Added /prune/ endpoint to removing "old" RPMs from a Repository.

closes #2909.

CHANGES/2909.feature

@@ -0,0 +1 @@
+Added /rpm/prune command to allow "pruning" old Packages from repositories.

pulp_rpm/app/serializers/__init__.py

@@ -27,6 +27,7 @@
     ModulemdObsoleteSerializer,
 )
 from .package import PackageSerializer, MinimalPackageSerializer  # noqa
+from .prune import PruneNEVRAsSerializer  # noqa
 from .repository import (  # noqa
     CopySerializer,
     RpmDistributionSerializer,

pulp_rpm/app/serializers/prune.py

@@ -0,0 +1,81 @@
+from gettext import gettext as _
+
+from rest_framework import fields, serializers
+
+from pulp_rpm.app.models import RpmRepository
+
+from pulpcore.plugin.serializers import ValidateFieldsMixin
+from pulpcore.plugin.util import get_domain
+
+
+class PruneNEVRAsSerializer(serializers.Serializer, ValidateFieldsMixin):
+    """
+    Serializer for prune-old-NEVRAs operation.
+    """
+
+    repo_hrefs = fields.ListField(
+        required=True,
+        help_text=_(
+            "Will prune old RPMs from the specified list of repos. Use ['*'] to specify all repos."
+        ),
+        child=serializers.CharField(),
+    )
+
+    keep_days = serializers.IntegerField(
+        help_text=_(
+            "Prune NEVRAs introduced *prior-to* this many days ago. "
+            "Default is 14. A value of 0 implies 'keep latest NEVRA only.'"
+        ),
+        required=False,
+        min_value=0,
+        default=14,
+    )
+
+    repo_concurrency = serializers.IntegerField(
+        help_text=(
+            "Number of concurrent workers to use to do the pruning. "
+            "If not set then the default value will be used."
+        ),
+        allow_null=True,
+        required=False,
+        min_value=1,
+        default=10,
+    )
+
+    dry_run = serializers.BooleanField(
+        help_text=_(
+            "Determine what would-be-pruned and log the list of NEVRAs. "
+            "Intended as a debugging aid."
+        ),
+        default=False,
+        required=False,
+    )
+
+    def validate_repo_hrefs(self, value):
+        """
+        Check that repo_hrefs is not an empty list and contains either valid hrefs or "*".
+        Args:
+            value (list): The list supplied by the user
+        Returns:
+            The list of RpmRepositories after validation
+        Raises:
+            ValidationError: If the list is empty or contains invalid hrefs.
+        """
+        if len(value) == 0:
+            raise serializers.ValidationError("Must not be [].")
+
+        # prune-all-repos is "*" - find all RPM repos in this domain
+        if "*" in value:
+            if len(value) != 1:
+                raise serializers.ValidationError("Can't specify specific HREFs when using '*'")
+            return RpmRepository.objects.filter(pulp_domain=get_domain())
+
+        from pulpcore.plugin.viewsets import NamedModelViewSet
+
+        # We're pruning a specific list of RPM repositories.
+        # Validate that they are for RpmRepositories.
+        hrefs_to_return = []
+        for href in value:
+            hrefs_to_return.append(NamedModelViewSet.get_resource(href, RpmRepository))
+
+        return hrefs_to_return

pulp_rpm/app/tasks/__init__.py

@@ -2,3 +2,4 @@
 from .synchronizing import synchronize  # noqa
 from .copy import copy_content  # noqa
 from .comps import upload_comps  # noqa
+from .prune import prune_nevras  # noqa

