Реализация rule-engine-service: RabbitMQ → правила из PostgreSQL → gRPC

Слушает telemetry.new_reading (ручной ack/nack, реквизишн только при транспортных ошибках gRPC), кэширует активные automation_rules в памяти с периодическим обновлением из PostgreSQL (JOIN с devices — резолвит внутренние ID в external_id, которым оперируют MQTT/Redis/gRPC). Условия правил (>,<,>=,<=,=,!=) оцениваются обобщённо, без привязки к конкретным типам устройств. При срабатывании вызывает device-control-service по gRPC и публикует automation.rule_triggered в RabbitMQ. Проверено сквозным тестом через docker compose на полном пайплайне: mosquitto_pub → ingest-service (ClickHouse + telemetry.new_reading) → rule-engine-service (совпадение правила) → device-control-service (gRPC → Redis desired_state + MQTT-команда) → automation.rule_triggered. Показание ниже порога проверено отдельно — правило корректно не срабатывает.
2026-07-26 21:18:59 +05:00
parent 3eb84aae6c
commit abc83faf05
20 changed files with 1334 additions and 10 deletions
@@ -0,0 +1,96 @@
+// Package engine orchestrates the core rule-engine loop: given one sensor
+// reading, find the rules that watch it, evaluate their conditions, and
+// dispatch the matching actions. It knows nothing about RabbitMQ, gRPC, or
+// PostgreSQL directly — those are injected as narrow interfaces so this
+// package is testable without any of them running.
+package engine
+
+import (
+	"context"
+	"log/slog"
+	"time"
+
+	"git.cactoz.su/cacto/home_automatization/services/rule-engine-service/internal/devicecontrolclient"
+	"git.cactoz.su/cacto/home_automatization/services/rule-engine-service/internal/rules"
+)
+
+// Reading is the generic sensor reading the engine reacts to — no
+// growbox/device-type-specific fields, matching the platform's abstractions.
+type Reading struct {
+	DeviceID   string
+	ZoneID     string
+	SensorType string
+	Value      float64
+	RecordedAt time.Time
+}
+
+// RuleSource is the subset of rulecache.Cache the engine needs.
+type RuleSource interface {
+	MatchingRules(deviceID, sensorType string) []rules.Rule
+}
+
+// Dispatcher is the subset of devicecontrolclient.Client the engine needs.
+type Dispatcher interface {
+	Dispatch(ctx context.Context, deviceID, actionType string, params map[string]any) (devicecontrolclient.Result, error)
+}
+
+// TriggerPublisher is the subset of rabbitmq.Client the engine needs.
+type TriggerPublisher interface {
+	PublishRuleTriggered(ctx context.Context, ruleID, zoneID int64, deviceID, actionType string, success bool, errMsg string) error
+}
+
+type Engine struct {
+	rules      RuleSource
+	dispatcher Dispatcher
+	publisher  TriggerPublisher
+	logger     *slog.Logger
+}
+
+func New(ruleSource RuleSource, dispatcher Dispatcher, publisher TriggerPublisher, logger *slog.Logger) *Engine {
+	return &Engine{rules: ruleSource, dispatcher: dispatcher, publisher: publisher, logger: logger}
+}
+
+// HandleReading evaluates every rule watching (r.DeviceID, r.SensorType) and
+// dispatches the ones whose condition matches. It returns a non-nil error
+// only when a dispatch failed at the transport level (e.g.
+// device-control-service unreachable) — that's the one case worth retrying
+// the whole reading for. A rule the device itself rejected (bad device_id,
+// unsupported action) is terminal and does not cause a retry.
+func (e *Engine) HandleReading(ctx context.Context, r Reading) error {
+	matched := e.rules.MatchingRules(r.DeviceID, r.SensorType)
+
+	var firstTransportErr error
+	for _, rule := range matched {
+		matchesCondition, err := rules.Evaluate(rule.ConditionOperator, r.Value, rule.ConditionValue)
+		if err != nil {
+			e.logger.Error("skipping rule with invalid condition", "rule_id", rule.ID, "error", err)
+			continue
+		}
+		if !matchesCondition {
+			continue
+		}
+
+		result, dispatchErr := e.dispatcher.Dispatch(ctx, rule.TargetDeviceID, rule.ActionType, rule.ActionParams)
+
+		success := dispatchErr == nil && result.Success
+		errMsg := result.Error
+		switch {
+		case dispatchErr != nil:
+			errMsg = dispatchErr.Error()
+			e.logger.Error("dispatch action failed", "rule_id", rule.ID, "target_device_id", rule.TargetDeviceID, "error", dispatchErr)
+			if firstTransportErr == nil {
+				firstTransportErr = dispatchErr
+			}
+		case !result.Success:
+			e.logger.Warn("device rejected command", "rule_id", rule.ID, "target_device_id", rule.TargetDeviceID, "error", result.Error)
+		default:
+			e.logger.Info("rule triggered", "rule_id", rule.ID, "target_device_id", rule.TargetDeviceID, "action_type", rule.ActionType)
+		}
+
+		if pubErr := e.publisher.PublishRuleTriggered(ctx, rule.ID, rule.ZoneID, rule.TargetDeviceID, rule.ActionType, success, errMsg); pubErr != nil {
+			e.logger.Error("publish rule_triggered event failed", "rule_id", rule.ID, "error", pubErr)
+		}
+	}
+
+	return firstTransportErr
+}