@dataclass(kw_only=True)
class Workflow(
ArgumentsMixin,
ContextMixin,
HookMixin,
VolumeMixin,
MetricsMixin,
ModelMapperMixin,
):
"""The base Workflow class for Hera.
Workflow implements the contextmanager interface so allows usage of `with`, under which
any `hera.workflows.protocol.Templatable` object instantiated under the context will be
added to the Workflow's list of templates.
Workflows can be created directly on your Argo cluster via `create`. They can also be dumped
to yaml via `to_yaml` or built according to the Argo schema via `build` to get an OpenAPI model
object.
"""
def _build_volume_claim_templates(self) -> Optional[List]:
return ((self.volume_claim_templates or []) + (self._build_persistent_volume_claims() or [])) or None
def _build_on_exit(self) -> Optional[str]:
if isinstance(self.on_exit, Templatable):
return self.on_exit._build_template().name # type: ignore
return self.on_exit
def _build_retry_strategy(self) -> Optional[ModelRetryStrategy]:
if self.retry_strategy is None:
return None
if isinstance(self.retry_strategy, RetryStrategy):
return self.retry_strategy.build()
return self.retry_strategy
def _build_templates(self) -> Optional[List[_ModelTemplate]]:
"""Builds the templates into an Argo schema."""
templates: List[_ModelTemplate] = []
for template in self.templates:
if isinstance(template, HookMixin):
template = template._dispatch_hooks()
if isinstance(template, Templatable):
templates.append(template._build_template())
elif isinstance(template, _ModelTemplate):
templates.append(template)
else:
raise InvalidType(f"{type(template)} is not a valid template type")
if isinstance(template, VolumeClaimable):
claims = template._build_persistent_volume_claims()
# If there are no claims, continue, nothing to add
if not claims:
continue
# If there are no volume claim templates, set them to the constructed claims
elif self.volume_claim_templates is None:
self.volume_claim_templates = claims
else:
# otherwise, we need to merge the two lists of volume claim templates. This prioritizes the
# already existing volume claim templates under the assumption that the user has already set
# a claim template on the workflow intentionally, or the user is sharing the same volumes across
# different templates
current_volume_claims_map = {}
for claim in self.volume_claim_templates:
assert claim.metadata is not None, "expected a workflow volume claim with metadata"
assert claim.metadata.name is not None, "expected a named workflow volume claim"
current_volume_claims_map[claim.metadata.name] = claim
new_volume_claims_map = {}
for claim in claims:
assert claim.metadata is not None, "expected a volume claim with metadata"
assert claim.metadata.name is not None, "expected a named volume claim"
new_volume_claims_map[claim.metadata.name] = claim
for claim_name, claim in new_volume_claims_map.items():
if claim_name not in current_volume_claims_map:
self.volume_claim_templates.append(claim)
return templates or None
# Workflow fields - https://argoproj.github.io/argo-workflows/fields/#workflow
api_version: Annotated[Optional[str], _WorkflowModelMapper("api_version")] = None
kind: Annotated[Optional[str], _WorkflowModelMapper("kind")] = None
status: Annotated[Optional[_ModelWorkflowStatus], _WorkflowModelMapper("status")] = None
# ObjectMeta fields - https://argoproj.github.io/argo-workflows/fields/#objectmeta
annotations: Annotated[Optional[Dict[str, str]], _WorkflowModelMapper("metadata.annotations")] = None
creation_timestamp: Annotated[Optional[Time], _WorkflowModelMapper("metadata.creation_timestamp")] = None
deletion_grace_period_seconds: Annotated[
Optional[int], _WorkflowModelMapper("metadata.deletion_grace_period_seconds")
] = None
