Crate ursa_engine 

Ursa’s SQL engine session.

new_session builds a DataFusion SessionContext on which a file is a table: SELECT * FROM 'edges.parquet', 'edges.csv.gz', 'edges/', 'edges/*.parquet', 'events.jsonl', and the same over s3://, gs://, az:// and https://, with object stores registered the first time a URL is read. read_parquet, read_csv and read_json read the same with options (read_functions).

Graphs are defined per session (graph). A session may have a .ursa database (database), which keeps its catalog between processes.

Re-exports

pub use error::user_message;

pub use stores::ensure_object_store;

pub use stores::ensure_object_store_for;

Modules

database

The .ursa database file: a session’s catalog, kept between processes.

error

Error messages for people, not programmers.

graph

Graphs in a session: their definitions and where they are kept.

options

Ursa’s session settings, set with SET ursa.<name> = … and shown with SHOW ursa.<name>: a DataFusion config extension with the prefix ursa.

paths

Turning a name in FROM into a file location.

read_functions

read_parquet(path), read_csv(path, header => false) and read_json(path): files as tables, as in DuckDB.

serial

Plans that run on one task, for the browser (#278).

stores

Object stores, registered the first time a query names a URL.

Structs

BlockingError

Why blocking work produced no result: it panicked or was cancelled. Only a native build can see one; inline work on wasm32 cannot fail this way.

FileOptions

How to read files, beyond what their extension says: the options of read_csv(...) and its siblings. None leaves a choice to the extension or to the session’s defaults.

Enums

FileFormat

A file format Ursa reads.

Functions

new_session

A session with Ursa’s defaults.

new_session_with_config

A session with Ursa’s file tables over config. information_schema is turned on, so SHOW TABLES and SHOW COLUMNS work.

open_file_table

Open location (a path, glob, directory or URL) as a listing table. The format and compression come from the file extension; Hive-style partition directories (year=2024/) become columns. A relative path is relative to the current directory, in a session with a database too: a stored definition’s paths are made relative to the database’s directory when it is stored, and resolved against it on replay (see crate::database).

open_files

Open locations (paths, globs, directories or URLs) as one listing table, read as options say. What options leaves open comes from the extension of the first file found, as DataFusion infers it; a directory or glob lists only the files with that extension. Each location must name a file.

run_blocking

Run f off the async workers (native), or inline (wasm32).