|
12 | 12 | // See the License for the specific language governing permissions and |
13 | 13 | // limitations under the License. |
14 | 14 |
|
| 15 | +use std::collections::HashSet; |
| 16 | +use std::collections::VecDeque; |
| 17 | +use std::fmt::Debug; |
| 18 | +use std::fmt::Formatter; |
| 19 | +use std::sync::Arc; |
| 20 | + |
| 21 | +use common_catalog::table::Table; |
| 22 | +use common_datavalues::DataField; |
| 23 | +use common_datavalues::DataType; |
| 24 | +use common_datavalues::DataTypeImpl; |
| 25 | +use common_datavalues::StructType; |
| 26 | +use common_datavalues::TypeID; |
| 27 | +use parking_lot::RwLock; |
| 28 | + |
15 | 29 | /// Planner use [`usize`] as it's index type. |
16 | 30 | /// |
17 | 31 | /// This type will be used across the whole planner. |
18 | 32 | pub type IndexType = usize; |
19 | 33 |
|
20 | 34 | /// Use IndexType::MAX to represent dummy table. |
21 | 35 | pub static DUMMY_TABLE_INDEX: IndexType = IndexType::MAX; |
| 36 | + |
| 37 | +/// ColumnSet represents a set of columns identified by its IndexType. |
| 38 | +pub type ColumnSet = HashSet<IndexType>; |
| 39 | + |
| 40 | +/// A Send & Send version of [`Metadata`]. |
| 41 | +/// |
| 42 | +/// Callers can clone this ref safely and cheaply. |
| 43 | +pub type MetadataRef = Arc<RwLock<Metadata>>; |
| 44 | + |
| 45 | +/// Metadata stores information about columns and tables used in a query. |
| 46 | +/// Tables and columns are identified with its unique index. |
| 47 | +/// Notice that index value of a column can be same with that of a table. |
| 48 | +#[derive(Clone, Debug, Default)] |
| 49 | +pub struct Metadata { |
| 50 | + tables: Vec<TableEntry>, |
| 51 | + columns: Vec<ColumnEntry>, |
| 52 | +} |
| 53 | + |
| 54 | +impl Metadata { |
| 55 | + pub fn table(&self, index: IndexType) -> &TableEntry { |
| 56 | + self.tables.get(index).expect("metadata must contain table") |
| 57 | + } |
| 58 | + |
| 59 | + pub fn tables(&self) -> &[TableEntry] { |
| 60 | + self.tables.as_slice() |
| 61 | + } |
| 62 | + |
| 63 | + pub fn table_index_by_column_indexes(&self, column_indexes: &ColumnSet) -> Option<IndexType> { |
| 64 | + self.columns |
| 65 | + .iter() |
| 66 | + .find(|v| column_indexes.contains(&v.column_index)) |
| 67 | + .and_then(|v| v.table_index) |
| 68 | + } |
| 69 | + |
| 70 | + pub fn column(&self, index: IndexType) -> &ColumnEntry { |
| 71 | + self.columns |
| 72 | + .get(index) |
| 73 | + .expect("metadata must contain column") |
| 74 | + } |
| 75 | + |
| 76 | + pub fn columns(&self) -> &[ColumnEntry] { |
| 77 | + self.columns.as_slice() |
| 78 | + } |
| 79 | + |
| 80 | + pub fn columns_by_table_index(&self, index: IndexType) -> Vec<ColumnEntry> { |
| 81 | + self.columns |
| 82 | + .iter() |
| 83 | + .filter(|v| v.table_index == Some(index)) |
| 84 | + .cloned() |
| 85 | + .collect() |
| 86 | + } |
| 87 | + |
| 88 | + pub fn add_column( |
| 89 | + &mut self, |
| 90 | + name: String, |
| 91 | + data_type: DataTypeImpl, |
| 92 | + table_index: Option<IndexType>, |
| 93 | + path_indices: Option<Vec<IndexType>>, |
| 94 | + ) -> IndexType { |
| 95 | + let column_index = self.columns.len(); |
| 96 | + let column_entry = |
| 97 | + ColumnEntry::new(name, data_type, column_index, table_index, path_indices); |
| 98 | + self.columns.push(column_entry); |
| 99 | + column_index |
| 100 | + } |
| 101 | + |
| 102 | + pub fn add_table( |
| 103 | + &mut self, |
| 104 | + catalog: String, |
| 105 | + database: String, |
| 106 | + table_meta: Arc<dyn Table>, |
| 107 | + ) -> IndexType { |
| 108 | + let table_name = table_meta.name().to_string(); |
| 109 | + let table_index = self.tables.len(); |
| 110 | + let table_entry = TableEntry { |
| 111 | + index: table_index, |
| 112 | + name: table_name, |
| 113 | + database, |
| 114 | + catalog, |
| 115 | + table: table_meta.clone(), |
| 116 | + }; |
| 117 | + self.tables.push(table_entry); |
| 118 | + let mut struct_fields = VecDeque::new(); |
| 119 | + for (i, field) in table_meta.schema().fields().iter().enumerate() { |
| 120 | + self.add_column( |
| 121 | + field.name().clone(), |
| 122 | + field.data_type().clone(), |
| 123 | + Some(table_index), |
| 124 | + None, |
| 125 | + ); |
| 126 | + if field.data_type().data_type_id() == TypeID::Struct { |
| 127 | + struct_fields.push_back((vec![i], field.clone())); |
| 128 | + } |
| 129 | + } |
| 130 | + // add inner columns of struct column |
| 131 | + while !struct_fields.is_empty() { |
| 132 | + let (path_indices, field) = struct_fields.pop_front().unwrap(); |
