sqlx-exasol

A database driver for Exasol to be used with the Rust sqlx framework, based on the Exasol Websocket API.
Inspired by Py-Exasol and based on the (now archived) rust-exasol sync driver.

MSRV: 1.74

Note

The crate's version resembles the sqlx version it is based on so that managing dependencies is simpler.

With that in mind, please favor using a fixed version of sqlx and sqlx-exasol in Cargo.toml to avoid issues, such as:

sqlx = "=0.7.4"
sqlx-exasol = "=0.7.4"

Crate Features flags

• etl - enables the usage ETL jobs without TLS encryption.
• etl_native_tls - enables the etl feature and adds TLS encryption through native-tls¹
• etl_rustls - enables the etl feature and adds TLS encryption through rustls¹
• compression - enables compression support (for both connections and ETL jobs)
• uuid - enables support for the uuid crate
• chrono - enables support for the chrono crate types
• rust_decimal - enables support for the rust_decimal type
• migrate - enables the use of migrations and testing (just like in other sqlx drivers).

Comparison to native sqlx drivers

Since the driver is used through sqlx and it implements the interfaces there, it can do all the drivers shipped with sqlx do, with some caveats:

• Limitations

  • no compile-time query check support²
  • no sqlx-cli support²
  • no locking migrations support³
  • no column nullability checks⁴
  • apart from migrations, only a single query per statement is allowed (including in fixtures)⁵

• Additions

  • array-like parameter binding in queries, thanks to the columnar nature of the Exasol database
  • performant & parallelizable ETL IMPORT/EXPORT jobs in CSV format through HTTP Transport

Connection string

The connection string is expected to be an URL with the exa:// scheme, e.g: exa://sys:exasol@localhost:8563.

Examples

Using the driver for regular database interactions:

use std::env;
use sqlx_exasol::*;

let pool = ExaPool::connect(&env::var("DATABASE_URL").unwrap()).await?;
let mut con = pool.acquire().await?;

sqlx::query("CREATE SCHEMA RUST_DOC_TEST")
    .execute(&mut *con)
    .await?;

Array-like parameter binding, also featuring the ExaIter adapter. An important thing to note is that the parameter sets must be of equal length, otherwise an error is thrown:

use std::{collections::HashSet, env};
use sqlx_exasol::*;

let pool = ExaPool::connect(&env::var("DATABASE_URL").unwrap()).await?;
let mut con = pool.acquire().await?;

let params1 = vec![1, 2, 3];
let params2 = HashSet::from([1, 2, 3]);

sqlx::query("INSERT INTO MY_TABLE VALUES (?, ?)")
    .bind(&params1)
    .bind(ExaIter::from(&params2))
    .execute(&mut *con)
    .await?;

An EXPORT - IMPORT ETL data pipe.

use std::env;
use futures_util::{
    future::{try_join, try_join3, try_join_all},
    AsyncReadExt, AsyncWriteExt, TryFutureExt,
};
use sqlx_exasol::{etl::*, *};

async fn pipe(mut reader: ExaExport, mut writer: ExaImport) -> anyhow::Result<()> {
    let mut buf = vec![0; 5120].into_boxed_slice();
    let mut read = 1;

    while read > 0 {
        // Readers return EOF when there's no more data.
        read = reader.read(&mut buf).await?;
        // Write data to Exasol
        writer.write_all(&buf[..read]).await?;
    }

    // Writes, unlike readers, MUST be closed to signal we won't send more data to Exasol
    writer.close().await?;
    Ok(())
}

let pool = ExaPool::connect(&env::var("DATABASE_URL").unwrap()).await?;
let mut con1 = pool.acquire().await?;
let mut con2 = pool.acquire().await?;

// Build EXPORT job
let (export_fut, readers) = ExportBuilder::new(ExportSource::Table("TEST_ETL"))
    .build(&mut con1)
    .await?;

// Build IMPORT job
let (import_fut, writers) = ImportBuilder::new("TEST_ETL").build(&mut con2).await?;

// Use readers and writers in some futures
let transport_futs = std::iter::zip(readers, writers).map(|(r, w)| pipe(r, w));

// Execute the EXPORT and IMPORT query futures along with the worker futures
let (export_res, import_res, _) = try_join3(
    export_fut.map_err(From::from),
    import_fut.map_err(From::from),
    try_join_all(transport_futs),
)
.await?;

assert_eq!(export_res.rows_affected(), import_res.rows_affected());

License

Licensed under either of

• Apache License, Version 2.0, (LICENSE-APACHE or https://www.apache.org/licenses/LICENSE-2.0)
• MIT license (LICENSE-MIT or https://opensource.org/licenses/MIT)

at your option.

Contributing

Contributions to this repository, unless explicitly stated otherwise, will be considered dual-licensed under MIT and Apache 2.0.
Bugs/issues encountered can be opened here

Footnotes

1: There is unfortunately no way to automagically choose a crate's feature flags based on its dependencies feature flags, so the TLS backend has to be manually selected. While nothing prevents you from using, say native-tls with sqlx and rustls with Exasol ETL jobs, it might be best to avoid compiling two different TLS backends. Therefore, consider choosing the sqlx and sqlx-exasol feature flags in a consistent manner.

2: The sqlx API powering the compile-time query checks and the sqlx-cli tool is not public. Even if it were, the drivers that are incorporated into sqlx are hardcoded in the part of the code that handles the compile-time driver decision logic.
The main problem from what I can gather is that there's no easy way of defining a plugin system in Rust at the moment, hence the hardcoding.

3: Exasol has no advisory or database locks and simple, unnested, transactions are unfortunately not enough to define a mechanism so that concurrent migrations do not collide. This does not pose a problem when migrations are run sequentially or do not act on the same database objects.

4: Exasol does not provide the information of whether a column is nullable or not, so the driver cannot implicitly decide whether a NULL value can go into a certain database column or not until it actually tries.

5: I didn't even know this (as I never even thought of doing it), but sqlx allows running multiple queries in a single statement. Due to limitations with the websocket API this driver is based on, sqlx-exasol can only run one query at a time.
This is only circumvented in migrations through a somewhat limited, convoluted and possibly costly workaround that tries to split queries by ;, which does not make it applicable for runtime queries at all.