void Table::serializeTo(SerializeOutput &serialOutput) {
// The table is serialized as:
// [(int) total size]
// [(int) header size] [num columns] [column types] [column names]
// [(int) num tuples] [tuple data]
/* NOTE:
VoltDBEngine uses a binary template to create tables of single integers.
It's called m_templateSingleLongTable and if you are seeing a serialization
bug in tables of single integers, make sure that's correct.
*/
// a placeholder for the total table size
std::size_t pos = serialOutput.position();
serialOutput.writeInt(-1);
serializeColumnHeaderTo(serialOutput);
// active tuple counts
serialOutput.writeInt(static_cast<int32_t>(m_tupleCount));
int64_t written_count = 0;
TableIterator titer = iterator();
TableTuple tuple(m_schema);
while (titer.next(tuple)) {
tuple.serializeTo(serialOutput);
++written_count;
}
assert(written_count == m_tupleCount);
// length prefix is non-inclusive
int32_t sz = static_cast<int32_t>(serialOutput.position() - pos - sizeof(int32_t));
assert(sz > 0);
serialOutput.writeIntAt(pos, sz);
}
void VarlenType::SerializeTo(const Value &val, SerializeOutput &out) const {
uint32_t len = GetLength(val);
out.WriteInt(len);
if (len > 0 && len < PELOTON_VALUE_NULL) {
out.WriteBytes(val.GetData(), len);
}
}
bool Table::serializeTo(int32_t offset, int32_t limit, SerializeOutput &serialize_io) {
// The table is serialized as:
// [(int) total size]
// [(int) header size] [num columns] [column types] [column names]
// [(int) num tuples] [tuple data]
/* NOTE:
VoltDBEngine uses a binary template to create tables of single integers.
It's called m_templateSingleLongTable and if you are seeing a serialization
bug in tables of single integers, make sure that's correct.
*/
// a placeholder for the total table size
std::size_t pos = serialize_io.position();
serialize_io.writeInt(-1);
if (!serializeColumnHeaderTo(serialize_io))
return false;
// active tuple counts
uint32_t output_size = m_tupleCount;
if (limit != -1 || offset != -1) {
if (offset == -1) {
output_size = (limit < m_tupleCount ? limit : m_tupleCount);
} else if (offset > m_tupleCount) {
output_size = 0;
} else {
output_size = m_tupleCount - offset;
if (limit != -1 && limit < output_size) output_size = limit;
}
}
serialize_io.writeInt(static_cast<int32_t>(output_size));
// fprintf(stderr, "SERIALIZE(output=%d, offset=%d, limit=%d, total=%d)\n", output_size, offset, limit, m_tupleCount);
int64_t written_count = 0;
int64_t read_count = 0;
TableIterator titer(this);
TableTuple tuple(m_schema);
while (titer.next(tuple)) {
if (offset == -1 || read_count >= offset) {
tuple.serializeTo(serialize_io);
if (limit != -1 && ++written_count == limit) break;
}
read_count++;
}
// assert(written_count == m_tupleCount);
// length prefix is non-inclusive
int32_t sz = static_cast<int32_t>(serialize_io.position() - pos - sizeof(int32_t));
assert(sz > 0);
serialize_io.writeIntAt(pos, sz);
return true;
}
void Tuple::SerializeTo(SerializeOutput &output) {
PL_ASSERT(tuple_schema_);
size_t start = output.ReserveBytes(4);
const int column_count = tuple_schema_->GetColumnCount();
for (int column_itr = 0; column_itr < column_count; column_itr++) {
type::Value value(GetValue(column_itr));
value.SerializeTo(output);
}
output.WriteIntAt(
start, static_cast<int32_t>(output.Position() - start - sizeof(int32_t)));
}
/**
* Serialized the table, but only includes the tuples specified (columns data and all).
* Used by the exception stuff Ariel put in.
*/
void Table::serializeTupleTo(SerializeOutput &serialOutput, voltdb::TableTuple *tuples, int numTuples) {
//assert(m_schema->equals(tuples[0].getSchema()));
std::size_t pos = serialOutput.position();
serialOutput.writeInt(-1);
assert(!tuples[0].isNullTuple());
serializeColumnHeaderTo(serialOutput);
serialOutput.writeInt(static_cast<int32_t>(numTuples));
for (int ii = 0; ii < numTuples; ii++) {
tuples[ii].serializeTo(serialOutput);
}
serialOutput.writeIntAt(pos, static_cast<int32_t>(serialOutput.position() - pos - sizeof(int32_t)));
}
void IntegerValue::SerializeTo(SerializeOutput &out) const {
switch(GetTypeId()) {
case Type::TINYINT:
out.WriteByte(value_.tinyint);
return;
case Type::SMALLINT:
out.WriteShort(value_.smallint);
return;
case Type::INTEGER:
case Type::PARAMETER_OFFSET:
out.WriteInt(value_.integer);
return;
case Type::BIGINT:
out.WriteLong(value_.bigint);
return;
default:
return;
}
throw Exception("type error");
}
void Tuple::SerializeWithHeaderTo(SerializeOutput &output) {
PL_ASSERT(tuple_schema_);
PL_ASSERT(tuple_data_);
size_t start = output.Position();
output.WriteInt(0); // reserve first 4 bytes for the total tuple size
const int column_count = tuple_schema_->GetColumnCount();
for (int column_itr = 0; column_itr < column_count; column_itr++) {
type::Value value = GetValue(column_itr);
value.SerializeTo(output);
}
int32_t serialized_size =
static_cast<int32_t>(output.Position() - start - sizeof(int32_t));
// write out the length of the tuple at start
output.WriteIntAt(start, serialized_size);
}
// Serialized only the tuples specified, along with header.
