Table maintenance

[maintenance]

​

Settings

​

target_file_size_bytes

Chooses the approximate output file size when compacting a table.

• Default: 134217728
• Type: integer (bytes); positive

Example 134217728 targets files around 128 MiB. Batch and row-group boundaries mean individual files may be larger or smaller.

​

max_group_size_bytes

Limits how many input bytes are grouped into one maintenance task.

• Default: 1073741824
• Type: integer (bytes); positive

Example 1073741824 groups at most 1 GiB of input at a time. It does not set the size of every output file.

​

min_input_files

Sets the minimum number of candidate files needed before compaction is planned.

• Default: 2
• Type: integer; positive

Example 2 avoids planning compaction for a single candidate file.

​

target_metadata_size_bytes

Sets the target size for metadata output produced by maintenance.

• Default: 8388608
• Type: integer (bytes); positive

Example 8388608 targets about 8 MiB of metadata output. This is separate from the target size of table data files.

​

Task limits

​

limits.max_input_files

Limits how many files one maintenance task may read.

• Default: 100000
• Type: integer; positive

Example 100000 bounds the input-file count for a task, independently of the total input bytes.

​

limits.max_input_bytes

Limits the combined input bytes admitted for one maintenance task.

• Default: 1099511627776
• Type: integer (bytes); positive

Example 1099511627776 is a 1 TiB input budget. A task must also fit the file-count and memory limits.

​

limits.max_output_files

Limits how many files one maintenance task may produce.

• Default: 100000
• Type: integer; positive

Example 100000 prevents maintenance from producing an unbounded number of small output files.

​

limits.max_memory_bytes

Limits the memory budget admitted for one maintenance task.

• Default: 268435456
• Type: integer (bytes); positive

Example 268435456 gives that maintenance work a 256 MiB memory budget. It does not resize the whole worker process.

​

limits.max_result_rows

Limits how many result rows a maintenance task may retain.

• Default: 100000
• Type: integer; positive

Example 100000 bounds a task’s returned maintenance result rows, not the rows stored in the whole table.

​

limits.max_result_bytes

Limits the memory occupied by retained maintenance result arrays before they are encoded for storage.

• Default: 16777216
• Type: integer (bytes); positive

Example 16777216 gives those Arrow arrays a 16 MiB budget. Their encoded result-object size is controlled separately.