datax-admin/app/core/airflow/task.py

import json
from app.core.airflow.uri import spark_result_tb_name
from app.schemas import AirflowTask
from jinja2 import Environment, PackageLoader, select_autoescape
from app.common.minio import FileHandler


class TaskCompiler:
    def __init__(self, item: AirflowTask):
        self.task = item
        self.default_image = None
        self.default_cmd = None

    @staticmethod
    def render_spark_script(parameters, template_file):
        env = Environment(
            loader=PackageLoader('app.core.airflow'),
            autoescape=select_autoescape()
        )
        template = env.get_template(template_file)
        return template.render(parameters)

    def translate(self, task_mode=1):
        return {'image': self.task.run_image or self.default_image,
                'cmds': ["/bin/bash", "-c", f"{self.task.cmd or self.default_cmd} "],
                'script': self.task.script,
                'id': f'{self.task.id}',
                'env': {**{"SCRIPT": self.task.script}, **self.task.envs},
                'operator_name': f'op_{self.task.id}',
                'name': self.task.name,
                'desc': ""
                }

    @staticmethod
    def write_to_oss(oss_path, context, bucket='mytest'):
        if isinstance(context, str):
            context = bytes(context, 'utf-8')
        minio_handler = FileHandler(bucket_name=bucket)
        return minio_handler.put_byte_file(file_name=oss_path, file_content=context)


class JavaTaskCompiler(TaskCompiler):
    def __init__(self, item: AirflowTask):
        super(JavaTaskCompiler, self).__init__(item)
        self.default_image = 'SXKJ:32775/java:1.0'
        self.default_cmd = "echo \"$SCRIPT\" > run.py && python run.py"
        self.task.cmd = self.task.cmd or self.default_cmd
        tar_name = self.task.file_urls[0].split('/')[-1].split('_')[-1]
        self.task.cmd = f'curl {"http://minio.default:9000"}/{self.task.file_urls[0]}  --output {tar_name} && {self.task.cmd}'


class PythonTaskCompiler(TaskCompiler):
    def __init__(self, item: AirflowTask):
        super(PythonTaskCompiler, self).__init__(item)
        self.default_image = 'SXKJ:32775/pod_python:1.1'
        self.default_cmd = "echo \"$SCRIPT\" > run.py && python run.py"


class DataXTaskCompiler(TaskCompiler):
    def __init__(self, item: AirflowTask):
        super(DataXTaskCompiler, self).__init__(item)
        self.default_image = 'SXKJ:32775/pod_datax:0.9'
        # self.default_cmd = f"cd datax/bin && echo $SCRIPT > config.json && echo " \
        #                    f"\"\'\"$HOME/conda/envs/py27/bin/python datax.py {self.task.cmd_parameters} config.json" \
        #                    f"\"\'\" |xargs bash -c "
        self.default_cmd = f"cd datax/bin && echo \"$SCRIPT\" > transform_datax.py &&cat  transform_datax.py  && python3 transform_datax.py && cat config.json && $HOME/conda/envs/py27/bin/python datax.py {self.task.cmd_parameters} config.json"

    def translate(self, task_mode=1):
        print(f'{self.task.envs}')
        script_str = self.render_spark_script(
            parameters={'script': self.task.script,
                        'first_begin_time': self.task.envs.get('first_begin_time', None),
                        'last_key': self.task.envs.get('last_key', None),
                        'current_key': self.task.envs.get('current_key', None),
                        'partition_key': self.task.envs.get('partition_key', None),
                        'partition_word': self.task.envs.get('partition_word', None),
                        'partition_format': self.task.envs.get('partition_format', None),
                        'partition_diff': self.task.envs.get('partition_diff', None),
                        },
            template_file="transform_datax.py.jinja2")
        # with open('./auto_generate_demo.py','w') as f:
        #     f.write(script_str)
        res = {'image': self.task.run_image or self.default_image,
               'cmds': ["/bin/bash", "-c", f"{self.task.cmd or self.default_cmd} "],
               'script': script_str,
               'id': f'{self.task.id}',
               'env': {**{"SCRIPT": script_str}, **self.task.envs},
               'operator_name': f'op_{self.task.id}',
               'name': self.task.name,
               'desc': ""
               }

