Ai/farm_alerts/services.py

from __future__ import annotations

import hashlib
import json
import logging
from typing import Any

from django.apps import apps
from django.core.serializers.json import DjangoJSONEncoder

from farm_data.services import get_farm_details
from farm_data.context import load_farm_context
from rag.api_provider import get_chat_client
from rag.chat import (
    _complete_audit_log,
    _create_audit_log,
    _fail_audit_log,
    _load_service_tone,
    build_rag_context,
)
from rag.config import RAGConfig, get_service_config, load_rag_config

from .models import FarmAlertNotification
from .alerts_tracker import build_farm_alerts_tracker

logger = logging.getLogger(__name__)

KB_NAME = "farm_alerts"
SERVICE_ID = "farm_alerts"

TRACKER_PROMPT = (
    "وضعیت هشدارهای مزرعه را فقط بر اساس داده های ساختاریافته، اطلاعات مزرعه، و متون بازیابی شده از پایگاه دانش تحلیل کن. "
    "پاسخ فقط JSON معتبر باشد و این کلیدها را داشته باشد: headline, overview, status_level, notifications. "
    "status_level فقط یکی از danger, warning, info باشد. "
    "notifications باید آرایه ای از آبجکت ها با کلیدهای level, title, message, suggested_action, source_alert_id, source_metric_type باشد. "
    "سطوح level فقط یکی از danger, warning, info باشند. "
    "اگر هشدار مهمی وجود ندارد، notifications را خالی برگردان."
)

TIMELINE_PROMPT = (
    "بر اساس داده های هشدار مزرعه، یک timeline عملیاتی بساز. "
    "پاسخ فقط JSON معتبر باشد و این کلیدها را داشته باشد: headline, overview, timeline, notifications. "
    "timeline باید آرایه ای از آبجکت ها با کلیدهای timestamp, level, title, description, source_alert_id, source_metric_type باشد. "
    "level فقط danger, warning, info باشد. "
    "notifications باید آرایه ای از آبجکت ها با کلیدهای level, title, message, suggested_action, source_alert_id, source_metric_type باشد."
)


def _json_dumps(value: Any) -> str:
    return json.dumps(value, ensure_ascii=False, indent=2, cls=DjangoJSONEncoder)


def _clean_json_response(raw: str) -> dict[str, Any]:
    cleaned = (raw or "").strip()
    if cleaned.startswith("```"):
        cleaned = cleaned.strip("`")
        if cleaned.startswith("json"):
            cleaned = cleaned[4:]
        cleaned = cleaned.strip()
    if not cleaned:
        return {}
    try:
        return json.loads(cleaned)
    except (json.JSONDecodeError, ValueError):
        logger.warning("Invalid JSON returned by farm_alerts LLM: %s", cleaned[:500])
        return {}


def _severity_to_level(severity: str) -> str:
    normalized = (severity or "").strip().lower()
    if normalized in {"critical", "high", "danger"}:
        return FarmAlertNotification.LEVEL_DANGER
    if normalized in {"medium", "warning"}:
        return FarmAlertNotification.LEVEL_WARNING
    return FarmAlertNotification.LEVEL_INFO


def _normalize_level(level: str | None) -> str:
    normalized = (level or "").strip().lower()
    if normalized in {
        FarmAlertNotification.LEVEL_DANGER,
        FarmAlertNotification.LEVEL_WARNING,
        FarmAlertNotification.LEVEL_INFO,
    }:
        return normalized
    if normalized in {"high", "critical"}:
        return FarmAlertNotification.LEVEL_DANGER
    if normalized in {"medium", "alert"}:
        return FarmAlertNotification.LEVEL_WARNING
    return FarmAlertNotification.LEVEL_INFO


def _alert_identifier(alert: dict[str, Any]) -> str:
    metric_type = alert.get("metric_type", "alert")
    timestamp = alert.get("timestamp", "")
    return f"{metric_type}:{timestamp}"


def _forecast_summary(context: dict[str, Any]) -> list[dict[str, Any]]:
    forecasts = context.get("forecasts", [])
    return [
        {
            "date": getattr(item, "forecast_date", None),
            "temperature_min": getattr(item, "temperature_min", None),
            "temperature_max": getattr(item, "temperature_max", None),
            "humidity_mean": getattr(item, "humidity_mean", None),
            "precipitation": getattr(item, "precipitation", None),
            "et0": getattr(item, "et0", None),
        }
        for item in forecasts[:7]
    ]


def _farm_profile(context: dict[str, Any], farm_uuid: str) -> dict[str, Any]:
    sensor = context.get("sensor")
    location = context.get("location")
    plants = context.get("plants", [])
    irrigation_method = getattr(sensor, "irrigation_method", None) if sensor else None
    return {
        "farm_uuid": farm_uuid,
        "location": {
            "latitude": float(location.latitude) if location else None,
            "longitude": float(location.longitude) if location else None,
        },
        "plant_names": [getattr(plant, "name", "") for plant in plants],
        "irrigation_method": getattr(irrigation_method, "name", None),
        "last_sensor_update": getattr(sensor, "updated_at", None),
    }


def _build_structured_context(farm_uuid: str) -> tuple[dict[str, Any], dict[str, Any]]:
    context = load_farm_context(farm_uuid)
    if context is None:
        raise ValueError("farm_uuid نامعتبر است یا اطلاعات هشدار مزرعه پیدا نشد.")

