# Copyright © The Debusine Developers
# See the AUTHORS file at the top-level directory of this distribution
#
# This file is part of Debusine. It is subject to the license terms
# in the LICENSE file found in the top-level directory of this
# distribution. No part of Debusine, including this file, may be copied,
# modified, propagated, or distributed except according to the terms
# contained in the LICENSE file.
"""Data models for db work requests."""
import datetime as dt
import enum
import logging
import re
from collections import defaultdict
from collections.abc import Generator, Sequence
from contextlib import contextmanager
from enum import Enum, auto
from itertools import chain
from typing import (
Any,
Final,
Literal,
Optional,
Self,
TYPE_CHECKING,
TypeAlias,
assert_never,
cast,
overload,
)
import pydantic
from django.contrib.postgres.fields import ArrayField
from django.core.exceptions import (
FieldError,
ImproperlyConfigured,
ValidationError,
)
from django.core.serializers.json import DjangoJSONEncoder
from django.db import connection, models, transaction
from django.db.models import (
CheckConstraint,
DateTimeField,
Exists,
ExpressionWrapper,
F,
IntegerField,
JSONField,
Max,
OuterRef,
Q,
QuerySet,
UniqueConstraint,
Value,
)
from django.db.models.functions import Coalesce, Greatest
from django.urls import reverse
from django.utils import timezone
from django_cte import CTEQuerySet, With
from prometheus_client import Histogram
from debusine.artifacts.models import (
BareDataCategory,
CollectionCategory,
DebusineHistoricalTaskRun,
TaskTypes,
WorkRequestResults,
)
from debusine.client.models import (
LookupChildType,
RuntimeParameter,
RuntimeParameters,
)
from debusine.db.context import context
from debusine.db.metrics import emit_metric
from debusine.db.models import Collection, CollectionItem, Scope, permissions
from debusine.db.models.permissions import (
Allow,
PermissionUser,
Role,
enforce,
permission_check,
permission_filter,
)
from debusine.db.models.worker_pools import WorkerPoolTaskExecutionStatistics
from debusine.db.models.workspaces import Workspace
from debusine.server import notifications
from debusine.tasks import TaskConfigError
from debusine.tasks.models import (
ActionRecordInTaskHistory,
ActionRetryWithDelays,
ActionSkipIfLookupResultChanged,
ActionTypes,
ActionUpdateCollectionWithArtifacts,
ActionUpdateCollectionWithData,
BaseDynamicTaskData,
BaseTaskData,
EventReaction,
EventReactions,
LookupMultiple,
LookupSingle,
NotificationDataEmail,
OutputData,
)
if TYPE_CHECKING:
from django.http import HttpRequest
from django_stubs_ext.db.models import TypedModelMeta
from debusine.db.models.auth import User
from debusine.db.models.tasks import DBTask
from debusine.db.models.workers import Worker
from debusine.server.collections.lookup import (
LookupResult,
LookupResultArtifact,
LookupResultBare,
LookupResultCollection,
)
from debusine.server.workflows.models import (
WorkRequestManualUnblockAction,
WorkRequestWorkflowData,
)
from debusine.web.views.ui.work_requests import (
WorkRequestUI,
WorkflowTemplateUI,
)
else:
TypedModelMeta = object
logger = logging.getLogger(__name__)
#: Limit on how many times a work request may be retried automatically.
MAX_AUTOMATIC_RETRIES = 3
# Lowest priority to be considered "normal" for metric labels: interactive
# tasks started by users who are likely to be waiting for the results.
MIN_PRIORITY_NORMAL: Final = 0
# Lowest priority to be considered "background" for metric labels: tasks
# where substantial delays are reasonable.
MIN_PRIORITY_BACKGROUND: Final = -100
# Everything below MIN_PRIORITY_BACKGROUND is considered "indefinite" for
# metric labels: the submitter is assumed to be OK with waiting
# indefinitely.
start_latency = Histogram(
"debusine_work_request_start_latency",
"Histogram of time that work requests spent waiting to start.",
labelnames=("task_type", "task_name", "priority"),
# Work request queues can potentially be quite long, depending on how
# busy the system is. Try to provide a reasonable range of buckets that
# distinguish between "was handled promptly", "took a while but not too
# bad", and "was backed up for ages".
buckets=(
1.0,
5.0,
15.0,
60.0, # 1 minute
300.0, # 5 minutes
900.0, # 15 minutes
3600.0, # 1 hour
10800.0, # 3 hours
21600.0, # 6 hours
43200.0, # 12 hours
86400.0, # 1 day
float("inf"),
),
)
class CannotRetry(Exception):
"""Exception raised if a work request cannot be retried."""
class CannotAbort(Exception):
"""Exception raised if a work request cannot be aborted."""
class CannotUnblock(Exception):
"""Exception raised if a work request cannot be unblocked."""
class InternalTaskError(Exception):
"""Exception raised when trying to instantiate an internal task."""
class SkipWorkRequest(Exception):
"""Exception raised when a work request should be skipped."""
class StatusChangeError(Exception):
"""Exception raised when a work request status change failed."""
class WorkRequestRetryReason(Enum):
"""The reason for retrying a work request."""
#: The worker that was running this work request failed or was
#: terminated. Only retry up to the limit set by
#: :py:const:`MAX_AUTOMATIC_RETRIES`, just in case this work request is
#: killing the worker, and increment the retry count.
WORKER_FAILED = auto()
#: The work request is being automatically retried due to a
#: ``retry-with-delays`` action. Increment the retry count, but ignore
#: :py:const:`MAX_AUTOMATIC_RETRIES`; the ``delays`` option sets a limit
#: on how many times the work request will be retried.
DELAY = auto()
#: A user requested a retry. Reset the (automatic) retry count.
MANUAL = auto()
class _WorkRequestStatuses(models.TextChoices):
"""Choices for WorkRequest.status."""
PENDING = "pending", "Pending"
RUNNING = "running", "Running"
COMPLETED = "completed", "Completed"
ABORTED = "aborted", "Aborted"
BLOCKED = "blocked", "Blocked"
class _UnblockStrategy(models.TextChoices):
DEPS = "deps", "Dependencies have completed"
MANUAL = "manual", "Manually unblocked"
class WorkRequestRoleBase(permissions.RoleBase):
"""WorkRequest role implementation."""
implied_by_scope_roles: frozenset[Scope.Roles]
implied_by_workspace_roles: frozenset[Workspace.Roles]
implied_by_public: bool
implied_by_creator: bool
def _setup(self) -> None:
"""Set up implications for a newly constructed role."""
implied_by_scope_roles: set[Scope.Roles] = set()
implied_by_workspace_roles: set[Workspace.Roles] = set()
implied_by_public: bool = False
implied_by_creator: bool = False
for i in self.implied_by:
match i:
case Workspace.Roles():
implied_by_scope_roles |= i.implied_by_scope_roles
implied_by_workspace_roles |= i.implied_by_workspace_roles
implied_by_public = implied_by_public or i.implied_by_public
case Role():
# Resolve a role passed during class definition into its
# enum instance
wr = self.__class__(i.value)
implied_by_scope_roles |= wr.implied_by_scope_roles
implied_by_workspace_roles |= wr.implied_by_workspace_roles
implied_by_public = (
implied_by_public or wr.implied_by_public
)
implied_by_creator = (
implied_by_creator or wr.implied_by_creator
)
case "creator":
implied_by_creator = True
case _:
raise ImproperlyConfigured(
"WorkRequest roles do not support"
f" implications by {i!r}"
)
self.implied_by_scope_roles = frozenset(implied_by_scope_roles)
self.implied_by_workspace_roles = frozenset(implied_by_workspace_roles)
self.implied_by_public = implied_by_public
self.implied_by_creator = implied_by_creator
def q(self, user: PermissionUser) -> Q:
"""Return a Q expression to select work requests with this role."""
assert user is not None
if not user.is_authenticated:
if self.implied_by_public:
return Q(workspace__public=True)
else:
return Q(pk__in=())
workspace_q = Q(
roles__group__users=user,
roles__role__in=self.implied_by_workspace_roles,
) | Q(
scope__in=Scope.objects.filter(
roles__group__users=user,
roles__role__in=self.implied_by_scope_roles,
)
)
if self.implied_by_public:
workspace_q |= Q(public=True)
q = Q(workspace__in=Workspace.objects.filter(workspace_q))
if self.implied_by_creator:
q |= Q(created_by=user)
return q
def implies(self, role: "WorkRequestRoles") -> bool:
"""Check if this role implies the given one."""
return (
self.implied_by_scope_roles <= role.implied_by_scope_roles
and self.implied_by_workspace_roles
<= role.implied_by_workspace_roles
and self.implied_by_public <= role.implied_by_public
and self.implied_by_creator <= role.implied_by_creator
)
class WorkRequestRoles(permissions.Roles, WorkRequestRoleBase, enum.ReprEnum):
"""Available roles for a Scope."""
OWNER = Role("owner", implied_by=[Workspace.Roles.OWNER])
CREATOR = Role("creator", implied_by=[OWNER, "creator"])
CONTRIBUTOR = Role(
"contributor", implied_by=[OWNER, Workspace.Roles.CONTRIBUTOR]
)
VIEWER = Role("viewer", implied_by=[CONTRIBUTOR, Workspace.Roles.VIEWER])
WorkRequestRoles.setup()
[docs]
class WellKnownWorkflowTemplate(enum.StrEnum):
"""
Track well-known workflow template names for use by Debusine itself.
Enumeration member names should be upper-cased versions of the
corresponding workflow task names.
"""
UPDATE_SUITES = "update-suites"
class WorkflowTemplateQuerySet[A](QuerySet["WorkflowTemplate", A]):
"""Custom QuerySet for WorkflowTemplate."""
def in_current_scope(self) -> "WorkflowTemplateQuerySet[A]":
"""Filter to work requests in the current scope."""
return self.filter(workspace__scope=context.require_scope())
def in_current_workspace(self) -> "WorkflowTemplateQuerySet[A]":
"""Filter to work requests in the current workspace."""
return self.filter(workspace=context.require_workspace())
@permission_filter(workers=Allow.ALWAYS, anonymous=Allow.PASS)
def can_display(
self, user: PermissionUser
) -> "WorkflowTemplateQuerySet[A]":
"""Keep only WorkflowTemplates that can be displayed."""
return self.filter(
workspace__in=Workspace.objects.with_role(
user, Workspace.Roles.VIEWER
)
).distinct()
@permission_filter()
def can_run(self, user: PermissionUser) -> "WorkflowTemplateQuerySet[A]":
"""Keep only WorkflowTemplates that can be started."""
return self.filter(
workspace__in=Workspace.objects.with_role(
user, Workspace.Roles.CONTRIBUTOR
)
).distinct()
@permission_filter()
def can_configure(
self, user: PermissionUser
) -> "WorkflowTemplateQuerySet[A]":
"""Keep only WorkflowTemplates that can be configured."""
return self.filter(
workspace__in=Workspace.objects.with_role(
user, Workspace.Roles.OWNER
)
).distinct()
class WorkflowTemplateManager(models.Manager["WorkflowTemplate"]):
"""Manager for WorkflowTemplate model."""
def get_queryset(self) -> WorkflowTemplateQuerySet[Any]:
"""Use the custom QuerySet."""
return WorkflowTemplateQuerySet(self.model, using=self._db)
[docs]
class WorkflowTemplate(models.Model):
"""
Database model for Workflow templates.
Workflow templates contain the information needed to instantiate a
workflow, with a Workflow orchestrator and mandatory parameters.
"""
name = models.CharField(max_length=255)
workspace = models.ForeignKey("Workspace", on_delete=models.PROTECT)
task_name = models.CharField(
max_length=100, verbose_name='Name of the Workflow orchestrator class'
)
static_parameters = JSONField(
default=dict, blank=True, encoder=DjangoJSONEncoder
)
runtime_parameters = JSONField(
default=dict, blank=True, encoder=DjangoJSONEncoder
)
priority = models.IntegerField(
default=0,
help_text="Base priority for work requests created from this template",
)
objects = WorkflowTemplateManager.from_queryset(WorkflowTemplateQuerySet)()
class Meta(TypedModelMeta):
base_manager_name = "objects"
constraints = [
UniqueConstraint(
fields=["name", "workspace"],
name="%(app_label)s_%(class)s_unique_name_workspace",
),
CheckConstraint(
check=(
~Q(task_name="update_suites")
| Q(name=WellKnownWorkflowTemplate.UPDATE_SUITES)
),
name="%(app_label)s_%(class)s_update_suites",
),
]
def __str__(self) -> str:
"""Return a string representation."""
return self.name
[docs]
def get_absolute_url(self) -> str:
"""Return the canonical display URL."""
from debusine.server.scopes import urlconf_scope
with urlconf_scope(self.workspace.scope.name):
return reverse(
"workspaces:workflow_templates:detail",
kwargs={"wname": self.workspace.name, "name": self.name},
)
[docs]
def clean(self) -> None:
"""
Ensure that task_name, static & runtime parameters are valid.
:raise ValidationError: for invalid data.
