void DistributionStatus::populateShardToChunksMap(const ShardStatisticsVector& allShards,
const ChunkManager& chunkMgr,
ShardToChunksMap* shardToChunksMap) {
// Makes sure there is an entry in shardToChunksMap for every shard.
for (const auto& stat : allShards) {
(*shardToChunksMap)[stat.shardId];
}
const ChunkMap& chunkMap = chunkMgr.getChunkMap();
for (ChunkMap::const_iterator it = chunkMap.begin(); it != chunkMap.end(); ++it) {
const ChunkPtr chunkPtr = it->second;
ChunkType chunk;
chunk.setNS(chunkMgr.getns());
chunk.setMin(chunkPtr->getMin().getOwned());
chunk.setMax(chunkPtr->getMax().getOwned());
chunk.setJumbo(chunkPtr->isJumbo()); // TODO: is this reliable?
const string shardName(chunkPtr->getShardId());
chunk.setShard(shardName);
(*shardToChunksMap)[shardName].push_back(chunk);
}
}
StatusWith<MigrateInfoVector> BalancerChunkSelectionPolicyImpl::_getMigrateCandidatesForCollection(
OperationContext* txn,
const NamespaceString& nss,
const ShardStatisticsVector& shardStats,
bool aggressiveBalanceHint) {
// Ensure the database exists
auto dbStatus = Grid::get(txn)->catalogCache()->getDatabase(txn, nss.db().toString());
if (!dbStatus.isOK()) {
return {dbStatus.getStatus().code(),
str::stream() << "Database " << nss.ns() << " was not found due to "
<< dbStatus.getStatus().toString()};
}
shared_ptr<DBConfig> db = dbStatus.getValue();
invariant(db);
// Ensure that the collection is sharded
shared_ptr<ChunkManager> cm = db->getChunkManagerIfExists(txn, nss.ns(), true);
if (!cm) {
return {ErrorCodes::NamespaceNotSharded,
str::stream() << "Collection " << nss.ns() << " does not exist or is not sharded."};
}
if (cm->getChunkMap().empty()) {
return {ErrorCodes::NamespaceNotSharded,
str::stream() << "Collection " << nss.ns() << " does not have any chunks."};
}
ShardToChunksMap shardToChunksMap;
std::set<BSONObj> allChunkMinimums;
for (const auto& entry : cm->getChunkMap()) {
const auto& chunkEntry = entry.second;
ChunkType chunk;
chunk.setMin(chunkEntry->getMin());
chunk.setMax(chunkEntry->getMax());
chunk.setJumbo(chunkEntry->isJumbo());
shardToChunksMap[chunkEntry->getShardId()].push_back(chunk);
allChunkMinimums.insert(chunkEntry->getMin());
}
for (const auto& stat : shardStats) {
// This loop just makes sure there is an entry in shardToChunksMap for every shard, which we
// plan to consider.
shardToChunksMap[stat.shardId];
}
DistributionStatus distStatus(shardStats, shardToChunksMap);
{
vector<TagsType> collectionTags;
Status status =
grid.catalogManager(txn)->getTagsForCollection(txn, nss.ns(), &collectionTags);
if (!status.isOK()) {
return status;
}
for (const auto& tagInfo : collectionTags) {
BSONObj min = cm->getShardKeyPattern().getKeyPattern().extendRangeBound(
tagInfo.getMinKey(), false);
if (!allChunkMinimums.count(min)) {
// This tag falls somewhere at the middle of a chunk. Therefore we must skip
// balancing this collection until it is split at the next iteration.
//
// TODO: We should be able to just skip chunks, which straddle tags and still make
// some progress balancing.
return {ErrorCodes::IllegalOperation,
str::stream()
<< "Tag boundaries " << tagInfo.toString()
<< " fall in the middle of an existing chunk. Balancing for collection "
<< nss.ns()
<< " will be postponed until the chunk is split appropriately."};
}
// TODO: TagRange contains all the information from TagsType except for the namespace,
// so maybe the two can be merged at some point in order to avoid the transformation
// below.
if (!distStatus.addTagRange(TagRange(tagInfo.getMinKey().getOwned(),
tagInfo.getMaxKey().getOwned(),
tagInfo.getTag()))) {
return {ErrorCodes::BadValue,
str::stream() << "Tag ranges are not valid for collection " << nss.ns()
<< ". Balancing for this collection will be skipped until "
"the ranges are fixed."};
}
}
}
unique_ptr<MigrateInfo> migrateInfo(
BalancerPolicy::balance(nss.ns(), distStatus, aggressiveBalanceHint));
if (migrateInfo) {
return MigrateInfoVector{*migrateInfo};
}
return MigrateInfoVector{};
}