void _applyOpToDataFiles( const string& database, FileOp &fo, bool afterAllocator, const string& path ) {
if ( afterAllocator )
FileAllocator::get()->waitUntilFinished();
string c = database;
c += '.';
boost::filesystem::path p(path);
if (storageGlobalParams.directoryperdb)
p /= database;
boost::filesystem::path q;
q = p / (c+"ns");
bool ok = false;
MONGO_ASSERT_ON_EXCEPTION( ok = fo.apply( q ) );
if ( ok ) {
LOG(2) << fo.op() << " file " << q.string() << endl;
}
int i = 0;
int extra = 10; // should not be necessary, this is defensive in case there are missing files
while ( 1 ) {
verify( i <= DiskLoc::MaxFiles );
stringstream ss;
ss << c << i;
q = p / ss.str();
MONGO_ASSERT_ON_EXCEPTION( ok = fo.apply(q) );
if ( ok ) {
if ( extra != 10 ) {
LOG(1) << fo.op() << " file " << q.string() << endl;
log() << " _applyOpToDataFiles() warning: extra == " << extra << endl;
}
}
else if ( --extra <= 0 )
break;
i++;
}
}
// generate a directory name for storing temp data files
Path uniqueReservedPath( const char *prefix ) {
Path repairPath = Path(storageGlobalParams.repairpath);
Path reservedPath;
int i = 0;
bool exists = false;
do {
stringstream ss;
ss << prefix << "_repairDatabase_" << i++;
reservedPath = repairPath / ss.str();
MONGO_ASSERT_ON_EXCEPTION( exists = boost::filesystem::exists( reservedPath ) );
}
while ( exists );
return reservedPath;
}
~RepairFileDeleter() {
if ( _success )
return;
log() << "cleaning up failed repair "
<< "db: " << _dbName << " path: " << _pathString;
try {
_txn->recoveryUnit()->syncDataAndTruncateJournal();
globalStorageEngine->flushAllFiles(true); // need both in case journaling is disabled
MONGO_ASSERT_ON_EXCEPTION( boost::filesystem::remove_all( _path ) );
}
catch ( DBException& e ) {
error() << "RepairFileDeleter failed to cleanup: " << e;
error() << "aborting";
fassertFailed( 17402 );
}
}
~RepairFileDeleter() {
if ( _success )
return;
log() << "cleaning up failed repair "
<< "db: " << _dbName << " path: " << _pathString;
try {
getDur().syncDataAndTruncateJournal();
MongoFile::flushAll(true); // need both in case journaling is disabled
{
Client::Context tempContext( _dbName, _pathString );
Database::closeDatabase( _dbName, _pathString );
}
MONGO_ASSERT_ON_EXCEPTION( boost::filesystem::remove_all( _path ) );
}
catch ( DBException& e ) {
error() << "RepairFileDeleter failed to cleanup: " << e;
error() << "aborting";
fassertFailed( 17402 );
}
}
void FileAllocator::run( FileAllocator * fa ) {
setThreadName( "FileAllocator" );
while( 1 ) {
{
scoped_lock lk( fa->_pendingMutex );
if ( fa->_pending.size() == 0 )
fa->_pendingUpdated.wait( lk.boost() );
}
while( 1 ) {
string name;
long size;
{
scoped_lock lk( fa->_pendingMutex );
if ( fa->_pending.size() == 0 )
break;
name = fa->_pending.front();
size = fa->_pendingSize[ name ];
}
string tmp;
long fd = 0;
try {
log() << "allocating new datafile " << name << ", filling with zeroes..." << endl;
boost::filesystem::path parent = ensureParentDirCreated(name);
tmp = makeTempFileName( parent );
ensureParentDirCreated(tmp);
fd = open(tmp.c_str(), O_CREAT | O_RDWR | O_NOATIME, S_IRUSR | S_IWUSR);
if ( fd <= 0 ) {
log() << "FileAllocator: couldn't create " << name << " (" << tmp << ") " << errnoWithDescription() << endl;
uasserted(10439, "");
}
#if defined(POSIX_FADV_DONTNEED)
if( posix_fadvise(fd, 0, size, POSIX_FADV_DONTNEED) ) {
log() << "warning: posix_fadvise fails " << name << " (" << tmp << ") " << errnoWithDescription() << endl;
}
#endif
Timer t;
/* make sure the file is the full desired length */
ensureLength( fd , size );
close( fd );
fd = 0;
if( rename(tmp.c_str(), name.c_str()) ) {
log() << "error: couldn't rename " << tmp << " to " << name << ' ' << errnoWithDescription() << endl;
uasserted(13653, "");
}
flushMyDirectory(name);
log() << "done allocating datafile " << name << ", "
<< "size: " << size/1024/1024 << "MB, "
<< " took " << ((double)t.millis())/1000.0 << " secs"
<< endl;
// no longer in a failed state. allow new writers.
