Arrow Flight RPC¶
Arrow Flight is an RPC framework for efficient transfer of Flight data over the network.
- Flight protocol documentation
Documentation of the Flight protocol, including how to use Flight conceptually.
- Flight API documentation
Python API documentation listing all of the various client and server classes.
- Python Cookbook
Recipes for using Arrow Flight in Python.
Writing a Flight Service¶
Servers are subclasses of
FlightServerBase. To implement
individual RPCs, override the RPC methods on this class.
import pyarrow.flight as flight class MyFlightServer(flight.FlightServerBase): def list_flights(self, context, criteria): info = flight.FlightInfo(...) yield info
Each RPC method always takes a
ServerCallContext for common
parameters. To indicate failure, raise an exception; Flight-specific
errors can be indicated by raising one of the subclasses of
To start a server, create a
Location to specify where to
listen, and create an instance of the server. (A string will be
converted into a location.) This will start the server, but won’t
block the rest of the program. Call
block until the server stops.
# Listen to all interfaces on a free port server = MyFlightServer("grpc://0.0.0.0:0") print("Server listening on port", server.port) server.serve()
Using the Flight Client¶
To connect to a Flight service, call
with a location.
Cancellation and Timeouts¶
When making a call, clients can optionally provide
FlightCallOptions. This allows clients to set a timeout on
calls or provide custom HTTP headers, among other features. Also, some
objects returned by client RPC calls expose a
cancel method which
allows terminating a call early.
On the server side, timeouts are transparent. For cancellation, the
server needs to manually poll
to check if the client has cancelled the call, and if so, break out of
any processing the server is currently doing.
TLS can be enabled when setting up a server by providing a certificate
and key pair to
On the client side, use
Location.for_grpc_tls() to construct the
Location to listen on.
Authentication is insecure without enabling TLS.
Handshake-based authentication can be enabled by implementing
ServerAuthHandler. Authentication consists of two parts: on
initial client connection, the server and client authentication
implementations can perform any negotiation needed; then, on each RPC
thereafter, the client provides a token. The server authentication
handler validates the token and provides the identity of the
client. This identity can be obtained from the
Servers and clients support custom middleware (or interceptors) that
are called on every request and can modify the request in a limited
fashion. These can be implemented by subclassing
providing them when creating the client or server.
Middleware are fairly limited, but they can add headers to a request/response. On the server, they can inspect incoming headers and fail the request; hence, they can be used to implement custom authentication methods.