123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123 |
- from airflow import DAG
- from airflow.operators.python import PythonOperator
- from airflow.operators.empty import EmptyOperator
- from airflow.sensors.external_task import ExternalTaskSensor
- from datetime import datetime, timedelta
- from utils import (
- get_enabled_tables, is_data_model_table, run_model_script, get_model_dependency_graph,
- check_table_relationship, process_model_tables
- )
- from config import NEO4J_CONFIG, TASK_RETRY_CONFIG
- import pendulum
- import logging
- import networkx as nx
- # 创建日志记录器
- logger = logging.getLogger(__name__)
- def generate_optimized_execution_order(table_names: list) -> list:
- """
- 生成优化的执行顺序,可处理循环依赖
- 参数:
- table_names: 表名列表
- 返回:
- list: 优化后的执行顺序列表
- """
- # 创建依赖图
- G = nx.DiGraph()
-
- # 添加所有节点
- for table_name in table_names:
- G.add_node(table_name)
-
- # 添加依赖边
- dependency_dict = get_model_dependency_graph(table_names)
- for target, upstreams in dependency_dict.items():
- for upstream in upstreams:
- if upstream in table_names: # 确保只考虑目标表集合中的表
- G.add_edge(upstream, target)
-
- # 检测循环依赖
- cycles = list(nx.simple_cycles(G))
- if cycles:
- logger.warning(f"检测到循环依赖,将尝试打破循环: {cycles}")
- # 打破循环依赖(简单策略:移除每个循环中的一条边)
- for cycle in cycles:
- # 移除循环中的最后一条边
- G.remove_edge(cycle[-1], cycle[0])
- logger.info(f"打破循环依赖: 移除 {cycle[-1]} -> {cycle[0]} 的依赖")
-
- # 生成拓扑排序
- try:
- execution_order = list(nx.topological_sort(G))
- return execution_order
- except Exception as e:
- logger.error(f"生成执行顺序失败: {str(e)}")
- # 返回原始列表作为备选
- return table_names
- def is_first_day():
- return True
- # 生产环境中应使用实际判断
- # return pendulum.now().day == 1
- with DAG("dag_data_model_monthly", start_date=datetime(2024, 1, 1), schedule_interval="@daily", catchup=False) as dag:
- logger.info("初始化 dag_data_model_monthly DAG")
-
- # 修改依赖关系:直接依赖于daily.py而不是weekly.py
- wait_for_daily = ExternalTaskSensor(
- task_id="wait_for_daily_model",
- external_dag_id="dag_data_model_daily",
- external_task_id="daily_processing_completed", # 指定完成标记任务
- mode="poke",
- timeout=3600,
- poke_interval=30
- )
- logger.info("创建日模型等待任务 - wait_for_daily_model")
-
- # 创建一个完成标记任务,确保即使没有处理任务也能标记DAG完成
- monthly_completed = EmptyOperator(
- task_id="monthly_processing_completed",
- dag=dag
- )
- logger.info("创建任务完成标记 - monthly_processing_completed")
-
- # 检查今天是否是月初
- if is_first_day():
- logger.info("今天是月初,开始处理月模型")
- # 获取启用的 monthly 模型表
- try:
- enabled_tables = get_enabled_tables("monthly")
-
- # 特别检查两个表之间的关系(这是monthly.py特有的)
- table_names = [t['table_name'] for t in enabled_tables if is_data_model_table(t['table_name'])]
- if 'book_sale_amt_yearly' in table_names and 'book_sale_amt_monthly' in table_names:
- logger.info("特别检查 book_sale_amt_yearly 和 book_sale_amt_monthly 之间的关系")
- relationship = check_table_relationship('book_sale_amt_yearly', 'book_sale_amt_monthly')
- logger.info(f"关系检查结果: {relationship}")
-
- # 定义monthly特有的任务选项
- task_options = {
- 'default': {
- 'retries': TASK_RETRY_CONFIG["retries"],
- 'retry_delay': timedelta(minutes=TASK_RETRY_CONFIG["retry_delay_minutes"])
- },
- 'book_sale_amt_monthly': {
- 'trigger_rule': "none_failed",
- 'retries': TASK_RETRY_CONFIG["retries"],
- 'retry_delay': timedelta(minutes=TASK_RETRY_CONFIG["retry_delay_minutes"])
- }
- }
-
- # 使用公共函数处理模型表 - 修改依赖任务名称
- process_model_tables(enabled_tables, "monthly", wait_for_daily, monthly_completed, dag, **task_options)
-
- except Exception as e:
- logger.error(f"获取 monthly 模型表时出错: {str(e)}")
- # 出错时也要确保完成标记被触发 - 修改依赖任务名称
- wait_for_daily >> monthly_completed
- raise
- else:
- # 如果不是月初,直接将等待任务与完成标记相连接,跳过处理 - 修改依赖任务名称
- logger.info("今天不是月初,跳过月模型处理")
- wait_for_daily >> monthly_completed
|