deletion_timestamp: Annotated[Optional[Time], _WorkflowModelMapper("metadata.deletion_timestamp")] = None
finalizers: Annotated[Optional[List[str]], _WorkflowModelMapper("metadata.finalizers")] = None
generate_name: Annotated[Optional[str], _WorkflowModelMapper("metadata.generate_name")] = None
generation: Annotated[Optional[int], _WorkflowModelMapper("metadata.generation")] = None
labels: Annotated[Optional[Dict[str, str]], _WorkflowModelMapper("metadata.labels")] = None
managed_fields: Annotated[Optional[List[ManagedFieldsEntry]], _WorkflowModelMapper("metadata.managed_fields")] = (
None
)
name: Annotated[Optional[str], _WorkflowModelMapper("metadata.name")] = None
namespace: Annotated[Optional[str], _WorkflowModelMapper("metadata.namespace")] = None
owner_references: Annotated[Optional[List[OwnerReference]], _WorkflowModelMapper("metadata.owner_references")] = (
None
)
resource_version: Annotated[Optional[str], _WorkflowModelMapper("metadata.resource_version")] = None
self_link: Annotated[Optional[str], _WorkflowModelMapper("metadata.self_link")] = None
uid: Annotated[Optional[str], _WorkflowModelMapper("metadata.uid")] = None
# WorkflowSpec fields - https://argoproj.github.io/argo-workflows/fields/#workflowspec
active_deadline_seconds: Annotated[Optional[int], _WorkflowModelMapper("spec.active_deadline_seconds")] = None
affinity: Annotated[Optional[Affinity], _WorkflowModelMapper("spec.affinity")] = None
archive_logs: Annotated[Optional[bool], _WorkflowModelMapper("spec.archive_logs")] = None
artifact_gc: Annotated[
Optional[ArtifactGC | WorkflowLevelArtifactGC], _WorkflowModelMapper("spec.artifact_gc")
] = None
artifact_repository_ref: Annotated[
Optional[ArtifactRepositoryRef], _WorkflowModelMapper("spec.artifact_repository_ref")
] = None
automount_service_account_token: Annotated[
Optional[bool], _WorkflowModelMapper("spec.automount_service_account_token")
] = None
dns_config: Annotated[Optional[PodDNSConfig], _WorkflowModelMapper("spec.dns_config")] = None
dns_policy: Annotated[Optional[str], _WorkflowModelMapper("spec.dns_policy")] = None
entrypoint: Annotated[Optional[str], _WorkflowModelMapper("spec.entrypoint")] = None
executor: Annotated[Optional[ExecutorConfig], _WorkflowModelMapper("spec.executor")] = None
hooks: Annotated[Optional[Dict[str, LifecycleHook]], _WorkflowModelMapper("spec.hooks")] = None
host_aliases: Annotated[Optional[List[HostAlias]], _WorkflowModelMapper("spec.host_aliases")] = None
host_network: Annotated[Optional[bool], _WorkflowModelMapper("spec.host_network")] = None
image_pull_secrets: Annotated[ImagePullSecretsT, _WorkflowModelMapper("spec.image_pull_secrets")] = None
node_selector: Annotated[Optional[Dict[str, str]], _WorkflowModelMapper("spec.node_selector")] = None
on_exit: Annotated[Optional[Union[str, Templatable]], _WorkflowModelMapper("spec.on_exit", _build_on_exit)] = None
parallelism: Annotated[Optional[int], _WorkflowModelMapper("spec.parallelism")] = None
pod_disruption_budget: Annotated[
Optional[PodDisruptionBudgetSpec], _WorkflowModelMapper("spec.pod_disruption_budget")
] = None
pod_gc: Annotated[Optional[PodGC], _WorkflowModelMapper("spec.pod_gc")] = None
pod_metadata: Annotated[Optional[Metadata], _WorkflowModelMapper("spec.pod_metadata")] = None
pod_priority_class_name: Annotated[Optional[str], _WorkflowModelMapper("spec.pod_priority_class_name")] = None
pod_spec_patch: Annotated[Optional[str], _WorkflowModelMapper("spec.pod_spec_patch")] = None
priority: Annotated[Optional[int], _WorkflowModelMapper("spec.priority")] = None
retry_strategy: Annotated[
Optional[Union[RetryStrategy, ModelRetryStrategy]],
_WorkflowModelMapper("spec.retry_strategy", _build_retry_strategy),
] = None
scheduler_name: Annotated[Optional[str], _WorkflowModelMapper("spec.scheduler_name")] = None
