gridfleet-testkit 0.14.0

Supported pytest and run-orchestration helpers for GridFleet integrations

GridFleet Testkit

testkit/ is the supported Python integration surface for external pytest/Appium suites that run through GridFleet.

What This Package Owns

• Stable import root: gridfleet_testkit
• Supported pytest plugin: gridfleet_testkit.pytest_plugin
• Supported pytest fixtures: appium_driver, gridfleet_client, gridfleet_client_config, device_test_data, device_handle, gridfleet_worker_id
• Supported public Appium helpers:
  • build_appium_options
  • create_appium_driver
  • get_device_test_data_for_driver
• Supported public client helpers:
  • GridFleetClient
  • HeartbeatThread
  • register_run_cleanup
• Supported public device/session helpers:
  • Device (return type of get_device / list_devices)
  • resolve_device_handle_from_driver
• Supported public result types:
  • CooldownResult
  • CooldownSetResult
  • CooldownEscalatedResult
• Manual hardware examples under testkit/examples/

What It Does Not Own

• Appium server installation or host-level driver setup
• WebDriver router lifecycle
• Device registration, verification, or readiness setup
• CI orchestration beyond the documented client helpers

The supported contract is the installable package and documented import pattern. The example scripts are onboarding aids, not CI-backed conformance tests.

Install

From PyPI:

pip install "gridfleet-testkit"

From a local checkout:

uv pip install -e ./testkit

From a copied testkit/ directory inside another repository:

uv pip install -e ./testkit

From a Git checkout or VCS URL that contains this package:

uv pip install "git+https://github.com/<org>/<repo>.git#subdirectory=testkit"

The package supports Python 3.10 through 3.14. Appium-Python-Client is installed as a runtime dependency because the pytest fixtures create real Appium sessions.

Environment

Variable	Default	Meaning
GRID_URL	http://localhost:4444	WebDriver router URL used by the pytest Appium fixture
GRIDFLEET_API_URL	http://localhost:8000/api	GridFleet API base used for session reporting, run helpers, and driver-pack catalog lookup
GRIDFLEET_TESTKIT_USERNAME	unset	Machine-auth username sent as HTTP Basic auth on every API call. Required when the manager runs with GRIDFLEET_AUTH_ENABLED=true. Use the same value as the manager's GRIDFLEET_MACHINE_AUTH_USERNAME.
GRIDFLEET_TESTKIT_PASSWORD	unset	Machine-auth password sent as HTTP Basic auth on every API call. Required when the manager runs with GRIDFLEET_AUTH_ENABLED=true. Use the same value as the manager's GRIDFLEET_MACHINE_AUTH_PASSWORD.
GRIDFLEET_TESTKIT_PACK_ID	unset	Optional default driver pack id for Appium option building
GRIDFLEET_TESTKIT_PLATFORM_ID	unset	Optional default platform id for Appium option building
GRIDFLEET_RUN_ID	unset	Optional run id. When set, drivers are created through the run-scoped grid endpoint GRID_URL/run/{id} so sessions land only on devices reserved for the run. Unset = free session on unreserved devices. Set this in the environment that launches pytest (e.g. the run launcher or CI step); the testkit reads it but does not set it.

The package assumes a running GridFleet API, a reachable WebDriver router, and platform-specific Appium driver setup on the registered hosts. When auth is disabled on the manager, leave GRIDFLEET_TESTKIT_USERNAME / GRIDFLEET_TESTKIT_PASSWORD unset and the testkit will send no Authorization header.

Pytest Plugin

Load the supported plugin from your test project:

pytest_plugins = ["gridfleet_testkit.pytest_plugin"]

Minimal usage:

import pytest

@pytest.mark.parametrize(
    "appium_driver",
    [{"pack_id": "appium-uiautomator2", "platform_id": "android_mobile"}],
    indirect=True,
)
def test_session_starts(appium_driver):
    assert appium_driver.session_id is not None

The plugin resolves pack_id and platform_id against the enabled driver-pack catalog, then injects Appium platformName, appium:automationName, appium:platform, and gridfleet:testName.

