Shard Shard::pick( const Shard& current ) {
vector<Shard> all;
staticShardInfo.getAllShards( all );
if ( all.size() == 0 ) {
staticShardInfo.reload();
staticShardInfo.getAllShards( all );
if ( all.size() == 0 )
return EMPTY;
}
// if current shard was provided, pick a different shard only if it is a better choice
ShardStatus best = all[0].getStatus();
if ( current != EMPTY ) {
best = current.getStatus();
}
for ( size_t i=0; i<all.size(); i++ ) {
ShardStatus t = all[i].getStatus();
if ( t < best )
best = t;
}
LOG(1) << "best shard for new allocation is " << best << endl;
return best.shard();
}
void DistributionStatus::populateShardInfoMap(const vector<Shard> allShards,
ShardInfoMap* shardInfo) {
for (vector<Shard>::const_iterator it = allShards.begin();
it != allShards.end(); ++it ) {
const Shard& shard = *it;
ShardStatus status = shard.getStatus();
shardInfo->insert(make_pair(shard.getName(),
ShardInfo(shard.getMaxSize(),
status.mapped(),
shard.isDraining(),
status.hasOpsQueued(),
shard.tags(),
status.mongoVersion())));
}
}
Shard Shard::pick(){
vector<Shard> all;
staticShardInfo.getAllShards( all );
if ( all.size() == 0 ){
staticShardInfo.reload();
staticShardInfo.getAllShards( all );
if ( all.size() == 0 )
return EMPTY;
}
ShardStatus best = all[0].getStatus();
for ( size_t i=1; i<all.size(); i++ ){
ShardStatus t = all[i].getStatus();
if ( t < best )
best = t;
}
log(1) << "picking shard: " << best << endl;
return best.shard();
}
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 Balancer::_doBalanceRound( DBClientBase& conn, vector<CandidateChunkPtr>* candidateChunks ){
assert( candidateChunks );
//
// 1. Check whether there is any sharded collection to be balanced by querying
// the ShardsNS::database collection
//
// { "_id" : "test", "partitioned" : true, "primary" : "shard0",
// "sharded" : {
// "test.images" : { "key" : { "_id" : 1 }, "unique" : false },
// ...
// }
// }
//
auto_ptr<DBClientCursor> cursor = conn.query( ShardNS::database , BSON( "partitioned" << true ) );
vector< string > collections;
while ( cursor->more() ){
BSONObj db = cursor->next();
// A database may be partitioned but not yet have a sharded collection.
// 'cursor' will point to docs that do not contain the "sharded" key. Since
// there'd be nothing to balance, we want to skip those here.
BSONElement shardedColls = db["sharded"];
if ( shardedColls.eoo() ){
log(2) << "balancer: skipping database with no sharded collection ("
<< db["_id"].str() << ")" << endl;
continue;
}
BSONObjIterator i( shardedColls.Obj() );
while ( i.more() ){
BSONElement e = i.next();
collections.push_back( e.fieldName() );
}
}
cursor.reset();
if ( collections.empty() ) {
log(1) << "balancer: 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) << "balancer: can't balance without more active shards" << endl;
return;
}
map< string, BSONObj > shardLimitsMap;
for ( vector<Shard>::const_iterator it = allShards.begin(); it != allShards.end(); ++it ){
const Shard& s = *it;
ShardStatus status = s.getStatus();
BSONObj limitsObj = BSON( ShardFields::maxSize( s.getMaxSize() ) <<
ShardFields::currSize( status.mapped() ) <<
ShardFields::draining( s.isDraining()) );
shardLimitsMap[ s.getName() ] = limitsObj;
}
//
// 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->next();
vector<BSONObj>& chunks = shardToChunksMap[chunk["shard"].String()];
chunks.push_back( chunk.getOwned() );
}
cursor.reset();
if (shardToChunksMap.empty()) {
log(1) << "balancer: 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();
}
CandidateChunk* p = _policy->balance( ns , shardLimitsMap , shardToChunksMap , _balancedLastTime );
