin << "', " diff --git a/src/mongo/db/fts/SConscript b/src/mongo/db/fts/SConscript index 6534b73bdc1..7fc9298de9b 100644 --- a/src/mongo/db/fts/SConscript +++ b/src/mongo/db/fts/SConscript @@ -47,7 +47,6 @@ env.StaticLibrary( 'server_common', [ env.StaticLibrary('ftsmongod', [ 'fts_command_mongod.cpp', - 'fts_index.cpp', 'fts_search.cpp', ], LIBDEPS=["base","server_common"]) diff --git a/src/mongo/db/fts/fts_index.cpp b/src/mongo/db/fts/fts_index.cpp deleted file mode 100644 index 0f6c3995fce..00000000000 --- a/src/mongo/db/fts/fts_index.cpp +++ /dev/null @@ -1,96 +0,0 @@ -// fts_index.cpp - -/** -* Copyright (C) 2012 10gen Inc. -* -* This program is free software: you can redistribute it and/or modify -* it under the terms of the GNU Affero General Public License, version 3, -* as published by the Free Software Foundation. -* -* This program is distributed in the hope that it will be useful, -* but WITHOUT ANY WARRANTY; without even the implied warranty of -* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the -* GNU Affero General Public License for more details. -* -* You should have received a copy of the GNU Affero General Public License -* along with this program. If not, see . -*/ - -#include "mongo/pch.h" - -#include "mongo/base/init.h" -#include "mongo/db/client.h" -#include "mongo/db/fts/fts_enabled.h" -#include "mongo/db/fts/fts_index.h" -#include "mongo/db/fts/fts_index_format.h" -#include "mongo/util/mongoutils/str.h" -#include "mongo/util/stringutils.h" -#include "mongo/util/timer.h" -#include "mongo/db/pdfile.h" - -namespace mongo { - - namespace fts { - - using namespace mongoutils; - - /* - * extrapolates the weights vector - * and extra information from the spec - * @param plugin the index plugin for FTS - * @param spec the index specification - */ - FTSIndex::FTSIndex( const IndexPlugin* plugin, const IndexSpec* spec ) - : IndexType( plugin, spec ), _ftsSpec( spec->info ) { - } - - void FTSIndex::getKeys( const BSONObj& obj, BSONObjSet& keys) const { - FTSIndexFormat::getKeys( _ftsSpec, obj, &keys ); - } - - - FTSIndexPlugin::FTSIndexPlugin() : IndexPlugin( INDEX_NAME ) {} - - - /* - * Adjusts spec by appending information relative to the - * FTS Index (such as weights, index name, etc) - * @param spec, specification object - * - */ - BSONObj FTSIndexPlugin::adjustIndexSpec( const BSONObj& spec ) const { - StringData desc = cc().desc(); - if ( desc.find( "conn" ) == 0 ) { - // this is to make sure we only complain for users - // if you do get a text index created an a primary - // want it to index on the secondary as well - massert( 16633, "text search not enabled", isTextSearchEnabled() ); - } - return FTSSpec::fixSpec( spec ); - } - - /* - * Generates an FTSIndex with a spec and this plugin - * @param spec, specification to be used - */ - IndexType* FTSIndexPlugin::generate( const IndexSpec* spec ) const { - return new FTSIndex( this, spec ); - } - - void FTSIndexPlugin::postBuildHook( const IndexSpec& spec ) const { - string ns = spec.getDetails()->parentNS(); - NamespaceDetails* nsd = nsdetails( ns ); - if ( nsd->setUserFlag( NamespaceDetails::Flag_UsePowerOf2Sizes ) ) { - nsd->syncUserFlags( ns ); - } - } - - FTSIndexPlugin* ftsPlugin; - MONGO_INITIALIZER(FTSIndexPlugin)(InitializerContext* context) { - ftsPlugin = new FTSIndexPlugin(); - return Status::OK(); - } - - } - -} diff --git a/src/mongo/db/fts/fts_index.h b/src/mongo/db/fts/fts_index.h deleted file mode 100644 index a6a1378672f..00000000000 --- a/src/mongo/db/fts/fts_index.h +++ /dev/null @@ -1,64 +0,0 @@ -// fts_index.h - -/** -* Copyright (C) 2012 10gen Inc. -* -* This program is free software: you can redistribute it and/or modify -* it under the terms of the GNU Affero General Public License, version 3, -* as published by the Free Software Foundation. -* -* This program is distributed in the hope that it will be useful, -* but WITHOUT ANY WARRANTY; without even the implied warranty of -* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the -* GNU Affero General Public License for more details. -* -* You should have received a copy of the GNU Affero General Public License -* along with this program. If not, see . -*/ - -#pragma once - -#include -#include - -#include "mongo/db/fts/fts_spec.h" -#include "mongo/db/fts/fts_util.h" -#include "mongo/db/fts/stemmer.h" -#include "mongo/db/fts/stop_words.h" -#include "mongo/db/fts/tokenizer.h" -#include "mongo/db/index.h" - -namespace mongo { - - namespace fts { - - class FTSIndex : public IndexType { - public: - - // index constructor, called when user enters ensureIndex command with fts flag - FTSIndex(const IndexPlugin *plugin, const IndexSpec* spec); - - void getKeys( const BSONObj& obj, BSONObjSet& keys) const; - - const FTSSpec& getFtsSpec() const { return _ftsSpec; } - - private: - - FTSSpec _ftsSpec; - }; - - - class FTSIndexPlugin : public IndexPlugin { - public: - FTSIndexPlugin(); - - IndexType* generate( const IndexSpec* spec ) const; - - BSONObj adjustIndexSpec( const BSONObj& spec ) const; - - void postBuildHook( const IndexSpec& spec ) const; - - }; - - } //namespace fts -} //namespace mongo diff --git a/src/mongo/db/fts/fts_search.h b/src/mongo/db/fts/fts_search.h index 6d6c4c09f5f..9bc2ede77c0 100644 --- a/src/mongo/db/fts/fts_search.h +++ b/src/mongo/db/fts/fts_search.h @@ -24,7 +24,6 @@ #include #include "mongo/base/disallow_copying.h" -#include "mongo/db/fts/fts_index.h" #include "mongo/db/fts/fts_matcher.h" #include "mongo/db/fts/fts_query.h" #include "mongo/db/fts/fts_util.h" diff --git a/src/mongo/db/geo/2d.cpp b/src/mongo/db/geo/2d.cpp deleted file mode 100644 index 21ecfac92ae..00000000000 --- a/src/mongo/db/geo/2d.cpp +++ /dev/null @@ -1,458 +0,0 @@ -/** -* Copyright (C) 2008 10gen Inc. -* -* This program is free software: you can redistribute it and/or modify -* it under the terms of the GNU Affero General Public License, version 3, -* as published by the Free Software Foundation. -* -* This program is distributed in the hope that it will be useful, -* but WITHOUT ANY WARRANTY; without even the implied warranty of -* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the -* GNU Affero General Public License for more details. -* -* You should have received a copy of the GNU Affero General Public License -* along with this program. If not, see . -*/ - -#include "pch.h" - -#include - -#include "mongo/db/auth/action_set.h" -#include "mongo/db/auth/action_type.h" -#include "mongo/db/auth/privilege.h" -#include "mongo/db/namespace-inl.h" -#include "mongo/db/jsobj.h" -#include "mongo/db/index.h" -#include "mongo/util/startup_test.h" -#include "mongo/db/commands.h" -#include "mongo/db/pdfile.h" -#include "mongo/db/btreecursor.h" -#include "mongo/db/curop-inl.h" -#include "mongo/db/matcher.h" -#include "mongo/db/queryutil.h" -#include "mongo/db/geo/core.h" -#include "mongo/db/geo/geonear.h" -#include "mongo/db/geo/hash.h" -#include "mongo/db/geo/shapes.h" -#include "mongo/util/timer.h" - -namespace mongo { - - enum GeoDistType { - GEO_PLAIN, - GEO_SPHERE - }; - - const string GEO2DNAME = "2d"; - - class Geo2dType : public IndexType { - public: - virtual ~Geo2dType() { } - - Geo2dType(const IndexPlugin *plugin, const IndexSpec* spec) : IndexType(plugin, spec) { - BSONObjIterator i(spec->keyPattern); - while (i.more()) { - BSONElement e = i.next(); - if (e.type() == String && GEO2DNAME == e.valuestr()) { - uassert(13022, "can't have 2 geo field", _geo.size() == 0); - uassert(13023, "2d has to be first in index", _other.size() == 0); - _geo = e.fieldName(); - } else { - int order = 1; - if (e.isNumber()) { - order = static_cast(e.Number()); - } - _other.push_back(make_pair(e.fieldName(), order)); - } - } - uassert(13024, "no geo field specified", _geo.size()); - - double bits = configValueWithDefault(spec, "bits", 26); // for lat/long, ~ 1ft - uassert(13028, "bits in geo index must be between 1 and 32", bits > 0 && bits <= 32); - - GeoHashConverter::Parameters params; - params.bits = static_cast(bits); - params.max = configValueWithDefault(spec, "max", 180.0); - params.min = configValueWithDefault(spec, "min", -180.0); - double numBuckets = (1024 * 1024 * 1024 * 4.0); - params.scaling = numBuckets / (params.max - params.min); - - _geoHashConverter.reset(new GeoHashConverter(params)); - } - - // XXX: what does this do - virtual BSONObj fixKey(const BSONObj& in) { - if (in.firstElement().type() == BinData) - return in; - - BSONObjBuilder b(in.objsize() + 16); - - if (in.firstElement().isABSONObj()) - _geoHashConverter->hash(in.firstElement().embeddedObject()).appendToBuilder(&b, ""); - else if (in.firstElement().type() == String) - GeoHash(in.firstElement().valuestr()).appendToBuilder(&b, ""); - else if (in.firstElement().type() == RegEx) - GeoHash(in.firstElement().regex()).appendToBuilder(&b, ""); - else - return in; - - BSONObjIterator i(in); - i.next(); - while (i.more()) - b.append(i.next()); - return b.obj(); - } - - /** Finds the key objects to put in an index */ - virtual void getKeys(const BSONObj& obj, BSONObjSet& keys) const { - getKeys(obj, &keys, NULL); - } - - /** Finds all locations in a geo-indexed object */ - // TODO: Can we just return references to the locs, if they won't change? - void getKeys(const BSONObj& obj, vector& locs) const { - getKeys(obj, NULL, &locs); - } - - /** Finds the key objects and/or locations for a geo-indexed object */ - void getKeys(const BSONObj &obj, BSONObjSet* keys, vector* locs) const { - BSONElementMSet bSet; - - // Get all the nested location fields, but don't return individual elements from - // the last array, if it exists. - obj.getFieldsDotted(_geo.c_str(), bSet, false); - - if (bSet.empty()) - return; - - for (BSONElementMSet::iterator setI = bSet.begin(); setI != bSet.end(); ++setI) { - BSONElement geo = *setI; - - GEODEBUG("Element " << geo << " found for query " << _geo.c_str()); - - if (geo.eoo() || !geo.isABSONObj()) - continue; - - // - // Grammar for location lookup: - // locs ::= [loc,loc,...,loc]|{:loc,:loc,...,:loc}|loc - // loc ::= { : #, : # }|[#, #]|{} - // - // Empty locations are ignored, preserving single-location semantics - // - - BSONObj embed = geo.embeddedObject(); - if (embed.isEmpty()) - continue; - - // Differentiate between location arrays and locations - // by seeing if the first element value is a number - bool singleElement = embed.firstElement().isNumber(); - - BSONObjIterator oi(embed); - - while (oi.more()) { - BSONObj locObj; - - if (singleElement) { - locObj = embed; - } else { - BSONElement locElement = oi.next(); - - uassert(13654, str::stream() << "location object expected, location " - "array not in correct format", - locElement.isABSONObj()); - - locObj = locElement.embeddedObject(); - if(locObj.isEmpty()) - continue; - } - - BSONObjBuilder b(64); - - // Remember the actual location object if needed - if (locs) - locs->push_back(locObj); - - // Stop if we don't need to get anything but location objects - if (!keys) { - if (singleElement) break; - else continue; - } - - _geoHashConverter->hash(locObj, &obj).appendToBuilder(&b, ""); - - // Go through all the other index keys - for (vector >::const_iterator i = _other.begin(); - i != _other.end(); ++i) { - // Get *all* fields for the index key - BSONElementSet eSet; - obj.getFieldsDotted(i->first, eSet); - - if (eSet.size() == 0) - b.appendAs(_spec->missingField(), ""); - else if (eSet.size() == 1) - b.appendAs(*(eSet.begin()), ""); - else { - // If we have more than one key, store as an array of the objects - BSONArrayBuilder aBuilder; - - for (BSONElementSet::iterator ei = eSet.begin(); ei != eSet.end(); - ++ei) { - aBuilder.append(*ei); - } - - b.append("", aBuilder.arr()); - } - } - keys->insert(b.obj()); - if(singleElement) break; - } - } - } - - const IndexDetails* getDetails() const { return _spec->getDetails(); } - - const GeoHashConverter& getConverter() const { return *_geoHashConverter; } - - // XXX: make private with a getter - string _geo; - vector > _other; - private: - double