void Balancer::_doBalanceRound( DBClientBase& conn, vector<CandidateChunkPtr>* candidateChunks ) {
verify( candidateChunks );
//
// 1. Check whether there is any sharded collection to be balanced by querying
// the ShardsNS::collections collection
//
auto_ptr<DBClientCursor> cursor = conn.query( ShardNS::collection , BSONObj() );
vector< string > collections;
while ( cursor->more() ) {
BSONObj col = cursor->nextSafe();
// sharded collections will have a shard "key".
if ( ! col["key"].eoo() && ! col["noBalance"].trueValue() ){
collections.push_back( col["_id"].String() );
}
else if( col["noBalance"].trueValue() ){
LOG(1) << "not balancing collection " << col["_id"].String() << ", explicitly disabled" << endl;
}
}
cursor.reset();
if ( collections.empty() ) {
LOG(1) << "no collections to balance" << endl;
return;
}
//
// 2. Get a list of all the shards that are participating in this balance round
// along with any maximum allowed quotas and current utilization. We get the
// latter by issuing db.serverStatus() (mem.mapped) to all shards.
//
// TODO: skip unresponsive shards and mark information as stale.
//
vector<Shard> allShards;
Shard::getAllShards( allShards );
if ( allShards.size() < 2) {
LOG(1) << "can't balance without more active shards" << endl;
return;
}
ShardInfoMap shardInfo;
for ( vector<Shard>::const_iterator it = allShards.begin(); it != allShards.end(); ++it ) {
const Shard& s = *it;
ShardStatus status = s.getStatus();
shardInfo[ s.getName() ] = ShardInfo( s.getMaxSize(),
status.mapped(),
s.isDraining(),
status.hasOpsQueued(),
s.tags()
);
}
//
// 3. For each collection, check if the balancing policy recommends moving anything around.
//
for (vector<string>::const_iterator it = collections.begin(); it != collections.end(); ++it ) {
const string& ns = *it;
map< string,vector<BSONObj> > shardToChunksMap;
cursor = conn.query( ShardNS::chunk , QUERY( "ns" << ns ).sort( "min" ) );
while ( cursor->more() ) {
BSONObj chunk = cursor->nextSafe();
if ( chunk["jumbo"].trueValue() )
continue;
vector<BSONObj>& chunks = shardToChunksMap[chunk["shard"].String()];
chunks.push_back( chunk.getOwned() );
}
cursor.reset();
if (shardToChunksMap.empty()) {
LOG(1) << "skipping empty collection (" << ns << ")";
continue;
}
for ( vector<Shard>::iterator i=allShards.begin(); i!=allShards.end(); ++i ) {
// this just makes sure there is an entry in shardToChunksMap for every shard
Shard s = *i;
shardToChunksMap[s.getName()].size();
}
DistributionStatus status( shardInfo, shardToChunksMap );
// load tags
conn.ensureIndex( ShardNS::tags, BSON( "ns" << 1 << "min" << 1 ), true );
cursor = conn.query( ShardNS::tags , QUERY( "ns" << ns ).sort( "min" ) );
while ( cursor->more() ) {
BSONObj tag = cursor->nextSafe();
uassert( 16356 , str::stream() << "tag ranges not valid for: " << ns ,
status.addTagRange( TagRange( tag["min"].Obj().getOwned(),
tag["max"].Obj().getOwned(),
tag["tag"].String() ) ) );
}
cursor.reset();
CandidateChunk* p = _policy->balance( ns, status, _balancedLastTime );
if ( p ) candidateChunks->push_back( CandidateChunkPtr( p ) );
}
}
void go( const string db , const boost::filesystem::path outdir ) {
log() << "DATABASE: " << db << "\t to \t" << outdir.string() << endl;
boost::filesystem::create_directories( outdir );
map <string, BSONObj> collectionOptions;
multimap <string, BSONObj> indexes;
vector <string> collections;
// Save indexes for database
string ins = db + ".system.indexes";
auto_ptr<DBClientCursor> cursor = conn( true ).query( ins.c_str() , Query() , 0 , 0 , 0 , QueryOption_SlaveOk | QueryOption_NoCursorTimeout );
while ( cursor->more() ) {
BSONObj obj = cursor->nextSafe();
const string name = obj.getField( "ns" ).valuestr();
indexes.insert( pair<string, BSONObj> (name, obj.getOwned()) );
}
string sns = db + ".system.namespaces";
cursor = conn( true ).query( sns.c_str() , Query() , 0 , 0 , 0 , QueryOption_SlaveOk | QueryOption_NoCursorTimeout );
while ( cursor->more() ) {
BSONObj obj = cursor->nextSafe();
const string name = obj.getField( "name" ).valuestr();
if (obj.hasField("options")) {
collectionOptions.insert( pair<string,BSONObj> (name, obj.getField("options").embeddedObject()) );
}
// skip namespaces with $ in them only if we don't specify a collection to dump
if ( _coll == "" && name.find( ".$" ) != string::npos ) {
log(1) << "\tskipping collection: " << name << endl;
continue;
}
const string filename = name.substr( db.size() + 1 );
//if a particular collections is specified, and it's not this one, skip it
if ( _coll != "" && db + "." + _coll != name && _coll != name )
continue;
// raise error before writing collection with non-permitted filename chars in the name
size_t hasBadChars = name.find_first_of("/\0");
if (hasBadChars != string::npos){
error() << "Cannot dump " << name << ". Collection has '/' or null in the collection name." << endl;
continue;
}
// Don't dump indexes
if ( endsWith(name.c_str(), ".system.indexes") ) {
continue;
}
if ( _coll != "" && db + "." + _coll != name && _coll != name )
continue;
collections.push_back(name);
}
for (vector<string>::iterator it = collections.begin(); it != collections.end(); ++it) {
string name = *it;
const string filename = name.substr( db.size() + 1 );
writeCollectionFile( name , outdir / ( filename + ".bson" ) );
writeMetadataFile( name, outdir / (filename + ".metadata.json"), collectionOptions, indexes);
}
}
void WriteErrorDetail::setErrInfo(const BSONObj& errInfo) {
_errInfo = errInfo.getOwned();
_isErrInfoSet = true;
}
PlanStage::StageState DeleteStage::work(WorkingSetID* out) {
++_commonStats.works;
// Adds the amount of time taken by work() to executionTimeMillis.
ScopedTimer timer(&_commonStats.executionTimeMillis);
if (isEOF()) {
return PlanStage::IS_EOF;
}
invariant(_collection); // If isEOF() returns false, we must have a collection.
// It is possible that after a delete was executed, a WriteConflictException occurred
// and prevented us from returning ADVANCED with the old version of the document.
if (_idReturning != WorkingSet::INVALID_ID) {
// We should only get here if we were trying to return something before.
invariant(_params.returnDeleted);
WorkingSetMember* member = _ws->get(_idReturning);
invariant(member->getState() == WorkingSetMember::OWNED_OBJ);
*out = _idReturning;
_idReturning = WorkingSet::INVALID_ID;
++_commonStats.advanced;
return PlanStage::ADVANCED;
}
// Either retry the last WSM we worked on or get a new one from our child.
WorkingSetID id;
if (_idRetrying != WorkingSet::INVALID_ID) {
id = _idRetrying;
_idRetrying = WorkingSet::INVALID_ID;
} else {
auto status = child()->work(&id);
switch (status) {
case PlanStage::ADVANCED:
break;
case PlanStage::FAILURE:
case PlanStage::DEAD:
*out = id;
// If a stage fails, it may create a status WSM to indicate why it failed, in which
// case 'id' is valid. If ID is invalid, we create our own error message.
if (WorkingSet::INVALID_ID == id) {
const std::string errmsg = "delete stage failed to read in results from child";
*out = WorkingSetCommon::allocateStatusMember(
_ws, Status(ErrorCodes::InternalError, errmsg));
}
return status;
case PlanStage::NEED_TIME:
++_commonStats.needTime;
return status;
case PlanStage::NEED_YIELD:
*out = id;
++_commonStats.needYield;
return status;
case PlanStage::IS_EOF:
return status;
default:
MONGO_UNREACHABLE;
}
}
// We advanced, or are retrying, and id is set to the WSM to work on.
WorkingSetMember* member = _ws->get(id);
// We want to free this member when we return, unless we need to retry it.
ScopeGuard memberFreer = MakeGuard(&WorkingSet::free, _ws, id);
if (!member->hasLoc()) {
// We expect to be here because of an invalidation causing a force-fetch.
// When we're doing a findAndModify with a sort, the sort will have a limit of 1, so will
// not produce any more results even if there is another matching document. Throw a WCE here
// so that these operations get another chance to find a matching document. The
// findAndModify command should automatically retry if it gets a WCE.
// TODO: this is not necessary if there was no sort specified.
if (_params.returnDeleted) {
throw WriteConflictException();
}
++_specificStats.nInvalidateSkips;
++_commonStats.needTime;
return PlanStage::NEED_TIME;
}
RecordId rloc = member->loc;
// Deletes can't have projections. This means that covering analysis will always add
// a fetch. We should always get fetched data, and never just key data.
invariant(member->hasObj());
try {
// If the snapshot changed, then we have to make sure we have the latest copy of the
// doc and that it still matches.
std::unique_ptr<SeekableRecordCursor> cursor;
if (getOpCtx()->recoveryUnit()->getSnapshotId() != member->obj.snapshotId()) {
cursor = _collection->getCursor(getOpCtx());
if (!WorkingSetCommon::fetch(getOpCtx(), _ws, id, cursor)) {
// Doc is already deleted. Nothing more to do.
++_commonStats.needTime;
return PlanStage::NEED_TIME;
}
// Make sure the re-fetched doc still matches the predicate.
if (_params.canonicalQuery &&
!_params.canonicalQuery->root()->matchesBSON(member->obj.value(), NULL)) {
// Doesn't match.
++_commonStats.needTime;
return PlanStage::NEED_TIME;
}
}
// Ensure that the BSONObj underlying the WorkingSetMember is owned because saveState()
// is allowed to free the memory.
if (_params.returnDeleted) {
// Save a copy of the document that is about to get deleted, but keep it in the
// LOC_AND_OBJ state in case we need to retry deleting it.
