Sets a DuckLake configuration option, either lake-wide or scoped to a schema or table. Options are persisted in the metadata catalog, so they survive detach/attach cycles and apply to every client of the lake.
Usage
set_ducklake_option(
option,
value,
table_name = NULL,
schema_name = NULL,
ducklake_name = NULL
)Arguments
- option
Name of the option, e.g.
"parquet_compression","target_file_size","sort_on_insert", or"data_inlining_row_limit". See https://ducklake.select/docs/stable/duckdb/usage/configuration for the full list.- value
The value to set. Logicals are rendered as
true/false, numbers as numeric literals, and everything else as a quoted string.- table_name
Optional table name to scope the option to one table, optionally qualified as
"schema.table".- schema_name
Optional schema name to scope the option to one schema (or, together with
table_name, to qualify the table).- ducklake_name
Optional name of the attached DuckLake catalog. If
NULL, the current database is used.
Details
Table-scoped settings override schema-scoped ones, which override the
lake-wide default. Runs CALL <lake>.set_option(...).
The options DuckLake 1.0 persists, with their defaults:
| Option | Default | What it controls |
auto_compact | true | Whether maintenance calls made without a table argument include the table |
data_inlining_row_limit | 10 | Rows below which an insert or delete is stored in the catalog instead of a file (see set_inlining_row_limit()) |
delete_older_than | unset | How long a released file waits before cleanup_old_files() and checkpoints delete it |
expire_older_than | unset | How old a snapshot must be before checkpoints expire it |
encrypted | false | Encrypt the Parquet files written to the data path (set at creation; see attach_ducklake(encrypted = )) |
hive_file_pattern | true | Write partitioned data in Hive-style directories |
parquet_compression | snappy | Codec: uncompressed, snappy, gzip, zstd, brotli, lz4, or lz4_raw |
parquet_compression_level | 3 | Level for codecs that have one |
parquet_row_group_size | 122880 | Rows per row group |
parquet_row_group_size_bytes | unset | Bytes per row group, as an alternative to rows |
parquet_version | 1 | Parquet format version, 1 or 2 |
per_thread_output | false | One output file per thread during a parallel insert |
require_commit_message | false | Refuse to commit a snapshot without a commit message |
rewrite_delete_threshold | 0.95 | Deleted fraction of a file above which rewrite_data_files() rewrites it |
sort_on_insert | true | Sort inserted rows by the table's sort keys (see set_table_sorting()) |
target_file_size | 512MB | Target data file size for inserts and compaction |
write_deletion_vectors | false | Write Iceberg V3 deletion vectors instead of positional delete files |
created_by, data_path, and version also appear in
get_ducklake_options() but describe the lake rather than configure it.
Retention settings (expire_older_than, delete_older_than) take
interval strings such as "90 days".
See also
get_ducklake_options(), set_inlining_row_limit()
Other options:
get_ducklake_options()
Examples
lake_dir <- tempfile("setopt_lake_")
dir.create(lake_dir)
attach_ducklake("setopt_lake", lake_path = lake_dir)
create_table(mtcars, "cars")
# Smaller files at some write cost, lake-wide
set_ducklake_option("parquet_compression", "zstd")
#> Option "parquet_compression" set to "zstd" for lake "setopt_lake".
# Skip one table during compaction
set_ducklake_option("auto_compact", FALSE, table_name = "cars")
#> Option "auto_compact" set to FALSE for table "cars".
# Make every snapshot carry a commit message (constrains later writes)
set_ducklake_option("require_commit_message", TRUE)
#> Option "require_commit_message" set to TRUE for lake "setopt_lake".
detach_ducklake("setopt_lake", shutdown = TRUE)
unlink(lake_dir, recursive = TRUE)