/**
* Stores ranges for a particular collection and shard starting from some version
*/
void storeCollectionRanges(const NamespaceString& nss,
const string& shardName,
const vector<KeyRange>& ranges,
const ChunkVersion& startVersion) {
// Get key pattern from first range
ASSERT_GREATER_THAN(ranges.size(), 0u);
CollectionType coll;
coll.setNs(nss);
coll.setKeyPattern(ranges.begin()->keyPattern);
coll.setEpoch(startVersion.epoch());
coll.setUpdatedAt(Date_t::fromMillisSinceEpoch(1));
ASSERT_OK(coll.validate());
DBDirectClient client(&_txn);
client.update(CollectionType::ConfigNS,
BSON(CollectionType::fullNs(coll.getNs().ns())),
coll.toBSON(),
true,
false);
ChunkVersion nextVersion = startVersion;
for (vector<KeyRange>::const_iterator it = ranges.begin(); it != ranges.end(); ++it) {
ChunkType chunk;
// TODO: We should not rely on the serialized ns, minkey being unique in the future,
// causes problems since it links string serialization to correctness.
chunk.setName(Chunk::genID(nss.ns(), it->minKey));
chunk.setShard(shardName);
chunk.setNS(nss.ns());
chunk.setVersion(nextVersion);
chunk.setMin(it->minKey);
chunk.setMax(it->maxKey);
nextVersion.incMajor();
client.insert(ChunkType::ConfigNS, chunk.toBSON());
}
}
void CollectionInfo::save(OperationContext* txn, const string& ns) {
CollectionType coll;
coll.setNs(NamespaceString{ns});
if (_cm) {
invariant(!_dropped);
coll.setEpoch(_cm->getVersion().epoch());
// TODO(schwerin): The following isn't really a date, but is stored as one in-memory and
// in config.collections, as a historical oddity.
coll.setUpdatedAt(Date_t::fromMillisSinceEpoch(_cm->getVersion().toLong()));
coll.setKeyPattern(_cm->getShardKeyPattern().toBSON());
coll.setDefaultCollation(_cm->getDefaultCollation());
coll.setUnique(_cm->isUnique());
} else {
invariant(_dropped);
coll.setDropped(true);
coll.setEpoch(ChunkVersion::DROPPED().epoch());
coll.setUpdatedAt(Date_t::now());
}
uassertStatusOK(grid.catalogClient(txn)->updateCollection(txn, ns, coll));
_dirty = false;
}
void ShardingCatalogManager::shardCollection(OperationContext* opCtx,
const NamespaceString& nss,
const boost::optional<UUID> uuid,
const ShardKeyPattern& fieldsAndOrder,
const BSONObj& defaultCollation,
bool unique,
const vector<BSONObj>& splitPoints,
bool isFromMapReduce,
const ShardId& dbPrimaryShardId) {
const auto shardRegistry = Grid::get(opCtx)->shardRegistry();
const auto primaryShard = uassertStatusOK(shardRegistry->getShard(opCtx, dbPrimaryShardId));
// Fail if there are partially written chunks from a previous failed shardCollection.
checkForExistingChunks(opCtx, nss);
// Prior to 4.0.5, zones cannot be taken into account at collection sharding time, so ignore
// them and let the balancer apply them later
const std::vector<TagsType> treatAsNoZonesDefined;
// Map/reduce with output to sharded collection ignores consistency checks and requires the
// initial chunks to be spread across shards unconditionally
const bool treatAsEmpty = isFromMapReduce;
// Record start in changelog
{
BSONObjBuilder collectionDetail;
collectionDetail.append("shardKey", fieldsAndOrder.toBSON());
collectionDetail.append("collection", nss.ns());
if (uuid)
uuid->appendToBuilder(&collectionDetail, "uuid");
collectionDetail.append("empty", treatAsEmpty);
collectionDetail.append("fromMapReduce", isFromMapReduce);
collectionDetail.append("primary", primaryShard->toString());
collectionDetail.append("numChunks", static_cast<int>(splitPoints.size() + 1));
uassertStatusOK(ShardingLogging::get(opCtx)->logChangeChecked(
opCtx,
"shardCollection.start",
nss.ns(),
collectionDetail.obj(),
ShardingCatalogClient::kMajorityWriteConcern));
}
// Construct the collection default collator.