bool Tile::SerializeTuplesTo(SerializeOutput &output, Tuple *tuples,
int num_tuples) {
std::size_t pos = output.Position();
output.WriteInt(-1);
PL_ASSERT(!tuples[0].IsNull());
// Serialize the header
if (!SerializeHeaderTo(output)) return false;
output.WriteInt(static_cast<int32_t>(num_tuples));
for (int tuple_itr = 0; tuple_itr < num_tuples; tuple_itr++) {
tuples[tuple_itr].SerializeTo(output);
}
// Length prefix is non-inclusive
output.WriteIntAt(
pos, static_cast<int32_t>(output.Position() - pos - sizeof(int32_t)));
return true;
}
bool Tile::SerializeTo(SerializeOutput &output, oid_t num_tuples) {
/**
* The table is serialized as:
*
* [(int) total size]
* [(int) header size] [num columns] [column types] [column names]
* [(int) num tuples] [tuple data]
*
*/
// A placeholder for the total table size written at the end
std::size_t pos = output.Position();
output.WriteInt(-1);
// Serialize the header
if (!SerializeHeaderTo(output)) return false;
// Active tuple count
output.WriteInt(static_cast<int>(num_tuples));
oid_t written_count = 0;
TupleIterator tile_itr(this);
Tuple tuple(&schema);
while (tile_itr.Next(tuple) && written_count < num_tuples) {
tuple.SerializeTo(output);
++written_count;
}
tuple.SetNull();
PL_ASSERT(written_count == num_tuples);
// Length prefix is non-inclusive
int32_t sz = static_cast<int32_t>(output.Position() - pos - sizeof(int32_t));
PL_ASSERT(sz > 0);
output.WriteIntAt(pos, sz);
return true;
}
void Table::serializeToWithoutTotalSize(SerializeOutput &serialOutput) {
serializeColumnHeaderTo(serialOutput);
// active tuple counts
serialOutput.writeInt(static_cast<int32_t>(m_tupleCount));
int64_t written_count = 0;
TableIterator titer = iterator();
TableTuple tuple(m_schema);
while (titer.next(tuple)) {
tuple.serializeTo(serialOutput);
++written_count;
}
assert(written_count == m_tupleCount);
}
bool Table::serializeToWithoutTotalSize(SerializeOutput &serialize_io) {
if (!serializeColumnHeaderTo(serialize_io))
return false;
// active tuple counts
serialize_io.writeInt(static_cast<int32_t>(m_tupleCount));
int64_t written_count = 0;
TableIterator titer = iterator();
TableTuple tuple(m_schema);
while (titer.next(tuple)) {
tuple.serializeTo(serialize_io);
++written_count;
}
assert(written_count == m_tupleCount);
return true;
}
bool SeqScanPlan::SerializeTo(SerializeOutput &output) {
// A placeholder for the total size written at the end
int start = output.Position();
output.WriteInt(-1);
// Write the SeqScanPlan type
PlanNodeType plan_type = GetPlanNodeType();
output.WriteByte(static_cast<int8_t>(plan_type));
// Write database id and table id
if (!GetTable()) {
// The plan is not completed
return false;
}
oid_t database_id = GetTable()->GetDatabaseOid();
oid_t table_id = GetTable()->GetOid();
output.WriteInt(static_cast<int>(database_id));
output.WriteInt(static_cast<int>(table_id));
// If column has 0 item, just write the columnid_count with 0
int columnid_count = GetColumnIds().size();
output.WriteInt(columnid_count);
// If column has 0 item, nothing happens here
for (int it = 0; it < columnid_count; it++) {
oid_t col_id = GetColumnIds()[it];
output.WriteInt(static_cast<int>(col_id));
}
// Write predicate
if (GetPredicate() == nullptr) {
// Write the type
output.WriteByte(static_cast<int8_t>(EXPRESSION_TYPE_INVALID));
} else {
// Write the expression type
ExpressionType expr_type = GetPredicate()->GetExpressionType();
output.WriteByte(static_cast<int8_t>(expr_type));
// Write predicate
//GetPredicate()->SerializeTo(output);
}
// Write parent, but parent seems never be set or used right now
if (GetParent() == nullptr) {
// Write the type
output.WriteByte(static_cast<int8_t>(PLAN_NODE_TYPE_INVALID));
} else {
// Write the parent type
PlanNodeType parent_type = GetParent()->GetPlanNodeType();
output.WriteByte(static_cast<int8_t>(parent_type));
// Write parent
GetParent()->SerializeTo(output);
}
// Write the total length
int32_t sz = static_cast<int32_t>(output.Position() - start - sizeof(int));
PL_ASSERT(sz > 0);
output.WriteIntAt(start, sz);
return true;
}
void TimestampType::SerializeTo(const Value& val, SerializeOutput &out) const {
out.WriteLong(val.value_.timestamp);
}
void BooleanType::SerializeTo(const Value& val, SerializeOutput& out) const {
out.WriteByte(val.value_.boolean);
return;
}
bool Table::serializeColumnHeaderTo(SerializeOutput &serialize_io) {
/* NOTE:
VoltDBEngine uses a binary template to create tables of single integers.
