bool run(OperationContext* txn,
const string& dbname,
BSONObj& cmdObj,
int,
string& errmsg,
BSONObjBuilder& result) {
int numSet = 0;
bool found = false;
const ServerParameter::Map& parameterMap = ServerParameterSet::getGlobal()->getMap();
// First check that we aren't setting the same parameter twice and that we actually are
// setting parameters that we have registered and can change at runtime
BSONObjIterator parameterCheckIterator(cmdObj);
// We already know that "setParameter" will be the first element in this object, so skip
// past that
parameterCheckIterator.next();
// Set of all the parameters the user is attempting to change
std::map<std::string, BSONElement> parametersToSet;
// Iterate all parameters the user passed in to do the initial validation checks,
// including verifying that we are not setting the same parameter twice.
while (parameterCheckIterator.more()) {
BSONElement parameter = parameterCheckIterator.next();
std::string parameterName = parameter.fieldName();
ServerParameter::Map::const_iterator foundParameter = parameterMap.find(parameterName);
// Check to see if this is actually a valid parameter
if (foundParameter == parameterMap.end()) {
errmsg = str::stream() << "attempted to set unrecognized parameter ["
<< parameterName << "], use help:true to see options ";
return false;
}
// Make sure we are allowed to change this parameter
if (!foundParameter->second->allowedToChangeAtRuntime()) {
errmsg = str::stream() << "not allowed to change [" << parameterName
<< "] at runtime";
return false;
}
// Make sure we are only setting this parameter once
if (parametersToSet.count(parameterName)) {
errmsg = str::stream()
<< "attempted to set parameter [" << parameterName
<< "] twice in the same setParameter command, "
<< "once to value: [" << parametersToSet[parameterName].toString(false)
<< "], and once to value: [" << parameter.toString(false) << "]";
return false;
}
parametersToSet[parameterName] = parameter;
}
// Iterate the parameters that we have confirmed we are setting and actually set them.
// Not that if setting any one parameter fails, the command will fail, but the user
// won't see what has been set and what hasn't. See SERVER-8552.
for (std::map<std::string, BSONElement>::iterator it = parametersToSet.begin();
it != parametersToSet.end();
++it) {
BSONElement parameter = it->second;
std::string parameterName = it->first;
ServerParameter::Map::const_iterator foundParameter = parameterMap.find(parameterName);
if (foundParameter == parameterMap.end()) {
errmsg = str::stream() << "Parameter: " << parameterName << " that was "
<< "avaliable during our first lookup in the registered "
<< "parameters map is no longer available.";
return false;
}
if (numSet == 0) {
foundParameter->second->append(txn, result, "was");
}
Status status = foundParameter->second->set(parameter);
if (status.isOK()) {
numSet++;
continue;
}
errmsg = status.reason();
result.append("code", status.code());
return false;
}
if (numSet == 0 && !found) {
errmsg = "no option found to set, use help:true to see options ";
return false;
}
return true;
}
void NetworkCounter::append(BSONObjBuilder& b) {
b.append("bytesIn", static_cast<long long>(_bytesIn.loadRelaxed()));
b.append("bytesOut", static_cast<long long>(_bytesOut.loadRelaxed()));
b.append("numRequests", static_cast<long long>(_requests.loadRelaxed()));
}
inline void kill_wrapper( ProcessId pid, int sig, int port, const BSONObj& opt ) {
#ifdef _WIN32
if (sig == SIGKILL || port == 0) {
verify( registry._handles.count(pid) );
TerminateProcess(registry._handles[pid], 1); // returns failure for "zombie" processes.
return;
}
std::string eventName = getShutdownSignalName(pid.asUInt32());
HANDLE event = OpenEventA(EVENT_MODIFY_STATE, FALSE, eventName.c_str());
if (event == NULL) {
int gle = GetLastError();
if (gle != ERROR_FILE_NOT_FOUND) {
warning() << "kill_wrapper OpenEvent failed: " << errnoWithDescription();
}
else {
log() << "kill_wrapper OpenEvent failed to open event to the process "
<< pid.asUInt32()
<< ". It has likely died already or server is running an older version."
<< " Attempting to shutdown through admin command.";
// Back-off to the old way of shutting down the server on Windows, in case we
// are managing a pre-2.6.0rc0 service, which did not have the event.
//
try {
DBClientConnection conn;
conn.connect("127.0.0.1:" + BSONObjBuilder::numStr(port));
BSONElement authObj = opt["auth"];
if (!authObj.eoo()){
string errMsg;
conn.auth("admin", authObj["user"].String(),
authObj["pwd"].String(), errMsg);
if (!errMsg.empty()) {
cout << "Failed to authenticate before shutdown: "
<< errMsg << endl;
}
}
BSONObj info;
BSONObjBuilder b;
b.append("shutdown", 1);
b.append("force", 1);
conn.runCommand("admin", b.done(), info);
}
catch (...) {
// Do nothing. This command never returns data to the client and the driver
// doesn't like that.
//
}
}
return;
}
ON_BLOCK_EXIT(CloseHandle, event);
bool result = SetEvent(event);
if (!result) {
error() << "kill_wrapper SetEvent failed: " << errnoWithDescription();
return;
}
#else
int x = kill( pid.toNative(), sig );
if ( x ) {
if ( errno == ESRCH ) {
}
else {
log() << "killFailed: " << errnoWithDescription() << endl;
verify( x == 0 );
}
}
#endif
}
virtual bool run(const string&, BSONObj& cmdObj, int, string& errmsg, BSONObjBuilder& result, bool fromRepl) {
result.append("argv", CmdLine::getArgvArray());
result.append("parsed", CmdLine::getParsedOpts());
return true;
}
static bool receivedQuery(Client& c, DbResponse& dbresponse, Message& m ) {
bool ok = true;
MSGID responseTo = m.header()->id;
DbMessage d(m);
QueryMessage q(d);
auto_ptr< Message > resp( new Message() );
CurOp& op = *(c.curop());
shared_ptr<AssertionException> ex;
try {
dbresponse.exhaust = runQuery(m, q, op, *resp);
verify( !resp->empty() );
}
catch ( SendStaleConfigException& e ){
ex.reset( new SendStaleConfigException( e.getns(), e.getInfo().msg, e.getVersionReceived(), e.getVersionWanted() ) );
ok = false;
}
catch ( AssertionException& e ) {
ex.reset( new AssertionException( e.getInfo().msg, e.getCode() ) );
ok = false;
}
if( ex ){
op.debug().exceptionInfo = ex->getInfo();
LOGWITHRATELIMIT {
log() << "assertion " << ex->toString() << " ns:" << q.ns << " query:" <<
(q.query.valid() ? q.query.toString() : "query object is corrupt") << endl;
if( q.ntoskip || q.ntoreturn )
log() << " ntoskip:" << q.ntoskip << " ntoreturn:" << q.ntoreturn << endl;
}
SendStaleConfigException* scex = NULL;
if ( ex->getCode() == SendStaleConfigCode ) scex = static_cast<SendStaleConfigException*>( ex.get() );
BSONObjBuilder err;
ex->getInfo().append( err );
if( scex ){
err.append( "ns", scex->getns() );
err.appendTimestamp( "vReceived", scex->getVersionReceived() );
err.appendTimestamp( "vWanted", scex->getVersionWanted() );
}
BSONObj errObj = err.done();
log() << errObj << endl;
BufBuilder b;
b.skip(sizeof(QueryResult));
b.appendBuf((void*) errObj.objdata(), errObj.objsize());
// todo: call replyToQuery() from here instead of this!!! see dbmessage.h
QueryResult * msgdata = (QueryResult *) b.buf();
b.decouple();
QueryResult *qr = msgdata;
qr->_resultFlags() = ResultFlag_ErrSet;
if( scex ) qr->_resultFlags() |= ResultFlag_ShardConfigStale;
qr->len = b.len();
qr->setOperation(opReply);
qr->cursorId = 0;
qr->startingFrom = 0;
qr->nReturned = 1;
resp.reset( new Message() );
resp->setData( msgdata, true );
}
op.debug().responseLength = resp->header()->dataLen();
dbresponse.response = resp.release();
dbresponse.responseTo = responseTo;
return ok;
}
bool run(const string& , BSONObj& cmdObj, int, string& errmsg, BSONObjBuilder& result, bool) {
if ( ! okForConfigChanges( errmsg ) )
return false;
ShardConnection::sync();
Timer t;
string ns = cmdObj.firstElement().valuestrsafe();
if ( ns.size() == 0 ) {
errmsg = "no ns";
return false;
}
DBConfigPtr config = grid.getDBConfig( ns );
if ( ! config->isSharded( ns ) ) {
config->reload();
if ( ! config->isSharded( ns ) ) {
errmsg = "ns not sharded. have to shard before we can move a chunk";
return false;
}
}
BSONObj find = cmdObj.getObjectField( "find" );
if ( find.isEmpty() ) {
errmsg = "need to specify find. see help";
return false;
}
string toString = cmdObj["to"].valuestrsafe();
if ( ! toString.size() ) {
errmsg = "you have to specify where you want to move the chunk";
return false;
}
Shard to = Shard::make( toString );
// so far, chunk size serves test purposes; it may or may not become a supported parameter
long long maxChunkSizeBytes = cmdObj["maxChunkSizeBytes"].numberLong();
if ( maxChunkSizeBytes == 0 ) {
maxChunkSizeBytes = Chunk::MaxChunkSize;
}
tlog() << "CMD: movechunk: " << cmdObj << endl;
ChunkManagerPtr info = config->getChunkManager( ns );
ChunkPtr c = info->findChunk( find );
const Shard& from = c->getShard();
if ( from == to ) {
errmsg = "that chunk is already on that shard";
return false;
}
BSONObj res;
if ( ! c->moveAndCommit( to , maxChunkSizeBytes , res ) ) {
errmsg = "move failed";
result.append( "cause" , res );
return false;
}
// preemptively reload the config to get new version info
config->getChunkManager( ns , true );
result.append( "millis" , t.millis() );
return true;
}
bool run(const string& , BSONObj& cmdObj, int, string& errmsg, BSONObjBuilder& result, bool) {
result.append("isdbgrid", 1);
result.append("hostname", getHostNameCached());
return true;
}
static BSONObj simpleKey( char c, int n = 1 ) {
BSONObjBuilder builder;
string val( n, c );
builder.append( "a", val );
return builder.obj();
}
virtual bool run(OperationContext* txn, const string& dbname, BSONObj& cmdObj, int, string& errmsg, BSONObjBuilder& result, bool fromRepl) {
if ( cmdObj.firstElement().type() != Array ) {
errmsg = "ops has to be an array";
return false;
}
BSONObj ops = cmdObj.firstElement().Obj();
{
// check input
BSONObjIterator i( ops );
while ( i.more() ) {
BSONElement e = i.next();
if (!_checkOperation(e, errmsg)) {
return false;
}
}
}
// SERVER-4328 todo : is global ok or does this take a long time? i believe multiple
// ns used so locking individually requires more analysis
ScopedTransaction scopedXact(txn, MODE_X);
Lock::GlobalWrite globalWriteLock(txn->lockState());
if (!fromRepl &&
!repl::getGlobalReplicationCoordinator()->canAcceptWritesForDatabase(dbname)) {
return appendCommandStatus(result, Status(ErrorCodes::NotMaster, str::stream()
<< "Not primary while applying ops to database " << dbname));
}
// Preconditions check reads the database state, so needs to be done locked
if ( cmdObj["preCondition"].type() == Array ) {
BSONObjIterator i( cmdObj["preCondition"].Obj() );
while ( i.more() ) {
BSONObj f = i.next().Obj();
DBDirectClient db( txn );
BSONObj realres = db.findOne( f["ns"].String() , f["q"].Obj() );
// Apply-ops would never have a $where matcher, so use the default callback,
// which will throw an error if $where is found.