fa->_failed = false;
}
catch ( ... ) {
if ( fd > 0 )
close( fd );
log() << "error failed to allocate new file: " << name
<< " size: " << size << ' ' << errnoWithDescription() << warnings;
log() << " will try again in 10 seconds" << endl; // not going to warning logs
try {
if ( tmp.size() )
MONGO_ASSERT_ON_EXCEPTION( boost::filesystem::remove( tmp ) );
MONGO_ASSERT_ON_EXCEPTION( boost::filesystem::remove( name ) );
}
catch ( ... ) {
}
scoped_lock lk( fa->_pendingMutex );
fa->_failed = true;
// not erasing from pending
fa->_pendingUpdated.notify_all();
sleepsecs(10);
continue;
}
{
scoped_lock lk( fa->_pendingMutex );
fa->_pendingSize.erase( name );
fa->_pending.pop_front();
fa->_pendingUpdated.notify_all();
}
}
}
}
Status MMAPV1Engine::repairDatabase( OperationContext* txn,
const std::string& dbName,
bool preserveClonedFilesOnFailure,
bool backupOriginalFiles ) {
// We must hold some form of lock here
invariant(txn->lockState()->threadState());
invariant( dbName.find( '.' ) == string::npos );
scoped_ptr<RepairFileDeleter> repairFileDeleter;
log() << "repairDatabase " << dbName << endl;
BackgroundOperation::assertNoBgOpInProgForDb(dbName);
txn->recoveryUnit()->syncDataAndTruncateJournal(); // Must be done before and after repair
intmax_t totalSize = dbSize( dbName );
intmax_t freeSize = File::freeSpace(storageGlobalParams.repairpath);
if ( freeSize > -1 && freeSize < totalSize ) {
return Status( ErrorCodes::OutOfDiskSpace,
str::stream() << "Cannot repair database " << dbName
<< " having size: " << totalSize
<< " (bytes) because free disk space is: " << freeSize << " (bytes)" );
}
txn->checkForInterrupt();
Path reservedPath =
uniqueReservedPath( ( preserveClonedFilesOnFailure || backupOriginalFiles ) ?
"backup" : "_tmp" );
MONGO_ASSERT_ON_EXCEPTION( boost::filesystem::create_directory( reservedPath ) );
string reservedPathString = reservedPath.string();
if ( !preserveClonedFilesOnFailure )
repairFileDeleter.reset( new RepairFileDeleter( txn,
dbName,
reservedPathString,
reservedPath ) );
{
Database* originalDatabase =
dbHolder().get(txn, dbName);
if (originalDatabase == NULL) {
return Status(ErrorCodes::NamespaceNotFound, "database does not exist to repair");
}
scoped_ptr<MMAPV1DatabaseCatalogEntry> dbEntry;
scoped_ptr<Database> tempDatabase;
{
dbEntry.reset( new MMAPV1DatabaseCatalogEntry( txn,
dbName,
reservedPathString,
storageGlobalParams.directoryperdb,
true ) );
invariant( !dbEntry->exists() );
tempDatabase.reset( new Database( txn,
dbName,
dbEntry.get() ) );
}
map<string,CollectionOptions> namespacesToCopy;
{
string ns = dbName + ".system.namespaces";
Client::Context ctx(txn, ns );
Collection* coll = originalDatabase->getCollection( txn, ns );
if ( coll ) {
scoped_ptr<RecordIterator> it( coll->getIterator( txn,
DiskLoc(),
false,
CollectionScanParams::FORWARD ) );
while ( !it->isEOF() ) {
DiskLoc loc = it->getNext();
BSONObj obj = coll->docFor( loc );
string ns = obj["name"].String();
NamespaceString nss( ns );
if ( nss.isSystem() ) {
if ( nss.isSystemDotIndexes() )
continue;
if ( nss.coll() == "system.namespaces" )
continue;
}
if ( !nss.isNormal() )
continue;
CollectionOptions options;
if ( obj["options"].isABSONObj() ) {
Status status = options.parse( obj["options"].Obj() );
if ( !status.isOK() )
return status;
}
namespacesToCopy[ns] = options;
}
}
}
for ( map<string,CollectionOptions>::const_iterator i = namespacesToCopy.begin();
i != namespacesToCopy.end();
++i ) {
string ns = i->first;
CollectionOptions options = i->second;
Collection* tempCollection = NULL;
{