BSONObj deletedDoc = member->obj.value();
member->obj.setValue(deletedDoc.getOwned());
}
// TODO: Do we want to buffer docs and delete them in a group rather than
// saving/restoring state repeatedly?
try {
if (supportsDocLocking()) {
// Doc-locking engines require this before saveState() since they don't use
// invalidations.
WorkingSetCommon::prepareForSnapshotChange(_ws);
}
child()->saveState();
} catch (const WriteConflictException& wce) {
std::terminate();
}
// Do the write, unless this is an explain.
if (!_params.isExplain) {
WriteUnitOfWork wunit(getOpCtx());
_collection->deleteDocument(getOpCtx(), rloc);
wunit.commit();
}
++_specificStats.docsDeleted;
} catch (const WriteConflictException& wce) {
// When we're doing a findAndModify with a sort, the sort will have a limit of 1, so will
// not produce any more results even if there is another matching document. Re-throw the WCE
// here so that these operations get another chance to find a matching document. The
// findAndModify command should automatically retry if it gets a WCE.
// TODO: this is not necessary if there was no sort specified.
if (_params.returnDeleted) {
throw;
}
_idRetrying = id;
memberFreer.Dismiss(); // Keep this member around so we can retry deleting it.
*out = WorkingSet::INVALID_ID;
_commonStats.needYield++;
return NEED_YIELD;
}
if (_params.returnDeleted) {
// After deleting the document, the RecordId associated with this member is invalid.
// Remove the 'loc' from the WorkingSetMember before returning it.
member->loc = RecordId();
member->transitionToOwnedObj();
}
// As restoreState may restore (recreate) cursors, cursors are tied to the
// transaction in which they are created, and a WriteUnitOfWork is a
// transaction, make sure to restore the state outside of the WritUnitOfWork.
try {
child()->restoreState();
} catch (const WriteConflictException& wce) {
// Note we don't need to retry anything in this case since the delete already
// was committed. However, we still need to return the deleted document
// (if it was requested).
if (_params.returnDeleted) {
// member->obj should refer to the deleted document.
invariant(member->getState() == WorkingSetMember::OWNED_OBJ);
_idReturning = id;
// Keep this member around so that we can return it on the next work() call.
memberFreer.Dismiss();
}
*out = WorkingSet::INVALID_ID;
_commonStats.needYield++;
return NEED_YIELD;
}
if (_params.returnDeleted) {
// member->obj should refer to the deleted document.
invariant(member->getState() == WorkingSetMember::OWNED_OBJ);
memberFreer.Dismiss(); // Keep this member around so we can return it.
*out = id;
++_commonStats.advanced;
return PlanStage::ADVANCED;
}
++_commonStats.needTime;
return PlanStage::NEED_TIME;
}
CollectionMetadata* CollectionMetadata::cloneSplit( const ChunkType& chunk,
const vector<BSONObj>& splitKeys,
const ChunkVersion& newShardVersion,
string* errMsg ) const {
// The error message string is optional.
string dummy;
if (errMsg == NULL) {
errMsg = &dummy;
}
// The version required in both resulting chunks could be simply an increment in the
// minor portion of the current version. However, we are enforcing uniqueness over the
// attributes <ns, version> of the configdb collection 'chunks'. So in practice, a
// migrate somewhere may force this split to pick up a version that has the major
// portion higher than the one that this shard has been using.
//
// TODO drop the uniqueness constraint and tighten the check below so that only the
// minor portion of version changes
if (newShardVersion <= _shardVersion) {
*errMsg = stream() << "cannot split chunk "
<< rangeToString( chunk.getMin(), chunk.getMax() )
<< ", new shard version "
<< newShardVersion.toString()
<< " is not greater than current version "
<< _shardVersion.toString();
warning() << *errMsg << endl;
return NULL;
}
// Check that we have the exact chunk that will be subtracted.
if ( !rangeMapContains( _chunksMap, chunk.getMin(), chunk.getMax() ) ) {
*errMsg = stream() << "cannot split chunk "
<< rangeToString( chunk.getMin(), chunk.getMax() )
<< ", this shard does not contain the chunk";
if ( rangeMapOverlaps( _chunksMap, chunk.getMin(), chunk.getMax() ) ) {
RangeVector overlap;
getRangeMapOverlap( _chunksMap, chunk.getMin(), chunk.getMax(), &overlap );
*errMsg += stream() << " and it overlaps " << overlapToString( overlap );
}
warning() << *errMsg << endl;
return NULL;
}
// Check that the split key is valid
for ( vector<BSONObj>::const_iterator it = splitKeys.begin(); it != splitKeys.end(); ++it )
{
if (!rangeContains(chunk.getMin(), chunk.getMax(), *it)) {
*errMsg = stream() << "cannot split chunk "
<< rangeToString( chunk.getMin(), chunk.getMax() ) << " at key "
<< *it;
warning() << *errMsg << endl;
return NULL;
}
}
auto_ptr<CollectionMetadata> metadata(new CollectionMetadata);
metadata->_keyPattern = this->_keyPattern;
metadata->_keyPattern.getOwned();
metadata->fillKeyPatternFields();
metadata->_pendingMap = this->_pendingMap;
metadata->_chunksMap = this->_chunksMap;
metadata->_shardVersion = newShardVersion; // will increment 2nd, 3rd,... chunks below
BSONObj startKey = chunk.getMin();
for ( vector<BSONObj>::const_iterator it = splitKeys.begin(); it != splitKeys.end();
++it ) {
BSONObj split = *it;
metadata->_chunksMap[chunk.getMin()] = split.getOwned();
metadata->_chunksMap.insert( make_pair( split.getOwned(), chunk.getMax().getOwned() ) );
metadata->_shardVersion.incMinor();
startKey = split;
}
metadata->_collVersion =
metadata->_shardVersion > _collVersion ? metadata->_shardVersion : _collVersion;
metadata->fillRanges();
dassert(metadata->isValid());
return metadata.release();
}
/**
* Run a query -- includes checking for and running a Command.
* @return points to ns if exhaust mode. 0=normal mode
* @locks the db mutex for reading (and potentially for writing temporarily to create a new db).
* @asserts on scan and order memory exhaustion and other cases.
*/
string runQuery(Message& m, QueryMessage& q, CurOp& curop, Message &result) {
shared_ptr<ParsedQuery> pq_shared( new ParsedQuery(q) );
ParsedQuery& pq( *pq_shared );
BSONObj jsobj = q.query;
int queryOptions = q.queryOptions;
const char *ns = q.ns;
uassert( 16332 , "can't have an empty ns" , ns[0] );
if( logLevel >= 2 )
log() << "runQuery called " << ns << " " << jsobj << endl;
curop.debug().ns = ns;
curop.debug().ntoreturn = pq.getNumToReturn();
curop.debug().query = jsobj;
curop.setQuery(jsobj);
uassert( 16256, str::stream() << "Invalid ns [" << ns << "]", NamespaceString::isValid(ns) );
// Run a command.
if ( pq.couldBeCommand() ) {
curop.markCommand();
BufBuilder bb;
bb.skip(sizeof(QueryResult));
BSONObjBuilder cmdResBuf;
if ( runCommands(ns, jsobj, curop, bb, cmdResBuf, false, queryOptions) ) {
curop.debug().iscommand = true;
curop.debug().query = jsobj;
auto_ptr< QueryResult > qr;
qr.reset( (QueryResult *) bb.buf() );
bb.decouple();
qr->setResultFlagsToOk();
qr->len = bb.len();
curop.debug().responseLength = bb.len();
qr->setOperation(opReply);
qr->cursorId = 0;
qr->startingFrom = 0;
qr->nReturned = 1;
result.setData( qr.release(), true );
}
else {
uasserted(13530, "bad or malformed command request?");
}
return "";
}
const bool explain = pq.isExplain();
const bool tailable = pq.hasOption(QueryOption_CursorTailable);
BSONObj order = pq.getOrder();
BSONObj query = pq.getFilter();
/* The ElemIter will not be happy if this isn't really an object. So throw exception
here when that is true.
(Which may indicate bad data from client.)
*/
if ( query.objsize() == 0 ) {
out() << "Bad query object?\n jsobj:";
out() << jsobj.toString() << "\n query:";
out() << query.toString() << endl;
uassert( 10110 , "bad query object", false);
}
// Tailable cursors need to read newly written entries from the tail
// of the collection. They manually arbitrate with the collection over
// what data is readable and when, so we choose read uncommited isolation.
OpSettings settings;
settings.setQueryCursorMode(DEFAULT_LOCK_CURSOR);
settings.setBulkFetch(true);
settings.setCappedAppendPK(pq.hasOption(QueryOption_AddHiddenPK));
cc().setOpSettings(settings);
// If our caller has a transaction, it's multi-statement.
const bool inMultiStatementTxn = cc().hasTxn();
if (tailable) {
// Because it's easier to disable this. It shouldn't be happening in a normal system.
uassert(16812, "May not perform a tailable query in a multi-statement transaction.",
!inMultiStatementTxn);
}
// Begin a read-only, snapshot transaction under normal circumstances.
// If the cursor is tailable, we need to be able to read uncommitted data.
const int txnFlags = (tailable ? DB_READ_UNCOMMITTED : DB_TXN_SNAPSHOT) | DB_TXN_READ_ONLY;
LOCK_REASON(lockReason, "query");
Client::ReadContext ctx(ns, lockReason);
scoped_ptr<Client::Transaction> transaction(!inMultiStatementTxn ?
new Client::Transaction(txnFlags) : NULL);
bool hasRetried = false;
while ( 1 ) {
try {
replVerifyReadsOk(&pq);
// Fast-path for primary key queries.
if (!explain && !tailable) {
replVerifyReadsOk(&pq);
if (_tryQueryByPKHack(ns, query, pq, curop, result)) {
if (transaction) {
transaction->commit();
}
return "";
}
}
// sanity check the query and projection
if (pq.getFields() != NULL) {
pq.getFields()->validateQuery( query );
}
if (tailable) {
Collection *cl = getCollection( ns );
if (cl != NULL && !(cl->isCapped() || str::equals(ns, rsoplog))) {
uasserted( 13051, "tailable cursor requested on non-capped, non-oplog collection" );
}
const BSONObj nat1 = BSON( "$natural" << 1 );
if ( order.isEmpty() ) {
order = nat1;
} else {
uassert( 13052, "only {$natural:1} order allowed for tailable cursor", order == nat1 );
}
}
// Run a regular query.