configValueWithDefault(const IndexSpec* spec, const string& name, double def) { - BSONElement e = spec->info[name]; - if (e.isNumber()) { - return e.numberDouble(); - } - return def; - } - - scoped_ptr _geoHashConverter; - }; - - class Geo2dPlugin : public IndexPlugin { - public: - Geo2dPlugin() : IndexPlugin(GEO2DNAME) { } - - virtual IndexType* generate(const IndexSpec* spec) const { - return new Geo2dType(this, spec); - } - } geo2dplugin; - - void __forceLinkGeoPlugin() { - geo2dplugin.getName(); - } - - struct GeoUnitTest : public StartupTest { - int round(double d) { - return (int)(.5 + (d * 1000)); - } - -#define GEOHEQ(a,b) if (a.toString() != b){ cout << "[" << a.toString() << "] != [" << b << "]" << endl; verify(a == GeoHash(b)); } - - void run() { - verify(!GeoHash::isBitSet(0, 0)); - verify(!GeoHash::isBitSet(0, 31)); - verify(GeoHash::isBitSet(1, 31)); - - IndexSpec i(BSON("loc" << "2d")); - Geo2dType g(&geo2dplugin, &i); - const GeoHashConverter &conv = g.getConverter(); - - { - double x = 73.01212; - double y = 41.352964; - BSONObj in = BSON("x" << x << "y" << y); - GeoHash h = conv.hash(in); - BSONObj out = conv.unhashToBSONObj(h); - verify(round(x) == round(out["x"].number())); - verify(round(y) == round(out["y"].number())); - verify(round(in["x"].number()) == round(out["x"].number())); - verify(round(in["y"].number()) == round(out["y"].number())); - } - { - double x = -73.01212; - double y = 41.352964; - BSONObj in = BSON("x" << x << "y" << y); - GeoHash h = conv.hash(in); - BSONObj out = conv.unhashToBSONObj(h); - verify(round(x) == round(out["x"].number())); - verify(round(y) == round(out["y"].number())); - verify(round(in["x"].number()) == round(out["x"].number())); - verify(round(in["y"].number()) == round(out["y"].number())); - } - { - GeoHash h("0000"); - h.move(0, 1); - GEOHEQ(h, "0001"); - h.move(0, -1); - GEOHEQ(h, "0000"); - - h = GeoHash("0001"); - h.move(0, 1); - GEOHEQ(h, "0100"); - h.move(0, -1); - GEOHEQ(h, "0001"); - - h = GeoHash("0000"); - h.move(1, 0); - GEOHEQ(h, "0010"); - } - { - Box b(5, 5, 2); - verify("(5,5) -->> (7,7)" == b.toString()); - } - { - GeoHash a = conv.hash(1, 1); - GeoHash b = conv.hash(4, 5); - verify(5 == (int)(conv.distanceBetweenHashes(a, b))); - a = conv.hash(50, 50); - b = conv.hash(42, 44); - verify(round(10) == round(conv.distanceBetweenHashes(a, b))); - } - { - GeoHash x("0000"); - verify(0 == x.getHash()); - x = GeoHash(0, 1, 32); - GEOHEQ(x, "0000000000000000000000000000000000000000000000000000000000000001") - - verify(GeoHash("1100").hasPrefix(GeoHash("11"))); - verify(!GeoHash("1000").hasPrefix(GeoHash("11"))); - } - { - GeoHash x("1010"); - GEOHEQ(x, "1010"); - GeoHash y = x + "01"; - GEOHEQ(y, "101001"); - } - { - GeoHash a = conv.hash(5, 5); - GeoHash b = conv.hash(5, 7); - GeoHash c = conv.hash(100, 100); - BSONObj oa = a.wrap(); - BSONObj ob = b.wrap(); - BSONObj oc = c.wrap(); - verify(oa.woCompare(ob) < 0); - verify(oa.woCompare(oc) < 0); - } - { - GeoHash x("000000"); - x.move(-1, 0); - GEOHEQ(x, "101010"); - x.move(1, -1); - GEOHEQ(x, "010101"); - x.move(0, 1); - GEOHEQ(x, "000000"); - } - { - GeoHash prefix("110011000000"); - GeoHash entry( "1100110000011100000111000001110000011100000111000001000000000000"); - verify(!entry.hasPrefix(prefix)); - entry = GeoHash("1100110000001100000111000001110000011100000111000001000000000000"); - verify(entry.toString().find(prefix.toString()) == 0); - verify(entry.hasPrefix(GeoHash("1100"))); - verify(entry.hasPrefix(prefix)); - } - { - GeoHash a = conv.hash(50, 50); - GeoHash b = conv.hash(48, 54); - verify(round(4.47214) == round(conv.distanceBetweenHashes(a, b))); - } - { - Box b(Point(29.762283, -95.364271), Point(29.764283000000002, -95.36227099999999)); - verify(b.inside(29.763, -95.363)); - verify(! b.inside(32.9570255, -96.1082497)); - verify(! b.inside(32.9570255, -96.1082497, .01)); - } - { - GeoHash a("11001111"); - verify(GeoHash("11") == a.commonPrefix(GeoHash("11"))); - verify(GeoHash("11") == a.commonPrefix(GeoHash("11110000"))); - } - { - int N = 10000; -#if 0 // XXX: we want to make sure the two unhash versions both work, but private. - { - Timer t; - for (int i = 0; i < N; i++) { - unsigned x = (unsigned)rand(); - unsigned y = (unsigned)rand(); - GeoHash h(x, y); - unsigned a, b; - h.unhash(&a, &b); - verify(a == x); - verify(b == y); - } - //cout << "slow: " << t.millis() << endl; - } -#endif - { - Timer t; - for (int i=0; i 2469 && dist < 2470); - } - { - Point BNA (-86.67, 36.12); - Point LAX (-118.40, 33.94); - Point JFK (-73.77694444, 40.63861111); - verify(spheredist_deg(BNA, BNA) < 1e-6); - verify(spheredist_deg(LAX, LAX) < 1e-6); - verify(spheredist_deg(JFK, JFK) < 1e-6); - - Point zero (0, 0); - Point antizero (0,-180); - - // these were known to cause NaN - verify(spheredist_deg(zero, zero) < 1e-6); - verify(fabs(M_PI-spheredist_deg(zero, antizero)) < 1e-6); - verify(fabs(M_PI-spheredist_deg(antizero, zero)) < 1e-6); - } - } - } - } geoUnitTest; -} diff --git a/src/mongo/db/geo/2d.h b/src/mongo/db/geo/2d.h deleted file mode 100644 index d54e1f766c6..00000000000 --- a/src/mongo/db/geo/2d.h +++ /dev/null @@ -1,23 +0,0 @@ -/** -* Copyright (C) 2012 10gen Inc. -* -* This program is free software: you can redistribute it and/or modify -* it under the terms of the GNU Affero General Public License, version 3, -* as published by the Free Software Foundation. -* -* This program is distributed in the hope that it will be useful, -* but WITHOUT ANY WARRANTY; without even the implied warranty of -* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the -* GNU Affero General Public License for more details. -* -* You should have received a copy of the GNU Affero General Public License -* along with this program. If not, see . -*/ - -namespace mongo { - // We need cmdObj and parsedArgs so we can print a useful error msg - // and pull other args out. - bool run2DGeoNear(const IndexDetails &id, const BSONObj& cmdObj, - const GeoNearArguments &parsedArgs, string& errmsg, - BSONObjBuilder& result); -} // namespace mongo diff --git a/src/mongo/db/geo/haystack.cpp b/src/mongo/db/geo/haystack.cpp index 6088d7c8f1e..84bb4daf4d2 100644 --- a/src/mongo/db/geo/haystack.cpp +++ b/src/mongo/db/geo/haystack.cpp @@ -1,8 +1,5 @@ -// XXX THIS FILE IS DEPRECATED. PLEASE DON'T MODIFY. -// db/geo/haystack.cpp - /** - * Copyright (C) 2008-2012 10gen Inc. + * Copyright (C) 2008-2013 10gen Inc. * * This program is free software: you can redistribute it and/or modify * it under the terms of the GNU Affero General Public License, version 3, @@ -17,8 +14,6 @@ * along with this program. If not, see . */ -#include "pch.h" - #include #include "mongo/db/auth/action_set.h" @@ -28,6 +23,7 @@ #include "mongo/db/index/haystack_access_method.h" #include "mongo/db/index/index_descriptor.h" #include "mongo/db/index/index_access_method.h" +#include "mongo/db/index_names.h" #include "mongo/db/namespace-inl.h" #include "mongo/db/jsobj.h" #include "mongo/db/index.h" @@ -42,268 +38,14 @@ #include "mongo/util/timer.h" /** - * Provides the geoHaystack index type and the command "geoSearch." * Examines all documents in a given radius of a given point. * Returns all documents that match a given search restriction. * See http://dochub.mongodb.org/core/haystackindexes * * Use when you want to look for restaurants within 25 miles with a certain name. - * Don't use when you want to find the closest open restaurants; see 2d.cpp for that. + * Don't use when you want to find the closest open restaurants. */ namespace mongo { - static const string GEOSEARCHNAME = "geoHaystack"; - - class GeoHaystackSearchHopper { - public: - /** - * Constructed with a point, a max distance from that point, and a max number of - * matched points to store. - * @param n The centroid that we're searching - * @param maxDistance The maximum distance to consider from that point - * @param limit The maximum number of results to return - * @param geoField Which field in the provided DiskLoc has the point to test. - */ - GeoHaystackSearchHopper(const BSONObj& nearObj, double maxDistance, unsigned limit, - const string& geoField) - : _near(nearObj), _maxDistance(maxDistance), _limit(limit), _geoField(geoField) { } - - // Consider the point in loc, and keep it if it's within _maxDistance (and we have space for - // it) - void consider(const DiskLoc& loc) { - if (limitReached()) return; - Point p(loc.obj().getFieldDotted(_geoField)); - if (distance(_near, p) > _maxDistance) - return; - _locs.push_back(loc); - } - - int appendResultsTo(BSONArrayBuilder* b) { - for (unsigned i = 0; i <_locs.size(); i++) - b->append(_locs[i].obj()); - return _locs.size(); - } - - // Have we stored as many points as we can? - const bool limitReached() const { - return _locs.size() >= _limit; - } - private: - Point _near; - double _maxDistance; - unsigned _limit; - const string _geoField; - vector _locs; - }; - - /** - * Provides the IndexType for geoSearch. - * Maps (lat, lng) to the bucketSize-sided square bucket that contains it. - * Usage: - * db.foo.ensureIndex({ pos : "geoHaystack", type : 1 }, { bucketSize : 1 }) - * pos is the name of the field to be indexed that has lat/lng data in an array. - * type is the name of the secondary field to be indexed. - * bucketSize specifies the dimension of the square bucket for the data in pos. - * ALL fields are mandatory. - */ - class GeoHaystackSearchIndex : public IndexType { - public: - GeoHaystackSearchIndex(const IndexPlugin* plugin, const IndexSpec* spec) - : IndexType(plugin, spec) { - - BSONElement e = spec->info["bucketSize"]; - uassert(13321, "need bucketSize", e.isNumber()); - _bucketSize = e.numberDouble(); - uassert(16455, "bucketSize cannot be zero", _bucketSize != 0.0); - - // Example: - // db.foo.ensureIndex({ pos : "geoHaystack", type : 1 }, { bucketSize : 1 }) - BSONObjIterator i(spec->keyPattern); - while (i.more()) { - BSONElement e = i.next(); - if (e.type() == String && GEOSEARCHNAME == e.valuestr()) { - uassert(13314, "can't have more than one geo field", _geoField.size() == 0); - uassert(13315, "the geo field has to be first in index", - _otherFields.size() == 0); - _geoField = e.fieldName(); - } else { - // TODO(hk): Do we want to do any checking on e.type and e.valuestr? - uassert(13326, "geoSearch can only have 1 non-geo field for now", - _otherFields.size() == 0); - _otherFields.push_back(e.fieldName()); - } - } - - uassert(13316, "no geo field specified", _geoField.size()); - // XXX: Fix documentation that says the other field is optional; code says it's mandatory. - uassert(13317, "no non-geo fields specified", _otherFields.size()); - } - - void getKeys(const BSONObj &obj, BSONObjSet &keys) const { - BSONElement loc = obj.getFieldDotted(_geoField); - if (loc.eoo()) - return; - - uassert(13323, "latlng not an array", loc.isABSONObj()); - string root; - { - BSONObjIterator i(loc.Obj()); - BSONElement x = i.next(); - BSONElement y = i.next(); - root = makeString(hash(x), hash(y)); - } - - verify(_otherFields.size() == 1); - - BSONElementSet all; - - // This is getFieldsDotted (plural not singular) since the object we're indexing - // may be an array. - obj.getFieldsDotted(_otherFields[0], all); - - if (all.size() == 0) { - // We're indexing a document that doesn't have the secondary non-geo field present. - // XXX: do we want to add this even if all.size() > 0? result:empty search terms - // match everything instead of only things w/empty search terms) - addKey(root, BSONElement(), keys); - } else { - // Ex:If our secondary field is type: "foo" or type: {a:"foo", b:"bar"}, - // all.size()==1. We can query on the complete field. - // Ex: If our secondary field is type: ["A", "B"] all.size()==2 and all has values - // "A" and "B". The query looks for any of the fields in the array. - for (BSONElementSet::iterator i = all.begin(); i != all.end(); ++i) { - addKey(root, *i, keys); - } - } - } - - void searchCommand(NamespaceDetails* nsd, - const BSONObj& n /*near*/, double maxDistance, const