Matcher m(f["res"].Obj());
if ( ! m.matches( realres ) ) {
result.append( "got" , realres );
result.append( "whatFailed" , f );
errmsg = "pre-condition failed";
return false;
}
}
}
// apply
int num = 0;
int errors = 0;
BSONObjIterator i( ops );
BSONArrayBuilder ab;
const bool alwaysUpsert = cmdObj.hasField("alwaysUpsert") ?
cmdObj["alwaysUpsert"].trueValue() : true;
while ( i.more() ) {
BSONElement e = i.next();
const BSONObj& temp = e.Obj();
// Ignore 'n' operations.
const char *opType = temp["op"].valuestrsafe();
if (*opType == 'n') continue;
const string ns = temp["ns"].String();
// Run operations under a nested lock as a hack to prevent yielding.
//
// The list of operations is supposed to be applied atomically; yielding
// would break atomicity by allowing an interruption or a shutdown to occur
// after only some operations are applied. We are already locked globally
// at this point, so taking a DBLock on the namespace creates a nested lock,
// and yields are disallowed for operations that hold a nested lock.
//
// We do not have a wrapping WriteUnitOfWork so it is possible for a journal
// commit to happen with a subset of ops applied.
// TODO figure out what to do about this.
Lock::GlobalWrite globalWriteLockDisallowTempRelease(txn->lockState());
// Ensures that yielding will not happen (see the comment above).
DEV {
Locker::LockSnapshot lockSnapshot;
invariant(!txn->lockState()->saveLockStateAndUnlock(&lockSnapshot));
};
OldClientContext ctx(txn, ns);
Status status(ErrorCodes::InternalError, "");
while (true) {
try {
// We assume that in the WriteConflict retry case, either the op rolls back
// any changes it makes or is otherwise safe to rerun.
status =
repl::applyOperation_inlock(txn, ctx.db(), temp, false, alwaysUpsert);
break;
}
catch (const WriteConflictException& wce) {
LOG(2) << "WriteConflictException in applyOps command, retrying.";
txn->recoveryUnit()->commitAndRestart();
continue;
}
}
ab.append(status.isOK());
if (!status.isOK()) {
errors++;
}
num++;
WriteUnitOfWork wuow(txn);
logOpForDbHash(txn, ns.c_str());
wuow.commit();
}
result.append( "applied" , num );
result.append( "results" , ab.arr() );
if ( ! fromRepl ) {
// We want this applied atomically on slaves
// so we re-wrap without the pre-condition for speed
string tempNS = str::stream() << dbname << ".$cmd";
// TODO: possibly use mutable BSON to remove preCondition field
// once it is available
BSONObjIterator iter(cmdObj);
BSONObjBuilder cmdBuilder;
while (iter.more()) {
BSONElement elem(iter.next());
if (strcmp(elem.fieldName(), "preCondition") != 0) {
cmdBuilder.append(elem);
}
}
const BSONObj cmdRewritten = cmdBuilder.done();
// We currently always logOp the command regardless of whether the individial ops
// succeeded and rely on any failures to also happen on secondaries. This isn't
// perfect, but it's what the command has always done and is part of its "correct"
// behavior.
while (true) {
try {
WriteUnitOfWork wunit(txn);
getGlobalEnvironment()->getOpObserver()->onApplyOps(txn,
tempNS,
cmdRewritten);
wunit.commit();
break;
}
catch (const WriteConflictException& wce) {
LOG(2) <<
"WriteConflictException while logging applyOps command, retrying.";
txn->recoveryUnit()->commitAndRestart();
continue;
}
}
}
if (errors != 0) {
return false;
}
return true;
}
bool wrappedRun(OperationContext* txn,
const string& dbname,
BSONObj& jsobj,
string& errmsg,
BSONObjBuilder& anObjBuilder) {
BSONElement e = jsobj.firstElement();
const string toDeleteNs = dbname + '.' + e.valuestr();
if (!serverGlobalParams.quiet) {
MONGO_TLOG(0) << "CMD: dropIndexes " << toDeleteNs << endl;
}
Client::Context ctx(toDeleteNs);
Database* db = ctx.db();
Collection* collection = db->getCollection( toDeleteNs );
if ( ! collection ) {
errmsg = "ns not found";
return false;
}
stopIndexBuilds(db, jsobj);
IndexCatalog* indexCatalog = collection->getIndexCatalog();
anObjBuilder.appendNumber("nIndexesWas", indexCatalog->numIndexesTotal() );
BSONElement f = jsobj.getField("index");
if ( f.type() == String ) {
string indexToDelete = f.valuestr();
if ( indexToDelete == "*" ) {
Status s = indexCatalog->dropAllIndexes(txn, false);
if ( !s.isOK() ) {
appendCommandStatus( anObjBuilder, s );
return false;
}
anObjBuilder.append("msg", "non-_id indexes dropped for collection");
return true;
}
IndexDescriptor* desc = collection->getIndexCatalog()->findIndexByName( indexToDelete );
if ( desc == NULL ) {
errmsg = str::stream() << "index not found with name [" << indexToDelete << "]";
return false;
}
if ( desc->isIdIndex() ) {
errmsg = "cannot drop _id index";
return false;
}
Status s = indexCatalog->dropIndex(txn, desc);
if ( !s.isOK() ) {
appendCommandStatus( anObjBuilder, s );
return false;
}
return true;
}
if ( f.type() == Object ) {
IndexDescriptor* desc = collection->getIndexCatalog()->findIndexByKeyPattern( f.embeddedObject() );
if ( desc == NULL ) {
errmsg = "can't find index with key:";
errmsg += f.embeddedObject().toString();
return false;
}
if ( desc->isIdIndex() ) {
errmsg = "cannot drop _id index";
return false;
}
Status s = indexCatalog->dropIndex(txn, desc);
if ( !s.isOK() ) {
appendCommandStatus( anObjBuilder, s );
return false;
}
return true;
}
errmsg = "invalid index name spec";
return false;
}
bool run(OperationContext* txn, const string& dbname , BSONObj& jsobj, int, string& errmsg, BSONObjBuilder& result, bool /*fromRepl*/) {
static DBDirectClient db;
BSONElement e = jsobj.firstElement();
string toDeleteNs = dbname + '.' + e.valuestr();
MONGO_TLOG(0) << "CMD: reIndex " << toDeleteNs << endl;
Lock::DBWrite dbXLock(dbname);
Client::Context ctx(toDeleteNs);
Collection* collection = ctx.db()->getCollection( toDeleteNs );
if ( !collection ) {
errmsg = "ns not found";
return false;
}
BackgroundOperation::assertNoBgOpInProgForNs( toDeleteNs );
std::vector<BSONObj> indexesInProg = stopIndexBuilds(ctx.db(), jsobj);
list<BSONObj> all;
auto_ptr<DBClientCursor> i = db.query( dbname + ".system.indexes" , BSON( "ns" << toDeleteNs ) , 0 , 0 , 0 , QueryOption_SlaveOk );
BSONObjBuilder b;
while ( i->more() ) {
const BSONObj spec = i->next().removeField("v").getOwned();
const BSONObj key = spec.getObjectField("key");
const Status keyStatus = validateKeyPattern(key);
if (!keyStatus.isOK()) {
errmsg = str::stream()
<< "Cannot rebuild index " << spec << ": " << keyStatus.reason()
<< " For more info see http://dochub.mongodb.org/core/index-validation";
return false;
}
b.append( BSONObjBuilder::numStr( all.size() ) , spec );
all.push_back( spec );
}
result.appendNumber( "nIndexesWas", collection->getIndexCatalog()->numIndexesTotal() );
Status s = collection->getIndexCatalog()->dropAllIndexes(txn, true);
if ( !s.isOK() ) {
errmsg = "dropIndexes failed";
return appendCommandStatus( result, s );
}
for ( list<BSONObj>::iterator i=all.begin(); i!=all.end(); i++ ) {
BSONObj o = *i;
LOG(1) << "reIndex ns: " << toDeleteNs << " index: " << o << endl;
Status s = collection->getIndexCatalog()->createIndex(txn, o, false);
if ( !s.isOK() )
return appendCommandStatus( result, s );
}
result.append( "nIndexes" , (int)all.size() );
result.appendArray( "indexes" , b.obj() );
IndexBuilder::restoreIndexes(indexesInProg);
return true;
}
UpdateResult _updateObjects( bool su,
const char* ns,
const BSONObj& updateobj,
const BSONObj& patternOrig,
bool upsert,
bool multi,
bool logop ,
OpDebug& debug,
RemoveSaver* rs,
bool fromMigrate,
const QueryPlanSelectionPolicy& planPolicy,
bool forReplication ) {
DEBUGUPDATE( "update: " << ns
<< " update: " << updateobj
<< " query: " << patternOrig
<< " upsert: " << upsert << " multi: " << multi );
Client& client = cc();
int profile = client.database()->profile;
debug.updateobj = updateobj;
// The idea with these here it to make them loop invariant for
// multi updates, and thus be a bit faster for that case. The
// pointers may be left invalid on a failed or terminal yield
// recovery.
NamespaceDetails* d = nsdetails(ns); // can be null if an upsert...