// these now may stored in a ClientCursor or somewhere else,
// so make sure we use a real copy
jsobj = jsobj.getOwned();
query = query.getOwned();
order = order.getOwned();
const ConfigVersion shardingVersionAtStart = shardingState.getVersion( ns );
const bool getCachedExplainPlan = ! hasRetried && explain && ! pq.hasIndexSpecifier();
const bool savedCursor = queryWithQueryOptimizer( queryOptions, ns, jsobj, curop, query,
order, pq_shared, shardingVersionAtStart,
getCachedExplainPlan, inMultiStatementTxn,
result );
// Did not save the cursor, so we can commit the transaction now if it exists.
if (transaction && !savedCursor) {
transaction->commit();
}
return curop.debug().exhaust ? ns : "";
}
catch ( const QueryRetryException & ) {
// In some cases the query may be retried if there is an in memory sort size assertion.
verify( ! hasRetried );
hasRetried = true;
}
}
}
bool runParsed(OperationContext* txn,
const NamespaceString& origNss,
const AggregationRequest& request,
BSONObj& cmdObj,
string& errmsg,
BSONObjBuilder& result) {
// For operations on views, this will be the underlying namespace.
const NamespaceString& nss = request.getNamespaceString();
// Set up the ExpressionContext.
intrusive_ptr<ExpressionContext> expCtx = new ExpressionContext(txn, request);
expCtx->tempDir = storageGlobalParams.dbpath + "/_tmp";
// Parse the pipeline.
auto statusWithPipeline = Pipeline::parse(request.getPipeline(), expCtx);
if (!statusWithPipeline.isOK()) {
return appendCommandStatus(result, statusWithPipeline.getStatus());
}
auto pipeline = std::move(statusWithPipeline.getValue());
auto resolvedNamespaces = resolveInvolvedNamespaces(txn, pipeline, expCtx);
if (!resolvedNamespaces.isOK()) {
return appendCommandStatus(result, resolvedNamespaces.getStatus());
}
expCtx->resolvedNamespaces = std::move(resolvedNamespaces.getValue());
unique_ptr<ClientCursorPin> pin; // either this OR the exec will be non-null
unique_ptr<PlanExecutor> exec;
auto curOp = CurOp::get(txn);
{
// This will throw if the sharding version for this connection is out of date. If the
// namespace is a view, the lock will be released before re-running the aggregation.
// Otherwise, the lock must be held continuously from now until we have we created both
// the output ClientCursor and the input executor. This ensures that both are using the
// same sharding version that we synchronize on here. This is also why we always need to
// create a ClientCursor even when we aren't outputting to a cursor. See the comment on
// ShardFilterStage for more details.
AutoGetCollectionOrViewForRead ctx(txn, nss);
Collection* collection = ctx.getCollection();
// If running $collStats on a view, we do not resolve the view since we want stats
// on this view namespace.
auto startsWithCollStats = [&pipeline]() {
const Pipeline::SourceContainer& sources = pipeline->getSources();
return !sources.empty() &&
dynamic_cast<DocumentSourceCollStats*>(sources.front().get());
};
// If this is a view, resolve it by finding the underlying collection and stitching view
// pipelines and this request's pipeline together. We then release our locks before
// recursively calling run, which will re-acquire locks on the underlying collection.
// (The lock must be released because recursively acquiring locks on the database will
// prohibit yielding.)
auto view = ctx.getView();
if (view && !startsWithCollStats()) {
auto viewDefinition =
ViewShardingCheck::getResolvedViewIfSharded(txn, ctx.getDb(), view);
if (!viewDefinition.isOK()) {
return appendCommandStatus(result, viewDefinition.getStatus());
}
if (!viewDefinition.getValue().isEmpty()) {
ViewShardingCheck::appendShardedViewStatus(viewDefinition.getValue(), &result);
return false;
}
auto resolvedView = ctx.getDb()->getViewCatalog()->resolveView(txn, nss);
if (!resolvedView.isOK()) {
return appendCommandStatus(result, resolvedView.getStatus());
}
// With the view resolved, we can relinquish locks.
ctx.releaseLocksForView();
// Parse the resolved view into a new aggregation request.
auto newCmd = resolvedView.getValue().asExpandedViewAggregation(request);
if (!newCmd.isOK()) {
return appendCommandStatus(result, newCmd.getStatus());
}
auto newNss = resolvedView.getValue().getNamespace();
auto newRequest = AggregationRequest::parseFromBSON(newNss, newCmd.getValue());
if (!newRequest.isOK()) {
return appendCommandStatus(result, newRequest.getStatus());
}
bool status = runParsed(
txn, origNss, newRequest.getValue(), newCmd.getValue(), errmsg, result);
{
// Set the namespace of the curop back to the view namespace so ctx records
// stats on this view namespace on destruction.
stdx::lock_guard<Client>(*txn->getClient());
curOp->setNS_inlock(nss.ns());
}
return status;
}
// If the pipeline does not have a user-specified collation, set it from the collection
// default.
if (request.getCollation().isEmpty() && collection &&
collection->getDefaultCollator()) {
invariant(!expCtx->getCollator());
expCtx->setCollator(collection->getDefaultCollator()->clone());
}
// Propagate the ExpressionContext throughout all of the pipeline's stages and
// expressions.
pipeline->injectExpressionContext(expCtx);
// The pipeline must be optimized after the correct collator has been set on it (by
// injecting the ExpressionContext containing the collator). This is necessary because
// optimization may make string comparisons, e.g. optimizing {$eq: [<str1>, <str2>]} to
// a constant.
pipeline->optimizePipeline();
if (kDebugBuild && !expCtx->isExplain && !expCtx->inShard) {
// Make sure all operations round-trip through Pipeline::serialize() correctly by
// re-parsing every command in debug builds. This is important because sharded
// aggregations rely on this ability. Skipping when inShard because this has
// already been through the transformation (and this un-sets expCtx->inShard).
pipeline = reparsePipeline(pipeline, request, expCtx);
}
// This does mongod-specific stuff like creating the input PlanExecutor and adding
// it to the front of the pipeline if needed.
PipelineD::prepareCursorSource(collection, pipeline);
// Create the PlanExecutor which returns results from the pipeline. The WorkingSet
// ('ws') and the PipelineProxyStage ('proxy') will be owned by the created
// PlanExecutor.
auto ws = make_unique<WorkingSet>();
auto proxy = make_unique<PipelineProxyStage>(txn, pipeline, ws.get());
auto statusWithPlanExecutor = (NULL == collection)
? PlanExecutor::make(
txn, std::move(ws), std::move(proxy), nss.ns(), PlanExecutor::YIELD_MANUAL)
: PlanExecutor::make(
txn, std::move(ws), std::move(proxy), collection, PlanExecutor::YIELD_MANUAL);
invariant(statusWithPlanExecutor.isOK());
exec = std::move(statusWithPlanExecutor.getValue());
{
auto planSummary = Explain::getPlanSummary(exec.get());
stdx::lock_guard<Client>(*txn->getClient());
curOp->setPlanSummary_inlock(std::move(planSummary));
}
if (collection) {
PlanSummaryStats stats;
Explain::getSummaryStats(*exec, &stats);
collection->infoCache()->notifyOfQuery(txn, stats.indexesUsed);
}
if (collection) {
const bool isAggCursor = true; // enable special locking behavior
ClientCursor* cursor =
new ClientCursor(collection->getCursorManager(),
exec.release(),
nss.ns(),
txn->recoveryUnit()->isReadingFromMajorityCommittedSnapshot(),
0,
cmdObj.getOwned(),
isAggCursor);
pin.reset(new ClientCursorPin(collection->getCursorManager(), cursor->cursorid()));
// Don't add any code between here and the start of the try block.
}
// At this point, it is safe to release the collection lock.
// - In the case where we have a collection: we will need to reacquire the
// collection lock later when cleaning up our ClientCursorPin.
// - In the case where we don't have a collection: our PlanExecutor won't be
// registered, so it will be safe to clean it up outside the lock.
invariant(!exec || !collection);
}
try {
// Unless set to true, the ClientCursor created above will be deleted on block exit.
bool keepCursor = false;
// Use of the aggregate command without specifying to use a cursor is deprecated.
// Applications should migrate to using cursors. Cursors are strictly more useful than
// outputting the results as a single document, since results that fit inside a single
// BSONObj will also fit inside a single batch.
//
// We occasionally log a deprecation warning.
if (!request.isCursorCommand()) {
RARELY {
warning()
<< "Use of the aggregate command without the 'cursor' "
"option is deprecated. See "
"http://dochub.mongodb.org/core/aggregate-without-cursor-deprecation.";
}
}
// If both explain and cursor are specified, explain wins.
if (expCtx->isExplain) {
result << "stages" << Value(pipeline->writeExplainOps());
} else if (request.isCursorCommand()) {
keepCursor = handleCursorCommand(txn,
origNss.ns(),
pin.get(),
pin ? pin->c()->getExecutor() : exec.get(),
request,
result);
} else {
pipeline->run(result);
}
if (!expCtx->isExplain) {
PlanSummaryStats stats;
Explain::getSummaryStats(pin ? *pin->c()->getExecutor() : *exec.get(), &stats);
curOp->debug().setPlanSummaryMetrics(stats);
curOp->debug().nreturned = stats.nReturned;
}
// Clean up our ClientCursorPin, if needed. We must reacquire the collection lock
// in order to do so.
if (pin) {
// We acquire locks here with DBLock and CollectionLock instead of using
// AutoGetCollectionForRead. AutoGetCollectionForRead will throw if the
// sharding version is out of date, and we don't care if the sharding version
// has changed.
Lock::DBLock dbLock(txn->lockState(), nss.db(), MODE_IS);
Lock::CollectionLock collLock(txn->lockState(), nss.ns(), MODE_IS);
if (keepCursor) {
pin->release();
} else {
pin->deleteUnderlying();
}
}
} catch (...) {
/**
* Do the initial sync for this member.