BSONObj& search, - BSONObjBuilder& result, unsigned limit) { - Timer t; - - LOG(1) << "SEARCH near:" << n << " maxDistance:" << maxDistance - << " search: " << search << endl; - int x, y; - { - BSONObjIterator i(n); - x = hash(i.next()); - y = hash(i.next()); - } - int scale = static_cast(ceil(maxDistance / _bucketSize)); - - GeoHaystackSearchHopper hopper(n, maxDistance, limit, _geoField); - - long long btreeMatches = 0; - - // TODO(hk): Consider starting with a (or b)=0, then going to a=+-1, then a=+-2, etc. - // Would want a HaystackKeyIterator or similar for this, but it'd be a nice - // encapsulation allowing us to S2-ify this trivially/abstract the key details. - for (int a = -scale; a <= scale && !hopper.limitReached(); ++a) { - for (int b = -scale; b <= scale && !hopper.limitReached(); ++b) { - BSONObjBuilder bb; - bb.append("", makeString(x + a, y + b)); - - for (unsigned i = 0; i < _otherFields.size(); i++) { - // See if the non-geo field we're indexing on is in the provided search term. - BSONElement e = search.getFieldDotted(_otherFields[i]); - if (e.eoo()) - bb.appendNull(""); - else - bb.appendAs(e, ""); - } - - BSONObj key = bb.obj(); - - GEOQUADDEBUG("KEY: " << key); - - // TODO(hk): this keeps a set of all DiskLoc seen in this pass so that we don't - // consider the element twice. Do we want to instead store a hash of the set? - // Is this often big? - set thisPass; - - // Lookup from key to key, inclusive. - scoped_ptr cursor(BtreeCursor::make(nsd, - *getDetails(), - key, - key, - true, - 1)); - while (cursor->ok() && !hopper.limitReached()) { - pair::iterator, bool> p = thisPass.insert(cursor->currLoc()); - // If a new element was inserted (haven't seen the DiskLoc before), p.second - // is true. - if (p.second) { - hopper.consider(cursor->currLoc()); - GEOQUADDEBUG("\t" << cursor->current()); - btreeMatches++; - } - cursor->advance(); - } - } - } - - BSONArrayBuilder arr(result.subarrayStart("results")); - int num = hopper.appendResultsTo(&arr); - arr.done(); - - { - BSONObjBuilder b(result.subobjStart("stats")); - b.append("time", t.millis()); - b.appendNumber("btreeMatches", btreeMatches); - b.append("n", num); - b.done(); - } - } - - const IndexDetails* getDetails() const { - return _spec->getDetails(); - } - private: - // TODO(hk): consider moving hash/unhash/makeString out - int hash(const BSONElement& e) const { - uassert(13322, "geo field is not a number", e.isNumber()); - return hash(e.numberDouble()); - } - - int hash(double d) const { - d += 180; - d /= _bucketSize; - return static_cast(d); - } - - string makeString(int hashedX, int hashedY) const { - stringstream ss; - ss << hashedX << "_" << hashedY; - return ss.str(); - } - - // Build a new BSONObj with root in it. If e is non-empty, append that to the key. Insert - // the BSONObj into keys. - void addKey(const string& root, const BSONElement& e, BSONObjSet& keys) const { - BSONObjBuilder buf; - buf.append("", root); - - if (e.eoo()) - buf.appendNull(""); - else - buf.appendAs(e, ""); - - keys.insert(buf.obj()); - } - - string _geoField; - vector _otherFields; - double _bucketSize; - }; - - class GeoHaystackSearchIndexPlugin : public IndexPlugin { - public: - GeoHaystackSearchIndexPlugin() : IndexPlugin(GEOSEARCHNAME) { } - - virtual IndexType* generate(const IndexSpec* spec) const { - return new GeoHaystackSearchIndex(this, spec); - } - } nameIndexPlugin; class GeoHaystackSearchCommand : public Command { public: @@ -330,7 +72,7 @@ namespace mongo { } vector idxs; - nsd->findIndexByType(GEOSEARCHNAME, idxs); + nsd->findIndexByType(IndexNames::GEO_HAYSTACK, idxs); if (idxs.size() == 0) { errmsg = "no geoSearch index"; return false; @@ -360,4 +102,5 @@ namespace mongo { return 1; } } nameSearchCommand; -} + +} // namespace mongo diff --git a/src/mongo/db/geo/s2index.cpp b/src/mongo/db/geo/s2index.cpp deleted file mode 100644 index 568a8db6145..00000000000 --- a/src/mongo/db/geo/s2index.cpp +++ /dev/null @@ -1,278 +0,0 @@ -// XXX THIS FILE IS DEPRECATED. PLEASE DON'T MODIFY. -/** -* Copyright (C) 2012 10gen Inc. -* -* This program is free software: you can redistribute it and/or modify -* it under the terms of the GNU Affero General Public License, version 3, -* as published by the Free Software Foundation. -* -* This program is distributed in the hope that it will be useful, -* but WITHOUT ANY WARRANTY; without even the implied warranty of -* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the -* GNU Affero General Public License for more details. -* -* You should have received a copy of the GNU Affero General Public License -* along with this program. If not, see . -*/ - -#include "mongo/db/namespace-inl.h" -#include "mongo/db/client.h" -#include "mongo/db/curop.h" -#include "mongo/db/jsobj.h" -#include "mongo/db/index.h" -#include "mongo/db/queryutil.h" -#include "mongo/db/geo/geonear.h" -#include "mongo/db/geo/geoparser.h" -#include "mongo/db/geo/geoquery.h" -#include "mongo/db/index/s2_common.h" -#include "third_party/s2/s2.h" -#include "third_party/s2/s2cell.h" -#include "third_party/s2/s2polygon.h" -#include "third_party/s2/s2polyline.h" -#include "third_party/s2/s2regioncoverer.h" - -namespace { - // Used in a handful of places in GeoSphere2DType below. - static void keysFromRegion(S2RegionCoverer *coverer, const S2Region ®ion, - vector *out) { - vector covering; - coverer->GetCovering(region, &covering); - for (size_t i = 0; i < covering.size(); ++i) { - out->push_back(covering[i].toString()); - } - } -} // namespace - -namespace mongo { - class S2IndexType : public IndexType { - public: - // We keep track of what fields we've indexed and if they're geo or not. - struct IndexedField { - enum Type { - GEO, - LITERAL - }; - - Type type; - string name; - IndexedField(Type t, const string& n) : type(t), name(n) { } - }; - - S2IndexType(const string& geoIdxName, const IndexPlugin *plugin, const IndexSpec *spec, - const S2IndexingParams ¶ms) : IndexType(plugin, spec), _params(params) { - int geoFields = 0; - // Categorize the fields we're indexing and make sure we have a geo field. - BSONObjIterator i(spec->keyPattern); - while (i.more()) { - BSONElement e = i.next(); - if (e.type() == String && geoIdxName == e.valuestr()) { - _fields.push_back(IndexedField(IndexedField::GEO, e.fieldName())); - ++geoFields; - } else { - _fields.push_back(IndexedField(IndexedField::LITERAL, e.fieldName())); - } - } - uassert(16450, "Expect at least one geo field, spec=" + spec->keyPattern.toString(), - geoFields >= 1); - } - - virtual ~S2IndexType() { } - - void getKeys(const BSONObj& obj, BSONObjSet& keys) const { - verify(_fields.size() >= 1); - - BSONObjSet keysToAdd; - // We output keys in the same order as the fields we index. - for (size_t i = 0; i < _fields.size(); ++i) { - const IndexedField &field = _fields[i]; - - // First, we get the keys that this field adds. Either they're added literally from - // the value of the field, or they're transformed if the field is geo. - BSONElementSet fieldElements; - // false means Don't expand the last array, duh. - obj.getFieldsDotted(field.name, fieldElements, false); - - BSONObjSet keysForThisField; - if (IndexedField::GEO == field.type) { - getGeoKeys(fieldElements, &keysForThisField); - } else if (IndexedField::LITERAL == field.type) { - getLiteralKeys(fieldElements, &keysForThisField); - } else { - verify(0); - } - - // We expect there to be _spec->_missingField() present in the keys if data is - // missing. So, this should be non-empty. - verify(!keysForThisField.empty()); - - // We take the Cartesian product of all of the keys. This requires that we have - // some keys to take the Cartesian product with. If keysToAdd.empty(), we - // initialize it. - if (keysToAdd.empty()) { - keysToAdd = keysForThisField; - continue; - } - - BSONObjSet updatedKeysToAdd; - for (BSONObjSet::const_iterator it = keysToAdd.begin(); it != keysToAdd.end(); - ++it) { - for (BSONObjSet::const_iterator newIt = keysForThisField.begin(); - newIt!= keysForThisField.end(); ++newIt) { - BSONObjBuilder b; - b.appendElements(*it); - b.append(newIt->firstElement()); - updatedKeysToAdd.insert(b.obj()); - } - } - keysToAdd = updatedKeysToAdd; - } - - if (keysToAdd.size() > _params.maxKeysPerInsert) { - warning() << "insert of geo object generated lots of keys (" << keysToAdd.size() - << ") consider creating larger buckets. obj=" - << obj; - } - - for (BSONObjSet::const_iterator it = keysToAdd.begin(); it != keysToAdd.end(); ++it) { - keys.insert(*it); - } - } - - const IndexDetails* getDetails() const { return _spec->getDetails(); } - - // These are used by the geoNear command. geoNear constructs its own cursor. - const S2IndexingParams& getParams() const { return _params; } - void getGeoFieldNames(vector *out) const { - for (size_t i = 0; i < _fields.size(); ++i) { - if (IndexedField::GEO == _fields[i].type) { - out->push_back(_fields[i].name); - } - } - } - private: - // 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) { - uassert(16700, "Can't parse geometry from element: " + i->toString(), - i->isABSONObj()); - const BSONObj &obj = i->Obj(); - - vector 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(16572, "Can't extract geo keys from object, malformed geometry?:" - + obj.toString()); - } - uassert(16673, "Unable to generate keys for (likely malformed) geometry: " - + obj.toString(), - cells.size() > 0); - - for (vector::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()); - } - } - - void getLiteralKeysArray(BSONObj obj, BSONObjSet *out) const { - BSONObjIterator objIt(obj); - if (!objIt.more()) { - // Empty arrays are indexed as undefined. - BSONObjBuilder b; - b.appendUndefined(""); - out->insert(b.obj()); - } else { - // Non-empty arrays are exploded. - while (objIt.more()) { - BSONObjBuilder b; - b.appendAs(objIt.next(), ""); - out->insert(b.obj()); - } - } - } - - void getOneLiteralKey(BSONElement elt, BSONObjSet *out) const { - if (Array == elt.type()) { - getLiteralKeysArray(elt.Obj(), out); - } else { - // One thing, not an array, index as-is. - BSONObjBuilder b; - b.appendAs(elt, ""); - out->insert(b.obj()); - } - } - - // elements is a non-geo field. Add the values literally, expanding arrays. - void getLiteralKeys(const BSONElementSet &elements, BSONObjSet *out) const { - if (0 == elements.size()) { - // Missing fields are indexed as null. - BSONObjBuilder b; - b.appendNull(""); - out->insert(b.obj()); - } else { - for (BSONElementSet::iterator i = elements.begin(); i != elements.end(); ++i) { - getOneLiteralKey(*i, out); - } - } - } - - vector _fields; - S2IndexingParams _params; - }; - - static const string SPHERE_2D_NAME = "2dsphere"; - class S2IndexPlugin : public IndexPlugin { - public: - S2IndexPlugin() : IndexPlugin(SPHERE_2D_NAME) { } - - virtual IndexType* generate(const IndexSpec* spec) const { - S2IndexingParams params; - params.maxKeysPerInsert = 200; - // This is advisory. - params.maxCellsInCovering = 50; - // Near distances are specified in meters...sometimes. - params.radius = S2IndexingParams::kRadiusOfEarthInMeters; - // These are not advisory. - params.finestIndexedLevel = configValueWithDefault(spec, "finestIndexedLevel", - S2::kAvgEdge.GetClosestLevel(500.0 / params.radius)); - params.coarsestIndexedLevel = configValueWithDefault(spec, "coarsestIndexedLevel", - S2::kAvgEdge.GetClosestLevel(100 * 1000.0 / params.radius)); - uassert(16687, "coarsestIndexedLevel must be >= 0", params.coarsestIndexedLevel >= 0); - uassert(16688, "finestIndexedLevel must be <= 30", params.finestIndexedLevel <= 30); - uassert(16689, "finestIndexedLevel must be >= coarsestIndexedLevel", - params.finestIndexedLevel >= params.coarsestIndexedLevel); - return new S2IndexType(SPHERE_2D_NAME, this, spec, params); - } - - int configValueWithDefault(const IndexSpec* spec, const string& name, int def) const { - BSONElement e = spec->info[name]; - if (e.isNumber()) { return e.numberInt(); } - return def; - } - } S2IndexPluginS2D; - -} // namespace mongo diff --git