std::unique_ptr<CollatorInterface> defaultCollator;
if (!defaultCollation.isEmpty()) {
defaultCollator = uassertStatusOK(CollatorFactoryInterface::get(opCtx->getServiceContext())
->makeFromBSON(defaultCollation));
}
const auto initialChunks = InitialSplitPolicy::createFirstChunks(opCtx,
nss,
fieldsAndOrder,
dbPrimaryShardId,
splitPoints,
treatAsNoZonesDefined,
treatAsEmpty);
InitialSplitPolicy::writeFirstChunksToConfig(opCtx, initialChunks);
{
CollectionType coll;
coll.setNs(nss);
if (uuid)
coll.setUUID(*uuid);
coll.setEpoch(initialChunks.collVersion().epoch());
coll.setUpdatedAt(Date_t::fromMillisSinceEpoch(initialChunks.collVersion().toLong()));
coll.setKeyPattern(fieldsAndOrder.toBSON());
coll.setDefaultCollation(defaultCollator ? defaultCollator->getSpec().toBSON() : BSONObj());
coll.setUnique(unique);
uassertStatusOK(ShardingCatalogClientImpl::updateShardingCatalogEntryForCollection(
opCtx, nss, coll, true /*upsert*/));
}
auto shard = uassertStatusOK(shardRegistry->getShard(opCtx, dbPrimaryShardId));
invariant(!shard->isConfig());
// Tell the primary mongod to refresh its data
SetShardVersionRequest ssv = SetShardVersionRequest::makeForVersioningNoPersist(
shardRegistry->getConfigServerConnectionString(),
dbPrimaryShardId,
primaryShard->getConnString(),
nss,
initialChunks.collVersion(),
true /* isAuthoritative */,
true /* forceRefresh */);
auto ssvResponse =
shard->runCommandWithFixedRetryAttempts(opCtx,
ReadPreferenceSetting{ReadPreference::PrimaryOnly},
"admin",
ssv.toBSON(),
Shard::RetryPolicy::kIdempotent);
auto status = ssvResponse.isOK() ? std::move(ssvResponse.getValue().commandStatus)
: std::move(ssvResponse.getStatus());
if (!status.isOK()) {
warning() << "could not update initial version of " << nss.ns() << " on shard primary "
<< dbPrimaryShardId << causedBy(redact(status));
}
ShardingLogging::get(opCtx)->logChange(
opCtx,
"shardCollection.end",
nss.ns(),
BSON("version" << initialChunks.collVersion().toString()),
ShardingCatalogClient::kMajorityWriteConcern);
}
Status ShardingCatalogManager::dropCollection(OperationContext* opCtx, const NamespaceString& nss) {
const Status logStatus =
ShardingLogging::get(opCtx)->logChangeChecked(opCtx,
"dropCollection.start",
nss.ns(),
BSONObj(),
ShardingCatalogClient::kMajorityWriteConcern);
if (!logStatus.isOK()) {
return logStatus;
}
const auto catalogClient = Grid::get(opCtx)->catalogClient();
const auto shardsStatus =
catalogClient->getAllShards(opCtx, repl::ReadConcernLevel::kLocalReadConcern);
if (!shardsStatus.isOK()) {
return shardsStatus.getStatus();
}
vector<ShardType> allShards = std::move(shardsStatus.getValue().value);
LOG(1) << "dropCollection " << nss.ns() << " started";
const auto dropCommandBSON = [opCtx, &nss] {
BSONObjBuilder builder;
builder.append("drop", nss.coll());
if (!opCtx->getWriteConcern().usedDefault) {
builder.append(WriteConcernOptions::kWriteConcernField,
opCtx->getWriteConcern().toBSON());
}
return builder.obj();
}();
std::map<std::string, BSONObj> errors;
auto* const shardRegistry = Grid::get(opCtx)->shardRegistry();
for (const auto& shardEntry : allShards) {
auto swShard = shardRegistry->getShard(opCtx, shardEntry.getName());
if (!swShard.isOK()) {
return swShard.getStatus();
}
const auto& shard = swShard.getValue();
auto swDropResult = shard->runCommandWithFixedRetryAttempts(
opCtx,
ReadPreferenceSetting{ReadPreference::PrimaryOnly},
nss.db().toString(),
dropCommandBSON,
Shard::RetryPolicy::kIdempotent);
if (!swDropResult.isOK()) {
return swDropResult.getStatus().withContext(
str::stream() << "Error dropping collection on shard " << shardEntry.getName());
}
auto& dropResult = swDropResult.getValue();
auto dropStatus = std::move(dropResult.commandStatus);
auto wcStatus = std::move(dropResult.writeConcernStatus);
if (!dropStatus.isOK() || !wcStatus.isOK()) {
if (dropStatus.code() == ErrorCodes::NamespaceNotFound && wcStatus.isOK()) {
// Generally getting NamespaceNotFound is okay to ignore as it simply means that
// the collection has already been dropped or doesn't exist on this shard.