Client::Context tempContext(txn, ns, tempDatabase );
WriteUnitOfWork wunit(txn);
tempCollection = tempDatabase->createCollection(txn, ns, options, true, false);
wunit.commit();
}
Client::Context readContext(txn, ns, originalDatabase);
Collection* originalCollection = originalDatabase->getCollection( txn, ns );
invariant( originalCollection );
// data
// TODO SERVER-14812 add a mode that drops duplicates rather than failing
MultiIndexBlock indexer(txn, tempCollection );
{
vector<BSONObj> indexes;
IndexCatalog::IndexIterator ii =
originalCollection->getIndexCatalog()->getIndexIterator( false );
while ( ii.more() ) {
IndexDescriptor* desc = ii.next();
indexes.push_back( desc->infoObj() );
}
Client::Context tempContext(txn, ns, tempDatabase);
Status status = indexer.init( indexes );
if ( !status.isOK() )
return status;
}
scoped_ptr<RecordIterator> iterator(
originalCollection->getIterator( txn, DiskLoc(), false,
CollectionScanParams::FORWARD ));
while ( !iterator->isEOF() ) {
DiskLoc loc = iterator->getNext();
invariant( !loc.isNull() );
BSONObj doc = originalCollection->docFor( loc );
Client::Context tempContext(txn, ns, tempDatabase);
WriteUnitOfWork wunit(txn);
StatusWith<DiskLoc> result = tempCollection->insertDocument(txn,
doc,
&indexer,
false);
if ( !result.isOK() )
return result.getStatus();
wunit.commit();
txn->checkForInterrupt(false);
}
Status status = indexer.doneInserting();
if (!status.isOK())
return status;
{
Client::Context tempContext(txn, ns, tempDatabase);
WriteUnitOfWork wunit(txn);
indexer.commit();
wunit.commit();
}
}
txn->recoveryUnit()->syncDataAndTruncateJournal();
globalStorageEngine->flushAllFiles(true); // need both in case journaling is disabled
txn->checkForInterrupt(false);
}
// at this point if we abort, we don't want to delete new files
// as they might be the only copies
if ( repairFileDeleter.get() )
repairFileDeleter->success();
dbHolder().close( txn, dbName );
if ( backupOriginalFiles ) {
_renameForBackup( dbName, reservedPath );
}
else {
// first make new directory before deleting data
Path newDir = Path(storageGlobalParams.dbpath) / dbName;
MONGO_ASSERT_ON_EXCEPTION(boost::filesystem::create_directory(newDir));
// this deletes old files
_deleteDataFiles( dbName );
if ( !boost::filesystem::exists(newDir) ) {
// we deleted because of directoryperdb
// re-create
MONGO_ASSERT_ON_EXCEPTION(boost::filesystem::create_directory(newDir));
}
}
_replaceWithRecovered( dbName, reservedPathString.c_str() );
if ( !backupOriginalFiles )
MONGO_ASSERT_ON_EXCEPTION( boost::filesystem::remove_all( reservedPath ) );
return Status::OK();
}
Status repairDatabase( string dbName,
bool preserveClonedFilesOnFailure,
bool backupOriginalFiles ) {
scoped_ptr<RepairFileDeleter> repairFileDeleter;
doingRepair dr;
dbName = nsToDatabase( dbName );
log() << "repairDatabase " << dbName << endl;
invariant( cc().database()->name() == dbName );
invariant( cc().database()->path() == storageGlobalParams.dbpath );
BackgroundOperation::assertNoBgOpInProgForDb(dbName);
getDur().syncDataAndTruncateJournal(); // Must be done before and after repair
intmax_t totalSize = dbSize( dbName );
intmax_t freeSize = File::freeSpace(storageGlobalParams.repairpath);
if ( freeSize > -1 && freeSize < totalSize ) {
return Status( ErrorCodes::OutOfDiskSpace,
str::stream() << "Cannot repair database " << dbName
<< " having size: " << totalSize
<< " (bytes) because free disk space is: " << freeSize << " (bytes)" );
}
killCurrentOp.checkForInterrupt();
Path reservedPath =
uniqueReservedPath( ( preserveClonedFilesOnFailure || backupOriginalFiles ) ?
