bool _userCreateNS(const char *ns, const BSONObj& options, string& err, bool *deferIdIndex) {
LOG(1) << "create collection " << ns << ' ' << options << endl;
if ( nsdetails(ns) ) {
err = "collection already exists";
return false;
}
long long size = Extent::initialSize(128);
{
BSONElement e = options.getField("size");
if ( e.isNumber() ) {
size = e.numberLong();
uassert( 10083 , "create collection invalid size spec", size >= 0 );
size += 0xff;
size &= 0xffffffffffffff00LL;
if ( size < Extent::minSize() )
size = Extent::minSize();
}
}
bool newCapped = false;
long long mx = 0;
if( options["capped"].trueValue() ) {
newCapped = true;
BSONElement e = options.getField("max");
if ( e.isNumber() ) {
mx = e.numberLong();
uassert( 16495,
"max in a capped collection has to be < 2^31 or not set",
NamespaceDetails::validMaxCappedDocs(&mx) );
}
}
cc().database()->createCollection( ns, options["capped"].trueValue(), &options );
Collection* collection = cc().database()->getCollection( ns );
verify( collection );
// $nExtents just for debug/testing.
BSONElement e = options.getField( "$nExtents" );
if ( e.type() == Array ) {
// We create one extent per array entry, with size specified
// by the array value.
BSONObjIterator i( e.embeddedObject() );
while( i.more() ) {
BSONElement e = i.next();
int size = int( e.number() );
verify( size <= 0x7fffffff );
// $nExtents is just for testing - always allocate new extents
// rather than reuse existing extents so we have some predictibility
// in the extent size used by our tests
collection->increaseStorageSize( (int)size, false );
}
}
else if ( int( e.number() ) > 0 ) {
// We create '$nExtents' extents, each of size 'size'.
int nExtents = int( e.number() );
verify( size <= 0x7fffffff );
for ( int i = 0; i < nExtents; ++i ) {
verify( size <= 0x7fffffff );
// $nExtents is just for testing - always allocate new extents
// rather than reuse existing extents so we have some predictibility
// in the extent size used by our tests
collection->increaseStorageSize( (int)size, false );
}
}
else {
// This is the non test case, where we don't have a $nExtents spec.
while ( size > 0 ) {
const int max = Extent::maxSize();
const int min = Extent::minSize();
int desiredExtentSize = static_cast<int> (size > max ? max : size);
desiredExtentSize = static_cast<int> (desiredExtentSize < min ? min : desiredExtentSize);
desiredExtentSize &= 0xffffff00;
Extent* e = collection->increaseStorageSize( (int)desiredExtentSize, true );
size -= e->length;
}
}
NamespaceDetails *d = nsdetails(ns);
verify(d);
bool ensure = true;
// respect autoIndexId if set. otherwise, create an _id index for all colls, except for
// capped ones in local w/o autoIndexID (reason for the exception is for the oplog and
// non-replicated capped colls)
if( options.hasField( "autoIndexId" ) ||
(newCapped && nsToDatabase( ns ) == "local" ) ) {
ensure = options.getField( "autoIndexId" ).trueValue();
}
if( ensure ) {
if( deferIdIndex )
*deferIdIndex = true;
else
ensureIdIndexForNewNs( ns );
}
if ( mx > 0 )
d->setMaxCappedDocs( mx );
if ( options["flags"].numberInt() ) {
d->replaceUserFlags( options["flags"].numberInt() );
}
return true;
}
DiskLoc DataFileMgr::insert(const char* ns,
const void* obuf,
int32_t len,
bool mayInterrupt,
bool god,
bool mayAddIndex,
bool* addedID) {
Database* database = cc().database();
bool wouldAddIndex = false;
massert( 10093 , "cannot insert into reserved $ collection", god || NamespaceString::normal( ns ) );
uassert( 10094 , str::stream() << "invalid ns: " << ns , isValidNS( ns ) );
{
const char *sys = strstr(ns, "system.");
if ( sys && !insert_checkSys(sys, ns, wouldAddIndex, obuf, god) )
return DiskLoc();
}
bool addIndex = wouldAddIndex && mayAddIndex;
Collection* collection = database->getCollection( ns );
if ( collection == NULL ) {
collection = database->createCollection( ns, false, NULL );
int ies = Extent::initialSize(len);
if( str::contains(ns, '$') &&
len + Record::HeaderSize >= BtreeData_V1::BucketSize - 256 &&
len + Record::HeaderSize <= BtreeData_V1::BucketSize + 256 ) {
// probably an index. so we pick a value here for the first extent instead of using
// initialExtentSize() which is more for user collections.
