264 lines
12 KiB
Python
264 lines
12 KiB
Python
import time
|
|
|
|
from server.metrics.MetricsManager import MetricsManager
|
|
|
|
class ServerMetricsCollector:
|
|
def __init__(self, metrics_manager:MetricsManager, game_state_local_cache:bool, metrics_backend:str, game_state_backend:str, stale_game_timeout_sec:int, game_last_seen_unix:dict, game_move_counts:dict,):
|
|
self._manager = metrics_manager
|
|
self._stale_game_timeout_sec = stale_game_timeout_sec
|
|
self._game_last_seen_unix = game_last_seen_unix
|
|
self._game_move_counts = game_move_counts
|
|
self._game_state_backend_is_redis = game_state_backend.strip().lower() == 'redis'
|
|
self._metrics = {
|
|
'games_started': 0,
|
|
'games_ended': 0,
|
|
'wins': 0,
|
|
'losses': 0,
|
|
'total_moves': 0,
|
|
'total_turns': 0,
|
|
'max_turn': 0,
|
|
'active_games_peak': 0,
|
|
'games_autocreated': 0,
|
|
'http_requests_total': 0,
|
|
# 'http_requests_by_endpoint': {
|
|
# 'info': 0,
|
|
# 'start': 0,
|
|
# 'move': 0,
|
|
# 'end': 0,
|
|
# },
|
|
# 'move_direction_counts': {
|
|
# 'up': 0,
|
|
# 'down': 0,
|
|
# 'left': 0,
|
|
# 'right': 0,
|
|
# 'unknown': 0,
|
|
# },
|
|
'move_response_time_ms_total': 0.0,
|
|
'move_response_time_ms_max': 0.0,
|
|
'last_game_start_unix': 0,
|
|
'last_game_end_unix': 0,
|
|
'last_move_unix': 0,
|
|
'games_stuck_removed': 0,
|
|
'game_state_local_cache_enabled': bool(game_state_local_cache),
|
|
'metrics_backend': metrics_backend,
|
|
}
|
|
|
|
# ── internal ──────────────────────────────────────────────────────────────
|
|
|
|
async def _auto_publish(self) -> None:
|
|
snapshot = self.build_local_snapshot(self._game_last_seen_unix, self._game_move_counts)
|
|
await self._manager.publish_only(snapshot)
|
|
|
|
# ── record helpers ────────────────────────────────────────────────────────
|
|
|
|
def record_http_request(self, endpoint:str) -> None:
|
|
self._metrics['http_requests_total'] += 1
|
|
# endpoint_counts = self._metrics['http_requests_by_endpoint']
|
|
# endpoint_counts[endpoint] = endpoint_counts.get(endpoint, 0) + 1
|
|
|
|
async def record_game_started(self, active_count:int) -> None:
|
|
self._metrics['games_started'] += 1
|
|
self._metrics['active_games_peak'] = max(
|
|
self._metrics['active_games_peak'],
|
|
active_count,
|
|
)
|
|
self._metrics['last_game_start_unix'] = int(time.time())
|
|
await self._auto_publish()
|
|
|
|
async def record_game_autocreated(self) -> None:
|
|
self._metrics['games_autocreated'] += 1
|
|
await self._auto_publish()
|
|
|
|
async def record_move(self, direction:str, elapsed_ms:float) -> None:
|
|
self._metrics['total_moves'] += 1
|
|
self._metrics['move_response_time_ms_total'] += elapsed_ms
|
|
self._metrics['move_response_time_ms_max'] = max(
|
|
self._metrics['move_response_time_ms_max'],
|
|
elapsed_ms,
|
|
)
|
|
# move_counts = self._metrics['move_direction_counts']
|
|
# if direction in move_counts:
|
|
# move_counts[direction] += 1
|
|
# else:
|
|
# move_counts['unknown'] += 1
|
|
# self._metrics['last_move_unix'] = int(time.time())
|
|
await self._auto_publish()
|
|
|
|
async def record_game_end(self, game_state:dict) -> None:
|
|
self._metrics['games_ended'] += 1
|
|
self._metrics['last_game_end_unix'] = int(time.time())
|
|
|
|
final_turn = int(game_state.get('turn', 0))
|
|
self._metrics['total_turns'] += final_turn
|
|
self._metrics['max_turn'] = max(self._metrics['max_turn'], final_turn)
|
|
|
|
you_id = game_state.get('you', {}).get('id')
|
|
alive_ids = {s.get('id') for s in game_state.get('board', {}).get('snakes', [])}
|
|
if you_id and you_id in alive_ids:
|
|
self._metrics['wins'] += 1
|
|
else:
|
|
self._metrics['losses'] += 1
|
|
await self._auto_publish()
|
|
|
|
async def record_stuck_removed(self) -> None:
|
|
if self._game_state_backend_is_redis:
|
|
return
|
|
self._metrics['games_stuck_removed'] += 1
|
|
await self._auto_publish()
|
|
|
|
# ── snapshot builders ─────────────────────────────────────────────────────
|
|
|
|
def _calc_active_game_stats(self, game_last_seen_unix:dict) -> tuple[int, int, int]:
|
|
"""Returns (report_active_games, report_stale_candidates, oldest_active_age_sec)."""
