Mongodb源码分析--插入记录及索引B树构建

     在之前的一篇文章中，介绍了assembleResponse函数(位于instance.cpp第224行)，它会根据op操作枚举类型来调用相应的crud操作，枚举类型定义如下：

     enum Operations {
        opReply = 1,     /* reply. responseTo is set. */
        dbMsg = 1000,    /* generic msg command followed by a string */
        dbUpdate = 2001, /* update object */
        dbInsert = 2002,
        //dbGetByOID = 2003,
        dbQuery = 2004,
        dbGetMore = 2005,
        dbDelete = 2006,
        dbKillCursors = 2007
    };

    可以看到dbInsert = 2002 为插入操作枚举值，下面我们看一下assembleResponse在确定是插入操作时调用的方法，如下：

assembleResponse( Message &m, DbResponse &dbresponse, const SockAddr &client ) {
    .....
            try {
                if ( op == dbInsert ) {  //添加记录操作
                    receivedInsert(m, currentOp);
                }
                else if ( op == dbUpdate ) { //更新记录
                    receivedUpdate(m, currentOp);
                }
                else if ( op == dbDelete ) { //删除记录
                    receivedDelete(m, currentOp);
                }
                else if ( op == dbKillCursors ) { //删除Cursors（游标）对象
                    currentOp.ensureStarted();
                    logThreshold = 10;
                    ss << "killcursors ";
                    receivedKillCursors(m);
                }
                else {
                    mongo::log() << "    operation isn't supported: " << op << endl;
                    currentOp.done();
                    log = true;
                }
            }
          .....
        }
    }

        从上面代码可以看出，系统在确定dbInsert操作时，调用了receivedInsert()方法（位于instance.cpp文件第570行），下面是该方法的定义：
   

void receivedInsert(Message& m, CurOp& op) {
        DbMessage d(m);//初始化数据库格式的消息
        const char *ns = d.getns();//获取名空间，用于接下来insert数据
        assert(*ns);
        uassert( 10058 ,  "not master", isMasterNs( ns ) );
        op.debug().str << ns;

        writelock lk(ns);//声明写锁
        
        if ( handlePossibleShardedMessage( m , 0 ) )//查看是不是sharding信息，如果是则处理
            return;

        Client::Context ctx(ns);
        int n = 0;
        while ( d.moreJSObjs() ) { //循环获取当前消息体中的BSONObj数据（数据库记录）
            BSONObj js = d.nextJsObj();
            uassert( 10059 , "object to insert too large", js.objsize() <= BSONObjMaxUserSize);
            {
                // 声明BSONObj迭代器，以查看里面元素是否有更新操作，如set inc push pull 等
                BSONObjIterator i( js );
                while ( i.more() ) {
                    BSONElement e = i.next();
                    uassert( 13511 , "object to insert can't have $ modifiers" , e.fieldName()[0] != '$' );
                }
            }
            //插入记录操作，god = false用于标识当前BSONObj对象为有效数据
            theDataFileMgr.insertWithObjMod(ns, js, false);
            logOp("i", ns, js);//日志操作，包括master状态下及sharding分片情况

            if( ++n % 4 == 0 ) {
                // 在插入一些数据后，进行持久化操作，有关持久化部分参见我的这篇文章
                // http://www.cnblogs.com/daizhj/archive/2011/03/21/1990344.html
                getDur().commitIfNeeded();
            }
        }
        globalOpCounters.incInsertInWriteLock(n);//在写锁环境下添加已插入记录数（n）,锁采用InterlockedIncrement实现数的原子性
    }

      上面的方法中，主要是在“写锁”环境下执行插入数据操作，并且在插入记录之前进行简单的数据对象检查，如长度和插入数据是否被修改，以确保数据的最终有效性。
      最终上面代码会调用 insertWithObjMod()方法（位于pdfile.cpp 文件第1432行），该方法定义如下：

   DiskLoc DataFileMgr::insertWithObjMod(const char *ns, BSONObj &o, bool god) {
        DiskLoc loc = insert( ns, o.objdata(), o.objsize(), god );
        if ( !loc.isNull() )//判断返回记录地址是否为空（记录是否插入成功）
            o = BSONObj( loc.rec() );//如有效，则用记录地地址上的记录(record类型指针)绑定到o上
        return loc;
   }

      该方法只是一个对插入操作及返回结果的封装，其中ns为数据对象的名空间，o就是要插入的数据对象（BSONObj），god用于标识当前BSONObj对象是否为有效数据（false=有效），这里之所以要传入god这个参数，是因为在接下来的insert方法里同时支持添加名空间(及索引)和插入记录操作（都会不断调用该方法），而在添加名空间时god=true。
    