// If, however, we get NamespaceNotFound but also have a write concern error then we
// can't confirm whether the fact that the namespace doesn't exist is actually
// committed. Thus we must still fail on NamespaceNotFound if there is also a write
// concern error. This can happen if we call drop, it succeeds but with a write
// concern error, then we retry the drop.
continue;
}
errors.emplace(shardEntry.getHost(), std::move(dropResult.response));
}
}
if (!errors.empty()) {
StringBuilder sb;
sb << "Dropping collection failed on the following hosts: ";
for (auto it = errors.cbegin(); it != errors.cend(); ++it) {
if (it != errors.cbegin()) {
sb << ", ";
}
sb << it->first << ": " << it->second;
}
return {ErrorCodes::OperationFailed, sb.str()};
}
LOG(1) << "dropCollection " << nss.ns() << " shard data deleted";
// Remove chunk data
Status result =
catalogClient->removeConfigDocuments(opCtx,
ChunkType::ConfigNS,
BSON(ChunkType::ns(nss.ns())),
ShardingCatalogClient::kMajorityWriteConcern);
if (!result.isOK()) {
return result;
}
LOG(1) << "dropCollection " << nss.ns() << " chunk data deleted";
// Remove tag data
result = catalogClient->removeConfigDocuments(opCtx,
TagsType::ConfigNS,
BSON(TagsType::ns(nss.ns())),
ShardingCatalogClient::kMajorityWriteConcern);
if (!result.isOK()) {
return result;
}
LOG(1) << "dropCollection " << nss.ns() << " tag data deleted";
// Mark the collection as dropped
CollectionType coll;
coll.setNs(nss);
coll.setDropped(true);
coll.setEpoch(ChunkVersion::DROPPED().epoch());
coll.setUpdatedAt(Grid::get(opCtx)->getNetwork()->now());
const bool upsert = false;
result = ShardingCatalogClientImpl::updateShardingCatalogEntryForCollection(
opCtx, nss, coll, upsert);
if (!result.isOK()) {
return result;
}
LOG(1) << "dropCollection " << nss.ns() << " collection marked as dropped";
for (const auto& shardEntry : allShards) {
auto swShard = shardRegistry->getShard(opCtx, shardEntry.getName());
if (!swShard.isOK()) {
return swShard.getStatus();
}
const auto& shard = swShard.getValue();
SetShardVersionRequest ssv = SetShardVersionRequest::makeForVersioningNoPersist(
shardRegistry->getConfigServerConnectionString(),
shardEntry.getName(),
fassert(28781, ConnectionString::parse(shardEntry.getHost())),
nss,
ChunkVersion::DROPPED(),
true /* isAuthoritative */,
true /* forceRefresh */);
auto ssvResult = shard->runCommandWithFixedRetryAttempts(
opCtx,
ReadPreferenceSetting{ReadPreference::PrimaryOnly},
"admin",
ssv.toBSON(),
Shard::RetryPolicy::kIdempotent);
if (!ssvResult.isOK()) {
return ssvResult.getStatus();
}
auto ssvStatus = std::move(ssvResult.getValue().commandStatus);
if (!ssvStatus.isOK()) {
return ssvStatus;
}
auto unsetShardingStatus = shard->runCommandWithFixedRetryAttempts(
opCtx,
ReadPreferenceSetting{ReadPreference::PrimaryOnly},
"admin",
BSON("unsetSharding" << 1),
Shard::RetryPolicy::kIdempotent);
if (!unsetShardingStatus.isOK()) {
return unsetShardingStatus.getStatus();
}
auto unsetShardingResult = std::move(unsetShardingStatus.getValue().commandStatus);
if (!unsetShardingResult.isOK()) {
return unsetShardingResult;
}
}
LOG(1) << "dropCollection " << nss.ns() << " completed";
ShardingLogging::get(opCtx)->logChange(
opCtx, "dropCollection", nss.ns(), BSONObj(), ShardingCatalogClient::kMajorityWriteConcern);
return Status::OK();
}
Status CatalogManagerReplicaSet::dropCollection(OperationContext* txn, const NamespaceString& ns) {
logChange(