*/
void ReplSetImpl::_syncDoInitialSync() {
replset::InitialSync init(replset::BackgroundSync::get());
sethbmsg("initial sync pending",0);
// if this is the first node, it may have already become primary
if ( box.getState().primary() ) {
sethbmsg("I'm already primary, no need for initial sync",0);
return;
}
const Member *source = getMemberToSyncTo();
if (!source) {
sethbmsg("initial sync need a member to be primary or secondary to do our initial sync", 0);
sleepsecs(15);
return;
}
string sourceHostname = source->h().toString();
init.setHostname(sourceHostname);
OplogReader r;
if( !r.connect(sourceHostname) ) {
sethbmsg( str::stream() << "initial sync couldn't connect to " << source->h().toString() , 0);
sleepsecs(15);
return;
}
BSONObj lastOp = r.getLastOp(rsoplog);
if( lastOp.isEmpty() ) {
sethbmsg("initial sync couldn't read remote oplog", 0);
sleepsecs(15);
return;
}
if (replSettings.fastsync) {
log() << "fastsync: skipping database clone" << rsLog;
// prime oplog
init.oplogApplication(lastOp, lastOp);
return;
}
else {
sethbmsg("initial sync drop all databases", 0);
dropAllDatabasesExceptLocal();
sethbmsg("initial sync clone all databases", 0);
list<string> dbs = r.conn()->getDatabaseNames();
if ( ! _syncDoInitialSync_clone( sourceHostname.c_str(), dbs, true ) ) {
veto(source->fullName(), 600);
sleepsecs(300);
return;
}
sethbmsg("initial sync data copy, starting syncup",0);
BSONObj minValid;
if ( ! _syncDoInitialSync_applyToHead( init, &r , source , lastOp , minValid ) ) {
return;
}
lastOp = minValid;
// its currently important that lastOp is equal to the last op we actually pulled
// this is because the background thread only pulls each op once now
// so if its now, we'll be waiting forever
{
// this takes whatever the last op the we got is
// and stores it locally before we wipe it out below
Lock::DBRead lk(rsoplog);
Helpers::getLast(rsoplog, lastOp);
lastOp = lastOp.getOwned();
}
// reset state, as that "didn't count"
emptyOplog();
lastOpTimeWritten = OpTime();
lastH = 0;
sethbmsg("initial sync building indexes",0);
if ( ! _syncDoInitialSync_clone( sourceHostname.c_str(), dbs, false ) ) {
veto(source->fullName(), 600);
sleepsecs(300);
return;
}
}
sethbmsg("initial sync query minValid",0);
BSONObj minValid;
if ( ! _syncDoInitialSync_applyToHead( init, &r, source, lastOp, minValid ) ) {
return;
}
// ---------
sethbmsg("initial sync finishing up",0);
verify( !box.getState().primary() ); // wouldn't make sense if we were.
{
Client::WriteContext cx( "local." );
cx.ctx().db()->flushFiles(true);
try {
log() << "replSet set minValid=" << minValid["ts"]._opTime().toString() << rsLog;
}
catch(...) { }
theReplSet->setMinValid(minValid);
cx.ctx().db()->flushFiles(true);
}
changeState(MemberState::RS_RECOVERING);
sethbmsg("initial sync done",0);
}
int ConfigDiffTracker<ValType,ShardType>::
calculateConfigDiff( DBClientCursorInterface& diffCursor )
{
MONGO_LOG_DEFAULT_COMPONENT_LOCAL(::mongo::logger::LogComponent::kSharding);
verifyAttached();
// Apply the chunk changes to the ranges and versions
//
// Overall idea here is to work in two steps :
// 1. For all the new chunks we find, increment the maximum version per-shard and
// per-collection, and remove any conflicting chunks from the ranges
// 2. For all the new chunks we're interested in (all of them for mongos, just chunks on the
// shard for mongod) add them to the ranges
//
vector<BSONObj> newTracked;
// Store epoch now so it doesn't change when we change max
OID currEpoch = _maxVersion->epoch();
_validDiffs = 0;
while( diffCursor.more() ){
BSONObj diffChunkDoc = diffCursor.next();
ChunkVersion chunkVersion = ChunkVersion::fromBSON(diffChunkDoc, ChunkType::DEPRECATED_lastmod());
if( diffChunkDoc[ChunkType::min()].type() != Object ||
diffChunkDoc[ChunkType::max()].type() != Object ||
diffChunkDoc[ChunkType::shard()].type() != String )
{
warning() << "got invalid chunk document " << diffChunkDoc
<< " when trying to load differing chunks" << endl;
continue;
}
if( ! chunkVersion.isSet() || ! chunkVersion.hasEqualEpoch( currEpoch ) ){
warning() << "got invalid chunk version " << chunkVersion << " in document " << diffChunkDoc
<< " when trying to load differing chunks at version "
<< ChunkVersion( _maxVersion->majorVersion(),
_maxVersion->minorVersion(),
currEpoch ) << endl;
// Don't keep loading, since we know we'll be broken here
return -1;
}
_validDiffs++;
// Get max changed version and chunk version
if( chunkVersion > *_maxVersion ) *_maxVersion = chunkVersion;
// Chunk version changes
ShardType shard = shardFor( diffChunkDoc[ChunkType::shard()].String() );
typename map<ShardType, ChunkVersion>::iterator shardVersionIt = _maxShardVersions->find( shard );
if( shardVersionIt == _maxShardVersions->end() || shardVersionIt->second < chunkVersion ){
(*_maxShardVersions)[ shard ] = chunkVersion;
}
// See if we need to remove any chunks we are currently tracking b/c of this chunk's changes
removeOverlapping(diffChunkDoc[ChunkType::min()].Obj(),
diffChunkDoc[ChunkType::max()].Obj());
// Figure out which of the new chunks we need to track
// Important - we need to actually own this doc, in case the cursor decides to getMore or unbuffer
if( isTracked( diffChunkDoc ) ) newTracked.push_back( diffChunkDoc.getOwned() );
}
LOG(3) << "found " << _validDiffs
<< " new chunks for collection " << _ns
<< " (tracking " << newTracked.size()
<< "), new version is " << *_maxVersion
<< endl;
for( vector<BSONObj>::iterator it = newTracked.begin(); it != newTracked.end(); it++ ){
BSONObj chunkDoc = *it;
// Important - we need to make sure we actually own the min and max here
BSONObj min = chunkDoc[ChunkType::min()].Obj().getOwned();
BSONObj max = chunkDoc[ChunkType::max()].Obj().getOwned();
// Invariant enforced by sharding
// It's possible to read inconsistent state b/c of getMore() and yielding, so we want
// to detect as early as possible.
// TODO: This checks for overlap, we also should check for holes here iff we're tracking
// all chunks
if( isOverlapping( min, max ) ) return -1;
_currMap->insert( rangeFor( chunkDoc, min, max ) );
}
return _validDiffs;
}
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() );
BSONObj res;
bool ok = false;
{
ShardConnection conn( theShard , "" );
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;
try {
ShardConnection conn( temp , "" );
ON_BLOCK_EXIT_OBJ( conn, &ShardConnection::done );
_addWriteBack( writebacks , conn->getLastErrorDetailed() );
}
catch( std::exception &e ){
warning() << "could not clear last error from shard " << temp << causedBy( e ) << endl;
}
}
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
verify( 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 );
boost::scoped_ptr<ShardConnection> conn;
BSONObj res;
bool ok = false;
try {
conn.reset( new ShardConnection( theShard , "" ) ); // constructor can throw if shard is down
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;
}
PlanStage::StageState IndexScan::work(WorkingSetID* out) {
++_commonStats.works;
// Adds the amount of time taken by work() to executionTimeMillis.
ScopedTimer timer(&_commonStats.executionTimeMillis);
// If we examined multiple keys in a prior work cycle, make up for it here by returning
// NEED_TIME. This is done for plan ranking. Refer to the comment for '_checkEndKeys'
// in the .h for details.
if (_checkEndKeys > 0) {
--_checkEndKeys;
++_commonStats.needTime;
return PlanStage::NEED_TIME;
}
if (NULL == _indexCursor.get()) {
// First call to work(). Perform possibly heavy init.
initIndexScan();
checkEnd();
++_commonStats.needTime;
return PlanStage::NEED_TIME;
}
else if (_yieldMovedCursor) {
_yieldMovedCursor = false;
// Note that we're not calling next() here. We got the next thing when we recovered
// from yielding.
}
if (isEOF()) { return PlanStage::IS_EOF; }
// Grab the next (key, value) from the index.
BSONObj keyObj = _indexCursor->getKey();
DiskLoc loc = _indexCursor->getValue();
// Move to the next result.
// The underlying IndexCursor points at the *next* thing we want to return. We do this so
// that if we're scanning an index looking for docs to delete we don't continually clobber
// the thing we're pointing at.
_indexCursor->next();
checkEnd();
if (_shouldDedup) {
++_specificStats.dupsTested;
if (_returned.end() != _returned.find(loc)) {
++_specificStats.dupsDropped;
++_commonStats.needTime;
return PlanStage::NEED_TIME;
}
else {
_returned.insert(loc);
}
}
if (Filter::passes(keyObj, _keyPattern, _filter)) {
if (NULL != _filter) {
++_specificStats.matchTested;
}
// We must make a copy of the on-disk data since it can mutate during the execution of
// this query.
BSONObj ownedKeyObj = keyObj.getOwned();
// Fill out the WSM.
WorkingSetID id = _workingSet->allocate();
WorkingSetMember* member = _workingSet->get(id);
member->loc = loc;
member->keyData.push_back(IndexKeyDatum(_keyPattern, ownedKeyObj));
member->state = WorkingSetMember::LOC_AND_IDX;
if (_params.addKeyMetadata) {
BSONObjBuilder bob;
bob.appendKeys(_keyPattern, ownedKeyObj);
member->addComputed(new IndexKeyComputedData(bob.obj()));
}
*out = id;
++_commonStats.advanced;
return PlanStage::ADVANCED;
}
++_commonStats.needTime;
return PlanStage::NEED_TIME;
}
bool run(OperationContext* txn,
const string& dbname,
BSONObj& jsobj,
int,
string& errmsg,
BSONObjBuilder& result) {
const std::string ns = parseNs(dbname, jsobj);
md5digest d;
md5_state_t st;
md5_init(&st);
int n = 0;
bool partialOk = jsobj["partialOk"].trueValue();
if (partialOk) {
// WARNING: This code depends on the binary layout of md5_state. It will not be
// compatible with different md5 libraries or work correctly in an environment with
// mongod's of different endians. It is ok for mongos to be a different endian since
// it just passes the buffer through to another mongod.
