Skip to content

airflow_pipeline_writer

AirflowPipelineWriter

Bases: BasePipelineWriter

Class for pipeline file writer. Corresponds to "AIRFLOW" framework.

Source code in lineapy/plugins/airflow_pipeline_writer.py
 52
 53
 54
 55
 56
 57
 58
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
class AirflowPipelineWriter(BasePipelineWriter):
    """
    Class for pipeline file writer. Corresponds to "AIRFLOW" framework.
    """

    @property
    def docker_template_name(self) -> str:
        return "airflow_dockerfile.jinja"

    def _write_dag(self) -> None:

        # Check if the given DAG flavor is a supported/valid one
        try:
            dag_flavor = AirflowDagFlavor[
                self.dag_config.get("dag_flavor", "PythonOperatorPerArtifact")
            ]
        except KeyError:
            raise ValueError(
                f'"{dag_flavor}" is an invalid airflow dag flavor.'
            )

        try:
            task_serialization = TaskSerializer[
                self.dag_config.get("task_serialization", "LocalPickle")
            ]
        except KeyError:
            raise ValueError(
                f'"{task_serialization}" is an invalid type of task serialization scheme.'
            )

        # Construct DAG text for the given flavor
        full_code = self._write_operators(dag_flavor, task_serialization)

        # Write out file
        file = self.output_dir / f"{self.pipeline_name}_dag.py"
        file.write_text(prettify(full_code))
        logger.info(f"Generated DAG file: {file}")

    def _write_operators(
        self,
        dag_flavor: AirflowDagFlavor,
        task_serialization: TaskSerializer,
    ) -> str:
        """
        This method implements Airflow DAG code generation corresponding
        to the following flavors

        - ``PythonOperatorPerSession`` flavor, where each session gets its
        own Python operator.
        - ``PythonOperatorPerArtifact`` flavor, where each artifact gets its own
        Python operator.

        Example of ``PythonOperatorPerSession`` if the two artifacts in our pipeline
        (e.g., model and prediction) were created in the same session.
        ``` python
            import pickle
            import g2_z_module
            ...
            def task_run_session_including_g2():
                artifacts = g2_z_module.run_session_including_g2()
                pickle.dump(artifacts["g2"], open("/tmp/g2_z/artifact_g2.pickle", "wb"))
                pickle.dump(artifacts["z"], open("/tmp/g2_z/artifact_z.pickle", "wb"))
            with DAG(...) as dag:
                run_session_including_g2 = PythonOperator(
                    task_id="run_session_including_g2_task",
                    python_callable=task_run_session_including_g2,
                )
        ```
        Example of ``PythonOperatorPerArtifact``, if the two artifacts in our pipeline
        (e.g., model and prediction) were created in the same session:
        ``` python
            import pickle
            import iris_module
            ...
            def task_iris_model():
                mod = iris_module.get_iris_model()
                pickle.dump(mod, open("/tmp/iris/variable_mod.pickle", "wb"))
            def task_iris_pred():
                mod = pickle.load(open("/tmp/iris/variable_mod.pickle", "rb"))
                pred = iris_module.get_iris_pred(mod)
                pickle.dump(
                    pred, open("/tmp/iris/variable_pred.pickle", "wb")
                )
            with DAG(...) as dag:
                iris_model = PythonOperator(
                    task_id="iris_model_task",
                    python_callable=task_iris_model,
                )
                iris_pred = PythonOperator(
                    task_id="iris_pred_task",
                    python_callable=task_iris_pred,
                )
            iris_model >> iris_pred
        ```
        This way, the generated Airflow DAG file opens room for engineers
        to control pipeline runs at a finer level and allows for further customization.
        """

        DAG_TEMPLATE = load_plugin_template("airflow_dag_PythonOperator.jinja")

        if dag_flavor == AirflowDagFlavor.PythonOperatorPerSession:
            task_breakdown = DagTaskBreakdown.TaskPerSession
        elif dag_flavor == AirflowDagFlavor.PythonOperatorPerArtifact:
            task_breakdown = DagTaskBreakdown.TaskPerArtifact

        # Get task definitions based on dag_flavor
        task_defs, task_graph = get_task_graph(
            self.artifact_collection,
            pipeline_name=self.pipeline_name,
            task_breakdown=task_breakdown,
        )

