@datumland/arrowflightjs

A lightweight Arrow Flight client for Node.js. Send Apache Arrow tables over gRPC with a fluent API.

Install

npm install @datumland/arrowflightjs

Peer requirement: Node.js >= 18.

Quick start

import { FlightClient, rowsToTable } from '@datumland/arrowflightjs';

const client = new FlightClient('localhost:50051');

// Upload rows as an Arrow table
await client
  .put([
    { id: 1, name: 'Alice' },
    { id: 2, name: 'Bob' },
  ])
  .toPath(['my', 'dataset'])
  .execute();

// Or pass an Arrow Table directly
import { tableFromArrays } from 'apache-arrow';

const table = tableFromArrays({ id: [1, 2], name: ['Alice', 'Bob'] });
await client.put(table).toPath(['my', 'dataset']).execute();

// Retrieve: describe a dataset, then fetch an endpoint's ticket
const info = await client.flights({ path: ['my', 'dataset'] }).execute();
const result = await client.get(info.endpoints()[0].ticket).execute();

result.table(); // apache-arrow Table
result.rows();  // [{ id: 1, name: 'Alice' }, { id: 2, name: 'Bob' }]

// Execute a server action
const results = await client
  .action('myAction')
  .withBody(Buffer.from('payload'))
  .execute();

await client.close();

API

new FlightClient(location, options?)

Creates a client connected to a Flight server. location is a host:port address, a grpc:// / grpc+tls:// URI, or a Flight Location object (as found on an endpoint). The scheme sets TLS (grpc+tls:// → on); a bare address falls back to the tls option. Only bare addresses and grpc:// / grpc+tls:// are accepted (case-insensitive); every other URI scheme — reuse-connection, http(s)://, grpc+unix://, … — throws.

Option	Type	Description
tls	boolean	Use TLS (default false)
headers	Record<string, string>	Headers sent with every RPC
middleware	ClientMiddleware[]	nice-grpc middleware chain

client.put(data)

Starts a DoPut operation. Accepts an Arrow Table or an array of row objects (converted via rowsToTable).

Returns a PutOperation builder:

client.put(rows)
  .toPath(['bucket', 'key'])   // or .toCmd(Buffer.from('...'))
  .withMetadata(Buffer.from('app meta'))
  .execute();                  // → Promise<PutResult[]>

client.flights(descriptor)

Starts a GetFlightInfo operation. descriptor is { path: string[] } or { cmd: Buffer }.

Returns a FlightsOperation builder whose execute() resolves to a FlightInfoResult:

const info = await client.flights({ path: ['bucket', 'key'] })  // or { cmd: Buffer.from('SELECT ...') }
  .withHeaders({ authorization: 'Bearer tok' })
  .execute();                  // → Promise<FlightInfoResult>

info.raw();        // the underlying FlightInfo (schema, totals, metadata)
info.endpoints();  // FlightEndpoint[] — each carries a ticket and locations

A flight may be partitioned across several endpoints; consume each one to read the whole flight. An endpoint's location says where its ticket can be redeemed: an empty list (or arrow-flight-reuse-connection) means the current client; a grpc:// / grpc+tls:// URI means another server, reached by opening a new client:

for (const endpoint of info.endpoints()) {
  const reuse = endpoint.location.length === 0;
  const target = reuse ? client : new FlightClient(endpoint.location[0]);
  const result = await target.get(endpoint.ticket).execute();
  // ... use result.table() / result.rows()
  if (!reuse) await target.close();
}

client.get(ticket)

Starts a DoGet operation. ticket is a Ticket (as found in FlightInfo.endpoint[].ticket) or raw ticket bytes (Buffer / Uint8Array).

Returns a GetOperation builder whose execute() resolves to a FlightResult:

const result = await client.get(ticket)
  .withHeaders({ ... })
  .execute();                  // → Promise<FlightResult>

result.table();  // apache-arrow Table
result.rows();   // Record<string, unknown>[]
result.raw();    // FlightData[] (the raw stream)

The whole stream is drained by execute(). table() and rows() are derived lazily and cached; reading an empty stream leaves raw() empty and makes table()/rows() throw.

client.action(type)

Starts a DoAction operation.

Returns an ActionOperation builder:

client.action('compact')
  .withBody(Buffer.from('{}'))
  .execute();                  // → Promise<Buffer[]>

rowsToTable(rows) / tableToRows(table)

Convert between an array of plain objects and a columnar Arrow Table.

const table = rowsToTable([{ x: 1 }, { x: 2 }]);
const rows = tableToRows(table);   // [{ x: 1 }, { x: 2 }]

tableToRows recurses into nested structs and lists, so rows are plain objects. Arrow types map to their JS equivalents — notably 64-bit integers come back as bigint (not JSON-serializable) and timestamps as Date.

Development

npm install
npm run generate   # generate gRPC stubs from Flight.proto
npm run build      # compile TypeScript
npm test           # run tests

License

MIT