txn, txn->getClient()->clientAddress(true), "dropCollection.start", ns.ns(), BSONObj());
vector<ShardType> allShards;
Status status = getAllShards(txn, &allShards);
if (!status.isOK()) {
return status;
}
LOG(1) << "dropCollection " << ns << " started";
// Lock the collection globally so that split/migrate cannot run
stdx::chrono::seconds waitFor(2);
MONGO_FAIL_POINT_BLOCK(setDropCollDistLockWait, customWait) {
const BSONObj& data = customWait.getData();
waitFor = stdx::chrono::seconds(data["waitForSecs"].numberInt());
}
const stdx::chrono::milliseconds lockTryInterval(500);
auto scopedDistLock = getDistLockManager()->lock(ns.ns(), "drop", waitFor, lockTryInterval);
if (!scopedDistLock.isOK()) {
return scopedDistLock.getStatus();
}
LOG(1) << "dropCollection " << ns << " locked";
std::map<string, BSONObj> errors;
auto* shardRegistry = grid.shardRegistry();
for (const auto& shardEntry : allShards) {
auto dropResult = shardRegistry->runCommandWithNotMasterRetries(
txn, shardEntry.getName(), ns.db().toString(), BSON("drop" << ns.coll()));
if (!dropResult.isOK()) {
return dropResult.getStatus();
}
auto dropStatus = getStatusFromCommandResult(dropResult.getValue());
if (!dropStatus.isOK()) {
if (dropStatus.code() == ErrorCodes::NamespaceNotFound) {
continue;
}
errors.emplace(shardEntry.getHost(), dropResult.getValue());
}
}
if (!errors.empty()) {
StringBuilder sb;
sb << "Dropping collection failed on the following hosts: ";
for (auto it = errors.cbegin(); it != errors.cend(); ++it) {
if (it != errors.cbegin()) {
sb << ", ";
}
sb << it->first << ": " << it->second;
}
return {ErrorCodes::OperationFailed, sb.str()};
}
LOG(1) << "dropCollection " << ns << " shard data deleted";
// Remove chunk data
Status result = remove(txn, ChunkType::ConfigNS, BSON(ChunkType::ns(ns.ns())), 0, nullptr);
if (!result.isOK()) {
return result;
}
LOG(1) << "dropCollection " << ns << " chunk data deleted";
// Mark the collection as dropped
CollectionType coll;
coll.setNs(ns);
coll.setDropped(true);
coll.setEpoch(ChunkVersion::DROPPED().epoch());
coll.setUpdatedAt(grid.shardRegistry()->getNetwork()->now());
result = updateCollection(txn, ns.ns(), coll);
if (!result.isOK()) {
return result;
}
LOG(1) << "dropCollection " << ns << " collection marked as dropped";
for (const auto& shardEntry : allShards) {
SetShardVersionRequest ssv = SetShardVersionRequest::makeForVersioningNoPersist(
grid.shardRegistry()->getConfigServerConnectionString(),
shardEntry.getName(),
fassertStatusOK(28781, ConnectionString::parse(shardEntry.getHost())),
ns,
ChunkVersion::DROPPED(),
true);
auto ssvResult = shardRegistry->runCommandWithNotMasterRetries(
txn, shardEntry.getName(), "admin", ssv.toBSON());
if (!ssvResult.isOK()) {
return ssvResult.getStatus();
}
auto ssvStatus = getStatusFromCommandResult(ssvResult.getValue());
if (!ssvStatus.isOK()) {
return ssvStatus;
}
auto unsetShardingStatus = shardRegistry->runCommandWithNotMasterRetries(
txn, shardEntry.getName(), "admin", BSON("unsetSharding" << 1));
if (!unsetShardingStatus.isOK()) {
return unsetShardingStatus.getStatus();
}
auto unsetShardingResult = getStatusFromCommandResult(unsetShardingStatus.getValue());
if (!unsetShardingResult.isOK()) {
return unsetShardingResult;
}
}
LOG(1) << "dropCollection " << ns << " completed";
logChange(txn, txn->getClient()->clientAddress(true), "dropCollection", ns.ns(), BSONObj());
return Status::OK();
}
Status CatalogManagerReplicaSet::dropCollection(OperationContext* txn, const NamespaceString& ns) {