When exactly one enabled pack provides a platform id, platform_id alone is accepted. For environment-portable tests, set GRIDFLEET_TESTKIT_PACK_ID and GRIDFLEET_TESTKIT_PLATFORM_ID, then parametrize with {}.

If you need raw Appium control instead, omit pack_id and platform_id, then pass platformName as a normal capability key.

Tuning the HTTP Transport

To tune the Appium HTTP transport for every appium_driver session — connection retries, timeouts, proxy, TLS — override the gridfleet_client_config fixture in your conftest.py. It defaults to None. The testkit still owns the endpoint, so any remote_server_addr you set is overwritten with the resolved grid URL:

# conftest.py
import pytest
from appium.webdriver.client_config import AppiumClientConfig
from urllib3.util.retry import Retry

@pytest.fixture
def gridfleet_client_config():
    return AppiumClientConfig(
        remote_server_addr="",  # overwritten by the testkit
        init_args_for_pool_manager={"init_args_for_pool_manager": {"retries": Retry(total=3, backoff_factor=0.5)}},
    )

Plugin Lifecycle

• Creates an Appium session through GRID_URL
• Injects gridfleet:testName with the pytest test name
• Resolves the WebDriver endpoint from GRIDFLEET_RUN_ID: run-scoped URL inside a reserved run, bare grid URL otherwise. No GridFleet identity is injected into capabilities.
• Reports final session status back to GRIDFLEET_API_URL
• Exposes device_test_data for post-session operator-attached test data using the runtime connection target
• Exposes device_handle (a typed Device) for the device the live session landed on
• Exposes gridfleet_worker_id which returns the pytest-xdist worker id, or "controller" for non-worker processes
• Relies on manager-owned runtime isolation for Appium driver sub-ports and XCUITest build paths

If Appium driver creation fails before a Grid session exists, the exception propagates directly to the test. The router/grid allocation flow owns session-row lifecycle; pre-session failures are not recorded by the testkit.

Direct Appium Usage

If you need to create a driver outside pytest, use the public Appium helpers:

from gridfleet_testkit import create_appium_driver, get_device_test_data_for_driver

driver = create_appium_driver(
    pack_id="appium-uiautomator2",
    platform_id="firetv_real",
    test_name="manual-smoke",
)

try:
    assert driver.session_id is not None
    test_data = get_device_test_data_for_driver(driver)
finally:
    driver.quit()

create_appium_driver(...) reuses the same driver-pack catalog resolver as the pytest fixture. Managed nodes still get their host-scoped runtime allocations from the manager, so callers should not hard-code systemPort, chromedriverPort, mjpegServerPort, wdaLocalPort, or derivedDataPath. get_device_test_data_for_driver(...) is the non-pytest equivalent of the device_test_data fixture. If you only need the options object, use build_appium_options(...).

To tune the HTTP transport — connection retries, timeouts, proxy, TLS — pass an AppiumClientConfig via client_config. The testkit still owns the endpoint, so any remote_server_addr you set is overwritten with the resolved grid URL:

from appium.webdriver.client_config import AppiumClientConfig
from urllib3.util.retry import Retry

client_config = AppiumClientConfig(
    remote_server_addr="",  # overwritten by the testkit
    init_args_for_pool_manager={"init_args_for_pool_manager": {"retries": Retry(total=3, backoff_factor=0.5)}},
)

driver = create_appium_driver(
    pack_id="appium-uiautomator2",
    platform_id="firetv_real",
    client_config=client_config,
)