"""
# Import here to prevent circular imports
from debusine.server.workflows import Workflow, WorkflowValidationError
if not isinstance(self.static_parameters, dict):
raise ValidationError(
{"static_parameters": "static parameters must be a dictionary"}
)
# Instantiate the orchestrator and use it to validate static_parameters
workflow_cls = Workflow.from_name(self.task_name)
for parameter, validator in (
("static_parameters", workflow_cls.validate_static_parameters),
("runtime_parameters", workflow_cls.validate_runtime_parameters),
):
try:
validator(getattr(self, parameter))
except (
KeyError,
ValueError,
RuntimeError,
WorkflowValidationError,
) as exc:
raise ValidationError({parameter: str(exc)})
[docs]
@permission_check(
"{user} cannot display {resource}",
workers=Allow.ALWAYS,
anonymous=Allow.PASS,
)
def can_display(self, user: PermissionUser) -> bool:
"""Check if the workflow template can be displayed."""
return self.workspace.has_role(user, Workspace.Roles.VIEWER)
[docs]
@permission_check("{user} cannot run {resource}")
def can_run(self, user: PermissionUser) -> bool:
"""Check if the workflow template can be run."""
return self.workspace.has_role(user, Workspace.Roles.CONTRIBUTOR)
[docs]
def completed_workflows(self) -> int:
"""Count completed workflows for this template in context.workspace."""
return (
self.workrequest_set.filter(workspace=context.workspace)
.filter(status=WorkRequest.Statuses.COMPLETED)
.count()
)
[docs]
def running_workflows(self) -> int:
"""Count completed workflows for this template in context.workspace."""
return (
self.workrequest_set.in_current_workspace()
.filter(status=WorkRequest.Statuses.RUNNING)
.count()
)
[docs]
def ui(self, request: "HttpRequest") -> "WorkflowTemplateUI":
"""Return a UI helper for this instance."""
from debusine.web.views.ui.work_requests import WorkflowTemplateUI
return WorkflowTemplateUI.get(request, self)
class WorkRequestQuerySet[A](
CTEQuerySet, # type: ignore[misc]
QuerySet["WorkRequest", A],
):
"""Custom QuerySet for WorkRequest."""
def in_current_scope(self) -> Self:
"""Filter to work requests in the current scope."""
return self.filter(workspace__scope=context.require_scope())
def in_current_workspace(self) -> Self:
"""Filter to work requests in the current workspace."""
return self.filter(workspace=context.require_workspace())
def with_role_annotations(self, user: PermissionUser) -> Self:
"""
Annotate the queryset with a bool for each role, with its presence.
The returned annotations are in the form ``"is_{role}"`` and are True
if the user has the given role on the annotated resource.
A further ``roles_for`` annotation marks the stringified public key of
the user used to compute annotations, or the empty string if the
annotations are for the anonymous user.
"""
queryset = self.annotate(
roles_for=Value(
str(user.pk) if user and user.is_authenticated else ""
)
)
for role in WorkRequestRoles:
queryset = queryset.annotate(
**{
f"is_{role}": Exists(
WorkRequest.objects.filter(pk=OuterRef("pk")).with_role(
user, role
)
)
}
)
return queryset
def with_role(self, user: PermissionUser, role: WorkRequestRoles) -> Self:
"""Keep only resources where the user has the given role."""
return self.filter(
pk__in=self.model.objects.filter(role.q(user)).distinct()
)
def can_be_unblocked(self) -> "WorkRequestQuerySet[WorkRequest]":
"""Filter to work requests that can be unblocked at all."""
return cast(
WorkRequestQuerySet[WorkRequest],
self.filter(
# The work request must currently be blocked.
Q(status=WorkRequest.Statuses.BLOCKED)
# If this work request is in a workflow, then that workflow
# must be running.
& (
Q(parent__isnull=True)
| Q(parent__status=WorkRequest.Statuses.RUNNING)
)
),
)
def can_be_automatically_unblocked(
self,
) -> "WorkRequestQuerySet[WorkRequest]":
"""Filter to work requests that can be automatically unblocked."""
return self.can_be_unblocked().filter(
# DEPS work requests can be unblocked if and only if all their
# dependencies have finished. We don't know how to deal with
# other unblock strategies automatically.
Q(unblock_strategy=WorkRequest.UnblockStrategy.DEPS)
& ~Exists(
WorkRequest.objects.filter(
Q(reverse_dependencies=OuterRef("id"))
& ~Q(
status__in={
WorkRequest.Statuses.COMPLETED,
WorkRequest.Statuses.ABORTED,
}
)
)
)
)
@permission_filter(workers=Allow.ALWAYS, anonymous=Allow.PASS)
def can_display(self, user: PermissionUser) -> "WorkRequestQuerySet[A]":
"""Keep only WorkRequests that can be displayed."""
return self.with_role(user, WorkRequestRoles.VIEWER)
@permission_filter(workers=Allow.ALWAYS)
def can_retry(self, user: PermissionUser) -> "WorkRequestQuerySet[A]":
"""Keep only WorkRequests that can be retried."""
return self.with_role(user, WorkRequestRoles.CONTRIBUTOR)
@permission_filter()
def can_abort(self, user: PermissionUser) -> "WorkRequestQuerySet[A]":
"""Keep only WorkRequests that can be aborted."""
return self.with_role(user, WorkRequestRoles.CREATOR)
@permission_filter()
def can_unblock(self, user: PermissionUser) -> "WorkRequestQuerySet[A]":
"""Keep only WorkRequests that can be unblocked."""
return self.with_role(user, WorkRequestRoles.OWNER)
@permission_filter()
def can_confirm(self, user: PermissionUser) -> "WorkRequestQuerySet[A]":
"""Keep only WorkRequests that can be confirmed."""
return self.with_role(user, WorkRequestRoles.CREATOR)
def pending(
self, exclude_assigned: bool = False, worker: Optional["Worker"] = None
) -> "WorkRequestQuerySet[Any]":
"""
Return a QuerySet of tasks in WorkRequest.Statuses.PENDING status.
QuerySet is ordered by descending effective priority, then by
created_at.
Filter out the assigned pending ones if exclude_assigned=True,
and include only the WorkRequest for worker.
PENDING is the default status of a task on creation.
"""
if exclude_assigned and worker is not None:
raise ValueError("Cannot exclude_assigned and filter by worker")
qs = self.filter(status=WorkRequest.Statuses.PENDING)
if exclude_assigned:
qs = qs.exclude(worker__isnull=False)
if worker is not None:
qs = qs.filter(worker=worker)
qs = qs.order_by(
(F("priority_base") + F("priority_adjustment")).desc(), "created_at"
)
return qs
def running(
self, worker: Optional["Worker"] = None
) -> "WorkRequestQuerySet[WorkRequest]":
"""Return a QuerySet of tasks in running status."""
qs = self.filter(status=WorkRequest.Statuses.RUNNING)
if worker is not None:
qs = qs.filter(worker=worker)
return qs
def completed(self) -> "WorkRequestQuerySet[WorkRequest]":
"""Return a QuerySet of tasks in completed status."""
return self.filter(status=WorkRequest.Statuses.COMPLETED)
def aborted(self) -> "WorkRequestQuerySet[WorkRequest]":
"""Return a QuerySet of tasks in aborted status."""
return self.filter(status=WorkRequest.Statuses.ABORTED)
def aborted_or_failed(self) -> "WorkRequestQuerySet[WorkRequest]":
"""Return a QuerySet of tasks that aborted or failed."""
return self.filter(
Q(status=WorkRequest.Statuses.ABORTED)
| Q(
status=WorkRequest.Statuses.COMPLETED,
result=WorkRequest.Results.FAILURE,
)
)
def expired(self, at: dt.datetime) -> "WorkRequestQuerySet[Any]":
"""
Return queryset with work requests that have expired.
:param at: datetime to check if the work requests are expired.
:return: work requests which expire before the given datetime.
"""
return (
self.annotate(
_effective_expiration_delay=Coalesce(
"expiration_delay",
"workspace__default_expiration_delay",
)
)
.exclude(_effective_expiration_delay=dt.timedelta(0))
.filter(
# https://github.com/typeddjango/django-stubs/issues/1548
created_at__lte=(
at - F("_effective_expiration_delay") # type: ignore[operator] # noqa: E501
)
)
)
def reassignable(
self, worker: "Worker"
) -> "WorkRequestQuerySet[WorkRequest]":
"""
Return a QuerySet of tasks that can be re-assigned to another worker.
Assuming the worker is dead, if the task hasn't produced any output
yet, it can safely be reassigned to a different worker.
"""
return self.filter(
status__in=(
WorkRequest.Statuses.RUNNING,
WorkRequest.Statuses.PENDING,
),
worker=worker,
artifact__id__isnull=True,
)
def workflows(self) -> "WorkRequestQuerySet[A]":
"""Keep only workflows."""
return self.filter(task_type=TaskTypes.WORKFLOW)
def needs_input(self) -> "WorkRequestQuerySet[A]":
"""Keep only work requests in NEEDS_INPUT state."""
return self.filter(
workflow_runtime_status=WorkRequest.RuntimeStatuses.NEEDS_INPUT
)
def visible_in_workflow(self) -> "WorkRequestQuerySet[A]":
"""Keep only work requests that should be shown in workflow views."""
return self.annotate(
visible=Coalesce(
F("workflow_data_json__visible"),
Value("true"),
output_field=JSONField(),
)
).filter(visible=True)
def unsuperseded(self) -> "WorkRequestQuerySet[A]":
"""Keep only work requests that have not been superseded."""
return self.filter(superseded__isnull=True)
def terminated(self) -> "WorkRequestQuerySet[A]":
"""
Keep only work requests that have terminated in some way.
Work requests that are completed or aborted will not change any more
(except perhaps for being retried, which will create a new work
request).
"""
return self.filter(
status__in={
WorkRequest.Statuses.COMPLETED,
WorkRequest.Statuses.ABORTED,
}
)
def lock_workflows_for_update(
self, nowait: bool = False
) -> "WorkRequestQuerySet[A]":
"""
Take an update lock on all work requests in each of some workflows.
To minimize the chance of deadlocks, this should be used near the
beginning of any transaction that might update multiple work
requests (which can happen as a result of what appear to be status
changes to single work requests, due to effects propagating across
dependencies and around workflow trees), before any other update
locks are taken.
All work requests that are in the same workflow as any of the work
requests in this query set will be locked for update, as well as
these work requests themselves even if they are not in a workflow.
:param nowait: if True, raise an exception immediately if another
transaction has already acquired a lock on one of the selected
work requests; if False, block until the conflicting lock is
released.
"""
# The query set needs to actually select all the rows in question
# (`.exists()` isn't enough), and we have to evaluate it (hence
# `bool()`), but we try to pull back as little data as possible
# since we aren't actually interested in the results.
bool(
WorkRequest.objects.select_for_update(nowait=nowait, no_key=True)
.filter(
# Select any work requests in the same workflows as the
# input work requests.
Q(
workflow_root__in=self.filter(
workflow_root__isnull=False
).values("workflow_root")
)
# Always select the work requests themselves, even if they
# aren't in workflows.
| Q(id__in=self)
)
.values("id")
)
return self
# We don't have a good way to refer to the annotated return type here.
def workflow_flattened(
self,
*,
include_workflows: bool = False,
include_invisible: bool = False,
) -> "WorkRequestQuerySet[Any]":
"""
Return queryset of all the work requests that are part of the workflows.
Recursively include any children of the work requests in the current
queryset. The returned queryset will have an ``ancestor_id``
annotation, allowing callers to correlate children with the ancestor
from which they were found.
Note that if the current queryset includes work requests that are
ancestors of other work requests in the same queryset, then the
returned queryset may include duplicates.
:param work_request: The work request to start from. If it is a
workflow, recursively include its children.
:param include_workflows: If True, include work requests of type
workflow; otherwise, exclude them (i.e. only include leaf work
requests).
:param include_invisible: If True, include work requests whose
workflow data field has ``visible=False`` set, indicating that the
task should not be shown in the visual representation of the
workflow; otherwise, exclude them.