pulp_rpm/app/tasks/prune.py

@@ -0,0 +1,156 @@
+from datetime import datetime, timedelta
+from logging import getLogger, DEBUG
+
+from django.db.models import F, Subquery
+from django.utils import timezone
+
+from pulpcore.plugin.models import ProgressReport
+from pulpcore.plugin.constants import TASK_STATES
+from pulpcore.plugin.models import (
+    GroupProgressReport,
+    RepositoryContent,
+    TaskGroup,
+)
+from pulpcore.plugin.tasking import dispatch
+from pulp_rpm.app.models.package import Package
+from pulp_rpm.app.models.repository import RpmRepository
+
+log = getLogger(__name__)
+
+
+def prune_repo_nevras(repo_pk, keep_days, dry_run):
+    """
+    This task prunes old NEVRAs from the latest_version of the specified repository.
+
+    Args:
+        repo_pk (UUID): UUID of the RpmRepository to be pruned.
+        keep_days(int): Keep RepositoryContent created less than this many days ago.
+        dry_run (boolean): If True, don't actually do the prune, just log to-be-pruned NEVRAs.
+    """
+    repo = RpmRepository.objects.filter(pk=repo_pk).get()
+    curr_vers = repo.latest_version()
+    eldest_datetime = datetime.now(tz=timezone.utc) - timedelta(days=keep_days)
+    log.info(f"PRUNING REPOSITORY {repo.name}.")
+    log.debug(f">>> TOTAL RPMS: {curr_vers.get_content(Package.objects).count()}")
+
+    # We only care about RPM-Names that have more than one EVRA - "singles" are always kept.
+    rpm_by_name_age = (
+        curr_vers.get_content(Package.objects.with_age())
+        .filter(age__gt=1)
+        .order_by("name", "epoch", "version", "release", "arch")
+        .values("pk")
+    )
+    log.debug(f">>> NAME/AGE COUNT {rpm_by_name_age.count()}")
+    log.debug(
+        ">>> # NAME/ARCH w/ MULTIPLE EVRs: {}".format(
+            curr_vers.get_content(Package.objects)
+            .filter(pk__in=rpm_by_name_age)
+            .values("name", "arch")
+            .distinct()
+            .count()
+        )
+    )
+    log.debug(
+        ">>> # UNIQUE NAME/ARCHS: {}".format(
+            curr_vers.get_content(Package.objects).values("name", "arch").distinct().count()
+        )
+    )
+
+    # Find the RepositoryContents associated with the multi-EVR-names from above,
+    # whose maximum-pulp-created date is LESS THAN eldest_datetime.
+    #
+    # Note that we can "assume" the latest-date is an "add" with no "remove", since we're
+    # limiting ourselves to the list of ids that we know are in the repo's current latest-version!
+    target_ids_q = (
+        RepositoryContent.objects.filter(
+            content__in=Subquery(rpm_by_name_age), repository=repo, version_removed=None
+        )
+        .filter(pulp_created__lt=eldest_datetime)
+        .values("content_id")
+    )
+    log.debug(f">>> TARGET IDS: {target_ids_q.count()}.")
+    to_be_removed = target_ids_q.count()
+    # Use the progressreport to report back numbers. The prune happens as one
+    # action.
+    data = dict(
+        message=f"Pruning {repo.name}",
+        code="rpm.nevra.prune.repository",
+        total=to_be_removed,
+        state=TASK_STATES.COMPLETED,
+        done=0,
+    )
+
+    if dry_run:
+        if log.getEffectiveLevel() == DEBUG:  # Don't go through the loop unless debugging
+            log.debug(">>> Packages to be removed : ")
+            for p in (
+                Package.objects.filter(pk__in=target_ids_q)
+                .order_by("name", "epoch", "version", "release", "arch")
+                .values("name", "epoch", "version", "release", "arch")
+            ):
+                log.debug(f'{p["name"]}-{p["epoch"]}:{p["version"]}-{p["release"]}.{p["arch"]}')
+    else:
+        with repo.new_version(base_version=None) as new_version:
+            new_version.remove_content(target_ids_q)
+        data["done"] = to_be_removed
+
+    pb = ProgressReport(**data)
+    pb.save()
+
+    # Report back that this repo has completed.
+    gpr = TaskGroup.current().group_progress_reports.filter(code="rpm.nevra.prune")
+    gpr.update(done=F("done") + 1)
+
+
+def prune_nevras(
+    repo_pks,
+    keep_days=14,
+    repo_concurrency=10,
+    dry_run=False,
+):
+    """
+    This task prunes old NEVRAs from the latest_version of the specified list of repos.
+
+    "Old" in this context is defined by the RepositoryContent record that added a NEVRA
+    to the repository in question.
+
+    It will issue one task-per-repository.
+
+    Kwargs:
+        repo_pks (list): A list of repo pks the disk reclaim space is performed on.
+        keep_days(int): Keep RepositoryContent created less than this many days ago.
+        repo_concurrency (int): number of repos to prune at a time.
+        dry_run (boolean): If True, don't actually do the prune, just record to-be-pruned NEVRAs.
+    """
+
+    repos_to_prune = RpmRepository.objects.filter(pk__in=repo_pks)
+    task_group = TaskGroup.current()
+
+    gpr = GroupProgressReport(
+        message="Pruning old NEVRAs",
+        code="rpm.nevra.prune",
+        total=len(repo_pks),
+        done=0,
+        task_group=task_group,
+    )
+    gpr.save()
+
+    # Dispatch a task-per-repository.
+    # Lock on the the repository *and* to insure the max-concurrency specified.
+    # This will keep an "all repositories" prune from locking up all the workers
+    # until all repositories are completed.
+    for index, a_repo in enumerate(repos_to_prune):
+        worker_rsrc = f"rpm-prune-worker-{index % repo_concurrency}"
+        exclusive_resources = [worker_rsrc, a_repo]
+
+        dispatch(
+            prune_repo_nevras,
+            exclusive_resources=exclusive_resources,
+            args=(
+                a_repo.pk,
+                keep_days,
+                dry_run,
+            ),
+            task_group=task_group,
+        )
+    task_group.finish()

pulp_rpm/app/urls.py

@@ -1,7 +1,7 @@
 from django.conf import settings
 from django.urls import path
 
-from .viewsets import CopyViewSet, CompsXmlViewSet
+from .viewsets import CopyViewSet, CompsXmlViewSet, PruneNEVRAsViewSet
 