a/src/mongo/db/hashindex.cpp b/src/mongo/db/hashindex.cpp deleted file mode 100644 index a658ba1467f..00000000000 --- a/src/mongo/db/hashindex.cpp +++ /dev/null @@ -1,112 +0,0 @@ -// XXX THIS FILE IS DEPRECATED. PLEASE DON'T MODIFY. - -// mongo/db/hashindex.cpp - -/** -* Copyright (C) 2012 10gen Inc. -* -* This program is free software: you can redistribute it and/or modify -* it under the terms of the GNU Affero General Public License, version 3, -* as published by the Free Software Foundation. -* -* This program is distributed in the hope that it will be useful, -* but WITHOUT ANY WARRANTY; without even the implied warranty of -* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the -* GNU Affero General Public License for more details. -* -* You should have received a copy of the GNU Affero General Public License -* along with this program. If not, see . -*/ - -#include "mongo/db/hashindex.h" - -#include "mongo/db/btreecursor.h" -#include "mongo/db/json.h" -#include "mongo/db/queryutil.h" - -namespace mongo { - - const string HashedIndexType::HASHED_INDEX_TYPE_IDENTIFIER = "hashed"; - - HashedIndexType::HashedIndexType( const IndexPlugin* plugin , const IndexSpec* spec ) : - IndexType( plugin , spec ) , _keyPattern( spec->keyPattern ) { - - //change these if single-field limitation lifted later - uassert( 16241 , "Currently only single field hashed index supported." , - _keyPattern.toBSON().nFields() == 1 ); - uassert( 16242 , "Currently hashed indexes cannot guarantee uniqueness. Use a regular index." , - ! (spec->info).getField("unique").booleanSafe() ); - - //Default _seed to 0 if "seed" is not included in the index spec - //or if the value of "seed" is not a number - _seed = (spec->info).getField("seed").numberInt(); - - //Default _isSparse to false if "sparse" is not included in the index spec - //or if the value of "sparse" is not a boolean - _isSparse = (spec->info).getField("sparse").booleanSafe(); - - //In case we have hashed indexes based on other hash functions in - //the future, we store a hashVersion number. If hashVersion changes, - // "makeSingleKey" will need to change accordingly. - //Defaults to 0 if "hashVersion" is not included in the index spec - //or if the value of "hashversion" is not a number - _hashVersion = (spec->info).getField("hashVersion").numberInt(); - - //Get the hashfield name - BSONElement firstElt = spec->keyPattern.firstElement(); - massert( 16243 , "error: no hashed index field" , - firstElt.str().compare( HASHED_INDEX_TYPE_IDENTIFIER ) == 0 ); - _hashedField = firstElt.fieldName(); - - // Explicit null valued fields and missing fields are both represented in hashed indexes - // using the hash value of the null BSONElement. This is partly for historical reasons - // (hash of null was used in the initial release of hashed indexes and changing would alter - // the data format). Additionally, in certain places the hashed index code and the index - // bound calculation code assume null and missing are indexed identically. - BSONObj nullObj = BSON( "" << BSONNULL ); - _missingKey = BSON( "" << makeSingleKey( nullObj.firstElement(), _seed, _hashVersion ) ); - } - - HashedIndexType::~HashedIndexType() { } - - void HashedIndexType::getKeys( const BSONObj &obj, BSONObjSet &keys ) const { - string hashedFieldCopy = string( _hashedField ); - const char* hashedFieldCopyPtr = hashedFieldCopy.c_str(); - BSONElement fieldVal = obj.getFieldDottedOrArray( hashedFieldCopyPtr ); - - uassert( 16244 , "Error: hashed indexes do not currently support array values" , fieldVal.type() != Array ); - - if ( ! fieldVal.eoo() ) { - BSONObj key = BSON( "" << makeSingleKey( fieldVal , _seed , _hashVersion ) ); - keys.insert( key ); - } - else if (! _isSparse ) { - keys.insert( _missingKey.copy() ); - } - } - - /* This class registers HASHED_INDEX_NAME in a global map of special index types - * Using this pattern, any index with the pattern, {fieldname : HASHED_INDEX_NAME} - * will be recognized as a HashedIndexType and the associated methods will be used. - */ - class HashedIndexPlugin : public IndexPlugin { - public: - - HashedIndexPlugin() : IndexPlugin( HashedIndexType::HASHED_INDEX_TYPE_IDENTIFIER ) {} - - virtual IndexType* generate( const IndexSpec* spec ) const { - return new HashedIndexType( this , spec ); - } - - } hashedIndexPlugin; - - - long long int HashedIndexType::makeSingleKey( const BSONElement& e , - HashSeed seed , - HashVersion v ) { - massert( 16245 , "Only HashVersion 0 has been defined" , v == 0 ); - return BSONElementHasher::hash64( e , seed ); - } - -} - diff --git a/src/mongo/db/hashindex.h b/src/mongo/db/hashindex.h deleted file mode 100644 index f7bfa4b4552..00000000000 --- a/src/mongo/db/hashindex.h +++ /dev/null @@ -1,105 +0,0 @@ -// XXX THIS FILE IS DEPRECATED. PLEASE DON'T MODIFY WITHOUT TALKING TO HK - -// hashindex.h - -/** -* Copyright (C) 2012 10gen Inc. -* -* This program is free software: you can redistribute it and/or modify -* it under the terms of the GNU Affero General Public License, version 3, -* as published by the Free Software Foundation. -* -* This program is distributed in the hope that it will be useful, -* but WITHOUT ANY WARRANTY; without even the implied warranty of -* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the -* GNU Affero General Public License for more details. -* -* You should have received a copy of the GNU Affero General Public License -* along with this program. If not, see . -*/ - -#include "mongo/db/hasher.h" -#include "mongo/db/index.h" -#include "mongo/db/keypattern.h" -#include "mongo/db/matcher.h" -#include "mongo/db/namespace-inl.h" -#include "mongo/db/pdfile.h" - -namespace mongo { - - /* This is an index where the keys are hashes of a given field. - * - * Optional arguments: - * "seed" : int (default = 0, a seed for the hash function) - * "hashVersion : int (default = 0, determines which hash function to use) - * - * Example use in the mongo shell: - * > db.foo.ensureIndex({a : "hashed"}, {seed : 3, hashVersion : 0}) - * - * LIMITATION: Only works with a single field. The HashedIndexType - * constructor uses uassert to ensure that the spec has the form - * { : "hashed"}, and not, for example, - * { a : "hashed" , b : 1} - * - * LIMITATION: Cannot be used as a unique index. - * The HashedIndexType constructor uses uassert to ensure that - * the spec does not contain {"unique" : true} - * - * LIMITATION: Cannot be used to index arrays. - * The getKeys function uasserts that value being inserted - * is not an array. This index will not be built if any - * array values of the hashed field exist. - * - */ - class HashedIndexType : public IndexType{ - public: - - static const string HASHED_INDEX_TYPE_IDENTIFIER; - typedef int HashVersion; - - /* Creates a new HashedIndex around a HashedIndexPlugin - * and an IndexSpec. New HashedIndexTypes are created via - * a factory method in the HashedIndexPlugin class. - */ - HashedIndexType( const IndexPlugin* plugin , const IndexSpec* spec ); - virtual ~HashedIndexType(); - - /* The input is "obj" which should have a field corresponding to the hashedfield. - * The output is a BSONObj with a single BSONElement whose value is the hash - * Eg if this is an index on "a" we have - * obj is {a : 45} --> key becomes {"" : hash(45) } - * - * Limitation: arrays values are not currently supported. This function uasserts - * that the value is not an array, and errors out in that case. - */ - void getKeys( const BSONObj &obj, BSONObjSet &keys ) const; - - /* A field missing from a document is represented by the hash value of a null BSONElement. - */ - BSONElement missingField() const { return _missingKey.firstElement(); } - - /* Takes a BSONElement, seed and hashVersion, and outputs the - * 64-bit hash used for this index - * E.g. if the element is {a : 3} this outputs v1-hash(3) - * */ - static long long int makeSingleKey( const BSONElement& e , - HashSeed seed , - HashVersion v = 0 ); - - /* Since the keys for this index are hashes, documents are not stored in order, - * thus we will need to perform scanAndOrder whenever the "order" is non-empty. - */ - bool scanAndOrderRequired( const BSONObj& query , const BSONObj& order ) const { - return ! order.isEmpty(); - } - - private: - string _hashedField; - KeyPattern _keyPattern; - HashSeed _seed; //defaults to zero if not in the IndexSpec - HashVersion _hashVersion; //defaults to zero if not in the IndexSpec - bool _isSparse; - BSONObj _missingKey; - }; - -} diff --git a/src/mongo/db/index.cpp b/src/mongo/db/index.cpp index c5ff0cb80e0..3a9c2b1fa9f 100644 --- a/src/mongo/db/index.cpp +++ b/src/mongo/db/index.cpp @@ -26,6 +26,8 @@ #include "mongo/db/auth/authorization_manager.h" #include "mongo/db/background.h" #include "mongo/db/btree.h" +#include "mongo/db/fts/fts_enabled.h" +#include "mongo/db/fts/fts_spec.h" #include "mongo/db/index/index_access_method.h" #include "mongo/db/index/index_cursor.h" #include "mongo/db/index/index_descriptor.h" @@ -128,7 +130,7 @@ namespace mongo { } static bool needToUpgradeMinorVersion(const string& newPluginName) { - if (IndexPlugin::existedBefore24(newPluginName)) + if (IndexNames::existedBefore24(newPluginName)) return false; DataFileHeader* dfh = cc().database()->getFile(0)->getHeader(); @@ -146,8 +148,8 @@ namespace mongo { for ( ; cursor && cursor->ok(); cursor->advance()) { const BSONObj index = cursor->current(); const BSONObj key = index.getObjectField("key"); - const string plugin = IndexPlugin::findPluginName(key); - if (IndexPlugin::existedBefore24(plugin)) + const string plugin = KeyPattern::findPluginName(key); + if (IndexNames::existedBefore24(plugin)) continue; const string errmsg = str::stream() @@ -243,13 +245,11 @@ namespace mongo { return false; } - string pluginName = IndexPlugin::findPluginName( key ); - IndexPlugin * plugin = NULL; + string pluginName = KeyPattern::findPluginName( key ); if (pluginName.size()) { - plugin = IndexPlugin::get(pluginName); uassert(16734, str::stream() << "Unknown index plugin '" << pluginName << "' " - << "in index "<< key - , plugin); + << "in index "<< key, + IndexNames::isKnownName(pluginName)); if (needToUpgradeMinorVersion(pluginName)) upgradeMinorVersionOrAssert(pluginName); @@ -257,8 +257,15 @@ namespace mongo { { BSONObj o = io; - if ( plugin ) { - o = plugin->adjustIndexSpec(o); + if (IndexNames::TEXT == pluginName || IndexNames::TEXT_INTERNAL == pluginName) { + StringData desc = cc().desc(); + if ( desc.find( "conn" ) == 0 ) { + // this is to make sure we only complain for users + // if you do get a text index created an a primary + // want it to index on the secondary as well + massert(16808, "text search not enabled", fts::isTextSearchEnabled() ); + } + o = fts::FTSSpec::fixSpec(o); } BSONObjBuilder b; int v = DefaultIndexVersionNumber; @@ -295,23 +302,17 @@ namespace mongo { } void IndexSpec::reset(const IndexDetails * details) { - const DataFileHeader* dfh = cc().database()->getFile(0)->getHeader(); - IndexSpec::PluginRules rules = dfh->versionMinor == PDFILE_VERSION_MINOR_24_AND_NEWER - ? IndexSpec::RulesFor24 - : IndexSpec::RulesFor22 - ; - _details = details; - reset(details->info, rules); + reset(details->info); } - void IndexSpec::reset(const BSONObj& _info, PluginRules rules) { + void IndexSpec::reset(const BSONObj& _info) { info = _info; keyPattern = info["key"].embeddedObjectUserCheck(); if ( keyPattern.objsize() == 0 ) { out() << info.toString() << endl; verify(false); } - _init(rules); + _init(); } } diff --git a/src/mongo/db/index/2d_index_cursor.cpp b/src/mongo/db/index/2d_index_cursor.cpp index 1c63db7ab46..48d7bf72c64 100644 --- a/src/mongo/db/index/2d_index_cursor.cpp +++ b/src/mongo/db/index/2d_index_cursor.cpp @@ -322,7 +322,7 @@ namespace mongo { scoped_ptr _cursor; scoped_ptr _frs; - scoped_ptr _spec; + BSONObj _keyPattern; BSONObj key() { return _cursor->currKey(); } @@ -390,11 +390,11 @@ namespace mongo { } BSONObj iSpec = bob.obj(); - min._spec.reset(new IndexSpec(iSpec)); - max._spec.reset(new