yiguolei commented on code in PR #66472:
URL: https://github.com/apache/doris/pull/66472#discussion_r3757501820
##########
be/src/storage/schema.h:
##########
@@ -37,110 +39,181 @@
namespace doris {
-// The class is used to represent row's format in memory. Each row contains
-// multiple columns, some of which are key-columns (the rest are
value-columns).
-// NOTE: If both key-columns and value-columns exist, then the key-columns
-// must be placed before value-columns.
-//
-// To compare two rows whose schemas are different, but they are from the same
origin
-// we store all column schema maybe accessed here. And default access through
column id
-class Schema;
-using SchemaSPtr = std::shared_ptr<const Schema>;
-class Schema {
+class ReadSchema;
+class Block;
+using ReadSchemaSPtr = std::shared_ptr<ReadSchema>;
+class ReadSchema {
public:
- // All the columns of one table may exist in the columns param, but
col_ids is only a subset.
- Schema(const std::vector<TabletColumnPtr>& columns, const
std::vector<ColumnId>& col_ids) {
- size_t num_key_columns = 0;
- for (int i = 0; i < columns.size(); ++i) {
- if (columns[i]->is_key()) {
- ++num_key_columns;
- }
- if (columns[i]->name() == DELETE_SIGN) {
- _delete_sign_idx = i;
- }
- if (columns[i]->name() == BeConsts::ROWID_COL ||
- columns[i]->name().starts_with(BeConsts::GLOBAL_ROWID_COL)) {
- _rowid_col_idx = i;
- }
- if (columns[i]->name() == VERSION_COL) {
- _version_col_idx = i;
- }
- if (columns[i]->name() == BINLOG_TSO_COL) {
- _tso_col_idx = i;
- }
- if (columns[i]->name() == BINLOG_LSN_COL) {
- _lsn_col_idx = i;
- }
- if (columns[i]->name() == BINLOG_OP_COL) {
- _op_col_idx = i;
- }
- if (columns[i]->name() == COMMIT_TSO_COL) {
- _commit_tso_col_idx = i;
- }
- }
- _init(columns, col_ids, num_key_columns);
- }
+ using SequenceMap = std::unordered_map<ColumnId, std::vector<ColumnId>>;
- Schema(const Schema&);
- Schema& operator=(const Schema& other);
+ explicit ReadSchema(std::vector<TabletColumnPtr> columns);
- ~Schema();
+ explicit ReadSchema(const std::vector<TabletColumnPtr>& columns,
+ const std::vector<ColumnId>& cids);
- static DataTypePtr get_data_type_ptr(const TabletColumn& column);
+ // Initially every column is a caller-visible FE slot. Storage-only columns
+ // may be appended later without changing `num_block_columns()`.
+ explicit ReadSchema(std::vector<TabletColumnPtr> columns,
std::vector<DataTypePtr> read_types);
static IColumn::MutablePtr get_predicate_column_ptr(const DataTypePtr&
data_type,
const ReaderType
reader_type);
- const std::vector<TabletColumnPtr>& columns() const { return _cols; }
+ const std::vector<TabletColumnPtr>& columns() const { return
_read_columns; }
- const TabletColumn* column(ColumnId cid) const { return _cols[cid].get(); }
+ DataTypePtr data_type(size_t ordinal) const {
+ DCHECK_LT(ordinal, _read_types.size());
+ return _read_types[ordinal];
+ }
+
+ // Append a storage-only column without extending the caller-visible slot
prefix.
+ ColumnId append_column(TabletColumnPtr column) {
+ auto data_type = column->get_vec_type();
+ auto ordinal = cast_set<ColumnId>(_read_columns.size());
+ if (column->unique_id() >= 0) {
+ _uid_to_ordinal.emplace(column->unique_id(), ordinal);
+ }
+ _read_columns.emplace_back(std::move(column));
+ _read_types.emplace_back(std::move(data_type));
+ return ordinal;
+ }
+
+ // Create caller-visible Blocks from the FE-slot prefix.
+ Block create_read_block() const;
+
+ std::string read_columns_to_string() const;
+
+ Status init_sequence_map(const TabletSchema& tablet_schema);
+
+ const SequenceMap& sequence_map() const { return _sequence_map; }
+
+ ColumnId before_column_ordinal(ColumnId ordinal) const {
+ DCHECK_LT(ordinal, _before_column_ordinals.size());
+ return _before_column_ordinals[ordinal];
+ }
+
+ const TabletColumn* column(size_t ordinal) const { return
_read_columns[ordinal].get(); }
+
+ // Resolve by unique id when one exists. Name identity is only for legacy
+ // columns without a unique id.
+ int32_t ordinal_by_column(const TabletColumn& column) const {
+ if (column.unique_id() >= 0) {
+ return ordinal_by_uid(column.unique_id());
+ }
+ for (uint32_t ordinal = 0; ordinal < _read_columns.size(); ++ordinal) {
+ if (_read_columns[ordinal]->name() == column.name()) {
+ return static_cast<int32_t>(ordinal);
+ }
+ }
+ return -1;
+ }
+
+ // Total columns used inside storage, including appended storage-only
columns.
+ // Use this for per-column state and iteration over the complete
ReadSchema.
+ size_t num_read_columns() const { return _read_columns.size(); }
+
+ // Columns materialized in caller Blocks. They are the ReadSchema prefix
before
Review Comment:
这个read schema 里的注释，都得举个例子，否则这种英语很难理解
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