     下面我们看一下insert方法（pdfile.cpp 第1467行），因为其内容较长，请详见注释：

DiskLoc DataFileMgr::insert(const char *ns, const void *obuf, int len, bool god, const BSONElement &writeId, bool mayAddIndex) {
        bool wouldAddIndex = false;
        massert( 10093 , "cannot insert into reserved $ collection", god || isANormalNSName( ns ) );
        uassert( 10094 , str::stream() << "invalid ns: " << ns , isValidNS( ns ) );
        const char *sys = strstr(ns, "system.");
        if ( sys ) {//对插入记录的ns进行判断，是否要插入保留的数据库名(system)，如是则停止执行其它代码
            uassert( 10095 , "attempt to insert in reserved database name 'system'", sys != ns);
            if ( strstr(ns, ".system.") ) {
                // later:check for dba-type permissions here if have that at some point separate
                if ( strstr(ns, ".system.indexes" ) )//判断是否创建索引
                    wouldAddIndex = true;
                else if ( legalClientSystemNS( ns , true ) )
                    ;
                else if ( !god ) {//表示obuf有数据，但这就意味着要向system下插入数据（把system当成数据表了）
                    out() << "ERROR: attempt to insert in system namespace " << ns << endl;
                    return DiskLoc();
                }
            }
            else
                sys = 0;
        }

        bool addIndex = wouldAddIndex && mayAddIndex;//判断是否需要添加索引

        NamespaceDetails *d = nsdetails(ns);//获取ns的详细信息
        if ( d == 0 ) {
            addNewNamespaceToCatalog(ns);//向system catalog添加新的名空间，它会再次调用当前insert()方法
            /* todo: shouldn't be in the namespace catalog until after the allocations here work.
               also if this is an addIndex, those checks should happen before this!
            */
            // 创建第一个数据库文件.
            cc().database()->allocExtent(ns, Extent::initialSize(len), false);
            d = nsdetails(ns);
            if ( !god )
                ensureIdIndexForNewNs(ns);
        }
        d->paddingFits();

        NamespaceDetails *tableToIndex = 0;

        string tabletoidxns;
        BSONObj fixedIndexObject;
        if ( addIndex ) {
            assert( obuf );
            BSONObj io((const char *) obuf);
            //做索引准备工作，这里并不真正创建索引，只是进行参数检查，以及索引是否已存在等
            if( !prepareToBuildIndex(io, god, tabletoidxns, tableToIndex, fixedIndexObject ) )
                return DiskLoc();

            if ( ! fixedIndexObject.isEmpty() ) {
                obuf = fixedIndexObject.objdata();
                len = fixedIndexObject.objsize();
            }

        }

        const BSONElement *newId = &writeId;
        int addID = 0;
        if( !god ) {
            //检查对象 是否有_id字段，没有则添加
            //Note that btree buckets which we insert aren't BSONObj's, but in that case god==true.      
            BSONObj io((const char *) obuf);
            BSONElement idField = io.getField( "_id" );
            uassert( 10099 ,  "_id cannot be an array", idField.type() != Array );

            if( idField.eoo() /*判断是否是结束元素*/&& !wouldAddIndex && strstr(ns, ".local.") == 0 ) {
                addID = len;
                if ( writeId.eoo() ) {
                    // 初始化一个_id 随机值（因为_id可能是12 byte类型或其它类型）
                    idToInsert_.oid.init();
                    newId = &idToInsert;//绑定初始化的_id值
                }
                len += newId->size();
            }
            //如果io对象中有时间戳元素时，并用当前时间进行更新
            BSONElementManipulator::lookForTimestamps( io );
        }

        //兼容旧的数据文件
        DiskLoc extentLoc;
        int lenWHdr = len + Record::HeaderSize;
        lenWHdr = (int) (lenWHdr * d->paddingFactor);
        if ( lenWHdr == 0 ) {
            assert( d->paddingFactor == 0 );
            *getDur().writing(&d->paddingFactor) = 1.0;
            lenWHdr = len + Record::HeaderSize;
        }

        // 在对新的对象分配空间前检查数据是否会造成索引冲突（唯一索引）
        // capped标识是否是固定大小的集合类型，这种类型下系统会自动将过于陈旧的数据remove掉
        // 注：此cap与nosql中常说的cap无太大关联
        //     nosql cap即：一致性，有效性，分区容忍性
        //     参见这篇文章:http://blog.nosqlfan.com/html/1112.html,
        //                 http://blog.nosqlfan.com/html/96.html)
        if ( d->nIndexes && d->capped && !god ) {
            checkNoIndexConflicts( d, BSONObj( reinterpret_cast<const char *>( obuf ) ) );
        }

        DiskLoc loc = d->alloc(ns, lenWHdr, extentLoc);//为当前记录分配空间namespace.cpp __stdAlloc方法
        if ( loc.isNull() ) {//如果分配失效
            if ( d->capped == 0 ) { // cap大小未增加，即
                log(1) << "allocating new extent for " << ns << " padding:" << d->paddingFactor << " lenWHdr: " << lenWHdr << endl;
                //尝试从空闲空间列表中分配空间
                cc().database()->allocExtent(ns, Extent::followupSize(lenWHdr, d->lastExtentSize), false);
                //尝试再次为当前记录分配空间
                loc = d->alloc(ns, lenWHdr, extentLoc);
                if ( loc.isNull() ) {
                    log() << "WARNING: alloc() failed after allocating new extent. lenWHdr: " << lenWHdr << " last extent size:" << d->lastExtentSize << "; trying again\n";
                    for ( int zzz=0; zzz<10 && lenWHdr > d->lastExtentSize; zzz++ ) {//最多尝试循环10次分配空间
                        log() << "try #" << zzz << endl;
                        cc().database()->allocExtent(ns, Extent::followupSize(len, d->lastExtentSize), false);
                        loc = d->alloc(ns, lenWHdr, extentLoc);
                        if ( ! loc.isNull() )
                            break;
                    }
                }
            }
            if ( loc.isNull() ) {//最终未分配空间给对象
                log() << "insert: couldn't alloc space for object ns:" << ns << " capped:" << d->capped << endl;
                assert(d->capped);
                return DiskLoc();
            }
        }