security_context: Annotated[Optional[PodSecurityContext], _WorkflowModelMapper("spec.security_context")] = None
service_account_name: Annotated[Optional[str], _WorkflowModelMapper("spec.service_account_name")] = None
shutdown: Annotated[Optional[str], _WorkflowModelMapper("spec.shutdown")] = None
suspend: Annotated[Optional[bool], _WorkflowModelMapper("spec.suspend")] = None
synchronization: Annotated[Optional[Synchronization], _WorkflowModelMapper("spec.synchronization")] = None
template_defaults: Annotated[Optional[_ModelTemplate], _WorkflowModelMapper("spec.template_defaults")] = None
templates: Annotated[
List[Union[_ModelTemplate, Templatable]], _WorkflowModelMapper("spec.templates", _build_templates)
] = field(default_factory=list)
tolerations: Annotated[Optional[List[Toleration]], _WorkflowModelMapper("spec.tolerations")] = None
ttl_strategy: Annotated[Optional[TTLStrategy], _WorkflowModelMapper("spec.ttl_strategy")] = None
volume_claim_gc: Annotated[Optional[VolumeClaimGC], _WorkflowModelMapper("spec.volume_claim_gc")] = None
volume_claim_templates: Annotated[
Optional[List[PersistentVolumeClaim]],
_WorkflowModelMapper("spec.volume_claim_templates", _build_volume_claim_templates),
] = None
workflow_metadata: Annotated[Optional[WorkflowMetadata], _WorkflowModelMapper("spec.workflow_metadata")] = None
workflow_template_ref: Annotated[
Optional[WorkflowTemplateRef], _WorkflowModelMapper("spec.workflow_template_ref")
] = None
# Override types for mixin fields
arguments: Annotated[
ArgumentsT,
_WorkflowModelMapper("spec.arguments", ArgumentsMixin._build_arguments),
] = None
metrics: Annotated[
MetricsT,
_WorkflowModelMapper("spec.metrics", MetricsMixin._build_metrics),
] = None
volumes: Annotated[VolumesT, _WorkflowModelMapper("spec.volumes", VolumeMixin._build_volumes)] = None
# Hera-specific fields
workflows_service: Optional[Union[WorkflowsService, AsyncWorkflowsService]] = None
def __post_init__(self):
"""Set hooks via __post_init__ and perform validation."""
super().__post_init__()
if self.name is not None and len(self.name) > NAME_LIMIT:
raise ValueError(f"name must be no more than {NAME_LIMIT} characters: {self.name}")
if self.generate_name is not None and len(self.generate_name) > NAME_LIMIT:
raise ValueError(f"generate_name must be no more than {NAME_LIMIT} characters: {self.generate_name}")
if self.api_version is None:
self.api_version = global_config.api_version
if self.kind is None:
self.kind = self.__class__.__name__
if self.workflows_service is None:
self.workflows_service = WorkflowsService()
if self.namespace is None:
self.namespace = global_config.namespace
if self.service_account_name is None:
self.service_account_name = global_config.service_account_name
if self.image_pull_secrets is not None:
self.image_pull_secrets = self._validate_image_pull_secrets(self.image_pull_secrets)
@staticmethod
def _validate_image_pull_secrets(image_pull_secrets: ImagePullSecretsT) -> ImagePullSecretsT:
if isinstance(image_pull_secrets, str):
image_pull_secrets = [LocalObjectReference(name=image_pull_secrets)]
elif isinstance(image_pull_secrets, LocalObjectReference):
image_pull_secrets = [image_pull_secrets]
assert isinstance(image_pull_secrets, list), (
"`image_pull_secrets` expected to be either a `str`, a `LocalObjectReferences`, a list of `str`, "
"or a list of `LocalObjectReferences`"
)
result = []
for secret in image_pull_secrets:
if isinstance(secret, str):
result.append(LocalObjectReference(name=secret))
elif isinstance(secret, LocalObjectReference):
result.append(secret)
return result
def get_parameter(self, name: str) -> Parameter:
"""Attempts to find and return a `Parameter` of the specified name."""
