worker.py 3.9 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129
  1. from __future__ import annotations
  2. import os
  3. import socket
  4. import time
  5. import traceback
  6. from dataclasses import dataclass
  7. from math import ceil
  8. from uuid import uuid4
  9. from core_shared import try_build_redis_client
  10. from core_shared.task_queue import TEAM_RUN_QUEUE, build_task_queue_consumer
  11. from sqlalchemy.orm import Session, sessionmaker
  12. from app.application.services import build_team_application_service
  13. from app.bootstrap.settings import TeamServiceSettings
  14. from app.db.session import build_session_factory
  15. from app.domain.repositories import (
  16. TeamDefinitionRepository,
  17. TeamRunRepository,
  18. TeamVersionRepository,
  19. )
  20. @dataclass(frozen=True)
  21. class TeamWorkerStats:
  22. worker_key: str
  23. executed_count: int = 0
  24. idle_count: int = 0
  25. error_count: int = 0
  26. class TeamWorker:
  27. def __init__(
  28. self,
  29. *,
  30. settings: TeamServiceSettings,
  31. session_factory: sessionmaker[Session],
  32. worker_key: str) -> None:
  33. self.settings = settings
  34. self.session_factory = session_factory
  35. self.worker_key = worker_key
  36. self.redis_client = try_build_redis_client(settings.redis_url)
  37. self.task_queue = build_task_queue_consumer(
  38. client=self.redis_client,
  39. queue_name=TEAM_RUN_QUEUE)
  40. def run_forever(self) -> TeamWorkerStats:
  41. executed_count = 0
  42. idle_count = 0
  43. error_count = 0
  44. while True:
  45. try:
  46. executed = self.run_once()
  47. except Exception:
  48. error_count += 1
  49. traceback.print_exc()
  50. executed = False
  51. if executed:
  52. executed_count += 1
  53. idle_count = 0
  54. else:
  55. idle_count += 1
  56. if self.task_queue is None:
  57. time.sleep(self.settings.worker_poll_interval_seconds)
  58. if self.settings.worker_max_idle_cycles is not None:
  59. if idle_count >= self.settings.worker_max_idle_cycles:
  60. return TeamWorkerStats(
  61. worker_key=self.worker_key,
  62. executed_count=executed_count,
  63. idle_count=idle_count,
  64. error_count=error_count)
  65. def run_once(self) -> bool:
  66. self._wait_for_queue_signal()
  67. db = self.session_factory()
  68. try:
  69. service = build_team_application_service(
  70. team_repository=TeamDefinitionRepository(db),
  71. team_version_repository=TeamVersionRepository(db),
  72. team_run_repository=TeamRunRepository(db),
  73. settings=self.settings)
  74. result = service.execute_next_claimed_team_run(
  75. worker_key=self.worker_key,
  76. lease_seconds=self.settings.worker_lease_seconds,
  77. dry_run=self.settings.worker_dry_run)
  78. return result is not None
  79. finally:
  80. db.close()
  81. def _wait_for_queue_signal(self) -> None:
  82. if self.task_queue is None:
  83. return
  84. try:
  85. self.task_queue.dequeue(
  86. timeout_seconds=max(1, ceil(self.settings.worker_poll_interval_seconds)))
  87. except Exception:
  88. return
  89. def build_worker_key() -> str:
  90. configured_key = os.getenv("AGENT_PLATFORM_WORKER_KEY")
  91. if configured_key:
  92. return configured_key
  93. hostname = socket.gethostname()
  94. return f"{hostname}-{uuid4().hex[:8]}"
  95. def main() -> None:
  96. settings = TeamServiceSettings()
  97. worker = TeamWorker(
  98. settings=settings,
  99. session_factory=build_session_factory(settings),
  100. worker_key=build_worker_key())
  101. stats = worker.run_forever()
  102. print(
  103. "team-worker stopped "
  104. f"worker_key={stats.worker_key} "
  105. f"executed_count={stats.executed_count} "
  106. f"idle_count={stats.idle_count} "
  107. f"error_count={stats.error_count}",
  108. flush=True)
  109. if __name__ == "__main__":
  110. main()