BSONElement stateElem = jsobj["md5state"];
if (!stateElem.eoo()) {
int len;
const char* data = stateElem.binDataClean(len);
massert(16247, "md5 state not correct size", len == sizeof(st));
memcpy(&st, data, sizeof(st));
}
n = jsobj["startAt"].numberInt();
}
BSONObj query = BSON("files_id" << jsobj["filemd5"] << "n" << GTE << n);
BSONObj sort = BSON("files_id" << 1 << "n" << 1);
MONGO_WRITE_CONFLICT_RETRY_LOOP_BEGIN {
CanonicalQuery* cq;
if (!CanonicalQuery::canonicalize(ns, query, sort, BSONObj(), &cq).isOK()) {
uasserted(17240, "Can't canonicalize query " + query.toString());
return 0;
}
// Check shard version at startup.
// This will throw before we've done any work if shard version is outdated
// We drop and re-acquire these locks every document because md5'ing is expensive
unique_ptr<AutoGetCollectionForRead> ctx(new AutoGetCollectionForRead(txn, ns));
Collection* coll = ctx->getCollection();
PlanExecutor* rawExec;
if (!getExecutor(txn,
coll,
cq,
PlanExecutor::YIELD_MANUAL,
&rawExec,
QueryPlannerParams::NO_TABLE_SCAN).isOK()) {
uasserted(17241, "Can't get executor for query " + query.toString());
return 0;
}
unique_ptr<PlanExecutor> exec(rawExec);
// Process notifications when the lock is released/reacquired in the loop below
exec->registerExec();
BSONObj obj;
PlanExecutor::ExecState state;
while (PlanExecutor::ADVANCED == (state = exec->getNext(&obj, NULL))) {
BSONElement ne = obj["n"];
verify(ne.isNumber());
int myn = ne.numberInt();
if (n != myn) {
if (partialOk) {
break; // skipped chunk is probably on another shard
}
log() << "should have chunk: " << n << " have:" << myn << endl;
dumpChunks(txn, ns, query, sort);
uassert(10040, "chunks out of order", n == myn);
}
// make a copy of obj since we access data in it while yielding locks
BSONObj owned = obj.getOwned();
exec->saveState();
// UNLOCKED
ctx.reset();
int len;
const char* data = owned["data"].binDataClean(len);
// This is potentially an expensive operation, so do it out of the lock
md5_append(&st, (const md5_byte_t*)(data), len);
n++;
try {
// RELOCKED
ctx.reset(new AutoGetCollectionForRead(txn, ns));
} catch (const SendStaleConfigException& ex) {
LOG(1) << "chunk metadata changed during filemd5, will retarget and continue";
break;
}
// Have the lock again. See if we were killed.
if (!exec->restoreState(txn)) {
if (!partialOk) {
uasserted(13281, "File deleted during filemd5 command");
}
}
}
if (partialOk)
result.appendBinData("md5state", sizeof(st), BinDataGeneral, &st);
// This must be *after* the capture of md5state since it mutates st
md5_finish(&st, d);
result.append("numChunks", n);
result.append("md5", digestToString(d));
}
MONGO_WRITE_CONFLICT_RETRY_LOOP_END(txn, "filemd5", dbname);
return true;
}
void ParsedQuery::initFields( const BSONObj& fields ) {
if ( fields.isEmpty() )
return;
_fields.reset( new Projection() );
_fields->init( fields.getOwned() );
}
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();
}
bool _run(const string& , BSONObj& cmdObj, int, string& errmsg, BSONObjBuilder& result, bool fromRepl) {
if( cmdObj["replSetReconfig"].type() != Object ) {
errmsg = "no configuration specified";
return false;
}
// We might want to add the protocol version of theReplSet->config() if it exists,
// instead of just blindly adding our compiled-in CURRENT_PROTOCOL_VERSION. But for
// TokuMX 1.0 it doesn't matter.
BSONObj configObj = ReplSetConfig::addProtocolVersionIfMissing(cmdObj["replSetReconfig"].Obj());
bool force = cmdObj.hasField("force") && cmdObj["force"].trueValue();
if( force && !theReplSet ) {
replSettings.reconfig = configObj.getOwned();
result.append("msg", "will try this config momentarily, try running rs.conf() again in a few seconds");
return true;
}
if ( !check(errmsg, result) ) {
return false;
}
if( !force && !theReplSet->box.getState().primary() ) {
errmsg = "replSetReconfig command must be sent to the current replica set primary.";
return false;
}
{
// just make sure we can get a write lock before doing anything else. we'll reacquire one
// later. of course it could be stuck then, but this check lowers the risk if weird things
// are up - we probably don't want a change to apply 30 minutes after the initial attempt.
time_t t = time(0);
Lock::GlobalWrite lk;
if( time(0)-t > 20 ) {
errmsg = "took a long time to get write lock, so not initiating. Initiate when server less busy?";
return false;
}
}
try {
scoped_ptr<ReplSetConfig> newConfig
(ReplSetConfig::make(configObj, force));
log() << "replSet replSetReconfig config object parses ok, " <<
newConfig->members.size() << " members specified" << rsLog;
if( !ReplSetConfig::legalChange(theReplSet->getConfig(), *newConfig, errmsg) ) {
return false;
}
checkMembersUpForConfigChange(*newConfig, result, false);
log() << "replSet replSetReconfig [2]" << rsLog;
theReplSet->haveNewConfig(*newConfig, true);
ReplSet::startupStatusMsg.set("replSetReconfig'd");
}
catch( DBException& e ) {
log() << "replSet replSetReconfig exception: " << e.what() << rsLog;
throw;
}
catch( string& se ) {
log() << "replSet reconfig exception: " << se << rsLog;
errmsg = se;
return false;
}
resetSlaveCache();
return true;
}
void BatchedUpdateDocument::setQuery(const BSONObj& query) {
_query = query.getOwned();
_isQuerySet = true;
}
/**
* Run a query with a cursor provided by the query optimizer, or FindingStartCursor.
* @returns true if client cursor was saved, false if the query has completed.
*/
bool queryWithQueryOptimizer( int queryOptions, const string& ns,
const BSONObj &jsobj, CurOp& curop,
const BSONObj &query, const BSONObj &order,
const shared_ptr<ParsedQuery> &pq_shared,
const ConfigVersion &shardingVersionAtStart,
const bool getCachedExplainPlan,
const bool inMultiStatementTxn,
Message &result ) {
const ParsedQuery &pq( *pq_shared );
shared_ptr<Cursor> cursor;
QueryPlanSummary queryPlan;
const bool tailable = pq.hasOption( QueryOption_CursorTailable ) && pq.getNumToReturn() != 1;
LOG(1) << "query beginning read-only transaction. tailable: " << tailable << endl;
BSONObj oldPlan;
if (getCachedExplainPlan) {
scoped_ptr<MultiPlanScanner> mps( MultiPlanScanner::make( ns.c_str(), query, order ) );
oldPlan = mps->cachedPlanExplainSummary();
}
cursor = getOptimizedCursor( ns.c_str(), query, order, QueryPlanSelectionPolicy::any(),
pq_shared, false, &queryPlan );
verify( cursor );
// Tailable cursors must be marked as such before any use. This is so that
// the implementation knows that uncommitted data cannot be returned.
if ( tailable ) {
cursor->setTailable();
}
scoped_ptr<QueryResponseBuilder> queryResponseBuilder
( QueryResponseBuilder::make( pq, cursor, queryPlan, oldPlan ) );
bool saveClientCursor = false;
int options = QueryOption_NoCursorTimeout;
if (pq.hasOption( QueryOption_OplogReplay )) {
options |= QueryOption_OplogReplay;
}
// create a client cursor that does not create a cursorID.
// The cursor ID will be created if and only if we save
// the client cursor further below
ClientCursor::Holder ccPointer(
new ClientCursor( options, cursor, ns, BSONObj(), false, false )
);
// for oplog cursors, we check if we are reading data that is too old and might
// be stale.
bool opChecked = false;
bool slaveLocationUpdated = false;
BSONObj last;
bool lastBSONObjSet = false;
for ( ; cursor->ok(); cursor->advance() ) {
if ( pq.getMaxScan() && cursor->nscanned() > pq.getMaxScan() ) {
break;
}
if ( !queryResponseBuilder->addMatch() ) {
continue;
}
// Note slave's position in the oplog.
if ( pq.hasOption( QueryOption_OplogReplay ) ) {
BSONObj current = cursor->current();
last = current.copy();
lastBSONObjSet = true;
// the first row returned is equal to the last element that
// the slave has synced up to, so we might as well update
// the slave location
if (!slaveLocationUpdated) {
ccPointer->updateSlaveLocation(curop);
slaveLocationUpdated = true;
}
// check if data we are about to return may be too stale
if (!opChecked) {
ccPointer->storeOpForSlave(current);
uassert(16785, "oplog cursor reading data that is too old", !ccPointer->lastOpForSlaveTooOld());
opChecked = true;
}
}
if ( pq.isExplain() ) {
if ( queryResponseBuilder->enoughTotalResults() ) {
break;
}
}
else if ( queryResponseBuilder->enoughForFirstBatch() ) {
// if only 1 requested, no cursor saved for efficiency...we assume it is findOne()
if ( pq.wantMore() && pq.getNumToReturn() != 1 ) {
queryResponseBuilder->finishedFirstBatch();
if ( cursor->advance() ) {
saveClientCursor = true;
}
}
break;
}
}
// If the tailing request succeeded
if ( cursor->tailable() ) {
saveClientCursor = true;
}
if ( ! shardingState.getVersion( ns ).isWriteCompatibleWith( shardingVersionAtStart ) ) {
// if the version changed during the query
// we might be missing some data
// and its safe to send this as mongos can resend
// at this point
throw SendStaleConfigException( ns , "version changed during initial query", shardingVersionAtStart, shardingState.getVersion( ns ) );
}
int nReturned = queryResponseBuilder->handoff( result );
ccPointer.reset();
long long cursorid = 0;
if ( saveClientCursor ) {
// Create a new ClientCursor, with a default timeout.
