perf(battlesnake): defer ordered lifecycle bookkeeping
Build and Push Docker Container / build-and-push (push) Successful in 10m12s

- Acknowledge start and end without waiting for persistence or dashboards.
- Keep runtime creation and finished-game removal on the response path.
- Serialize start, turn, and end bookkeeping per game in Quart tasks.
- Capture turn thinking before scheduling deferred persistence.
- Defer start metrics, logging, and stale-game pruning.
- Run legacy history exports in a worker thread.
- Record results and end metrics even when history export fails.
- Drain pending bookkeeping before closing storage during shutdown.
- Test sub-500 ms responses with slow backends and real SQLite.
- Cover event ordering, export failures, and rejected user agents.
This commit is contained in:
2026-11-03 10:38:20 +01:00
parent d4d6375e01
commit 4403978537
6 changed files with 242 additions and 33 deletions
+3
View File
@@ -21,6 +21,7 @@ from server.blueprints import (
create_metrics_blueprint, create_metrics_blueprint,
create_dashboard_blueprint, create_dashboard_blueprint,
) )
from server.services.game_lifecycle import GameLifecycleService
from server.services import ( from server.services import (
DashboardEventsService, DashboardEventsService,
DashboardWebSocketHub, DashboardWebSocketHub,
@@ -106,6 +107,7 @@ class Server:
self.app = Quart('Battlesnake', template_folder=os.path.join(data_path, 'templates', 'side'), static_folder=os.path.join(data_path, 'templates', 'files')) self.app = Quart('Battlesnake', template_folder=os.path.join(data_path, 'templates', 'side'), static_folder=os.path.join(data_path, 'templates', 'files'))
self.game_lifecycle = GameLifecycleService(self)
self.app.register_blueprint(create_battlesnake_blueprint(self)) self.app.register_blueprint(create_battlesnake_blueprint(self))
self.app.register_blueprint(create_metrics_blueprint(self)) self.app.register_blueprint(create_metrics_blueprint(self))
self.app.register_blueprint(create_dashboard_blueprint(self)) self.app.register_blueprint(create_dashboard_blueprint(self))
@@ -143,6 +145,7 @@ class Server:
@self.app.after_serving @self.app.after_serving
async def shutdown_state_storage(): async def shutdown_state_storage():
await self.game_lifecycle.drain()
await self.dashboard_events_service.stop_listener() await self.dashboard_events_service.stop_listener()
await self.metrics_collector.close() await self.metrics_collector.close()
if self.gameplay_database is not None: if self.gameplay_database is not None:
+17 -30
View File
@@ -5,7 +5,6 @@ from quart import Blueprint, request, jsonify
from quart_common.web.decorators import require_user_agent from quart_common.web.decorators import require_user_agent
from quart_common.web.logger import await_log from quart_common.web.logger import await_log
from server.database import StorageLoader
from snakes import DEFAULT_SNAKE_CONFIG from snakes import DEFAULT_SNAKE_CONFIG
from server.GameBoard import GameBoard from server.GameBoard import GameBoard
from server.Files import read_file from server.Files import read_file
@@ -47,14 +46,13 @@ def create_battlesnake_blueprint(server:'Server') -> Blueprint:
@require_user_agent(("BattlesnakeEngine", "BattlesnakeArena"), abort_code=404) @require_user_agent(("BattlesnakeEngine", "BattlesnakeArena"), abort_code=404)
async def on_start(): async def on_start():
server.metrics_collector.record_http_request('start') server.metrics_collector.record_http_request('start')
await server.game_runtime.prune_stale_games()
game_state = await request.get_json() game_state = await request.get_json()
game_board = await server.game_runtime.create_game_board(game_state) game_board = await server.game_runtime.create_game_board(game_state, publish_metrics=False)
await server.gameplay_tracking.record_gameplay_start(game_state, game_board) active_count = len(server.game_runtime.running_games)
await server.dashboard_query.push_dashboard_games_update( server.game_lifecycle.submit(
game_state, trigger='game_started', game_state['game']['id'],
lambda: server.game_lifecycle.started(game_state, game_board, active_count),
) )
await await_log(server.logger.info(f'GAME START: {game_state['game']}'))
return 'ok' return 'ok'
@blueprint.post('/move') @blueprint.post('/move')
@@ -79,8 +77,10 @@ def create_battlesnake_blueprint(server:'Server') -> Blueprint:
except TimeoutError: except TimeoutError:
await await_log(server.logger.warning(f'MOVE TIMEOUT: turn={game_state.get("turn")}, game={game_id}, returning fallback {next_move!r}')) await await_log(server.logger.warning(f'MOVE TIMEOUT: turn={game_state.get("turn")}, game={game_id}, returning fallback {next_move!r}'))
await server.gameplay_tracking.record_gameplay_turn(game_state, next_move, game_board) thinking = server.gameplay_tracking._extract_latest_snake_thinking(game_board)
await server.dashboard_query.push_dashboard_game_replay_update(game_id) server.game_lifecycle.submit(
game_id, lambda: server.game_lifecycle.turn(game_state, next_move, thinking),
)
elapsed_ms = (time.perf_counter() - move_started) * 1000.0 elapsed_ms = (time.perf_counter() - move_started) * 1000.0
await server.metrics_collector.record_move(next_move, elapsed_ms) await server.metrics_collector.record_move(next_move, elapsed_ms)
@@ -93,29 +93,16 @@ def create_battlesnake_blueprint(server:'Server') -> Blueprint:
@require_user_agent(("BattlesnakeEngine", "BattlesnakeArena"), abort_code=404) @require_user_agent(("BattlesnakeEngine", "BattlesnakeArena"), abort_code=404)
async def on_end(): async def on_end():
server.metrics_collector.record_http_request('end') server.metrics_collector.record_http_request('end')
await server.game_runtime.prune_stale_games()
game_state = await request.get_json() game_state = await request.get_json()
if server.store_game_state: game_id = game_state['game']['id']
game_board = cast(GameBoard, await server.game_runtime.get_game_board(game_state, end=True)) game_board = server.game_runtime.running_games.get(game_id)
if server.check_tls_security: if game_board is not None:
await game_board.save( game_board.read_game_data(game_state)
StorageLoader.build(server.storage_type), game_board.end_game(game_state)
file_path=os.path.join(server.data_path, 'data'),
database=os.getenv('EDGEDB_DATABASE', None),
tls_security=None,
)
else:
await game_board.save(
StorageLoader.build(server.storage_type),
file_path=os.path.join(server.data_path, 'data'),
database=os.getenv('EDGEDB_DATABASE', None),
)
await server.gameplay_tracking.record_gameplay_end(game_state)
await server.dashboard_query.push_dashboard_games_update(game_state)
await await_log(server.logger.info(f'GAME ENDED: Winner is {[x['name'] for x in game_state['board']['snakes']]}'))
await server.game_runtime.delete_game_board(game_state) await server.game_runtime.delete_game_board(game_state)
await server.metrics_collector.record_game_end(game_state) server.game_lifecycle.submit(
game_id, lambda: server.game_lifecycle.ended(game_state, game_board),
)
return 'ok' return 'ok'
# @blueprint.get('/cleanup') # @blueprint.get('/cleanup')
+75
View File
@@ -0,0 +1,75 @@
"""Ordered, application-managed game bookkeeping outside HTTP response budgets."""
import asyncio
from collections.abc import Awaitable, Callable
import os
from quart_common.web.logger import await_log
from server.database import StorageLoader
class GameLifecycleService:
def __init__(self, server):
self.server = server
self.pending: dict[str, asyncio.Future] = {}
def submit(self, game_id:str, operation:Callable[[], Awaitable[None]]) -> None:
previous = self.pending.get(game_id)
completed = asyncio.get_running_loop().create_future()
self.pending[game_id] = completed
async def run():
try:
if previous is not None:
await asyncio.shield(previous)
await operation()
except Exception:
self.server.app.logger.exception('Game bookkeeping failed for %s', game_id)
finally:
if not completed.done():
completed.set_result(None)
if self.pending.get(game_id) is completed:
self.pending.pop(game_id, None)
self.server.app.add_background_task(run)
async def drain(self) -> None:
while self.pending:
await asyncio.gather(*(asyncio.shield(task) for task in list(self.pending.values())))
async def started(self, game_state:dict, game_board, active_count:int) -> None:
server = self.server
await server.metrics_collector.record_game_started(active_count)
await server.gameplay_tracking.record_gameplay_start(game_state, game_board)
await server.dashboard_query.push_dashboard_games_update(game_state, trigger='game_started')
await await_log(server.logger.info(f'GAME START: {game_state["game"]}'))
await server.game_runtime.prune_stale_games()
async def ended(self, game_state:dict, game_board) -> None:
server = self.server
try:
if server.store_game_state and game_board is not None:
kwargs = {
'file_path': os.path.join(server.data_path, 'data'),
'database': os.getenv('EDGEDB_DATABASE'),
}
if server.check_tls_security:
kwargs['tls_security'] = None
# Legacy exporters include synchronous serialization and database calls.
# Keep those off the request event loop as well as off the response path.
def export():
asyncio.run(game_board.save(StorageLoader.build(server.storage_type), **kwargs))
await asyncio.to_thread(export)
finally:
# A history export failure must not suppress the game result or metrics.
await server.gameplay_tracking.record_gameplay_end(game_state)
await server.metrics_collector.record_game_end(game_state)
await server.dashboard_query.push_dashboard_games_update(game_state)
await await_log(server.logger.info(
f'GAME ENDED: Winner is {[snake["name"] for snake in game_state["board"]["snakes"]]}'
))
async def turn(self, game_state:dict, move:str|None, thinking:dict|None) -> None:
server = self.server
await server.gameplay_tracking.record_gameplay_turn_snapshot(game_state, move, thinking)
await server.dashboard_query.push_dashboard_game_replay_update(game_state['game']['id'])
+2 -2
View File
@@ -18,7 +18,7 @@ class GameRuntimeService:
def attach_metrics_collector(self, metrics_collector:MetricsCollector) -> None: def attach_metrics_collector(self, metrics_collector:MetricsCollector) -> None:
self.metrics_collector = metrics_collector self.metrics_collector = metrics_collector
async def create_game_board(self, game_state:dict) -> GameBoard: async def create_game_board(self, game_state:dict, publish_metrics:bool=True) -> GameBoard:
game_id = game_state['game']['id'] game_id = game_state['game']['id']
new_game_board = GameBoard( new_game_board = GameBoard(
game_id=game_id, game_id=game_id,
@@ -34,7 +34,7 @@ class GameRuntimeService:
self.running_games[game_id] = new_game_board self.running_games[game_id] = new_game_board
self.game_move_counts[game_id] = 0 self.game_move_counts[game_id] = 0
self.game_last_seen_unix[game_id] = int(time.time()) self.game_last_seen_unix[game_id] = int(time.time())
if self.metrics_collector is not None: if publish_metrics and self.metrics_collector is not None:
await self.metrics_collector.record_game_started(len(self.game_last_seen_unix)) await self.metrics_collector.record_game_started(len(self.game_last_seen_unix))
return new_game_board return new_game_board
+4 -1
View File
@@ -23,10 +23,13 @@ class GameplayTrackingService:
await await_log(self.logger.warning(f"Gameplay DB start record failed:{error}")) await await_log(self.logger.warning(f"Gameplay DB start record failed:{error}"))
async def record_gameplay_turn(self, game_state:dict, my_move:str, game_board:GameBoard) -> None: async def record_gameplay_turn(self, game_state:dict, my_move:str, game_board:GameBoard) -> None:
thinking = self._extract_latest_snake_thinking(game_board)
await self.record_gameplay_turn_snapshot(game_state, my_move, thinking)
async def record_gameplay_turn_snapshot(self, game_state:dict, my_move:str|None, thinking:dict|None) -> None:
if self.gameplay_database is None: if self.gameplay_database is None:
return return
try: try:
thinking = self._extract_latest_snake_thinking(game_board)
await self.gameplay_database.record_turn(game_state, my_move, thinking) await self.gameplay_database.record_turn(game_state, my_move, thinking)
except Exception as error: except Exception as error:
await await_log(self.logger.warning(f"Gameplay DB turn record failed:{error}")) await await_log(self.logger.warning(f"Gameplay DB turn record failed:{error}"))
+141
View File
@@ -0,0 +1,141 @@
import asyncio
from copy import deepcopy
import os
import tempfile
from time import perf_counter
import unittest
from unittest.mock import AsyncMock, Mock, patch
from server.Server import Server
class TestBattlesnakeLifecycle(unittest.IsolatedAsyncioTestCase):
def setUp(self):
self.server = Server(
data_path=os.path.dirname(os.path.dirname(__file__)),
snake_type='TemplateSnake', storage_type='LocalStorage',
gameplay_db_enabled=False,
)
snake = {
'id': 'snake-1', 'name': 'test', 'health': 100, 'length': 3,
'head': {'x': 1, 'y': 1},
'body': [{'x': 1, 'y': 1}] * 3,
}
self.state = {
'game': {
'id': 'game-1', 'source': 'test', 'map': 'standard', 'timeout': 500,
'ruleset': {'name': 'standard', 'version': 'v1'},
},
'turn': 0, 'you': snake,
'board': {'width': 11, 'height': 11, 'food': [], 'hazards': [], 'snakes': [snake]},
}
self.headers = {'User-Agent': 'BattlesnakeArena'}
async def test_slow_io_does_not_delay_acknowledgements_or_reorder_events(self):
events = []
gate = asyncio.Event()
async def slow_start(*args):
await gate.wait()
await asyncio.sleep(0.55)
events.append('start')
async def slow_end(*args):
await asyncio.sleep(0.55)
events.append('end')
async def record_turn(*args):
events.append('turn')
server = self.server
server.logger = Mock()
server.logger.info.return_value = None
server.gameplay_tracking.record_gameplay_start = slow_start
server.gameplay_tracking.record_gameplay_end = slow_end
server.gameplay_tracking.record_gameplay_turn_snapshot = record_turn
server.dashboard_query.push_dashboard_games_update = AsyncMock()
server.dashboard_query.push_dashboard_game_replay_update = AsyncMock()
server.metrics_collector.record_game_started = AsyncMock()
server.metrics_collector.record_game_end = AsyncMock()
client = server.app.test_client()
started = perf_counter()
response = await client.post('/start', json=self.state, headers=self.headers)
start_ms = (perf_counter() - started) * 1000
self.assertEqual(response.status_code, 200)
self.assertEqual(await response.get_data(), b'ok')
self.assertLess(start_ms, 500)
self.assertIn('game-1', server.game_runtime.running_games)
self.assertEqual(events, [])