IndexSpec(iSpec)); + min._keyPattern = iSpec; + max._keyPattern = iSpec; - shared_ptr frvMin(new FieldRangeVector(*min._frs, *min._spec, -1)); - shared_ptr frvMax(new FieldRangeVector(*max._frs, *max._spec, 1)); + shared_ptr frvMin(new FieldRangeVector(*min._frs, min._keyPattern, -1)); + shared_ptr frvMax(new FieldRangeVector(*max._frs, max._keyPattern, 1)); min._cursor.reset(BtreeCursor::make(nsdetails(descriptor->parentNS()), descriptor->getOnDisk(), frvMin, 0, -1)); diff --git a/src/mongo/db/index/catalog_hack.h b/src/mongo/db/index/catalog_hack.h index 17eaacf78e5..2b306b75462 100644 --- a/src/mongo/db/index/catalog_hack.h +++ b/src/mongo/db/index/catalog_hack.h @@ -27,6 +27,7 @@ #include "mongo/db/index/s2_access_method.h" #include "mongo/db/index_names.h" #include "mongo/db/keypattern.h" +#include "mongo/db/pdfile.h" namespace mongo { @@ -35,12 +36,61 @@ namespace mongo { */ class CatalogHack { public: + static bool shouldOverridePlugin(const BSONObj& keyPattern) { + string pluginName = KeyPattern::findPluginName(keyPattern); + bool known = IndexNames::isKnownName(pluginName); + + if (NULL == cc().database()) { + return false; + } + + const DataFileHeader* dfh = cc().database()->getFile(0)->getHeader(); + + if (dfh->versionMinor == PDFILE_VERSION_MINOR_24_AND_NEWER) { + // RulesFor24 + // This assert will be triggered when downgrading from a future version that + // supports an index plugin unsupported by this version. + uassert(16736, str::stream() << "Invalid index type '" << pluginName << "' " + << "in index " << keyPattern, + known); + return false; + } else { + // RulesFor22 + if (!known) { + log() << "warning: can't find plugin [" << pluginName << "]" << endl; + return true; + } + + if (!IndexNames::existedBefore24(pluginName)) { + warning() << "Treating index " << keyPattern << " as ascending since " + << "it was created before 2.4 and '" << pluginName << "' " + << "was not a valid type at that time." + << endl; + return true; + } else { + return false; + } + } + } + + static string findPluginName(const BSONObj& keyPattern) { + if (shouldOverridePlugin(keyPattern)) { + return ""; + } else { + return KeyPattern::findPluginName(keyPattern); + } + } + static IndexDescriptor* getDescriptor(NamespaceDetails* nsd, int idxNo) { IndexDetails& id = nsd->idx(idxNo); return new IndexDescriptor(nsd, idxNo, &id, id.info.obj()); } static BtreeBasedAccessMethod* getBtreeBasedIndex(IndexDescriptor* desc) { + if (shouldOverridePlugin(desc->keyPattern())) { + return new BtreeAccessMethod(desc); + } + string type = KeyPattern::findPluginName(desc->keyPattern()); if (IndexNames::HASHED == type) { return new HashAccessMethod(desc); @@ -62,6 +112,10 @@ namespace mongo { } static IndexAccessMethod* getIndex(IndexDescriptor* desc) { + if (shouldOverridePlugin(desc->keyPattern())) { + return new BtreeAccessMethod(desc); + } + string type = KeyPattern::findPluginName(desc->keyPattern()); if (IndexNames::HASHED == type) { return new HashAccessMethod(desc); diff --git a/src/mongo/db/index/hash_access_method.cpp b/src/mongo/db/index/hash_access_method.cpp index d547f7fce1b..13a966fecf0 100644 --- a/src/mongo/db/index/hash_access_method.cpp +++ b/src/mongo/db/index/hash_access_method.cpp @@ -28,6 +28,20 @@ namespace mongo { return BSONElementHasher::hash64(e, seed); } + BSONObj HashAccessMethod::getMissingField(const IndexDetails& details) { + BSONObj infoObj = details.info.obj(); + int hashVersion = infoObj["hashVersion"].numberInt(); + HashSeed seed = infoObj["seed"].numberInt(); + + // Explicit null valued fields and missing fields are both represented in hashed indexes + // using the hash value of the null BSONElement. This is partly for historical reasons + // (hash of null was used in the initial release of hashed indexes and changing would alter + // the data format). Additionally, in certain places the hashed index code and the index + // bound calculation code assume null and missing are indexed identically. + BSONObj nullObj = BSON("" << BSONNULL); + return BSON("" << makeSingleKey(nullObj.firstElement(), seed, hashVersion)); + } + HashAccessMethod::HashAccessMethod(IndexDescriptor* descriptor) : BtreeBasedAccessMethod(descriptor) { @@ -56,11 +70,6 @@ namespace mongo { firstElt.str().compare(HASHED_INDEX_TYPE_IDENTIFIER) == 0); _hashedField = firstElt.fieldName(); - // Explicit null valued fields and missing fields are both represented in hashed indexes - // using the hash value of the null BSONElement. This is partly for historical reasons - // (hash of null was used in the initial release of hashed indexes and changing would alter - // the data format). Additionally, in certain places the hashed index code and the index - // bound calculation code assume null and missing are indexed identically. BSONObj nullObj = BSON("" << BSONNULL); _missingKey = BSON("" << makeSingleKey(nullObj.firstElement(), _seed, _hashVersion)); } diff --git a/src/mongo/db/index/hash_access_method.h b/src/mongo/db/index/hash_access_method.h index 5a49b69bfe7..19060cf3f96 100644 --- a/src/mongo/db/index/hash_access_method.h +++ b/src/mongo/db/index/hash_access_method.h @@ -43,6 +43,10 @@ namespace mongo { return Status::OK(); } + // Our missing field is different than the default missing field, this needs to be + // exposed in s/d_split.cpp. That's the only thing that calls this. + static BSONObj getMissingField(const IndexDetails& details); + /** * Hashing function used by both this class and the cursors we create. */ @@ -60,9 +64,6 @@ namespace mongo { // _hashVersion defaults to zero. int _hashVersion; - // What key do we insert when the field is missing? - // TODO: fix migration code to do the right thing. - // TODO: see http://codereview.10gen.com/9497028/patch/3001/4007 BSONObj _missingKey; }; diff --git a/src/mongo/db/index/hash_index_cursor.cpp b/src/mongo/db/index/hash_index_cursor.cpp index 7909f5c843f..682ce9f8ca3 100644 --- a/src/mongo/db/index/hash_index_cursor.cpp +++ b/src/mongo/db/index/hash_index_cursor.cpp @@ -67,7 +67,6 @@ namespace mongo { inObj.done(); BSONObj newQuery = newQueryBuilder.obj(); - // FieldRangeVector needs an IndexSpec so we make it one. BSONObjBuilder specBuilder; BSONObjIterator it(_descriptor->keyPattern()); while (it.more()) { @@ -75,12 +74,11 @@ namespace mongo { specBuilder.append(e.fieldName(), 1); } BSONObj spec = specBuilder.obj(); - IndexSpec specForFRV(spec); //Use the point-intervals of the new query to create a Btree cursor FieldRangeSet newfrs( "" , newQuery , true, true ); shared_ptr newVector( - new FieldRangeVector( newfrs , specForFRV, 1 ) ); + new FieldRangeVector( newfrs , spec, 1 ) ); _oldCursor.reset( BtreeCursor::make(nsdetails(_descriptor->parentNS()), diff --git a/src/mongo/db/index/s2_access_method.cpp b/src/mongo/db/index/s2_access_method.cpp index 619775566c5..ca4fc60d2c8 100644 --- a/src/mongo/db/index/s2_access_method.cpp +++ b/src/mongo/db/index/s2_access_method.cpp @@ -110,7 +110,7 @@ namespace mongo { getLiteralKeys(fieldElements, &keysForThisField); } - // We expect there to be _spec->_missingField() present in the keys if data is + // We expect there to be the missing field element present in the keys if data is // missing. So, this should be non-empty. verify(!keysForThisField.empty()); diff --git a/src/mongo/db/index/s2_near_cursor.cpp b/src/mongo/db/index/s2_near_cursor.cpp index 7983ee4a915..780735e4b37 100644 --- a/src/mongo/db/index/s2_near_cursor.cpp +++ b/src/mongo/db/index/s2_near_cursor.cpp @@ -53,8 +53,7 @@ namespace mongo { BSONElement e = specIt.next(); specBuilder.append(e.fieldName(), 1); } - BSONObj spec = specBuilder.obj(); - _specForFRV = IndexSpec(spec); + _specForFRV = specBuilder.obj(); specIt = BSONObjIterator(_descriptor->keyPattern()); while (specIt.more()) { diff --git a/src/mongo/db/index/s2_near_cursor.h b/src/mongo/db/index/s2_near_cursor.h index 860b92a3a16..5d922db18c1 100644 --- a/src/mongo/db/index/s2_near_cursor.h +++ b/src/mongo/db/index/s2_near_cursor.h @@ -117,7 +117,7 @@ namespace mongo { S2IndexingParams _params; // We also pass this to the FieldRangeVector ctor. - IndexSpec _specForFRV; + BSONObj _specForFRV; // Geo-related variables. // What's the max distance (arc length) we're willing to look for results? diff --git a/src/mongo/db/index/s2_simple_cursor.cpp b/src/mongo/db/index/s2_simple_cursor.cpp index e6a6a4e9250..25c88aa8f3f 100644 --- a/src/mongo/db/index/s2_simple_cursor.cpp +++ b/src/mongo/db/index/s2_simple_cursor.cpp @@ -42,7 +42,6 @@ namespace mongo { // false means we want to filter OUT geoFieldsToNuke, not filter to include only that. _filteredQuery = query.filterFieldsUndotted(geoFieldsToNuke.obj(), false); - // FieldRangeVector needs an IndexSpec so we make it one. BSONObjBuilder specBuilder; BSONObjIterator i(_descriptor->keyPattern()); while (i.more()) { @@ -50,7 +49,6 @@ namespace mongo { specBuilder.append(e.fieldName(), 1); } BSONObj spec = specBuilder.obj(); - IndexSpec specForFRV(spec); BSONObj frsObj; @@ -74,7 +72,7 @@ namespace mongo { frsObj = frsObjBuilder.obj(); FieldRangeSet frs(_descriptor->parentNS().c_str(), frsObj, false, false); - shared_ptr frv(new FieldRangeVector(frs, specForFRV, 1)); + shared_ptr frv(new FieldRangeVector(frs, spec, 1)); _btreeCursor.reset(BtreeCursor::make(nsdetails(_descriptor->parentNS()), _descriptor->getOnDisk(), frv, 0, 1)); next(); diff --git a/src/mongo/db/index_names.h b/src/mongo/db/index_names.h index db16ab0067a..c2eef144ad9 100644 --- a/src/mongo/db/index_names.h +++ b/src/mongo/db/index_names.h @@ -34,6 +34,28 @@ namespace mongo { static const string TEXT; static const string TEXT_INTERNAL; static const string HASHED; + + /** + * True if is a regular (non-plugin) index or uses a plugin that existed before 2.4. + * These plugins are grandfathered in and allowed to exist in DBs with + * PDFILE_MINOR_VERSION_22_AND_OLDER + */ + static bool existedBefore24(const string& name) { + return name.empty() + || name == IndexNames::GEO_2D + || name == IndexNames::GEO_HAYSTACK + || name == IndexNames::HASHED; + } + + static bool isKnownName(const string& name) { + return name.empty() + || name == IndexNames::GEO_2D + || name == IndexNames::GEO_2DSPHERE + || name == IndexNames::GEO_HAYSTACK + || name == IndexNames::TEXT + || name == IndexNames::TEXT_INTERNAL + || name == IndexNames::HASHED; + } }; } // namespace mongo diff --git a/src/mongo/db/index_selection.cpp b/src/mongo/db/index_selection.cpp index deaa130d255..431d51f8e0a 100644 --- a/src/mongo/db/index_selection.cpp +++ b/src/mongo/db/index_selection.cpp @@ -16,6 +16,7 @@ #include "mongo/db/index_selection.h" +#include "mongo/db/index/catalog_hack.h" #include "mongo/db/index_names.h" #include "mongo/db/jsobj.h" #include "mongo/db/keypattern.h" @@ -27,68 +28,10 @@ namespace mongo { const FieldRangeSet& queryConstraints, const BSONObj& order) { - string type = KeyPattern::findPluginName(keyPattern); + string type = CatalogHack::findPluginName(keyPattern); BSONObj query = queryConstraints.originalQuery(); - if (IndexNames::HASHED == type) { - /* This index is only considered "HELPFUL" for a query - * if it's the union of at least one equality constraint on the - * hashed field. Otherwise it's considered USELESS. - * Example queries (supposing the indexKey is {a : "hashed"}): - * {a : 3} HELPFUL - * {a : 3 , b : 3} HELPFUL - * {a : {$in : [3,4]}} HELPFUL - * {a : {$gte : 3, $lte : 3}} HELPFUL - * {} USELESS - * {b : 3} USELESS - * {a : {$gt : 3}} USELESS - */ - BSONElement firstElt = keyPattern.firstElement(); - if (queryConstraints.isPointIntervalSet(firstElt.fieldName())) { - return HELPFUL; - } else { - return USELESS; - } - } else if (IndexNames::GEO_2DSPHERE == type) { - BSONObjIterator i(keyPattern); - while (i.more()) { - BSONElement ie = i.next(); - - if (ie.type() != String || IndexNames::GEO_2DSPHERE != ie.valuestr()) { - continue; - } - - BSONElement e = query.getFieldDotted(ie.fieldName()); - // Some locations are given to us as arrays. Sigh. - if (Array == e.type()) { return HELPFUL; } - if (Object != e.type()) { continue; } - // getGtLtOp is horribly misnamed and really means get the operation. - switch (e.embeddedObject().firstElement().getGtLtOp()) { - case BSONObj::opNEAR: - return OPTIMAL; - case BSONObj::opWITHIN: { - BSONElement elt = e.embeddedObject().firstElement(); - if (Object != elt.type()) { continue; } - const char* fname = elt.embeddedObject().firstElement().fieldName(); - if (mongoutils::str::equals("$geometry", fname) - || mongoutils::str::equals("$centerSphere", fname)) { - return OPTIMAL; - } else { - return USELESS; - } - } - case BSONObj::opGEO_INTERSECTS: - return OPTIMAL; - default: - return USELESS; - } - } - return USELESS; - } else if (IndexNames::TEXT == type || IndexNames::TEXT_INTERNAL == type) { - return USELESS; - } else if (IndexNames::GEO_HAYSTACK == type) { - return USELESS; - } else if ("" == type) { + if ("" == type) { // This is a quick first pass to determine the suitability of the index. It produces // some false positives (returns HELPFUL for some indexes which are not particularly). // When we return HELPFUL a more precise determination of utility is done by the query @@ -160,6 +103,64 @@ namespace mongo { default: return USELESS; } + } else if (IndexNames::HASHED == type) { + /* This index is only considered "HELPFUL" for a query + * if it's the union of at least one equality constraint on the + * hashed field. Otherwise it's considered USELESS. + * Example queries (supposing the indexKey is {a : "hashed"}): + * {a : 3} HELPFUL + * {a : 3 , b : 3} HELPFUL + * {a : {$in : [3,4]}} HELPFUL + * {a : {$gte : 3, $lte : 3}} HELPFUL + * {} USELESS + * {b : 3} USELESS + * {a : {$gt : 3}} USELESS + */ + BSONElement firstElt = keyPattern.firstElement(); + if (queryConstraints.isPointIntervalSet(firstElt.fieldName())) { + return HELPFUL; + } else { + return USELESS; + } + } else if (IndexNames::GEO_2DSPHERE == type) { + BSONObjIterator i(keyPattern); + while (i.more()) { + BSONElement ie = i.next(); + + if (ie.type() != String || IndexNames::GEO_2DSPHERE != ie.valuestr()) { + continue; + } + + BSONElement e = query.getFieldDotted(ie.fieldName()); + // Some locations are given to us as arrays. Sigh. + if (Array == e.type()) { return HELPFUL; } + if (Object != e.type()) { continue; } + // getGtLtOp is horribly misnamed and really means get the operation. + switch (e.embeddedObject().firstElement().getGtLtOp()) { + case BSONObj::opNEAR: + return OPTIMAL; + case BSONObj::opWITHIN: { + BSONElement elt = e.embeddedObject().firstElement(); + if (Object != elt.type()) { continue; } + const char* fname = elt.embeddedObject().firstElement().fieldName(); + if (mongoutils::str::equals("$geometry", fname) + || mongoutils::str::equals("$centerSphere", fname)) { + return OPTIMAL; + } else { + return USELESS; + } + } + case BSONObj::opGEO_INTERSECTS: + return OPTIMAL; + default: + return USELESS; + } + } + return USELESS; + } else if (IndexNames::TEXT == type || IndexNames::TEXT_INTERNAL == type) { + return USELESS; + } else if (IndexNames::GEO_HAYSTACK == type) { + return USELESS; } else { cout << "Can't find index for keypattern " << keyPattern << endl; verify(0); diff --git a/src/mongo/db/indexkey.cpp b/src/mongo/db/indexkey.cpp index b0203713ab4..56de9511ec7 100644 --- a/src/mongo/db/indexkey.cpp +++ b/src/mongo/db/indexkey.cpp @@ -25,6 +25,7 @@ #include "../util/text.h" #include "mongo/db/client.h" #include "mongo/db/database.h" +#include "mongo/db/index/catalog_hack.h" #include "mongo/db/pdfile.h" #include "mongo/db/queryutil.h" @@ -35,53 +36,7 @@ namespace mongo { */ const int DefaultIndexVersionNumber = 1; - map * IndexPlugin::_plugins; - - IndexType::IndexType( const IndexPlugin * plugin , const IndexSpec * spec ) - : _plugin( plugin ) , _spec( spec ) { - - } - - IndexType::~IndexType() { - } - - BSONElement IndexType::missingField() const { - return _spec->_nullElt; - } - - const BSONObj& IndexType::keyPattern() const { - return _spec->keyPattern; - } - - IndexPlugin::IndexPlugin( const string& name ) - : _name( name ) { - if ( ! _plugins ) - _plugins = new map(); - (*_plugins)[name] = this; - } - - string IndexPlugin::findPluginName( const BSONObj& keyPattern ) { - string pluginName = ""; - - BSONObjIterator i( keyPattern ); - - while( i.more() ) { - BSONElement e = i.next(); - if ( e.type() != String ) - continue; - - uassert( 13007 , "can only have 1 index plugin / bad index key pattern" , pluginName.size() == 0 || pluginName == e.String() ); - pluginName = e.String(); - } - - return pluginName; - } - - int IndexType::compare( const BSONObj& l , const BSONObj& r ) const { - return l.woCompare( r , _spec->keyPattern ); - } - - void IndexSpec::_init(PluginRules rules) { + void IndexSpec::_init() { verify( keyPattern.objsize() ); // some basics @@ -120,53 +75,14 @@ namespace mongo { _undefinedObj = b.obj(); _undefinedElt = _undefinedObj.firstElement(); } - - { - // handle plugins - string pluginName = IndexPlugin::findPluginName( keyPattern ); - if ( pluginName.size() ) { - IndexPlugin * plugin = IndexPlugin::get( pluginName ); - - switch (rules) { - case NoPlugins: - uasserted(16735, - str::stream() - << "Attempting to use index type '" << pluginName << "' " - << "where index types are not allowed (1 or -1 only)."); - break; - - case RulesFor22: { - if ( ! plugin ) { - log() << "warning: can't find plugin [" << pluginName << "]" << endl; - } - - if (!IndexPlugin::existedBefore24(pluginName)) { - warning() << "Treating index " << info << " as ascending since " - << "it was created before 2.4 and '" << pluginName << "' " - << "was not a valid type at that time." - << endl; - - plugin = NULL; - } - break; - } - case RulesFor24: - // This assert will be triggered when downgrading from a future version that - // supports an index plugin unsupported by this version. - uassert(16736, str::stream() << "Invalid index type '" << pluginName << "' " - << "in index " << info - , plugin); - break; - } - - if (plugin) - _indexType.reset( plugin->generate( this ) ); - } - } _finishedInit = true; } + string IndexSpec::getTypeName() const { + return CatalogHack::findPluginName(_details->keyPattern()); + } + string IndexSpec::toString() const { stringstream s; s << "IndexSpec @ " << hex << this << dec << ", " @@ -177,298 +93,7 @@ namespace mongo { << "Info: " << info; return s.str(); } - - void assertParallelArrays( const char *first, const char *second ) { - stringstream ss; - ss << "cannot index parallel arrays [" << first << "] [" << second << "]"; - uasserted( ParallelArraysCode , ss.str() ); - } - - class KeyGeneratorV0 { - public: - KeyGeneratorV0( const IndexSpec &spec ) : _spec( spec ) {} - - void getKeys( const BSONObj &obj, BSONObjSet &keys ) const { - if ( _spec._indexType.get() ) { //plugin (eg geo) - _spec._indexType->getKeys( obj , keys ); - return; - } - vector fieldNames( _spec._fieldNames ); - vector fixed( _spec._fixed ); - _getKeys( fieldNames , fixed , obj, keys ); - if ( keys.empty() && ! _spec._sparse ) - keys.insert( _spec._nullKey ); - } - - private: - void _getKeys( vector fieldNames , vector fixed , const BSONObj &obj, BSONObjSet &keys ) const { - BSONElement arrElt; - unsigned arrIdx = ~0; - int numNotFound = 0; - - for( unsigned i = 0; i < fieldNames.size(); ++i ) { - if ( *fieldNames[ i ] == '\0' ) - continue; - - BSONElement e = obj.getFieldDottedOrArray( fieldNames[ i ] ); - - if ( e.eoo() ) { - e = _spec._nullElt; // no matching field - numNotFound++; - } - - if ( e.type() != Array ) - fieldNames[ i ] = ""; // no matching field or non-array match - - if ( *fieldNames[ i ] == '\0' ) - fixed[ i ] = e; // no need for further object expansion (though array expansion still possible) - - if ( e.type() == Array && arrElt.eoo() ) { // we only expand arrays on a single path -- track the path here - arrIdx = i; - arrElt = e; - } - - // enforce single array path here - if ( e.type() == Array && e.rawdata() != arrElt.rawdata() ) { - assertParallelArrays( e.fieldName(), arrElt.fieldName() ); - } - } - - bool allFound = true; // have we found elements for all field names in the key spec? - for( vector::const_iterator i = fieldNames.begin(); i != fieldNames.end(); ++i ) { - if ( **i != '\0' ) { - allFound = false; - break; - } - } - - if ( _spec._sparse && numNotFound == _spec._nFields ) { - // we didn't find any fields - // so we're not going to index this document - return; - } - - bool insertArrayNull = false; - - if ( allFound ) { - if ( arrElt.eoo() ) { - // no terminal array element to expand - BSONObjBuilder b(_spec._sizeTracker); - for( vector< BSONElement >::iterator i = fixed.begin(); i != fixed.end(); ++i ) - b.appendAs( *i, "" ); - keys.insert( b.obj() ); - } - else { - // terminal array element to expand, so generate all keys - BSONObjIterator i( arrElt.embeddedObject() ); - if ( i.more() ) { - while( i.more() ) { - BSONObjBuilder b(_spec._sizeTracker); - for( unsigned j = 0; j < fixed.size(); ++j ) { - if ( j == arrIdx ) - b.appendAs( i.next(), "" ); - else - b.appendAs( fixed[ j ], "" ); - } - keys.insert( b.obj() ); - } - } - else if ( fixed.size() > 1 ) { - insertArrayNull = true; - } - } - } - else { - // nonterminal array element to expand, so recurse - verify( !arrElt.eoo() ); - BSONObjIterator i( arrElt.embeddedObject() ); - if ( i.more() ) { - while( i.more() ) { - BSONElement e = i.next(); - if ( e.type() == Object ) { - _getKeys( fieldNames, fixed, e.embeddedObject(), keys ); - } - } - } - else { - insertArrayNull = true; - } - } - - if ( insertArrayNull ) { - // x : [] - need to insert undefined - BSONObjBuilder b(_spec._sizeTracker); - for( unsigned j = 0; j < fixed.size(); ++j ) { - if ( j == arrIdx ) { - b.appendUndefined( "" ); - } - else { - BSONElement e = fixed[j]; - if ( e.eoo() ) - b.appendNull( "" ); - else - b.appendAs( e , "" ); - } - } - keys.insert( b.obj() ); - } - } - - const IndexSpec &_spec; - }; - - class KeyGeneratorV1 { - public: - KeyGeneratorV1( const IndexSpec &spec ) : _spec( spec ) {} - - void getKeys( const BSONObj &obj, BSONObjSet &keys ) const { - if ( _spec._indexType.get() ) { //plugin (eg geo) - _spec._indexType->getKeys( obj , keys ); - return; - } - vector fieldNames( _spec._fieldNames ); - vector fixed( _spec._fixed ); - _getKeys( fieldNames , fixed , obj, keys ); - if ( keys.empty() && ! _spec._sparse ) - keys.insert( _spec._nullKey ); - } - - private: - /** - * @param arrayNestedArray - set if the returned element is an array nested directly within arr. - */ - BSONElement extractNextElement( const BSONObj &obj, const BSONObj &arr, const char *&field, bool &arrayNestedArray ) const { - string firstField = mongoutils::str::before( field, '.' ); - bool haveObjField = !obj.getField( firstField ).eoo(); - BSONElement arrField = arr.getField( firstField ); - bool haveArrField = !arrField.eoo(); - - // An index component field name cannot exist in both a document array and one of that array's children. - uassert( 15855, - mongoutils::str::stream() << - "Ambiguous field name found in array (do not use numeric field names in " - "embedded elements in an array), field: '" << arrField.fieldName() << - "' for array: " << arr, - !haveObjField || !haveArrField ); - - arrayNestedArray = false; - if ( haveObjField ) { - return obj.getFieldDottedOrArray( field ); - } - else if ( haveArrField ) { - if ( arrField.type() == Array ) { - arrayNestedArray = true; - } - return arr.getFieldDottedOrArray( field ); - } - return BSONElement(); - } - - void _getKeysArrEltFixed( vector &fieldNames , vector &fixed , const BSONElement &arrEntry, BSONObjSet &keys, int numNotFound, const BSONElement &arrObjElt, const set< unsigned > &arrIdxs, bool mayExpandArrayUnembedded ) const { - // set up any terminal array values - for( set::const_iterator j = arrIdxs.begin(); j != arrIdxs.end(); ++j ) { - if ( *fieldNames[ *j ] == '\0' ) { - fixed[ *j ] = mayExpandArrayUnembedded ? arrEntry : arrObjElt; - } - } - // recurse - _getKeys( fieldNames, fixed, ( arrEntry.type() == Object ) ? arrEntry.embeddedObject() : BSONObj(), keys, numNotFound, arrObjElt.embeddedObject() ); - } - - /** - * @param fieldNames - fields to index, may be postfixes in recursive calls - * @param fixed - values that have already been identified for their index fields - * @param obj - object from which keys should be extracted, based on names in fieldNames - * @param keys - set where index keys are written - * @param numNotFound - number of index fields that have already been identified as missing - * @param array - array from which keys should be extracted, based on names in fieldNames - * If obj and array are both nonempty, obj will be one of the elements of array. - */ - void _getKeys( vector fieldNames , vector fixed , const BSONObj &obj, BSONObjSet &keys, int numNotFound = 0, const BSONObj &array = BSONObj() ) const { - BSONElement arrElt; - set arrIdxs; - bool mayExpandArrayUnembedded = true; - for( unsigned i = 0; i < fieldNames.size(); ++i ) { - if ( *fieldNames[ i ] == '\0' ) { - continue; - } - - bool arrayNestedArray; - // Extract element matching fieldName[ i ] from object xor array. - BSONElement e = extractNextElement( obj, array, fieldNames[ i ], arrayNestedArray ); - - if ( e.eoo() ) { - // if field not present, set to null - fixed[ i ] = _spec._nullElt; - // done expanding this field name - fieldNames[ i ] = ""; - numNotFound++; - } - else if ( e.type() == Array ) { - arrIdxs.insert( i ); - if ( arrElt.eoo() ) { - // we only expand arrays on a single path -- track the path here - arrElt = e; - } - else if ( e.rawdata() != arrElt.rawdata() ) { - // enforce single array path here - assertParallelArrays( e.fieldName(), arrElt.fieldName() ); - } - if ( arrayNestedArray ) { - mayExpandArrayUnembedded = false; - } - } - else { - // not an array - no need for further expansion - fixed[ i ] = e; - } - } - - if ( arrElt.eoo() ) { - // No array, so generate a single key. - if ( _spec._sparse && numNotFound == _spec._nFields ) { - return; - } - BSONObjBuilder b(_spec._sizeTracker); - for( vector< BSONElement >::iterator i = fixed.begin(); i != fixed.end(); ++i ) { - b.appendAs( *i, "" ); - } - keys.insert( b.obj() ); - } - else if ( arrElt.embeddedObject().firstElement().eoo() ) { - // Empty array, so set matching fields to undefined. - _getKeysArrEltFixed( fieldNames, fixed, _spec._undefinedElt, keys, numNotFound, arrElt, arrIdxs, true ); - } - else { - // Non empty array that can be expanded, so generate a key for each member. - BSONObj arrObj = arrElt.embeddedObject(); - BSONObjIterator i( arrObj ); - while( i.more() ) { - _getKeysArrEltFixed( fieldNames, fixed, i.next(), keys, numNotFound, arrElt, arrIdxs, mayExpandArrayUnembedded ); - } - } - } - - const IndexSpec &_spec; - }; - void IndexSpec::getKeys( const BSONObj &obj, BSONObjSet &keys ) const { - switch( indexVersion() ) { - case 0: { - KeyGeneratorV0 g( *this ); - g.getKeys( obj, keys ); - break; - } - case 1: { - KeyGeneratorV1 g( *this ); - g.getKeys( obj, keys ); - break; - } - default: - massert( 15869, "Invalid index version for key generation.", false ); - } - } - int IndexSpec::indexVersion() const { if ( !info.hasField( "v" ) ) { return DefaultIndexVersionNumber; @@ -476,8 +101,4 @@ namespace mongo { return IndexDetails::versionForIndexObj( info ); } - bool IndexType::scanAndOrderRequired( const BSONObj& query , const BSONObj& order ) const { - return ! order.isEmpty(); - } - } diff --git a/src/mongo/db/indexkey.h b/src/mongo/db/indexkey.h index 2acd38e998d..7fed236db7e 100644 --- a/src/mongo/db/indexkey.h +++ b/src/mongo/db/indexkey.h @@ -22,6 +22,7 @@ #include "mongo/db/diskloc.h" #include "mongo/db/index_names.h" +#include "mongo/db/keypattern.h" #include "mongo/db/jsobj.h" namespace mongo { @@ -32,126 +33,16 @@ namespace mongo { class Cursor; class IndexSpec; - class IndexType; // TODO: this name sucks - class IndexPlugin; class IndexDetails; class FieldRangeSet; enum IndexSuitability { USELESS = 0 , HELPFUL = 1 , OPTIMAL = 2 }; - /** - * this represents an instance of a index plugin - * done this way so parsing, etc... can be cached - * so if there is a FTS IndexPlugin, for each index using FTS - * there will be 1 of these, and it can have things pre-parsed, etc... - */ - class IndexType : boost::noncopyable { - public: - IndexType( const IndexPlugin * plugin , const IndexSpec * spec ); - virtual ~IndexType(); - - virtual void getKeys( const BSONObj &obj, BSONObjSet &keys ) const = 0; - - /** - * Returns the element placed in an index key when indexing a field absent from a document. - * By default this is a null BSONElement. - */ - virtual BSONElement missingField() const; - - /** optional op : changes query to match what's in the index */ - virtual BSONObj fixKey( const BSONObj& in ) { return in; } - - /** optional op : compare 2 objects with regards to this index */ - virtual int compare( const BSONObj& l , const BSONObj& r ) const; - - /** @return plugin */ - const IndexPlugin * getPlugin() const { return _plugin; } - - const BSONObj& keyPattern() const; - - virtual bool scanAndOrderRequired( const BSONObj& query , const BSONObj& order ) const ; - - protected: - const IndexPlugin * _plugin; - const IndexSpec * _spec; - }; - - /** - * this represents a plugin - * a plugin could be something like full text search, sparse index, etc... - * 1 of these exists per type of index per server - * 1 IndexType is created per index using this plugin - */ - class IndexPlugin : boost::noncopyable { - public: - IndexPlugin( const string& name ); - virtual ~IndexPlugin() {} - - virtual IndexType* generate( const IndexSpec * spec ) const = 0; - - string getName() const { return _name; } - - /** - * @return new keyPattern - * if nothing changes, should return keyPattern - */ - virtual BSONObj adjustIndexSpec( const BSONObj& spec ) const { return spec; } - - /** - * Hook function to run after an index that uses this plugin is built. - * - * This will be called with an active write context (and lock) on the database. - * - * @param spec The IndexSpec of the newly built index. - */ - virtual void postBuildHook( const IndexSpec& spec ) const { } - - // ------- static below ------- - - static IndexPlugin* get( const string& name ) { - if ( ! _plugins ) - return 0; - map::iterator i = _plugins->find( name ); - if ( i == _plugins->end() ) - return 0; - return i->second; - } - - /** - * @param keyPattern { x : "fts" } - * @return "" or the name - */ - static string findPluginName( const BSONObj& keyPattern ); - - /** - * True if is a regular (non-plugin) index or uses a plugin that existed before 2.4. - * These plugins are grandfathered in and allowed to exist in DBs with - * PDFILE_MINOR_VERSION_22_AND_OLDER - */ - static bool existedBefore24(const string& name) { - return name.empty() - || name == IndexNames::GEO_2D - || name == IndexNames::GEO_HAYSTACK - || name == IndexNames::HASHED - ; - } - - private: - string _name; - static map * _plugins; - }; - /* precomputed details about an index, used for inserting keys on updates stored/cached in NamespaceDetailsTransient, or can be used standalone */ class IndexSpec { public: - enum PluginRules { - NoPlugins, - RulesFor22, // if !IndexPlugin::existedBefore24() treat as ascending - RulesFor24, // allow new plugins but error if unknown - }; - BSONObj keyPattern; // e.g., { name : 1 } BSONObj info; // this is the same as IndexDetails::info.obj() @@ -159,47 +50,26 @@ namespace mongo { : _details(0) , _finishedInit(false) { } - explicit IndexSpec(const BSONObj& k, const BSONObj& m=BSONObj(), - PluginRules rules=RulesFor24) + explicit IndexSpec(const BSONObj& k, const BSONObj& m=BSONObj()) : keyPattern(k) , info(m) , _details(0) , _finishedInit(false) { - _init(rules); + _init(); } /** this is a DiscLoc of an IndexDetails info should have a key field */ - explicit IndexSpec(const DiskLoc& loc, PluginRules rules=RulesFor24) { - reset(loc, rules); + explicit IndexSpec(const DiskLoc& loc) { + reset(loc); } - void reset(const BSONObj& info, PluginRules rules=RulesFor24); + void reset(const BSONObj& info); void reset(const IndexDetails * details); // determines rules based on pdfile version - void reset(const DiskLoc& infoLoc, PluginRules rules=RulesFor24) { - reset(infoLoc.obj(), rules); + void reset(const DiskLoc& infoLoc) { + reset(infoLoc.obj()); } - void getKeys( const BSONObj &obj, BSONObjSet &keys ) const; - - /** - * Returns the element placed in an index key when indexing a field absent from a document. - * By default this is a null BSONElement. - */ - BSONElement missingField() const { - if ( _indexType.get() ) - return _indexType->missingField(); - return _nullElt; - } - - string getTypeName() const { - if ( _indexType.get() ) - return _indexType->getPlugin()->getName(); - return ""; - } - - IndexType* getType() const { - return _indexType.get(); - } + string getTypeName() const; const IndexDetails * getDetails() const { return _details; @@ -226,12 +96,10 @@ namespace mongo { int _nFields; // number of fields in the index bool _sparse; // if the index is sparse - shared_ptr _indexType; const IndexDetails * _details; - void _init(PluginRules rules); + void _init(); - friend class IndexType; friend class KeyGeneratorV0; friend class KeyGeneratorV1; public: diff --git a/src/mongo/db/pdfile.cpp b/src/mongo/db/pdfile.cpp index bda45e2047c..d592991a014 100644 --- a/src/mongo/db/pdfile.cpp +++ b/src/mongo/db/pdfile.cpp @@ -45,6 +45,7 @@ _ disallow system* manipulations from the database. #include "mongo/db/db.h" #include "mongo/db/dbhelpers.h" #include "mongo/db/extsort.h" +#include "mongo/db/index_names.h" #include "mongo/db/index_update.h" #include "mongo/db/index/catalog_hack.h" #include "mongo/db/index/index_descriptor.h" @@ -1530,12 +1531,13 @@ namespace mongo { tableToIndex->addIndex(tabletoidxns.c_str()); getDur().writingInt(tableToIndex->indexBuildsInProgress) -= 1; - IndexType* indexType = idx.getSpec().getType(); - const IndexPlugin *plugin = indexType ? indexType->getPlugin() : NULL; - if (plugin) { - plugin->postBuildHook( idx.getSpec() ); + // If it's an FTS index, we want to set the power of 2 flag. + string pluginName = KeyPattern::findPluginName(idx.keyPattern()); + if (IndexNames::TEXT == pluginName || IndexNames::TEXT_INTERNAL == pluginName) { + if (tableToIndex->setUserFlag(NamespaceDetails::Flag_UsePowerOf2Sizes)) { + tableToIndex->syncUserFlags(idx.parentNS()); + } } - } catch (...) { // Generally, this will be called as an exception from building the index bubbles up. diff --git a/src/mongo/db/query_optimizer_internal.cpp b/src/mongo/db/query_optimizer_internal.cpp index 0bed4fbdaf8..dcd7516b357 100644 --- a/src/mongo/db/query_optimizer_internal.cpp +++ b/src/mongo/db/query_optimizer_internal.cpp @@ -20,6 +20,7 @@ #include "mongo/client/dbclientinterface.h" #include "mongo/db/db.h" +#include "mongo/db/index/catalog_hack.h" #include "mongo/db/index_selection.h" #include "mongo/db/pagefault.h" #include "mongo/db/parsed_query.h" @@ -543,12 +544,13 @@ namespace mongo { while( i.more() ) { int j = i.pos(); IndexDetails& ii = i.next(); - const IndexSpec& spec = ii.getSpec(); - if (special.has(spec.getTypeName()) && - (USELESS != IndexSelection::isSuitableFor(spec.keyPattern, + BSONObj keyPattern = ii.keyPattern(); + string pluginName = CatalogHack::findPluginName(keyPattern); + if (special.has(pluginName) && + (USELESS != IndexSelection::isSuitableFor(keyPattern, _qps.frsp().frsForIndex(d, j), _qps.order()))) { uassert( 16330, "'special' query operator not allowed", _allowSpecial ); - _qps.setSinglePlan( newPlan( d, j, BSONObj(), BSONObj(), spec.getTypeName())); + _qps.setSinglePlan( newPlan( d, j, BSONObj(), BSONObj(), pluginName)); return true; } } @@ -1495,7 +1497,8 @@ namespace mongo { while( i.more() ) { IndexDetails& ii = i.next(); if ( indexWorks( ii.keyPattern(), min.isEmpty() ? max : min, ret.first, ret.second ) ) { - if ( ii.getSpec().getType() == 0 ) { + if ( ii.getSpec().getTypeName().empty()) { + //if ( ii.getSpec().getType() == 0 ) { id = ⅈ keyPattern = ii.keyPattern(); break; diff --git a/src/mongo/db/query_plan.cpp b/src/mongo/db/query_plan.cpp index c3ee9d63cb5..6bb157369a6 100644 --- a/src/mongo/db/query_plan.cpp +++ b/src/mongo/db/query_plan.cpp @@ -88,7 +88,6 @@ namespace mongo { _endKeyInclusive(), _utility( Helpful ), _special( special ), - _type( 0 ), _startOrEndSpec() { } @@ -112,24 +111,24 @@ namespace mongo { return; } + _descriptor.reset(CatalogHack::getDescriptor(_d, _idxNo)); _index = &_d->idx(_idxNo); // If the parsing or index indicates this is a special query, don't continue the processing if (!