"""
work_request_queryset: WorkRequestQuerySet[WorkRequest] = (
WorkRequestQuerySet(WorkRequest, using=WorkRequest.objects._db)
)
def workflow_flattened_cte(
cte: With,
) -> WorkRequestQuerySet[WorkRequest]:
recursive = self.values(
rec_ancestor=F("id"),
rec_parent=F("parent"),
rec_child=F("id"),
).union(
cte.join(WorkRequest, parent=cte.col.rec_child).values(
rec_ancestor=cte.col.rec_ancestor,
rec_parent=F("parent"),
rec_child=F("id"),
)
)
assert isinstance(recursive, WorkRequestQuerySet)
return recursive
cte = With.recursive(workflow_flattened_cte)
flattened = (
cte.join(work_request_queryset, id=cte.col.rec_child)
.with_cte(cte)
.annotate(ancestor_id=cte.col.rec_ancestor)
)
if not include_workflows:
flattened = flattened.exclude(task_type=TaskTypes.WORKFLOW)
if not include_invisible:
flattened = flattened.visible_in_workflow()
assert isinstance(flattened, WorkRequestQuerySet)
return flattened
class WorkRequestManager(models.Manager["WorkRequest"]):
"""Manager for WorkRequest model."""
def get_queryset(self) -> WorkRequestQuerySet[Any]:
"""Use the custom QuerySet."""
return WorkRequestQuerySet(self.model, using=self._db)
def create_worker(
self,
*,
task_name: str,
task_data: BaseTaskData | dict[str, Any] | None = None,
workspace: Workspace,
created_by: "User",
started_at: dt.datetime | None = None,
parent: Optional["WorkRequest"] = None,
unblock_strategy: str = _UnblockStrategy.DEPS,
priority_base: int = 0,
expiration_delay: int | dt.timedelta | None = None,
event_reactions: EventReactions | None = None,
workflow_data: Optional["WorkRequestWorkflowData"] = None,
) -> "WorkRequest":
"""Create a WorkRequest with type=WORKER."""
if isinstance(task_data, BaseTaskData):
task_data = task_data.model_dump(mode="json", exclude_unset=True)
else:
task_data = task_data or {}
kwargs: dict[str, Any] = {}
if event_reactions is not None:
kwargs["event_reactions_json"] = event_reactions.model_dump(
exclude_unset=True
)
if workflow_data is not None:
kwargs["workflow_data_json"] = workflow_data.model_dump(
exclude_unset=True
)
wr = self.create(
workspace=workspace,
task_type=TaskTypes.WORKER,
task_name=task_name,
task_data=task_data,
created_by=created_by,
# We don't currently need to allow independent control of
# pending_at here, but if we're simulating a work request being
# started at a particular time then it also makes sense to treat
# it as if it had been made pending at that time.
pending_at=started_at,
started_at=started_at,
status=WorkRequest.Statuses.BLOCKED,
parent=parent,
workflow_root=None if parent is None else parent.workflow_root,
unblock_strategy=unblock_strategy,
priority_base=priority_base,
expiration_delay=expiration_delay,
**kwargs,
)
wr.full_clean()
# Worker tasks are not configured on creation, because configuring them
# requires knowing their assigned worker
return wr
def create_server(
self,
*,
task_name: str,
task_data: BaseTaskData | dict[str, Any] | None = None,
workspace: Workspace,
created_by: "User",
parent: Optional["WorkRequest"] = None,
priority_base: int = 0,
workflow_data: Optional["WorkRequestWorkflowData"] = None,
) -> "WorkRequest":
"""Create a WorkRequest with type=SERVER."""
if isinstance(task_data, BaseTaskData):
task_data = task_data.model_dump(mode="json", exclude_unset=True)
else:
task_data = task_data or {}
kwargs: dict[str, Any] = {}
if workflow_data is not None:
kwargs["workflow_data_json"] = workflow_data.model_dump(
exclude_unset=True
)
wr = self.create(
workspace=workspace,
task_type=TaskTypes.SERVER,
task_name=task_name,
task_data=task_data,
created_by=created_by,
status=WorkRequest.Statuses.BLOCKED,
parent=parent,
workflow_root=None if parent is None else parent.workflow_root,
priority_base=priority_base,
**kwargs,
)
wr.full_clean()
return wr
@staticmethod
def _serialize_task_data(
task_data: BaseTaskData | dict[str, Any] | None = None,
) -> dict[str, Any]:
"""Serialize an input task_data to be stored in a JSONField."""
if isinstance(task_data, BaseTaskData):
return task_data.model_dump(mode="json", exclude_unset=True)
else:
return task_data or {}
@staticmethod
def _enforce_runtime_parameters(
task_data: dict[str, Any], runtime_parameters: RuntimeParameters
) -> None:
"""Ensure that task_data fits within runtime_parameters."""
if runtime_parameters == RuntimeParameter.ANY:
return
assert isinstance(runtime_parameters, dict)
for key, value in task_data.items():
if key in runtime_parameters:
permitted = runtime_parameters.get(key)
if permitted == RuntimeParameter.ANY:
continue
assert isinstance(permitted, list)
if value not in permitted:
raise ValidationError(
{
"task_data": (
f"runtime_parameters do not permit {key!r}: "
f"{value!r}, only {permitted!r}"
)
}
)
else:
raise ValidationError(
{
"task_data": (
f"runtime_parameters do not permit {key!r}"
),
}
)
def create_signing(
self,
*,
task_name: str,
task_data: BaseTaskData | dict[str, Any] | None = None,
workspace: Workspace,
created_by: "User",
parent: Optional["WorkRequest"] = None,
priority_base: int = 0,
workflow_data: Optional["WorkRequestWorkflowData"] = None,
) -> "WorkRequest":
"""Create a WorkRequest with type=SIGNING."""
kwargs: dict[str, Any] = {}
if workflow_data is not None:
kwargs["workflow_data_json"] = workflow_data.model_dump(
exclude_unset=True
)
wr = self.create(
workspace=workspace,
task_type=TaskTypes.SIGNING,
task_name=task_name,
task_data=self._serialize_task_data(task_data),
created_by=created_by,
status=WorkRequest.Statuses.BLOCKED,
parent=parent,
workflow_root=None if parent is None else parent.workflow_root,
priority_base=priority_base,
**kwargs,
)
wr.full_clean()
return wr
def create_internal(
self,
*,
task_name: str,
task_data: BaseTaskData | dict[str, Any] | None = None,
workspace: Workspace,
created_by: "User",
parent: Optional["WorkRequest"] = None,
priority_base: int = 0,
workflow_data: Optional["WorkRequestWorkflowData"] = None,
) -> "WorkRequest":
"""Create a WorkRequest with type=INTERNAL."""
kwargs: dict[str, Any] = {}
if workflow_data is not None:
kwargs["workflow_data_json"] = workflow_data.model_dump(
exclude_unset=True
)
wr = self.create(
workspace=workspace,
task_type=TaskTypes.INTERNAL,
task_name=task_name,
task_data=self._serialize_task_data(task_data),
created_by=created_by,
status=WorkRequest.Statuses.BLOCKED,
parent=parent,
workflow_root=None if parent is None else parent.workflow_root,
priority_base=priority_base,
**kwargs,
)
wr.full_clean()
return wr
def create_wait(
self,
*,
task_name: str,
task_data: BaseTaskData | dict[str, Any] | None = None,
workspace: Workspace,
created_by: "User",
parent: Optional["WorkRequest"] = None,
priority_base: int = 0,
workflow_data: "WorkRequestWorkflowData",
) -> "WorkRequest":
"""Create a WorkRequest with type=WAIT."""
wr = self.create(
workspace=workspace,
task_type=TaskTypes.WAIT,
task_name=task_name,
task_data=self._serialize_task_data(task_data),
created_by=created_by,
status=WorkRequest.Statuses.BLOCKED,
parent=parent,
workflow_root=None if parent is None else parent.workflow_root,
workflow_data_json=workflow_data.model_dump(exclude_unset=True),
priority_base=priority_base,
)
wr.full_clean()
return wr
def create_workflow_callback(
self,
*,
parent: "WorkRequest",
step: str,
display_name: str | None = None,
visible: bool = True,
) -> "WorkRequest":
"""
Create a workflow callback WorkRequest.
A parent is always required, as a callback only makes sense as part of
a workflow.
:param step: string set by the workflow to identify this callback
"""
# Import here to prevent circular imports
from debusine.server.workflows.models import WorkRequestWorkflowData
return self.create_internal(
workspace=parent.workspace,
parent=parent,
created_by=parent.created_by,
task_name="workflow",
priority_base=parent.priority_effective,
workflow_data=WorkRequestWorkflowData(
step=step, display_name=display_name or step, visible=visible
),
)
def create_workflow(
self,
*,
template: WorkflowTemplate,
data: BaseTaskData | dict[str, Any] | None = None,
parent: Optional["WorkRequest"] = None,
) -> "WorkRequest":
"""Create a workflow from a template and user-provided data."""
# Import here to prevent circular imports
from debusine.server.workflows.models import WorkRequestWorkflowData
enforce(template.can_run)
created_by = context.require_user()
assert created_by.is_authenticated
# Merge user provided data into template data
task_data = template.static_parameters
user_data = self._serialize_task_data(data)
self._enforce_runtime_parameters(user_data, template.runtime_parameters)
task_data.update(user_data)
# Build the WorkRequest
work_request = self.create(
workspace=template.workspace,
parent=parent,
workflow_root=None if parent is None else parent.workflow_root,
workflow_template=template,
workflow_data_json=WorkRequestWorkflowData(
workflow_template_name=template.name
).model_dump(exclude_unset=True),
created_by=created_by,
status=WorkRequest.Statuses.BLOCKED,
task_type=TaskTypes.WORKFLOW,
task_name=template.task_name,
task_data=task_data,
priority_base=template.priority,
)
if parent is None:
# A workflow with no parent is its own root.
work_request.workflow_root = work_request
# Root work requests need an internal collection
# (:collection:`debusine:workflow-internal`)
if parent is None:
work_request.internal_collection = Collection.objects.create(
name=f"workflow-{work_request.id}",
category=CollectionCategory.WORKFLOW_INTERNAL,
workspace=work_request.workspace,
retains_artifacts=Collection.RetainsArtifacts.WORKFLOW,
)
work_request.save()
work_request.full_clean()
work_request._validate_workflow()
work_request.save()
return work_request
def create_synchronization_point(
self,
*,
parent: "WorkRequest",
step: str,
display_name: str | None = None,
) -> "WorkRequest":
"""
Create a synchronization point WorkRequest.
A parent is always required, as a synchronization point only makes
sense as part of a workflow.
"""
# Import here to prevent circular imports
from debusine.server.workflows.models import WorkRequestWorkflowData
return self.create_internal(
workspace=parent.workspace,
parent=parent,
created_by=parent.created_by,
task_name="synchronization_point",
priority_base=parent.priority_effective,
workflow_data=WorkRequestWorkflowData(
step=step, display_name=display_name or step
),
)
class _RuntimeStatuses(models.TextChoices):
NEEDS_INPUT = "needs_input", "Needs Input"
RUNNING = "running", "Running"
WAITING = "waiting", "Waiting"
PENDING = "pending", "Pending"
ABORTED = "aborted", "Aborted"
COMPLETED = "completed", "Completed"
BLOCKED = "blocked", "Blocked"
class _DynamicTaskDataExtraFields(BaseDynamicTaskData):
model_config = BaseDynamicTaskData.model_config | pydantic.ConfigDict(
extra="allow"
)
[docs]
class WorkRequest(models.Model):
"""
Database model of a request to execute a task.
Time-consuming operations offloaded to Workers and using
Artifacts (and associated Files) as input and output.
Submission API needs to check if the request is valid using
ontological rules - e.g. whether the specified distribution for
a build task exists.
Avoid exposing the status of tasks to the admin interface to avoid
runaway changes whilst the scheduler process is running.
The WorkRequest uses the non-Django tasks module to do the checks on
whether a task can run on a particular worker.
WorkRequest State Machine
=========================
New WorkRequest database entries default to
``WorkRequest.Statuses.BLOCKED``.
Once the WorkRequest is unblocked (which may happen almost immediately
if it has no dependencies), the status is changed to
``WorkRequest.Statuses.PENDING``.
Once the WorkRequest is assigned to a worker and is running,
the status is changed to ``WorkRequest.Statuses.RUNNING``.
If the WorkRequest is aborted, the status is
``WorkRequest.Statuses.ABORTED``.
If the task finishes on the Worker, the WorkRequest status will be
``WorkRequest.Statuses.COMPLETED`` and a WorkRequest.Result is then set,
``WorkRequest.Results.SUCCESS`` or ``WorkRequest.Results.FAILURE``.
.. graphviz::
digraph {
Statuses_BLOCKED -> Statuses_PENDING -> Statuses_RUNNING
-> Statuses_COMPLETED;
Statuses_BLOCKED -> Statuses_PENDING -> Statuses_COMPLETED;
Statuses_BLOCKED -> Statuses_PENDING -> Statuses_ABORTED;
Statuses_BLOCKED -> Statuses_PENDING -> Statuses_RUNNING
-> Statuses_ABORTED;
}
``WorkRequest.pending_at`` is set when the WorkRequest moves from
``WorkRequest.Statuses.BLOCKED`` to ``WorkRequest.Statuses.PENDING``.
``WorkRequest.started_at`` is set when the WorkRequest moves from
``WorkRequest.Statuses.PENDING`` to ``WorkRequest.Statuses.RUNNING``.
``WorkRequest.completed_at`` is set when the Task moves from
``WorkRequest.Statuses.RUNNING`` to ``WorkRequest.Statuses.COMPLETED``
or ``WorkRequest.Statuses.ABORTED``.