arguments = self._build_arguments()
if arguments is None:
raise KeyError("Workflow has no arguments set")
if arguments.parameters is None:
raise KeyError("Workflow has no argument parameters set")
parameters = arguments.parameters
if next((p for p in parameters if p.name == name), None) is None:
raise KeyError(f"`{name}` is not a valid workflow parameter")
return Parameter(name=name, value=f"{{{{workflow.parameters.{name}}}}}")
def build(self) -> TWorkflow:
"""Builds the Workflow and its components into an Argo schema Workflow object."""
self = self._dispatch_hooks()
model_workflow = _ModelWorkflow(
metadata=ObjectMeta(),
spec=_ModelWorkflowSpec(),
)
return _WorkflowModelMapper.build_model(Workflow, self, model_workflow)
def to_dict(self) -> Any:
"""Builds the Workflow as an Argo schema Workflow object and returns it as a dictionary."""
return self.build().model_dump(exclude_none=True, by_alias=True)
def __eq__(self, other) -> bool:
"""Verifies equality of `self` with the specified `other`."""
if other.__class__ is self.__class__:
return self.to_dict() == other.to_dict()
return False
def to_yaml(self, *args, **kwargs) -> str:
"""Builds the Workflow as an Argo schema Workflow object and returns it as yaml string."""
def human_readable_ordering(kv: tuple) -> int:
"""Key ordering function for ordering in a more human-readable fashion.
Ordering is:
1. "name" keys always first (if present)
2. Primitives (not dicts/lists)
3. lists
4. dict
"""
k, v = kv
if k == "name" and isinstance(v, str):
return 0
if not isinstance(v, (dict, list)):
return 1
if isinstance(v, list):
return 2
return 3
def order_dict(d: dict) -> dict[str, Any]:
"""Recursively orders `d` by the custom_ordering function by inserting them into a copy of the dict in order."""
d_copy: dict[str, Any] = dict()
for k, v in sorted(d.items(), key=lambda x: (human_readable_ordering(x), x[0])):
if isinstance(v, dict):
d_copy[k] = order_dict(v)
elif isinstance(v, list):
if v and isinstance(v[0], dict):
d_copy[k] = [order_dict(i) if isinstance(i, dict) else i for i in v]
elif v and isinstance(v[0], list):
d_copy[k] = [[order_dict(i) for i in sublist] for sublist in v]
else:
d_copy[k] = v
else:
d_copy[k] = v
return d_copy
human_ordered_dict = order_dict(self.to_dict())
return _yaml.dump(human_ordered_dict, *args, **kwargs)
def create(self, wait: bool = False, poll_interval: int = 5) -> TWorkflow:
"""Creates the Workflow on the Argo cluster.
Parameters
----------
wait: bool = False
If false then the workflow is created and the function returns immediately after the server
creates the Workflow.
If true then the workflow is created and the function blocks until the workflow is done executing.
poll_interval: int = 5
The interval in seconds to poll the workflow status if wait is true. Ignored when wait is false.
"""
assert isinstance(self.workflows_service, WorkflowsService), "workflows service not initialized"
assert self.namespace, "workflow namespace not defined"
wf = self.workflows_service.create_workflow(
WorkflowCreateRequest(workflow=self.build()), # type: ignore
namespace=self.namespace,
)
# set the workflow name to the name returned by the API, which helps cover the case of users relying on
# `generate_name=True`
self.name = wf.metadata.name
if wait:
return self.wait(poll_interval=poll_interval)
return wf
def wait(self, poll_interval: int = 5) -> TWorkflow:
"""Waits for the Workflow to complete execution.
