def start_non_interactive_pipeline_run(
self,
job_uuid,
project_uuid,
pipeline_definition: PipelineDefinition,
run_config: Dict[str, Union[str, Dict[str, str]]],
) -> str:
"""Starts a non-interactive pipeline run.
It is a pipeline run that is part of a job.
Args:
job_uuid: UUID of the job.
project_uuid: UUID of the project.
pipeline_definition: A json description of the pipeline.
run_config: Configuration of the run for the compute backend.
Example: {
'host_user_dir': '/home/../userdir',
'project_dir': '/home/../pipelines/uuid',
'env_uuid_docker_id_mappings': {
'b6527b0b-bfcc-4aff-91d1-37f9dfd5d8e8':
'sha256:61f82126945bb25dd85d6a5b122a1815df1c0c5f91621089cde0938be4f698d4'
}
}
Returns:
Status of the pipeline run. "FAILURE" or "SUCCESS".
"""
pipeline_uuid = pipeline_definition["uuid"]
job_dir = os.path.join("/userdir", "jobs", project_uuid, pipeline_uuid, job_uuid)
snapshot_dir = os.path.join(job_dir, "snapshot")
run_dir = os.path.join(job_dir, self.request.id)
# TODO: It should not copy all directories, e.g. not "data".
# Copy the contents of `snapshot_dir` to the new (not yet existing
# folder) `run_dir` (that will then be created by `copytree`).
# copytree(snapshot_dir, run_dir)
os.system('cp -R "%s" "%s"' % (snapshot_dir, run_dir))
# Update the `run_config` for the interactive pipeline run. The
# pipeline run should execute on the `run_dir` as its
# `project_dir`. Note that the `project_dir` inside the
# `run_config` has to be the abs path w.r.t. the host because it is
# used by the `docker.sock` when mounting the dir to the container
# of a step.
host_base_user_dir = os.path.split(run_config["host_user_dir"])[0]
# To join the paths, the `run_dir` cannot start with `/userdir/...`
# but should start as `userdir/...`
run_config["project_dir"] = os.path.join(host_base_user_dir, run_dir[1:])
run_config["run_endpoint"] = f"jobs/{job_uuid}"
run_config["pipeline_uuid"] = pipeline_uuid
run_config["project_uuid"] = project_uuid
# Overwrite the `pipeline.json`, that was copied from the snapshot,
# with the new `pipeline.json` that contains the new parameters for
# every step.
pipeline_json = os.path.join(run_dir, run_config["pipeline_path"])
with open(pipeline_json, "w") as f:
json.dump(pipeline_definition, f, indent=4, sort_keys=True)
with launch_noninteractive_session(
docker_client,
self.request.id,
project_uuid,
run_config["pipeline_path"],
run_config["project_dir"],
):
status = run_pipeline(
pipeline_definition, project_uuid, run_config, task_id=self.request.id
)
return status
def start_non_interactive_pipeline_run(
self,
job_uuid,
project_uuid,
pipeline_definition: PipelineDefinition,
run_config: Dict[str, Union[str, Dict[str, str]]],
) -> str:
"""Starts a non-interactive pipeline run.
It is a pipeline run that is part of a job.
Args:
job_uuid: UUID of the job.
project_uuid: UUID of the project.
pipeline_definition: A json description of the pipeline.
run_config: Configuration of the run for the compute backend.
Example: {
'userdir_pvc': 'userdir-pvc',
'project_dir': 'pipelines/uuid',
'env_uuid_to_image': {
'b6527b0b-bfcc-4aff-91d1-37f9dfd5d8e8':
'sha256:61f82126945bb25dd85d6a5b122a1815df1c0c5f91621089cde0938be4f698d4'
}
}
Returns:
Status of the pipeline run. "FAILURE" or "SUCCESS".
"""
pipeline_uuid = pipeline_definition["uuid"]
job_dir = os.path.join("/userdir", "jobs", project_uuid, pipeline_uuid, job_uuid)
snapshot_dir = os.path.join(job_dir, "snapshot")
run_dir = os.path.join(job_dir, self.request.id)
# Copy the contents of `snapshot_dir` to the new (not yet existing
# folder) `run_dir`. No need to use_gitignore since the snapshot
# was copied with use_gitignore=True.
copytree(snapshot_dir, run_dir, use_gitignore=False)
# Update the `run_config` for the interactive pipeline run. The
# pipeline run should execute on the `run_dir` as its
# `project_dir`. Note that the `project_dir` inside the
# `run_config` has to be relative to userdir_pvc as it is used
# by k8s as a subpath of userdir_pvc
userdir_pvc = run_config["userdir_pvc"]
# For non interactive runs the session uuid is equal to the task
# uuid, which is actually the pipeline run uuid.
session_uuid = self.request.id
run_config["session_uuid"] = session_uuid
run_config["session_type"] = "noninteractive"
run_config["pipeline_uuid"] = pipeline_uuid
run_config["project_uuid"] = project_uuid
run_config["project_dir"] = run_dir
run_config["run_endpoint"] = f"jobs/{job_uuid}"
# Overwrite the `pipeline.json`, that was copied from the snapshot,
# with the new `pipeline.json` that contains the new parameters for
# every step.
pipeline_json = os.path.join(run_dir, run_config["pipeline_path"])
with open(pipeline_json, "w") as f:
json.dump(pipeline_definition, f, indent=4, sort_keys=True)
# Note that run_config contains user_env_variables, which is of
# interest for the session_config.
session_config = copy.deepcopy(run_config)
session_config.pop("env_uuid_to_image")
session_config.pop("run_endpoint")
session_config["userdir_pvc"] = userdir_pvc
session_config["services"] = pipeline_definition.get("services", {})
session_config["env_uuid_to_image"] = run_config["env_uuid_to_image"]
with launch_noninteractive_session(
session_uuid,
session_config,
lambda: AbortableAsyncResult(session_uuid).is_aborted(),
):
status = run_pipeline(
pipeline_definition,
run_config,
session_uuid,
task_id=self.request.id,
)
return status