|
|
now = int(time.time())
|
|
stale_candidates = sum(
|
|
1
|
|
for last_seen in game_last_seen_unix.values()
|
|
if now - last_seen >= self._stale_game_timeout_sec
|
|
)
|
|
|
|
if self._game_state_backend_is_redis:
|
|
# Redis auto-expires stale keys via TTL, so stale games are already gone from the
|
|
# server's perspective. We exclude them from all metrics so we only report games
|
|
# that are actually still alive in Redis.
|
|
report_active_games = len(game_last_seen_unix) - stale_candidates
|
|
report_stale_candidates = 0
|
|
# Only include non-stale timestamps when calculating the oldest active game age,
|
|
# so a game that Redis already deleted doesn't inflate the age metric.
|
|
active_last_seen = [
|
|
last_seen
|
|
for last_seen in game_last_seen_unix.values()
|
|
if now - last_seen < self._stale_game_timeout_sec
|
|
]
|
|
else:
|
|
report_active_games = len(game_last_seen_unix)
|
|
report_stale_candidates = stale_candidates
|
|
active_last_seen = list(game_last_seen_unix.values())
|
|
|
|
oldest_active_age = max(0, now - min(active_last_seen)) if active_last_seen else 0
|
|
return report_active_games, report_stale_candidates, oldest_active_age
|
|
|
|
def build_local_snapshot(self, game_last_seen_unix:dict, game_move_counts:dict) -> dict:
|
|
games_ended = self._metrics['games_ended']
|
|
total_moves = self._metrics['total_moves']
|
|
avg_turns = self._metrics['total_turns'] / games_ended if games_ended else 0.0
|
|
win_rate = self._metrics['wins'] / games_ended if games_ended else 0.0
|
|
avg_move_ms = (
|
|
self._metrics['move_response_time_ms_total'] / total_moves if total_moves else 0.0
|
|
)
|
|
|
|
report_active_games, report_stale_candidates, oldest_active_age = self._calc_active_game_stats(game_last_seen_unix)
|
|
|
|
return {
|
|
**self._metrics,
|
|
'active_games': report_active_games,
|
|
'tracked_games': len(game_move_counts),
|
|
'avg_turns_per_game': round(avg_turns, 2),
|
|
'win_rate': round(win_rate, 4),
|
|
'avg_move_response_ms': round(avg_move_ms, 2),
|
|
# 'http_requests_by_endpoint': dict(self._metrics['http_requests_by_endpoint']),
|
|
# 'move_direction_counts': dict(self._metrics['move_direction_counts']),
|
|
'oldest_active_game_age_sec': oldest_active_age,
|
|
'stale_game_timeout_sec': self._stale_game_timeout_sec,
|
|
'active_games_stale': report_stale_candidates,
|
|
}
|
|
|
|
async def build_snapshot(self, game_last_seen_unix:dict, game_move_counts:dict) -> dict:
|
|
local_snapshot = self.build_local_snapshot(game_last_seen_unix, game_move_counts)
|
|
return await self._manager.snapshot(local_snapshot)
|
|
|
|
def build_prometheus_metrics(self, snapshot:dict) -> str:
|
|
lines = [
|
|
'# HELP snake_games_started_total Total games started by snake server.',
|
|
'# TYPE snake_games_started_total counter',
|
|
f'snake_games_started_total {snapshot["games_started"]}',
|
|
'# HELP snake_games_ended_total Total games ended by snake server.',
|
|
'# TYPE snake_games_ended_total counter',
|
|
f'snake_games_ended_total {snapshot["games_ended"]}',
|
|
'# HELP snake_wins_total Total games won by this snake.',
|
|
'# TYPE snake_wins_total counter',
|
|
f'snake_wins_total {snapshot["wins"]}',
|
|
'# HELP snake_losses_total Total games lost by this snake.',
|
|
'# TYPE snake_losses_total counter',
|
|
f'snake_losses_total {snapshot["losses"]}',
|
|
'# HELP snake_moves_total Total move decisions served by /move.',
|
|
'# TYPE snake_moves_total counter',
|
|
f'snake_moves_total {snapshot["total_moves"]}',
|
|
'# HELP snake_turns_total Total turns across all ended games.',
|
|
'# TYPE snake_turns_total counter',
|
|
f'snake_turns_total {snapshot["total_turns"]}',
|
|
'# HELP snake_active_games Currently active games in memory.',
|
|
'# TYPE snake_active_games gauge',
|
|
f'snake_active_games {snapshot["active_games"]}',
|
|
'# HELP snake_tracked_games Currently tracked game IDs for move counters.',
|
|
'# TYPE snake_tracked_games gauge',
|
|
f'snake_tracked_games {snapshot["tracked_games"]}',
|
|