"backup" : "_tmp" );
MONGO_ASSERT_ON_EXCEPTION( boost::filesystem::create_directory( reservedPath ) );
string reservedPathString = reservedPath.string();
if ( !preserveClonedFilesOnFailure )
repairFileDeleter.reset( new RepairFileDeleter( dbName,
reservedPathString,
reservedPath ) );
{
Database* originalDatabase = dbHolder().get( dbName, storageGlobalParams.dbpath );
if ( originalDatabase == NULL )
return Status( ErrorCodes::NamespaceNotFound, "database does not exist to repair" );
Database* tempDatabase = NULL;
{
bool justCreated = false;
tempDatabase = dbHolderW().getOrCreate( dbName, reservedPathString, justCreated );
invariant( justCreated );
}
map<string,CollectionOptions> namespacesToCopy;
{
string ns = dbName + ".system.namespaces";
Client::Context ctx( ns );
Collection* coll = originalDatabase->getCollection( ns );
if ( coll ) {
scoped_ptr<CollectionIterator> it( coll->getIterator( DiskLoc(),
false,
CollectionScanParams::FORWARD ) );
while ( !it->isEOF() ) {
DiskLoc loc = it->getNext();
BSONObj obj = coll->docFor( loc );
string ns = obj["name"].String();
NamespaceString nss( ns );
if ( nss.isSystem() ) {
if ( nss.isSystemDotIndexes() )
continue;
if ( nss.coll() == "system.namespaces" )
continue;
}
if ( !nss.isNormal() )
continue;
CollectionOptions options;
if ( obj["options"].isABSONObj() ) {
Status status = options.parse( obj["options"].Obj() );
if ( !status.isOK() )
return status;
}
namespacesToCopy[ns] = options;
}
}
}
for ( map<string,CollectionOptions>::const_iterator i = namespacesToCopy.begin();
i != namespacesToCopy.end();
++i ) {
string ns = i->first;
CollectionOptions options = i->second;
Collection* tempCollection = NULL;
{
Client::Context tempContext( ns, tempDatabase );
tempCollection = tempDatabase->createCollection( ns, options, true, false );
}
Client::Context readContext( ns, originalDatabase );
Collection* originalCollection = originalDatabase->getCollection( ns );
invariant( originalCollection );
// data
MultiIndexBlock indexBlock( tempCollection );
{
vector<BSONObj> indexes;
IndexCatalog::IndexIterator ii =
originalCollection->getIndexCatalog()->getIndexIterator( false );
while ( ii.more() ) {
IndexDescriptor* desc = ii.next();
indexes.push_back( desc->infoObj() );
}
Client::Context tempContext( ns, tempDatabase );
Status status = indexBlock.init( indexes );
if ( !status.isOK() )
return status;
}
scoped_ptr<CollectionIterator> iterator( originalCollection->getIterator( DiskLoc(),
false,
CollectionScanParams::FORWARD ) );
while ( !iterator->isEOF() ) {
DiskLoc loc = iterator->getNext();
invariant( !loc.isNull() );
BSONObj doc = originalCollection->docFor( loc );
Client::Context tempContext( ns, tempDatabase );
StatusWith<DiskLoc> result = tempCollection->insertDocument( doc, indexBlock );
if ( !result.isOK() )
return result.getStatus();
getDur().commitIfNeeded();
killCurrentOp.checkForInterrupt(false);
}
{
Client::Context tempContext( ns, tempDatabase );
Status status = indexBlock.commit();
if ( !status.isOK() )
return status;
}
}
getDur().syncDataAndTruncateJournal();
MongoFile::flushAll(true); // need both in case journaling is disabled
killCurrentOp.checkForInterrupt(false);
Client::Context tempContext( dbName, reservedPathString );
Database::closeDatabase( dbName, reservedPathString );
}
// at this point if we abort, we don't want to delete new files
// as they might be the only copies
if ( repairFileDeleter.get() )
repairFileDeleter->success();
Client::Context ctx( dbName );
Database::closeDatabase(dbName, storageGlobalParams.dbpath);
if ( backupOriginalFiles ) {
_renameForBackup( dbName, reservedPath );
}
else {
// first make new directory before deleting data
Path newDir = Path(storageGlobalParams.dbpath) / dbName;
MONGO_ASSERT_ON_EXCEPTION(boost::filesystem::create_directory(newDir));
// this deletes old files
_deleteDataFiles( dbName );
if ( !boost::filesystem::exists(newDir) ) {
// we deleted because of directoryperdb
// re-create
MONGO_ASSERT_ON_EXCEPTION(boost::filesystem::create_directory(newDir));
}
}
_replaceWithRecovered( dbName, reservedPathString.c_str() );
if ( !backupOriginalFiles )
MONGO_ASSERT_ON_EXCEPTION( boost::filesystem::remove_all( reservedPath ) );
return Status::OK();
}