Parameters
----------
poll_interval: int = 5
The interval in seconds to poll the workflow status.
"""
assert isinstance(self.workflows_service, WorkflowsService), "workflows service not initialized"
assert self.namespace is not None, "workflow namespace not defined"
assert self.name is not None, "workflow name not defined"
# here we use the sleep interval to wait for the workflow post creation. This is to address a potential
# race conditions such as:
# 1. Argo server says "workflow was accepted" but the workflow is not yet created
# 2. Hera wants to verify the status of the workflow, but it's not yet defined because it's not created
# 3. Argo finally creates the workflow
# 4. Hera throws an `AssertionError` because the phase assertion fails
time.sleep(poll_interval)
wf = self.workflows_service.get_workflow(self.name, namespace=self.namespace)
assert wf.metadata.name is not None, f"workflow name not defined for workflow {self.name}"
assert wf.status is not None, f"workflow status not defined for workflow {wf.metadata.name}"
assert wf.status.phase is not None, f"workflow phase not defined for workflow status {wf.status}"
status = WorkflowStatus.from_argo_status(wf.status.phase)
# keep polling for workflow status until completed, at the interval dictated by the user
while status == WorkflowStatus.running:
time.sleep(poll_interval)
wf = self.workflows_service.get_workflow(wf.metadata.name, namespace=self.namespace)
assert wf.metadata.name is not None
assert wf.status is not None, f"workflow status not defined for workflow {wf.metadata.name}"
assert wf.status.phase is not None, f"workflow phase not defined for workflow status {wf.status}"
status = WorkflowStatus.from_argo_status(wf.status.phase)
return wf
def lint(self) -> TWorkflow:
"""Lints the Workflow using the Argo cluster."""
assert isinstance(self.workflows_service, WorkflowsService), "workflows service not initialized"
assert self.namespace, "workflow namespace not defined"
return self.workflows_service.lint_workflow(
WorkflowLintRequest(workflow=self.build()), # type: ignore
namespace=self.namespace,
)
async def async_create(self, wait: bool = True, poll_interval: int = 5) -> TWorkflow:
"""Creates the Workflow on the Argo cluster. Note that `wait` is `True` by default as this is an async function.
Parameters
----------
wait: bool = True
If false then the workflow is created and the function returns immediately.
If true then the workflow is created and the function blocks until the workflow is done executing.
poll_interval: int = 5
The interval in seconds to poll the workflow status if wait is true. Ignored when wait is false.
"""
assert isinstance(self.workflows_service, AsyncWorkflowsService), "workflows service not initialized"
assert self.namespace, "workflow namespace not defined"
wf = await self.workflows_service.create_workflow(
WorkflowCreateRequest(workflow=self.build()), # type: ignore
namespace=self.namespace,
)
# set the workflow name to the name returned by the API, which helps cover the case of users relying on
# `generate_name=True`
self.name = wf.metadata.name
if wait:
return await self.async_wait(poll_interval=poll_interval)
return wf
async def async_wait(self, poll_interval: int = 5) -> TWorkflow:
"""Waits for the Workflow to complete execution.
Parameters
----------
poll_interval: int = 5
The interval in seconds to poll the workflow status.