// TODO: we could look at the # of records in the parent collection to be smarter here.
ies = (32+4) * 1024;
}
collection->increaseStorageSize( ies, false);
if ( !god )
ensureIdIndexForNewNs(ns);
}
NamespaceDetails* d = collection->details();
string tabletoidxns;
Collection* collectionToIndex = 0;
NamespaceDetails* tableToIndex = 0;
BSONObj fixedIndexObject;
if ( addIndex ) {
verify( obuf );
BSONObj io((const char *) obuf);
tabletoidxns = io.getStringField( "ns" );
uassert(10096, "invalid ns to index", tabletoidxns.find( '.' ) != string::npos);
massert(10097,
str::stream() << "trying to create index on wrong db "
<< " db: " << database->name() << " collection: " << tabletoidxns,
database->ownsNS( tabletoidxns ) );
collectionToIndex = database->getCollection( tabletoidxns );
if ( !collectionToIndex ) {
collectionToIndex = database->createCollection( tabletoidxns, false, NULL );
verify( collectionToIndex );
if ( !god )
ensureIdIndexForNewNs( tabletoidxns.c_str() );
}
tableToIndex = collectionToIndex->details();
Status status = collectionToIndex->getIndexCatalog()->okToAddIndex( io );
if ( status.code() == ErrorCodes::IndexAlreadyExists ) {
// dup index, we ignore
return DiskLoc();
}
uassert( 17199,
str::stream() << "cannot build index on " << tabletoidxns
<< " because of " << status.toString(),
status.isOK() );
if( !prepareToBuildIndex(io,
mayInterrupt,
god,
tabletoidxns ) ) {
// prepare creates _id itself, or this indicates to fail the build silently (such
// as if index already exists)
return DiskLoc();
}
fixedIndexObject = IndexCatalog::fixIndexSpec( io );
obuf = fixedIndexObject.objdata();
len = fixedIndexObject.objsize();
}
IDToInsert idToInsert; // only initialized if needed
if( !god ) {
/* Check if we have an _id field. If we don't, we'll add it.
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 );
// we don't add _id for capped collections in local as they don't have an _id index
if( idField.eoo() &&
!wouldAddIndex &&
nsToDatabase( ns ) != "local" &&
d->haveIdIndex() ) {
if( addedID )
*addedID = true;
idToInsert.init();
len += idToInsert.size();
}
BSONElementManipulator::lookForTimestamps( io );
}
int lenWHdr = d->getRecordAllocationSize( len + Record::HeaderSize );
fassert( 16440, lenWHdr >= ( len + Record::HeaderSize ) );
// If the collection is capped, check if the new object will violate a unique index
// constraint before allocating space.
if ( d->isCapped() && !god) {
BSONObj temp = BSONObj( reinterpret_cast<const char *>( obuf ) );
Status ret = collection->getIndexCatalog()->checkNoIndexConflicts( temp );
uassert(12582, "duplicate key insert for unique index of capped collection", ret.isOK() );
}
DiskLoc loc = allocateSpaceForANewRecord(ns, d, lenWHdr, god);
if ( loc.isNull() ) {
log() << "insert: couldn't alloc space for object ns:" << ns
<< " capped:" << d->isCapped() << endl;
verify(d->isCapped());
return DiskLoc();
}
Record *r = loc.rec();
{
verify( r->lengthWithHeaders() >= lenWHdr );
r = (Record*) getDur().writingPtr(r, lenWHdr);
if( idToInsert.needed() ) {
/* a little effort was made here to avoid a double copy when we add an ID */
int originalSize = *((int*) obuf);
((int&)*r->data()) = originalSize + idToInsert.size();
memcpy(r->data()+4, idToInsert.rawdata(), idToInsert.size());
memcpy(r->data()+4+idToInsert.size(), ((char*)obuf)+4, originalSize-4);
}
else {
if( obuf ) // obuf can be null from internal callers
memcpy(r->data(), obuf, len);
}
}
addRecordToRecListInExtent(r, loc);
d->incrementStats( r->netLength(), 1 );
// we don't bother resetting query optimizer stats for the god tables - also god is true when adding a btree bucket
if ( !god )
collection->infoCache()->notifyOfWriteOp();
if ( tableToIndex ) {
insert_makeIndex(collectionToIndex, loc, mayInterrupt);
}
/* add this record to our indexes */
if ( d->getTotalIndexCount() > 0 ) {
try {
BSONObj obj(r->data());
collection->getIndexCatalog()->indexRecord(obj, loc);
}
catch( AssertionException& e ) {
// should be a dup key error on _id index
if( tableToIndex || d->isCapped() ) {
massert( 12583, "unexpected index insertion failure on capped collection", !d->isCapped() );
string s = e.toString();
s += " : on addIndex/capped - collection and its index will not match";
setLastError(0, s.c_str());
error() << s << endl;
}
else {
// normal case -- we can roll back
_deleteRecord(d, ns, r, loc);
throw;
}
}
}
d->paddingFits();
return loc;
}
bool Cloner::go(const char *masterHost, string& errmsg, const string& fromdb, bool logForRepl, bool slaveOk, bool useReplAuth, bool snapshot) {
massert( 10289 , "useReplAuth is not written to replication log", !useReplAuth || !logForRepl );
string todb = cc().database()->name;
stringstream a,b;
a << "localhost:" << cmdLine.port;
b << "127.0.0.1:" << cmdLine.port;
bool masterSameProcess = ( a.str() == masterHost || b.str() == masterHost );
if ( masterSameProcess ) {
if ( fromdb == todb && cc().database()->path == dbpath ) {
// guard against an "infinite" loop
/* if you are replicating, the local.sources config may be wrong if you get this */
errmsg = "can't clone from self (localhost).";
return false;
}
}
/* todo: we can put these releases inside dbclient or a dbclient specialization.