NamespaceDetailsTransient* nsdt = &NamespaceDetailsTransient::get(ns);
auto_ptr<ModSet> mods;
bool isOperatorUpdate = updateobj.firstElementFieldName()[0] == '$';
int modsIsIndexed = false; // really the # of indexes
if ( isOperatorUpdate ) {
if( d && d->indexBuildInProgress ) {
set<string> bgKeys;
d->inProgIdx().keyPattern().getFieldNames(bgKeys);
mods.reset( new ModSet(updateobj, nsdt->indexKeys(), &bgKeys, forReplication) );
}
else {
mods.reset( new ModSet(updateobj, nsdt->indexKeys(), NULL, forReplication) );
}
modsIsIndexed = mods->isIndexed();
}
if( planPolicy.permitOptimalIdPlan() && !multi && isSimpleIdQuery(patternOrig) && d &&
!modsIsIndexed ) {
int idxNo = d->findIdIndex();
if( idxNo >= 0 ) {
debug.idhack = true;
UpdateResult result = _updateById( isOperatorUpdate,
idxNo,
mods.get(),
profile,
d,
nsdt,
su,
ns,
updateobj,
patternOrig,
logop,
debug,
fromMigrate);
if ( result.existing || ! upsert ) {
return result;
}
else if ( upsert && ! isOperatorUpdate && ! logop) {
// this handles repl inserts
checkNoMods( updateobj );
debug.upsert = true;
BSONObj no = updateobj;
theDataFileMgr.insertWithObjMod(ns, no, false, su);
return UpdateResult( 0 , 0 , 1 , no );
}
}
}
int numModded = 0;
debug.nscanned = 0;
shared_ptr<Cursor> c =
NamespaceDetailsTransient::getCursor( ns, patternOrig, BSONObj(), planPolicy );
d = nsdetails(ns);
nsdt = &NamespaceDetailsTransient::get(ns);
bool autoDedup = c->autoDedup();
if( c->ok() ) {
set<DiskLoc> seenObjects;
MatchDetails details;
auto_ptr<ClientCursor> cc;
do {
if ( cc.get() == 0 &&
client.allowedToThrowPageFaultException() &&
! c->currLoc().isNull() &&
! c->currLoc().rec()->likelyInPhysicalMemory() ) {
throw PageFaultException( c->currLoc().rec() );
}
bool atomic = c->matcher() && c->matcher()->docMatcher().atomic();
if ( ! atomic && debug.nscanned > 0 ) {
// we need to use a ClientCursor to yield
if ( cc.get() == 0 ) {
shared_ptr< Cursor > cPtr = c;
cc.reset( new ClientCursor( QueryOption_NoCursorTimeout , cPtr , ns ) );
}
bool didYield;
if ( ! cc->yieldSometimes( ClientCursor::WillNeed, &didYield ) ) {
cc.release();
break;
}
if ( !c->ok() ) {
break;
}
if ( didYield ) {
d = nsdetails(ns);
if ( ! d )
break;
nsdt = &NamespaceDetailsTransient::get(ns);
if ( mods.get() && ! mods->isIndexed() ) {
// we need to re-check indexes
set<string> bgKeys;
if ( d->indexBuildInProgress )
d->inProgIdx().keyPattern().getFieldNames(bgKeys);
mods->updateIsIndexed( nsdt->indexKeys() , &bgKeys );
modsIsIndexed = mods->isIndexed();
}
}
} // end yielding block
debug.nscanned++;
if ( mods.get() && mods->hasDynamicArray() ) {
// The Cursor must have a Matcher to record an elemMatchKey. But currently
// a modifier on a dynamic array field may be applied even if there is no
// elemMatchKey, so a matcher cannot be required.
//verify( c->matcher() );
details.requestElemMatchKey();
}
if ( !c->currentMatches( &details ) ) {
c->advance();
continue;
}
Record* r = c->_current();
DiskLoc loc = c->currLoc();
if ( c->getsetdup( loc ) && autoDedup ) {
c->advance();
continue;
}
BSONObj js = BSONObj::make(r);
BSONObj pattern = patternOrig;
if ( logop ) {
BSONObjBuilder idPattern;
BSONElement id;
// NOTE: If the matching object lacks an id, we'll log
// with the original pattern. This isn't replay-safe.
// It might make sense to suppress the log instead
// if there's no id.
if ( js.getObjectID( id ) ) {
idPattern.append( id );
pattern = idPattern.obj();
}
else {
uassert( 10157 , "multi-update requires all modified objects to have an _id" , ! multi );
}
}
/* look for $inc etc. note as listed here, all fields to inc must be this type, you can't set some
regular ones at the moment. */
if ( isOperatorUpdate ) {
if ( multi ) {
// go to next record in case this one moves
c->advance();
// Update operations are deduped for cursors that implement their own
// deduplication. In particular, some geo cursors are excluded.
if ( autoDedup ) {
if ( seenObjects.count( loc ) ) {
continue;
}
// SERVER-5198 Advance past the document to be modified, provided
// deduplication is enabled, but see SERVER-5725.
while( c->ok() && loc == c->currLoc() ) {
c->advance();
}
}
}
const BSONObj& onDisk = loc.obj();
ModSet* useMods = mods.get();
auto_ptr<ModSet> mymodset;
if ( details.hasElemMatchKey() && mods->hasDynamicArray() ) {
useMods = mods->fixDynamicArray( details.elemMatchKey() );
mymodset.reset( useMods );
}
auto_ptr<ModSetState> mss = useMods->prepare( onDisk );
bool willAdvanceCursor = multi && c->ok() && ( modsIsIndexed || ! mss->canApplyInPlace() );
if ( willAdvanceCursor ) {
if ( cc.get() ) {
cc->setDoingDeletes( true );
}
c->prepareToTouchEarlierIterate();
}
if ( modsIsIndexed <= 0 && mss->canApplyInPlace() ) {
mss->applyModsInPlace( true );// const_cast<BSONObj&>(onDisk) );
DEBUGUPDATE( "\t\t\t doing in place update" );
if ( profile && !multi )
debug.fastmod = true;
if ( modsIsIndexed ) {
seenObjects.insert( loc );
}
d->paddingFits();
}
else {
if ( rs )
rs->goingToDelete( onDisk );
BSONObj newObj = mss->createNewFromMods();
checkTooLarge(newObj);
DiskLoc newLoc = theDataFileMgr.updateRecord(ns,
d,
nsdt,
r,
loc,
newObj.objdata(),
newObj.objsize(),
debug);
if ( newLoc != loc || modsIsIndexed ){
// log() << "Moved obj " << newLoc.obj()["_id"] << " from " << loc << " to " << newLoc << endl;
// object moved, need to make sure we don' get again
seenObjects.insert( newLoc );
}
}
if ( logop ) {
DEV verify( mods->size() );
BSONObj logObj = mss->getOpLogRewrite();
DEBUGUPDATE( "\t rewrite update: " << logObj );
// It is possible that the entire mod set was a no-op over this
// document. We would have an empty log record in that case. If we
// call logOp, with an empty record, that would be replicated as "clear
// this record", which is not what we want. Therefore, to get a no-op
// in the replica, we simply don't log.
if ( logObj.nFields() ) {
logOp("u", ns, logObj , &pattern, 0, fromMigrate );
}
}
numModded++;
if ( ! multi )
return UpdateResult( 1 , 1 , numModded , BSONObj() );
if ( willAdvanceCursor )
c->recoverFromTouchingEarlierIterate();
getDur().commitIfNeeded();
continue;
}
uassert( 10158 , "multi update only works with $ operators" , ! multi );
BSONElementManipulator::lookForTimestamps( updateobj );
checkNoMods( updateobj );
theDataFileMgr.updateRecord(ns, d, nsdt, r, loc , updateobj.objdata(), updateobj.objsize(), debug, su);
if ( logop ) {
DEV wassert( !su ); // super used doesn't get logged, this would be bad.
