The lightweight `DELETE` statement removes rows from the table `[db.]table` that match the expression `expr`. It is only available for the *MergeTree table engine family.
It is called "lightweight `DELETE`" to contrast it to the [ALTER TABLE ... DELETE](/en/sql-reference/statements/alter/delete) command, which is a heavyweight process.
Lightweight `DELETE` is implemented as a [mutation](/en/sql-reference/statements/alter#mutations) that marks rows as deleted but does not immediately physically delete them.
By default, `DELETE` statements wait until marking the rows as deleted is completed before returning. This can take a long time if the amount of data is large. Alternatively, you can run it asynchronously in the background using the setting [`lightweight_deletes_sync`](/en/operations/settings/settings#lightweight_deletes_sync). If disabled, the `DELETE` statement is going to return immediately, but the data can still be visible to queries until the background mutation is finished.
The mutation does not physically delete the rows that have been marked as deleted, this will only happen during the next merge. As a result, it is possible that for an unspecified period, data is not actually deleted from storage and is only marked as deleted.
If you need to guarantee that your data is deleted from storage in a predictable time, consider using the table setting [`min_age_to_force_merge_seconds`](https://clickhouse.com/docs/en/operations/settings/merge-tree-settings#min_age_to_force_merge_seconds). Or you can use the [ALTER TABLE ... DELETE](/en/sql-reference/statements/alter/delete) command. Note that deleting data using `ALTER TABLE ... DELETE` may consume significant resources as it recreates all affected parts.
Large deletes can negatively affect ClickHouse performance. If you are attempting to delete all rows from a table, consider using the [`TRUNCATE TABLE`](/en/sql-reference/statements/truncate) command.
If you anticipate frequent deletes, consider using a [custom partitioning key](/en/engines/table-engines/mergetree-family/custom-partitioning-key). You can then use the [`ALTER TABLE ... DROP PARTITION`](/en/sql-reference/statements/alter/partition#drop-partitionpart) command to quickly drop all rows associated with that partition.
By default, `DELETE` does not work for tables with projections. This is because rows in a projection may be affected by a `DELETE` operation. But there is a [MergeTree setting](https://clickhouse.com/docs/en/operations/settings/merge-tree-settings) `lightweight_mutation_projection_mode` to change the behavior.
## Performance considerations when using lightweight `DELETE`
**Deleting large volumes of data with the lightweight `DELETE` statement can negatively affect SELECT query performance.**
The following can also negatively impact lightweight `DELETE` performance:
- A heavy `WHERE` condition in a `DELETE` query.
- If the mutations queue is filled with many other mutations, this can possibly lead to performance issues as all mutations on a table are executed sequentially.
When a `DELETE FROM table ...` query is executed, ClickHouse saves a mask where each row is marked as either “existing” or as “deleted”. Those “deleted” rows are omitted for subsequent queries. However, rows are actually only removed later by subsequent merges. Writing this mask is much more lightweight than what is done by an `ALTER TABLE ... DELETE` query.
The mask is implemented as a hidden `_row_exists` system column that stores `True` for all visible rows and `False` for deleted ones. This column is only present in a part if some rows in the part were deleted. This column does not exist when a part has all values equal to `True`.
When a masked column is used in a query, the `SELECT ... FROM table WHERE condition` query internally is extended by the predicate on `_row_exists` and is transformed to:
```sql
SELECT ... FROM table PREWHERE _row_exists WHERE condition
```
At execution time, the column `_row_exists` is read to determine which rows should not be returned. If there are many deleted rows, ClickHouse can determine which granules can be fully skipped when reading the rest of the columns.
2. Based on the commands above, affected parts are then mutated, and hardlinks are created for unaffected parts. In the case of wide parts, the `_row_exists` column for each row is updated, and all other columns' files are hardlinked. For compact parts, all columns are re-written because they are all stored together in one file.
From the steps above, we can see that lightweight `DELETE` using the masking technique improves performance over traditional `ALTER TABLE ... DELETE` because it does not re-write all the columns' files for affected parts.