    tracker = build_farm_alerts_tracker(sensor_id=farm_uuid, context=context, ai_bundle=None)
    structured = {
        "farm_profile": _farm_profile(context, farm_uuid),
        "tracker": tracker,
        "forecasts": _forecast_summary(context),
    }
    return context, structured


def _validate_tracker_response(payload: dict[str, Any]) -> dict[str, Any]:
    required_keys = {"headline", "overview", "status_level", "notifications"}
    missing = [key for key in required_keys if key not in payload]
    if missing:
        raise ValueError(
            "Farm alerts tracker response is missing required fields: "
            + ", ".join(missing)
        )
    if not isinstance(payload.get("notifications"), list):
        raise ValueError("Farm alerts tracker notifications must be a list.")
    return payload


def _validate_timeline_response(payload: dict[str, Any]) -> dict[str, Any]:
    required_keys = {"headline", "overview", "timeline", "notifications"}
    missing = [key for key in required_keys if key not in payload]
    if missing:
        raise ValueError(
            "Farm alerts timeline response is missing required fields: "
            + ", ".join(missing)
        )
    if not isinstance(payload.get("timeline"), list):
        raise ValueError("Farm alerts timeline must be a list.")
    if not isinstance(payload.get("notifications"), list):
        raise ValueError("Farm alerts timeline notifications must be a list.")
    return payload


def _notification_fingerprint(
    *,
    farm_uuid: str,
    endpoint: str,
    level: str,
    title: str,
    source_alert_id: str,
    source_metric_type: str,
) -> str:
    raw = "|".join([
        str(farm_uuid),
        endpoint,
        level,
        source_alert_id or "-",
        source_metric_type or "-",
        title.strip(),
    ])
    return hashlib.sha256(raw.encode("utf-8")).hexdigest()


def _save_notifications(
    *,
    farm_uuid: str,
    endpoint: str,
    notifications: list[dict[str, Any]],
) -> list[FarmAlertNotification]:
    saved: list[FarmAlertNotification] = []
    for item in notifications:
        level = _normalize_level(item.get("level"))
        title = (item.get("title") or "هشدار مزرعه").strip()
        source_alert_id = (item.get("source_alert_id") or "").strip()
        source_metric_type = (item.get("source_metric_type") or "").strip()
        fingerprint = _notification_fingerprint(
            farm_uuid=farm_uuid,
            endpoint=endpoint,
            level=level,
            title=title,
            source_alert_id=source_alert_id,
            source_metric_type=source_metric_type,
        )
        payload = item.get("payload") if isinstance(item.get("payload"), dict) else {}
        notification, _ = FarmAlertNotification.objects.update_or_create(
            fingerprint=fingerprint,
            defaults={
                "farm_uuid": farm_uuid,
                "endpoint": endpoint,
                "level": level,
                "title": title,
                "message": item.get("message") or "",
                "suggested_action": item.get("suggested_action") or "",
                "source_alert_id": source_alert_id,
                "source_metric_type": source_metric_type,
                "payload": payload,
            },
        )
        saved.append(notification)
    return saved


def _serialize_notification(notification: FarmAlertNotification) -> dict[str, Any]:
    return {
        "id": notification.id,
        "farm_uuid": str(notification.farm_uuid),
        "endpoint": notification.endpoint,
        "level": notification.level,
        "title": notification.title,
        "message": notification.message,
        "suggested_action": notification.suggested_action,
        "source_alert_id": notification.source_alert_id,
        "source_metric_type": notification.source_metric_type,
        "payload": notification.payload,
        "created_at": notification.created_at.isoformat(),
        "updated_at": notification.updated_at.isoformat(),
    }


def _build_service_config(cfg: RAGConfig, service_id: str) -> tuple[Any, Any, str, Any]:
    service = get_service_config(service_id, cfg)
    service_cfg = RAGConfig(
        embedding=cfg.embedding,
        qdrant=cfg.qdrant,
        chunking=cfg.chunking,
        llm=service.llm,
        knowledge_bases=cfg.knowledge_bases,
        services=cfg.services,
        chromadb=cfg.chromadb,
    )
    client = get_chat_client(service_cfg)
    model = service.llm.model
    return service, service_cfg, model, client