or just wait until we get rid of global lock anyway.
*/
string ns = fromdb + ".system.namespaces";
list<BSONObj> toClone;
{
dbtemprelease r;
auto_ptr<DBClientCursor> c;
{
if ( conn.get() ) {
// nothing to do
} else if ( !masterSameProcess ) {
auto_ptr< DBClientConnection > c( new DBClientConnection() );
if ( !c->connect( masterHost, errmsg ) )
return false;
if( !replAuthenticate(c.get()) )
return false;
conn = c;
} else {
conn.reset( new DBDirectClient() );
}
c = conn->query( ns.c_str(), BSONObj(), 0, 0, 0, slaveOk ? QueryOption_SlaveOk : 0 );
}
if ( c.get() == 0 ) {
errmsg = "query failed " + ns;
return false;
}
while ( c->more() ){
BSONObj collection = c->next();
log(2) << "\t cloner got " << collection << endl;
BSONElement e = collection.getField("name");
if ( e.eoo() ) {
string s = "bad system.namespaces object " + collection.toString();
massert( 10290 , s.c_str(), false);
}
assert( !e.eoo() );
assert( e.type() == String );
const char *from_name = e.valuestr();
if( strstr(from_name, ".system.") ) {
/* system.users is cloned -- but nothing else from system. */
if( legalClientSystemNS( from_name , true ) == 0 ){
log(2) << "\t\t not cloning because system collection" << endl;
continue;
}
}
if( ! nsDollarCheck( from_name ) ){
log(2) << "\t\t not cloning because has $ " << endl;
continue;
}
toClone.push_back( collection.getOwned() );
}
}
for ( list<BSONObj>::iterator i=toClone.begin(); i != toClone.end(); i++ ){
{
dbtemprelease r;
}
BSONObj collection = *i;
log(2) << " really will clone: " << collection << endl;
const char * from_name = collection["name"].valuestr();
BSONObj options = collection.getObjectField("options");
/* change name "<fromdb>.collection" -> <todb>.collection */
const char *p = strchr(from_name, '.');
assert(p);
string to_name = todb + p;
bool wantIdIndex = false;
{
string err;
const char *toname = to_name.c_str();
/* we defer building id index for performance - building it in batch is much faster */
userCreateNS(toname, options, err, logForRepl, &wantIdIndex);
}
log(1) << "\t\t cloning " << from_name << " -> " << to_name << endl;
Query q;
if( snapshot )
q.snapshot();
copy(from_name, to_name.c_str(), false, logForRepl, masterSameProcess, slaveOk, q);
if( wantIdIndex ) {
/* we need dropDups to be true as we didn't do a true snapshot and this is before applying oplog operations
that occur during the initial sync. inDBRepair makes dropDups be true.
*/
bool old = inDBRepair;
try {
inDBRepair = true;
ensureIdIndexForNewNs(to_name.c_str());
inDBRepair = old;
}
catch(...) {
inDBRepair = old;
throw;
}
}
}
// now build the indexes
string system_indexes_from = fromdb + ".system.indexes";
string system_indexes_to = todb + ".system.indexes";
/* [dm]: is the ID index sometimes not called "_id_"? There is other code in the system that looks for a "_id" prefix
rather than this exact value. we should standardize. OR, remove names - which is in the bugdb. Anyway, this
is dubious here at the moment.
*/
copy(system_indexes_from.c_str(), system_indexes_to.c_str(), true, logForRepl, masterSameProcess, slaveOk, BSON( "name" << NE << "_id_" ) );
return true;
}