_special.empty() || - ( _index->getSpec().getType() && (USELESS != - IndexSelection::isSuitableFor(_index->getSpec().keyPattern, _frs, _order)))) { + ( ("" != CatalogHack::findPluginName(_descriptor->keyPattern())) && (USELESS != + IndexSelection::isSuitableFor(_descriptor->keyPattern(), _frs, _order)))) { - _type = _index->getSpec().getType(); - if (_special.empty()) _special = _index->getSpec().getType()->getPlugin()->getName(); + _specialIndexName = CatalogHack::findPluginName(_descriptor->keyPattern()); + if (_special.empty()) _special = _specialIndexName; - massert( 13040 , (string)"no type for special: " + _special , _type ); + massert( 13040 , (string)"no type for special: " + _special , "" != _specialIndexName); // hopefully safe to use original query in these contexts; // don't think we can mix special with $or clause separation yet - _scanAndOrderRequired = _type->scanAndOrderRequired( _originalQuery , _order ); + _scanAndOrderRequired = !_order.isEmpty(); return; } - const IndexSpec &idxSpec = _index->getSpec(); BSONObjIterator o( _order ); BSONObjIterator k( idxKey ); if ( !o.moreWithEOO() ) @@ -192,7 +191,7 @@ doneCheckOrder: if ( !_scanAndOrderRequired && ( optimalIndexedQueryCount == _frs.numNonUniversalRanges() ) ) _utility = Optimal; - _frv.reset( new FieldRangeVector( _frs, idxSpec, _direction ) ); + _frv.reset( new FieldRangeVector( _frs, _descriptor->keyPattern(), _direction ) ); if ( // If all field range constraints are on indexed fields and ... _utility == Optimal && @@ -208,7 +207,7 @@ doneCheckOrder: if ( originalFrsp ) { _originalFrv.reset( new FieldRangeVector( originalFrsp->frsForIndex( _d, _idxNo ), - idxSpec, + _descriptor->keyPattern(), _direction ) ); } else { @@ -231,7 +230,7 @@ doneCheckOrder: _utility = Unhelpful; } - if ( idxSpec.isSparse() && hasPossibleExistsFalsePredicate() ) { + if ( _descriptor->isSparse() && hasPossibleExistsFalsePredicate() ) { _utility = Disallowed; } @@ -244,7 +243,7 @@ doneCheckOrder: shared_ptr QueryPlan::newCursor( const DiskLoc& startLoc, bool requestIntervalCursor ) const { - if ( _type ) { + if ("" != _specialIndexName) { // hopefully safe to use original query in these contexts - don't think we can mix type // with $or clause separation yet int numWanted = 0; @@ -253,6 +252,7 @@ doneCheckOrder: numWanted = _parsedQuery->getSkip() + _parsedQuery->getNumToReturn(); } + // Why do we get new objects here? Because EmulatedCursor takes ownership of them. IndexDescriptor* descriptor = CatalogHack::getDescriptor(_d, _idxNo); IndexAccessMethod* iam = CatalogHack::getIndex(descriptor); return shared_ptr(EmulatedCursor::make(descriptor, iam, _originalQuery, @@ -284,7 +284,7 @@ doneCheckOrder: _direction >= 0 ? 1 : -1 ) ); } - if ( _index->getSpec().getType() ) { + if ( !_index->getSpec().getTypeName().empty()) { //_index->getSpec().getType() ) { return shared_ptr( BtreeCursor::make( _d, *_index, _frv->startKey(), diff --git a/src/mongo/db/query_plan.h b/src/mongo/db/query_plan.h index 62cd4c7c567..3c6350cc180 100644 --- a/src/mongo/db/query_plan.h +++ b/src/mongo/db/query_plan.h @@ -17,6 +17,7 @@ #pragma once #include "mongo/db/diskloc.h" +#include "mongo/db/index/index_descriptor.h" #include "mongo/db/indexkey.h" #include "mongo/db/jsobj.h" #include "mongo/db/matcher.h" @@ -29,7 +30,6 @@ namespace mongo { class FieldRangeSet; class FieldRangeSetPair; class IndexDetails; - class IndexType; class NamespaceDetails; class ParsedQuery; struct QueryPlanSummary; @@ -176,10 +176,11 @@ namespace mongo { bool _endKeyInclusive; Utility _utility; string _special; - IndexType* _type; bool _startOrEndSpec; shared_ptr _keyFieldsOnly; mutable shared_ptr _matcher; // Lazy initialization. + auto_ptr _descriptor; + string _specialIndexName; }; std::ostream &operator<< ( std::ostream& out, const QueryPlan::Utility& utility ); diff --git a/src/mongo/db/query_runner.cpp b/src/mongo/db/query_runner.cpp index d94ff3efa23..c649bca957e 100644 --- a/src/mongo/db/query_runner.cpp +++ b/src/mongo/db/query_runner.cpp @@ -15,6 +15,7 @@ */ #include "mongo/db/query_runner.h" + #include "mongo/db/btree.h" #include "mongo/db/index.h" #include "mongo/db/jsobj.h" diff --git a/src/mongo/db/queryutil.cpp b/src/mongo/db/queryutil.cpp index dffb52c34bd..6ac3433caf4 100644 --- a/src/mongo/db/queryutil.cpp +++ b/src/mongo/db/queryutil.cpp @@ -1160,15 +1160,26 @@ namespace mongo { return true; } - FieldRangeVector::FieldRangeVector( const FieldRangeSet &frs, const IndexSpec &indexSpec, + FieldRangeVector::FieldRangeVector( const FieldRangeSet &frs, BSONObj keyPattern, int direction ) : - _indexSpec( indexSpec ), + _keyPattern(keyPattern), _direction( direction >= 0 ? 1 : -1 ), _hasAllIndexedRanges( true ) { - verify( frs.matchPossibleForIndex( _indexSpec.keyPattern ) ); + verify( frs.matchPossibleForIndex( keyPattern)); _queries = frs._queries; - BSONObjIterator i( _indexSpec.keyPattern ); + + // For key generation + BSONObjIterator it(_keyPattern); + while (it.more()) { + BSONElement elt = it.next(); + _fieldNames.push_back(elt.fieldName()); + _fixed.push_back(BSONElement()); + } + + _keyGenerator.reset(new BtreeKeyGeneratorV1(_fieldNames, _fixed, false)); + map topFieldElemMatchContexts; + BSONObjIterator i(keyPattern); while( i.more() ) { BSONElement e = i.next(); const FieldRange *range = &frs.range( e.fieldName() ); @@ -1262,7 +1273,7 @@ namespace mongo { BSONObj FieldRangeVector::startKey() const { BSONObjBuilder b; - BSONObjIterator keys( _indexSpec.keyPattern ); + BSONObjIterator keys(_keyPattern); vector::const_iterator i = _ranges.begin(); for( ; i != _ranges.end(); ++i, ++keys ) { // Append lower bounds until an exclusive bound is found. @@ -1300,7 +1311,7 @@ namespace mongo { BSONObj FieldRangeVector::endKey() const { BSONObjBuilder b; - BSONObjIterator keys( _indexSpec.keyPattern ); + BSONObjIterator keys(_keyPattern); vector::const_iterator i = _ranges.begin(); for( ; i != _ranges.end(); ++i, ++keys ) { // Append upper bounds until an exclusive bound is found. @@ -1338,7 +1349,7 @@ namespace mongo { BSONObj FieldRangeVector::obj() const { BSONObjBuilder b; - BSONObjIterator k( _indexSpec.keyPattern ); + BSONObjIterator k(_keyPattern); for( int i = 0; i < (int)_ranges.size(); ++i ) { BSONArrayBuilder a( b.subarrayStart( k.next().fieldName() ) ); for( vector::const_iterator j = _ranges[ i ].intervals().begin(); @@ -1514,7 +1525,7 @@ namespace mongo { bool FieldRangeVector::matchesKey( const BSONObj &key ) const { BSONObjIterator j( key ); - BSONObjIterator k( _indexSpec.keyPattern ); + BSONObjIterator k(_keyPattern); for( int l = 0; l < (int)_ranges.size(); ++l ) { int number = (int) k.next().number(); bool forward = ( number >= 0 ? 1 : -1 ) * ( _direction >= 0 ? 1 : -1 ) > 0; @@ -1526,16 +1537,16 @@ namespace mongo { } bool FieldRangeVector::matches( const BSONObj &obj ) const { - bool ok = false; + BSONObjSet keys; + _keyGenerator->getKeys(obj, &keys); + // TODO The representation of matching keys could potentially be optimized // more for the case at hand. (For example, we can potentially consider // fields individually instead of constructing several bson objects using // multikey arrays.) But getKeys() canonically defines the key set for a // given object and for now we are using it as is. - BSONObjSet keys; - _indexSpec.getKeys( obj, keys ); for( BSONObjSet::const_iterator i = keys.begin(); i != keys.end(); ++i ) { if ( matchesKey( *i ) ) { ok = true; @@ -1551,8 +1562,9 @@ namespace mongo { BSONObj FieldRangeVector::firstMatch( const BSONObj &obj ) const { // NOTE Only works in forward direction. verify( _direction >= 0 ); - BSONObjSet keys( BSONObjCmp( _indexSpec.keyPattern ) ); - _indexSpec.getKeys( obj, keys ); + BSONObjCmp oc(_keyPattern); + BSONObjSet keys(oc); + _keyGenerator->getKeys(obj, &keys); for( BSONObjSet::const_iterator i = keys.begin(); i != keys.end(); ++i ) { if ( matchesKey( *i ) ) { return *i; @@ -1563,7 +1575,7 @@ namespace mongo { string FieldRangeVector::toString() const { BSONObjBuilder bob; - BSONObjIterator i( _indexSpec.keyPattern ); + BSONObjIterator i(_keyPattern); for( vector::const_iterator r = _ranges.begin(); r != _ranges.end() && i.more(); ++r ) { BSONElement e = i.next(); @@ -1585,7 +1597,7 @@ namespace mongo { // TODO optimize more SERVER-5450. int FieldRangeVectorIterator::advance( const BSONObj &curr ) { BSONObjIterator j( curr ); - BSONObjIterator o( _v._indexSpec.keyPattern ); + BSONObjIterator o( _v._keyPattern); // track first field for which we are not at the end of the valid values, // since we may need to advance from the key prefix ending with this field int latestNonEndpoint = -1; diff --git a/src/mongo/db/queryutil.h b/src/mongo/db/queryutil.h index 381c1a14d1a..c53082c06af 100644 --- a/src/mongo/db/queryutil.h +++ b/src/mongo/db/queryutil.h @@ -17,9 +17,10 @@ #include "jsobj.h" #include "indexkey.h" +#include "mongo/db/index/btree_key_generator.h" namespace mongo { - + //maximum number of intervals produced by $in queries. static const unsigned MAX_IN_COMBINATIONS = 4000000; @@ -433,7 +434,7 @@ namespace mongo { * @param indexSpec The index spec (key pattern and info) * @param direction The direction of index traversal */ - FieldRangeVector( const FieldRangeSet &frs, const IndexSpec &indexSpec, int direction ); + FieldRangeVector( const FieldRangeSet &frs, BSONObj keyPattern, int direction ); /** * Methods for identifying compound start and end btree bounds describing this field range @@ -499,8 +500,6 @@ namespace mongo { /** @return a client readable representation of 'this' */ BSONObj obj() const; - const IndexSpec& getSpec(){ return _indexSpec; } - /** * @return true iff the provided document matches valid ranges on all * of this FieldRangeVector's fields, which is the case iff this document @@ -541,11 +540,15 @@ namespace mongo { int matchingLowElement( const BSONElement &e, int i, bool direction, bool &lowEquality ) const; bool matchesElement( const BSONElement &e, int i, bool direction ) const; vector _ranges; - const IndexSpec _indexSpec; + BSONObj _keyPattern; int _direction; vector _queries; // make sure mem owned bool _hasAllIndexedRanges; friend class FieldRangeVectorIterator; + + vector _fieldNames; + vector _fixed; + scoped_ptr _keyGenerator; }; /** diff --git a/src/mongo/db/scanandorder.cpp b/src/mongo/db/scanandorder.cpp index 03cf5fd0346..d3ed2beffa2 100644 --- a/src/mongo/db/scanandorder.cpp +++ b/src/mongo/db/scanandorder.cpp @@ -103,7 +103,7 @@ namespace mongo { void ScanAndOrder::_addIfBetter(const BSONObj& k, const BSONObj& o, const BestMap::iterator& i, const DiskLoc* loc) { const BSONObj& worstBestKey = i->first; - int cmp = worstBestKey.woCompare(k, _order._spec.keyPattern); + int cmp = worstBestKey.woCompare(k, _order._keyPattern); if ( cmp > 0 ) { // k is better, 'upgrade' _validateAndUpdateApproxSize( -i->first.objsize() + -i->second.objsize() ); diff --git a/src/mongo/db/scanandorder.h b/src/mongo/db/scanandorder.h index c3a9e834d5a..4bcfb4e44f9 100644 --- a/src/mongo/db/scanandorder.h +++ b/src/mongo/db/scanandorder.h @@ -32,12 +32,11 @@ namespace mongo { class KeyType : boost::noncopyable { public: - IndexSpec _spec; + BSONObj _keyPattern; FieldRangeVector _keyCutter; public: - KeyType(const BSONObj &pattern, const FieldRangeSet &frs): - _spec(pattern, BSONObj(), IndexSpec::NoPlugins), - _keyCutter(frs, _spec, 1) { + KeyType(const BSONObj &pattern, const FieldRangeSet &frs) + : _keyPattern(pattern), _keyCutter(frs, pattern, 1) { verify(!pattern.isEmpty()); } -- cgit v1.2.1