"""
Roles: TypeAlias = WorkRequestRoles
Statuses: TypeAlias = _WorkRequestStatuses
RuntimeStatuses: TypeAlias = _RuntimeStatuses
Results: TypeAlias = WorkRequestResults
UnblockStrategy: TypeAlias = _UnblockStrategy
workspace = models.ForeignKey("Workspace", on_delete=models.PROTECT)
created_at = models.DateTimeField(auto_now_add=True)
pending_at = models.DateTimeField(blank=True, null=True)
started_at = models.DateTimeField(blank=True, null=True)
completed_at = models.DateTimeField(blank=True, null=True)
# workflow_last_activity_at was introduced to support UI changes and
# could be dropped in the future to have a different way to support
# UI (or, if the UI changes and is not needed anymore: this field could
# be deleted).
#
# Use it only in the context of UI, and bear in mind it is a strong
# candidate for refactoring.
#
# See also #656
workflow_last_activity_at = models.DateTimeField(blank=True, null=True)
created_by = models.ForeignKey("User", on_delete=models.PROTECT)
# Usually null, but may be set for WAIT tasks that were explicitly
# completed by a user action.
completed_by = models.ForeignKey(
"User",
blank=True,
null=True,
on_delete=models.SET_NULL,
related_name="completed_work_requests",
)
aborted_by = models.ForeignKey(
"User",
blank=True,
null=True,
on_delete=models.SET_NULL,
related_name="aborted_work_requests",
)
status = models.CharField(
max_length=9,
choices=Statuses.choices,
default=Statuses.BLOCKED,
editable=False,
)
# workflow_runtime_status was introduced to support UI changes and
# could be dropped in the future to have a different way to support
# UI (or, if the UI changes and is not needed anymore: this field could
# be deleted).
#
# Use it only in the context of UI, and bear in mind it is a strong
# candidate for refactoring.
#
# See also #656
#
# TODO: ruff is correct that we shouldn't use null=True on a CharField,
# but fixing that retroactively is non-trivial.
workflow_runtime_status = models.CharField( # noqa: DJ001
max_length=11,
choices=RuntimeStatuses.choices,
editable=False,
blank=True,
null=True,
)
# Flag indicating that debusine.server.celery.update_workflows needs to
# update workflow_last_activity_at and workflow_runtime_status on
# workflows containing this work request.
workflows_need_update = models.BooleanField(default=False)
result = models.CharField(
max_length=7,
choices=Results.choices,
default=Results.NONE,
editable=False,
)
# any one work request can only be on one worker
# even if the worker can handle multiple work request.
worker = models.ForeignKey(
"Worker",
null=True,
blank=True,
on_delete=models.CASCADE,
related_name="assigned_work_requests",
)
task_type = models.CharField(
max_length=8,
choices=TaskTypes.choices,
default=TaskTypes.WORKER,
editable=False,
verbose_name="Type of task to execute",
)
task_name = models.CharField(
max_length=100, verbose_name='Name of the task to execute'
)
task_data = JSONField(default=dict, blank=True, encoder=DjangoJSONEncoder)
dynamic_task_data = JSONField(
null=True, blank=True, encoder=DjangoJSONEncoder
)
priority_base = models.IntegerField(
default=0, help_text="Base priority of this work request"
)
priority_adjustment = models.IntegerField(
default=0,
help_text=(
"Administrator adjustment to the priority of this work request"
),
)
output_data_json = JSONField(
null=True,
blank=True,
db_column="output_data",
encoder=DjangoJSONEncoder,
)
# Workflows
unblock_strategy = models.CharField(
max_length=6,
choices=UnblockStrategy.choices,
default=UnblockStrategy.DEPS,
editable=False,
)
# workflow hierarchy tree (optional, null if stand-alone)
parent = models.ForeignKey(
"self",
# WorkRequests are only deleted through expiration, use CASCADE
on_delete=models.CASCADE,
blank=True,
null=True,
related_name="children",
)
workflow_root = models.ForeignKey(
"self",
# Handled by CASCADE on WorkRequest.parent above.
on_delete=models.DO_NOTHING,
blank=True,
null=True,
related_name="descendants",
)
# order of execution within a workflow hierarchy (optional)
dependencies = models.ManyToManyField(
"self", symmetrical=False, related_name="reverse_dependencies"
)
workflow_template = models.ForeignKey(
"WorkflowTemplate", on_delete=models.SET_NULL, null=True, blank=True
)
workflow_data_json = JSONField(
default=dict,
blank=True,
db_column="workflow_data",
encoder=DjangoJSONEncoder,
)
event_reactions_json = JSONField(
default=dict,
blank=True,
db_column="event_reactions",
encoder=DjangoJSONEncoder,
)
internal_collection = models.OneToOneField(
"Collection",
related_name="workflow",
null=True,
blank=True,
on_delete=models.PROTECT,
)
expiration_delay = models.DurationField(blank=True, null=True)
supersedes = models.OneToOneField(
"WorkRequest",
on_delete=models.SET_NULL,
related_name="superseded",
null=True,
blank=True,
)
configured_task_data = JSONField(
null=True,
blank=True,
encoder=DjangoJSONEncoder,
help_text="task_data with task configuration applied",
)
version = models.IntegerField(
default=0,
help_text="version of the code that computed the dynamic task data",
)
scheduler_tags_provided = ArrayField(
models.CharField(max_length=255),
default=list,
blank=True,
help_text="final set of task-provided tags used for scheduling",
)
scheduler_tags_required = ArrayField(
models.CharField(max_length=255),
default=list,
blank=True,
help_text="final set of task-required tags used for scheduling",
)
objects = WorkRequestManager.from_queryset(WorkRequestQuerySet)()
class Meta(TypedModelMeta):
base_manager_name = "objects"
constraints = [
# All work requests with parents have a workflow root.
CheckConstraint(
check=Q(parent__isnull=True) | Q(workflow_root__isnull=False),
name="%(app_label)s_%(class)s_workflow_has_root",
),
]
indexes = [
# Handles the main scheduler query.
models.Index(
(F("priority_base") + F("priority_adjustment")).desc(),
"created_at",
name="%(app_label)s_%(class)s_pending_idx",
condition=Q(status=_WorkRequestStatuses.PENDING),
),
# Handles queries from workers.
models.Index(
"worker",
name="%(app_label)s_%(class)s_worker_idx",
condition=Q(
status__in=(
_WorkRequestStatuses.PENDING,
_WorkRequestStatuses.RUNNING,
)
),
),
# Handles debusine.server.celery.update_workflows.
models.Index(
"workflows_need_update",
name="%(app_label)s_%(class)s_wf_update_idx",
),
]
permissions = [
(
"manage_workrequest_priorities",
"Can set positive priority adjustments on work requests",
)
]
ordering = ["id"]
def __str__(self) -> str:
"""Return the id of the WorkRequest."""
return str(self.id)
[docs]
def save(self, *args: Any, **kwargs: Any) -> None:
"""Save the current instance."""
if self._state.adding:
self.process_event_reactions("on_creation")
super().save(*args, **kwargs)
[docs]
def get_absolute_url(self) -> str:
"""Return an absolute URL to this work request."""
from debusine.server.scopes import urlconf_scope
with urlconf_scope(self.workspace.scope.name):
return reverse(
"workspaces:work_requests:detail",
kwargs={"wname": self.workspace.name, "pk": self.pk},
)
[docs]
def get_absolute_url_retry(self) -> str:
"""Return an absolute URL to retry this work request."""
from debusine.server.scopes import urlconf_scope
with urlconf_scope(self.workspace.scope.name):
return reverse(
"workspaces:work_requests:retry",
kwargs={"wname": self.workspace.name, "pk": self.pk},
)
[docs]
def get_absolute_url_abort(self) -> str:
"""Return an absolute URL to abort this work request."""
from debusine.server.scopes import urlconf_scope
with urlconf_scope(self.workspace.scope.name):
return reverse(
"workspaces:work_requests:abort",
kwargs={"wname": self.workspace.name, "pk": self.pk},
)
[docs]
def get_absolute_url_unblock(self) -> str:
"""Return an absolute URL to unblock this work request."""
from debusine.server.scopes import urlconf_scope
with urlconf_scope(self.workspace.scope.name):
return reverse(
"workspaces:work_requests:unblock",
kwargs={"wname": self.workspace.name, "pk": self.pk},
)
[docs]
def get_absolute_url_confirm(self) -> str:
"""Return an absolute URL to confirm this work request."""
from debusine.server.scopes import urlconf_scope
with urlconf_scope(self.workspace.scope.name):
return reverse(
"workspaces:work_requests:confirm",
kwargs={"wname": self.workspace.name, "pk": self.pk},
)
[docs]
def ui(self, request: "HttpRequest") -> "WorkRequestUI":
"""Return a UI helper for this instance."""
from debusine.web.views.ui.work_requests import WorkRequestUI
return WorkRequestUI.get(request, self)
[docs]
def has_role(self, user: PermissionUser, role: WorkRequestRoles) -> bool:
"""Check if the user has the given role on this Workspace."""
# Honor implied_by_public immediately as it's independent from the user
if role.implied_by_public and self.workspace.public:
return True
if not user or not user.is_authenticated:
return False
if role.implied_by_creator and user == self.created_by:
return True
if context.user == user and context.workspace == self.workspace:
# Allowed as implied by scope roles the user has
if not context.scope_roles.isdisjoint(role.implied_by_scope_roles):
return True
# Allowed as implied by workspace roles the user has
if not context.workspace_roles.isdisjoint(
role.implied_by_workspace_roles
):
return True
# The context has enough information to know that the user does not
# have the given role on this workspace
return False
return (
WorkRequest.objects.with_role(user, role)
.filter(pk=self.pk)
.exists()
)
[docs]
def get_roles(self, user: PermissionUser) -> frozenset[WorkRequestRoles]:
"""
Get the effective roles of the user on this workspace.
If you invoke this method on a whole queryset, you greatly reduce the
number of queries by annotating the queryset using
``with_role_annotations``.
"""
roles: list[WorkRequestRoles] = []
if ((roles_for := getattr(self, "roles_for", None)) is not None) and (
(
roles_for
and user
and user.is_authenticated
and int(roles_for) == user.pk
)
or (not roles_for and (not user or not user.is_authenticated))
):
# Use roles from annotations
for role in WorkRequestRoles:
if getattr(self, f"is_{role}", False):
roles.append(role)
else:
# Fallback on querying each role
for role in WorkRequestRoles:
if self.has_role(user, role):
roles.append(role)
return WorkRequestRoles.from_iterable(roles)
[docs]
@permission_check(
"{user} cannot display {resource}",
workers=Allow.ALWAYS,
anonymous=Allow.PASS,
)
def can_display(self, user: PermissionUser) -> bool:
"""Check if the work request can be displayed."""
return self.has_role(user, WorkRequest.Roles.VIEWER)
[docs]
@permission_check("{user} cannot retry {resource}", workers=Allow.ALWAYS)
def can_retry(self, user: PermissionUser) -> bool:
"""Check if the work request can be retried."""
return self.has_role(user, WorkRequestRoles.CONTRIBUTOR)
[docs]
@permission_check("{user} cannot abort {resource}")
def can_abort(self, user: PermissionUser) -> bool:
"""Check if the work request can be aborted."""
return self.has_role(user, WorkRequestRoles.CREATOR)
[docs]
@permission_check("{user} cannot unblock {resource}")
def can_unblock(self, user: PermissionUser) -> bool:
"""Check if the work request can be unblocked."""
return self.has_role(user, WorkRequestRoles.OWNER)
[docs]
@permission_check("{user} cannot confirm {resource}")
def can_confirm(self, user: PermissionUser) -> bool:
"""Check if the work request can be confirmed."""
return self.has_role(user, WorkRequestRoles.CREATOR)
[docs]
def set_current(self) -> None:
"""Set the execution context to this scope/user/workspace."""
context.set_scope(self.workspace.scope)
context.set_user(self.created_by)
self.workspace.set_current()
@property
def used_task_data(self) -> Any:
"""Return the task data used to run this task."""
if self.configured_task_data is not None:
return self.configured_task_data
else:
return self.task_data
@property
def output_data(self) -> OutputData | None:
"""Access output_data_json as a pydantic structure."""
return (
None
if self.output_data_json is None
else OutputData(**self.output_data_json)
)
@output_data.setter
def output_data(self, value: OutputData | None) -> None:
"""Set output_data_json from a pydantic structure."""
self.output_data_json = (
None if value is None else value.model_dump(exclude_unset=True)
)
@property
def workflow_data(self) -> "WorkRequestWorkflowData":
"""Access workflow_data_json as a python structure."""
# Import here to avoid a circular dependency
from debusine.server.workflows.models import WorkRequestWorkflowData
return WorkRequestWorkflowData(**self.workflow_data_json)
@workflow_data.setter
def workflow_data(self, value: "WorkRequestWorkflowData") -> None:
"""Set workflow_data_json from a python structure."""
self.workflow_data_json = value.model_dump(
mode="json", exclude_unset=True
)
@property
def event_reactions(self) -> EventReactions:
"""Access event_reactions_json as a pydantic structure."""
return EventReactions(**self.event_reactions_json)
@event_reactions.setter
def event_reactions(self, value: EventReactions) -> None:
"""Set event_reactions_json from a pydantic structure."""
self.event_reactions_json = value.model_dump(exclude_unset=True)
[docs]
def add_event_reaction(
self,
event_name: Literal[
"on_creation",
"on_unblock",
"on_assignment",
"on_success",
"on_failure",
],
action: EventReaction,
) -> None:
"""Add an event reaction."""
event_reactions = self.event_reactions
event_reactions.add_reaction(event_name, action)
self.event_reactions = event_reactions
self.save()
[docs]
@contextmanager
def scheduling_disabled(self) -> Generator[None]:
"""Temporarily disable scheduling on changes to this work request."""
try:
# See debusine.server.scheduler._work_request_changed.
self._disable_signals = True
yield
finally:
del self._disable_signals
[docs]
@contextmanager
def running_workflow_orchestrator(self) -> Generator[None]:
"""
Temporarily indicate that we are running this workflow's orchestrator.
:py:meth:`maybe_finish_workflow` does nothing while this is the
case, since the workflow may not yet have been completely populated.