Client Helpers

Helper	Purpose
GridFleetClient.list_devices(*, pack_id=None, status=None, host_id=None, ...)	List devices using backend keyword filters (pack_id, platform_id, status, host_id, connection_target, tags, ...); returns a list of typed Device objects
GridFleetClient.get_device(device_id)	Fetch one device as a typed Device (curated base fields) by backend device id
GridFleetClient.get_device_test_data(device_id)	Fetch operator-attached free-form test_data for a device
GridFleetClient.get_run(run_id)	Fetch one run detail row by backend run id
GridFleetClient.replace_device_test_data(device_id, body)	Replace test_data with the supplied object
GridFleetClient.merge_device_test_data(device_id, body)	Deep-merge into device test_data
GridFleetClient.get_driver_pack_catalog()	Fetch enabled driver-pack catalog data for Appium platform selection
GridFleetClient.reserve_devices(...)	Create a run/reservation and return the manager response
GridFleetClient.signal_ready(run_id)	Signal that a run is ready
GridFleetClient.signal_active(run_id)	Move a run to active
GridFleetClient.heartbeat(run_id)	Send a run heartbeat and read current state
GridFleetClient.report_preparation_failure(run_id, device_id, message, source="ci_preparation")	Exclude one reserved device after setup fails
GridFleetClient.update_session_status(session_id, status)	Report final session status
GridFleetClient.complete_run(run_id)	Complete a run
GridFleetClient.cancel_run(run_id)	Cancel a run
GridFleetClient.cooldown_device(run_id, device_id, reason=..., ttl_seconds=...)	Exclude a reserved device from the run with a cooldown TTL
GridFleetClient.start_heartbeat(run_id, interval=30)	Start a background heartbeat thread
get_device_id_from_driver(driver)	Resolve the backend device id from a live driver's gridfleet:deviceId session capability
resolve_device_handle_from_driver(driver, client)	Resolve the assigned device as a typed Device from a running Appium session
get_device_test_data_for_driver(driver, gridfleet_client=None)	Fetch test_data for a live Appium driver
register_run_cleanup(client, run_id, heartbeat_thread=None)	Register atexit cleanup callable and return it; stops the heartbeat thread on exit but does not complete or cancel the run by default

Targeting Devices by Tag

GridFleet injects device tags into node stereotypes as gridfleet:tag:<key> capabilities, so the router's backend allocation can route sessions to devices matching specific tags.

@pytest.mark.parametrize(
    "appium_driver",
    [
        {
            "pack_id": "appium-uiautomator2",
            "platform_id": "android_mobile",
            "gridfleet:tag:screen_type": "4k",
        }
    ],
    indirect=True,
)
def test_4k_display(appium_driver):
    ...

The same capability works for free sessions:

driver = create_appium_driver(
    pack_id="appium-uiautomator2",
    platform_id="android_mobile",
    capabilities={"gridfleet:tag:screen_type": "4k"},
)

When an operator edits device tags, GridFleet marks the device for re-verification. The next verification restarts the Appium node and re-registers it with the updated Grid stereotype.

Worker Identity

The gridfleet_worker_id fixture is informational only: it returns the pytest-xdist worker id (normally gw0, gw1, and so on), or "controller" for non-worker processes. It is never transmitted to the manager; use it client-side for local sharding or log correlation. For run attribution, pass the created_by argument to GridFleetClient.reserve_devices — that is the only run-attribution field the reservation request carries.

Reservation Flow

GridFleet runs are router-routed: once devices are reserved, the manager tags matching nodes with the run id, and the router routes new Appium sessions to those nodes automatically when they arrive through the run-scoped endpoint (GRID_URL/run/{run_id}). There are no per-worker claim or release calls.

from gridfleet_testkit import GridFleetClient, register_run_cleanup

client = GridFleetClient("http://manager-ip:8000/api")

run = client.reserve_devices(
    name="my-test-run",
    requirements=[
        {
            "pack_id": "appium-uiautomator2",
            "platform_id": "firetv_real",
            "os_version": "8",
            "allocation": "all_available",
            "min_count": 1,
        }
    ],
    ttl_minutes=45,
    created_by="local-dev",
)

run_id = run["id"]
worker_count = len(run["devices"])
heartbeat_thread = client.start_heartbeat(run_id, interval=30)
cleanup = register_run_cleanup(client, run_id, heartbeat_thread)
# cleanup() runs at process exit; call client.complete_run(run_id) on success
# or client.cancel_run(run_id) on failure to set the run state explicitly.

# If one reserved device fails setup:
client.report_preparation_failure(
    run_id,
    device_id="device-123",
    message="Driver bootstrap timed out during CI setup",
    source="local-dev",
)

client.signal_ready(run_id)
client.signal_active(run_id)

Use count for exact reservations. Use allocation: "all_available" when CI should reserve every currently eligible matching device and size its worker pool from len(run["devices"]).

For pytest-xdist controller/worker orchestration, see Testkit xdist recipe. The recipe is copyable guidance, not a public testkit abstraction.