'# HELP snake_max_turn Highest final turn seen in an ended game.',
|
|
'# TYPE snake_max_turn gauge',
|
|
f'snake_max_turn {snapshot["max_turn"]}',
|
|
'# HELP snake_active_games_peak Highest active game count observed.',
|
|
'# TYPE snake_active_games_peak gauge',
|
|
f'snake_active_games_peak {snapshot["active_games_peak"]}',
|
|
'# HELP snake_games_autocreated_total Games created on /move or /end due to missing /start.',
|
|
'# TYPE snake_games_autocreated_total counter',
|
|
f'snake_games_autocreated_total {snapshot["games_autocreated"]}',
|
|
'# HELP snake_http_requests_total Total HTTP requests handled by this process.',
|
|
'# TYPE snake_http_requests_total counter',
|
|
f'snake_http_requests_total {snapshot["http_requests_total"]}',
|
|
'# HELP snake_move_response_ms_total Total move endpoint compute time in milliseconds.',
|
|
'# TYPE snake_move_response_ms_total counter',
|
|
f'snake_move_response_ms_total {round(snapshot["move_response_time_ms_total"], 3)}',
|
|
'# HELP snake_move_response_ms_max Maximum move endpoint compute time in milliseconds.',
|
|
'# TYPE snake_move_response_ms_max gauge',
|
|
f'snake_move_response_ms_max {round(snapshot["move_response_time_ms_max"], 3)}',
|
|
'# HELP snake_avg_turns_per_game Average final turn per ended game.',
|
|
'# TYPE snake_avg_turns_per_game gauge',
|
|
f'snake_avg_turns_per_game {snapshot["avg_turns_per_game"]}',
|
|
'# HELP snake_avg_move_response_ms Average move endpoint compute time in milliseconds.',
|
|
'# TYPE snake_avg_move_response_ms gauge',
|
|
f'snake_avg_move_response_ms {snapshot["avg_move_response_ms"]}',
|
|
'# HELP snake_win_rate Win ratio from ended games (0.0 - 1.0).',
|
|
'# TYPE snake_win_rate gauge',
|
|
f'snake_win_rate {snapshot["win_rate"]}',
|
|
'# HELP snake_last_game_start_unix Unix timestamp of most recent /start request.',
|
|
'# TYPE snake_last_game_start_unix gauge',
|
|
f'snake_last_game_start_unix {snapshot["last_game_start_unix"]}',
|
|
'# HELP snake_last_game_end_unix Unix timestamp of most recent /end request.',
|
|
'# TYPE snake_last_game_end_unix gauge',
|
|
f'snake_last_game_end_unix {snapshot["last_game_end_unix"]}',
|
|
'# HELP snake_last_move_unix Unix timestamp of most recent /move response.',
|
|
'# TYPE snake_last_move_unix gauge',
|
|
f'snake_last_move_unix {snapshot["last_move_unix"]}',
|
|
'# HELP snake_games_stuck_removed_total Active games auto-removed due to inactivity timeout.',
|
|
'# TYPE snake_games_stuck_removed_total counter',
|
|
f'snake_games_stuck_removed_total {snapshot["games_stuck_removed"]}',
|
|
'# HELP snake_oldest_active_game_age_sec Age in seconds of the oldest active game.',
|
|
'# TYPE snake_oldest_active_game_age_sec gauge',
|
|
f'snake_oldest_active_game_age_sec {snapshot["oldest_active_game_age_sec"]}',
|
|
'# HELP snake_stale_game_timeout_sec Configured inactivity timeout for stale games.',
|
|
'# TYPE snake_stale_game_timeout_sec gauge',
|
|
f'snake_stale_game_timeout_sec {snapshot["stale_game_timeout_sec"]}',
|
|
'# HELP snake_active_games_stale Active games currently beyond stale timeout.',
|
|
'# TYPE snake_active_games_stale gauge',
|
|
f'snake_active_games_stale {snapshot["active_games_stale"]}',
|
|
]
|
|
|
|
# lines.extend([
|
|
# '# HELP snake_http_requests_by_endpoint_total Requests served grouped by endpoint.',
|
|
# '# TYPE snake_http_requests_by_endpoint_total counter',
|
|
# ])
|
|
# for endpoint, count in snapshot['http_requests_by_endpoint'].items():
|
|
# lines.append(f'snake_http_requests_by_endpoint_total{{endpoint="{endpoint}"}} {count}')
|
|
|
|
# lines.extend([
|
|
# '# HELP snake_moves_by_direction_total Move responses grouped by direction.',
|
|
# '# TYPE snake_moves_by_direction_total counter',
|
|
# ])
|
|
# for direction, count in snapshot['move_direction_counts'].items():
|
|
# lines.append(f'snake_moves_by_direction_total{{direction="{direction}"}} {count}')
|
|
|
|
return '\n'.join(lines) + '\n'
|
|
|
|
async def close(self) -> None:
|
|
await self._manager.close()
|