void SysParamInit( void ) //system param init
{
char* pdst = NULL;
int iRet;
/* BEGIN: Added by wupm, 2013/3/12 */
#ifdef HONESTECH
ReadHTClassParams();
#endif
InitSystemParam();
Textout("Read Config File , UUID = [%s], MAC = [%s]",
bparam.stIEBaseParam.dwDeviceID,
bparam.stIEBaseParam.szMac );
Textout("ApiLisense = [%s]",bparam.stIEBaseParam.dwApiLisense);
nvram_init( RT2860_NVRAM );
IPCStart(); //ipc init
#ifndef NOT_READ_MAC
ReadFactoryMac(); //read mac
Textout("ReadFactoryMac, UUID = [%s], MAC = [%s]",
bparam.stIEBaseParam.dwDeviceID,
bparam.stIEBaseParam.szMac );
Textout("ApiLisense = [%s]",bparam.stIEBaseParam.dwApiLisense);
#endif
ReadRactoryID(); //read deviceid
Textout("ReadRactoryID, UUID = [%s], MAC = [%s]",
bparam.stIEBaseParam.dwDeviceID,
bparam.stIEBaseParam.szMac );
Textout("ApiLisense = [%s]",bparam.stIEBaseParam.dwApiLisense);
//ReadDnsFactory(); //read dns factory
ConfigVersion(); //config system version
DateInit();
FactoryUpdateInit();
}
const ConfigVersion ShardedConnectionInfo::getVersion( const string& ns ) const {
NSVersionMap::const_iterator it = _versions.find( ns );
if ( it != _versions.end() ) {
return it->second;
}
else {
return ConfigVersion( 0, OID() );
}
}
const ConfigVersion ShardingState::getVersion( const string& ns ) const {
scoped_lock lk(_mutex);
ChunkManagersMap::const_iterator it = _chunks.find( ns );
if ( it != _chunks.end() ) {
ShardChunkManagerPtr p = it->second;
return p->getVersion();
}
else {
return ConfigVersion( 0, OID() );
}
}
bool run(const string& , BSONObj& cmdObj, int, string& errmsg, BSONObjBuilder& result, bool) {
// Steps
// 1. check basic config
// 2. extract params from command
// 3. fast check
// 4. slow check (LOCKS)
// step 1
lastError.disableForCommand();
ShardedConnectionInfo* info = ShardedConnectionInfo::get( true );
// make sure we have the mongos id for writebacks
if ( ! checkMongosID( info , cmdObj["serverID"] , errmsg ) )
return false;
bool authoritative = cmdObj.getBoolField( "authoritative" );
// check config server is ok or enable sharding
if ( ! checkConfigOrInit( cmdObj["configdb"].valuestrsafe() , authoritative , errmsg , result ) )
return false;
// check shard name/hosts are correct
if ( cmdObj["shard"].type() == String ) {
shardingState.gotShardName( cmdObj["shard"].String() );
shardingState.gotShardHost( cmdObj["shardHost"].String() );
}
// Handle initial shard connection
if( cmdObj["version"].eoo() && cmdObj["init"].trueValue() ) {
result.append( "initialized", true );
return true;
}
// we can run on a slave up to here
if ( ! isMaster( "admin" ) ) {
result.append( "errmsg" , "not master" );
result.append( "note" , "from post init in setShardVersion" );
return false;
}
// step 2
string ns = cmdObj["setShardVersion"].valuestrsafe();
if ( ns.size() == 0 ) {
errmsg = "need to specify namespace";
return false;
}
const ConfigVersion version = ConfigVersion( extractVersion( cmdObj["version"] , errmsg ), OID() );
if ( errmsg.size() )
return false;
// step 3
const ConfigVersion oldVersion = info->getVersion(ns);
const ConfigVersion globalVersion = shardingState.getVersion(ns);
oldVersion.addToBSON( result, "oldVersion" );
if ( globalVersion.isSet() && version.isSet() ) {
// this means there is no reset going on an either side
// so its safe to make some assumptions
if ( version.isEquivalentTo( globalVersion ) ) {
// mongos and mongod agree!
if ( ! oldVersion.isEquivalentTo( version ) ) {
if ( oldVersion < globalVersion ) {
info->setVersion( ns , version );
}
else if ( authoritative ) {
// this means there was a drop and our version is reset
info->setVersion( ns , version );
}
else {
result.append( "ns" , ns );
result.appendBool( "need_authoritative" , true );
errmsg = "verifying drop on '" + ns + "'";
return false;
}
}
return true;
}
}
// step 4
// this is because of a weird segfault I saw and I can't see why this should ever be set
massert( 13647 , str::stream() << "context should be empty here, is: " << cc().getContext()->ns() , cc().getContext() == 0 );
Lock::GlobalWrite setShardVersionLock; // TODO: can we get rid of this??