        Record *r = loc.rec();
        {
            assert( r->lengthWithHeaders >= lenWHdr );
            r = (Record*) getDur().writingPtr(r, lenWHdr);//持久化插入记录信息
            if( addID ) {
                /* a little effort was made here to avoid a double copy when we add an ID */
                ((int&)*r->data) = *((int*) obuf) + newId->size();
                memcpy(r->data+4, newId->rawdata(), newId->size());//拷贝_id字段到指定记录内存空间
                memcpy(r->data+4+newId->size(), ((char *)obuf)+4, addID-4);//拷贝数据到指定内存空间
            }
            else {
                if( obuf )
                    memcpy(r->data, obuf, len);//直接拷贝数据到记录字段r
            }
        }

        {
            Extent *e = r->myExtent(loc);
            if ( e->lastRecord.isNull() ) {//如果未尾记录为空，本人理解：即之前未插入过记录
                Extent::FL *fl = getDur().writing(e->fl());
                fl->firstRecord = fl->lastRecord = loc;
                r->prevOfs = r->nextOfs = DiskLoc::NullOfs;
            }
            else {
                Record *oldlast = e->lastRecord.rec();//否则将新记录添加到最后一条记录的后面
                r->prevOfs = e->lastRecord.getOfs();
                r->nextOfs = DiskLoc::NullOfs;
                getDur().writingInt(oldlast->nextOfs) = loc.getOfs();
                getDur().writingDiskLoc(e->lastRecord) = loc;
            }
        }

        /* 持久化操作并更新相应统计信息 */
        {
            NamespaceDetails::Stats *s = getDur().writing(&d->stats);
            s->datasize += r->netLength();
            s->nrecords++;
        }

        // 在god时会清空stats信息，同时会添加一个 btree bucket（占据存储空间）
        if ( !god )
            NamespaceDetailsTransient::get_w( ns ).notifyOfWriteOp();//在写操作时清空缓存，优化查询优化

        if ( tableToIndex ) {
            uassert( 13143 , "can't create index on system.indexes" , tabletoidxns.find( ".system.indexes" ) == string::npos );

            BSONObj info = loc.obj();
            bool background = info["background"].trueValue();
            if( background && cc().isSyncThread() ) {
                /* don't do background indexing on slaves.  there are nuances.  this could be added later but requires more code.*/
                log() << "info: indexing in foreground on this replica; was a background index build on the primary" << endl;
                background = false;
            }

            int idxNo = tableToIndex->nIndexes;
            IndexDetails& idx = tableToIndex->addIndex(tabletoidxns.c_str(), !background); // 清空临时缓存信息; 同时递增索引数量
            getDur().writingDiskLoc(idx.info) = loc;
            try {
                buildAnIndex(tabletoidxns, tableToIndex, idx, idxNo, background);//创建索引
            }
            catch( DBException& e ) {
                // 保存异常信息，并执行dropIndexes
                LastError *le = lastError.get();
                int savecode = 0;
                string saveerrmsg;
                if ( le ) {
                    savecode = le->code;
                    saveerrmsg = le->msg;
                }
                else {
                    savecode = e.getCode();
                    saveerrmsg = e.what();
                }

                //回滚索引操作(drop索引)
                string name = idx.indexName();
                BSONObjBuilder b;
                string errmsg;
                bool ok = dropIndexes(tableToIndex, tabletoidxns.c_str(), name.c_str(), errmsg, b, true);
                if( !ok ) {
                    log() << "failed to drop index after a unique key error building it: " << errmsg << ' ' << tabletoidxns << ' ' << name << endl;
                }

                assert( le && !saveerrmsg.empty() );
                raiseError(savecode,saveerrmsg.c_str());
                throw;
            }
        }

        /* 将记录数据添加到索引信息（btree）中 */
        if ( d->nIndexes ) {
            try {
                BSONObj obj(r->data);
                indexRecord(d, obj, loc);
            }
            catch( AssertionException& e ) {
                // _id index 键值重复
                if( tableToIndex || d->capped ) {
                    massert( 12583, "unexpected index insertion failure on capped collection", !d->capped );
                    string s = e.toString();
                    s += " : on addIndex/capped - collection and its index will not match";
                    uassert_nothrow(s.c_str());
                    error() << s << endl;
                }
                else {
                    // 回滚上述操作
                    _deleteRecord(d, ns, r, loc);
                    throw;
                }
            }
        }

        //  out() << "   inserted at loc:" << hex << loc.getOfs() << " lenwhdr:" << hex << lenWHdr << dec << ' ' << ns << endl;
        return loc;
    }

     正如之前所说，该方法会完成添加名空间，添加索引，添加数据记录（memcpy调用）。其中名空间的添加方法addNewNamespaceToCatalog比较简单，下面主要介绍一下索引的创建过程，这里分为了两步：