def _build_messages(
    *,
    prompt: str,
    service: Any,
    cfg: RAGConfig,
    query: str,
    rag_context: str,
    structured_context: dict[str, Any],
) -> tuple[str, list[dict[str, str]]]:
    tone = _load_service_tone(service, cfg)
    system_parts = [tone] if tone else []
    if service.system_prompt:
        system_parts.append(service.system_prompt)
    system_parts.append(prompt)
    system_parts.append("[کانتکست ساختاریافته هشدار مزرعه]\n" + _json_dumps(structured_context))
    if rag_context:
        system_parts.append(rag_context)
    system_prompt = "\n\n".join(part for part in system_parts if part)
    messages = [
        {"role": "system", "content": system_prompt},
        {"role": "user", "content": query},
    ]
    return system_prompt, messages


def _llm_response(
    *,
    farm_uuid: str,
    service_id: str,
    prompt: str,
    query: str,
    structured_context: dict[str, Any],
) -> tuple[dict[str, Any], str, str]:
    cfg = load_rag_config()
    service, service_cfg, model, client = _build_service_config(cfg, service_id)
    farm_details = get_farm_details(farm_uuid)
    rag_context = build_rag_context(
        query=query,
        sensor_uuid=farm_uuid,
        config=cfg,
        kb_name=KB_NAME,
        service_id=service_id,
        farm_details=farm_details,
    )
    system_prompt, messages = _build_messages(
        prompt=prompt,
        service=service,
        cfg=cfg,
        query=query,
        rag_context=rag_context,
        structured_context=structured_context,
    )
    audit_log = _create_audit_log(
        farm_uuid=farm_uuid,
        service_id=service_id,
        model=model,
        query=query,
        system_prompt=system_prompt,
        messages=messages,
    )
    try:
        response = client.chat.completions.create(model=model, messages=messages)
        raw = response.choices[0].message.content.strip()
        parsed = _clean_json_response(raw)
        if not parsed:
            raise ValueError("farm_alerts LLM returned an empty or invalid JSON payload.")
        _complete_audit_log(audit_log, raw)
        return parsed, raw, service.tone_file or ""
    except Exception as exc:
        logger.error("farm_alerts llm error for %s: %s", farm_uuid, exc)
        _fail_audit_log(audit_log, str(exc))
        raise RuntimeError(f"Farm alerts generation failed for farm {farm_uuid}.") from exc


def get_farm_alerts_tracker(*, farm_uuid: str, query: str | None = None) -> dict[str, Any]:
    _, structured_context = _build_structured_context(farm_uuid)
    tracker = structured_context["tracker"]
    user_query = query or "وضعیت فعلی هشدارهای مزرعه را ارزیابی کن و اگر لازم است notification بساز."
    llm_result, raw_response, tone_file = _llm_response(
        farm_uuid=farm_uuid,
        service_id=SERVICE_ID,
        prompt=TRACKER_PROMPT,
        query=user_query,
        structured_context=structured_context,
    )
    llm_result = _validate_tracker_response(llm_result)

    notifications_input = llm_result["notifications"]
    saved_notifications = _save_notifications(
        farm_uuid=farm_uuid,
        endpoint=FarmAlertNotification.ENDPOINT_TRACKER,
        notifications=notifications_input,
    )
    return {
        "farm_uuid": farm_uuid,
        "service_id": SERVICE_ID,
        "tracker": tracker,
        "headline": llm_result["headline"],
        "overview": llm_result["overview"],
        "status_level": _normalize_level(llm_result.get("status_level")),
        "notifications": [_serialize_notification(item) for item in saved_notifications],
        "raw_llm_response": raw_response or None,
        "structured_context": structured_context,
    }


def get_farm_alerts_timeline(*, farm_uuid: str, query: str | None = None) -> dict[str, Any]:
    _, structured_context = _build_structured_context(farm_uuid)
    tracker = structured_context["tracker"]
    user_query = query or "برای هشدارهای مزرعه یک timeline عملیاتی بساز و اگر لازم است notification ثبت کن."
    llm_result, raw_response, tone_file = _llm_response(
        farm_uuid=farm_uuid,
        service_id=SERVICE_ID,
        prompt=TIMELINE_PROMPT,
        query=user_query,
        structured_context=structured_context,
    )
    llm_result = _validate_timeline_response(llm_result)

    timeline = llm_result["timeline"]
    notifications_input = llm_result["notifications"]

    saved_notifications = _save_notifications(
        farm_uuid=farm_uuid,
        endpoint=FarmAlertNotification.ENDPOINT_TIMELINE,
        notifications=notifications_input,
    )
    return {
        "farm_uuid": farm_uuid,
        "service_id": SERVICE_ID,
        "tracker": tracker,
        "headline": llm_result["headline"],
        "overview": llm_result["overview"],
        "timeline": timeline,
        "notifications": [_serialize_notification(item) for item in saved_notifications],
        "raw_llm_response": raw_response or None,
        "structured_context": structured_context,
    }