Add safe_in() utility to prevent PostgreSQL 65K parameter limit errors

PostgreSQL's wire protocol limits bind parameters to 65,535 per
statement.  When Django ORM's filter(field__in=python_list) generates
WHERE field IN ($1, $2, ..., $65536+), it exceeds this limit when
using server-side cursors (.iterator()).

This introduces a safe_in() utility that uses a custom Django lookup
(= ANY(%s)) for large lists, passing the entire list as a single
PostgreSQL array parameter regardless of size.  For small lists, the
standard __in lookup is used unchanged.

Applied safe_in() to all vulnerable code paths in pulpcore:
- RepositoryVersion.get_content(), added(), removed()
- import_repository_version() content mapping

Also updated the test to use .iterator() so it reliably exercises the
server-side cursor path that triggers the parameter limit.

Assisted-By: claude-opus-4.6

Author: dralley

CHANGES/+safe-in-parameter-limit.bugfix

@@ -0,0 +1 @@
+Avoid exceeding PostgreSQL's 65,535 query parameter limit when filtering by large lists of IDs. This fixes `OperationalError` crashes during large import and copy operations involving more than 65,535 content units.

CHANGES/plugin_api/+safe-in-parameter-limit.feature

@@ -0,0 +1 @@
+Added `safe_in()` to the plugin API for building `Q` objects that are safe for arbitrarily large value lists, avoiding PostgreSQL's 65,535 query parameter limit.

pulpcore/app/models/repository.py

@@ -25,6 +25,7 @@
     get_prn,
     get_view_name_for_model,
     reverse,
+    safe_in,
 )
 from pulpcore.cache import Cache
 from pulpcore.constants import ALL_KNOWN_CONTENT_CHECKSUMS, PROTECTED_REPO_VERSION_MESSAGE
@@ -981,7 +982,7 @@ def get_content(self, content_qs=None):
         Args:
             content_qs (django.db.models.QuerySet): The queryset for Content that will be
                 restricted further to the content present in this repository version. If not given,
-                ``Content.objects.all()`` is used (to return over all content types present in the
+                `Content.objects.all()` is used (to return over all content types present in the
                 repository version).
 
         Returns:
@@ -997,15 +998,7 @@ def get_content(self, content_qs=None):
         if content_qs is None:
             content_qs = Content.objects
 
-        content_ids = self.content_ids
-        if len(content_ids) >= 65535:
-            # Workaround for PostgreSQL's limit on the number of parameters in a query
-            content_ids = (
-                RepositoryVersion.objects.filter(pk=self.pk)
-                .annotate(cids=Func(F("content_ids"), function="unnest"))
-                .values_list("cids", flat=True)
-            )
-        return content_qs.filter(pk__in=content_ids)
+        return content_qs.filter(safe_in("pk", self.content_ids))
 
     @property
     def content(self):
@@ -1049,14 +1042,14 @@ def content_batch_qs(self, content_qs=None, order_by_params=("pk",), batch_size=
         Args:
             content_qs (django.db.models.QuerySet) The queryset for Content that will be
                 restricted further to the content present in this repository version. If not given,
-                ``Content.objects.all()`` is used (to iterate over all content present in the
+                `Content.objects.all()` is used (to iterate over all content present in the
                 repository version). A plugin may want to use a specific subclass of
-                [pulpcore.plugin.models.Content][] or use e.g. ``filter()`` to select
+                [pulpcore.plugin.models.Content][] or use e.g. `filter()` to select
                 a subset of the repository version's content.
-            order_by_params (tuple of str): The parameters for the ``order_by`` clause
-                for the content. The Default is ``("pk",)``. This needs to
+            order_by_params (tuple of str): The parameters for the `order_by` clause
+                for the content. The Default is `("pk",)`. This needs to
                 specify a stable order. For example, if you want to iterate by
-                decreasing creation time stamps use ``("-pulp_created", "pk")`` to
+                decreasing creation time stamps use `("-pulp_created", "pk")` to
                 ensure that content records are still sorted by primary key even
                 if their creation timestamp happens to be equal.
             batch_size (int): The maximum batch size.
@@ -1065,8 +1058,8 @@ def content_batch_qs(self, content_qs=None, order_by_params=("pk",), batch_size=
             [django.db.models.QuerySet][]: A QuerySet representing a slice of the content.
 
