summaryrefslogtreecommitdiff
path: root/s/chunk.cpp
diff options
context:
space:
mode:
Diffstat (limited to 's/chunk.cpp')
-rw-r--r--s/chunk.cpp114
1 files changed, 68 insertions, 46 deletions
diff --git a/s/chunk.cpp b/s/chunk.cpp
index 09dc994..066265e 100644
--- a/s/chunk.cpp
+++ b/s/chunk.cpp
@@ -208,7 +208,7 @@ namespace mongo {
// no split points means there isn't enough data to split on
// 1 split point means we have between half the chunk size to full chunk size
// so we shouldn't split
- LOG(1) << "chunk not full enough to trigger auto-split" << endl;
+ LOG(1) << "chunk not full enough to trigger auto-split " << ( candidates.size() == 0 ? "no split entry" : candidates[0].toString() ) << endl;
return BSONObj();
}
@@ -342,6 +342,12 @@ namespace mongo {
if ( _dataWritten < splitThreshold / 5 )
return false;
+
+ if ( ! getManager()->_splitTickets.tryAcquire() ) {
+ LOG(1) << "won't auto split becaue not enough tickets: " << getManager()->getns() << endl;
+ return false;
+ }
+ TicketHolderReleaser releaser( &getManager()->_splitTickets );
// this is a bit ugly
// we need it so that mongos blocks for the writes to actually be committed
@@ -352,8 +358,6 @@ namespace mongo {
LOG(1) << "about to initiate autosplit: " << *this << " dataWritten: " << _dataWritten << " splitThreshold: " << splitThreshold << endl;
- _dataWritten = 0; // reset so we check often enough
-
BSONObj res;
BSONObj splitPoint = singleSplit( false /* does not force a split if not enough data */ , res );
if ( splitPoint.isEmpty() ) {
@@ -361,6 +365,15 @@ namespace mongo {
_dataWritten = 0; // this means there wasn't enough data to split, so don't want to try again until considerable more data
return false;
}
+
+ if ( maxIsInf() || minIsInf() ) {
+ // we don't want to reset _dataWritten since we kind of want to check the other side right away
+ }
+ else {
+ _dataWritten = 0; // we're splitting, so should wait a bit
+ }
+
+
log() << "autosplitted " << _manager->getns() << " shard: " << toString()
<< " on: " << splitPoint << "(splitThreshold " << splitThreshold << ")"
@@ -497,7 +510,9 @@ namespace mongo {
// The shard versioning mechanism hinges on keeping track of the number of times we reloaded ChunkManager's.
// Increasing this number here will prompt checkShardVersion() to refresh the connection-level versions to
// the most up to date value.
- _sequenceNumber(++NextSequenceNumber)
+ _sequenceNumber(++NextSequenceNumber),
+
+ _splitTickets( 5 )
{
int tries = 3;
@@ -611,44 +626,65 @@ namespace mongo {
return _key.hasShardKey( obj );
}
- void ChunkManager::createFirstChunk( const Shard& shard ) const {
+ void ChunkManager::createFirstChunks( const Shard& shard ) const {
// TODO distlock?