logOp("u", ns, updateobj, &pattern, 0, fromMigrate );
}
return UpdateResult( 1 , 0 , 1 , BSONObj() );
} while ( c->ok() );
} // endif
if ( numModded )
return UpdateResult( 1 , 1 , numModded , BSONObj() );
if ( upsert ) {
if ( updateobj.firstElementFieldName()[0] == '$' ) {
// upsert of an $operation. build a default object
BSONObj newObj = mods->createNewFromQuery( patternOrig );
checkNoMods( newObj );
debug.fastmodinsert = true;
theDataFileMgr.insertWithObjMod(ns, newObj, false, su);
if ( logop )
logOp( "i", ns, newObj, 0, 0, fromMigrate );
return UpdateResult( 0 , 1 , 1 , newObj );
}
uassert( 10159 , "multi update only works with $ operators" , ! multi );
checkNoMods( updateobj );
debug.upsert = true;
BSONObj no = updateobj;
theDataFileMgr.insertWithObjMod(ns, no, false, su);
if ( logop )
logOp( "i", ns, no, 0, 0, fromMigrate );
return UpdateResult( 0 , 0 , 1 , no );
}
return UpdateResult( 0 , isOperatorUpdate , 0 , BSONObj() );
}
Status MMAPV1DatabaseCatalogEntry::_renameSingleNamespace( OperationContext* txn,
const StringData& fromNS,
const StringData& toNS,
bool stayTemp ) {
// some sanity checking
NamespaceDetails* fromDetails = _namespaceIndex.details( fromNS );
if ( !fromDetails )
return Status( ErrorCodes::BadValue, "from namespace doesn't exist" );
if ( _namespaceIndex.details( toNS ) )
return Status( ErrorCodes::BadValue, "to namespace already exists" );
_removeFromCache( fromNS );
// at this point, we haven't done anything destructive yet
// ----
// actually start moving
// ----
// this could throw, but if it does we're ok
_namespaceIndex.add_ns( txn, toNS, fromDetails );
NamespaceDetails* toDetails = _namespaceIndex.details( toNS );
try {
toDetails->copyingFrom(txn,
toNS,
_namespaceIndex,
fromDetails); // fixes extraOffset
}
catch( DBException& ) {
// could end up here if .ns is full - if so try to clean up / roll back a little
_namespaceIndex.kill_ns( txn, toNS );
throw;
}
// at this point, code .ns stuff moved
_namespaceIndex.kill_ns( txn, fromNS );
fromDetails = NULL;
// fix system.namespaces
BSONObj newSpec;
DiskLoc oldSpecLocation;
{
BSONObj oldSpec;
{
RecordStoreV1Base* rs = _getNamespaceRecordStore();
scoped_ptr<RecordIterator> it( rs->getIterator(txn) );
while ( !it->isEOF() ) {
DiskLoc loc = it->getNext();
BSONObj entry = it->dataFor( loc ).toBson();
if ( fromNS == entry["name"].String() ) {
oldSpecLocation = loc;
oldSpec = entry.getOwned();
break;
}
}
}
invariant( !oldSpec.isEmpty() );
invariant( !oldSpecLocation.isNull() );
BSONObjBuilder b;
BSONObjIterator i( oldSpec.getObjectField( "options" ) );
while( i.more() ) {
BSONElement e = i.next();
if ( strcmp( e.fieldName(), "create" ) != 0 ) {
if (stayTemp || (strcmp(e.fieldName(), "temp") != 0))
b.append( e );
}
else {
b << "create" << toNS;
}
}
newSpec = b.obj();
}
_addNamespaceToNamespaceCollection( txn, toNS, newSpec.isEmpty() ? 0 : &newSpec );
_getNamespaceRecordStore()->deleteRecord( txn, oldSpecLocation );
boost::mutex::scoped_lock lk( _collectionsLock );
Entry*& entry = _collections[toNS.toString()];
invariant( entry == NULL );
entry = new Entry();
_fillInEntry_inlock( txn, toNS, entry );
return Status::OK();
}
Status MMAPV1DatabaseCatalogEntry::renameCollection( OperationContext* txn,
const StringData& fromNS,
const StringData& toNS,
bool stayTemp ) {
Status s = _renameSingleNamespace( txn, fromNS, toNS, stayTemp );
if ( !s.isOK() )
return s;
NamespaceDetails* details = _namespaceIndex.details( toNS );
invariant( details );
RecordStoreV1Base* systemIndexRecordStore = _getIndexRecordStore();
scoped_ptr<RecordIterator> it( systemIndexRecordStore->getIterator(txn) );
while ( !it->isEOF() ) {
DiskLoc loc = it->getNext();
BSONObj oldIndexSpec = it->dataFor( loc ).toBson();
if ( fromNS != oldIndexSpec["ns"].valuestrsafe() )
continue;
BSONObj newIndexSpec;
{
BSONObjBuilder b;
BSONObjIterator i( oldIndexSpec );
while( i.more() ) {
BSONElement e = i.next();
if ( strcmp( e.fieldName(), "ns" ) != 0 )
b.append( e );
else
b << "ns" << toNS;
}
newIndexSpec = b.obj();
}
StatusWith<DiskLoc> newIndexSpecLoc =
systemIndexRecordStore->insertRecord( txn,
newIndexSpec.objdata(),
newIndexSpec.objsize(),
-1 );
if ( !newIndexSpecLoc.isOK() )
return newIndexSpecLoc.getStatus();
const string& indexName = oldIndexSpec.getStringField( "name" );
{
// fix IndexDetails pointer
NamespaceDetailsCollectionCatalogEntry ce( toNS,
details,
systemIndexRecordStore,
this );
int indexI = ce._findIndexNumber( indexName );
IndexDetails& indexDetails = details->idx(indexI);
*txn->recoveryUnit()->writing(&indexDetails.info) = newIndexSpecLoc.getValue(); // XXX: dur
}
{
// move underlying namespac
string oldIndexNs = IndexDescriptor::makeIndexNamespace( fromNS, indexName );
string newIndexNs = IndexDescriptor::makeIndexNamespace( toNS, indexName );
Status s = _renameSingleNamespace( txn, oldIndexNs, newIndexNs, false );
if ( !s.isOK() )
return s;
}
systemIndexRecordStore->deleteRecord( txn, loc );
}
return Status::OK();
}
bool handleSpecialNamespaces( Request& r , QueryMessage& q ) {
const char * ns = r.getns();
ns = strstr( r.getns() , ".$cmd.sys." );
if ( ! ns )
return false;
ns += 10;
BSONObjBuilder b;
vector<Shard> shards;
if ( strcmp( ns , "inprog" ) == 0 ) {
Shard::getAllShards( shards );
BSONArrayBuilder arr( b.subarrayStart( "inprog" ) );
for ( unsigned i=0; i<shards.size(); i++ ) {
Shard shard = shards[i];
ScopedDbConnection conn( shard );
BSONObj temp = conn->findOne( r.getns() , BSONObj() );
if ( temp["inprog"].isABSONObj() ) {
BSONObjIterator i( temp["inprog"].Obj() );
while ( i.more() ) {
BSONObjBuilder x;
BSONObjIterator j( i.next().Obj() );
while( j.more() ) {
BSONElement e = j.next();
if ( str::equals( e.fieldName() , "opid" ) ) {
stringstream ss;
ss << shard.getName() << ':' << e.numberInt();
x.append( "opid" , ss.str() );
}
else if ( str::equals( e.fieldName() , "client" ) ) {
x.appendAs( e , "client_s" );
}
else {
x.append( e );
}
}
arr.append( x.obj() );
}
}
conn.done();
}
arr.done();
}
else if ( strcmp( ns , "killop" ) == 0 ) {
BSONElement e = q.query["op"];
if ( strstr( r.getns() , "admin." ) != 0 ) {
b.append( "err" , "unauthorized" );
}
else if ( e.type() != String ) {
b.append( "err" , "bad op" );
b.append( e );
}
else {
b.append( e );
string s = e.String();
string::size_type i = s.find( ':' );
if ( i == string::npos ) {
b.append( "err" , "bad opid" );
}
else {
string shard = s.substr( 0 , i );
int opid = atoi( s.substr( i + 1 ).c_str() );
b.append( "shard" , shard );
b.append( "shardid" , opid );
log() << "want to kill op: " << e << endl;
Shard s(shard);
ScopedDbConnection conn( s );
conn->findOne( r.getns() , BSON( "op" << opid ) );
conn.done();
}
}
}
else if ( strcmp( ns , "unlock" ) == 0 ) {
b.append( "err" , "can't do unlock through mongos" );
}
else {
log( LL_WARNING ) << "unknown sys command [" << ns << "]" << endl;
return false;
}
BSONObj x = b.done();
replyToQuery(0, r.p(), r.m(), x);
return true;
}
bool OpDebug::append(const CurOp& curop, BSONObjBuilder& b, size_t maxSize) const {
b.append( "op" , iscommand ? "command" : opToString( op ) );
b.append( "ns" , ns.toString() );
int queryUpdateObjSize = 0;
if (!query.isEmpty()) {
queryUpdateObjSize += query.objsize();
}
else if (!iscommand && curop.haveQuery()) {
queryUpdateObjSize += curop.query()["query"].size();
}
if (!updateobj.isEmpty()) {
queryUpdateObjSize += updateobj.objsize();
}
if (static_cast<size_t>(queryUpdateObjSize) > maxSize) {
if (!query.isEmpty()) {
// Use 60 since BSONObj::toString can truncate strings into 150 chars
// and we want to have enough room for both query and updateobj when
// the entire document is going to be serialized into a string
const string abbreviated(query.toString(false, false), 0, 60);
b.append(iscommand ? "command" : "query", abbreviated + "...");
}
else if (!iscommand && curop.haveQuery()) {
const string abbreviated(curop.query()["query"].toString(false, false), 0, 60);
b.append("query", abbreviated + "...");
}
if (!updateobj.isEmpty()) {
const string abbreviated(updateobj.toString(false, false), 0, 60);
b.append("updateobj", abbreviated + "...");
}
return false;
}
if (!query.isEmpty()) {
b.append(iscommand ? "command" : "query", query);
}
else if (!iscommand && curop.haveQuery()) {
curop.appendQuery(b, "query");
}
if (!updateobj.isEmpty()) {
b.append("updateobj", updateobj);
}
const bool moved = (nmoved >= 1);
OPDEBUG_APPEND_NUMBER( cursorid );
OPDEBUG_APPEND_NUMBER( ntoreturn );
OPDEBUG_APPEND_NUMBER( ntoskip );
OPDEBUG_APPEND_BOOL( exhaust );
OPDEBUG_APPEND_NUMBER( nscanned );
OPDEBUG_APPEND_BOOL( idhack );
OPDEBUG_APPEND_BOOL( scanAndOrder );
OPDEBUG_APPEND_BOOL( moved );
OPDEBUG_APPEND_NUMBER( nmoved );
OPDEBUG_APPEND_NUMBER( nupdated );
OPDEBUG_APPEND_NUMBER( ninserted );
OPDEBUG_APPEND_NUMBER( ndeleted );
OPDEBUG_APPEND_BOOL( fastmod );
OPDEBUG_APPEND_BOOL( fastmodinsert );
OPDEBUG_APPEND_BOOL( upsert );
OPDEBUG_APPEND_NUMBER( keyUpdates );
b.appendNumber( "numYield" , curop.numYields() );
b.append( "lockStats" , curop.lockStat().report() );
if ( ! exceptionInfo.empty() )
exceptionInfo.append( b , "exception" , "exceptionCode" );
OPDEBUG_APPEND_NUMBER( nreturned );
OPDEBUG_APPEND_NUMBER( responseLength );
b.append( "millis" , executionTime );
return true;
}
bool run(const string& , BSONObj& cmdObj, int, string& errmsg, BSONObjBuilder& result, bool) {
if ( ! okForConfigChanges( errmsg ) )
return false;
ShardConnection::sync();
string ns = cmdObj.firstElement().valuestrsafe();
if ( ns.size() == 0 ) {
errmsg = "no ns";
return false;
}
DBConfigPtr config = grid.getDBConfig( ns );
if ( ! config->isSharded( ns ) ) {
config->reload();
if ( ! config->isSharded( ns ) ) {
errmsg = "ns not sharded. have to shard before can split";
return false;
}
}
BSONObj find = cmdObj.getObjectField( "find" );
if ( find.isEmpty() ) {
find = cmdObj.getObjectField( "middle" );
if ( find.isEmpty() ) {
errmsg = "need to specify find or middle";
return false;
}
}
ChunkManagerPtr info = config->getChunkManager( ns );
ChunkPtr chunk = info->findChunk( find );
BSONObj middle = cmdObj.getObjectField( "middle" );
assert( chunk.get() );