     1.创建索引树（b树）

     2.将数据（主要是地址）添加到索引（树）中

     先看一下创建索引过程：

static void buildAnIndex(string ns, NamespaceDetails *d, IndexDetails& idx, int idxNo, bool background) {
        tlog() << "building new index on " << idx.keyPattern() << " for " << ns << ( background ? " background" : "" ) << endl;
        Timer t;
        unsigned long long n;

        if( background ) {
            log(2) << "buildAnIndex: background=true\n";
        }

        assert( !BackgroundOperation::inProgForNs(ns.c_str()) ); // should have been checked earlier, better not be...
        assert( d->indexBuildInProgress == 0 );
        assertInWriteLock();
        RecoverableIndexState recoverable( d );
        if( inDBRepair || !background ) {//当数据库在repair时或非后台工作方式下
            n = fastBuildIndex(ns.c_str(), d, idx, idxNo);//创建索引
            assert( !idx.head.isNull() );
        }
        else {
            BackgroundIndexBuildJob j(ns.c_str());//以后台方式创建索引
            n = j.go(ns, d, idx, idxNo);
        }
        tlog() << "done for " << n << " records " << t.millis() / 1000.0 << "secs" << endl;
    }

    创建索引方法会要据创建方式（是否是后台线程等），使用不同的方法，这里主要讲解非后台方式，也就是上面的fastBuildIndex方法（pdfile.cpp第1101行），其定义如下（内容详见注释）：
    
   

unsigned long long fastBuildIndex(const char *ns, NamespaceDetails *d, IndexDetails& idx, int idxNo) {
        CurOp * op = cc().curop();//设置当前操作指针，用于设置操作信息

        Timer t;

        tlog(1) << "fastBuildIndex " << ns << " idxNo:" << idxNo << ' ' << idx.info.obj().toString() << endl;

        bool dupsAllowed = !idx.unique();
        bool dropDups = idx.dropDups() || inDBRepair;
        BSONObj order = idx.keyPattern();

        getDur().writingDiskLoc(idx.head).Null();

        if ( logLevel > 1 ) printMemInfo( "before index start" );

        /* 获取并排序所有键值 ----- */
        unsigned long long n = 0;
        shared_ptr<Cursor> c = theDataFileMgr.findAll(ns);
        BSONObjExternalSorter sorter(order);
        sorter.hintNumObjects( d->stats.nrecords );
        unsigned long long nkeys = 0;
        ProgressMeterHolder pm( op->setMessage( "index: (1/3) external sort" , d->stats.nrecords , 10 ) );
        while ( c->ok() ) {
            BSONObj o = c->current();
            DiskLoc loc = c->currLoc();

            BSONObjSetDefaultOrder keys;
            idx.getKeysFromObject(o, keys);//从对象中获取键值信息
            int k = 0;
            for ( BSONObjSetDefaultOrder::iterator i=keys.begin(); i != keys.end(); i++ ) {
                if( ++k == 2 ) {//是否是多键索引
                    d->setIndexIsMultikey(idxNo);
                }
                sorter.add(*i, loc);//向排序器添加键值和记录位置信息
                nkeys++;
            }

            c->advance();
            n++;
            pm.hit();
            if ( logLevel > 1 && n % 10000 == 0 ) {
                printMemInfo( "\t iterating objects" );
            }

        };
        pm.finished();

        if ( logLevel > 1 ) printMemInfo( "before final sort" );
        sorter.sort();
        if ( logLevel > 1 ) printMemInfo( "after final sort" );

        log(t.seconds() > 5 ? 0 : 1) << "\t external sort used : " << sorter.numFiles() << " files " << " in " << t.seconds() << " secs" << endl;

        list<DiskLoc> dupsToDrop;

        /* 创建索引 */
        {
            BtreeBuilder btBuilder(dupsAllowed, idx);//实例化b树索引对象
            //BSONObj keyLast;
            auto_ptr<BSONObjExternalSorter::Iterator> i = sorter.iterator();//初始化迭代器用于下面遍历
            assert( pm == op->setMessage( "index: (2/3) btree bottom up" , nkeys , 10 ) );
            while( i->more() ) {
                RARELY killCurrentOp.checkForInterrupt();//检查冲突如shutdown或kill指令
                BSONObjExternalSorter::Data d = i->next();

                try {
                    btBuilder.addKey(d.first, d.second);//向b树索引对象中添加索引键值和记录位置信息
                }
                catch( AssertionException& e ) {
                    if ( dupsAllowed ) {
                        // unknow exception??
                        throw;
                    }

                    if( e.interrupted() )
                        throw;

                    if ( ! dropDups )
                        throw;

                    /* we could queue these on disk, but normally there are very few dups, so instead we
                       keep in ram and have a limit.
                    */
                    dupsToDrop.push_back(d.second);
                    uassert( 10092 , "too may dups on index build with dropDups=true", dupsToDrop.size() < 1000000 );
                }
                pm.hit();
            }
            pm.finished();
            op->setMessage( "index: (3/3) btree-middle" );
            log(t.seconds() > 10 ? 0 : 1 ) << "\t done building bottom layer, going to commit" << endl;
            btBuilder.commit();//提交创建索引操作，该方法会完成最终构造Btree索引操作
            wassert( btBuilder.getn() == nkeys || dropDups );
        }

        log(1) << "\t fastBuildIndex dupsToDrop:" << dupsToDrop.size() << endl;
        //删除索引中已出现的重复记录
        for( list<DiskLoc>::iterator i = dupsToDrop.begin(); i != dupsToDrop.end(); i++ )
            theDataFileMgr.deleteRecord( ns, i->rec(), *i, false, true );

        return n;
    }

  

