Status getOplogStartHack(OperationContext* txn,
Collection* collection,
CanonicalQuery* cq,
PlanExecutor** execOut) {
invariant(cq);
auto_ptr<CanonicalQuery> autoCq(cq);
if ( collection == NULL )
return Status(ErrorCodes::InternalError,
"getOplogStartHack called with a NULL collection" );
// A query can only do oplog start finding if it has a top-level $gt or $gte predicate over
// the "ts" field (the operation's timestamp). Find that predicate and pass it to
// the OplogStart stage.
MatchExpression* tsExpr = NULL;
if (MatchExpression::AND == cq->root()->matchType()) {
// The query has an AND at the top-level. See if any of the children
// of the AND are $gt or $gte predicates over 'ts'.
for (size_t i = 0; i < cq->root()->numChildren(); ++i) {
MatchExpression* me = cq->root()->getChild(i);
if (isOplogTsPred(me)) {
tsExpr = me;
break;
}
}
}
else if (isOplogTsPred(cq->root())) {
// The root of the tree is a $gt or $gte predicate over 'ts'.
tsExpr = cq->root();
}
if (NULL == tsExpr) {
return Status(ErrorCodes::OplogOperationUnsupported,
"OplogReplay query does not contain top-level "
"$gt or $gte over the 'ts' field.");
}
DiskLoc startLoc = DiskLoc().setInvalid();
// See if the RecordStore supports the oplogStartHack
const BSONElement tsElem = extractOplogTsOptime(tsExpr);
if (tsElem.type() == Timestamp) {
StatusWith<DiskLoc> goal = oploghack::keyForOptime(tsElem._opTime());
if (goal.isOK()) {
startLoc = collection->getRecordStore()->oplogStartHack(txn, goal.getValue());
}
}
if (startLoc.isValid()) {
LOG(3) << "Using direct oplog seek";
}
else {
LOG(3) << "Using OplogStart stage";
// Fallback to trying the OplogStart stage.
WorkingSet* oplogws = new WorkingSet();
OplogStart* stage = new OplogStart(txn, collection, tsExpr, oplogws);
PlanExecutor* rawExec;
// Takes ownership of oplogws and stage.
Status execStatus = PlanExecutor::make(txn, oplogws, stage, collection,
PlanExecutor::YIELD_AUTO, &rawExec);
invariant(execStatus.isOK());
scoped_ptr<PlanExecutor> exec(rawExec);
// The stage returns a DiskLoc of where to start.
PlanExecutor::ExecState state = exec->getNext(NULL, &startLoc);
// This is normal. The start of the oplog is the beginning of the collection.
if (PlanExecutor::IS_EOF == state) {
return getExecutor(txn, collection, autoCq.release(), PlanExecutor::YIELD_AUTO,
execOut);
}
// This is not normal. An error was encountered.
if (PlanExecutor::ADVANCED != state) {
return Status(ErrorCodes::InternalError,
"quick oplog start location had error...?");
}
}
// cout << "diskloc is " << startLoc.toString() << endl;
// Build our collection scan...
CollectionScanParams params;
params.collection = collection;
params.start = startLoc;
params.direction = CollectionScanParams::FORWARD;
params.tailable = cq->getParsed().getOptions().tailable;
WorkingSet* ws = new WorkingSet();
CollectionScan* cs = new CollectionScan(txn, params, ws, cq->root());
// Takes ownership of 'ws', 'cs', and 'cq'.
return PlanExecutor::make(txn, ws, cs, autoCq.release(), collection,
PlanExecutor::YIELD_AUTO, execOut);
}
Status getOplogStartHack(Collection* collection, CanonicalQuery* cq, Runner** runnerOut) {
if ( collection == NULL )
return Status(ErrorCodes::InternalError,
"getOplogStartHack called with a NULL collection" );
// A query can only do oplog start finding if it has a top-level $gt or $gte predicate over
// the "ts" field (the operation's timestamp). Find that predicate and pass it to
// the OplogStart stage.
MatchExpression* tsExpr = NULL;
if (MatchExpression::AND == cq->root()->matchType()) {
// The query has an AND at the top-level. See if any of the children
// of the AND are $gt or $gte predicates over 'ts'.
for (size_t i = 0; i < cq->root()->numChildren(); ++i) {
MatchExpression* me = cq->root()->getChild(i);
if (isOplogTsPred(me)) {
tsExpr = me;
break;
}
}
}
else if (isOplogTsPred(cq->root())) {
// The root of the tree is a $gt or $gte predicate over 'ts'.
tsExpr = cq->root();
}
if (NULL == tsExpr) {
return Status(ErrorCodes::OplogOperationUnsupported,
"OplogReplay query does not contain top-level "
"$gt or $gte over the 'ts' field.");
}
// Make an oplog start finding stage.
WorkingSet* oplogws = new WorkingSet();
OplogStart* stage = new OplogStart(cq->ns(), tsExpr, oplogws);
// Takes ownership of ws and stage.
auto_ptr<InternalRunner> runner(new InternalRunner(collection, stage, oplogws));
runner->setYieldPolicy(Runner::YIELD_AUTO);
// The stage returns a DiskLoc of where to start.
DiskLoc startLoc;
Runner::RunnerState state = runner->getNext(NULL, &startLoc);
// This is normal. The start of the oplog is the beginning of the collection.
if (Runner::RUNNER_EOF == state) { return getRunner(cq, runnerOut); }
// This is not normal. An error was encountered.
if (Runner::RUNNER_ADVANCED != state) {
return Status(ErrorCodes::InternalError,
"quick oplog start location had error...?");
}
// cout << "diskloc is " << startLoc.toString() << endl;
// Build our collection scan...
CollectionScanParams params;
params.ns = cq->ns();
params.start = startLoc;
params.direction = CollectionScanParams::FORWARD;
params.tailable = cq->getParsed().hasOption(QueryOption_CursorTailable);
WorkingSet* ws = new WorkingSet();
CollectionScan* cs = new CollectionScan(params, ws, cq->root());
// Takes ownership of cq, cs, ws.
*runnerOut = new SingleSolutionRunner(collection, cq, NULL, cs, ws);
return Status::OK();
}
