feat: {add|alter|drop}_columns APIs (#1015)

Initial work for #959. This exposes the basic functionality for each in
all of the APIs. Will add user guide documentation in a later PR.
This commit is contained in:
Will Jones
2024-02-26 11:04:53 -08:00
committed by GitHub
parent 8a52619bc0
commit 5af74b5aca
14 changed files with 660 additions and 11 deletions

View File

@@ -17,7 +17,7 @@ import * as path from "path";
import * as fs from "fs";
import { connect } from "../dist";
import { Schema, Field, Float32, Int32, FixedSizeList } from "apache-arrow";
import { Schema, Field, Float32, Int32, FixedSizeList, Int64, Float64 } from "apache-arrow";
import { makeArrowTable } from "../dist/arrow";
describe("Test creating index", () => {
@@ -214,4 +214,69 @@ describe("Read consistency interval", () => {
expect(await table2.countRows()).toEqual(2n);
}
});
});
describe('schema evolution', function () {
let tmpDir: string;
beforeEach(() => {
tmpDir = fs.mkdtempSync(path.join(os.tmpdir(), "schema-evolution-"));
});
// Create a new sample table
it('can add a new column to the schema', async function () {
const con = await connect(tmpDir)
const table = await con.createTable('vectors', [
{ id: 1n, vector: [0.1, 0.2] }
])
await table.addColumns([{ name: 'price', valueSql: 'cast(10.0 as float)' }])
const expectedSchema = new Schema([
new Field('id', new Int64(), true),
new Field('vector', new FixedSizeList(2, new Field('item', new Float32(), true)), true),
new Field('price', new Float32(), false)
])
expect(await table.schema()).toEqual(expectedSchema)
});
it('can alter the columns in the schema', async function () {
const con = await connect(tmpDir)
const schema = new Schema([
new Field('id', new Int64(), true),
new Field('vector', new FixedSizeList(2, new Field('item', new Float32(), true)), true),
new Field('price', new Float64(), false)
])
const table = await con.createTable('vectors', [
{ id: 1n, vector: [0.1, 0.2] }
])
// Can create a non-nullable column only through addColumns at the moment.
await table.addColumns([{ name: 'price', valueSql: 'cast(10.0 as double)' }])
expect(await table.schema()).toEqual(schema)
await table.alterColumns([
{ path: 'id', rename: 'new_id' },
{ path: 'price', nullable: true }
])
const expectedSchema = new Schema([
new Field('new_id', new Int64(), true),
new Field('vector', new FixedSizeList(2, new Field('item', new Float32(), true)), true),
new Field('price', new Float64(), true)
])
expect(await table.schema()).toEqual(expectedSchema)
});
it('can drop a column from the schema', async function () {
const con = await connect(tmpDir)
const table = await con.createTable('vectors', [
{ id: 1n, vector: [0.1, 0.2] }
])
await table.dropColumns(['vector'])
const expectedSchema = new Schema([
new Field('id', new Int64(), true)
])
expect(await table.schema()).toEqual(expectedSchema)
});
});

View File

@@ -12,6 +12,38 @@ export const enum MetricType {
Cosine = 1,
Dot = 2
}
/**
* A definition of a column alteration. The alteration changes the column at
* `path` to have the new name `name`, to be nullable if `nullable` is true,
* and to have the data type `data_type`. At least one of `rename` or `nullable`
* must be provided.
*/
export interface ColumnAlteration {
/**
* The path to the column to alter. This is a dot-separated path to the column.
* If it is a top-level column then it is just the name of the column. If it is
* a nested column then it is the path to the column, e.g. "a.b.c" for a column
* `c` nested inside a column `b` nested inside a column `a`.
*/
path: string
/**
* The new name of the column. If not provided then the name will not be changed.
* This must be distinct from the names of all other columns in the table.
*/
rename?: string
/** Set the new nullability. Note that a nullable column cannot be made non-nullable. */
nullable?: boolean
}
/** A definition of a new column to add to a table. */
export interface AddColumnsSql {
/** The name of the new column. */
name: string
/**
* The values to populate the new column with, as a SQL expression.
* The expression can reference other columns in the table.
*/
valueSql: string
}
export interface ConnectionOptions {
uri: string
apiKey?: string
@@ -89,4 +121,7 @@ export class Table {
delete(predicate: string): Promise<void>
createIndex(): IndexBuilder
query(): Query
addColumns(transforms: Array<AddColumnsSql>): Promise<void>
alterColumns(alterations: Array<ColumnAlteration>): Promise<void>
dropColumns(columns: Array<string>): Promise<void>
}