"""
assert isinstance(self.workflows_service, AsyncWorkflowsService), "workflows service not initialized"
assert self.namespace is not None, "workflow namespace not defined"
assert self.name is not None, "workflow name not defined"
# here we use the sleep interval to wait for the workflow post creation. This is to address a potential
# race conditions such as:
# 1. Argo server says "workflow was accepted" but the workflow is not yet created
# 2. Hera wants to verify the status of the workflow, but it's not yet defined because it's not created
# 3. Argo finally creates the workflow
# 4. Hera throws an `AssertionError` because the phase assertion fails
await asyncio.sleep(poll_interval)
wf = await self.workflows_service.get_workflow(self.name, namespace=self.namespace)
assert wf.metadata.name is not None, f"workflow name not defined for workflow {self.name}"
assert wf.status is not None, f"workflow status not defined for workflow {wf.metadata.name}"
assert wf.status.phase is not None, f"workflow phase not defined for workflow status {wf.status}"
status = WorkflowStatus.from_argo_status(wf.status.phase)
# keep polling for workflow status until completed, at the interval dictated by the user
while status == WorkflowStatus.running:
await asyncio.sleep(poll_interval)
wf = await self.workflows_service.get_workflow(wf.metadata.name, namespace=self.namespace)
assert wf.metadata.name is not None
assert wf.status is not None, f"workflow status not defined for workflow {wf.metadata.name}"
assert wf.status.phase is not None, f"workflow phase not defined for workflow status {wf.status}"
status = WorkflowStatus.from_argo_status(wf.status.phase)
return wf
async def async_lint(self) -> TWorkflow:
"""Lints the Workflow using the Argo cluster."""
assert isinstance(self.workflows_service, AsyncWorkflowsService), "workflows service not initialized"
assert self.namespace, "workflow namespace not defined"
return await self.workflows_service.lint_workflow(
WorkflowLintRequest(workflow=self.build()), # type: ignore
namespace=self.namespace,
)
def _add_sub(self, node: Any):
"""Adds the given node (expected to satisfy the `Templatable` protocol) to the context."""
if not isinstance(node, (Templatable, _ModelTemplate)):
raise InvalidType(type(node))
self.templates.append(node)
def to_file(self, output_directory: Union[Path, str] = ".", name: str = "", *args, **kwargs) -> Path:
"""Writes the Workflow as an Argo schema Workflow object to a YAML file and returns the path to the file.
Args:
output_directory: The directory to write the file to. Defaults to the current working directory.
name: The name of the file to write without the file extension. Defaults to the Workflow's name or a
generated name.
*args: Additional arguments to pass to `yaml.dump`.
**kwargs: Additional keyword arguments to pass to `yaml.dump`.
"""
workflow_name = self.name or (self.generate_name or "workflow").rstrip("-")
name = name or workflow_name
output_directory = Path(output_directory)
output_path = Path(output_directory) / f"{name}.yaml"
output_directory.mkdir(parents=True, exist_ok=True)
output_path.write_text(self.to_yaml(*args, **kwargs))
return output_path.absolute()
@classmethod
def from_dict(cls, model_dict: Dict) -> ModelMapperMixin:
"""Create a Workflow from a Workflow contained in a dict.
Examples:
>>> my_workflow = Workflow(name="my-workflow")
>>> my_workflow == Workflow.from_dict(my_workflow.to_dict())
True
"""
return cls._from_dict(model_dict, _ModelWorkflow)
@classmethod
def from_yaml(cls, yaml_str: str) -> ModelMapperMixin:
"""Create a Workflow from a Workflow contained in a YAML string.
Examples:
>>> my_workflow = Workflow.from_yaml(yaml_str)
"""
return cls._from_yaml(yaml_str, _ModelWorkflow)
@classmethod
def from_file(cls, yaml_file: Union[Path, str]) -> ModelMapperMixin:
"""Create a Workflow from a Workflow contained in a YAML file.
Examples:
>>> yaml_file = Path(...)
>>> my_workflow = Workflow.from_file(yaml_file)
"""
return cls._from_file(yaml_file, _ModelWorkflow)
def get_workflow_link(self) -> str:
"""Returns the workflow link for the workflow."""
assert self.workflows_service is not None, "Cannot fetch a workflow link without a service"
assert self.name is not None, "Cannot fetch a workflow link without a workflow name"
return self.workflows_service.get_workflow_link(self.name)