log() << "splitting: " << ns << " shard: " << chunk << endl;
BSONObj res;
bool worked;
if ( middle.isEmpty() ) {
BSONObj ret = chunk->singleSplit( true /* force a split even if not enough data */ , res );
worked = !ret.isEmpty();
}
else {
// sanity check if the key provided is a valid split point
if ( ( middle == chunk->getMin() ) || ( middle == chunk->getMax() ) ) {
errmsg = "cannot split on initial or final chunk's key";
return false;
}
if (!fieldsMatch(middle, info->getShardKey().key())){
errmsg = "middle has different fields (or different order) than shard key";
return false;
}
vector<BSONObj> splitPoints;
splitPoints.push_back( middle );
worked = chunk->multiSplit( splitPoints , res );
}
if ( !worked ) {
errmsg = "split failed";
result.append( "cause" , res );
return false;
}
config->getChunkManager( ns , true );
return true;
}
void ReplSetImpl::_summarizeStatus(BSONObjBuilder& b) const {
vector<BSONObj> v;
const Member *_self = this->_self;
verify( _self );
MemberState myState = box.getState();
// add self
{
BSONObjBuilder bb;
bb.append("_id", (int) _self->id());
bb.append("name", _self->fullName());
bb.append("health", 1.0);
bb.append("state", (int)myState.s);
bb.append("stateStr", myState.toString());
bb.append("uptime", (unsigned)(time(0) - cmdLine.started));
if (!_self->config().arbiterOnly) {
GTID lastLive;
GTID lastUnapplied;
GTID minLive;
GTID minUnapplied;
gtidManager->getGTIDs(
&lastLive,
&lastUnapplied,
&minLive,
&minUnapplied
);
bb.appendDate("optimeDate", gtidManager->getCurrTimestamp());
bb.append("lastGTID", lastLive.toString());
bb.append("lastUnappliedGTID", lastUnapplied.toString());
bb.append("minLiveGTID", minLive.toString());
bb.append("minUnappliedGTID", minUnapplied.toString());
bb.append("oplogVersion", ReplSetConfig::OPLOG_VERSION);
}
bb.append("highestKnownPrimaryInReplSet", theReplSet->getHighestKnownPrimaryAcrossSet());
int maintenance = _maintenanceMode;
if (maintenance) {
bb.append("maintenanceMode", maintenance);
}
if (theReplSet) {
string s = theReplSet->hbmsg();
if( !s.empty() )
bb.append("errmsg", s);
}
bb.append("self", true);
v.push_back(bb.obj());
}
Member *m =_members.head();
while( m ) {
BSONObjBuilder bb;
bb.append("_id", (int) m->id());
bb.append("name", m->fullName());
double h = m->hbinfo().health;
bb.append("health", h);
bb.append("state", (int) m->state().s);
if( h == 0 ) {
// if we can't connect the state info is from the past and could be confusing to show
bb.append("stateStr", "(not reachable/healthy)");
}
else {
bb.append("stateStr", m->state().toString());
}
bb.append("uptime", (unsigned) (m->hbinfo().upSince ? (time(0)-m->hbinfo().upSince) : 0));
if (!m->config().arbiterOnly) {
bb.appendDate("optimeDate", m->hbinfo().opTime);
bb.append("lastGTID", m->hbinfo().gtid.toString());
bb.append("lastUnappliedGTID", m->hbinfo().lastUnappliedGTID.toString());
bb.append("minLiveGTID", m->hbinfo().minLiveGTID.toString());
bb.append("minUnappliedGTID", m->hbinfo().minUnappliedGTID.toString());
bb.append("oplogVersion", m->hbinfo().oplogVersion);
}
bb.append("highestKnownPrimaryInReplSet", m->hbinfo().highestKnownPrimaryInSet);
bb.appendTimeT("lastHeartbeat", m->hbinfo().lastHeartbeat);
bb.appendTimeT("lastHeartbeatRecv", m->getLastRecvHeartbeat());
bb.append("pingMs", m->hbinfo().ping);
string s = m->lhb();
if( !s.empty() )
bb.append("lastHeartbeatMessage", s);
if (m->hbinfo().authIssue) {
bb.append("authenticated", false);
}
string syncingTo = m->hbinfo().syncingTo;
if (!syncingTo.empty()) {
bb.append("syncingTo", syncingTo);
}
v.push_back(bb.obj());
m = m->next();
}
sort(v.begin(), v.end());
b.append("set", name());
b.appendTimeT("date", time(0));
b.append("myState", myState.s);
const Member *syncTarget = BackgroundSync::get()->getSyncTarget();
if ( syncTarget &&
(myState != MemberState::RS_PRIMARY) &&
(myState != MemberState::RS_SHUNNED) ) {
b.append("syncingTo", syncTarget->fullName());
}
b.append("members", v);
if( replSetBlind )
b.append("blind",true); // to avoid confusion if set...normally never set except for testing.
}
bool run(const string& , BSONObj& cmdObj, int, string& errmsg, BSONObjBuilder& result, bool) {
string target = cmdObj.firstElement().valuestrsafe();
Shard s = Shard::make( target );
if ( ! grid.knowAboutShard( s.getConnString() ) ) {
errmsg = "unknown shard";
return false;
}
ScopedDbConnection conn( configServer.getPrimary() );
// If the server is not yet draining chunks, put it in draining mode.
BSONObj searchDoc = BSON( "_id" << s.getName() );
BSONObj drainingDoc = BSON( "_id" << s.getName() << ShardFields::draining(true) );
BSONObj shardDoc = conn->findOne( "config.shards", drainingDoc );
if ( shardDoc.isEmpty() ) {
// TODO prevent move chunks to this shard.
log() << "going to start draining shard: " << s.getName() << endl;
BSONObj newStatus = BSON( "$set" << BSON( ShardFields::draining(true) ) );
conn->update( "config.shards" , searchDoc , newStatus, false /* do no upsert */);
errmsg = conn->getLastError();
if ( errmsg.size() ) {
log() << "error starting remove shard: " << s.getName() << " err: " << errmsg << endl;
return false;
}
Shard::reloadShardInfo();
result.append( "msg" , "draining started successfully" );
result.append( "state" , "started" );
result.append( "shard" , s.getName() );
conn.done();
return true;
}
// If the server has been completely drained, remove it from the ConfigDB.
// Check not only for chunks but also databases.
BSONObj shardIDDoc = BSON( "shard" << shardDoc[ "_id" ].str() );
long long chunkCount = conn->count( "config.chunks" , shardIDDoc );
BSONObj primaryDoc = BSON( "primary" << shardDoc[ "_id" ].str() );
long long dbCount = conn->count( "config.databases" , primaryDoc );
if ( ( chunkCount == 0 ) && ( dbCount == 0 ) ) {
log() << "going to remove shard: " << s.getName() << endl;
conn->remove( "config.shards" , searchDoc );
errmsg = conn->getLastError();
if ( errmsg.size() ) {
log() << "error concluding remove shard: " << s.getName() << " err: " << errmsg << endl;
return false;
}
Shard::removeShard( shardDoc[ "_id" ].str() );
Shard::reloadShardInfo();
result.append( "msg" , "removeshard completed successfully" );
result.append( "state" , "completed" );
result.append( "shard" , s.getName() );
conn.done();
return true;
}
// If the server is already in draining mode, just report on its progress.
// Report on databases (not just chunks) that are left too.
result.append( "msg" , "draining ongoing" );
result.append( "state" , "ongoing" );
BSONObjBuilder inner;
inner.append( "chunks" , chunkCount );
inner.append( "dbs" , dbCount );
result.append( "remaining" , inner.obj() );
conn.done();
return true;
}
virtual bool run(OperationContext* txn,
const string& dbname,
BSONObj& cmdObj,
int,
string& errmsg,
BSONObjBuilder& result,
bool fromRepl) {
ScopedTransaction transaction(txn, MODE_X);
Lock::GlobalWrite globalWriteLock(txn->lockState());
string source = cmdObj.getStringField( name.c_str() );
string target = cmdObj.getStringField( "to" );
if (!fromRepl &&
!repl::getGlobalReplicationCoordinator()->canAcceptWritesForDatabase(dbname)) {
return appendCommandStatus(result, Status(ErrorCodes::NotMaster, str::stream()
<< "Not primary while renaming collection " << source << " to " << target));
}
// We stay in source context the whole time. This is mostly to set the CurOp namespace.
Client::Context ctx(txn, source);
if ( !NamespaceString::validCollectionComponent(target.c_str()) ) {
errmsg = "invalid collection name: " + target;
return false;
}
if ( source.empty() || target.empty() ) {
errmsg = "invalid command syntax";
return false;
}
if ((repl::getGlobalReplicationCoordinator()->getReplicationMode() !=
repl::ReplicationCoordinator::modeNone)) {
if (NamespaceString(source).isOplog()) {
errmsg = "can't rename live oplog while replicating";
return false;
}
if (NamespaceString(target).isOplog()) {
errmsg = "can't rename to live oplog while replicating";
return false;
}
}
if (NamespaceString::oplog(source) != NamespaceString::oplog(target)) {
errmsg =
"If either the source or target of a rename is an oplog name, both must be";
return false;
}
if (!fromRepl) { // If it got through on the master, need to allow it here too
Status sourceStatus = userAllowedWriteNS(source);
if (!sourceStatus.isOK()) {
errmsg = "error with source namespace: " + sourceStatus.reason();
return false;
}
Status targetStatus = userAllowedWriteNS(target);
if (!targetStatus.isOK()) {
errmsg = "error with target namespace: " + targetStatus.reason();
return false;
}
}
if (NamespaceString(source).coll() == "system.indexes"
|| NamespaceString(target).coll() == "system.indexes") {
errmsg = "renaming system.indexes is not allowed";
return false;
}
Database* const sourceDB = dbHolder().get(txn, nsToDatabase(source));
Collection* const sourceColl = sourceDB ? sourceDB->getCollection(source)
: NULL;
if (!sourceColl) {
errmsg = "source namespace does not exist";
return false;
}
{
// Ensure that collection name does not exceed maximum length.
// Ensure that index names do not push the length over the max.
// Iterator includes unfinished indexes.
IndexCatalog::IndexIterator sourceIndIt =
sourceColl->getIndexCatalog()->getIndexIterator( txn, true );
int longestIndexNameLength = 0;
while ( sourceIndIt.more() ) {
int thisLength = sourceIndIt.next()->indexName().length();
if ( thisLength > longestIndexNameLength )
longestIndexNameLength = thisLength;
}
unsigned int longestAllowed =
min(int(NamespaceString::MaxNsCollectionLen),
int(NamespaceString::MaxNsLen) - 2/*strlen(".$")*/ - longestIndexNameLength);
if (target.size() > longestAllowed) {
StringBuilder sb;
sb << "collection name length of " << target.size()
<< " exceeds maximum length of " << longestAllowed
<< ", allowing for index names";
errmsg = sb.str();
return false;
}
}
const std::vector<BSONObj> indexesInProg = stopIndexBuilds(txn, sourceDB, cmdObj);
// Dismissed on success
ScopeGuard indexBuildRestorer = MakeGuard(IndexBuilder::restoreIndexes,
txn,
indexesInProg);
Database* const targetDB = dbHolder().openDb(txn, nsToDatabase(target));
{
WriteUnitOfWork wunit(txn);
// Check if the target namespace exists and if dropTarget is true.
