Files
Artemis-Eapil/apps/admin_console/services/exploration_recovery.py
T

96 lines
4.2 KiB
Python

"""Reconcile saved exploration metadata with the existing queue and worker owners."""
import json
import time
from apps.admin_console.core.state import state
from apps.admin_console.database.repositories.session_repository import SessionRepository
from artemis.exploration.store import ExplorationConflict, ExplorationStore, explorations
from artemis.runtime.process_probe import pid_is_alive
from artemis.runtime.task_control import read_state
from artemis.runtime.task_result import read_worker_outcome
from artemis.runtime.case_result import read_case_result
def reconcile_explorations(repository: ExplorationStore = explorations) -> None:
for kind in ("exploration_generation", "exploration_discussion"):
with repository._transaction() as connection:
for row in connection.execute(
"SELECT id, payload FROM automation_records WHERE kind=? AND json_extract(payload,'$.status')='running'",
(kind,),
).fetchall():
job = json.loads(row[1])
if (
not pid_is_alive(job.get("owner_pid"))
or job.get("lease_until", 0) < time.time()
):
repository._put(
connection,
kind,
row[0],
{
**job,
"status": "failed",
"lease_until": 0,
"error": "生成进程已中断;已有结果保留,可重试。",
},
)
with repository.database.connection() as connection:
rows = connection.execute(
"SELECT payload FROM automation_records WHERE kind='exploration' "
"AND json_extract(payload,'$.active_run_id') IS NOT NULL"
).fetchall()
history = SessionRepository(repository.database.path)
queued = {
str(item.get("session_id"))
for item in state.queue_items
if isinstance(item, dict) and item.get("status") in {"pending", "running", "queued"}
}
for row in rows:
session = json.loads(row[0])
run_id = session["active_run_id"]
run = repository.run(run_id)
if not run:
continue
outcome = read_worker_outcome(run_id)
if outcome:
repository.finish_run(run_id, {**outcome, "status": outcome["task_status"]})
continue
control = read_state(run_id)
if run["kind"] == "validation" and pid_is_alive(control.get("pid")):
observed = {"running": "running", "task_paused": "paused", "resuming": "queued"}.get(
control.get("status")
)
if observed and observed != run["status"]:
try:
repository.set_run_state(run_id, observed, reason=control.get("pause_reason"))
except ExplorationConflict:
continue # The owner finalized while this tick read its control file.
if run_id in queued or run_id in state.active_runs:
continue
if control.get("pid") and pid_is_alive(control["pid"]):
continue
execution = history.get_session_by_id(run_id) or {}
if execution.get("pid") and history.process_is_alive(execution["pid"]):
continue
job = repository.database.job(run_id) or {}
terminal = execution.get("status") or job.get("terminal_status") or job.get("status")
if terminal in {"completed", "failed", "cancelled", "interrupted", "skipped"}:
repository.finish_run(
run_id,
{
"status": terminal,
"error": execution.get("error") or job.get("error") or "",
"test_result": execution.get("test_result")
or read_case_result(run_id, terminal),
},
)
elif time.time() - run["created_at"] > 300:
repository.finish_run(
run_id,
{
"status": "interrupted",
"error": "探索工作进程已结束;原证据已保留,可继续探索并重新观察当前设备。",
},
)