Warning
This documentation is actively being updated as the project evolves and may not be complete in all areas.
Drivers¶
Jumpstarter uses a modular driver model to build abstractions around the interfaces used to interact with target devices, both physical hardware and virtual systems.
An Exporter uses Drivers to “export” these interfaces from a host machine to the clients via gRPC. Drivers can be thought of as a simplified API for an interface or device type.
Architecture¶
Drivers in Jumpstarter follow a client/server architecture where:
Driver implementations run on the exporter side and interact directly with hardware or virtual devices
Driver clients run on the client side and communicate with drivers via gRPC
Interface classes define the contract between implementations and clients
The architecture follows a pattern with these key components:
Interface Class - An abstract base class using Python’s ABCMeta to define the contract (methods and their signatures) that driver implementations must fulfill. The interface also specifies the client class through the
client()
class method.Driver Class - Inherits from both the Interface and the base
Driver
class, implementing the logic to configure and use hardware interfaces. Driver methods are marked with the@export
decorator to expose them over the network.Driver Client - Provides a user-friendly interface that can be used by clients to interact with the driver either locally or remotely over the network.
When a client requests a lease and connects to an exporter, a session is created
for all tests the client needs to execute. Within this session, the specified
Driver
subclass is instantiated for each configured interface. These driver
instances live throughout the session’s duration, maintaining state and
executing setup/teardown logic.
On the client side, a DriverClient
subclass is instantiated for each exported
interface. Since clients may run on different machines than exporters,
DriverClient
classes are loaded dynamically when specified in the allowed
packages list.
To maintain compatibility, avoid making breaking changes to interfaces. Add new methods when needed but preserve existing signatures. If breaking changes are required, create new interface, client, and driver versions within the same module.
Drivers are often used with Adapters, which transform driver connections into different forms or interfaces for specific use cases.
Types¶
The API reference of the documentation provides a complete list of all standard drivers, you can find it here: Driver API Reference.
Some categories of drivers include:
System Control: Control power to devices, or general control.
Communication: Provide protocols for network communication, such as TCP/IP, Serial, CAN bus, etc.
Data and Storage: Control storage devices, such as SD cards or USB drives, and data.
Media: Provide interfaces for media capture and playback, such as video or audio.
Debug and Programming: Provide interfaces for debugging and programming devices, such as JTAG or SWD, remote flashing, emulation, etc.
Utility: Provide utility functions, such as shell driver commands on a exporter.
Composite Drivers¶
Composite drivers combine multiple lower-level drivers to create higher-level abstractions or specialized workflows. For example, a composite driver might coordinate power cycling, storage re-flashing, and serial communication to automate a device initialization process.
In Jumpstarter, drivers are organized in a tree structure which allows for the representation of complex device configurations that may be found in your environment.
Here’s an example of a composite driver tree:
MyHarness # Custom composite driver for the entire target device harness
├─ TcpNetwork # TCP Network driver to tunnel port 8000
├─ MyPower # Custom power driver to control device power
├─ SDWire # SD Wire storage emulator to enable re-flash on demand
├─ DigitalOutput # GPIO pin control to send signals to the device
└─ MyDebugger # Custom debugger interface composite driver
└─ PySerial # Serial debugger with PySerial
Configuration¶
Drivers are configured using a YAML Exporter config file, which specifies the drivers to load and the parameters for each. Drivers are distributed as Python packages making it easy to develop and install your own drivers.
Here is an example exporter config that loads drivers for both physical and virtual devices:
apiVersion: jumpstarter.dev/v1alpha1
kind: ExporterConfig
metadata:
namespace: default
name: demo
endpoint: grpc.jumpstarter.example.com:443
token: xxxxx
export:
# Physical hardware drivers
power:
type: jumpstarter_driver_yepkit.driver.Ykush
config:
serial: "YK25838"
port: "1"
serial:
type: "jumpstarter_driver_pyserial.driver.PySerial"
config:
url: "/dev/ttyUSB0"
baudrate: 115200
# Virtual device drivers
qemu:
type: "jumpstarter_driver_qemu.driver.QEMU"
config:
image_path: "/var/lib/jumpstarter/images/vm.qcow2"
memory: "1G"
cpu_cores: 2
Communication¶
Drivers use two primary methods to communicate between client and exporter:
Messages¶
Commands are sent as messages from driver clients to driver implementations,
allowing the client to trigger actions or retrieve information from the device.