// If target exists and dropTarget is not true, return false.
if (targetDB->getCollection(target)) {
if (!cmdObj["dropTarget"].trueValue()) {
errmsg = "target namespace exists";
return false;
}
Status s = targetDB->dropCollection(txn, target);
if ( !s.isOK() ) {
errmsg = s.toString();
return false;
}
}
// If we are renaming in the same database, just
// rename the namespace and we're done.
if (sourceDB == targetDB) {
Status s = targetDB->renameCollection(txn,
source,
target,
cmdObj["stayTemp"].trueValue() );
if (!s.isOK()) {
return appendCommandStatus(result, s);
}
if (!fromRepl) {
repl::logOp(txn, "c", (dbname + ".$cmd").c_str(), cmdObj);
}
wunit.commit();
indexBuildRestorer.Dismiss();
return true;
}
wunit.commit();
}
// If we get here, we are renaming across databases, so we must copy all the data and
// indexes, then remove the source collection.
// Create the target collection. It will be removed if we fail to copy the collection.
// TODO use a temp collection and unset the temp flag on success.
Collection* targetColl = NULL;
{
CollectionOptions options;
options.setNoIdIndex();
if (sourceColl->isCapped()) {
const CollectionOptions sourceOpts =
sourceColl->getCatalogEntry()->getCollectionOptions(txn);
options.capped = true;
options.cappedSize = sourceOpts.cappedSize;
options.cappedMaxDocs = sourceOpts.cappedMaxDocs;
}
WriteUnitOfWork wunit(txn);
// No logOp necessary because the entire renameCollection command is one logOp.
targetColl = targetDB->createCollection(txn, target, options);
if (!targetColl) {
errmsg = "Failed to create target collection.";
return false;
}
wunit.commit();
}
// Dismissed on success
ScopeGuard targetCollectionDropper = MakeGuard(dropCollection, txn, targetDB, target);
MultiIndexBlock indexer(txn, targetColl);
indexer.allowInterruption();
// Copy the index descriptions from the source collection, adjusting the ns field.
{
std::vector<BSONObj> indexesToCopy;
IndexCatalog::IndexIterator sourceIndIt =
sourceColl->getIndexCatalog()->getIndexIterator( txn, true );
while (sourceIndIt.more()) {
const BSONObj currIndex = sourceIndIt.next()->infoObj();
// Process the source index.
BSONObjBuilder newIndex;
newIndex.append("ns", target);
newIndex.appendElementsUnique(currIndex);
indexesToCopy.push_back(newIndex.obj());
}
indexer.init(indexesToCopy);
}
{
// Copy over all the data from source collection to target collection.
boost::scoped_ptr<RecordIterator> sourceIt(sourceColl->getIterator(txn));
while (!sourceIt->isEOF()) {
txn->checkForInterrupt();
const Snapshotted<BSONObj> obj = sourceColl->docFor(txn, sourceIt->getNext());
WriteUnitOfWork wunit(txn);
// No logOp necessary because the entire renameCollection command is one logOp.
Status status =
targetColl->insertDocument(txn, obj.value(), &indexer, true).getStatus();
if (!status.isOK())
return appendCommandStatus(result, status);
wunit.commit();
}
}
Status status = indexer.doneInserting();
if (!status.isOK())
return appendCommandStatus(result, status);
{
// Getting here means we successfully built the target copy. We now remove the
// source collection and finalize the rename.
WriteUnitOfWork wunit(txn);
Status status = sourceDB->dropCollection(txn, source);
if (!status.isOK())
return appendCommandStatus(result, status);
indexer.commit();
if (!fromRepl) {
repl::logOp(txn, "c", (dbname + ".$cmd").c_str(), cmdObj);
}
wunit.commit();
}
indexBuildRestorer.Dismiss();
targetCollectionDropper.Dismiss();
return true;
}
virtual bool run(const string& , BSONObj& cmdObj, int, string& errmsg, BSONObjBuilder& result, bool) {
result.appendBool("ismaster", true );
result.append("msg", "isdbgrid");
result.appendNumber("maxBsonObjectSize", BSONObjMaxUserSize);
return true;
}
bool ReplSetImpl::forceSyncFrom(const string& host, string& errmsg, BSONObjBuilder& result) {
lock lk(this);
// initial sanity check
if (iAmArbiterOnly()) {
errmsg = "arbiters don't sync";
return false;
}
if (box.getState().primary()) {
errmsg = "primaries don't sync";
return false;
}
if (_self != NULL && host == _self->fullName()) {
errmsg = "I cannot sync from myself";
return false;
}
// find the member we want to sync from
Member *newTarget = 0;
for (Member *m = _members.head(); m; m = m->next()) {
if (m->fullName() == host) {
newTarget = m;
break;
}
}
// do some more sanity checks
if (!newTarget) {
// this will also catch if someone tries to sync a member from itself, as _self is not
// included in the _members list.
errmsg = "could not find member in replica set";
return false;
}
if (newTarget->config().arbiterOnly) {
errmsg = "I cannot sync from an arbiter";
return false;
}
if (!newTarget->config().buildIndexes && myConfig().buildIndexes) {
errmsg = "I cannot sync from a member who does not build indexes";
return false;
}
if (newTarget->hbinfo().authIssue) {
errmsg = "I cannot authenticate against the requested member";
return false;
}
if (newTarget->hbinfo().health == 0) {
errmsg = "I cannot reach the requested member";
return false;
}
if (newTarget->hbinfo().opTime.getSecs()+10 < lastOpTimeWritten.getSecs()) {
log() << "attempting to sync from " << newTarget->fullName()
<< ", but its latest opTime is " << newTarget->hbinfo().opTime.getSecs()
<< " and ours is " << lastOpTimeWritten.getSecs() << " so this may not work"
<< rsLog;
result.append("warning", "requested member is more than 10 seconds behind us");
// not returning false, just warning
}
// record the previous member we were syncing from
const Member *prev = replset::BackgroundSync::get()->getSyncTarget();
if (prev) {
result.append("prevSyncTarget", prev->fullName());
}
// finally, set the new target
_forceSyncTarget = newTarget;
return true;
}
virtual bool run(const string& dbname, BSONObj& cmdObj, int, string& errmsg, BSONObjBuilder& result, bool fromRepl) {
string source = cmdObj.getStringField( name.c_str() );
string target = cmdObj.getStringField( "to" );
if ( !NamespaceString::validCollectionComponent(target.c_str()) ) {
errmsg = "invalid collection name: " + target;
return false;
}
if ( source.empty() || target.empty() ) {
errmsg = "invalid command syntax";
return false;
}
if (!fromRepl) { // If it got through on the master, need to allow it here too
Status sourceStatus = userAllowedWriteNS(source);
if (!sourceStatus.isOK()) {
errmsg = "error with source namespace: " + sourceStatus.reason();
return false;
}
Status targetStatus = userAllowedWriteNS(target);
if (!targetStatus.isOK()) {
errmsg = "error with target namespace: " + targetStatus.reason();
return false;
}
}
string sourceDB = nsToDatabase(source);
string targetDB = nsToDatabase(target);
bool capped = false;
long long size = 0;
std::vector<BSONObj> indexesInProg;
{
Client::Context srcCtx( source );
Collection* sourceColl = srcCtx.db()->getCollection( source );
if ( !sourceColl ) {
errmsg = "source namespace does not exist";
return false;
}
// Ensure that collection name does not exceed maximum length.
// Ensure that index names do not push the length over the max.
// Iterator includes unfinished indexes.
IndexCatalog::IndexIterator sourceIndIt =
sourceColl->getIndexCatalog()->getIndexIterator( true );
int longestIndexNameLength = 0;
while ( sourceIndIt.more() ) {
int thisLength = sourceIndIt.next()->indexName().length();
if ( thisLength > longestIndexNameLength )
longestIndexNameLength = thisLength;
}
unsigned int longestAllowed =
min(int(Namespace::MaxNsColletionLen),
int(Namespace::MaxNsLen) - 2/*strlen(".$")*/ - longestIndexNameLength);
if (target.size() > longestAllowed) {
StringBuilder sb;
sb << "collection name length of " << target.size()
<< " exceeds maximum length of " << longestAllowed
<< ", allowing for index names";
errmsg = sb.str();
return false;
}
{
const NamespaceDetails *nsd = nsdetails( source );
indexesInProg = stopIndexBuilds( dbname, cmdObj );
capped = nsd->isCapped();
if ( capped )
for( DiskLoc i = nsd->firstExtent(); !i.isNull(); i = i.ext()->xnext )
size += i.ext()->length;
}
}
{
Client::Context ctx( target );
// Check if the target namespace exists and if dropTarget is true.
// If target exists and dropTarget is not true, return false.
if ( ctx.db()->getCollection( target ) ) {
if ( !cmdObj["dropTarget"].trueValue() ) {
errmsg = "target namespace exists";
return false;
}
Status s = cc().database()->dropCollection( target );
if ( !s.isOK() ) {
errmsg = s.toString();
restoreIndexBuildsOnSource( indexesInProg, source );
return false;
}
}
// If we are renaming in the same database, just
// rename the namespace and we're done.
if ( sourceDB == targetDB ) {
Status s = ctx.db()->renameCollection( source, target,
cmdObj["stayTemp"].trueValue() );
if ( !s.isOK() ) {
errmsg = s.toString();
restoreIndexBuildsOnSource( indexesInProg, source );
return false;
}
return true;
}
// Otherwise, we are enaming across databases, so we must copy all
// the data and then remove the source collection.
// Create the target collection.
Collection* targetColl = NULL;
if ( capped ) {
BSONObjBuilder spec;
spec.appendBool( "capped", true );
spec.append( "size", double( size ) );
spec.appendBool( "autoIndexId", false );
userCreateNS( target.c_str(), spec.obj(), errmsg, false );
targetColl = ctx.db()->getCollection( target );
}
else {
CollectionOptions options;
options.setNoIdIndex();
// No logOp necessary because the entire renameCollection command is one logOp.
targetColl = ctx.db()->createCollection( target, options );
}
if ( !targetColl ) {
errmsg = "Failed to create target collection.";
restoreIndexBuildsOnSource( indexesInProg, source );
return false;
}
}
// Copy over all the data from source collection to target collection.
bool insertSuccessful = true;
boost::scoped_ptr<CollectionIterator> sourceIt;
{
Client::Context srcCtx( source );
Collection* sourceColl = srcCtx.db()->getCollection( source );
sourceIt.reset( sourceColl->getIterator( DiskLoc(), false, CollectionScanParams::FORWARD ) );
}
Collection* targetColl = NULL;
while ( !sourceIt->isEOF() ) {
BSONObj o;
{
Client::Context srcCtx( source );
o = sourceIt->getNext().obj();
}
// Insert and check return status of insert.