         Example:
-            The following code could be used to loop over all ``FileContent`` in
-            ``repository_version``. It prefetches the related
+            The following code could be used to loop over all `FileContent` in
+            `repository_version`. It prefetches the related
             [pulpcore.plugin.models.ContentArtifact][] instances for every batch::
 
                 repository_version = ...
@@ -1119,8 +1112,8 @@ def added(self, base_version=None):
         if not base_version:
             return Content.objects.filter(version_memberships__version_added=self)
 
-        return Content.objects.filter(pk__in=self.content_ids).exclude(
-            pk__in=base_version.content_ids
+        return Content.objects.filter(safe_in("pk", self.content_ids)).exclude(
+            safe_in("pk", base_version.content_ids)
         )
 
     def removed(self, base_version=None):
@@ -1134,8 +1127,8 @@ def removed(self, base_version=None):
         if not base_version:
             return Content.objects.filter(version_memberships__version_removed=self)
 
-        return Content.objects.filter(pk__in=base_version.content_ids).exclude(
-            pk__in=self.content_ids
+        return Content.objects.filter(safe_in("pk", base_version.content_ids)).exclude(
+            safe_in("pk", self.content_ids)
         )
 
     def contains(self, content):

pulpcore/app/tasks/importer.py

@@ -39,6 +39,7 @@
     compute_file_hash,
     get_domain,
     get_domain_pk,
+    safe_in,
 )
 from pulpcore.constants import TASK_STATES
 from pulpcore.exceptions.plugin import MissingPlugin
@@ -417,14 +418,14 @@ def import_repository_version(
             for repo_name, content_ids in mapping.items():
                 repo_name = _get_destination_repo_name(importer, repo_name)
                 dest_repo = Repository.objects.get(name=repo_name)
-                content = Content.objects.filter(upstream_id__in=content_ids)
+                content = Content.objects.filter(safe_in("upstream_id", content_ids))
                 content_count += len(content_ids)
                 with dest_repo.new_version() as new_version:
                     new_version.set_content(content)
         else:
             # just map all the content to our destination repo
             dest_repo = Repository.objects.get(pk=dest_repo_pk)
-            content = Content.objects.filter(pk__in=resulting_content_ids)
+            content = Content.objects.filter(safe_in("pk", resulting_content_ids))
             content_count += len(resulting_content_ids)
             with dest_repo.new_version() as new_version:
                 new_version.set_content(content)

pulpcore/app/util.py

@@ -16,7 +16,7 @@
 from django.apps import apps
 from django.conf import settings
 from django.db import connection
-from django.db.models import Model, UUIDField
+from django.db.models import Field, Lookup, Model, Q, UUIDField
 from rest_framework.reverse import reverse as drf_reverse
 from rest_framework.serializers import ValidationError
 
@@ -26,6 +26,47 @@
 from pulpcore.app.loggers import deprecation_logger
 from pulpcore.exceptions.validation import InvalidSignatureError
 
+POSTGRES_MAX_QUERY_PARAMS = 65535
+
+
+class AnyArray(Lookup):
+    """PostgreSQL `= ANY(%s)` lookup that passes a list as a single array parameter.
+
+    psycopg3 adapts the Python list into a PostgreSQL array, so the entire list
+    counts as **one** bind parameter regardless of size.  This avoids the
+    protocol-level 65535-parameter limit that `IN ($1, $2, …)` hits.
+    """
+
+    lookup_name = "any_array"
+
+    def get_prep_lookup(self):
+        return [self.lhs.output_field.get_prep_value(v) for v in self.rhs]
+
+    def as_sql(self, compiler, connection):
+        lhs, lhs_params = self.process_lhs(compiler, connection)
+        return f"{lhs} = ANY(%s)", lhs_params + [list(self.rhs)]
+
+
+Field.register_lookup(AnyArray)
+
+
+def safe_in(field_name, values):
+    """Build a `Q` object for `field__in` that is safe for arbitrarily large lists.
+
+    * If *values* is already a queryset (or other non-collection type), the
+      normal `__in` lookup is used — Django turns it into a subquery.
+    * If the collection has fewer than 65 535 items, `__in` is used as-is.
+    * Otherwise `__any_array` is used so the whole list travels as a single
+      PostgreSQL array parameter.
+    """
+    if not isinstance(values, (list, set, tuple, frozenset)):
+        return Q(**{f"{field_name}__in": values})
+    values = list(values)
+    if len(values) < POSTGRES_MAX_QUERY_PARAMS:
+        return Q(**{f"{field_name}__in": values})
+    return Q(**{f"{field_name}__any_array": values})
+
+
 # a little cache so viewset_for_model doesn't have to iterate over every app every time
 _model_viewset_cache = {}
 

pulpcore/plugin/util.py

@@ -27,6 +27,7 @@
     raise_for_unknown_content_units,
     resolve_prn,
     reverse,
+    safe_in,
     set_current_user,
     set_domain,
 )
@@ -59,5 +60,6 @@
     "reverse",
     "set_current_user",
     "resolve_prn",
+    "safe_in",
     "cache_key",
 ]