| 133 | + let struct_type: StructType = field.data_type().clone().try_into().unwrap(); |
| 134 | + |
| 135 | + let inner_types = struct_type.types(); |
| 136 | + let inner_names = match struct_type.names() { |
| 137 | + Some(inner_names) => inner_names |
| 138 | + .iter() |
| 139 | + .map(|name| format!("{}:{}", field.name(), name)) |
| 140 | + .collect::<Vec<_>>(), |
| 141 | + None => (0..inner_types.len()) |
| 142 | + .map(|i| format!("{}:{}", field.name(), i)) |
| 143 | + .collect::<Vec<_>>(), |
| 144 | + }; |
| 145 | + for ((i, inner_name), inner_type) in |
| 146 | + inner_names.into_iter().enumerate().zip(inner_types.iter()) |
| 147 | + { |
| 148 | + let mut inner_path_indices = path_indices.clone(); |
| 149 | + inner_path_indices.push(i); |
| 150 | + |
| 151 | + self.add_column( |
| 152 | + inner_name.clone(), |
| 153 | + inner_type.clone(), |
| 154 | + Some(table_index), |
| 155 | + Some(inner_path_indices.clone()), |
| 156 | + ); |
| 157 | + if inner_type.data_type_id() == TypeID::Struct { |
| 158 | + let inner_field = DataField::new(&inner_name, inner_type.clone()); |
| 159 | + struct_fields.push_back((inner_path_indices, inner_field)); |
| 160 | + } |
| 161 | + } |
| 162 | + } |
| 163 | + table_index |
| 164 | + } |
| 165 | + |
| 166 | + pub fn find_smallest_column_within(&self, indices: &[usize]) -> usize { |
| 167 | + let entries = indices |
| 168 | + .iter() |
| 169 | + .map(|i| self.column(*i).clone()) |
| 170 | + .collect::<Vec<_>>(); |
| 171 | + find_smallest_column(entries.as_slice()) |
| 172 | + } |
| 173 | +} |
| 174 | + |
| 175 | +#[derive(Clone)] |
| 176 | +pub struct TableEntry { |
| 177 | + pub index: IndexType, |
| 178 | + pub name: String, |
| 179 | + pub catalog: String, |
| 180 | + pub database: String, |
| 181 | + |
| 182 | + pub table: Arc<dyn Table>, |
| 183 | +} |
| 184 | + |
| 185 | +impl Debug for TableEntry { |
| 186 | + fn fmt(&self, f: &mut Formatter<'_>) -> std::fmt::Result { |
| 187 | + f.debug_struct("TableEntry") |
| 188 | + .field("index", &self.index) |
| 189 | + .field("name", &self.name) |
| 190 | + .field("catalog", &self.catalog) |
| 191 | + .field("database", &self.database) |
| 192 | + .finish_non_exhaustive() |
| 193 | + } |
| 194 | +} |
| 195 | + |
| 196 | +impl TableEntry { |
| 197 | + pub fn new( |
| 198 | + index: IndexType, |
| 199 | + name: String, |
| 200 | + catalog: String, |
| 201 | + database: String, |
| 202 | + table: Arc<dyn Table>, |
| 203 | + ) -> Self { |
| 204 | + TableEntry { |
| 205 | + index, |
| 206 | + name, |
| 207 | + catalog, |
| 208 | + database, |
| 209 | + table, |
| 210 | + } |
| 211 | + } |
| 212 | +} |
| 213 | + |
| 214 | +#[derive(Clone, Debug)] |
| 215 | +pub struct ColumnEntry { |
| 216 | + pub column_index: IndexType, |
| 217 | + pub name: String, |
| 218 | + pub data_type: DataTypeImpl, |
| 219 | + |
| 220 | + /// Table index of column entry. None if column is derived from a subquery. |
| 221 | + pub table_index: Option<IndexType>, |
| 222 | + /// Path indices for inner column of struct data type. |
| 223 | + pub path_indices: Option<Vec<IndexType>>, |
| 224 | +} |
| 225 | + |
| 226 | +impl ColumnEntry { |
| 227 | + pub fn new( |
| 228 | + name: String, |
| 229 | + data_type: DataTypeImpl, |
| 230 | + column_index: IndexType, |
| 231 | + table_index: Option<IndexType>, |
| 232 | + path_indices: Option<Vec<IndexType>>, |
| 233 | + ) -> Self { |
| 234 | + ColumnEntry { |
| 235 | + column_index, |
| 236 | + name, |
| 237 | + data_type, |
| 238 | + table_index, |
| 239 | + path_indices, |
| 240 | + } |
| 241 | + } |
| 242 | +} |
| 243 | + |
| 244 | +/// TODO(xuanwo): migrate this as a function of metadata. |
| 245 | +pub fn find_smallest_column(entries: &[ColumnEntry]) -> usize { |
| 246 | + debug_assert!(!entries.is_empty()); |
| 247 | + let mut column_indexes = entries |
| 248 | + .iter() |
| 249 | + .map(|entry| entry.column_index) |
| 250 | + .collect::<Vec<IndexType>>(); |
| 251 | + column_indexes.sort(); |
| 252 | + let mut smallest_index = column_indexes[0]; |
| 253 | + let mut smallest_size = usize::MAX; |
| 254 | + for (idx, column_entry) in entries.iter().enumerate() { |
| 255 | + if let Ok(bytes) = column_entry.data_type.data_type_id().numeric_byte_size() { |
| 256 | + if smallest_size > bytes { |
| 257 | + smallest_size = bytes; |
| 258 | + smallest_index = entries[idx].column_index; |
| 259 | + } |
| 260 | + } |
| 261 | + } |
| 262 | + smallest_index |
| 263 | +} |
0 commit comments