Methods marked with the @export
decorator are made available over the network.
Streams¶
Drivers can establish streams for continuous data exchange, such as for serial
communication or video streaming. This enables real-time interaction with both
physical and virtual interfaces across the network. Methods marked with the
@exportstream
decorator create streams for bidirectional communication.
Authentication and Security¶
Driver access is controlled through Jumpstarter’s authentication mechanisms:
Local Mode Authentication¶
In local mode, drivers are accessible to any process that can connect to the local Unix socket. This is typically restricted by file system permissions. When running tests locally, authentication is simplified since everything runs in the same user context.
Distributed Mode Authentication¶
In distributed mode, authentication is handled through JWT tokens:
Client Authentication: Clients authenticate to the controller using JWT tokens, which establishes their identity and permissions
Exporter Authentication: Similarly, exporters authenticate to the controller with their own tokens
Driver Access Control: The controller enforces access control by only allowing authorized clients to acquire leases on exporters and their drivers
Driver Allowlist: Client configurations can specify which driver packages are allowed to be loaded, preventing unintended execution of untrusted code
Driver Package Security¶
When using distributed mode, driver security considerations include:
Package Verification: Clients can verify that only trusted driver packages are loaded by configuring allowlists
Capability Restrictions: Access to specific driver functionality can be restricted based on client permissions
Session Isolation: Each client session operates with its own driver instances to prevent interference between users
Custom Drivers¶
While Jumpstarter comes with drivers for many basic interfaces, custom drivers can be developed for specialized hardware interfaces, emulated environments, or to provide domain-specific abstractions for your use case. Custom drivers follow the same architecture pattern as built-in drivers and can be integrated into the system through the exporter configuration.
Example Implementation¶
from sys import modules
from types import SimpleNamespace
from anyio import connect_tcp, sleep
from contextlib import asynccontextmanager
from collections.abc import Generator, AsyncGenerator
from abc import ABCMeta, abstractmethod
from jumpstarter.driver import Driver, export, exportstream
from jumpstarter.client import DriverClient
from jumpstarter.common.utils import serve
# Define an interface with ABCMeta
class GenericInterface(metaclass=ABCMeta):
@classmethod
def client(cls) -> str:
return "example.GenericClient"
@abstractmethod
def query(self, param: str) -> str: ...
@abstractmethod
def get_data(self) -> Generator[dict, None, None]: ...
@abstractmethod
def create_stream(self): ...
# Implement the interface with the Driver base class
class GenericDriver(GenericInterface, Driver):
@export
def query(self, param: str) -> str:
# This could be any device-specific command
return f"Response for {param}"
# driver calls can be either sync or async
@export
async def async_query(self, param: str) -> str:
# Example of an async operation with delay
await sleep(1)
return f"Async response for {param}"
@export
def get_data(self) -> Generator[dict, None, None]:
# Example of a streaming response - could be sensor data, logs, etc.
for i in range(3):
yield {"type": "data", "value": i, "timestamp": f"2023-04-0{i+1}"}
# stream constructor has to be an AsyncContextManager
# that yield an anyio.abc.ObjectStream
@exportstream
@asynccontextmanager
async def create_stream(self):
# This could be any stream connection to a device
async with await connect_tcp(remote_host="example.com", remote_port=80) as stream:
yield stream
class GenericClient(DriverClient):
# client methods are sync
def query(self, param: str) -> str:
return self.call("query", param)
def async_query(self, param: str) -> str:
# async driver methods can be invoked the same way
return self.call("async_query", param)
def get_data(self) -> Generator[dict, None, None]:
yield from self.streamingcall("get_data")
# Streams can be used for bidirectional communication
def with_stream(self, callback):
with self.stream("create_stream") as stream:
callback(stream)
modules["example"] = SimpleNamespace(GenericClient=GenericClient)
with serve(GenericDriver()) as client:
assert client.query("test") == "Response for test"
assert client.async_query("async test") == "Async response for async test"
data = list(client.get_data())
assert len(data) == 3