ccPointer.reset( new ClientCursor( queryOptions, cursor, ns,
jsobj.getOwned(), inMultiStatementTxn ) );
cursorid = ccPointer->cursorid();
DEV tlog(2) << "query has more, cursorid: " << cursorid << endl;
if ( !ccPointer->ok() && ccPointer->c()->tailable() ) {
DEV tlog() << "query has no more but tailable, cursorid: " << cursorid << endl;
}
if( queryOptions & QueryOption_Exhaust ) {
curop.debug().exhaust = true;
}
// Set attributes for getMore.
ccPointer->setChunkManager( queryResponseBuilder->chunkManager() );
ccPointer->setPos( nReturned );
ccPointer->pq = pq_shared;
ccPointer->fields = pq.getFieldPtr();
if (pq.hasOption( QueryOption_OplogReplay ) && lastBSONObjSet) {
ccPointer->storeOpForSlave(last);
}
if (!inMultiStatementTxn) {
// This cursor is not part of a multi-statement transaction, so
// we pass off the current client's transaction stack to the
// cursor so that it may be live as long as the cursor.
cc().swapTransactionStack(ccPointer->transactions);
verify(!cc().hasTxn());
}
ccPointer.release();
}
QueryResult *qr = (QueryResult *) result.header();
qr->cursorId = cursorid;
curop.debug().cursorid = ( cursorid == 0 ? -1 : qr->cursorId );
qr->setResultFlagsToOk();
// qr->len is updated automatically by appendData()
curop.debug().responseLength = qr->len;
qr->setOperation(opReply);
qr->startingFrom = 0;
qr->nReturned = nReturned;
curop.debug().nscanned = ( cursor ? cursor->nscanned() : 0LL );
curop.debug().ntoskip = pq.getSkip();
curop.debug().nreturned = nReturned;
return saveClientCursor;
}
void BatchedUpdateDocument::setUpdateExpr(const BSONObj& updateExpr) {
_updateExpr = updateExpr.getOwned();
_isUpdateExprSet = true;
}
/**
* Invariant: idObj should belong to a document that is part of the active chunk being migrated
*/
LogOpForShardingHandler(MigrationSourceManager* migrateSourceManager,
const BSONObj& idObj,
const char op)
: _migrationSourceManager(migrateSourceManager), _idObj(idObj.getOwned()), _op(op) {}
void BatchedUpdateDocument::setCollation(const BSONObj& collation) {
_collation = collation.getOwned();
_isCollationSet = true;
}
void operator()( DBClientCursorBatchIterator &i ) {
Lock::GlobalWrite lk;
context.relocked();
bool createdCollection = false;
Collection* collection = NULL;
while( i.moreInCurrentBatch() ) {
if ( numSeen % 128 == 127 /*yield some*/ ) {
collection = NULL;
time_t now = time(0);
if( now - lastLog >= 60 ) {
// report progress
if( lastLog )
log() << "clone " << to_collection << ' ' << numSeen << endl;
lastLog = now;
}
mayInterrupt( _mayBeInterrupted );
dbtempreleaseif t( _mayYield );
}
if ( isindex == false && collection == NULL ) {
collection = context.db()->getCollection( to_collection );
if ( !collection ) {
massert( 17321,
str::stream()
<< "collection dropped during clone ["
<< to_collection << "]",
!createdCollection );
createdCollection = true;
collection = context.db()->createCollection( to_collection );
verify( collection );
}
}
BSONObj tmp = i.nextSafe();
/* assure object is valid. note this will slow us down a little. */
const Status status = validateBSON(tmp.objdata(), tmp.objsize());
if (!status.isOK()) {
out() << "Cloner: skipping corrupt object from " << from_collection
<< ": " << status.reason();
continue;
}
++numSeen;
BSONObj js = tmp;
if ( isindex ) {
verify(nsToCollectionSubstring(from_collection) == "system.indexes");
js = fixindex(tmp);
indexesToBuild->push_back( js.getOwned() );
continue;
}
verify(nsToCollectionSubstring(from_collection) != "system.indexes");
StatusWith<DiskLoc> loc = collection->insertDocument( js, true );
if ( !loc.isOK() ) {
error() << "error: exception cloning object in " << from_collection
<< ' ' << loc.toString() << " obj:" << js;
}
uassertStatusOK( loc.getStatus() );
if ( logForRepl )
logOp("i", to_collection, js);
getDur().commitIfNeeded();
RARELY if ( time( 0 ) - saveLast > 60 ) {
log() << numSeen << " objects cloned so far from collection " << from_collection;
saveLast = time( 0 );
}
}
}
void appendReplicationInfo(OperationContext* txn, BSONObjBuilder& result, int level) {
ReplicationCoordinator* replCoord = getGlobalReplicationCoordinator();
if (replCoord->getSettings().usingReplSets()) {
IsMasterResponse isMasterResponse;
replCoord->fillIsMasterForReplSet(&isMasterResponse);
result.appendElements(isMasterResponse.toBSON());
if (level) {
replCoord->appendSlaveInfoData(&result);
}
return;
}
// TODO(dannenberg) replAllDead is bad and should be removed when master slave is removed
if (replAllDead) {
result.append("ismaster", 0);
string s = string("dead: ") + replAllDead;
result.append("info", s);
} else {
result.appendBool("ismaster",
getGlobalReplicationCoordinator()->isMasterForReportingPurposes());
}
if (level) {
BSONObjBuilder sources(result.subarrayStart("sources"));
int n = 0;
list<BSONObj> src;
{
const char* localSources = "local.sources";
AutoGetCollectionForRead ctx(txn, localSources);
unique_ptr<PlanExecutor> exec(InternalPlanner::collectionScan(
txn, localSources, ctx.getCollection(), PlanExecutor::YIELD_MANUAL));
BSONObj obj;
PlanExecutor::ExecState state;
while (PlanExecutor::ADVANCED == (state = exec->getNext(&obj, NULL))) {
src.push_back(obj.getOwned());
}
// Non-yielding collection scans from InternalPlanner will never error.
invariant(PlanExecutor::IS_EOF == state);
}
for (list<BSONObj>::const_iterator i = src.begin(); i != src.end(); i++) {
BSONObj s = *i;
BSONObjBuilder bb;
bb.append(s["host"]);
string sourcename = s["source"].valuestr();
if (sourcename != "main")
bb.append(s["source"]);
{
BSONElement e = s["syncedTo"];
BSONObjBuilder t(bb.subobjStart("syncedTo"));
t.appendDate("time", e.timestampTime());
t.append("inc", e.timestampInc());
t.done();
}
if (level > 1) {
wassert(!txn->lockState()->isLocked());
// note: there is no so-style timeout on this connection; perhaps we should have
// one.
ScopedDbConnection conn(s["host"].valuestr());
DBClientConnection* cliConn = dynamic_cast<DBClientConnection*>(&conn.conn());
if (cliConn && replAuthenticate(cliConn)) {
BSONObj first = conn->findOne((string) "local.oplog.$" + sourcename,
Query().sort(BSON("$natural" << 1)));
BSONObj last = conn->findOne((string) "local.oplog.$" + sourcename,
Query().sort(BSON("$natural" << -1)));
bb.appendDate("masterFirst", first["ts"].timestampTime());
bb.appendDate("masterLast", last["ts"].timestampTime());
const auto lag = (last["ts"].timestampTime() - s["syncedTo"].timestampTime());
bb.append("lagSeconds", durationCount<Milliseconds>(lag) / 1000.0);
}
conn.done();
}
sources.append(BSONObjBuilder::numStr(n++), bb.obj());
}
sources.done();
replCoord->appendSlaveInfoData(&result);
}
}
bool ClientInfo::getLastError( const string& dbName,
const BSONObj& options,
BSONObjBuilder& result,
string& errmsg,
bool fromWriteBackListener)
{
scoped_ptr<TimerHolder> gleTimerHolder;
if ( ! fromWriteBackListener ) {
bool doTiming = false;
const BSONElement& e = options["w"];
if ( e.isNumber() ) {
doTiming = e.numberInt() > 1;
}
else if ( e.type() == String ) {
doTiming = true;
}
if ( doTiming ) {
gleTimerHolder.reset( new TimerHolder( &gleWtimeStats ) );
}
}
set<string> * shards = getPrev();
if ( shards->size() == 0 ) {
result.appendNull( "err" );
return true;
}
vector<WBInfo> writebacks;
//
// TODO: These branches should be collapsed into a single codepath
//
// handle single server
if ( shards->size() == 1 ) {
string theShard = *(shards->begin() );
BSONObj res;
bool ok = false;
{
LOG(5) << "gathering response for gle from: " << theShard << endl;
ShardConnection conn( theShard , "" );
try {
ok = conn->runCommand( dbName , options , res );
}
catch( std::exception &e ) {
string message =
str::stream() << "could not get last error from shard " << theShard
<< causedBy( e );
warning() << message << endl;
errmsg = message;
// 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, true );
LOG(4) << "gathering writebacks from " << sinceLastGetError().size() << " hosts for"
<< " gle (" << theShard << ")" << endl;
// 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;
LOG(5) << "gathering writebacks for single-shard gle from: " << temp << endl;
try {
ShardConnection conn( temp , "" );
ON_BLOCK_EXIT_OBJ( conn, &ShardConnection::done );
_addWriteBack( writebacks, conn->getLastErrorDetailed(), false );
}
catch( std::exception &e ){
warning() << "could not clear last error from shard " << temp << causedBy( e ) << endl;
}
}
clearSinceLastGetError();
LOG(4) << "checking " << writebacks.size() << " writebacks for"
<< " gle (" << theShard << ")" << endl;
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
verify( v.size() >= 1 );
if ( res["writebackSince"].numberInt() > 0 ) {
// got writeback from older op
// ignore the result from it, just needed to wait
result.appendElements( res );
}
else if ( writebacks[0].fromLastOperation ) {
result.appendElements( v[0] );
result.appendElementsUnique( res );
result.append( "writebackGLE" , v[0] );
result.append( "initialGLEHost" , theShard );
result.append( "initialGLE", res );
}
else {
// there was a writeback
// but its from an old operations
// so all that's important is that we block, not that we return stats
result.appendElements( res );
}
}
}
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 );
LOG(5) << "gathering a response for gle from: " << theShard << endl;
boost::scoped_ptr<ShardConnection> conn;
BSONObj res;
bool ok = false;
try {
conn.reset( new ShardConnection( theShard , "" ) ); // constructor can throw if shard is down
ok = (*conn)->runCommand( dbName , 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.
