Skip to content

Upgrading from Hera v6 to v7

Hera v7 targets Argo Workflows v4 instead of v3.7. Most code keeps working unchanged because the models and services are regenerated automatically. Where Argo Workflows v4 dropped a field that Hera’s hand-written wrapper still accepts, Hera v7 emits a DeprecationWarning and rewrites the call to the v4 equivalent.

The shims will be removed in Hera v8. Migrate when convenient.

CronWorkflow.schedule -> schedules

Argo Workflows v4 removed CronWorkflowSpec.schedule in favour of the already-supported schedules list (introduced in v3.6).

# v6 / Argo v3 (still works, emits DeprecationWarning)
from hera.workflows import CronWorkflow

with CronWorkflow(
    name="daily-cleanup",
    schedule="0 3 * * *",
) as cw:
    ...

# v7 / Argo v4
from hera.workflows import CronWorkflow

with CronWorkflow(
    name="daily-cleanup",
    schedules=["0 3 * * *"],
) as cw:
    ...

If both schedule and schedules are set the legacy value is appended to the list (schedule="X" plus schedules=["Y"] produces schedules=["Y", "X"]), so callers that switched to schedules while leaving a stale schedule argument in place still get a single, deterministic schedule list on the wire.

WorkflowSpec.mutex / semaphore -> synchronization.mutexes / semaphores

Argo v4 removed the singular synchronization.mutex and synchronization.semaphore fields, which were already deprecated in v3.6, in favour of synchronization.mutexes and synchronization.semaphores (both lists).

Hera never exposed mutex / semaphore on its hand-written Workflow or Template wrappers, so most users are unaffected. If you previously reached into Synchronization directly through the generated models you will need to move from the singular to the plural form:

# v6 / Argo v3
from hera.workflows import Workflow
from hera.workflows.models import Mutex, Synchronization

with Workflow(
    generate_name="locked-",
    synchronization=Synchronization(mutex=Mutex(name="db")),
) as w:
    ...

# v7 / Argo v4
from hera.workflows import Workflow
from hera.workflows.models import Mutex, Synchronization

with Workflow(
    generate_name="locked-",
    synchronization=Synchronization(mutexes=[Mutex(name="db")]),
) as w:
    ...

The same applies anywhere Synchronization appears in a Hera workflow: template-level synchronization, step-level synchronization, etc.

WorkflowSpec.podPriority -> priorityClassName

Argo v4 removed WorkflowSpec.podPriority (an integer that was applied to each Pod). The replacement is WorkflowSpec.priorityClassName, which references a Kubernetes PriorityClass.

Hera dropped pod_priority from its hand-written wrapper surface in v6, so most callers do not need to change anything. If you were reaching through the generated model directly, switch to a named priority class:

# v6 / Argo v3
from hera.workflows.models import WorkflowSpec

spec = WorkflowSpec(pod_priority=100, ...)

# v7 / Argo v4
from hera.workflows.models import WorkflowSpec

spec = WorkflowSpec(priority_class_name="high", ...)

New: PluginArtifact

Argo v4 introduces a pluggable artifact driver. Hera v7 ships a sugar wrapper in hera.workflows.artifact matching the existing artifact backend wrappers:

from hera.workflows import PluginArtifact, Workflow, script


@script()
def consumer(message: str) -> None:
    print(message)


with Workflow(generate_name="plugin-artifacts-", entrypoint="consumer") as w:
    consumer(
        arguments={
            "message": PluginArtifact(
                name="payload",
                plugin_name="my-driver",
                key="some/object/path",
                configuration='{"timeout": 10}',
            ),
        },
    )

The driver plugin name is exposed as plugin_name (not name) to avoid shadowing the artifact identifier inherited from Artifact.name.

See Argo’s artifact plugin documentation for the full list of available drivers and how to author your own.

Comments