{
Client::Context ctx( target );
if ( !targetColl )
targetColl = ctx.db()->getCollection( target );
// No logOp necessary because the entire renameCollection command is one logOp.
Status s = targetColl->insertDocument( o, true ).getStatus();
if ( !s.isOK() ) {
insertSuccessful = false;
errmsg = s.toString();
break;
}
}
}
// If inserts were unsuccessful, drop the target collection and return false.
if ( !insertSuccessful ) {
Client::Context ctx( target );
Status s = ctx.db()->dropCollection( target );
if ( !s.isOK() )
errmsg = s.toString();
restoreIndexBuildsOnSource( indexesInProg, source );
return false;
}
// Copy over the indexes to temp storage and then to the target..
vector<BSONObj> copiedIndexes;
bool indexSuccessful = true;
{
Client::Context srcCtx( source );
Collection* sourceColl = srcCtx.db()->getCollection( source );
IndexCatalog::IndexIterator sourceIndIt =
sourceColl->getIndexCatalog()->getIndexIterator( true );
while ( sourceIndIt.more() ) {
BSONObj currIndex = sourceIndIt.next()->infoObj();
// Process the source index.
BSONObjBuilder b;
BSONObjIterator i( currIndex );
while( i.moreWithEOO() ) {
BSONElement e = i.next();
if ( e.eoo() )
break;
else if ( strcmp( e.fieldName(), "ns" ) == 0 )
b.append( "ns", target );
else
b.append( e );
}
BSONObj newIndex = b.obj();
copiedIndexes.push_back( newIndex );
}
}
{
Client::Context ctx( target );
if ( !targetColl )
targetColl = ctx.db()->getCollection( target );
for ( vector<BSONObj>::iterator it = copiedIndexes.begin();
it != copiedIndexes.end(); ++it ) {
Status s = targetColl->getIndexCatalog()->createIndex( *it, true );
if ( !s.isOK() ) {
indexSuccessful = false;
errmsg = s.toString();
break;
}
}
// If indexes were unsuccessful, drop the target collection and return false.
if ( !indexSuccessful ) {
Status s = ctx.db()->dropCollection( target );
if ( !s.isOK() )
errmsg = s.toString();
restoreIndexBuildsOnSource( indexesInProg, source );
return false;
}
}
// Drop the source collection.
{
Client::Context srcCtx( source );
Status s = srcCtx.db()->dropCollection( source );
if ( !s.isOK() ) {
errmsg = s.toString();
restoreIndexBuildsOnSource( indexesInProg, source );
return false;
}
}
return true;
}
bool ClientInfo::getLastError( const BSONObj& options , BSONObjBuilder& result , bool fromWriteBackListener ) {
set<string> * shards = getPrev();
if ( shards->size() == 0 ) {
result.appendNull( "err" );
return true;
}
vector<WBInfo> writebacks;
// handle single server
if ( shards->size() == 1 ) {
string theShard = *(shards->begin() );
ShardConnection conn( theShard , "" );
BSONObj res;
bool ok = false;
try {
ok = conn->runCommand( "admin" , options , res );
}
catch( std::exception &e ) {
warning() << "could not get last error from shard " << theShard << causedBy( e ) << endl;
// Catch everything that happens here, since we need to ensure we return our connection when we're
// finished.
conn.done();
return false;
}
res = res.getOwned();
conn.done();
_addWriteBack( writebacks , res );
// hit other machines just to block
for ( set<string>::const_iterator i=sinceLastGetError().begin(); i!=sinceLastGetError().end(); ++i ) {
string temp = *i;
if ( temp == theShard )
continue;
ShardConnection conn( temp , "" );
try {
_addWriteBack( writebacks , conn->getLastErrorDetailed() );
}
catch( std::exception &e ) {
warning() << "could not clear last error from shard " << temp << causedBy( e ) << endl;
}
conn.done();
}
clearSinceLastGetError();
if ( writebacks.size() ) {
vector<BSONObj> v = _handleWriteBacks( writebacks , fromWriteBackListener );
if ( v.size() == 0 && fromWriteBackListener ) {
// ok
}
else {
// this will usually be 1
// it can be greater than 1 if a write to a different shard
// than the last write op had a writeback
// all we're going to report is the first
// since that's the current write
// but we block for all
assert( v.size() >= 1 );
result.appendElements( v[0] );
result.appendElementsUnique( res );
result.append( "writebackGLE" , v[0] );
result.append( "initialGLEHost" , theShard );
}
}
else {
result.append( "singleShard" , theShard );
result.appendElements( res );
}
return ok;
}
BSONArrayBuilder bbb( result.subarrayStart( "shards" ) );
BSONObjBuilder shardRawGLE;
long long n = 0;
int updatedExistingStat = 0; // 0 is none, -1 has but false, 1 has true
// hit each shard
vector<string> errors;
vector<BSONObj> errorObjects;
for ( set<string>::iterator i = shards->begin(); i != shards->end(); i++ ) {
string theShard = *i;
bbb.append( theShard );
ShardConnection conn( theShard , "" );
BSONObj res;
bool ok = false;
try {
ok = conn->runCommand( "admin" , options , res );
shardRawGLE.append( theShard , res );
}
catch( std::exception &e ) {
// Safe to return here, since we haven't started any extra processing yet, just collecting
// responses.
warning() << "could not get last error from a shard " << theShard << causedBy( e ) << endl;
conn.done();
return false;
}
_addWriteBack( writebacks, res );
string temp = DBClientWithCommands::getLastErrorString( res );
if ( conn->type() != ConnectionString::SYNC && ( ok == false || temp.size() ) ) {
errors.push_back( temp );
errorObjects.push_back( res );
}
n += res["n"].numberLong();
if ( res["updatedExisting"].type() ) {
if ( res["updatedExisting"].trueValue() )
updatedExistingStat = 1;
else if ( updatedExistingStat == 0 )
updatedExistingStat = -1;
}
conn.done();
}
bbb.done();
result.append( "shardRawGLE" , shardRawGLE.obj() );
result.appendNumber( "n" , n );
if ( updatedExistingStat )
result.appendBool( "updatedExisting" , updatedExistingStat > 0 );
// hit other machines just to block
for ( set<string>::const_iterator i=sinceLastGetError().begin(); i!=sinceLastGetError().end(); ++i ) {
string temp = *i;
if ( shards->count( temp ) )
continue;
ShardConnection conn( temp , "" );
try {
_addWriteBack( writebacks, conn->getLastErrorDetailed() );
}
catch( std::exception &e ) {
warning() << "could not clear last error from a shard " << temp << causedBy( e ) << endl;
}
conn.done();
}
clearSinceLastGetError();
if ( errors.size() == 0 ) {
result.appendNull( "err" );
_handleWriteBacks( writebacks , fromWriteBackListener );
return true;
}
result.append( "err" , errors[0].c_str() );
{
// errs
BSONArrayBuilder all( result.subarrayStart( "errs" ) );
for ( unsigned i=0; i<errors.size(); i++ ) {
all.append( errors[i].c_str() );
}
all.done();
}
{
// errObjects
BSONArrayBuilder all( result.subarrayStart( "errObjects" ) );
for ( unsigned i=0; i<errorObjects.size(); i++ ) {
all.append( errorObjects[i] );
}
all.done();
}
_handleWriteBacks( writebacks , fromWriteBackListener );
return true;
}
void Model::save( bool safe ){
ScopedDbConnection conn( modelServer() );
BSONObjBuilder b;
serialize( b );
BSONElement myId;
{
BSONObjIterator i = b.iterator();
while ( i.more() ){
BSONElement e = i.next();
if ( strcmp( e.fieldName() , "_id" ) == 0 ){
myId = e;
break;
}
}
}
if ( myId.type() ){
if ( _id.isEmpty() ){
_id = myId.wrap();
}
else if ( myId.woCompare( _id.firstElement() ) ){
stringstream ss;
ss << "_id from serialize and stored differ: ";
ss << "[" << myId << "] != ";
ss << "[" << _id.firstElement() << "]";
throw UserException( 13121 , ss.str() );
}
}
if ( _id.isEmpty() ){
OID oid;
oid.init();
b.appendOID( "_id" , &oid );
BSONObj o = b.obj();
conn->insert( getNS() , o );
_id = o["_id"].wrap().getOwned();
log(4) << "inserted new model " << getNS() << " " << o << endl;
}
else {
if ( myId.eoo() ){
myId = _id["_id"];
b.append( myId );
}
assert( ! myId.eoo() );
BSONObjBuilder qb;
qb.append( myId );
BSONObj q = qb.obj();
BSONObj o = b.obj();
log(4) << "updated model" << getNS() << " " << q << " " << o << endl;
conn->update( getNS() , q , o , true );
}
string errmsg = "";
if ( safe )
errmsg = conn->getLastError();
conn.done();
if ( safe && errmsg.size() )
throw UserException( 9003 , (string)"error on Model::save: " + errmsg );
}
Status MMAPV1DatabaseCatalogEntry::renameCollection(OperationContext* txn,
StringData fromNS,
StringData toNS,
bool stayTemp) {
Status s = _renameSingleNamespace(txn, fromNS, toNS, stayTemp);
if (!s.isOK())
return s;
NamespaceDetails* details = _namespaceIndex.details(toNS);
invariant(details);
RecordStoreV1Base* systemIndexRecordStore = _getIndexRecordStore();
auto cursor = systemIndexRecordStore->getCursor(txn);
while (auto record = cursor->next()) {
BSONObj oldIndexSpec = record->data.releaseToBson();
if (fromNS != oldIndexSpec["ns"].valuestrsafe())
continue;
BSONObj newIndexSpec;
{
BSONObjBuilder b;
BSONObjIterator i(oldIndexSpec);
while (i.more()) {
BSONElement e = i.next();
if (strcmp(e.fieldName(), "ns") != 0)
b.append(e);
else
b << "ns" << toNS;
}
newIndexSpec = b.obj();
}
StatusWith<RecordId> newIndexSpecLoc = systemIndexRecordStore->insertRecord(
txn, newIndexSpec.objdata(), newIndexSpec.objsize(), false);
if (!newIndexSpecLoc.isOK())
return newIndexSpecLoc.getStatus();
const std::string& indexName = oldIndexSpec.getStringField("name");
{
// Fix the IndexDetails pointer.
