Split into Files

Data > Split into files... (Ctrl+Shift+Z) writes the active table as several files: one per value of a column, at most so many rows each, at most so much size each, or a mix.

The three limits

Choose at least one of these; they combine.

• Column: one file per distinct value, named after the value. Splitting a sales table by region gives north.csv, south.csv, and so on. Leave it at (none) to split the whole table by rows or size only.
• Max rows per file: no file holds more rows than this. Type a whole number; 10,000 works.
• Max size per file (MB): no file is bigger than this (1 MB = 1,000,000 bytes; decimals such as 2.5 work). Octa writes a file, measures it and rewrites it with fewer rows if it came out too big, so every format works, compressed Parquet included. Files land a little under the limit rather than packed to the byte.

With a column and a limit, each value is split further. With only a limit the files are named after the source file.

Names

Without a row or size limit the names are the plain ones in the table below. With a limit every file gets a part number: flat files become berlin_part_0001.csv, berlin_part_0002.csv, ..., and the folder layouts number the files inside their folder (city=Berlin/part-0001.csv). The numbers count per folder, so each folder starts at part-0001. The dialog previews the first file of the first few values with the same naming code the writer uses.

Big tabs

If the tab holds only part of its source, the dialog says so and offers Load the rest; a split writes only the rows that are loaded. A tab that shows one page of a very large file is refused with a pointer to octa --split-rows and octa --split-size, which stream the whole file.

Running

The split runs in the background with a progress line and Cancel. Cancel stops after the file being written; files already written stay. The original table is not changed.

Command line and MCP

Also available as octa --partition-by, --split-rows and --split-size (see the --partition-by reference) and as the partition_table MCP / assistant tool.

Choosing how the pieces are named

The Layout option offers four shapes. Taking city with values New York, Berlin and Sao Paulo, and writing CSV:

Layout	On disk
Flat files	new_york.csv, berlin.csv, sao_paulo.csv
Folder per value	New York/part-0001.csv, Berlin/part-0001.csv, ...
Hive folders	city=New York/data.csv, city=Berlin/data.csv, ...
Hive folders, numbered files	city=New York/part-0001.csv, ...

All four hold the same rows, and all four reopen as one table with File > Open table folder..., because the column you split on is written into every file whichever you pick. The choice is only about the names - but the names are not merely cosmetic:

• Flat is the only lossy one. The value goes through the same tidy-up a SQL identifier gets: capitals fold to lower case and anything that is not a letter or digit becomes an underscore. So New York, new-york and NEW_YORK all arrive as new_york, and the second and third become new_york_2.csv and new_york_3.csv. No rows are lost, but the name no longer tells you which value is inside. An empty value becomes table.csv and one starting with a digit gains a t_ prefix.
• The three folder layouts keep the value intact, replacing only characters that cannot appear in a path at all.
• Hive (column=value) is what Spark, Athena, DuckDB and pandas expect from a partitioned dataset. Choose it when the files are going into another tool.
• The numbered variants name the file part-0001 instead of after the value or data, matching what those tools write themselves. Some pipelines expect that shape.

You do not have to work any of this out from the descriptions: the dialog shows a live preview of the first few paths, built from your own column's values with the same code that writes the files. Change the column, the format or the layout and the preview follows.

Flat is the default, so nothing changes unless you pick otherwise.