        return res


class SparksTaskCompiler(TaskCompiler):
    def __init__(self, item: AirflowTask):
        super(SparksTaskCompiler, self).__init__(item)
        self.default_image = 'SXKJ:32775/jupyter:0.96'
        parameters = {"master": "yarn",
                      "deploy-mode": "cluster",
                      "driver-memory": "2g",
                      "driver-cores ": 1,
                      "executor-memory": "2g",
                      "executor-cores": 4,
                      "num-executors": 1,
                      "archives": "/home/sxkj/bigdata/py37.zip#python3env"
                      }
        spark_config = {'spark.default.parallelism': 2,
                        "spark.executor.memoryOverhead": "4g",
                        "spark.driver.memoryOverhead": "2g",
                        "spark.yarn.maxAppAttempts": 3,
                        "spark.yarn.submit.waitAppCompletion": "true",
                        "spark.pyspark.driver.python": "python3env/py37/bin/python",
                        "spark.yarn.appMasterEnv.PYSPARK_PYTHON": "python3env/py37/bin/python",
                        "spark.pyspark.python": "python3env/py37/bin/python"
                        }
        param_str = ' '.join([f'--{k} {v}' for k, v in parameters.items()])
        param_str += ''.join([f' --conf {k}={v}' for k, v in spark_config.items()])
        basic_cmds = "cd /home/sxkj/bigdata && echo \"$SCRIPT\" > run.py && ${SPARK_HOME}/bin/spark-submit"
        self.cmd_str = lambda name: f"{basic_cmds} --name {name} {param_str} run.py"

    def translate(self, task_mode=1):
        # dag_script = {
        #     "sub_nodes": [
        #         {
        #             "id": "1",
        #             "name": "SqlNode1",
        #             "op": "sql",
        #             "script": "select * from train",
        #         },
        #         {
        #             "id": "2",
        #             "name": "SqlNode1",
        #             "op": "sql",
        #             "script": "select * from test",
        #         },
        #         {
        #             "id": "3",
        #             "name": "PysparkNode1",
        #             "op": "pyspark",  # or python
        #             "inputs": {'train': ("1", 0),
        #                        'test': ("2", 0)
        #                        },
        #             "script": "import os\n ...",
        #         },
        #     ],
        #     "edges": [
        #         ("1", "3"),
        #         ("2", "3")
        #     ]
        # }
        infos = json.loads(self.task.script)
        sub_nodes = []
        for info in infos['sub_nodes']:
            if info['op'] == 'sql':
                inputs = {}
                template_file = 'sql_script_template.py.jinja2'
            elif info['op'] == 'pyspark':
                inputs = {k: spark_result_tb_name(self.task.id, *v, task_mode) for k, v in info['inputs'].items()}
                template_file = 'pyspark_script_template.py.jinja2'
            else:
                continue
            outputs = [spark_result_tb_name(self.task.id, info['id'], 0, task_mode)]
            sub_node = {
                'id': f'{self.task.id}_{info["id"]}',
                'name': info['name'],
                'env': {
                    'SCRIPT': self.render_spark_script(
                        parameters={'script': info['script'], 'inputs': inputs, 'outputs': outputs},
                        template_file=template_file),
                },
                'cmds': ['/bin/bash', '-c', self.cmd_str(name=f'spark_{self.task.id}_{info["id"]}')],
                'image': "SXKJ:32775/jupyter:0.96",
            }
            sub_nodes.append(sub_node)

        edges = [(f'{self.task.id}_{source}', f'{self.task.id}_{sink}') for (source, sink) in infos['edges']]
        return {
            "id": self.task.id,
            "sub_nodes": sub_nodes,
            "edges": edges,
            'name': self.task.name,
            'desc': "first spark dag task"
        }