if ( oldVersion.isSet() && ! globalVersion.isSet() ) {
// this had been reset
info->setVersion( ns , ShardChunkVersion( 0, OID() ) );
}
if ( ! version.isSet() && ! globalVersion.isSet() ) {
// this connection is cleaning itself
info->setVersion( ns , ShardChunkVersion( 0, OID() ) );
return true;
}
if ( ! version.isSet() && globalVersion.isSet() ) {
if ( ! authoritative ) {
result.appendBool( "need_authoritative" , true );
result.append( "ns" , ns );
globalVersion.addToBSON( result, "globalVersion" );
errmsg = "dropping needs to be authoritative";
return false;
}
log() << "wiping data for: " << ns << endl;
globalVersion.addToBSON( result, "beforeDrop" );
// only setting global version on purpose
// need clients to re-find meta-data
shardingState.resetVersion( ns );
info->setVersion( ns , ShardChunkVersion( 0, OID() ) );
return true;
}
if ( version < oldVersion ) {
errmsg = "this connection already had a newer version of collection '" + ns + "'";
result.append( "ns" , ns );
version.addToBSON( result, "newVersion" );
globalVersion.addToBSON( result, "globalVersion" );
return false;
}
if ( version < globalVersion ) {
while ( shardingState.inCriticalMigrateSection() ) {
dbtemprelease r;
sleepmillis(2);
OCCASIONALLY log() << "waiting till out of critical section" << endl;
}
errmsg = "shard global version for collection is higher than trying to set to '" + ns + "'";
result.append( "ns" , ns );
version.addToBSON( result, "version" );
globalVersion.addToBSON( result, "globalVersion" );
result.appendBool( "reloadConfig" , true );
return false;
}
if ( ! globalVersion.isSet() && ! authoritative ) {
// Needed b/c when the last chunk is moved off a shard, the version gets reset to zero, which
// should require a reload.
// TODO: Maybe a more elegant way of doing this
while ( shardingState.inCriticalMigrateSection() ) {
dbtemprelease r;
sleepmillis(2);
OCCASIONALLY log() << "waiting till out of critical section for version reset" << endl;
}
// need authoritative for first look
result.append( "ns" , ns );
result.appendBool( "need_authoritative" , true );
errmsg = "first time for collection '" + ns + "'";
return false;
}
Timer relockTime;
{
dbtemprelease unlock;
ShardChunkVersion currVersion = version;
if ( ! shardingState.trySetVersion( ns , currVersion ) ) {
errmsg = str::stream() << "client version differs from config's for collection '" << ns << "'";
result.append( "ns" , ns );
version.addToBSON( result, "version" );
globalVersion.addToBSON( result, "globalVersion" );
return false;
}
}
if ( relockTime.millis() >= ( cmdLine.slowMS - 10 ) ) {
log() << "setShardVersion - relocking slow: " << relockTime.millis() << endl;
}
info->setVersion( ns , version );
return true;
}
bool ShardingState::trySetVersion( const string& ns , ConfigVersion& version /* IN-OUT */ ) {
// Currently this function is called after a getVersion(), which is the first "check", and the assumption here
// is that we don't do anything nearly as long as a remote query in a thread between then and now.
// Otherwise it may be worth adding an additional check without the _configServerMutex below, since then it
// would be likely that the version may have changed in the meantime without waiting for or fetching config results.
// TODO: Mutex-per-namespace?
LOG( 2 ) << "trying to set shard version of " << version.toString() << " for '" << ns << "'" << endl;
_configServerTickets.waitForTicket();
TicketHolderReleaser needTicketFrom( &_configServerTickets );
// fast path - double-check if requested version is at the same version as this chunk manager before verifying
// against config server
//
// This path will short-circuit the version set if another thread already managed to update the version in the
// meantime. First check is from getVersion().