View File

@@ -13,7 +13,7 @@
// limitations under the License.
import { Schema, tableFromIPC } from "apache-arrow";
import { Table as _NativeTable } from "./native";
import { AddColumnsSql, ColumnAlteration, Table as _NativeTable } from "./native";
import { toBuffer, Data } from "./arrow";
import { Query } from "./query";
import { IndexBuilder } from "./indexer";
@@ -150,4 +150,42 @@ export class Table {
}
return q;
}
// TODO: Support BatchUDF
/**
* Add new columns with defined values.
*
* @param newColumnTransforms pairs of column names and the SQL expression to use
* to calculate the value of the new column. These
* expressions will be evaluated for each row in the
* table, and can reference existing columns in the table.
*/
async addColumns(newColumnTransforms: AddColumnsSql[]): Promise<void> {
await this.inner.addColumns(newColumnTransforms);
}
/**
* Alter the name or nullability of columns.
*
* @param columnAlterations One or more alterations to apply to columns.
*/
async alterColumns(columnAlterations: ColumnAlteration[]): Promise<void> {
await this.inner.alterColumns(columnAlterations);
}
/**
* Drop one or more columns from the dataset
*
* This is a metadata-only operation and does not remove the data from the
* underlying storage. In order to remove the data, you must subsequently
* call ``compact_files`` to rewrite the data without the removed columns and
* then call ``cleanup_files`` to remove the old files.
*
* @param columnNames The names of the columns to drop. These can be nested
* column references (e.g. "a.b.c") or top-level column
* names (e.g. "a").
*/
async dropColumns(columnNames: string[]): Promise<void> {
await this.inner.dropColumns(columnNames);
}
}

View File

@@ -13,8 +13,11 @@
// limitations under the License.
use arrow_ipc::writer::FileWriter;
use lancedb::table::AddDataOptions;
use lancedb::{ipc::ipc_file_to_batches, table::TableRef};
use lance::dataset::ColumnAlteration as LanceColumnAlteration;
use lancedb::{
ipc::ipc_file_to_batches,
table::{AddDataOptions, TableRef},
};
use napi::bindgen_prelude::*;
use napi_derive::napi;
@@ -93,4 +96,104 @@ impl Table {
pub fn query(&self) -> Query {
Query::new(self)
}
#[napi]
pub async fn add_columns(&self, transforms: Vec<AddColumnsSql>) -> napi::Result<()> {
let transforms = transforms
.into_iter()
.map(|sql| (sql.name, sql.value_sql))
.collect::<Vec<_>>();
let transforms = lance::dataset::NewColumnTransform::SqlExpressions(transforms);
self.table
.add_columns(transforms, None)
.await
.map_err(|err| {
napi::Error::from_reason(format!(
"Failed to add columns to table {}: {}",
self.table, err
))
})?;
Ok(())
}
#[napi]
pub async fn alter_columns(&self, alterations: Vec<ColumnAlteration>) -> napi::Result<()> {
for alteration in &alterations {
if alteration.rename.is_none() && alteration.nullable.is_none() {
return Err(napi::Error::from_reason(
"Alteration must have a 'rename' or 'nullable' field.",
));
}
}
let alterations = alterations
.into_iter()
.map(LanceColumnAlteration::from)
.collect::<Vec<_>>();
self.table
.alter_columns(&alterations)
.await
.map_err(|err| {
napi::Error::from_reason(format!(
"Failed to alter columns in table {}: {}",
self.table, err
))
})?;
Ok(())
}
#[napi]
pub async fn drop_columns(&self, columns: Vec<String>) -> napi::Result<()> {
let col_refs = columns.iter().map(String::as_str).collect::<Vec<_>>();
self.table.drop_columns(&col_refs).await.map_err(|err| {
napi::Error::from_reason(format!(
"Failed to drop columns from table {}: {}",
self.table, err
))
})?;
Ok(())
}
}
/// A definition of a column alteration. The alteration changes the column at
/// `path` to have the new name `name`, to be nullable if `nullable` is true,
/// and to have the data type `data_type`. At least one of `rename` or `nullable`
/// must be provided.
#[napi(object)]
pub struct ColumnAlteration {
/// The path to the column to alter. This is a dot-separated path to the column.
/// If it is a top-level column then it is just the name of the column. If it is
/// a nested column then it is the path to the column, e.g. "a.b.c" for a column
/// `c` nested inside a column `b` nested inside a column `a`.
pub path: String,
/// The new name of the column. If not provided then the name will not be changed.
/// This must be distinct from the names of all other columns in the table.
pub rename: Option<String>,
/// Set the new nullability. Note that a nullable column cannot be made non-nullable.
pub nullable: Option<bool>,
}
impl From<ColumnAlteration> for LanceColumnAlteration {
fn from(js: ColumnAlteration) -> Self {
let ColumnAlteration {
path,
rename,
nullable,
} = js;
Self {
path,
rename,
nullable,
}
}
}
/// A definition of a new column to add to a table.
#[napi(object)]
pub struct AddColumnsSql {
/// The name of the new column.
pub name: String,
/// The values to populate the new column with, as a SQL expression.
/// The expression can reference other columns in the table.
pub value_sql: String,
}