void addOp(const string& op, BSONObj o, BSONObj* o2 = NULL, const char* coll = NULL,
int version = 0) {
OpTime ts(getNextGlobalOptime());
BSONObjBuilder b;
b.appendTimestamp("ts", ts.asLL());
if (version != 0) {
b.append("v", version);
}
b.append("op", op);
b.append("o", o);
if (o2) {
b.append("o2", *o2);
}
if (coll) {
b.append("ns", coll);
}
else {
b.append("ns", ns());
}
_bgsync->addDoc(b.done());
}
static void _logOpOld(OperationContext* txn,
const char *opstr,
const char *ns,
const char *logNS,
const BSONObj& obj,
BSONObj *o2,
bool *bb,
bool fromMigrate ) {
Lock::DBWrite lk(txn->lockState(), "local");
WriteUnitOfWork wunit(txn);
static BufBuilder bufbuilder(8*1024); // todo there is likely a mutex on this constructor
if ( strncmp(ns, "local.", 6) == 0 ) {
if ( strncmp(ns, "local.slaves", 12) == 0 ) {
resetSlaveCache();
}
return;
}
mutex::scoped_lock lk2(newOpMutex);
OpTime ts(getNextGlobalOptime());
newOptimeNotifier.notify_all();
/* we jump through a bunch of hoops here to avoid copying the obj buffer twice --
instead we do a single copy to the destination position in the memory mapped file.
*/
bufbuilder.reset();
BSONObjBuilder b(bufbuilder);
b.appendTimestamp("ts", ts.asDate());
b.append("op", opstr);
b.append("ns", ns);
if (fromMigrate)
b.appendBool("fromMigrate", true);
if ( bb )
b.appendBool("b", *bb);
if ( o2 )
b.append("o2", *o2);
BSONObj partial = b.done(); // partial is everything except the o:... part.
if( logNS == 0 ) {
logNS = "local.oplog.$main";
}
if ( localOplogMainCollection == 0 ) {
Client::Context ctx(txn, logNS);
localDB = ctx.db();
verify( localDB );
localOplogMainCollection = localDB->getCollection(txn, logNS);
verify( localOplogMainCollection );
}
Client::Context ctx(txn, logNS , localDB);
OplogDocWriter writer( partial, obj );
checkOplogInsert( localOplogMainCollection->insertDocument( txn, &writer, false ) );
ctx.getClient()->setLastOp( ts );
wunit.commit();
}
Status ModifierCurrentDate::apply() const {
const bool destExists = (_preparedState->elemFound.ok() &&
_preparedState->idxFound == (_updatePath.numParts() - 1));
mutablebson::Document& doc = _preparedState->doc;
StringData lastPart = _updatePath.getPart(_updatePath.numParts() - 1);
// If the element exists and is the same type, then that is what we want to work with
mutablebson::Element elemToSet = destExists ? _preparedState->elemFound : doc.end();
if (!destExists) {
// Creates the final element that's going to be $set in 'doc'.
// fills in the value with place-holder/empty
elemToSet = _typeIsDate ? doc.makeElementDate(lastPart, Date_t())
: doc.makeElementTimestamp(lastPart, OpTime());
if (!elemToSet.ok()) {
return Status(ErrorCodes::InternalError, "can't create new element");
}
// Now, we can be in two cases here, as far as attaching the element being set goes:
// (a) none of the parts in the element's path exist, or (b) some parts of the path
// exist but not all.
if (!_preparedState->elemFound.ok()) {
_preparedState->elemFound = doc.root();
_preparedState->idxFound = 0;
} else {
_preparedState->idxFound++;
}
// createPathAt() will complete the path and attach 'elemToSet' at the end of it.
