Output node
Parquet File
Write the incoming rows to a columnar Parquet file.
What it does
Writes the incoming rows to a compressed, columnar Parquet file that you download from the run afterwards. Parquet stores the schema alongside the data, so every column comes back with the type it left with: nothing is re-guessed and no leading zeros are lost. Pick a compression codec to trade file size against write speed.
Inputs and outputs
| Inputs | One. |
| Outputs | None. This node ends a pipeline. |
When to use it
- Store a large result efficiently
- Hand off to a data lake, a warehouse loader, or a later pipeline
- Keep exact column types across runs, where CSV would lose them
Examples
Default
Compression: snappy
Fast to write and well compressed: the right answer unless you know otherwise.
Smaller file
Compression: gzip
Noticeably smaller, noticeably slower. Worth it when you're paying to move the file.
Uncompressed
Compression: none
Only for a reader that can't decompress. Rare.
Between pipelines
Parquet File → a later run's Parquet Files input
Types survive the round trip exactly, which CSV can't promise.
Tips
- Find the file on the run's detail page once the run finishes.
- A Test run leaves you nothing to download; press Run for a real file.
- A run producing ZERO rows writes no file at all: Parquet needs a schema to write its footer and there is nothing to take one from, so an empty result is an absent file rather than an empty one.
- Snappy is the best default. gzip only pays off when you're moving the file over a slow link or paying for the storage.
- This is the format to use between Trinium pipelines: it's the only one here that carries types.