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();
}
/** write an op to the oplog that is already built.
todo : make _logOpRS() call this so we don't repeat ourself?
*/
void _logOpObjRS(OperationContext* txn, const BSONObj& op) {
Lock::DBWrite lk(txn->lockState(), "local");
// XXX soon this needs to be part of an outer WUOW not its own.
// We can't do this yet due to locking limitations.
WriteUnitOfWork wunit(txn);
const OpTime ts = op["ts"]._opTime();
long long h = op["h"].numberLong();
{
if ( localOplogRSCollection == 0 ) {
Client::Context ctx(txn, rsoplog);
localDB = ctx.db();
verify( localDB );
localOplogRSCollection = localDB->getCollection(txn, rsoplog);
massert(13389,
"local.oplog.rs missing. did you drop it? if so restart server",
localOplogRSCollection);
}
Client::Context ctx(txn, rsoplog, localDB);
checkOplogInsert(localOplogRSCollection->insertDocument(txn, op, 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 = h;
ctx.getClient()->setLastOp( ts );
BackgroundSync::notify();
}
}
setNewOptime(ts);
wunit.commit();
}
/** write an op to the oplog that is already built.
todo : make _logOpRS() call this so we don't repeat ourself?
*/
void _logOpObjRS(const BSONObj& op) {
OperationContextImpl txn;
Lock::DBWrite lk(txn.lockState(), "local");
const OpTime ts = op["ts"]._opTime();
long long h = op["h"].numberLong();
{
if ( localOplogRSCollection == 0 ) {
Client::Context ctx(rsoplog, storageGlobalParams.dbpath);
localDB = ctx.db();
verify( localDB );
localOplogRSCollection = localDB->getCollection( &txn, rsoplog );
massert(13389,
"local.oplog.rs missing. did you drop it? if so restart server",
localOplogRSCollection);
}
Client::Context ctx(rsoplog, localDB);
checkOplogInsert( localOplogRSCollection->insertDocument( &txn, op, 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;
std::string errmsg;
BSONObjBuilder result;
if (!theReplSet->forceSyncFrom(theReplSet->box.getPrimary()->fullName(),
errmsg, result)) {
log() << "Can't sync from primary: " << errmsg << endl;
}
}
theReplSet->lastOpTimeWritten = ts;
theReplSet->lastH = h;
ctx.getClient()->setLastOp( ts );
BackgroundSync::notify();
}
}
setNewOptime(ts);
}
/** write an op to the oplog that is already built.
todo : make _logOpRS() call this so we don't repeat ourself?
*/
OpTime _logOpObjRS(OperationContext* txn, const BSONObj& op) {
Lock::DBLock lk(txn->lockState(), "local", newlm::MODE_X);
// XXX soon this needs to be part of an outer WUOW not its own.
// We can't do this yet due to locking limitations.
WriteUnitOfWork wunit(txn);
const OpTime ts = op["ts"]._opTime();
long long hash = op["h"].numberLong();
{
if ( localOplogRSCollection == 0 ) {
Client::Context ctx(txn, rsoplog);
localDB = ctx.db();
verify( localDB );
localOplogRSCollection = localDB->getCollection(txn, rsoplog);
massert(13389,
"local.oplog.rs missing. did you drop it? if so restart server",
localOplogRSCollection);
}
Client::Context ctx(txn, rsoplog, localDB);
checkOplogInsert(localOplogRSCollection->insertDocument(txn, op, false));
ReplicationCoordinator* replCoord = getGlobalReplicationCoordinator();
OpTime myLastOptime = replCoord->getMyLastOptime();
if (!(myLastOptime < ts)) {
severe() << "replication oplog stream went back in time. previous timestamp: "
<< myLastOptime << " newest timestamp: " << ts;
fassertFailedNoTrace(18905);
}
BackgroundSync* bgsync = BackgroundSync::get();
// Keep this up-to-date, in case we step up to primary.
bgsync->setLastAppliedHash(hash);
ctx.getClient()->setLastOp( ts );
replCoord->setMyLastOptime(txn, ts);
bgsync->notify();
}
setNewOptime(ts);
wunit.commit();
return ts;
}
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();
}
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();
}