Cooling Down an Unstable Device

If a reserved device becomes unstable during a test, you can put it on cooldown so it is excluded from the run for a TTL. If the same device is cooled down too many times in the same run, it is escalated to maintenance automatically.

from gridfleet_testkit import GridFleetClient

client = GridFleetClient()

result = client.cooldown_device(
    run_id="run-123",
    device_id="device-456",
    reason="Connection dropped mid-test",
    ttl_seconds=120,
)

if result["status"] == "cooldown_set":
    print(f"Device on cooldown until {result['excluded_until']}")
elif result["status"] == "maintenance_escalated":
    print(f"Escalated after {result['cooldown_count']} cooldowns")

The manager enforces a maximum TTL via the general.device_cooldown_max_sec setting. The default is 3600 seconds. An httpx.HTTPStatusError with status 422 is raised if ttl_seconds exceeds the maximum.

Typed Device Reads

get_device and list_devices return typed Device objects so callers see the available fields in their IDE instead of guessing dict keys.

from gridfleet_testkit import GridFleetClient, resolve_device_handle_from_driver

client = GridFleetClient()

# After creating an Appium session, resolve the assigned device row
device = resolve_device_handle_from_driver(driver, client=client)
assert device.id
assert device.platform_label in {"Android", "iOS", "tvOS", "Roku", None}

# Or fetch / list directly
one = client.get_device(device.id)
available = client.list_devices(status="available")

Device carries the curated base field set both endpoints emit (id, identity_value, connection_target, name, pack_id, platform_id, platform_label, os_version, os_version_display, host_id, device_type, connection_type, manufacturer, model, tags, operational_state, is_reserved). Volatile long-tail fields (battery, telemetry, readiness, health summary, ...) are intentionally not surfaced. For operator-attached free-form data, use client.get_device_test_data(device.id) or the device_test_data fixture.

Run Cleanup Policy

register_run_cleanup(...) registers an atexit cleanup callable and returns it. By default it stops the heartbeat thread but does not complete or cancel the run, because process exit alone does not prove test success. Prefer explicit client.complete_run(run_id) after successful orchestration and client.cancel_run(run_id) for known failures. Pass on_exit="complete" or on_exit="cancel" only when that policy is correct for your script. Signal handlers are opt-in with install_signal_handlers=True; signal cleanup defaults to cancellation.

Device Test Data

The device_test_data fixture returns the operator-attached free-form test_data for the device assigned to the current test:

def test_uses_operator_data(appium_driver, device_test_data):
    assert "account" in device_test_data

Outside of pytest, use the client directly:

test_data = client.get_device_test_data(device_id)

Or use the driver helper:

from gridfleet_testkit import get_device_test_data_for_driver

test_data = get_device_test_data_for_driver(driver)

Errors and Result Types

• CooldownResult: union response type from cooldown_device, with status equal to "cooldown_set" or "maintenance_escalated". CooldownSetResult and CooldownEscalatedResult are the concrete TypedDict variants.

Examples

Baseline screenshot examples:

• examples/test_android_mobile_screenshot.py
• examples/test_android_tv_screenshot.py
• examples/test_firetv_screenshot.py
• examples/test_ios_simulator_screenshot.py
• examples/test_tvos_screenshot.py
• examples/test_roku_screenshot.py

Advanced example:

• examples/test_roku_sideload_screenshot.py

The baseline examples share the same flow:

1. Create a session through the WebDriver router
2. Print the resolved connection context
3. Save a screenshot
4. Assert that the screenshot file exists and is non-empty

Platform Notes

• Android Mobile / Android TV / Fire TV:
  • require the UiAutomator2 driver
  • rely on Grid routing hints generated from GridFleet metadata
• Fire TV:
  • baseline example supports optional appium:os_version filtering when you need a specific Fire OS release
• iOS simulator:
  • baseline example intentionally targets the simulator lane with appium:device_type=simulator
• tvOS:
  • baseline example intentionally targets a real device and assumes the host already satisfies XCUITest and WebDriverAgent prerequisites
• Roku:
  • screenshot examples install and activate the bundled sample dev app before capture
  • both Roku examples depend on Roku dev credentials