def test_dynamic_graph(self):
class DynTaskA(Task):
def __init__(self):
self._output_datakeys = []
def output_datakeys(self):
return self._output_datakeys
def main(self, ds: DataSet):
self._output_datakeys = ["DynTaskA"]
return DataSet()
class DynTaskB(Task):
def input_datakeys(self):
return ["DynTaskA"]
def main(self, ds: DataSet):
return DataSet()
class DynTaskC(Task):
def input_datakeys(self):
return ["DynTaskX"]
def main(self, ds: DataSet):
return DataSet()
#
g = Graph()
g.append(DynTaskA())
gtb = g.append(DynTaskB())
gtc = g.append(DynTaskC())
g.run()
self.assertEqual(TaskStatus.COMPLETED, gtb.status)
self.assertEqual(TaskStatus.INIT, gtc.status)
def test_auto_resolver(self):
g = Graph()
g.append(TaskA())
g.append(TaskB())
taskC = g.append(TaskC())
g.autoresolve_dependencies()
self.assertEqual(2, len(taskC.dependencies))
def test_error_handler(self):
def value_error_handler(e, ds):
print(str(e))
print(str(ds))
g = Graph()
g.append(ErrorTask())
g.add_error_handler(ValueError, value_error_handler)
g.run()
def test_auto_resolver_error(self):
g = Graph()
g.append(TaskA())
g.append(TaskB())
g.append(TaskC())
g.append(TaskD())
with self.assertRaises(ValueError):
g.autoresolve_dependencies()
def test_default_ds(self):
class TestTask(Task):
def main(self, ds: DataSet):
if "default" not in ds.keys():
raise ValueError()
return DataSet()
#
default_ds = DataSet().put("default", JsonData({}))
g = Graph()
g.append(TestTask())
g.run(default_ds)
def test_catalog_ds(self):
class TestTask(Task):
def main(self, ds: DataSet):
if "catalog" not in ds.keys():
raise ValueError()
return DataSet()
catalog_ds = DataSet().put("catalog", JsonData({}))
g = Graph(catalog_ds=catalog_ds)
g.append(TestTask())
g.run()
def add_tasks_to_graph(graph: Graph, tasks: list):
added_tasks = {}
def exist_all_dpes(at, t):
ds = t["dependencies"]
return all([d in at for d in ds])
def all_deps(at, t):
ds = t["dependencies"]
deps = [at[d] for d in ds]
return deps
while len(added_tasks) < len(tasks):
for k, t in tasks.items():
if k in added_tasks:
# 追加済
continue
# 追加可能なら追加
if exist_all_dpes(added_tasks, t):
gt = graph.append(t["instance"], all_deps(added_tasks, t))
added_tasks[k] = gt
break
else:
# 依存を解決できないタスクがあった
for k, t in tasks.items():
if k not in added_tasks:
logger.error(f"{t['instance'].__class__.__name__} can't resolve dependencies")
raise ValueError("build graph failed")
def test_hook(self):
g = Graph()
gt = g.append(DumbTask())
global hook_called
hook_called = 0
def hook(ds):
global hook_called
hook_called += 1
gt.pre_run_hook = hook
gt.post_run_hook = hook
g.run()
self.assertEqual(2, hook_called)
logger.info("execute TaskC")
return DataSet()
if __name__ == "__main__":
basicConfig(level=DEBUG)
parser = argparse.ArgumentParser()
parser.add_argument("-g", "--graph", type=str, default="normal", help="Graph {normal/thread/process}")
args = parser.parse_args()
if args.graph == "normal":
graph = Graph()
logger.info("use Graph")
elif args.graph == "thread":
graph = ConcurrentGraph(ThreadPoolExecutor())
logger.info("use ConcurrentGraph(ThreadPoolExecutor)")
elif args.graph == "process":
graph = ConcurrentGraph(ProcessPoolExecutor())
logger.info("use ConcurrentGraph(ProcessPoolExecutor)")
else:
logger.error("unknown graph,")
sys.exit(-1)
# Graphで処理する
graph.append(TaskA())
graph.append(TaskB())
graph.append(TaskC())
graph.run()
print(f"PlantUML Diagraph : {PlantUML.graph_to_url(graph)}")
rds = DataSet()
rds.put("titanic", DataFrameData(df))
return rds
# DAG定義
default_args = {
"owner": "airflow",
"depends_on_past": False,
"start_date": datetime(2020, 6, 11),
"email": ["*****@*****.**"],
"email_on_failure": False,
"email_on_retry": False,
"retries": 1,
"retry_delay": timedelta(minutes=5),
}
dag = DAG("graph_to_dag", default_args=default_args, schedule_interval=None)
# Graph作成
graph = Graph()
readcsv = graph.append(ReadCsv())
fill_age = graph.append(FillNaMedian("Age"), [readcsv])
sex_to_code = graph.append(SexToCode(), [readcsv])
embarked_to_code = graph.append(EmbarkedToCode(), [readcsv])
graph.append(Merge(), [readcsv, fill_age, sex_to_code, embarked_to_code])
# Graphのタスクを AirFlowに登録
af = AirFlow(graph)
af.to_dag(dag)
def test_no_handler(self):
g = Graph()
g.append(ErrorTask())
with self.assertRaises(ValueError):
g.run()
ds = DataSet()
ds.put("titanic", DataFrameData(df))
return ds
if __name__ == "__main__":
basicConfig(level=DEBUG)
# データセットの読み込み
ds = DataSet()
repo = LocalFileRepository(
Path(os.path.dirname(__file__)) / Path("../titanic.csv"))
titanic_data = DataFrameData.load(repo)
ds.put("titanic", titanic_data)
#
print("## Original data")
print(ds.get("titanic").content)
# Graphで処理する
# Age欠損埋め -> 性別のコード化 -> 乗船した港 のコード化 の順で処理
graph = Graph()
fill_age = graph.append(FillNaMedian("Age"))
sex_to_code = graph.append(SexToCode(), [fill_age])
graph.append(EmbarkedToCode(), [sex_to_code])
ds = graph.run(ds)
print("## Processed data")
print(ds.get("titanic").content)
sys.exit(-1)
# データセットの読み込み
ds = DataSet()
repo = LocalFileRepository(
Path(os.path.dirname(__file__)) / Path("../titanic.csv"))
titanic_data = DataFrameData.load(repo)
ds.put("titanic", titanic_data)
#
print("## Original data")
print(ds.get("titanic").content)
# Graphで処理する
# Age欠損埋め・性別のコード化・乗船した港のコード化 を順不同で実行してマージ
passthrough = graph.append(PassThrough())
fill_age = graph.append(FillNaMedian("Age"))
sex_to_code = graph.append(SexToCode())
embarked_to_code = graph.append(EmbarkedToCode())
graph.append(Merge(),
[passthrough, fill_age, sex_to_code, embarked_to_code])
ds = graph.run(ds)
print("## Processed data")
print(ds.get("titanic").content)
print(f"PlantUML Diagraph : {PlantUML.graph_to_url(graph)}")