if ( p ) candidateChunks->push_back( CandidateChunkPtr( p ) );
}
}
bool ShardStatus::operator<(const ShardStatus& other) const {
return dataSizeBytes() < other.dataSizeBytes();
}
void Balancer::_doBalanceRound( DBClientBase& conn, vector<CandidateChunkPtr>* candidateChunks ) {
assert( 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() )
collections.push_back( col["_id"].String() );
}
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;
}
map< string, BSONObj > shardLimitsMap;
for ( vector<Shard>::const_iterator it = allShards.begin(); it != allShards.end(); ++it ) {
const Shard& s = *it;
ShardStatus status = s.getStatus();
BSONObj limitsObj = BSON( ShardFields::maxSize( s.getMaxSize() ) <<
LimitsFields::currSize( status.mapped() ) <<
ShardFields::draining( s.isDraining() ) <<
LimitsFields::hasOpsQueued( status.hasOpsQueued() )
);
shardLimitsMap[ s.getName() ] = limitsObj;
}
//
// 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();
}
CandidateChunk* p = _policy->balance( ns , shardLimitsMap , shardToChunksMap , _balancedLastTime );
if ( p ) candidateChunks->push_back( CandidateChunkPtr( p ) );
}
}
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(CollectionType::ConfigNS, BSONObj());
vector< string > collections;
while ( cursor->more() ) {
BSONObj col = cursor->nextSafe();
// sharded collections will have a shard "key".
if ( ! col[CollectionType::keyPattern()].eoo() &&
! col[CollectionType::noBalance()].trueValue() ){
collections.push_back( col[CollectionType::ns()].String() );
}
else if( col[CollectionType::noBalance()].trueValue() ){
LOG(1) << "not balancing collection " << col[CollectionType::ns()].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(ChunkType::ConfigNS,
QUERY(ChunkType::ns(ns)).sort(ChunkType::min()));
set<BSONObj> allChunkMinimums;
while ( cursor->more() ) {
BSONObj chunk = cursor->nextSafe().getOwned();
vector<BSONObj>& chunks = shardToChunksMap[chunk[ChunkType::shard()].String()];
allChunkMinimums.insert( chunk[ChunkType::min()].Obj() );
chunks.push_back( chunk );
}
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(TagsType::ConfigNS,
BSON(TagsType::ns() << 1 << TagsType::min() << 1),
true);
cursor = conn.query(TagsType::ConfigNS,
QUERY(TagsType::ns(ns)).sort(TagsType::min()));
vector<TagRange> ranges;
while ( cursor->more() ) {
BSONObj tag = cursor->nextSafe();
TagRange tr(tag[TagsType::min()].Obj().getOwned(),
tag[TagsType::max()].Obj().getOwned(),
tag[TagsType::tag()].String());
ranges.push_back(tr);
uassert(16356,
str::stream() << "tag ranges not valid for: " << ns,
status.addTagRange(tr) );
}
cursor.reset();
DBConfigPtr cfg = grid.getDBConfig( ns );
verify( cfg );
ChunkManagerPtr cm = cfg->getChunkManager( ns );
verify( cm );
// loop through tags to make sure no chunk spans tags; splits on tag min. for all chunks
bool didAnySplits = false;
for ( unsigned i = 0; i < ranges.size(); i++ ) {
BSONObj min = ranges[i].min;
min = cm->getShardKey().extendRangeBound( min, false );
if ( allChunkMinimums.count( min ) > 0 )
continue;
didAnySplits = true;
log() << "ns: " << ns << " need to split on "
<< min << " because there is a range there" << endl;
ChunkPtr c = cm->findIntersectingChunk( min );
vector<BSONObj> splitPoints;
splitPoints.push_back( min );
BSONObj res;
if ( !c->multiSplit( splitPoints, res ) ) {
error() << "split failed: " << res << endl;
}
else {
LOG(1) << "split worked: " << res << endl;
}
break;
}
if ( didAnySplits ) {
// state change, just wait till next round
continue;
}
CandidateChunk* p = _policy->balance( ns, status, _balancedLastTime );
if ( p ) candidateChunks->push_back( CandidateChunkPtr( p ) );
}
}