      上面方法主要对要创建的索引信息进行提取，并封装到一个BtreeBuilder中，顾名思义，该对象用于进行b树的创建(因为索引也是一个b树),当信息收集排序完成后，就开始创建索引，如下：

    btree.cpp 1842行
    void BtreeBuilder::commit() {
        buildNextLevel(first);
        committed = true;
    }

     void BtreeBuilder::buildNextLevel(DiskLoc loc) {
        int levels = 1;
        while( 1 ) {
            if( loc.btree()->tempNext().isNull() ) {
                // 在当前层级上只有一个 bucket
                getDur().writingDiskLoc(idx.head) = loc;
                break;
            }
            levels++;

            DiskLoc upLoc = BtreeBucket::addBucket(idx);//添加bucket并实例化上一层DiskLoc
            DiskLoc upStart = upLoc;
            BtreeBucket *up = upLoc.btreemod();//获取上一层的bucket指针

            DiskLoc xloc = loc;
            while( !xloc.isNull() ) {
                RARELY {
                    getDur().commitIfNeeded();
                    b = cur.btreemod();
                    up = upLoc.btreemod();
                }

                BtreeBucket *x = xloc.btreemod();
                BSONObj k;
                DiskLoc r;
                x->popBack(r,k);//弹出当前bucket中最右边的键
                bool keepX = ( x->n != 0 );//当前bucket中元素个数是否为0
                DiskLoc keepLoc = keepX ? xloc : x->nextChild;

                //压入上面弹出的最右边的键值，该键值为当前up（bucket）中最大值
                if ( ! up->_pushBack(r, k, ordering, keepLoc) )
                {
                    // 当前 bucket 已满，则新创建一个addBucket
                    DiskLoc n = BtreeBucket::addBucket(idx);
                    up->tempNext() = n;
                    upLoc = n;
                    up = upLoc.btreemod();
                    up->pushBack(r, k, ordering, keepLoc);
                }

                DiskLoc nextLoc = x->tempNext(); //get next in chain at current level
                if ( keepX ) {//表示当前结点非顶层结点，则设置它的父结点
                    x->parent = upLoc;
                }
                else {
                    if ( !x->nextChild.isNull() )
                        x->nextChild.btreemod()->parent = upLoc;
                    x->deallocBucket( xloc, idx );//删除xloc bucket
                }
                xloc = nextLoc;//指向当前层的下个元素
            }

            loc = upStart;//升级当前结点
            mayCommitProgressDurably();
        }

        if( levels > 1 )
            log(2) << "btree levels: " << levels << endl;
    }

    上面的buildNextLevel方法自下而上根据之前抽取的键值逐层构造一个b树。这里有一个问题需要注意一下，因为mongodb使用bucket来作为b树中的一个层次结点或叶子结点容器（如下图），bucket最大尺寸为8192字节,c。有关b树索引的文章可以参见这篇文章：，
    mongodb目前关于B树索引的文档 ：http://blog.nosqlfan.com/html/758.html

      当初始化了b树索引及空间信息之后，下面就会将数据绑定到相应信息结点上了，也就是DataFileMgr::insert方法(pdfile.cpp文件)的如下代码：
    

/* 将记录数据添加到索引信息（btree）中 */

        if ( d->nIndexes ) {
            try {
                BSONObj obj(r->data);
                indexRecord(d, obj, loc);
            }
            ......
        }  

     上面的indexRecord方法会将键值和数据（包括存储位置）添加到索引中(其中参数d包括之前创建的B树索引信息), 该方法定义如下（pdfile.cpp 第1355行）:

  

/* 将键值和数据（包括存储位置）添加到索引中*/
    static void indexRecord(NamespaceDetails *d, BSONObj obj, DiskLoc loc) {
        int n = d->nIndexesBeingBuilt();//获取已（及正在）构建的索引数
        for ( int i = 0; i < n; i++ ) {
            try {
                bool unique = d->idx(i).unique();
                //内联函数（inline）:将索引和记录相关信息初始化到btree中
                _indexRecord(d, i/*索引顺序位*/, obj, loc, /*dupsAllowed*/!unique);
            }
            catch( DBException& ) {
                /* 如果发生异常，则进行回滚操作
                   note <= i (not < i) is important here as the index we were just attempted
                   may be multikey and require some cleanup.
                */
                for( int j = 0; j <= i; j++ ) {
                    try {
                        _unindexRecord(d->idx(j), obj, loc, false);
                    }
                    catch(...) {
                        log(3) << "unindex fails on rollback after unique failure\n";
                    }
                }
                throw;
            }
        }
    }

    上面的_indexRecord为内联函数（pdfile.cpp）（inline关键字参见C++说明），该参数声明如下：

  

 static inline void  _indexRecord(NamespaceDetails *d, int idxNo, BSONObj& obj, DiskLoc recordLoc, bool dupsAllowed) {
        IndexDetails& idx = d->idx(idxNo);//
        BSONObjSetDefaultOrder keys;
        idx.getKeysFromObject(obj, keys);//从对象信息中获取键属性信息
        BSONObj order = idx.keyPattern();
        Ordering ordering = Ordering::make(order);//初始化排序方式用于下面传参
        int n = 0;
        for ( BSONObjSetDefaultOrder::iterator i=keys.begin(); i != keys.end(); i++ ) {
            if( ++n == 2 ) {
                d->setIndexIsMultikey(idxNo);//设置多键值索引
            }
            assert( !recordLoc.isNull() );
            try {
                idx.head/*DiskLoc*/.btree()/*BtreeBucket*/->bt_insert(idx.head, recordLoc, //执行向btree中添加记录和绑定索引信息的操作
                                            *i, ordering, dupsAllowed, idx);
            }
            catch (AssertionException& e) {
                if( e.getCode() == 10287 && idxNo == d->nIndexes ) {
                    DEV log() << "info: caught key already in index on bg indexing (ok)" << endl;
                    continue;
                }
                if( !dupsAllowed ) {
                    // 重复键值异常
                    throw;
                }
                problem() << " caught assertion _indexRecord " << idx.indexNamespace() << endl;
            }
        }
    }

