Skip to content

kubeflow_pipeline_writer

KubeflowPipelineWriter

Bases: BasePipelineWriter

Source code in lineapy/plugins/kubeflow_pipeline_writer.py
 42
 43
 44
 45
 46
 47
 48
 49
 50
 51
 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
class KubeflowPipelineWriter(BasePipelineWriter):
    def _write_dag(self) -> None:
        # Check if the given DAG flavor is a supported/valid one
        try:
            dag_flavor = KubeflowDagFlavor[
                self.dag_config.get("dag_flavor", "ComponentPerArtifact")
            ]
        except KeyError:
            raise ValueError(
                f'"{dag_flavor}" is an invalid kubeflow dag flavor.'
            )

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

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

    def _write_operators(
        self,
        dag_flavor: KubeflowDagFlavor,
    ) -> str:
        """
        Returns a code block containing all the operators for a Kubeflow DAG.
        """

        DAG_TEMPLATE = load_plugin_template("kubeflow_dag.jinja")

        if dag_flavor == KubeflowDagFlavor.ComponentPerSession:
            task_breakdown = DagTaskBreakdown.TaskPerSession
        elif dag_flavor == KubeflowDagFlavor.ComponentPerArtifact:
            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,
        )

        task_defs["setup"] = get_noop_setup_task_definition(self.pipeline_name)
        task_defs["teardown"] = get_noop_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")

        task_names = list(task_defs.keys())
        task_defs = {tn: task_defs[tn] for tn in task_names}

        (
            rendered_task_defs,
            task_loading_blocks,
        ) = self.get_rendered_task_definitions(task_defs)

        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

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

        full_code = DAG_TEMPLATE.render(
            DAG_NAME=self.pipeline_name,
            HOST_URL=self.dag_config.get("host_url", "http://localhost:3000"),
            dag_params=input_parameters_dict,
            task_definitions=rendered_task_defs,
            tasks=task_defs,
            task_loading_blocks=task_loading_blocks,
            task_dependencies=task_dependencies,
        )

        return prettify(full_code)

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

    def get_rendered_task_definitions(
        self,
        task_defs: Dict[str, TaskDefinition],
    ) -> Tuple[List[str], Dict[str, str]]:
        """
        Returns rendered tasks for the pipeline tasks along with a dictionary to lookup
        previous task outputs.

        The returned dictionary is used by the DAG to connect the right input files to
        output files for inter task communication.
        """
        TASK_FUNCTION_TEMPLATE = load_plugin_template(
            "task/task_function.jinja"
        )
        rendered_task_defs: List[str] = []
        task_loading_blocks: Dict[str, str] = {}

        for task_name, task_def in task_defs.items():
            loading_blocks, dumping_blocks = render_task_io_serialize_blocks(
                task_def, TaskSerializer.ParametrizedPickle
            )

            input_vars = task_def.user_input_variables

            input_paths = [
                f"variable_{loaded_input_variable}_path: kfp.components.InputPath(str)"
                for loaded_input_variable in task_def.loaded_input_variables
            ]

            output_paths = [
                f"variable_{return_variable}_path: kfp.components.OutputPath(str)"
                for return_variable in task_def.return_vars
            ]

            # this task will output variables to a file that other tasks can access
            # through KFP's task.outputs attribute
            for return_variable in task_def.return_vars:
                task_loading_blocks[
                    return_variable
                ] = f'task_{task_name}.outputs["variable_{return_variable}"]'

            task_def_rendered = TASK_FUNCTION_TEMPLATE.render(
                MODULE_NAME=self.pipeline_name + "_module",
                function_name=task_name,
                user_input_variables=", ".join(
                    input_vars + input_paths + output_paths
                ),
                typing_blocks=task_def.typing_blocks,
                loading_blocks=loading_blocks,
                pre_call_block="",
                call_block=task_def.call_block,
                post_call_block=task_def.post_call_block,
                dumping_blocks=dumping_blocks,
                include_imports_locally=True,
                return_block="",
            )
            rendered_task_defs.append(task_def_rendered)

        return rendered_task_defs, task_loading_blocks

get_rendered_task_definitions(task_defs)

Returns rendered tasks for the pipeline tasks along with a dictionary to lookup previous task outputs.

The returned dictionary is used by the DAG to connect the right input files to output files for inter task communication.

Source code in lineapy/plugins/kubeflow_pipeline_writer.py
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
def get_rendered_task_definitions(
    self,
    task_defs: Dict[str, TaskDefinition],
) -> Tuple[List[str], Dict[str, str]]:
    """
    Returns rendered tasks for the pipeline tasks along with a dictionary to lookup
    previous task outputs.

    The returned dictionary is used by the DAG to connect the right input files to
    output files for inter task communication.
    """
    TASK_FUNCTION_TEMPLATE = load_plugin_template(
        "task/task_function.jinja"
    )
    rendered_task_defs: List[str] = []
    task_loading_blocks: Dict[str, str] = {}

    for task_name, task_def in task_defs.items():
        loading_blocks, dumping_blocks = render_task_io_serialize_blocks(
            task_def, TaskSerializer.ParametrizedPickle
        )

        input_vars = task_def.user_input_variables

        input_paths = [
            f"variable_{loaded_input_variable}_path: kfp.components.InputPath(str)"
            for loaded_input_variable in task_def.loaded_input_variables
        ]

        output_paths = [
            f"variable_{return_variable}_path: kfp.components.OutputPath(str)"
            for return_variable in task_def.return_vars
        ]

        # this task will output variables to a file that other tasks can access
        # through KFP's task.outputs attribute
        for return_variable in task_def.return_vars:
            task_loading_blocks[
                return_variable
            ] = f'task_{task_name}.outputs["variable_{return_variable}"]'

        task_def_rendered = TASK_FUNCTION_TEMPLATE.render(
            MODULE_NAME=self.pipeline_name + "_module",
            function_name=task_name,
            user_input_variables=", ".join(
                input_vars + input_paths + output_paths
            ),
            typing_blocks=task_def.typing_blocks,
            loading_blocks=loading_blocks,
            pre_call_block="",
            call_block=task_def.call_block,
            post_call_block=task_def.post_call_block,
            dumping_blocks=dumping_blocks,
            include_imports_locally=True,
            return_block="",
        )
        rendered_task_defs.append(task_def_rendered)

    return rendered_task_defs, task_loading_blocks

Was this helpful?

Help us improve docs with your feedback!