It's called m_templateSingleLongTable and if you are seeing a serialization
bug in tables of single integers, make sure that's correct.
*/
// skip header position
std::size_t start;
// use a cache
if (m_columnHeaderData) {
assert(m_columnHeaderSize != -1);
serialize_io.writeBytes(m_columnHeaderData, m_columnHeaderSize);
return true;
}
assert(m_columnHeaderSize == -1);
start = serialize_io.position();
// skip header position
serialize_io.writeInt(-1);
//status code
serialize_io.writeByte(-128);
// column counts as a short
serialize_io.writeShort(static_cast<int16_t>(m_columnCount));
// write an array of column types as bytes
for (int i = 0; i < m_columnCount; ++i) {
ValueType type = m_schema->columnType(i);
serialize_io.writeByte(static_cast<int8_t>(type));
}
// write the array of column names as voltdb strings
// NOTE: strings are ASCII only in metadata (UTF-8 in table storage)
for (int i = 0; i < m_columnCount; ++i) {
// column name: write (offset, length) for column definition, and string to string table
const string& name = columnName(i);
// column names can't be null, so length must be >= 0
int32_t length = static_cast<int32_t>(name.size());
assert(length >= 0);
// this is standard string serialization for voltdb
serialize_io.writeInt(length);
serialize_io.writeBytes(name.data(), length);
}
// write the header size which is a non-inclusive int
size_t position = serialize_io.position();
m_columnHeaderSize = static_cast<int32_t>(position - start);
int32_t nonInclusiveHeaderSize = static_cast<int32_t>(m_columnHeaderSize - sizeof(int32_t));
serialize_io.writeIntAt(start, nonInclusiveHeaderSize);
// cache the results
m_columnHeaderData = new char[m_columnHeaderSize];
memcpy(m_columnHeaderData, static_cast<const char*>(serialize_io.data()) + start, m_columnHeaderSize);
return true;
}
bool Tile::SerializeHeaderTo(SerializeOutput &output) {
std::size_t start;
// Use the cache if possible
if (column_header != NULL) {
PL_ASSERT(column_header_size != INVALID_OID);
output.WriteBytes(column_header, column_header_size);
return true;
}
PL_ASSERT(column_header_size == INVALID_OID);
// Skip header position
start = output.Position();
output.WriteInt(-1);
// Status code
output.WriteByte(-128);
// Column counts as a short
output.WriteShort(static_cast<int16_t>(column_count));
// Write an array of column types as bytes
for (oid_t column_itr = 0; column_itr < column_count; ++column_itr) {
type::Type::TypeId type = schema.GetType(column_itr);
output.WriteByte(static_cast<int8_t>(type));
}
// Write the array of column names as strings
// NOTE: strings are ASCII only in metadata (UTF-8 in table storage)
for (oid_t column_itr = 0; column_itr < column_count; ++column_itr) {
// Column name: Write (offset, length) for column definition, and string to
// string table
const std::string &name = GetColumnName(column_itr);
// Column names can't be null, so length must be >= 0
int32_t length = static_cast<int32_t>(name.size());
PL_ASSERT(length >= 0);
// this is standard string serialization for voltdb
output.WriteInt(length);
output.WriteBytes(name.data(), length);
}
// Write the header size which is a non-inclusive int
size_t Position = output.Position();
column_header_size = static_cast<int32_t>(Position - start);
int32_t non_inclusive_header_size =
static_cast<int32_t>(column_header_size - sizeof(int32_t));
output.WriteIntAt(start, non_inclusive_header_size);
// Cache the column header
column_header = new char[column_header_size];
PL_MEMCPY(column_header, static_cast<const char *>(output.Data()) + start,
column_header_size);
return true;
}