//
// cases:
// + this shard updated the version for a migrate's commit (FROM side)
// a client reloaded chunk state from config and picked the newest version
// + two clients reloaded
// one triggered the 'slow path' (below)
// when the second's request gets here, the version is already current
ConfigVersion storedVersion;
ShardChunkManagerPtr currManager;
{
scoped_lock lk( _mutex );
ChunkManagersMap::const_iterator it = _chunks.find( ns );
if( it == _chunks.end() ){
// TODO: We need better semantic distinction between *no manager found* and
// *manager of version zero found*
log() << "no current chunk manager found for this shard, will initialize" << endl;
}
else{
currManager = it->second;
if( ( storedVersion = it->second->getVersion() ).isEquivalentTo( version ) )
return true;
}
}
LOG( 2 ) << "verifying cached version " << storedVersion.toString() << " and new version " << version.toString() << " for '" << ns << "'" << endl;
// slow path - requested version is different than the current chunk manager's, if one exists, so must check for
// newest version in the config server
//
// cases:
// + a chunk moved TO here
// (we don't bump up the version on the TO side but the commit to config does use higher version)
// a client reloads from config an issued the request
// + there was a take over from a secondary
// the secondary had no state (managers) at all, so every client request will fall here
// + a stale client request a version that's not current anymore
// Can't lock default mutex while creating ShardChunkManager, b/c may have to create a new connection to myself
const string c = (_configServer == _shardHost) ? "" /* local */ : _configServer;
// If our epochs aren't compatible, it's not useful to use the old manager for chunk diffs
if( currManager && ! currManager->getCollVersion().hasCompatibleEpoch( version ) ){
warning() << "detected incompatible version epoch in new version " << version
<< ", old version was " << currManager->getCollVersion() << endl;
currManager.reset();
}
ShardChunkManagerPtr p( ShardChunkManager::make( c , ns , _shardName, currManager ) );
// Handle the case where the collection isn't sharded more gracefully
if( p->getKey().isEmpty() ){
version = ConfigVersion( 0, OID() );
// There was an error getting any data for this collection, return false
return false;
}
{
// NOTE: This lock prevents the ns version from changing while a write operation occurs.
Lock::DBRead readLk(ns);
// This lock prevents simultaneous metadata changes using the same map
scoped_lock lk( _mutex );
// since we loaded the chunk manager unlocked, other thread may have done the same
// make sure we keep the freshest config info only
ChunkManagersMap::const_iterator it = _chunks.find( ns );
if ( it == _chunks.end() || p->getVersion() >= it->second->getVersion() ) {
_chunks[ns] = p;
}
ChunkVersion oldVersion = version;
version = p->getVersion();
return oldVersion.isEquivalentTo( version );
}
}
bool _handlePossibleShardedMessage( Message &m, DbResponse* dbresponse ) {
DEV verify( shardingState.enabled() );
int op = m.operation();
if ( op < 2000
|| op >= 3000
|| op == dbGetMore // cursors are weird
)
return false;
DbMessage d(m);
const char *ns = d.getns();
string errmsg;
// We don't care about the version here, since we're returning it later in the writeback
ConfigVersion received, wanted;
if ( shardVersionOk( ns , errmsg, received, wanted ) ) {
return false;
}
LOG(1) << "connection meta data too old - will retry ns:(" << ns << ") op:(" << opToString(op) << ") " << errmsg << endl;
if ( doesOpGetAResponse( op ) ) {
verify( dbresponse );
BufBuilder b( 32768 );
b.skip( sizeof( QueryResult ) );
{
BSONObj obj = BSON( "$err" << errmsg << "ns" << ns );
b.appendBuf( obj.objdata() , obj.objsize() );
}
QueryResult *qr = (QueryResult*)b.buf();
qr->_resultFlags() = ResultFlag_ErrSet | ResultFlag_ShardConfigStale;
qr->len = b.len();
qr->setOperation( opReply );
qr->cursorId = 0;
qr->startingFrom = 0;
qr->nReturned = 1;
b.decouple();
Message * resp = new Message();
resp->setData( qr , true );
dbresponse->response = resp;
dbresponse->responseTo = m.header()->id;
return true;
}
uassert( 9517 , "writeback" , ( d.reservedField() & DbMessage::Reserved_FromWriteback ) == 0 );
OID writebackID;
writebackID.init();
lastError.getSafe()->writeback( writebackID );
const OID& clientID = ShardedConnectionInfo::get(false)->getID();
massert( 10422 , "write with bad shard config and no server id!" , clientID.isSet() );
LOG(1) << "got write with an old config - writing back ns: " << ns << endl;
LOG(1) << m.toString() << endl;
BSONObjBuilder b;
b.appendBool( "writeBack" , true );
b.append( "ns" , ns );
b.append( "id" , writebackID );
b.append( "connectionId" , cc().getConnectionId() );
b.append( "instanceIdent" , prettyHostName() );
shardingState.getVersion( ns ).addToBSON( b );
ShardedConnectionInfo* info = ShardedConnectionInfo::get( false );
( info ? info->getVersion(ns) : ConfigVersion( 0, OID() ) ).addToBSON( b, "yourVersion" );
b.appendBinData( "msg" , m.header()->len , bdtCustom , (char*)(m.singleData()) );
LOG(2) << "writing back msg with len: " << m.header()->len << " op: " << m.operation() << endl;
writeBackManager.queueWriteBack( clientID.str() , b.obj() );
return true;
}