Status s = pathsupport::createPathAt(
_updatePath, _preparedState->idxFound, _preparedState->elemFound, elemToSet);
if (!s.isOK())
return s;
}
dassert(elemToSet.ok());
// By the time we are here the element is in place and we just need to update the value
if (_typeIsDate) {
const mongo::Date_t now = mongo::jsTime();
Status s = elemToSet.setValueDate(now);
if (!s.isOK())
return s;
} else {
Status s = elemToSet.setValueTimestamp(getNextGlobalOptime());
if (!s.isOK())
return s;
}
// Set the elemFound, idxFound to the changed element for oplog logging.
_preparedState->elemFound = elemToSet;
_preparedState->idxFound = (_updatePath.numParts() - 1);
return Status::OK();
}
void insertSucceed() {
BSONObjBuilder b;
OpTime ts(getNextGlobalOptime());
b.appendTimestamp("ts", ts.asLL());
b.append("op", "i");
b.append("o", BSON("_id" << 123 << "x" << 456));
b.append("ns", cappedNs());
verify(apply(b.obj()));
}
void updateSucceed() {
BSONObjBuilder b;
OpTime ts(getNextGlobalOptime());
b.appendTimestamp("ts", ts.asLL());
b.append("op", "u");
b.append("o", BSON("$set" << BSON("x" << 789)));
b.append("o2", BSON("x" << 456));
b.append("ns", cappedNs());
verify(apply(b.obj()));
}
BSONObj updateFail() {
BSONObjBuilder b;
OpTime ts(getNextGlobalOptime());
b.appendTimestamp("ts", ts.asLL());
b.append("op", "u");
b.append("o", BSON("$set" << BSON("x" << 456)));
b.append("o2", BSON("_id" << 123 << "x" << 123));
b.append("ns", _cappedNs);
BSONObj o = b.obj();
verify(!apply(o));
return o;
}
void run() {
OpTime o(getNextGlobalOptime());
BSONObjBuilder b;
b.append("ns","dummy");
b.appendTimestamp("ts", o.asLL());
BSONObj obj = b.obj();
MockInitialSync mock;
// all three should succeed
std::vector<BSONObj> ops;
ops.push_back(obj);
repl::multiInitialSyncApply(ops, &mock);
mock.failOnStep = MockInitialSync::FAIL_FIRST_APPLY;
repl::multiInitialSyncApply(ops, &mock);
mock.retry = false;
repl::multiInitialSyncApply(ops, &mock);
drop();
}
void run() {
OpTime o(getNextGlobalOptime());
BSONObjBuilder b;
b.appendTimestamp("ts", o.asLL());
b.append("op", "u");
b.append("o", BSON("$set" << BSON("x" << 456)));
b.append("o2", BSON("_id" << 123));
b.append("ns", ns());
BSONObj obj = b.obj();
SyncTest2 sync2;
std::vector<BSONObj> ops;
ops.push_back(obj);
sync2.insertOnRetry = true;
// succeeds
multiInitialSyncApply(ops, &sync2);
BSONObj fin = findOne();
verify(fin["x"].Number() == 456);
drop();
}
static void _logOpRS(OperationContext* txn,
const char *opstr,
const char *ns,
const char *logNS,
const BSONObj& obj,
BSONObj *o2,
bool *bb,
bool fromMigrate ) {
Lock::DBWrite lk1(txn->lockState(), "local");
WriteUnitOfWork wunit(txn);
if ( strncmp(ns, "local.", 6) == 0 ) {
if ( strncmp(ns, "local.slaves", 12) == 0 )
resetSlaveCache();
return;
}
mutex::scoped_lock lk2(newOpMutex);
OpTime ts(getNextGlobalOptime());
newOptimeNotifier.notify_all();
long long hashNew;
if( theReplSet ) {
if (!theReplSet->box.getState().primary()) {
log() << "replSet error : logOp() but not primary";
fassertFailed(17405);
}
hashNew = (theReplSet->lastH * 131 + ts.asLL()) * 17 + theReplSet->selfId();
}
else {
// must be initiation
verify( *ns == 0 );
hashNew = 0;
}
/* we jump through a bunch of hoops here to avoid copying the obj buffer twice --
instead we do a single copy to the destination position in the memory mapped file.