    上面方法最终会执行b树插入方法bt_insert（btree.cpp文件1622行），如下（详情见注释）：
 

   int BtreeBucket::bt_insert(const DiskLoc thisLoc, const DiskLoc recordLoc,
                               const BSONObj& key, const Ordering &order, bool dupsAllowed,
                               IndexDetails& idx, bool toplevel) const {
        if ( toplevel ) {//如果是顶级节点（如果是通过构造索引方式调用 ，则toplevel=true）
            //判断键值是否过界（因为其会存储在system.indexs中），其中：KeyMax = 8192 / 10 .mongodb开发团队可能会在更高版本中扩大该值
            if ( key.objsize() > KeyMax ) {
                problem() << "Btree::insert: key too large to index, skipping " << idx.indexNamespace() << ' ' << key.objsize() << ' ' << key.toString() << endl;
                return 3;
            }
        }
        //执行添加操作
        int x = _insert(thisLoc, recordLoc, key, order, dupsAllowed, DiskLoc(), DiskLoc(), idx);
        assertValid( order );//assert排序方式是否有效

        return x;
    }

    上面代码紧接着会调用btree.cpp文件的内部方法_insert（btree.cpp文件 1554行）：

 

   int BtreeBucket::_insert(const DiskLoc thisLoc, const DiskLoc recordLoc,
                             const BSONObj& key, const Ordering &order, bool dupsAllowed,
                             const DiskLoc lChild, const DiskLoc rChild, IndexDetails& idx) const {
        if ( key.objsize() > KeyMax ) {
            problem() << "ERROR: key too large len:" << key.objsize() << " max:" << KeyMax << ' ' << key.objsize() << ' ' << idx.indexNamespace() << endl;
            return 2;
        }
        assert( key.objsize() > 0 );

        int pos;
        //在btree bucket中使用二分查询，查看键值是否已在所索引信息中
        bool found = find(idx, key, recordLoc, order, pos /*返回该索引信息所在或应该在的位置*/, !dupsAllowed);
        if ( insert_debug ) {
            out() << "  " << thisLoc.toString() << '.' << "_insert " <<
                  key.toString() << '/' << recordLoc.toString() <<
                  " l:" << lChild.toString() << " r:" << rChild.toString() << endl;
            out() << "    found:" << found << " pos:" << pos << " n:" << n << endl;
        }

        if ( found ) {
            const _KeyNode& kn = k(pos);//获取指定磁盘位置的节点信息，_KeyNode
            if ( kn.isUnused() ) {//查看已存在的键结点是否已使用
                log(4) << "btree _insert: reusing unused key" << endl;
                massert( 10285 , "_insert: reuse key but lchild is not null", lChild.isNull());
                massert( 10286 , "_insert: reuse key but rchild is not null", rChild.isNull());
                kn.writing().setUsed();
                return 0;
            }

            DEV {
                log() << "_insert(): key already exists in index (ok for background:true)\n";
                log() << "  " << idx.indexNamespace() << " thisLoc:" << thisLoc.toString() << '\n';
                log() << "  " << key.toString() << '\n';
                log() << "  " << "recordLoc:" << recordLoc.toString() << " pos:" << pos << endl;
                log() << "  old l r: " << childForPos(pos).toString() << ' ' << childForPos(pos+1).toString() << endl;
                log() << "  new l r: " << lChild.toString() << ' ' << rChild.toString() << endl;
            }
            alreadyInIndex();//提示键值结点已在索引中，不必再创建，并抛出异常
        }

        DEBUGGING out() << "TEMP: key: " << key.toString() << endl;
        DiskLoc child = childForPos(pos);//查询当前pos的子结点信息，以寻找插入位置
        if ( insert_debug )
            out() << "    getChild(" << pos << "): " << child.toString() << endl;
        if ( child.isNull() || !rChild.isNull() /* 在当前buckets中插入，即 'internal' 插入 */ ) {
            insertHere(thisLoc, pos, recordLoc, key, order, lChild, rChild, idx);//在当前buckets中插入
            return 0;
        }
        //如果有子结点，则在子结点上执行插入操作
        return child.btree()->bt_insert(child, recordLoc, key, order, dupsAllowed, idx, /*toplevel*/false);
    }

    上面_insert方法首先会使用二分法查找要插入的记录是否已存在于索引中，同时会返回一个插入点（pos），如不存在则会进一步在插入点位置查看找元素以决定是在当前bucket中插入，还是在当前pos位置的（右）子结点(bucket)上插入（这会再次递归调用上面的bt_insert方法），这里我们假定在当前bucket插入，则会执行insertHere方法（btree.cpp文件1183行），它的定义如下：
    
  

