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
+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, {})