assert( _chunkMap.size() == 0 );
- Chunk c (this, _key.globalMin(), _key.globalMax(), shard);
+ unsigned long long numObjects = 0;
+ {
+ // get stats to see if there is any data
+ ScopedDbConnection shardConn( shard.getConnString() );
+ numObjects = shardConn->count( getns() );
+ shardConn.done();
+ }
// this is the first chunk; start the versioning from scratch
ShardChunkVersion version;
version.incMajor();
- // build update for the chunk collection
- BSONObjBuilder chunkBuilder;
- c.serialize( chunkBuilder , version );
- BSONObj chunkCmd = chunkBuilder.obj();
+ Chunk c(this, _key.globalMin(), _key.globalMax(), shard);
- log() << "about to create first chunk for: " << _ns << endl;
+ vector<BSONObj> splitPoints;
+ if ( numObjects > 0 )
+ c.pickSplitVector( splitPoints , Chunk::MaxChunkSize );
+
+ log() << "going to create " << splitPoints.size() + 1 << " chunk(s) for: " << _ns << endl;
+
- ScopedDbConnection conn( configServer.modelServer() );
- BSONObj res;
- conn->update( Chunk::chunkMetadataNS, QUERY( "_id" << c.genID() ), chunkCmd, true, false );
+ ScopedDbConnection conn( configServer.modelServer() );
+
+ for ( unsigned i=0; i<=splitPoints.size(); i++ ) {
+ BSONObj min = i == 0 ? _key.globalMin() : splitPoints[i-1];
+ BSONObj max = i < splitPoints.size() ? splitPoints[i] : _key.globalMax();
+
+ Chunk temp( this , min , max , shard );
+
+ BSONObjBuilder chunkBuilder;
+ temp.serialize( chunkBuilder , version );
+ BSONObj chunkObj = chunkBuilder.obj();
+
+ conn->update( Chunk::chunkMetadataNS, QUERY( "_id" << temp.genID() ), chunkObj, true, false );
+
+ version.incMinor();
+ }
string errmsg = conn->getLastError();
if ( errmsg.size() ) {
- stringstream ss;
- ss << "saving first chunk failed. cmd: " << chunkCmd << " result: " << errmsg;
- log( LL_ERROR ) << ss.str() << endl;
- msgasserted( 13592 , ss.str() );
+ string ss = str::stream() << "creating first chunks failed. result: " << errmsg;
+ error() << ss << endl;
+ msgasserted( 15903 , ss );
}
-
+
conn.done();
- // the ensure index will have the (desired) indirect effect of creating the collection on the
- // assigned shard, as it sets up the index over the sharding keys.
- ScopedDbConnection shardConn( c.getShard().getConnString() );
- shardConn->ensureIndex( getns() , getShardKey().key() , _unique , "" , false /* do not cache ensureIndex SERVER-1691 */ );
- shardConn.done();
+ if ( numObjects == 0 ) {
+ // the ensure index will have the (desired) indirect effect of creating the collection on the
+ // assigned shard, as it sets up the index over the sharding keys.
+ ScopedDbConnection shardConn( c.getShard().getConnString() );
+ shardConn->ensureIndex( getns() , getShardKey().key() , _unique , "" , false ); // do not cache ensureIndex SERVER-1691
+ shardConn.done();
+ }
- log() << "successfully created first chunk for " << c.toString() << endl;
}
ChunkPtr ChunkManager::findChunk( const BSONObj & obj ) const {
@@ -836,26 +872,6 @@ namespace mongo {
configServer.logChange( "dropCollection" , _ns , BSONObj() );
}
- void ChunkManager::maybeChunkCollection() const {
- uassert( 13346 , "can't pre-split already splitted collection" , (_chunkMap.size() == 1) );
-
- ChunkPtr soleChunk = _chunkMap.begin()->second;
- vector<BSONObj> splitPoints;
- soleChunk->pickSplitVector( splitPoints , Chunk::MaxChunkSize );
- if ( splitPoints.empty() ) {
- LOG(1) << "not enough data to warrant chunking " << getns() << endl;
- return;
- }
-
- BSONObj res;
- ChunkPtr p;
- bool worked = soleChunk->multiSplit( splitPoints , res );
- if (!worked) {
- log( LL_WARNING ) << "could not split '" << getns() << "': " << res << endl;
- return;
- }
- }
-
ShardChunkVersion ChunkManager::getVersion( const Shard& shard ) const {
ShardVersionMap::const_iterator i = _shardVersions.find( shard );
if ( i == _shardVersions.end() )
@@ -956,7 +972,13 @@ namespace mongo {
int nc = numChunks();
- if ( nc < 10 ) {
+ if ( nc <= 1 ) {
+ return 1024;
+ }
+ else if ( nc < 3 ) {
+ return minChunkSize / 2;
+ }
+ else if ( nc < 10 ) {
splitThreshold = max( splitThreshold / 4 , minChunkSize );
}
else if ( nc < 20 ) {
generated by cgit v1.2.3 (git 2.39.1) at 2025-06-13 09:45:39 +0000