// Return false if the tuple's table (tile group) is dropped.
// In such case, this recycled tuple can not be added to the recycled_list.
// Since no one will use it any more, keeping track of it is useless.
// Note that, if we drop a single tile group without dropping the whole table,
// such assumption is problematic.
bool GCManager::ResetTuple(const TupleMetadata &tuple_metadata) {
auto &manager = catalog::Manager::GetInstance();
auto tile_group = manager.GetTileGroup(tuple_metadata.tile_group_id);
// During the resetting, a table may deconstruct because of the DROP TABLE request
if (tile_group == nullptr) {
LOG_TRACE("Garbage tuple(%u, %u) in table %u no longer exists",
tuple_metadata.tile_group_id, tuple_metadata.tuple_slot_id,
tuple_metadata.table_id);
return false;
}
// From now on, the tile group shared pointer is held by us
// It's safe to set headers from now on.
auto tile_group_header = tile_group->GetHeader();
// Reset the header
tile_group_header->SetTransactionId(tuple_metadata.tuple_slot_id,
INVALID_TXN_ID);
tile_group_header->SetBeginCommitId(tuple_metadata.tuple_slot_id, MAX_CID);
tile_group_header->SetEndCommitId(tuple_metadata.tuple_slot_id, MAX_CID);
tile_group_header->SetPrevItemPointer(tuple_metadata.tuple_slot_id,
INVALID_ITEMPOINTER);
tile_group_header->SetNextItemPointer(tuple_metadata.tuple_slot_id,
INVALID_ITEMPOINTER);
PL_MEMSET(
tile_group_header->GetReservedFieldRef(tuple_metadata.tuple_slot_id), 0,
storage::TileGroupHeader::GetReservedSize());
LOG_TRACE("Garbage tuple(%u, %u) in table %u is reset",
tuple_metadata.tile_group_id, tuple_metadata.tuple_slot_id,
tuple_metadata.table_id);
return true;
}
TileGroupHeader::TileGroupHeader(BackendType backend_type, int tuple_count)
: backend_type(backend_type),
data(nullptr),
num_tuple_slots(tuple_count),
next_tuple_slot(0) {
header_size = num_tuple_slots * header_entry_size;
// allocate storage space for header
auto &storage_manager = storage::StorageManager::GetInstance();
data = reinterpret_cast<char *>(
storage_manager.Allocate(backend_type, header_size));
assert(data != nullptr);
// zero out the data
std::memset(data, 0, header_size);
// Set MVCC Initial Value
for (oid_t tuple_slot_id = START_OID; tuple_slot_id < num_tuple_slots;
tuple_slot_id++) {
SetTransactionId(tuple_slot_id, INVALID_TXN_ID);
SetBeginCommitId(tuple_slot_id, MAX_CID);
SetEndCommitId(tuple_slot_id, MAX_CID);
SetInsertCommit(tuple_slot_id, false);
SetDeleteCommit(tuple_slot_id, false);
}
}
TileGroupHeader::TileGroupHeader(const BackendType &backend_type,
const int &tuple_count)
: backend_type(backend_type),
data(nullptr),
num_tuple_slots(tuple_count),
next_tuple_slot(0),
tile_header_lock() {
header_size = num_tuple_slots * header_entry_size;
// allocate storage space for header
auto &storage_manager = storage::StorageManager::GetInstance();
data = reinterpret_cast<char *>(
storage_manager.Allocate(backend_type, header_size));
PL_ASSERT(data != nullptr);
// zero out the data
PL_MEMSET(data, 0, header_size);
// Set MVCC Initial Value
for (oid_t tuple_slot_id = START_OID; tuple_slot_id < num_tuple_slots;
tuple_slot_id++) {
SetTransactionId(tuple_slot_id, INVALID_TXN_ID);
SetBeginCommitId(tuple_slot_id, MAX_CID);
SetEndCommitId(tuple_slot_id, MAX_CID);
SetNextItemPointer(tuple_slot_id, INVALID_ITEMPOINTER);
SetPrevItemPointer(tuple_slot_id, INVALID_ITEMPOINTER);
SetInsertCommit(tuple_slot_id, false); // unused
SetDeleteCommit(tuple_slot_id, false); // unused
}
}
bool TransactionLevelGCManager::ResetTuple(const ItemPointer &location) {
auto &manager = catalog::Manager::GetInstance();
auto tile_group = manager.GetTileGroup(location.block).get();
auto tile_group_header = tile_group->GetHeader();
// Reset the header
tile_group_header->SetTransactionId(location.offset, INVALID_TXN_ID);
tile_group_header->SetBeginCommitId(location.offset, MAX_CID);
tile_group_header->SetEndCommitId(location.offset, MAX_CID);
tile_group_header->SetPrevItemPointer(location.offset, INVALID_ITEMPOINTER);
tile_group_header->SetNextItemPointer(location.offset, INVALID_ITEMPOINTER);
PL_MEMSET(