 /**
     * insert a key in this bucket, splitting if necessary.
     * @keypos - where to insert the key in range 0..n.  0=make leftmost, n=make rightmost.
     * NOTE this function may free some data, and as a result the value passed for keypos may
     * be invalid after calling insertHere()
     */
    void BtreeBucket::insertHere( const DiskLoc thisLoc, int keypos,
                                  const DiskLoc recordLoc, const BSONObj& key, const Ordering& order,
                                  const DiskLoc lchild, const DiskLoc rchild, IndexDetails& idx) const {
        if ( insert_debug )
            out() << "   " << thisLoc.toString() << ".insertHere " << key.toString() << '/' << recordLoc.toString() << ' '
                  << lchild.toString() << ' ' << rchild.toString() << " keypos:" << keypos << endl;

        DiskLoc oldLoc = thisLoc;
        //根据keypos插入相应位置并将数据memcpy到内存指定位置
        if ( !basicInsert(thisLoc, keypos, recordLoc, key, order) ) {
            //如果插入无效，表示当前bucket已满，则分割记录并放到新创建的bucket中
            thisLoc.btreemod()->split(thisLoc, keypos, recordLoc, key, order, lchild, rchild, idx);
            return;
        }

        {//持久化当前thisLoc的结点信息并根据插入位置（是否最后一个key），来更新当前thisLoc（及后面key结点）的子结点信息
            const _KeyNode *_kn = &k(keypos);
            _KeyNode *kn = (_KeyNode *) getDur().alreadyDeclared((_KeyNode*) _kn); // already declared intent in basicInsert()
            if ( keypos+1 == n ) { // n为pack（打包后）存储的记录数，这里"判断等于n"表示为最后(last)一个key
                if ( nextChild != lchild ) {//如果是最后元素，那么"当前最高键值的右子结点应该与要插入的左子结点相同
                    out() << "ERROR nextChild != lchild" << endl;
                    out() << "  thisLoc: " << thisLoc.toString() << ' ' << idx.indexNamespace() << endl;
                    out() << "  keyPos: " << keypos << " n:" << n << endl;
                    out() << "  nextChild: " << nextChild.toString() << " lchild: " << lchild.toString() << endl;
                    out() << "  recordLoc: " << recordLoc.toString() << " rchild: " << rchild.toString() << endl;
                    out() << "  key: " << key.toString() << endl;
                    dump();
                    assert(false);
                }
                kn->prevChildBucket = nextChild;//"当前最高键值的右子结点”绑定到持久化结点的左子结点
                assert( kn->prevChildBucket == lchild );
                nextChild.writing() = rchild;//持久化"当前最高键值的右子结点”，并将“要插入结点”的右子结点绑定到
                if ( !rchild.isNull() )//如果有右子结点，则更新右子结点的父结点信息为当前thisLoc
                    rchild.btree()->parent.writing() = thisLoc;
            }
            else {
                //如果keypos位置不是最后一个
                kn->prevChildBucket = lchild;//将左子结点绑定到keypos位置结点的左子结点上
                if ( k(keypos+1).prevChildBucket != lchild ) {//这时左子结点应该与下一个元素的左子结点相同
                    out() << "ERROR k(keypos+1).prevChildBucket != lchild" << endl;
                    out() << "  thisLoc: " << thisLoc.toString() << ' ' << idx.indexNamespace() << endl;
                    out() << "  keyPos: " << keypos << " n:" << n << endl;
                    out() << "  k(keypos+1).pcb: " << k(keypos+1).prevChildBucket.toString() << " lchild: " << lchild.toString() << endl;
                    out() << "  recordLoc: " << recordLoc.toString() << " rchild: " << rchild.toString() << endl;
                    out() << "  key: " << key.toString() << endl;
                    dump();
                    assert(false);
                }
                const DiskLoc *pc = &k(keypos+1).prevChildBucket;//获取keypos后面元素的左子结点信息
                *getDur().alreadyDeclared((DiskLoc*) pc) = rchild; // 将右子结点绑定到下一个元素（keypos+1）的左子结点上declared in basicInsert()
                if ( !rchild.isNull() )//如果有右子结点，则更新右子结点的父结点信息为当前thisLoc
                    rchild.btree()->parent.writing() = thisLoc;
            }
            return;
        }
    }

       该方法中会调用一个叫basicInsert的方法，它主要会在当前bucket中指定位置（keypos）添加记录信息，同时持久化该结点信息，如下：

   

//tree.cpp 1183
     bool BucketBasics::basicInsert(const DiskLoc thisLoc, int &keypos, const DiskLoc recordLoc, const BSONObj& key, const Ordering &order) const {
        assert( keypos >= 0 && keypos <= n );
        //判断bucket剩余的空间是否满足当前数据需要的存储空间
        int bytesNeeded = key.objsize() + sizeof(_KeyNode);
        if ( bytesNeeded > emptySize ) {
            _pack(thisLoc, order, keypos);//如不够用，进行一次整理打包操作，以为bucket中整理更多空间
            if ( bytesNeeded > emptySize )//如还不够用，则返回
                return false;
        }

        BucketBasics *b;//声明Bucket管理对象指针，该对象提供了Bucket存储管理的基本操作和属性，如insert,_pack等
        {
            const char *p = (const char *) &k(keypos);
            const char *q = (const char *) &k(n+1);
            // declare that we will write to [k(keypos),k(n)]
            // todo: this writes a medium amount to the journal.  we may want to add a verb "shift" to the redo log so
            //       we can log a very small amount.
            b = (BucketBasics*) getDur().writingAtOffset((void *) this, p-(char*)this, q-p);
            //如已有3个结点，目前要插到第三个结点之间，则对每三个元素进行迁移，
            // e.g. n==3, keypos==2
            // 1 4 9
            // ->
            // 1 4 _ 9
            for ( int j = n; j > keypos; j-- ) // make room
                b->k(j) = b->k(j-1);
        }
        getDur().declareWriteIntent(&b->emptySize, 12); // [b->emptySize..b->n] is 12 bytes and we are going to write those
        b->emptySize -= sizeof(_KeyNode);//将当前bucket中的剩余空闲空间减少
        b->n++;//已有结点数加1

