Delete files contain the positions of rows that are deleted within a data file. Each data file can have its own delete file if any deletes are present for this data file.

• delete_file_id is the numeric identifier of the delete file. It is a primary key. delete_file_id is incremented from next_file_id in the ducklake_snapshot table.
• table_id refers to a table_id from the ducklake_table table.
• begin_snapshot refers to a snapshot_id from the ducklake_snapshot table. The delete file is part of the table starting with this snapshot id. For a partial delete file, this is the earliest deletion snapshot it contains.
• end_snapshot refers to a snapshot_id from the ducklake_snapshot table. The delete file is part of the table up to but not including this snapshot id. If end_snapshot is NULL, the delete file is currently part of the table.
• data_file_id refers to a data_file_id from the ducklake_data_file table.
• path is the file name of the delete file, e.g., my_file-deletes.parquet for a relative path.
• path_is_relative whether the path is relative to the path of the table (true) or an absolute path (false).
• format is the storage format of the delete file. Supported values are parquet (positional delete file) and puffin (Iceberg V3 deletion vector, experimental).
• delete_count is the number of distinct deleted row positions at the latest snapshot represented in the file. For cumulative Puffin vectors, this is the latest vector's count, not the sum of the vectors.
• file_size_bytes is the size of the file in bytes.
• footer_size is the size of the file metadata footer, in the case of Parquet the Thrift data. This is an optimization that allows for faster reading of the file. For a Puffin container, it includes the JSON payload and the 16-byte footer structure. A bare deletion vector has no footer and stores 0.
• encryption_key contains the encryption for the file if encryption is enabled. Puffin files are written without encryption, even if this field is populated.
• partial_max is the maximum snapshot id stored in a partial deletion file. When multiple deletes target the same data file across different snapshots, the deletion file is rewritten as a partial deletion file that tracks which rows were deleted in which snapshot. This column stores the highest snapshot id present in such a file. It is NULL for standard (non-partial) delete files.
• row_group_count is the number of Parquet row groups in the delete file, or NULL for Puffin files or when unknown.

DuckLake format version 1.1 supports Puffin files with cumulative deletion vectors for several snapshots. Each vector's ducklake-snapshot-id property identifies its snapshot. Readers select the vector with the greatest snapshot id that does not exceed the requested snapshot. A file for a single deletion snapshot can contain a bare deletion-vector-v1 blob, whose snapshot is given by begin_snapshot.

A replacement delete file preserves earlier deletions and their snapshot information. The old row is removed from ducklake_delete_file, and its file is added to ducklake_files_scheduled_for_deletion.