        # Add setup and teardown if local pickle serializer is selected
        if task_serialization == TaskSerializer.LocalPickle:
            task_defs["setup"] = get_localpickle_setup_task_definition(
                self.pipeline_name
            )
            task_defs["teardown"] = get_localpickle_teardown_task_definition(
                self.pipeline_name
            )
            # insert in order to task_names so that setup runs first and teardown runs last
            task_graph.insert_setup_task("setup")
            task_graph.insert_teardown_task("teardown")

        rendered_task_defs = self.get_rendered_task_definitions(
            task_defs, task_serialization
        )

        task_dependencies = sorted(
            [f"{task0} >> {task1}" for task0, task1 in task_graph.graph.edges]
        )

        # Get DAG parameters for an Airflow pipeline
        input_parameters_dict: Dict[str, Any] = {}
        for parameter_name, input_spec in super().get_pipeline_args().items():
            input_parameters_dict[parameter_name] = input_spec.value

        full_code = DAG_TEMPLATE.render(
            DAG_NAME=self.pipeline_name,
            MODULE_NAME=self.pipeline_name + "_module",
            OWNER=self.dag_config.get("owner", "airflow"),
            RETRIES=self.dag_config.get("retries", 2),
            START_DATE=self.dag_config.get("start_date", "days_ago(1)"),
            SCHEDULE_INTERVAL=self.dag_config.get(
                "schedule_interval", "*/15 * * * *"
            ),
            MAX_ACTIVE_RUNS=self.dag_config.get("max_active_runs", 1),
            CATCHUP=self.dag_config.get("catchup", "False"),
            dag_params=input_parameters_dict,
            task_definitions=rendered_task_defs,
            tasks=task_defs,
            task_dependencies=task_dependencies,
        )

        return full_code

    def get_rendered_task_definitions(
        self,
        task_defs: Dict[str, TaskDefinition],
        task_serialization: TaskSerializer,
    ) -> List[str]:
        """
        Returns rendered tasks for the pipeline tasks.
        """
        TASK_FUNCTION_TEMPLATE = load_plugin_template(
            "task/task_function.jinja"
        )
        rendered_task_defs: List[str] = []
        for task_name, task_def in task_defs.items():
            loading_blocks, dumping_blocks = render_task_io_serialize_blocks(
                task_def, task_serialization
            )
            task_def_rendered = TASK_FUNCTION_TEMPLATE.render(
                function_name=task_name,
                user_input_variables=", ".join(task_def.user_input_variables),
                typing_blocks=task_def.typing_blocks,
                loading_blocks=loading_blocks,
                pre_call_block=task_def.pre_call_block,
                call_block=task_def.call_block,
                post_call_block=task_def.post_call_block,
                dumping_blocks=dumping_blocks,
                return_block="",
                include_imports_locally=False,
            )
            rendered_task_defs.append(task_def_rendered)

        # sort here to maintain deterministic behavior of writing
        return sorted(rendered_task_defs)

get_rendered_task_definitions(task_defs, task_serialization)

Returns rendered tasks for the pipeline tasks.

Source code in lineapy/plugins/airflow_pipeline_writer.py
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
def get_rendered_task_definitions(
    self,
    task_defs: Dict[str, TaskDefinition],
    task_serialization: TaskSerializer,
) -> List[str]:
    """
    Returns rendered tasks for the pipeline tasks.
    """
    TASK_FUNCTION_TEMPLATE = load_plugin_template(
        "task/task_function.jinja"
    )
    rendered_task_defs: List[str] = []
    for task_name, task_def in task_defs.items():
        loading_blocks, dumping_blocks = render_task_io_serialize_blocks(
            task_def, task_serialization
        )
        task_def_rendered = TASK_FUNCTION_TEMPLATE.render(
            function_name=task_name,
            user_input_variables=", ".join(task_def.user_input_variables),
            typing_blocks=task_def.typing_blocks,
            loading_blocks=loading_blocks,
            pre_call_block=task_def.pre_call_block,
            call_block=task_def.call_block,
            post_call_block=task_def.post_call_block,
            dumping_blocks=dumping_blocks,
            return_block="",
            include_imports_locally=False,
        )
        rendered_task_defs.append(task_def_rendered)

    # sort here to maintain deterministic behavior of writing
    return sorted(rendered_task_defs)

Was this helpful?

Help us improve docs with your feedback!