# A real move request must not overtake the pending start DB write.
response = await client.post('/move', json=self.state, headers=self.headers)
self.assertEqual(response.status_code, 200)
self.assertIn((await response.get_json())['move'], ('up', 'down', 'left', 'right'))
end_state = deepcopy(self.state)
end_state['turn'] = 1
started = perf_counter()
response = await client.post('/end', json=end_state, headers=self.headers)
end_ms = (perf_counter() - started) * 1000
self.assertEqual(response.status_code, 200)
self.assertLess(end_ms, 500)
self.assertNotIn('game-1', server.game_runtime.running_games)
self.assertEqual(events, [])
gate.set()
await server.game_lifecycle.drain()
self.assertEqual(events, ['start', 'turn', 'end'])
self.assertEqual(server.game_lifecycle.pending, {})
server.metrics_collector.record_game_started.assert_awaited_once_with(1)
server.metrics_collector.record_game_end.assert_awaited_once_with(end_state)
print(f'Lifecycle HTTP timings with >500 ms I/O: start={start_ms:.1f} ms, end={end_ms:.1f} ms')
async def test_export_failure_still_records_result_and_cleans_runtime(self):
server = self.server
server.logger = Mock()
server.logger.info.return_value = None
server.gameplay_tracking.record_gameplay_start = AsyncMock()
server.gameplay_tracking.record_gameplay_end = AsyncMock()
server.dashboard_query.push_dashboard_games_update = AsyncMock()
server.metrics_collector.record_game_end = AsyncMock()
client = server.app.test_client()
await client.post('/start', json=self.state, headers=self.headers)
await server.game_lifecycle.drain()
board = server.game_runtime.running_games['game-1']
board.save = AsyncMock(side_effect=OSError('export failed'))
server.store_game_state = True
with patch('server.services.game_lifecycle.StorageLoader.build', return_value=object):
response = await client.post('/end', json=self.state, headers=self.headers)
await server.game_lifecycle.drain()
self.assertEqual(response.status_code, 200)
self.assertNotIn('game-1', server.game_runtime.running_games)
board.save.assert_awaited_once()
server.gameplay_tracking.record_gameplay_end.assert_awaited_once()
server.metrics_collector.record_game_end.assert_awaited_once()
async def test_real_sqlite_lifecycle_persists_after_acknowledgements(self):
with tempfile.TemporaryDirectory() as folder:
server = Server(
data_path=folder, snake_type='TemplateSnake', storage_type='LocalStorage',
gameplay_db_path=os.path.join(folder, 'gameplay.sqlite3'),
)
server.logger = Mock()
server.logger.info.return_value = None
async with server.app.test_app():
client = server.app.test_client()
timings = {}
for endpoint in ('start', 'end'):
started = perf_counter()
response = await client.post('/' + endpoint, json=self.state, headers=self.headers)
timings[endpoint] = (perf_counter() - started) * 1000
self.assertEqual(response.status_code, 200)
self.assertLess(timings[endpoint], 500)
await server.game_lifecycle.drain()
replay = await server.gameplay_database.get_game_replay('game-1')
self.assertIsNotNone(replay)
self.assertNotEqual(replay['game']['status'], 'running')
print(f'Real SQLite HTTP timings: {timings}')
async def test_invalid_agent_does_not_schedule_work(self):
response = await self.server.app.test_client().post('/start', json=self.state)
self.assertEqual(response.status_code, 404)
self.assertEqual(self.server.game_lifecycle.pending, {})