Function query_parquet

datafusion-examples/examples/sql_ops/query.rs:113–217

The simplest way to query parquet files is to use the [`SessionContext::read_parquet`] API For more control, you can use the lower level [`ListingOptions`] and [`ListingTable`] APIS This example shows how to use relative and absolute paths. [`ListingTable`]: datafusion::datasource::listing::ListingTable

()

///
/// [`ListingTable`]: datafusion::datasource::listing::ListingTable
async fn query_parquet() -> Result<()> {
    // create local execution context
    let ctx = SessionContext::new();

    // Convert the CSV input into a temporary Parquet directory for querying
    let dataset = ExampleDataset::Cars;
    let parquet_temp = write_csv_to_parquet(&ctx, &dataset.path()).await?;

    // Configure listing options
    let file_format = ParquetFormat::default().with_enable_pruning(true);
    let listing_options =
        ListingOptions::new(Arc::new(file_format)).with_file_extension(".parquet");

    let table_path = parquet_temp.file_uri()?;

    // First example were we use an absolute path, which requires no additional setup.
    ctx.register_listing_table(
        "my_table",
        &table_path,
        listing_options.clone(),
        None,
        None,
    )
    .await?;

    // execute the query
    let df = ctx
        .sql(
            "SELECT * \
        FROM my_table \
        ORDER BY speed \
        LIMIT 1",
        )
        .await?;

    // print the results
    let results = df.collect().await?;
    assert_batches_eq!(
        [
            "+-----+-------+---------------------+",
            "| car | speed | time                |",
            "+-----+-------+---------------------+",
            "| red | 0.0   | 1996-04-12T12:05:15 |",
            "+-----+-------+---------------------+",
        ],
        &results
    );

    // Second example where we change the current working directory and explicitly
    // register a local filesystem object store. This demonstrates how listing tables
    // resolve paths via an ObjectStore, even when using filesystem-backed data.
    let cur_dir = std::env::current_dir()?;
    let test_data_path_parent = parquet_temp
        .tmp_dir
        .path()
        .parent()
        .ok_or(exec_datafusion_err!("test_data path needs a parent"))?;