*/
logopbufbuilder.reset();
BSONObjBuilder b(logopbufbuilder);
b.appendTimestamp("ts", ts.asDate());
b.append("h", hashNew);
b.append("v", OPLOG_VERSION);
b.append("op", opstr);
b.append("ns", ns);
if (fromMigrate)
b.appendBool("fromMigrate", true);
if ( bb )
b.appendBool("b", *bb);
if ( o2 )
b.append("o2", *o2);
BSONObj partial = b.done();
DEV verify( logNS == 0 ); // check this was never a master/slave master
if ( localOplogRSCollection == 0 ) {
Client::Context ctx(txn, rsoplog);
localDB = ctx.db();
verify( localDB );
localOplogRSCollection = localDB->getCollection( txn, rsoplog );
massert(13347, "local.oplog.rs missing. did you drop it? if so restart server", localOplogRSCollection);
}
Client::Context ctx(txn, rsoplog, localDB);
OplogDocWriter writer( partial, obj );
checkOplogInsert( localOplogRSCollection->insertDocument( txn, &writer, false ) );
/* todo: now() has code to handle clock skew. but if the skew server to server is large it will get unhappy.
this code (or code in now() maybe) should be improved.
*/
if( theReplSet ) {
if( !(theReplSet->lastOpTimeWritten<ts) ) {
log() << "replication oplog stream went back in time. previous timestamp: "
<< theReplSet->lastOpTimeWritten << " newest timestamp: " << ts
<< ". attempting to sync directly from primary." << endl;
BSONObjBuilder result;
Status status = theReplSet->forceSyncFrom(theReplSet->box.getPrimary()->fullName(),
&result);
if (!status.isOK()) {
log() << "Can't sync from primary: " << status;
}
}
theReplSet->lastOpTimeWritten = ts;
theReplSet->lastH = hashNew;
ctx.getClient()->setLastOp( ts );
}
wunit.commit();
}
void Consensus::_electSelf() {
if( time(0) < steppedDown )
return;
{
const OpTime ord = theReplSet->lastOpTimeWritten;
if( ord == 0 ) {
log() << "replSet info not trying to elect self, do not yet have a complete set of data from any point in time" << rsLog;
return;
}
}
bool allUp;
int nTies;
if( !_weAreFreshest(allUp, nTies) ) {
return;
}
rs.sethbmsg("",9);
if (!allUp && time(0) - serverGlobalParams.started < 60 * 5) {
/* the idea here is that if a bunch of nodes bounce all at once, we don't want to drop data
if we don't have to -- we'd rather be offline and wait a little longer instead
todo: make this configurable.
*/
rs.sethbmsg("not electing self, not all members up and we have been up less than 5 minutes");
return;
}
Member& me = *rs._self;
if( nTies ) {
/* tie? we then randomly sleep to try to not collide on our voting. */
/* todo: smarter. */
if( me.id() == 0 || _sleptLast ) {
// would be fine for one node not to sleep
// todo: biggest / highest priority nodes should be the ones that get to not sleep
}
else {
verify( !rs.lockedByMe() ); // bad to go to sleep locked
unsigned ms = ((unsigned) rand()) % 1000 + 50;
DEV log() << "replSet tie " << nTies << " sleeping a little " << ms << "ms" << rsLog;
_sleptLast = true;
sleepmillis(ms);
throw RetryAfterSleepException();
}
}
_sleptLast = false;
time_t start = time(0);
unsigned meid = me.id();
int tally = _yea( meid );
bool success = false;
try {
log() << "replSet info electSelf " << meid << rsLog;
BSONObj electCmd = BSON(
"replSetElect" << 1 <<
"set" << rs.name() <<
"who" << me.fullName() <<
"whoid" << me.hbinfo().id() <<
"cfgver" << rs._cfg->version <<
"round" << OID::gen() /* this is just for diagnostics */
);
int configVersion;