"""
try:
self._running_workflow_orchestrator = True
yield
finally:
del self._running_workflow_orchestrator
def _child_task_data(
self, task_data: BaseTaskData | None = None
) -> dict[str, Any]:
"""Annotate task_data with the current task_configuration value."""
if task_data is not None:
task_data_json = task_data.model_dump(
mode="json", exclude_unset=True
)
else:
task_data_json = {}
if (
# If this workflow does not define a special task configuration,
# we have nothing to do
"task_configuration" in self.task_data
and self.task_data["task_configuration"]
!= BaseTaskData.DEFAULT_TASK_CONFIGURATION_LOOKUP
) and (
# If task_data does define a special task configuration, leave it be
"task_configuration" not in task_data_json
or task_data_json["task_configuration"]
== BaseTaskData.DEFAULT_TASK_CONFIGURATION_LOOKUP
):
task_data_json["task_configuration"] = self.task_data[
"task_configuration"
]
return task_data_json
[docs]
def create_child_worker(
self,
task_name: str,
task_data: BaseTaskData | None = None,
workflow_data: Optional["WorkRequestWorkflowData"] = None,
event_reactions: EventReactions | None = None,
relative_priority: int = 0,
) -> "WorkRequest":
"""Create a child WORKER WorkRequest."""
if self.task_type != TaskTypes.WORKFLOW:
raise ValueError("Only workflows may have child work requests.")
child = WorkRequest.objects.create_worker(
workspace=self.workspace,
parent=self,
created_by=self.created_by,
task_name=task_name,
task_data=self._child_task_data(task_data),
priority_base=self.priority_effective + relative_priority,
workflow_data=workflow_data,
event_reactions=event_reactions,
)
workflow_status = compute_workflow_runtime_status(self)
self.workflow_runtime_status = workflow_status
self.save()
return child
[docs]
def create_child_server(
self,
task_name: str,
task_data: BaseTaskData | None = None,
workflow_data: Optional["WorkRequestWorkflowData"] = None,
) -> "WorkRequest":
"""Create a child SERVER WorkRequest."""
if self.task_type != TaskTypes.WORKFLOW:
raise ValueError("Only workflows may have child work requests.")
child = WorkRequest.objects.create_server(
workspace=self.workspace,
parent=self,
created_by=self.created_by,
task_name=task_name,
task_data=self._child_task_data(task_data),
priority_base=self.priority_effective,
workflow_data=workflow_data,
)
workflow_status = compute_workflow_runtime_status(self)
self.workflow_runtime_status = workflow_status
self.save()
return child
[docs]
def create_child_internal(
self,
task_name: str,
workflow_data: Optional["WorkRequestWorkflowData"] = None,
relative_priority: int = 0,
) -> "WorkRequest":
"""Create a child INTERNAL WorkRequest."""
if self.task_type != TaskTypes.WORKFLOW:
raise ValueError("Only workflows may have child work requests.")
child = WorkRequest.objects.create_internal(
workspace=self.workspace,
parent=self,
created_by=self.created_by,
task_name=task_name,
task_data=self._child_task_data(None),
priority_base=self.priority_effective + relative_priority,
workflow_data=workflow_data,
)
workflow_status = compute_workflow_runtime_status(self)
self.workflow_runtime_status = workflow_status
self.save()
return child
[docs]
def create_child_workflow(
self,
task_name: str,
task_data: BaseTaskData | None = None,
workflow_data: Optional["WorkRequestWorkflowData"] = None,
event_reactions: EventReactions | None = None,
relative_priority: int = 0,
) -> "WorkRequest":
"""Create a child WORKFLOW WorkRequest."""
if self.task_type != TaskTypes.WORKFLOW:
raise ValueError("Only workflows may have child work requests.")
child = WorkRequest.objects.create(
workspace=self.workspace,
parent=self,
workflow_root=self.workflow_root,
created_by=self.created_by,
status=WorkRequest.Statuses.BLOCKED,
task_type=TaskTypes.WORKFLOW,
task_name=task_name,
task_data=self._child_task_data(task_data),
priority_base=self.priority_effective + relative_priority,
workflow_data_json=(
workflow_data.model_dump(exclude_unset=True)
if workflow_data
else {}
),
event_reactions_json=(
event_reactions.model_dump() if event_reactions else {}
),
)
workflow_status = compute_workflow_runtime_status(self)
self.workflow_runtime_status = workflow_status
self.save()
return child
[docs]
def create_child_signing(
self,
task_name: str,
task_data: BaseTaskData | None = None,
workflow_data: Optional["WorkRequestWorkflowData"] = None,
) -> "WorkRequest":
"""Create a child SIGNING WorkRequest."""
if self.task_type != TaskTypes.WORKFLOW:
raise ValueError("Only workflows may have child work requests.")
child = WorkRequest.objects.create_signing(
workspace=self.workspace,
parent=self,
created_by=self.created_by,
task_name=task_name,
task_data=self._child_task_data(task_data),
priority_base=self.priority_effective,
workflow_data=workflow_data,
)
workflow_status = compute_workflow_runtime_status(self)
self.workflow_runtime_status = workflow_status
self.save()
return child
[docs]
def create_child_wait(
self,
task_name: str,
workflow_data: "WorkRequestWorkflowData",
task_data: BaseTaskData | None = None,
) -> "WorkRequest":
"""Create a child WAIT WorkRequest."""
if self.task_type != TaskTypes.WORKFLOW:
raise ValueError("Only workflows may have child work requests.")
child = WorkRequest.objects.create_wait(
workspace=self.workspace,
parent=self,
created_by=self.created_by,
task_name=task_name,
task_data=self._child_task_data(task_data),
workflow_data=workflow_data,
priority_base=self.priority_effective,
)
workflow_status = compute_workflow_runtime_status(self)
self.workflow_runtime_status = workflow_status
self.save()
return child
@overload
def lookup_single(
self,
lookup: "LookupSingle",
*,
workspace: Workspace | None = None,
default_category: CollectionCategory | None = None,
expect_type: Literal[LookupChildType.BARE],
) -> "LookupResultBare": ...
@overload
def lookup_single(
self,
lookup: "LookupSingle",
*,
workspace: Workspace | None = None,
default_category: CollectionCategory | None = None,
expect_type: Literal[LookupChildType.ARTIFACT],
) -> "LookupResultArtifact": ...
@overload
def lookup_single(
self,
lookup: "LookupSingle",
*,
workspace: Workspace | None = None,
default_category: CollectionCategory | None = None,
expect_type: Literal[LookupChildType.COLLECTION],
) -> "LookupResultCollection": ...
@overload
def lookup_single(
self,
lookup: "LookupSingle",
*,
workspace: Workspace | None = None,
default_category: CollectionCategory | None = None,
expect_type: LookupChildType = LookupChildType.ANY,
) -> "LookupResult": ...
[docs]
def lookup_single(
self,
lookup: "LookupSingle",
*,
workspace: Workspace | None = None,
default_category: CollectionCategory | None = None,
expect_type: LookupChildType = LookupChildType.ANY,
) -> "LookupResult":
"""
Look up a collection item for this work request using a single lookup.
See :ref:`lookup-single`.
:raises KeyError: if the lookup does not resolve to an item.
:raises LookupError: if the lookup is invalid in some way, or does
not resolve to the expected collection item type.
"""
# Import here to prevent circular imports.
from debusine.server.collections.lookup import lookup_single
return lookup_single(
lookup,
workspace=workspace or self.workspace,
user=self.created_by,
default_category=default_category,
workflow_root=self.workflow_root,
expect_type=expect_type,
)
@overload
def lookup_multiple(
self,
lookup: "LookupMultiple",
*,
workspace: Workspace | None = None,
default_category: CollectionCategory | None = None,
expect_type: Literal[LookupChildType.ARTIFACT],
) -> Sequence["LookupResultArtifact"]: ...
@overload
def lookup_multiple(
self,
lookup: "LookupMultiple",
*,
workspace: Workspace | None = None,
default_category: CollectionCategory | None = None,
expect_type: Literal[LookupChildType.COLLECTION],
) -> Sequence["LookupResultCollection"]: ...
@overload
def lookup_multiple(
self,
lookup: "LookupMultiple",
*,
workspace: Workspace | None = None,
default_category: CollectionCategory | None = None,
expect_type: LookupChildType = LookupChildType.ANY,
) -> Sequence["LookupResult"]: ...
[docs]
def lookup_multiple(
self,
lookup: "LookupMultiple",
*,
workspace: Workspace | None = None,
default_category: CollectionCategory | None = None,
expect_type: LookupChildType = LookupChildType.ANY,
) -> Sequence["LookupResult"]:
"""
Look up collection items for this work request using a multiple lookup.
See :ref:`lookup-multiple`.
:raises KeyError: if any of the lookups does not resolve to an item.
:raises LookupError: if any of the lookups is invalid in some way,
or does not resolve to the expected collection item type.
"""
# Import here to prevent circular imports.
from debusine.server.collections.lookup import lookup_multiple
return lookup_multiple(
lookup,
workspace=workspace or self.workspace,
user=self.created_by,
default_category=default_category,
workflow_root=self.workflow_root,
expect_type=expect_type,
)
[docs]
def can_satisfy_dynamic_data(self) -> bool:
"""Check if dynamic data can still be looked up."""
# Import here to prevent an import loop
from debusine.db.models.task_database import TaskDatabase
try:
task = self.get_task()
except TaskConfigError:
return False
try:
task.compute_dynamic_data(TaskDatabase(self))
except LookupError:
return False
return True
[docs]
def is_aborted_failed(self) -> bool:
"""Check if the work request was aborted or failed."""
return self.status == WorkRequest.Statuses.ABORTED or (
self.status == WorkRequest.Statuses.COMPLETED
and self.result == WorkRequest.Results.FAILURE
)
def _verify_retry(self) -> None:
"""
Check if a work request can be retried.
:raises CannotRetry: raised with an explanation of why it cannot be
retried
"""
if self.task_type not in {
TaskTypes.WORKFLOW,
TaskTypes.WORKER,
TaskTypes.INTERNAL,
TaskTypes.SIGNING,
}:
raise CannotRetry(
"Only workflow, worker, internal, or signing tasks"
" can be retried"
)
if hasattr(self, "superseded"):
raise CannotRetry("Cannot retry old superseded tasks")
if not self.is_aborted_failed():
raise CannotRetry("Only aborted or failed tasks can be retried")
if self.task_type in {TaskTypes.INTERNAL, TaskTypes.WORKFLOW}:
return
if not self.can_satisfy_dynamic_data():
raise CannotRetry("Task dependencies cannot be satisfied")
[docs]
def verify_retry(self) -> bool:
"""Check if this work request can be retried."""
if not self.can_retry(context.user):
return False
try:
self._verify_retry()
except CannotRetry:
return False
return True
def _retry_workflow(self, reason: WorkRequestRetryReason) -> Self:
"""Retry logic for workflows."""
# Import here to prevent circular imports
from debusine.server.workflows.base import orchestrate_workflow
# Set the workflow back to running, so that further actions execute in
# a running workflow
self.status = self.Statuses.PENDING
self.mark_running()
# Update the retry count
workflow_data = self.workflow_data
if reason in {
WorkRequestRetryReason.WORKER_FAILED,
WorkRequestRetryReason.DELAY,
}:
workflow_data.retry_count += 1
else:
workflow_data.retry_count = 0
self.workflow_data = workflow_data
self.save()
try:
# TODO: workflow orchestrators may be slow; this should move to
# a Celery task
# Give the workflow a chance to update its graph
orchestrate_workflow(self)
# Retry failed child work requests
for child in (
WorkRequest.objects.aborted_or_failed()
.filter(parent=self)
.order_by("id")
):
try:
child.retry(reason=reason)
except CannotRetry as exc:
raise CannotRetry(
f"child work request {child.pk} cannot be retried"
) from exc
finally:
self.maybe_finish_workflow()
return self
def _retry_supersede(self, reason: WorkRequestRetryReason) -> "WorkRequest":
"""Retry logic superseding work requests."""
work_request = WorkRequest.objects.create(
workspace=self.workspace,
created_by=self.created_by,
status=WorkRequest.Statuses.BLOCKED,
task_type=self.task_type,
task_name=self.task_name,
task_data=self.task_data,
priority_base=self.priority_base,
priority_adjustment=self.priority_adjustment,
unblock_strategy=self.unblock_strategy,
parent=self.parent,
workflow_root=self.workflow_root,
workflow_template=self.workflow_template,
workflow_data_json=self.workflow_data_json,
event_reactions_json=self.event_reactions_json,
expiration_delay=self.expiration_delay,
supersedes=self,
)
# Copy forward dependencies
for dep in self.dependencies.all():
work_request.add_dependency(dep)
# Update any reverse-dependencies of the old work request to point to
# the new one, and move them from pending to blocked
for dep in self.reverse_dependencies.all():
dep.dependencies.remove(self)
dep.add_dependency(work_request)
# If the task was part of an aborted workflow, set it back to running
if (workflow := work_request.parent) and workflow.is_aborted_failed():
# We cannot use mark_running, as it would fail to run when the
# workflow is failed or aborted
workflow.status = WorkRequest.Statuses.RUNNING
workflow.result = WorkRequest.Results.NONE
workflow.completed_at = None
workflow.save()
# Update the retry count
workflow_data = work_request.workflow_data
if reason in {
WorkRequestRetryReason.WORKER_FAILED,
WorkRequestRetryReason.DELAY,
}:
workflow_data.retry_count += 1
else:
workflow_data.retry_count = 0
work_request.workflow_data = workflow_data
work_request.save()
if work_request.can_be_automatically_unblocked():
work_request.mark_pending()
# Update promises associated with the previous work request
if (
self.parent is not None
and self.parent.internal_collection is not None
):
for (
debusine_promise
) in self.parent.internal_collection.child_items.filter(
category=BareDataCategory.PROMISE,
data__promise_work_request_id=self.id,
):
debusine_promise.data["promise_work_request_id"] = (
work_request.id
)
debusine_promise.save()
self.schedule_workflow_update()
return work_request
[docs]
def retry(
self, reason: WorkRequestRetryReason = WorkRequestRetryReason.MANUAL
) -> "WorkRequest":
"""
Create a WorkRequest that supersedes this one.