tile_group_header->GetReservedFieldRef(location.offset), 0,
storage::TileGroupHeader::GetReservedSize());
// Reclaim the varlen pool
CheckAndReclaimVarlenColumns(tile_group, location.offset);
LOG_TRACE("Garbage tuple(%u, %u) is reset", location.block, location.offset);
return true;
}
Result PessimisticTxnManager::AbortTransaction() {
LOG_TRACE("Aborting peloton txn : %lu ", current_txn->GetTransactionId());
auto &manager = catalog::Manager::GetInstance();
auto &rw_set = current_txn->GetRWSet();
for (auto &tile_group_entry : rw_set) {
oid_t tile_group_id = tile_group_entry.first;
auto tile_group = manager.GetTileGroup(tile_group_id);
auto tile_group_header = tile_group->GetHeader();
for (auto &tuple_entry : tile_group_entry.second) {
auto tuple_slot = tuple_entry.first;
if (tuple_entry.second == RW_TYPE_READ) {
if (pessimistic_released_rdlock.find(tile_group_id) ==
pessimistic_released_rdlock.end() ||
pessimistic_released_rdlock[tile_group_id].find(tuple_slot) ==
pessimistic_released_rdlock[tile_group_id].end()) {
ReleaseReadLock(tile_group_header, tuple_slot);
pessimistic_released_rdlock[tile_group_id].insert(tuple_slot);
}
} else if (tuple_entry.second == RW_TYPE_UPDATE) {
ItemPointer new_version =
tile_group_header->GetNextItemPointer(tuple_slot);
auto new_tile_group_header =
manager.GetTileGroup(new_version.block)->GetHeader();
new_tile_group_header->SetBeginCommitId(new_version.offset, MAX_CID);
new_tile_group_header->SetEndCommitId(new_version.offset, MAX_CID);
COMPILER_MEMORY_FENCE;
tile_group_header->SetEndCommitId(tuple_slot, MAX_CID);
COMPILER_MEMORY_FENCE;
new_tile_group_header->SetTransactionId(new_version.offset,
INVALID_TXN_ID);
tile_group_header->SetTransactionId(tuple_slot, INITIAL_TXN_ID);
} else if (tuple_entry.second == RW_TYPE_DELETE) {
ItemPointer new_version =
tile_group_header->GetNextItemPointer(tuple_slot);
auto new_tile_group_header =
manager.GetTileGroup(new_version.block)->GetHeader();
new_tile_group_header->SetBeginCommitId(new_version.offset, MAX_CID);
new_tile_group_header->SetEndCommitId(new_version.offset, MAX_CID);
COMPILER_MEMORY_FENCE;
tile_group_header->SetEndCommitId(tuple_slot, MAX_CID);
COMPILER_MEMORY_FENCE;
new_tile_group_header->SetTransactionId(new_version.offset,
INVALID_TXN_ID);
tile_group_header->SetTransactionId(tuple_slot, INITIAL_TXN_ID);
} else if (tuple_entry.second == RW_TYPE_INSERT) {
tile_group_header->SetEndCommitId(tuple_slot, MAX_CID);
tile_group_header->SetBeginCommitId(tuple_slot, MAX_CID);
COMPILER_MEMORY_FENCE;
tile_group_header->SetTransactionId(tuple_slot, INVALID_TXN_ID);
} else if (tuple_entry.second == RW_TYPE_INS_DEL) {
tile_group_header->SetEndCommitId(tuple_slot, MAX_CID);
tile_group_header->SetBeginCommitId(tuple_slot, MAX_CID);
COMPILER_MEMORY_FENCE;
tile_group_header->SetTransactionId(tuple_slot, INVALID_TXN_ID);
}
}
}
EndTransaction();
pessimistic_released_rdlock.clear();
return Result::RESULT_ABORTED;
}
Result PessimisticTxnManager::CommitTransaction() {
LOG_TRACE("Committing peloton txn : %lu ", current_txn->GetTransactionId());
auto &manager = catalog::Manager::GetInstance();
auto &rw_set = current_txn->GetRWSet();
//*****************************************************
// we can optimize read-only transaction.
if (current_txn->IsReadOnly() == true) {
// validate read set.
for (auto &tile_group_entry : rw_set) {
oid_t tile_group_id = tile_group_entry.first;
auto tile_group = manager.GetTileGroup(tile_group_id);
auto tile_group_header = tile_group->GetHeader();
for (auto &tuple_entry : tile_group_entry.second) {
auto tuple_slot = tuple_entry.first;
// if this tuple is not newly inserted.
if (tuple_entry.second == RW_TYPE_READ) {
// Release read locks
if (pessimistic_released_rdlock.find(tile_group_id) ==
pessimistic_released_rdlock.end() ||
pessimistic_released_rdlock[tile_group_id].find(tuple_slot) ==
pessimistic_released_rdlock[tile_group_id].end()) {
ReleaseReadLock(tile_group_header, tuple_slot);
pessimistic_released_rdlock[tile_group_id].insert(tuple_slot);
}
} else {
assert(tuple_entry.second == RW_TYPE_INS_DEL);
}
}
}
// is it always true???