logChange(
txn, txn->getClient()->clientAddress(true), "dropCollection.start", ns.ns(), BSONObj());
vector<ShardType> allShards;
Status status = getAllShards(txn, &allShards);
if (!status.isOK()) {
return status;
}
LOG(1) << "dropCollection " << ns << " started";
// Lock the collection globally so that split/migrate cannot run
auto scopedDistLock = getDistLockManager()->lock(ns.ns(), "drop");
if (!scopedDistLock.isOK()) {
return scopedDistLock.getStatus();
}
LOG(1) << "dropCollection " << ns << " locked";
std::map<string, BSONObj> errors;
auto* shardRegistry = grid.shardRegistry();
for (const auto& shardEntry : allShards) {
auto dropResult = shardRegistry->runCommandWithNotMasterRetries(
txn, shardEntry.getName(), ns.db().toString(), BSON("drop" << ns.coll()));
if (!dropResult.isOK()) {
return dropResult.getStatus();
}
auto dropStatus = getStatusFromCommandResult(dropResult.getValue());
if (!dropStatus.isOK()) {
if (dropStatus.code() == ErrorCodes::NamespaceNotFound) {
continue;
}
errors.emplace(shardEntry.getHost(), dropResult.getValue());
}
}
if (!errors.empty()) {
StringBuilder sb;
sb << "Dropping collection failed on the following hosts: ";
for (auto it = errors.cbegin(); it != errors.cend(); ++it) {
if (it != errors.cbegin()) {
sb << ", ";
}
sb << it->first << ": " << it->second;
}
return {ErrorCodes::OperationFailed, sb.str()};
}
LOG(1) << "dropCollection " << ns << " shard data deleted";
// Remove chunk data
Status result = remove(txn, ChunkType::ConfigNS, BSON(ChunkType::ns(ns.ns())), 0, nullptr);
if (!result.isOK()) {
return result;
}
LOG(1) << "dropCollection " << ns << " chunk data deleted";
// Mark the collection as dropped
CollectionType coll;
coll.setNs(ns);
coll.setDropped(true);
coll.setEpoch(ChunkVersion::DROPPED().epoch());
coll.setUpdatedAt(grid.shardRegistry()->getNetwork()->now());
result = updateCollection(txn, ns.ns(), coll);
if (!result.isOK()) {
return result;
}
LOG(1) << "dropCollection " << ns << " collection marked as dropped";
// We just called updateCollection above and this would have advanced the config op time, so use
// the latest value. On the MongoD side, we need to load the latest config metadata, which
// indicates that the collection was dropped.
const ChunkVersionAndOpTime droppedVersion(ChunkVersion::DROPPED(),
grid.shardRegistry()->getConfigOpTime());
for (const auto& shardEntry : allShards) {
SetShardVersionRequest ssv = SetShardVersionRequest::makeForVersioningNoPersist(
grid.shardRegistry()->getConfigServerConnectionString(),
shardEntry.getName(),
fassertStatusOK(28781, ConnectionString::parse(shardEntry.getHost())),
ns,
droppedVersion,
true);
auto ssvResult = shardRegistry->runCommandWithNotMasterRetries(
txn, shardEntry.getName(), "admin", ssv.toBSON());
if (!ssvResult.isOK()) {
return ssvResult.getStatus();
}
auto ssvStatus = getStatusFromCommandResult(ssvResult.getValue());
if (!ssvStatus.isOK()) {
return ssvStatus;
}
auto unsetShardingStatus = shardRegistry->runCommandWithNotMasterRetries(
txn, shardEntry.getName(), "admin", BSON("unsetSharding" << 1));
if (!unsetShardingStatus.isOK()) {
return unsetShardingStatus.getStatus();
}
auto unsetShardingResult = getStatusFromCommandResult(unsetShardingStatus.getValue());
if (!unsetShardingResult.isOK()) {
return unsetShardingResult;
}
}
LOG(1) << "dropCollection " << ns << " completed";
logChange(txn, txn->getClient()->clientAddress(true), "dropCollection", ns.ns(), BSONObj());
return Status::OK();
}