 if settings.DOMAIN_ENABLED:
     V3_API_ROOT = settings.V3_DOMAIN_API_ROOT_NO_FRONT_SLASH
@@ -11,4 +11,5 @@
 urlpatterns = [
     path(f"{V3_API_ROOT}rpm/copy/", CopyViewSet.as_view({"post": "create"})),
     path(f"{V3_API_ROOT}rpm/comps/", CompsXmlViewSet.as_view({"post": "create"})),
+    path(f"{V3_API_ROOT}rpm/prune/", PruneNEVRAsViewSet.as_view({"post": "prune_nevras"})),
 ]

pulp_rpm/app/viewsets/__init__.py

@@ -11,6 +11,7 @@
 from .distribution import DistributionTreeViewSet  # noqa
 from .modulemd import ModulemdViewSet, ModulemdDefaultsViewSet, ModulemdObsoleteViewSet  # noqa
 from .package import PackageViewSet  # noqa
+from .prune import PruneNEVRAsViewSet  # noqa
 from .repository import (  # noqa
     RpmRepositoryViewSet,
     RpmRepositoryVersionViewSet,

pulp_rpm/app/viewsets/prune.py

@@ -0,0 +1,73 @@
+from drf_spectacular.utils import extend_schema
+from django.conf import settings
+from rest_framework.viewsets import ViewSet
+
+from pulpcore.plugin.viewsets import TaskGroupOperationResponse
+from pulpcore.plugin.models import TaskGroup
+from pulpcore.plugin.serializers import TaskGroupOperationResponseSerializer
+from pulp_rpm.app.serializers import PruneNEVRAsSerializer
+from pulp_rpm.app.tasks import prune_nevras
+from pulpcore.plugin.tasking import dispatch
+
+
+class PruneNEVRAsViewSet(ViewSet):
+    """
+    Viewset for prune-old-NEVRAs endpoint.
+    """
+
+    serializer_class = PruneNEVRAsSerializer
+
+    DEFAULT_ACCESS_POLICY = {
+        "statements": [
+            {
+                "action": ["prune_nevras"],
+                "principal": "authenticated",
+                "effect": "allow",
+                "condition": [
+                    "has_repository_model_or_domain_or_obj_perms:rpm.modify_content_rpmrepository",
+                    "has_repository_model_or_domain_or_obj_perms:rpm.view_rpmrepository",
+                ],
+            },
+        ],
+    }
+
+    @extend_schema(
+        description="Trigger an asynchronous old-NEVRA-prune operation.",
+        responses={202: TaskGroupOperationResponseSerializer},
+    )
+    def prune_nevras(self, request):
+        """
+        Triggers an asynchronous old-NEVRA-purge operation.
+
+        This returns a task-group that contains a "master" task that dispatches one task
+        per repo being pruned. This allows repositories to become available for other
+        processing as soon as their task completes, rather than having to wait for *all*
+        repositories to be pruned.
+        """
+        serializer = PruneNEVRAsSerializer(data=request.data)
+        serializer.is_valid(raise_exception=True)
+
+        repos = serializer.validated_data.get("repo_hrefs", [])
+        repos_to_prune_pks = []
+        for repo in repos:
+            repos_to_prune_pks.append(repo.pk)
+
+        uri = "/api/v3/rpm/prune/"
+        if settings.DOMAIN_ENABLED:
+            uri = f"/{request.pulp_domain.name}{uri}"
+        exclusive_resources = [uri, f"pdrn:{request.pulp_domain.pulp_id}:rpm:prune"]
+
+        task_group = TaskGroup.objects.create(description="Prune old NEVRAs.")
+
+        dispatch(
+            prune_nevras,
+            exclusive_resources=exclusive_resources,
+            task_group=task_group,
+            kwargs={
+                "repo_pks": repos_to_prune_pks,
+                "keep_days": serializer.validated_data["keep_days"],
+                "repo_concurrency": serializer.validated_data["repo_concurrency"],
+                "dry_run": serializer.validated_data["dry_run"],
+            },
+        )
+        return TaskGroupOperationResponse(task_group, request)

pulp_rpm/tests/conftest.py

@@ -10,6 +10,7 @@
     RemotesRpmApi,
     RepositoriesRpmApi,
     RepositoriesRpmVersionsApi,
+    RpmPruneApi,
     RpmRepositorySyncURL,
 )
 
@@ -36,6 +37,12 @@ def rpm_distribution_api(rpm_client):
     return DistributionsRpmApi(rpm_client)
 
 
+@pytest.fixture(scope="session")
+def rpm_prune_api(rpm_client):
+    """Fixture for RPM Prune API."""
+    return RpmPruneApi(rpm_client)
+
+
 @pytest.fixture(scope="session")
 def rpm_client(bindings_cfg):
     """Fixture for RPM client."""