:param reason: The reason for retrying this work request. If it is
``WORKER_FAILED``, then only allow the retry if the retry count is
less than :py:const:`MAX_AUTOMATIC_RETRIES`. If it is either
``WORKER_FAILED`` or ``DELAY``, then increment the work request's
``retry_count``; otherwise, set it back to 0.
"""
enforce(self.can_retry)
self._verify_retry()
if (
reason == WorkRequestRetryReason.WORKER_FAILED
and self.workflow_data.retry_count >= MAX_AUTOMATIC_RETRIES
):
raise CannotRetry("Maximum number of automatic retries exceeded")
if self.task_type == TaskTypes.WORKFLOW:
return self._retry_workflow(reason=reason)
else:
return self._retry_supersede(reason=reason)
def _verify_abort(self) -> None:
"""
Check if this work request can be aborted.
:raises CannotAbort: if the work request cannot be aborted.
"""
if hasattr(self, "superseded"):
raise CannotAbort("Cannot abort old superseded tasks")
if self.status not in {
WorkRequest.Statuses.PENDING,
WorkRequest.Statuses.BLOCKED,
WorkRequest.Statuses.RUNNING,
}:
raise CannotAbort(
"Only pending, blocked, or running tasks can be aborted"
)
[docs]
def verify_abort(self) -> bool:
"""Check if this work request can be aborted."""
if not self.can_abort(context.user):
return False
try:
self._verify_abort()
except CannotAbort:
return False
return True
[docs]
def abort(self, *, output_data: OutputData | None = None) -> None:
"""
Abort a work request.
If the work request is a workflow, then recursively abort its child
work requests too.
"""
enforce(self.can_abort)
user = context.require_user()
# Enforcing `can_abort` ensures that the user can't be anonymous by
# this point.
assert user.is_authenticated
self._verify_abort()
for work_request in WorkRequest.objects.filter(
id__in=WorkRequest.objects.filter(id=self.id)
.workflow_flattened(include_workflows=True, include_invisible=True)
.filter(
superseded__isnull=True,
status__in={
WorkRequest.Statuses.PENDING,
WorkRequest.Statuses.BLOCKED,
WorkRequest.Statuses.RUNNING,
},
)
.values_list("id", flat=True)
).select_for_update():
work_request.mark_aborted(user=user, output_data=output_data)
# TODO: If the work request was assigned to a worker, we should
# tell the worker so that it can abandon execution
# (https://salsa.debian.org/freexian-team/debusine/-/issues/824).
work_request.worker = None
work_request.save()
[docs]
def clean(self) -> None:
"""
Ensure that task data is valid for this task name.
:raise ValidationError: for invalid data.
"""
from debusine.db.models.tasks import DBTask
if not isinstance(self.task_data, dict):
raise ValidationError(
{"task_data": "task data must be a dictionary"}
)
match self.task_type:
case (
TaskTypes.WORKER
| TaskTypes.SERVER
| TaskTypes.SIGNING
| TaskTypes.WAIT
):
try:
DBTask.class_from_name(self.task_type, self.task_name)
except (KeyError, ValueError) as e:
raise ValidationError(
{
"task_name": f"{self.task_name}:"
f" invalid {self.task_type} task name"
}
) from e
try:
self.get_task()
except TaskConfigError as e:
raise ValidationError(
{
"task_data": f"invalid {self.task_type}"
f" task data: {e}"
}
) from e
case TaskTypes.WORKFLOW:
# Import here to prevent circular imports
from debusine.server.workflows import Workflow
try:
Workflow.from_name(self.task_name)
except (KeyError, ValueError) as e:
raise ValidationError(
{
"task_name": f"{self.task_name}:"
f" invalid workflow name"
}
) from e
try:
self.get_task()
except TaskConfigError as e:
raise ValidationError(
{"task_data": f"invalid workflow data: {e}"}
) from e
# TODO: do we want to run expensive tests here
# (Workflow.validate_input), like looking up the types of
# references artifacts to validate them?
case TaskTypes.INTERNAL:
if self.task_name not in ("synchronization_point", "workflow"):
raise ValidationError(
{
"task_name": f"{self.task_name}:"
" invalid task name for internal task"
}
)
case _:
raise NotImplementedError(
f"task type {self.task_type} not yet supported"
)
match self.task_type:
case TaskTypes.WAIT:
if self.workflow_data.needs_input is None:
raise ValidationError("WAIT tasks must specify needs_input")
def _validate_workflow(self) -> None:
"""Perform validation of workflow data."""
# Import here to prevent circular imports
from debusine.server.workflows import Workflow
if self.workflow_template is None:
raise AssertionError(
"workflow_template is unset in a workflow WorkRequest"
)
# Lookup the orchestrator
workflow_cls = Workflow.from_name(self.workflow_template.task_name)
# Instantiate the orchestrator
orchestrator = workflow_cls(self)
# Thorough input validation
orchestrator.validate_input()
[docs]
def get_task(self, worker: "Worker | None" = None) -> "DBTask[Any, Any]":
"""
Instantiate the Task for this work request.
:param worker: if set, the worker that this work request is intended
to run on; `worker_native_architecture` will be set to its native
architecture
:raise InternalTaskError: if the work request is for an internal
task other than a workflow callback
"""
# Import here to prevent circular imports
from debusine.db.models.tasks import DBTask
from debusine.server.workflows import Workflow
match (self.task_type, self.task_name):
case (
(TaskTypes.WORKER, _)
| (TaskTypes.SERVER, _)
| (TaskTypes.SIGNING, _)
| (TaskTypes.WAIT, _)
):
task_cls = DBTask.class_from_name(
TaskTypes(self.task_type), self.task_name
)
task = task_cls(work_request=self)
case (TaskTypes.WORKFLOW, _):
workflow_cls = Workflow.from_name(self.task_name)
task = workflow_cls(work_request=self)
case (TaskTypes.INTERNAL, "workflow"):
if (workflow := self.parent) is None:
raise InternalTaskError(
"Workflow callback is not contained in a workflow"
)
assert workflow.task_type == TaskTypes.WORKFLOW
task = workflow.get_task()
case (TaskTypes.INTERNAL, _):
raise InternalTaskError(
"Internal tasks other than workflow callbacks cannot be "
"instantiated"
)
case _ as unreachable: # noqa: F841
raise NotImplementedError(
f"task type {self.task_type} not yet supported"
)
if worker is not None:
task.set_worker(worker)
return task
[docs]
def mark_running(self) -> None:
"""Worker has begun executing the task."""
if (
self.task_type
in {TaskTypes.WORKER, TaskTypes.SERVER, TaskTypes.SIGNING}
and self.worker is None
):
raise StatusChangeError(
f"Cannot mark WorkRequest {self.pk} as running:"
" it does not have an assigned worker"
)
if self.status == self.Statuses.RUNNING:
# It was already running - nothing to do
return
if self.status != self.Statuses.PENDING:
raise StatusChangeError(
f"Cannot mark WorkRequest {self.pk} as running:"
f" current status is {self.status}"
)
if self.worker is not None:
work_requests_running_for_worker = WorkRequest.objects.running(
worker=self.worker
)
# There is a possible race condition here. This check (and
# other checks in this class) currently help to avoid
# development mistakes not database full integrity
if (
work_requests_running_for_worker.count()
>= self.worker.concurrency
):
raise StatusChangeError(
f"Cannot mark WorkRequest {self.pk} as running:"
f" the assigned worker {self.worker} is running"
" too many WorkRequests: "
+ str(list(work_requests_running_for_worker.order_by("id")))
)
if (
self.configured_task_data is None
and self.task_type != TaskTypes.INTERNAL
):
raise StatusChangeError(
f"Unconfigured {self.task_type} task marked running"
)
self.started_at = timezone.now()
self.status = self.Statuses.RUNNING
self.save()
if self.pending_at is not None:
emit_metric(
metric_type="histogram",
name="debusine_work_request_start_latency",
labels={
"task_type": str(self.task_type),
"task_name": self.task_name,
"priority": self.priority_for_metrics,
},
value=(self.started_at - self.pending_at).total_seconds(),
)
self.schedule_workflow_update()
logger.debug("Marked WorkRequest %s as running", self.pk)
[docs]
def mark_completed(
self,
result: "WorkRequest.Results",
*,
user: "User | None" = None,
output_data: OutputData | None = None,
) -> None:
"""
Mark the task as completed.
This is normally because a worker has finished executing it, though
some ``WAIT`` tasks may also be completed by the scheduler or by a
user action.
"""
if self.status not in (self.Statuses.PENDING, self.Statuses.RUNNING):
raise StatusChangeError(
f"Cannot mark WorkRequest {self.pk} as completed:"
f" current status is {self.status}"
)
if result == self.Results.NONE:
raise StatusChangeError("result cannot be NONE")
self.result = result
self.completed_by = user
self.completed_at = timezone.now()
self.status = self.Statuses.COMPLETED
if output_data is not None:
if self.output_data is None:
self.output_data = output_data
else:
self.output_data = self.output_data.merge(output_data)
self.save()
if (
output_data is not None
and output_data.runtime_statistics is not None
and output_data.runtime_statistics.duration is not None
and self.worker is not None
and self.worker.worker_pool is not None
):
WorkerPoolTaskExecutionStatistics.objects.create(
worker_pool=self.worker.worker_pool,
worker=self.worker,
scope=self.workspace.scope,
runtime=output_data.runtime_statistics.duration,
)
self.schedule_workflow_update()
logger.debug("Marked WorkRequest %s as completed", self.pk)
match self.result:
case self.Results.SUCCESS:
self.process_event_reactions("on_success")
case self.Results.FAILURE | self.Results.ERROR:
self.process_event_reactions("on_failure")
case self.Results.SKIPPED:
pass
case _ as unreachable:
assert_never(unreachable)
# TODO: This may cascade to other work requests in the same
# workflow, so WorkRequestQuerySet.lock_workflows_for_update must
# have been called first in this transaction. We should check this.
# https://salsa.debian.org/freexian-team/debusine/-/merge_requests/2381#note_689749
self.unblock_reverse_dependencies()
if self.parent is not None:
self.parent.maybe_finish_workflow()
[docs]
def process_event_reactions(
self,
event_name: Literal[
"on_creation",
"on_unblock",
"on_assignment",
"on_success",
"on_failure",
],
) -> None:
"""Process list of actions to perform."""
actions = self.get_triggered_actions(event_name)
self.process_skip_if_lookup_result_changed(
actions[ActionTypes.SKIP_IF_LOOKUP_RESULT_CHANGED]
)
if event_name in {"on_success", "on_failure"}:
notifications.notify_work_request_completed(
self, actions[ActionTypes.SEND_NOTIFICATION]
)
self.process_update_collection_with_artifacts(
actions[ActionTypes.UPDATE_COLLECTION_WITH_ARTIFACTS]
)
self.process_update_collection_with_data(
actions[ActionTypes.UPDATE_COLLECTION_WITH_DATA]
)
self.process_retry_with_delays(actions[ActionTypes.RETRY_WITH_DELAYS])
self.process_record_in_task_history(
actions[ActionTypes.RECORD_IN_TASK_HISTORY]
)
[docs]
def get_triggered_actions(
self,
event_name: Literal[
"on_creation",
"on_unblock",
"on_assignment",
"on_success",
"on_failure",
],
) -> dict[str, list[EventReaction]]:
"""
Filter events to trigger, grouped by type.
:param event_name: the name of the event being triggered.
"""
actions: dict[str, list[EventReaction]] = {
action: [] for action in ActionTypes
}
try:
task = self.get_task()
except Exception:
task_event_reactions = []
else:
task_event_reactions = task.get_event_reactions(event_name)
for action in chain(
task_event_reactions, getattr(self.event_reactions, event_name)
):
action_type = action.action
actions[action_type].append(action)
return actions
[docs]
def process_skip_if_lookup_result_changed(
self, actions: list[EventReaction]
) -> None:
"""Skip this work request if the result of a lookup has changed."""
workflow_root = self.workflow_root
should_skip = False
for skip in actions:
assert isinstance(skip, ActionSkipIfLookupResultChanged)
try:
result = self.lookup_single(skip.lookup)
except KeyError:
result = None
except LookupError as e:
logger.error("%s in WorkRequest %s", e, self.pk) # noqa: G200
continue
new_collection_item_id = (
None
if result is None or result.collection_item is None
else result.collection_item.id
)
if new_collection_item_id == skip.collection_item_id:
continue
# Mark ourselves as complete, with a note about why.
should_skip = True
self.mark_completed(
WorkRequest.Results.SKIPPED,
output_data=OutputData(
skip_reason=f"Result of lookup {skip.lookup!r} changed"
),
)
if (
skip.promise_name is not None
and workflow_root is not None
and result is not None
and result.collection_item is not None
):
assert workflow_root.internal_collection is not None
# Use the lookup result to resolve the promise.
if result.artifact is not None:
workflow_root.internal_collection.manager.add_artifact(
result.artifact,
user=self.created_by,
workflow=self.parent,
variables=result.collection_item.data,
name=skip.promise_name,
replace=True,
)
else:
workflow_root.internal_collection.manager.add_bare_data(
BareDataCategory(result.collection_item.category),
user=self.created_by,
workflow=self.parent,
data=result.collection_item.data,
name=skip.promise_name,
replace=True,
)
if should_skip:
raise SkipWorkRequest
[docs]
def process_update_collection_with_artifacts(
self, actions: list[EventReaction]
) -> None:
"""Update collection with artifacts following event_reactions."""