int indexI = getCollectionCatalogEntry(toNS)->_findIndexNumber(txn, indexName);
IndexDetails& indexDetails = details->idx(indexI);
*txn->recoveryUnit()->writing(&indexDetails.info) =
DiskLoc::fromRecordId(newIndexSpecLoc.getValue());
}
{
// Move the underlying namespace.
std::string oldIndexNs = IndexDescriptor::makeIndexNamespace(fromNS, indexName);
std::string newIndexNs = IndexDescriptor::makeIndexNamespace(toNS, indexName);
Status s = _renameSingleNamespace(txn, oldIndexNs, newIndexNs, false);
if (!s.isOK())
return s;
}
systemIndexRecordStore->deleteRecord(txn, record->id);
}
return Status::OK();
}
void ParallelSortClusteredCursor::_finishCons() {
_numServers = _servers.size();
_lastFrom = 0;
_cursors = 0;
if (!_qSpec.isEmpty()) {
_needToSkip = _qSpec.ntoskip();
_cursors = 0;
_sortKey = _qSpec.sort();
_fields = _qSpec.fields();
}
// Partition sort key fields into (a) text meta fields and (b) all other fields.
set<string> textMetaSortKeyFields;
set<string> normalSortKeyFields;
// Transform _sortKey fields {a:{$meta:"textScore"}} into {a:-1}, in order to apply the
// merge sort for text metadata in the correct direction.
BSONObjBuilder transformedSortKeyBuilder;
BSONObjIterator sortKeyIt(_sortKey);
while (sortKeyIt.more()) {
BSONElement e = sortKeyIt.next();
if (LiteParsedQuery::isTextScoreMeta(e)) {
textMetaSortKeyFields.insert(e.fieldName());
transformedSortKeyBuilder.append(e.fieldName(), -1);
} else {
normalSortKeyFields.insert(e.fieldName());
transformedSortKeyBuilder.append(e);
}
}
_sortKey = transformedSortKeyBuilder.obj();
// Verify that that all text metadata sort fields are in the projection. For all other sort
// fields, copy them into the projection if they are missing (and if projection is
// negative).
if (!_sortKey.isEmpty() && !_fields.isEmpty()) {
BSONObjBuilder b;
bool isNegative = false;
{
BSONObjIterator i(_fields);
while (i.more()) {
BSONElement e = i.next();
b.append(e);
string fieldName = e.fieldName();
if (LiteParsedQuery::isTextScoreMeta(e)) {
textMetaSortKeyFields.erase(fieldName);
} else {
// exact field
bool found = normalSortKeyFields.erase(fieldName);
// subfields
set<string>::const_iterator begin =
normalSortKeyFields.lower_bound(fieldName + ".\x00");
set<string>::const_iterator end =
normalSortKeyFields.lower_bound(fieldName + ".\xFF");
normalSortKeyFields.erase(begin, end);
if (!e.trueValue()) {
uassert(13431,
"have to have sort key in projection and removing it",
!found && begin == end);
} else if (!e.isABSONObj()) {
isNegative = true;
}
}
}
}
if (isNegative) {
for (set<string>::const_iterator it(normalSortKeyFields.begin()),
end(normalSortKeyFields.end());
it != end;
++it) {
b.append(*it, 1);
}
}
_fields = b.obj();
}
if (!_qSpec.isEmpty()) {
_qSpec.setFields(_fields);
}
uassert(
17306, "have to have all text meta sort keys in projection", textMetaSortKeyFields.empty());
}
Status MMAPV1DatabaseCatalogEntry::_renameSingleNamespace(OperationContext* txn,
StringData fromNS,
StringData toNS,
bool stayTemp) {
// some sanity checking
NamespaceDetails* fromDetails = _namespaceIndex.details(fromNS);
if (!fromDetails)
return Status(ErrorCodes::BadValue, "from namespace doesn't exist");
if (_namespaceIndex.details(toNS))
return Status(ErrorCodes::BadValue, "to namespace already exists");
// at this point, we haven't done anything destructive yet
// ----
// actually start moving
// ----
// this could throw, but if it does we're ok
_namespaceIndex.add_ns(txn, toNS, fromDetails);
NamespaceDetails* toDetails = _namespaceIndex.details(toNS);
try {
toDetails->copyingFrom(txn, toNS, _namespaceIndex, fromDetails); // fixes extraOffset
} catch (DBException&) {
// could end up here if .ns is full - if so try to clean up / roll back a little
_namespaceIndex.kill_ns(txn, toNS);
throw;
}
// at this point, code .ns stuff moved
_namespaceIndex.kill_ns(txn, fromNS);
fromDetails = NULL;
// fix system.namespaces
BSONObj newSpec;
RecordId oldSpecLocation = getCollectionCatalogEntry(fromNS)->getNamespacesRecordId();
invariant(!oldSpecLocation.isNull());
{
BSONObj oldSpec = _getNamespaceRecordStore()->dataFor(txn, oldSpecLocation).releaseToBson();
invariant(!oldSpec.isEmpty());
BSONObjBuilder b;
BSONObjIterator i(oldSpec.getObjectField("options"));
while (i.more()) {
BSONElement e = i.next();
if (strcmp(e.fieldName(), "create") != 0) {
if (stayTemp || (strcmp(e.fieldName(), "temp") != 0))
b.append(e);
} else {
b << "create" << toNS;
}
}
newSpec = b.obj();
}
RecordId rid = _addNamespaceToNamespaceCollection(txn, toNS, newSpec.isEmpty() ? 0 : &newSpec);
_getNamespaceRecordStore()->deleteRecord(txn, oldSpecLocation);
Entry*& entry = _collections[toNS.toString()];
invariant(entry == NULL);
txn->recoveryUnit()->registerChange(new EntryInsertion(toNS, this));
entry = new Entry();
_removeFromCache(txn->recoveryUnit(), fromNS);
_insertInCache(txn, toNS, rid, entry);
return Status::OK();
}
bool run(const string& dbname, BSONObj& cmdObj, int, string& errmsg, BSONObjBuilder& result, bool fromRepl ) {
int s = 0;
bool found = false;
// TODO: convert to ServerParameters -- SERVER-10515
if( cmdObj.hasElement("journalCommitInterval") ) {
if (isMongos()) {
errmsg = "cannot set journalCommitInterval on a mongos";
return false;
}
if(!isJournalingEnabled()) {
errmsg = "journaling is off";
return false;
}
int x = (int) cmdObj["journalCommitInterval"].Number();
verify( x > 1 && x < 500 );
setJournalCommitInterval(x);
log() << "setParameter journalCommitInterval=" << x << endl;
s++;
}
if( cmdObj.hasElement( "traceExceptions" ) ) {
if( s == 0 ) result.append( "was", DBException::traceExceptions );
DBException::traceExceptions = cmdObj["traceExceptions"].Bool();
s++;
}
if( cmdObj.hasElement( "replMonitorMaxFailedChecks" ) ) {
if( s == 0 ) result.append( "was", ReplicaSetMonitor::maxConsecutiveFailedChecks );
ReplicaSetMonitor::maxConsecutiveFailedChecks =
cmdObj["replMonitorMaxFailedChecks"].numberInt();
s++;
}
const ServerParameter::Map& m = ServerParameterSet::getGlobal()->getMap();
BSONObjIterator i( cmdObj );
i.next(); // skip past command name
while ( i.more() ) {
BSONElement e = i.next();
ServerParameter::Map::const_iterator j = m.find( e.fieldName() );
if ( j == m.end() )
continue;
if ( ! j->second->allowedToChangeAtRuntime() ) {
errmsg = str::stream()
<< "not allowed to change ["
<< e.fieldName()
<< "] at runtime";
return false;
}
if ( s == 0 )
j->second->append( result, "was" );
Status status = j->second->set( e );
if ( status.isOK() ) {
s++;
continue;
}
errmsg = status.reason();
result.append( "code", status.code() );
return false;
}
if( s == 0 && !found ) {
errmsg = "no option found to set, use help:true to see options ";
return false;
}
return true;
}
void WiredTigerSizeStorer::syncCache(bool syncToDisk) {
boost::mutex::scoped_lock cursorLock( _cursorMutex );
_checkMagic();
Map myMap;
{
boost::mutex::scoped_lock lk( _entriesMutex );
for ( Map::iterator it = _entries.begin(); it != _entries.end(); ++it ) {
std::string uriKey = it->first;
Entry& entry = it->second;
if ( entry.rs ) {
if ( entry.dataSize != entry.rs->dataSize( NULL ) ) {
entry.dataSize = entry.rs->dataSize( NULL );
entry.dirty = true;
}
if ( entry.numRecords != entry.rs->numRecords( NULL ) ) {
entry.numRecords = entry.rs->numRecords( NULL );
entry.dirty = true;
}
}
if ( !entry.dirty )
continue;
myMap[uriKey] = entry;
}
}
if (myMap.empty())
return; // Nothing to do.
WT_SESSION* session = _session.getSession();
invariantWTOK(session->begin_transaction(session, syncToDisk ? "sync=true" : ""));
ScopeGuard rollbacker = MakeGuard(session->rollback_transaction, session, "");
for ( Map::iterator it = myMap.begin(); it != myMap.end(); ++it ) {
string uriKey = it->first;
Entry& entry = it->second;
BSONObj data;
{
BSONObjBuilder b;
b.append( "numRecords", entry.numRecords );
b.append( "dataSize", entry.dataSize );
data = b.obj();
}
LOG(2) << "WiredTigerSizeStorer::storeInto " << uriKey << " -> " << data;
WiredTigerItem key( uriKey.c_str(), uriKey.size() );
WiredTigerItem value( data.objdata(), data.objsize() );
_cursor->set_key( _cursor, key.Get() );
_cursor->set_value( _cursor, value.Get() );
invariantWTOK( _cursor->insert(_cursor) );
}
invariantWTOK(_cursor->reset(_cursor));
rollbacker.Dismiss();
invariantWTOK(session->commit_transaction(session, NULL));
{
boost::mutex::scoped_lock lk( _entriesMutex );
for (Map::iterator it = _entries.begin(); it != _entries.end(); ++it) {
it->second.dirty = false;
}
}
}