Result ret = current_txn->GetResult();
EndTransaction();
return ret;
}
//*****************************************************
// generate transaction id.
cid_t end_commit_id = GetNextCommitId();
auto &log_manager = logging::LogManager::GetInstance();
log_manager.LogBeginTransaction(end_commit_id);
// install everything.
for (auto &tile_group_entry : rw_set) {
oid_t tile_group_id = tile_group_entry.first;
auto tile_group = manager.GetTileGroup(tile_group_id);
auto tile_group_header = tile_group->GetHeader();
for (auto &tuple_entry : tile_group_entry.second) {
auto tuple_slot = tuple_entry.first;
if (tuple_entry.second == RW_TYPE_READ) {
// Release read locks
if (pessimistic_released_rdlock.find(tile_group_id) ==
pessimistic_released_rdlock.end() ||
pessimistic_released_rdlock[tile_group_id].find(tuple_slot) ==
pessimistic_released_rdlock[tile_group_id].end()) {
ReleaseReadLock(tile_group_header, tuple_slot);
pessimistic_released_rdlock[tile_group_id].insert(tuple_slot);
}
} else if (tuple_entry.second == RW_TYPE_UPDATE) {
// we must guarantee that, at any time point, only one version is
// visible.
ItemPointer new_version =
tile_group_header->GetNextItemPointer(tuple_slot);
ItemPointer old_version(tile_group_id, tuple_slot);
// logging.
log_manager.LogUpdate(current_txn, end_commit_id, old_version,
new_version);
auto new_tile_group_header =
manager.GetTileGroup(new_version.block)->GetHeader();
new_tile_group_header->SetEndCommitId(new_version.offset, MAX_CID);
new_tile_group_header->SetBeginCommitId(new_version.offset,
end_commit_id);
COMPILER_MEMORY_FENCE;
tile_group_header->SetEndCommitId(tuple_slot, end_commit_id);
COMPILER_MEMORY_FENCE;
new_tile_group_header->SetTransactionId(new_version.offset,
INITIAL_TXN_ID);
tile_group_header->SetTransactionId(tuple_slot, INITIAL_TXN_ID);
} else if (tuple_entry.second == RW_TYPE_DELETE) {
ItemPointer new_version =
tile_group_header->GetNextItemPointer(tuple_slot);
ItemPointer delete_location(tile_group_id, tuple_slot);
// logging.
log_manager.LogDelete(end_commit_id, delete_location);
// we do not change begin cid for old tuple.
auto new_tile_group_header =
manager.GetTileGroup(new_version.block)->GetHeader();
new_tile_group_header->SetEndCommitId(new_version.offset, MAX_CID);
new_tile_group_header->SetBeginCommitId(new_version.offset,
end_commit_id);
COMPILER_MEMORY_FENCE;
tile_group_header->SetEndCommitId(tuple_slot, end_commit_id);
COMPILER_MEMORY_FENCE;
new_tile_group_header->SetTransactionId(new_version.offset,
INVALID_TXN_ID);
tile_group_header->SetTransactionId(tuple_slot, INITIAL_TXN_ID);
} else if (tuple_entry.second == RW_TYPE_INSERT) {
assert(tile_group_header->GetTransactionId(tuple_slot) ==
current_txn->GetTransactionId());
// set the begin commit id to persist insert
ItemPointer insert_location(tile_group_id, tuple_slot);
log_manager.LogInsert(current_txn, end_commit_id, insert_location);
tile_group_header->SetEndCommitId(tuple_slot, MAX_CID);
tile_group_header->SetBeginCommitId(tuple_slot, end_commit_id);
COMPILER_MEMORY_FENCE;
tile_group_header->SetTransactionId(tuple_slot, INITIAL_TXN_ID);
} else if (tuple_entry.second == RW_TYPE_INS_DEL) {
assert(tile_group_header->GetTransactionId(tuple_slot) ==
current_txn->GetTransactionId());
tile_group_header->SetEndCommitId(tuple_slot, MAX_CID);
tile_group_header->SetBeginCommitId(tuple_slot, MAX_CID);
COMPILER_MEMORY_FENCE;
// set the begin commit id to persist insert
tile_group_header->SetTransactionId(tuple_slot, INVALID_TXN_ID);
}
}
}
log_manager.LogCommitTransaction(end_commit_id);
EndTransaction();
pessimistic_released_rdlock.clear();
return Result::RESULT_SUCCESS;
}