string message =
str::stream() << "could not get last error from a shard " << theShard
<< causedBy( e );
warning() << message << endl;
errmsg = message;
if (conn)
conn->done();
return false;
}
_addWriteBack( writebacks, res, true );
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 );
LOG(4) << "gathering writebacks from " << sinceLastGetError().size() << " hosts for"
<< " gle (" << shards->size() << " shards)" << endl;
// 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;
LOG(5) << "gathering writebacks for multi-shard gle from: " << temp << endl;
ShardConnection conn( temp , "" );
try {
_addWriteBack( writebacks, conn->getLastErrorDetailed(), false );
}
catch( std::exception &e ){
warning() << "could not clear last error from a shard " << temp << causedBy( e ) << endl;
}
conn.done();
}
clearSinceLastGetError();
LOG(4) << "checking " << writebacks.size() << " writebacks for"
<< " gle (" << shards->size() << " shards)" << endl;
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;
}
bool run(OperationContext* txn,
const string& dbname,
BSONObj& cmdObj,
int,
string& errmsg,
BSONObjBuilder& result) {
BSONElement first = cmdObj.firstElement();
uassert(28528,
str::stream() << "Argument to listIndexes must be of type String, not "
<< typeName(first.type()),
first.type() == String);
StringData collectionName = first.valueStringData();
uassert(28529,
str::stream() << "Argument to listIndexes must be a collection name, "
<< "not the empty string",
!collectionName.empty());
const NamespaceString ns(dbname, collectionName);
const long long defaultBatchSize = std::numeric_limits<long long>::max();
long long batchSize;
Status parseCursorStatus = parseCommandCursorOptions(cmdObj, defaultBatchSize, &batchSize);
if (!parseCursorStatus.isOK()) {
return appendCommandStatus(result, parseCursorStatus);
}
AutoGetCollectionForRead autoColl(txn, ns);
if (!autoColl.getDb()) {
return appendCommandStatus(result,
Status(ErrorCodes::NamespaceNotFound, "no database"));
}
const Collection* collection = autoColl.getCollection();
if (!collection) {
return appendCommandStatus(result,
Status(ErrorCodes::NamespaceNotFound, "no collection"));
}
const CollectionCatalogEntry* cce = collection->getCatalogEntry();
invariant(cce);
vector<string> indexNames;
MONGO_WRITE_CONFLICT_RETRY_LOOP_BEGIN {
indexNames.clear();
cce->getAllIndexes(txn, &indexNames);
}
MONGO_WRITE_CONFLICT_RETRY_LOOP_END(txn, "listIndexes", ns.ns());
auto ws = make_unique<WorkingSet>();
auto root = make_unique<QueuedDataStage>(txn, ws.get());
for (size_t i = 0; i < indexNames.size(); i++) {
BSONObj indexSpec;
MONGO_WRITE_CONFLICT_RETRY_LOOP_BEGIN {
indexSpec = cce->getIndexSpec(txn, indexNames[i]);
}
MONGO_WRITE_CONFLICT_RETRY_LOOP_END(txn, "listIndexes", ns.ns());
WorkingSetID id = ws->allocate();
WorkingSetMember* member = ws->get(id);
member->keyData.clear();
member->loc = RecordId();
member->obj = Snapshotted<BSONObj>(SnapshotId(), indexSpec.getOwned());
member->transitionToOwnedObj();
root->pushBack(id);
}
std::string cursorNamespace = str::stream() << dbname << ".$cmd." << name << "."
<< ns.coll();
dassert(NamespaceString(cursorNamespace).isValid());
dassert(NamespaceString(cursorNamespace).isListIndexesCursorNS());
dassert(ns == NamespaceString(cursorNamespace).getTargetNSForListIndexes());
auto statusWithPlanExecutor = PlanExecutor::make(
txn, std::move(ws), std::move(root), cursorNamespace, PlanExecutor::YIELD_MANUAL);
if (!statusWithPlanExecutor.isOK()) {
return appendCommandStatus(result, statusWithPlanExecutor.getStatus());
}
unique_ptr<PlanExecutor> exec = std::move(statusWithPlanExecutor.getValue());
BSONArrayBuilder firstBatch;
const int byteLimit = FindCommon::kMaxBytesToReturnToClientAtOnce;
for (long long objCount = 0; objCount < batchSize && firstBatch.len() < byteLimit;
objCount++) {
BSONObj next;
PlanExecutor::ExecState state = exec->getNext(&next, NULL);
if (state == PlanExecutor::IS_EOF) {
break;
}
invariant(state == PlanExecutor::ADVANCED);
firstBatch.append(next);
}
CursorId cursorId = 0LL;
if (!exec->isEOF()) {
exec->saveState();
exec->detachFromOperationContext();
ClientCursor* cursor =
new ClientCursor(CursorManager::getGlobalCursorManager(),
exec.release(),
cursorNamespace,
txn->recoveryUnit()->isReadingFromMajorityCommittedSnapshot());
cursorId = cursor->cursorid();
}
appendCursorResponseObject(cursorId, cursorNamespace, firstBatch.arr(), &result);
return true;
}
void WriteBackListener::run() {
int secsToSleep = 0;
scoped_ptr<ShardChunkVersion> lastNeededVersion;
int lastNeededCount = 0;
while ( ! inShutdown() ) {
if ( ! Shard::isAShardNode( _addr ) ) {
LOG(1) << _addr << " is not a shard node" << endl;
sleepsecs( 60 );
continue;
}
try {
scoped_ptr<ScopedDbConnection> conn(
ScopedDbConnection::getInternalScopedDbConnection( _addr ) );
BSONObj result;
{
BSONObjBuilder cmd;
cmd.appendOID( "writebacklisten" , &serverID ); // Command will block for data
if ( ! conn->get()->runCommand( "admin" , cmd.obj() , result ) ) {
result = result.getOwned();
log() << "writebacklisten command failed! " << result << endl;
conn->done();
continue;
}
}
conn->done();
LOG(1) << "writebacklisten result: " << result << endl;
BSONObj data = result.getObjectField( "data" );
if ( data.getBoolField( "writeBack" ) ) {
string ns = data["ns"].valuestrsafe();
ConnectionIdent cid( "" , 0 );
OID wid;
if ( data["connectionId"].isNumber() && data["id"].type() == jstOID ) {
string s = "";
if ( data["instanceIdent"].type() == String )
s = data["instanceIdent"].String();
cid = ConnectionIdent( s , data["connectionId"].numberLong() );
wid = data["id"].OID();
}
else {
warning() << "mongos/mongod version mismatch (1.7.5 is the split)" << endl;
}
int len; // not used, but needed for next call
Message msg( (void*)data["msg"].binData( len ) , false );
massert( 10427 , "invalid writeback message" , msg.header()->valid() );
DBConfigPtr db = grid.getDBConfig( ns );
ShardChunkVersion needVersion = ShardChunkVersion::fromBSON( data, "version" );
//
// TODO: Refactor the sharded strategy to correctly handle all sharding state changes itself,
// we can't rely on WBL to do this for us b/c anything could reset our state in-between.
// We should always reload here for efficiency when possible, but staleness is also caught in the
// loop below.
//
ChunkManagerPtr manager;
ShardPtr primary;
db->getChunkManagerOrPrimary( ns, manager, primary );
ShardChunkVersion currVersion;
if( manager ) currVersion = manager->getVersion();
LOG(1) << "connectionId: " << cid << " writebackId: " << wid << " needVersion : " << needVersion.toString()
<< " mine : " << currVersion.toString() << endl;
LOG(1) << msg.toString() << endl;
//
// We should reload only if we need to update our version to be compatible *and* we
// haven't already done so. This avoids lots of reloading when we remove/add a sharded collection
//
bool alreadyReloaded = lastNeededVersion &&
lastNeededVersion->isEquivalentTo( needVersion );
if( alreadyReloaded ){
LOG(1) << "wbl already reloaded config information for version "
<< needVersion << ", at version " << currVersion << endl;
}
else if( lastNeededVersion ) {
log() << "new version change detected to " << needVersion.toString()
<< ", " << lastNeededCount << " writebacks processed at "
<< lastNeededVersion->toString() << endl;
lastNeededCount = 0;
}
//
// Set our lastNeededVersion for next time
//
lastNeededVersion.reset( new ShardChunkVersion( needVersion ) );
lastNeededCount++;
//
// Determine if we should reload, if so, reload
//
bool shouldReload = ! needVersion.isWriteCompatibleWith( currVersion ) &&
! alreadyReloaded;
if( shouldReload && currVersion.isSet()
&& needVersion.isSet()
&& currVersion.hasCompatibleEpoch( needVersion ) )
{
//
// If we disagree about versions only, reload the chunk manager
//
db->getChunkManagerIfExists( ns, true );
}
else if( shouldReload ){
//
// If we disagree about anything else, reload the full db
//
warning() << "reloading config data for " << db->getName() << ", "
<< "wanted version " << needVersion.toString()
<< " but currently have version " << currVersion.toString() << endl;
db->reload();
}
// do request and then call getLastError
// we have to call getLastError so we can return the right fields to the user if they decide to call getLastError
BSONObj gle;
int attempts = 0;
while ( true ) {
attempts++;
try {
Request r( msg , 0 );
r.init();
r.d().reservedField() |= Reserved_FromWriteback;
ClientInfo * ci = r.getClientInfo();
if (!noauth) {
// TODO: Figure out why this is 'admin' instead of 'local'.