list<Target> L;
rs.getTargets(L, configVersion);
_multiCommand(electCmd, L);
{
for( list<Target>::iterator i = L.begin(); i != L.end(); i++ ) {
LOG(1) << "replSet elect res: " << i->result.toString() << rsLog;
if( i->ok ) {
int v = i->result["vote"].Int();
tally += v;
}
}
if( tally*2 <= _totalVotes() ) {
log() << "replSet couldn't elect self, only received " << tally << " votes" << rsLog;
}
else if( time(0) - start > 30 ) {
// defensive; should never happen as we have timeouts on connection and operation for our conn
log() << "replSet too much time passed during our election, ignoring result" << rsLog;
}
else if( configVersion != rs.config().version ) {
log() << "replSet config version changed during our election, ignoring result" << rsLog;
}
else {
/* succeeded. */
LOG(1) << "replSet election succeeded, assuming primary role" << rsLog;
success = true;
setElectionTime(getNextGlobalOptime());
rs.assumePrimary();
}
}
}
catch( std::exception& ) {
if( !success ) _electionFailed(meid);
throw;
}
if( !success ) _electionFailed(meid);
}
static void _logOpRS(OperationContext* txn,
const char *opstr,
const char *ns,
const char *logNS,
const BSONObj& obj,
BSONObj *o2,
bool *bb,
bool fromMigrate ) {
Lock::DBLock lk1(txn->lockState(), "local", newlm::MODE_X);
WriteUnitOfWork wunit(txn);
if ( strncmp(ns, "local.", 6) == 0 ) {
if ( strncmp(ns, "local.slaves", 12) == 0 )
resetSlaveCache();
return;
}
ReplicationCoordinator* replCoord = getGlobalReplicationCoordinator();
mutex::scoped_lock lk2(newOpMutex);
OpTime ts(getNextGlobalOptime());
newOptimeNotifier.notify_all();
long long hashNew = BackgroundSync::get()->getLastAppliedHash();
// Check to make sure logOp() is legal at this point.
if (*opstr == 'n') {
// 'n' operations are always logged
invariant(*ns == '\0');
// 'n' operations do not advance the hash, since they are not rolled back
}
else {
if (!replCoord->canAcceptWritesForDatabase(nsToDatabaseSubstring(ns))) {
severe() << "replSet error : logOp() but can't accept write to collection " << ns;
fassertFailed(17405);
}
// Advance the hash
hashNew = (hashNew * 131 + ts.asLL()) * 17 + replCoord->getMyId();
}
/* we jump through a bunch of hoops here to avoid copying the obj buffer twice --
instead we do a single copy to the destination position in the memory mapped file.
*/
logopbufbuilder.reset();
BSONObjBuilder b(logopbufbuilder);
b.appendTimestamp("ts", ts.asDate());
b.append("h", hashNew);
b.append("v", OPLOG_VERSION);
b.append("op", opstr);
b.append("ns", ns);
if (fromMigrate)
b.appendBool("fromMigrate", true);
if ( bb )
b.appendBool("b", *bb);
if ( o2 )
b.append("o2", *o2);
BSONObj partial = b.done();
DEV verify( logNS == 0 ); // check this was never a master/slave master
if ( localOplogRSCollection == 0 ) {
Client::Context ctx(txn, rsoplog);
localDB = ctx.db();
verify( localDB );
localOplogRSCollection = localDB->getCollection( txn, rsoplog );
massert(13347, "local.oplog.rs missing. did you drop it? if so restart server", localOplogRSCollection);
}
Client::Context ctx(txn, rsoplog, localDB);
OplogDocWriter writer( partial, obj );
checkOplogInsert( localOplogRSCollection->insertDocument( txn, &writer, false ) );
BackgroundSync::get()->setLastAppliedHash(hashNew);
ctx.getClient()->setLastOp( ts );
replCoord->setMyLastOptime(txn, ts);
wunit.commit();
}