        _KeyNode& kn = b->k(keypos);
        kn.prevChildBucket.Null();//设置当前结点的左子结点为空
        kn.recordLoc = recordLoc;//绑定结点记录信息
        kn.setKeyDataOfs((short) b->_alloc(key.objsize()) );//设置结点数据偏移信息
        char *p = b->dataAt(kn.keyDataOfs());//实例化指向磁盘数据(journal文件)位置（含偏移量）的指针
        getDur().declareWriteIntent(p, key.objsize());//持久化结点数据信息
        memcpy(p, key.objdata(), key.objsize());//将当前结点信息复制到p指向的地址空间
        return true;
    }

    如果上面方法调用失效，则意味着当前 bucket中已有可用空间插入新记录，这时系统会调用 split（btree.cpp文件 1240行）方法来进行bucket分割，以创建新的bucket并将信息塞入其中，如下：

    void BtreeBucket::split(const DiskLoc thisLoc, int keypos, const DiskLoc recordLoc, const BSONObj& key, const Ordering& order, const DiskLoc lchild, const DiskLoc rchild, IndexDetails& idx) {
        assertWritable();

        if ( split_debug )
            out() << "    " << thisLoc.toString() << ".split" << endl;

        int split = splitPos( keypos );//找到要迁移的数据位置
        DiskLoc rLoc = addBucket(idx);//添加一个新的BtreeBucket
        BtreeBucket *r = rLoc.btreemod();
        if ( split_debug )
            out() << "     split:" << split << ' ' << keyNode(split).key.toString() << " n:" << n << endl;
        for ( int i = split+1; i < n; i++ ) {
            KeyNode kn = keyNode(i);
            r->pushBack(kn.recordLoc, kn.key, order, kn.prevChildBucket);//向新bucket中迁移过剩数据
        }
        r->nextChild = nextChild;//绑定新bucket的右子结点
        r->assertValid( order );

        if ( split_debug )
            out() << "     new rLoc:" << rLoc.toString() << endl;
        r = 0;
        rLoc.btree()->fixParentPtrs(rLoc);//设置当前bucket树的父指针信息

        {
            KeyNode splitkey = keyNode(split);//获取内存中分割点位置所存储的数据
            nextChild = splitkey.prevChildBucket; // 提升splitkey 键，它的子结点将会是 thisLoc (l) 和 rLoc (r)
            if ( split_debug ) {
                out() << "    splitkey key:" << splitkey.key.toString() << endl;
            }

            // 将 splitkey 提升为父结点
            if ( parent.isNull() ) {
                // 如果无父结点时，则创建一个，并将
                DiskLoc L = addBucket(idx);
                BtreeBucket *p = L.btreemod();
                p->pushBack(splitkey.recordLoc, splitkey.key, order, thisLoc);
                p->nextChild = rLoc;//将分割的bucket为了当前
                p->assertValid( order );
                parent = idx.head.writing() = L;//将splitkey 提升为父结点
                if ( split_debug )
                    out() << "    we were root, making new root:" << hex << parent.getOfs() << dec << endl;
                rLoc.btree()->parent.writing() = parent;
            }
            else {
                // set this before calling _insert - if it splits it will do fixParent() logic and change the value.
                rLoc.btree()->parent.writing() = parent;
                if ( split_debug )
                    out() << "    promoting splitkey key " << splitkey.key.toString() << endl;
                //提升splitkey键，它的左子结点 thisLoc， 右子点rLoc
                parent.btree()->_insert(parent, splitkey.recordLoc, splitkey.key, order, /*dupsallowed*/true, thisLoc, rLoc, idx);
            }
        }

        int newpos = keypos;
        // 打包压缩数据(pack,移除无用数据)，以提供更多空间
        truncateTo(split, order, newpos);  // note this may trash splitkey.key.  thus we had to promote it before finishing up here.

        // add our new key, there is room now
        {
            if ( keypos <= split ) {//如果还有空间存储新键
                if ( split_debug )
                    out() << "  keypos<split, insertHere() the new key" << endl;
                insertHere(thisLoc, newpos, recordLoc, key, order, lchild, rchild, idx);//再次向当前bucket中添加记录
            }
            else {//如压缩之后依旧无可用空间，则向新创建的bucket中添加节点
                int kp = keypos-split-1;
                assert(kp>=0);
                rLoc.btree()->insertHere(rLoc, kp, recordLoc, key, order, lchild, rchild, idx);
            }
        }

        if ( split_debug )
            out() << "     split end " << hex << thisLoc.getOfs() << dec << endl;
    }

    好了，今天的内容到这里就告一段落了，在接下来的文章中，将会介绍客户端发起Delete操作时，Mongodb的执行流程和相应实现部分。

    原文链接:http://www.cnblogs.com/daizhj/archive/2011/03/30/1999699.html
    作者: daizhj, 代震军   
    微博: http://t.sina.com.cn/daizhj
    Tags: mongodb,c++,btree