# local import to avoid circular dependency
from debusine.server.collections import (
CollectionManagerInterface,
ItemAdditionError,
ItemRemovalError,
)
for update in actions:
assert isinstance(update, ActionUpdateCollectionWithArtifacts)
try:
collection = self.lookup_single(
update.collection, expect_type=LookupChildType.COLLECTION
).collection
except LookupError as e:
logger.error("%s in WorkRequest %s", e, self.pk) # noqa: G200
continue
manager = CollectionManagerInterface.get_manager_for(collection)
try:
artifacts_to_add = self.artifact_set.filter(
**update.artifact_filters
)
except FieldError:
logger.exception(
"Invalid update-collection-with-artifacts"
" artifact_filters in WorkRequest %s",
self.pk,
)
continue
for artifact in artifacts_to_add:
try:
expanded_variables = CollectionItem.expand_variables(
update.variables or {}, artifact.data
)
except (KeyError, ValueError):
logger.exception(
"Invalid update-collection-with-artifacts variables "
"in WorkRequest %s",
self.pk,
)
continue
if update.name_template is not None:
item_name = CollectionItem.expand_name(
update.name_template, expanded_variables
)
else:
item_name = None
try:
manager.add_artifact(
artifact,
user=self.created_by,
workflow=self.parent,
variables=expanded_variables,
name=item_name,
replace=True,
created_at=update.created_at,
)
except (ItemAdditionError, ItemRemovalError):
logger.exception(
"Cannot replace or add artifact %s to collection %s"
" from WorkRequest %s",
artifact,
collection,
self.pk,
)
[docs]
def process_update_collection_with_data(
self, actions: list[EventReaction]
) -> None:
"""Update collection with bare data following event_reactions."""
# local import to avoid circular dependency
from debusine.server.collections import (
CollectionManagerInterface,
ItemAdditionError,
ItemRemovalError,
)
for update in actions:
assert isinstance(update, ActionUpdateCollectionWithData)
try:
collection = self.lookup_single(
update.collection, expect_type=LookupChildType.COLLECTION
).collection
except LookupError as e:
logger.error("%s in WorkRequest %s", e, self.pk) # noqa: G200
continue
manager = CollectionManagerInterface.get_manager_for(collection)
data = update.data or {}
if update.name_template is not None:
item_name = CollectionItem.expand_name(
update.name_template, data
)
else:
item_name = None
try:
manager.add_bare_data(
update.category,
user=self.created_by,
workflow=self.parent,
data=data,
name=item_name,
replace=True,
created_at=update.created_at,
)
except (ItemAdditionError, ItemRemovalError):
logger.exception(
"Cannot replace or add bare data of category %s to"
" collection %s from WorkRequest %s",
update.category,
collection,
self.pk,
)
[docs]
@context.disable_permission_checks()
def process_retry_with_delays(self, actions: list[EventReaction]) -> None:
"""Retry a work request with delays."""
# Import here to prevent circular imports
from debusine.server.tasks.wait.models import DelayData
from debusine.server.workflows.models import WorkRequestWorkflowData
for action in actions:
assert isinstance(action, ActionRetryWithDelays)
if self.parent is None:
raise CannotRetry(
"retry-with-delays action may only be used in a workflow"
)
retry_count = self.workflow_data.retry_count
if retry_count >= len(action.delays):
continue
m = ActionRetryWithDelays._delay_re.match(
action.delays[retry_count]
)
# Checked by ActionRetryWithDelays.validate_delays.
assert m is not None
match m.group(2):
case "m":
delay = dt.timedelta(minutes=int(m.group(1)))
case "h":
delay = dt.timedelta(hours=int(m.group(1)))
case "d":
delay = dt.timedelta(days=int(m.group(1)))
case "w":
delay = dt.timedelta(weeks=int(m.group(1)))
case _ as unreachable:
raise AssertionError(
f"Unexpected delay unit: {unreachable}"
)
wr_delay = self.parent.create_child_wait(
task_name="delay",
task_data=DelayData(delay_until=timezone.now() + delay),
workflow_data=WorkRequestWorkflowData(needs_input=False),
)
wr_delay.mark_pending()
wr_retried = self.retry(reason=WorkRequestRetryReason.DELAY)
wr_retried.add_dependency(wr_delay)
[docs]
def process_record_in_task_history(
self, actions: list[EventReaction]
) -> None:
"""Record a task run in a task-history collection."""
if (
actions
and self.started_at is not None
and self.dynamic_task_data is not None
and self.output_data is not None
and self.output_data.runtime_statistics is not None
):
try:
collection = self.workspace.get_singleton_collection(
user=self.created_by,
category=CollectionCategory.TASK_HISTORY,
)
except Collection.DoesNotExist:
# No task-history collection; skip these actions.
return
default_subject: str | None = None
default_context: str | None = None
try:
task = self.get_task()
except Exception:
pass
else:
assert task.dynamic_data is not None
default_subject = task.dynamic_data.subject
default_context = task.dynamic_data.runtime_context
for action in actions:
assert isinstance(action, ActionRecordInTaskHistory)
collection.manager.add_bare_data(
BareDataCategory.HISTORICAL_TASK_RUN,
user=self.created_by,
workflow=self.parent,
data=DebusineHistoricalTaskRun(
task_type=TaskTypes(self.task_type),
task_name=self.task_name,
subject=action.subject or default_subject,
context=action.context or default_context,
timestamp=int(self.started_at.timestamp()),
work_request_id=self.id,
result=WorkRequestResults(self.result),
runtime_statistics=self.output_data.runtime_statistics,
),
)
[docs]
def mark_pending(self) -> None:
"""Worker is ready for execution."""
if self.status != self.Statuses.BLOCKED:
raise StatusChangeError(
f"Cannot mark WorkRequest {self.pk} as pending:"
f" current status is {self.status}"
)
self.pending_at = timezone.now()
self.status = self.Statuses.PENDING
if self.task_type not in (TaskTypes.INTERNAL, TaskTypes.WORKER):
# WORKER tasks require a worker to configure
# INTERNAL tasks do not support configuration
self.apply_task_configuration()
self.save()
self.schedule_workflow_update()
logger.debug("Marked WorkRequest %s as pending", self.pk)
self.process_event_reactions("on_unblock")
[docs]
def add_dependency(
self,
dependency: "WorkRequest",
*,
skip_same_workflow_check: bool = False,
) -> None:
"""
Make this work request depend on another one.
:param dependency: the work request to add as a new dependency; if
this work request is part of a workflow, then the new dependency
must be part of the same workflow.
:param skip_same_workflow_check: when adding a large number of
dependencies, checking that they are all part of the same workflow
may be slow; if this is True, the caller promises that it has
already ensured that.
"""
if not skip_same_workflow_check and self.is_part_of_workflow:
# Work requests in a workflow may only depend on other work
# requests in the same workflow.
assert self.workflow_root is not None
if dependency.workflow_root != self.workflow_root:
raise ValueError(
"Work requests in a workflow may not depend on other work "
"requests outside that workflow"
)
self.dependencies.add(dependency)
if (
self.status == WorkRequest.Statuses.PENDING
and self.unblock_strategy == WorkRequest.UnblockStrategy.DEPS
and dependency.status
not in {
WorkRequest.Statuses.COMPLETED,
WorkRequest.Statuses.ABORTED,
}
):
self.status = WorkRequest.Statuses.BLOCKED
self.save()
[docs]
def maybe_finish_workflow(self) -> bool:
"""Update workflow status if its children are no longer in progress."""
# Do nothing if this workflow's orchestrator is running, since in
# that case the workflow may not have been completely populated.
if hasattr(self, "_running_workflow_orchestrator"):
return False
# Only relevant for running workflows where all work requests have been
# either completed or aborted
if (
self.task_type != TaskTypes.WORKFLOW
or self.status != WorkRequest.Statuses.RUNNING
or self.has_children_in_progress
):
return False
# If there are aborted child requests, abort the workflow
if (
not self.workflow_data.allow_dependency_failures
and self.children.filter(superseded__isnull=True)
.exclude(
Q(status=WorkRequest.Statuses.COMPLETED)
| Q(workflow_data_json__contains={"allow_failure": True})
)
.exists()
):
self.mark_aborted()
return True
# If there are failed/errored child requests, fail the workflow
if (
not self.workflow_data.allow_dependency_failures
and self.children.filter(superseded__isnull=True)
.exclude(
Q(
result__in={
WorkRequest.Results.SUCCESS,
WorkRequest.Results.SKIPPED,
}
)
| Q(workflow_data_json__contains={"allow_failure": True})
)
.exists()
):
self.mark_completed(WorkRequest.Results.FAILURE)
return True
# All child requests succeeded
self.mark_completed(WorkRequest.Results.SUCCESS)
return True
[docs]
def can_be_unblocked(self) -> bool:
"""Return True iff this work request can be unblocked at all."""
return (
WorkRequest.objects.can_be_unblocked().filter(id=self.id).exists()
)
[docs]
def can_be_automatically_unblocked(self) -> bool:
"""Return True iff this work request can be automatically unblocked."""
return (
WorkRequest.objects.can_be_automatically_unblocked()
.filter(id=self.id)
.exists()
)
[docs]
def unblock_reverse_dependencies(self) -> None:
"""Unblock reverse dependencies."""
assert self.status in {
WorkRequest.Statuses.COMPLETED,
WorkRequest.Statuses.ABORTED,
}
automatically_unblockable = (
self.reverse_dependencies.can_be_automatically_unblocked()
)
# Look up work requests that depend on the completed work request,
# and unblock or abort them as appropriate.
for rdep in self.reverse_dependencies.filter(
status=WorkRequest.Statuses.BLOCKED
):
if (
self.result
in {WorkRequest.Results.SUCCESS, WorkRequest.Results.SKIPPED}
or self.workflow_data.allow_failure
or rdep.workflow_data.allow_dependency_failures
):
if rdep in automatically_unblockable:
rdep.mark_pending()
else:
rdep.mark_aborted()
[docs]
def unblock_workflow_children(self) -> None:
"""Unblock children of a workflow that has just started running."""
for child in self.children.can_be_automatically_unblocked():
child.mark_pending()
[docs]
def review_manual_unblock(
self,
*,
user: "User",
notes: str,
action: "WorkRequestManualUnblockAction | None",
) -> None:
"""Review a work request awaiting manual unblocking."""
# Import here to prevent circular imports
from debusine.server.workflows.models import (
WorkRequestManualUnblockAction,
WorkRequestManualUnblockData,
WorkRequestManualUnblockLog,
)
enforce(self.can_unblock)
if self.unblock_strategy != WorkRequest.UnblockStrategy.MANUAL:
raise CannotUnblock(
f"Work request {self.id} cannot be manually unblocked"
)
if (
not self.is_part_of_workflow
or (workflow_data := self.workflow_data) is None
):
raise CannotUnblock(
f"Work request {self.id} is not part of a workflow"
)
if not self.can_be_unblocked():
raise CannotUnblock(f"Work request {self.id} cannot be unblocked")
if workflow_data.manual_unblock is None:
workflow_data.manual_unblock = WorkRequestManualUnblockData()
manual_unblock_log = list(workflow_data.manual_unblock.log)
manual_unblock_log.append(
WorkRequestManualUnblockLog(
user_id=user.id,
timestamp=timezone.now(),
notes=notes,
action=action,
)
)
workflow_data.manual_unblock.log = manual_unblock_log
self.workflow_data = workflow_data
self.save()
match action:
case WorkRequestManualUnblockAction.ACCEPT:
# Always succeeds, since we already checked that the current
# status is BLOCKED (via `work_request.can_be_unblocked()`).
self.mark_pending()
case WorkRequestManualUnblockAction.REJECT:
# The current implementation always aborts the work request.
self.mark_aborted()
@property
def requires_signature(self) -> bool:
"""True if this task is blocked on `debusine provide-signature`."""
# TODO: At some point it may be worth having this delegate to a
# property on the task, but for now this is good enough.
return (
self.task_type == TaskTypes.WAIT
and self.task_name == "externaldebsign"
and self.status == self.Statuses.RUNNING
)
@property
def requires_confirmation(self) -> bool:
"""True if this task is blocked on user confirmation."""
# TODO: At some point it may be worth having this delegate to a
# property on the task, but for now this is good enough.
return (
self.task_type == TaskTypes.WAIT
and self.task_name == "confirm"
and self.status == self.Statuses.RUNNING
)
[docs]
def mark_aborted(
self,
*,
user: "User | None" = None,
output_data: OutputData | None = None,
) -> None:
"""
Worker has aborted the task after request from UI.
Task will typically be in CREATED or RUNNING status.