ci->getAuthenticationInfo()->authorize("admin", internalSecurity.user);
}
ci->noAutoSplit();
r.process( attempts );
ci->newRequest(); // this so we flip prev and cur shards
BSONObjBuilder b;
string errmsg;
if ( ! ci->getLastError( "admin",
BSON( "getLastError" << 1 ),
b,
errmsg,
true ) )
{
b.appendBool( "commandFailed" , true );
if( ! b.hasField( "errmsg" ) ){
b.append( "errmsg", errmsg );
gle = b.obj();
}
else if( errmsg.size() > 0 ){
// Rebuild GLE object with errmsg
// TODO: Make this less clumsy by improving GLE interface
gle = b.obj();
if( gle["errmsg"].type() == String ){
BSONObj gleNoErrmsg =
gle.filterFieldsUndotted( BSON( "errmsg" << 1 ),
false );
BSONObjBuilder bb;
bb.appendElements( gleNoErrmsg );
bb.append( "errmsg", gle["errmsg"].String() +
" ::and:: " +
errmsg );
gle = bb.obj().getOwned();
}
}
}
else{
gle = b.obj();
}
log() << "GLE is " << gle << endl;
if ( gle["code"].numberInt() == 9517 ) {
log() << "new version change detected, "
<< lastNeededCount << " writebacks processed previously" << endl;
lastNeededVersion.reset();
lastNeededCount = 1;
log() << "writeback failed because of stale config, retrying attempts: " << attempts << endl;
LOG(1) << "writeback error : " << gle << endl;
//
// Bringing this in line with the similar retry logic elsewhere
//
// TODO: Reloading the chunk manager may not help if we dropped a
// collection, but we don't actually have that info in the writeback
// error
//
if( attempts <= 2 ){
db->getChunkManagerIfExists( ns, true );
}
else{
versionManager.forceRemoteCheckShardVersionCB( ns );
sleepsecs( attempts - 1 );
}
uassert( 15884, str::stream()
<< "Could not reload chunk manager after "
<< attempts << " attempts.", attempts <= 4 );
continue;
}
ci->clearSinceLastGetError();
}
catch ( DBException& e ) {
error() << "error processing writeback: " << e << endl;
BSONObjBuilder b;
e.getInfo().append( b, "err", "code" );
gle = b.obj();
}
break;
}
{
scoped_lock lk( _seenWritebacksLock );
WBStatus& s = _seenWritebacks[cid];
s.id = wid;
s.gle = gle;
}
}
else if ( result["noop"].trueValue() ) {
// no-op
}
else {
log() << "unknown writeBack result: " << result << endl;
}
secsToSleep = 0;
continue;
}
catch ( std::exception& e ) {
if ( inShutdown() ) {
// we're shutting down, so just clean up
return;
}
log() << "WriteBackListener exception : " << e.what() << endl;
// It's possible this shard was removed
Shard::reloadShardInfo();
}
catch ( ... ) {
log() << "WriteBackListener uncaught exception!" << endl;
}
secsToSleep++;
sleepsecs(secsToSleep);
if ( secsToSleep > 10 )
secsToSleep = 0;
}
log() << "WriteBackListener exiting : address no longer in cluster " << _addr;
}
void operator()( DBClientCursorBatchIterator &i ) {
Lock::GlobalWrite lk;
if ( context ) {
context->relocked();
}
while( i.moreInCurrentBatch() ) {
if ( n % 128 == 127 /*yield some*/ ) {
time_t now = time(0);
if( now - lastLog >= 60 ) {
// report progress
if( lastLog )
log() << "clone " << to_collection << ' ' << n << endl;
lastLog = now;
}
mayInterrupt( _mayBeInterrupted );
dbtempreleaseif t( _mayYield );
}
BSONObj tmp = i.nextSafe();
/* assure object is valid. note this will slow us down a little. */
if ( !tmp.valid() ) {
stringstream ss;
ss << "Cloner: skipping corrupt object from " << from_collection;
BSONElement e = tmp.firstElement();
try {
e.validate();
ss << " firstElement: " << e;
}
catch( ... ) {
ss << " firstElement corrupt";
}
out() << ss.str() << endl;
continue;
}
++n;
BSONObj js = tmp;
if ( isindex ) {
verify( strstr(from_collection, "system.indexes") );
js = fixindex(tmp);
storedForLater->push_back( js.getOwned() );
continue;
}
try {
theDataFileMgr.insertWithObjMod(to_collection, js);
if ( logForRepl )
logOp("i", to_collection, js);
getDur().commitIfNeeded();
}
catch( UserException& e ) {
log() << "warning: exception cloning object in " << from_collection << ' ' << e.what() << " obj:" << js.toString() << '\n';
}
RARELY if ( time( 0 ) - saveLast > 60 ) {
log() << n << " objects cloned so far from collection " << from_collection << endl;
saveLast = time( 0 );
}
}
}
void BatchedInsertRequest::setWriteConcern(const BSONObj& writeConcern) {
_writeConcern = writeConcern.getOwned();
_isWriteConcernSet = true;
}
MigrateInfo* BalancerPolicy::balance( const string& ns,
const DistributionStatus& distribution,
int balancedLastTime ) {
// 1) check for shards that policy require to us to move off of:
// draining only
// 2) check tag policy violations
// 3) then we make sure chunks are balanced for each tag
// ----
// 1) check things we have to move
{
const set<string>& shards = distribution.shards();
for ( set<string>::const_iterator z = shards.begin(); z != shards.end(); ++z ) {
string shard = *z;
const ShardInfo& info = distribution.shardInfo( shard );
if ( ! info.isDraining() )
continue;
if ( distribution.numberOfChunksInShard( shard ) == 0 )
continue;
// now we know we need to move to chunks off this shard
// we will if we are allowed
if ( info.hasOpsQueued() ) {
warning() << "want to shed load from " << shard << " but can't because it has ops queued" << endl;
continue;
}
const vector<BSONObj>& chunks = distribution.getChunks( shard );
unsigned numJumboChunks = 0;
// since we have to move all chunks, lets just do in order
for ( unsigned i=0; i<chunks.size(); i++ ) {
BSONObj chunkToMove = chunks[i];
if ( _isJumbo( chunkToMove ) ) {
numJumboChunks++;
continue;
}
string tag = distribution.getTagForChunk( chunkToMove );
string to = distribution.getBestReceieverShard( tag );
if ( to.size() == 0 ) {
warning() << "want to move chunk: " << chunkToMove << "(" << tag << ") "
<< "from " << shard << " but can't find anywhere to put it" << endl;
continue;
}
log() << "going to move " << chunkToMove << " from " << shard << "(" << tag << ")" << " to " << to << endl;
return new MigrateInfo( ns, to, shard, chunkToMove.getOwned() );
}
warning() << "can't find any chunk to move from: " << shard
<< " but we want to. "
<< " numJumboChunks: " << numJumboChunks
<< endl;
}
}
// 2) tag violations
if ( distribution.tags().size() > 0 ) {
const set<string>& shards = distribution.shards();
for ( set<string>::const_iterator i = shards.begin(); i != shards.end(); ++i ) {
string shard = *i;
const ShardInfo& info = distribution.shardInfo( shard );
const vector<BSONObj>& chunks = distribution.getChunks( shard );
for ( unsigned j = 0; j < chunks.size(); j++ ) {
string tag = distribution.getTagForChunk( chunks[j] );
if ( info.hasTag( tag ) )
continue;
// uh oh, this chunk is in the wrong place
log() << "chunk " << chunks[j]
<< " is not on a shard with the right tag: "
<< tag << endl;
if ( _isJumbo( chunks[j] ) ) {
warning() << "chunk " << chunks[j] << " is jumbo, so cannot be moved" << endl;
continue;
}
string to = distribution.getBestReceieverShard( tag );
if ( to.size() == 0 ) {
log() << "no where to put it :(" << endl;
continue;
}
verify( to != shard );
log() << " going to move to: " << to << endl;
return new MigrateInfo( ns, to, shard, chunks[j].getOwned() );
}
}
}
// 3) for each tag balance
int threshold = 8;
if ( balancedLastTime || distribution.totalChunks() < 20 )
threshold = 2;
else if ( distribution.totalChunks() < 80 )
threshold = 4;
// randomize the order in which we balance the tags
// this is so that one bad tag doesn't prevent others from getting balanced
vector<string> tags;
{
set<string> t = distribution.tags();
for ( set<string>::const_iterator i = t.begin(); i != t.end(); ++i )
tags.push_back( *i );
tags.push_back( "" );
std::random_shuffle( tags.begin(), tags.end() );
}
for ( unsigned i=0; i<tags.size(); i++ ) {
string tag = tags[i];
string from = distribution.getMostOverloadedShard( tag );
if ( from.size() == 0 )
continue;
unsigned max = distribution.numberOfChunksInShardWithTag( from, tag );
if ( max == 0 )
continue;
string to = distribution.getBestReceieverShard( tag );
if ( to.size() == 0 ) {
log() << "no available shards to take chunks for tag [" << tag << "]" << endl;
return NULL;
}
unsigned min = distribution.numberOfChunksInShardWithTag( to, tag );
const int imbalance = max - min;
LOG(1) << "collection : " << ns << endl;
LOG(1) << "donor : " << from << " chunks on " << max << endl;
LOG(1) << "receiver : " << to << " chunks on " << min << endl;
LOG(1) << "threshold : " << threshold << endl;
if ( imbalance < threshold )
continue;
const vector<BSONObj>& chunks = distribution.getChunks( from );
unsigned numJumboChunks = 0;
for ( unsigned j = 0; j < chunks.size(); j++ ) {
if ( distribution.getTagForChunk( chunks[j] ) != tag )
continue;
if ( _isJumbo( chunks[j] ) ) {
numJumboChunks++;
continue;
}
log() << " ns: " << ns << " going to move " << chunks[j]
<< " from: " << from << " to: " << to << " tag [" << tag << "]"
<< endl;
return new MigrateInfo( ns, to, from, chunks[j] );
}
if ( numJumboChunks ) {
error() << "shard: " << from << "ns: " << ns
<< "has too many chunks, but they are all jumbo "
<< " numJumboChunks: " << numJumboChunks
<< endl;
continue;
}
verify( false ); // should be impossible
}
// Everything is balanced here!
return NULL;
}
IndexBuilder::IndexBuilder(const BSONObj& index) :
BackgroundJob(true /* self-delete */), _index(index.getOwned()),
_name(str::stream() << "repl index builder " << _indexBuildCount.addAndFetch(1)) {
}
void StorageEngineMetadata::setStorageEngineOptions(const BSONObj& storageEngineOptions) {
_storageEngineOptions = storageEngineOptions.getOwned();
}