"""
self.aborted_by = user
self.completed_at = timezone.now()
previous_status = self.status
self.status = self.Statuses.ABORTED
if output_data is not None:
if self.output_data is None:
self.output_data = output_data
else:
self.output_data = self.output_data.merge(output_data)
self.save()
# TODO: This may cascade to other work requests in the same
# workflow, so WorkRequestQuerySet.lock_workflows_for_update must
# have been called first in this transaction. We should check this.
# https://salsa.debian.org/freexian-team/debusine/-/merge_requests/2381#note_689749
self.unblock_reverse_dependencies()
if self.parent is not None:
self.parent.maybe_finish_workflow()
self.schedule_workflow_update()
logger.debug(
"Marked WorkRequest %s as aborted (from status %s)",
self.pk,
previous_status,
)
[docs]
def apply_task_configuration(
self, worker: Optional["Worker"] = None
) -> None:
"""Apply task configuration, filling configured_task_data."""
task = self.get_task(worker=worker)
task.apply_task_configuration()
[docs]
def assign_worker(self, worker: "Worker") -> None:
"""Assign worker and save it."""
self.configure_for_worker(worker)
self.worker = worker
self.save()
notifications.notify_work_request_assigned(self)
self.process_event_reactions("on_assignment")
[docs]
def de_assign_worker(self) -> bool:
"""
De-assign a worker from this work request.
Only allowed if the status is RUNNING or PENDING.
"""
if self.status not in {
WorkRequest.Statuses.RUNNING,
WorkRequest.Statuses.PENDING,
}:
logger.warning(
"WorkRequest %d cannot be de-assigned: current status: %s",
self.pk,
self.status,
)
return False
self.worker = None
self.started_at = None
self.status = WorkRequest.Statuses.PENDING
self.save()
return True
@property
def duration(self) -> float | None:
"""Return duration in seconds between started_at and completed_at."""
if self.started_at and self.completed_at:
return (self.completed_at - self.started_at).total_seconds()
else:
return None
@property
def priority_effective(self) -> int:
"""The effective priority of this work request."""
return self.priority_base + self.priority_adjustment
@property
def priority_for_metrics(self) -> str:
"""A low-cardinality label for the effective priority."""
if self.priority_effective >= MIN_PRIORITY_NORMAL:
return "normal"
elif self.priority_effective >= MIN_PRIORITY_BACKGROUND:
return "background"
else:
return "indefinite"
@property
def is_workflow(self) -> bool:
"""Return whether this work request is a workflow."""
return self.task_type == TaskTypes.WORKFLOW
@property
def is_part_of_workflow(self) -> bool:
"""Return whether this work request is part of a workflow."""
return self.parent is not None
@property
def is_workflow_root(self) -> bool:
"""Return whether this work request is a root workflow."""
return self.is_workflow and self.parent is None
@property
def workflow_display_name(self) -> str:
"""Return this work request's name for display in a workflow."""
if (
self.workflow_data is not None
and self.workflow_data.display_name is not None
):
return self.workflow_data.display_name
else:
return self.task_name
@property
def workflow_display_name_parameters(self) -> str | None:
"""Return this workflow template name with useful information."""
assert self.is_workflow
if (template_name := self.workflow_data.workflow_template_name) is None:
return None
if (
self.dynamic_task_data
and (
parameter_summary := self.dynamic_task_data.get(
"parameter_summary"
)
)
is not None
):
return f"{template_name}({parameter_summary})"
else:
return template_name
[docs]
def schedule_workflow_update(self) -> None:
"""
Schedule updates of expensive properties of containing workflows.
The ``workflow_last_activity_at`` and ``workflow_runtime_status``
fields of workflows are used in the web UI and need to be kept up to
date, but updating them requires scanning the workflow graph and can
be expensive. When making changes to a large workflow graph, much
of this work can be amortized: for example, we only need to update
the root workflow once even if many work requests under it have
changed. We therefore defer that work until the end of the
transaction and run it in a Celery task.
"""
from debusine.server.celery import update_workflows
if self.parent:
self.workflows_need_update = True
self.save()
# Run the update on commit, but only once per savepoint. This
# uses undocumented Django internals, but if it goes wrong
# without an obvious unit test failure, then the worst case
# should be some extra workflow updates.
if (set(connection.savepoint_ids), update_workflows.delay) not in [
hook[:2] for hook in connection.run_on_commit
]:
transaction.on_commit(update_workflows.delay)
@property
def has_children_in_progress(self) -> bool:
"""Return whether child work requests are still in progress."""
return self.children.exclude(
status__in={
WorkRequest.Statuses.COMPLETED,
WorkRequest.Statuses.ABORTED,
}
).exists()
[docs]
def effective_expiration_delay(self) -> dt.timedelta:
"""Return expiration_delay, inherited if None."""
expiration_delay = self.expiration_delay
if self.expiration_delay is None: # inherit
expiration_delay = self.workspace.default_expiration_delay
assert expiration_delay is not None
return expiration_delay
@property
def expire_at(self) -> dt.datetime | None:
"""Return computed expiration date."""
delay = self.effective_expiration_delay()
if delay == dt.timedelta(0):
return None
return self.created_at + delay
[docs]
def get_label(self) -> str:
"""
Return a label for this work request: "{name}: {parameter_summary}".
Where "parameter_summary" is optional.
- "name":
* workflow: `workflow_template_name` else `task_name`
* task: `task_name`
"""
name = None
if self.is_workflow:
name = self.workflow_data.workflow_template_name
if name is None:
name = self.task_name
if self.dynamic_task_data is None:
return name
dynamic_task_data = _DynamicTaskDataExtraFields(
**self.dynamic_task_data
)
if dynamic_task_data.parameter_summary is None:
return name
return f"{name}: {dynamic_task_data.parameter_summary}"
#: Regexp matching the structure of notification channel names
notification_channel_name_regex = re.compile(r"^[A-Za-z][A-Za-z0-9+._-]*$")
def is_valid_notification_channel_name(value: str) -> bool:
"""Check if value is a valid scope name."""
return bool(notification_channel_name_regex.match(value))
def validate_notification_channel_name(value: str) -> None:
"""Validate notification channel names."""
if not is_valid_notification_channel_name(value):
raise ValidationError(
"%(value)r is not a valid notification channel name",
params={"value": value},
)
[docs]
class NotificationChannel(models.Model):
"""Model to store notification configuration."""
[docs]
class Methods(models.TextChoices):
EMAIL = "email", "Email"
data_validators = {Methods.EMAIL: NotificationDataEmail}
name = models.CharField(
max_length=20,
unique=True,
validators=[validate_notification_channel_name],
)
method = models.CharField(max_length=10, choices=Methods.choices)
data = models.JSONField(default=dict, blank=True)
def __str__(self) -> str:
"""Return name."""
return self.name
[docs]
def save(self, *args: Any, **kwargs: Any) -> None:
"""Run validators and save the instance."""
self.full_clean()
return super().save(*args, **kwargs)
[docs]
def clean(self) -> None:
"""
Ensure that data is valid for the specific method.
:raise ValidationError: for invalid data.
"""
try:
self.data_validators[self.Methods(self.method)](**self.data)
except (TypeError, ValueError) as exc:
raise ValidationError(
f"NotificationChannel data is not valid: {exc}"
)
return super().clean()
def compute_workflow_last_activity(workflow: WorkRequest) -> dt.datetime | None:
"""
Compute the last activity timestamp of a workflow.
This is the latest relevant timestamp of its descendants.
"""
last_activity = (
WorkRequest.objects.filter(id=workflow.id)
.workflow_flattened(include_invisible=True)
.aggregate(
last_activity=Max(
Greatest("pending_at", "started_at", "completed_at"),
output_field=DateTimeField(blank=True, null=True),
)
)["last_activity"]
)
assert last_activity is None or isinstance(last_activity, dt.datetime)
return last_activity
def compute_workflow_runtime_status(
workflow: WorkRequest,
) -> WorkRequest.RuntimeStatuses:
"""
Compute the runtime status of the workflow.
This is recursively computed based on the status of all
the work requests in the workflow and its sub-workflows's work requests.
"""
assert workflow.task_type == TaskTypes.WORKFLOW
flattened = WorkRequest.objects.filter(id=workflow.id).workflow_flattened(
include_invisible=True
)
if flattened.filter(
task_type=TaskTypes.WAIT,
status=WorkRequest.Statuses.RUNNING,
workflow_data_json__needs_input=True,
).exists():
return _RuntimeStatuses.NEEDS_INPUT
if (
flattened.exclude(task_type=TaskTypes.WAIT)
.filter(status=WorkRequest.Statuses.RUNNING)
.exists()
):
return _RuntimeStatuses.RUNNING
if flattened.filter(
task_type=TaskTypes.WAIT,
status=WorkRequest.Statuses.RUNNING,
workflow_data_json__needs_input=False,
).exists():
return _RuntimeStatuses.WAITING
if flattened.filter(status=WorkRequest.Statuses.PENDING).exists():
return _RuntimeStatuses.PENDING
if flattened.filter(status=WorkRequest.Statuses.BLOCKED).exists():
return _RuntimeStatuses.BLOCKED
if flattened.filter(status=WorkRequest.Statuses.ABORTED).exists():
return _RuntimeStatuses.ABORTED
if not flattened.exclude(status=WorkRequest.Statuses.COMPLETED).exists():
return _RuntimeStatuses.COMPLETED
# Any possibility should have been dealt with
raise AssertionError(
f"Could not compute runtime status for workflow {workflow}"
)
def workflow_ancestors(
work_requests: QuerySet[WorkRequest],
) -> QuerySet[WorkRequest]:
"""
Return workflows containing any of these work requests.
This returns any workflows that are one of the given work requests, or
direct or indirect ancestors.
"""
work_request_queryset = CTEQuerySet(
WorkRequest, using=WorkRequest.objects._db
)
def workflow_ancestors_cte(cte: With) -> QuerySet[WorkRequest]:
recursive = (
work_request_queryset.filter(id__in=work_requests)
.values(rec_parent=F("parent"), rec_child=F("id"))
.union(
cte.join(WorkRequest, id=cte.col.rec_parent).values(
rec_parent=F("parent"), rec_child=F("id")
)
)
)
assert isinstance(recursive, QuerySet)
return recursive
cte = With.recursive(workflow_ancestors_cte)
ancestors = (
cte.join(work_request_queryset, id=cte.col.rec_child)
.with_cte(cte)
.distinct("id")
.filter(task_type=TaskTypes.WORKFLOW)
)
assert isinstance(ancestors, QuerySet)
return ancestors
class DeleteUnused:
"""
Delete work requests not currently in use.
Delete work requests that don't have unexpired child work requests.
Delete also the work requests' internal collection if present.
"""
#: Work requests to delete
work_requests: set[WorkRequest]
#: Internal collections to delete
collections: set["Collection"]
def __init__(self, queryset: QuerySet[WorkRequest]) -> None:
"""
Initialize object.
:param work_requests: candidate work_requests to delete
"""
self.queryset = queryset
self.scan_work_requests()
self.scan_collections()
def scan_work_requests(self) -> None:
"""Look for work requests to delete, filling self.work_requests."""
# to_delete: expired work requests excluding the ones referenced
# by a CollectionItem.created_by_workflow/removed_by_workflow
# that are not in the CollectionItem.parent_collection
#
# In other words: if a WorkRequest is referenced by a
# CollectionItem.created_by_workflow/removed_by_workflow but the
# CollectionItem.parent_collection is WorkRequest.internal_collection
# the WorkRequest can be deleted.
to_delete = set(
self.queryset.exclude(
Exists(
# See https://stackoverflow.com/questions/67775006/django-queryset-matching-null-value-with-null-outerref # noqa: E501
# and https://code.djangoproject.com/ticket/31714
CollectionItem.objects.annotate(
internal_collection_id=ExpressionWrapper(
OuterRef("internal_collection_id"),
output_field=IntegerField(),
)
).filter(
(
Q(created_by_workflow_id=OuterRef("id"))
| Q(removed_by_workflow_id=OuterRef("id"))
)
& (
Q(internal_collection_id__isnull=True)
| ~Q(
parent_collection_id=F("internal_collection_id")
)
)
)
)
)
)
# Keep work requests whose parent is not in the queryset
to_keep: set[WorkRequest]
while to_delete:
to_keep = set()
for wr in to_delete:
if wr.parent is not None and wr.parent not in to_delete:
to_keep.add(wr)
to_delete -= to_keep
if not to_keep:
break
# Keep work requests that have children not in the queryset
# Precache children of all nodes in to_delete
children = defaultdict(list)
for child in WorkRequest.objects.filter(
parent__in=to_delete
).select_related("parent"):
children[child.parent].append(child)
while to_delete:
to_keep = set()
for wr in to_delete:
for child in children[wr]:
if child not in to_delete:
to_keep.add(wr)
break
to_delete -= to_keep
if not to_keep:
break
self.work_requests = to_delete
def scan_collections(self) -> None:
"""Look for collections to delete, filling self.collections."""
collections: set[Collection] = set()
for wr in self.work_requests:
if wr.internal_collection:
collections.add(wr.internal_collection)
self.collections = collections
def perform_deletions(self) -> None:
"""Delete objects found in self.work_requests and self.collections."""
CollectionItem.objects.filter(
parent_collection__in=self.collections
).delete()
WorkRequest.objects.filter(
id__in=[wr.id for wr in self.work_requests]
).delete()
Collection.objects.filter(
id__in=[c.id for c in self.collections]
).delete()