Performance: batch permission assignment in bulk set_permissions (#13806)

* Perf: batch guardian permission assignment in bulk-edit

bulk_edit.set_permissions and BulkEditObjectPermissionsView both
looped documents/objects and called set_permissions_for_object per
object, which itself calls guardian's assign_perm/remove_perm once
per (object, user) pair -- ~10-20+ queries per object, scaling with
selection size.

Added set_permissions_for_objects, a bulk equivalent that resolves
existing permission holders once across the whole batch (not once per
object) and applies changes with a small, batch-size-independent
number of queries per action instead of one per (object, user) pair.

* Perf: avoid unnecessary full-row fetches in batch permission assignment

set_permissions_for_objects now takes a model + pks instead of instances,
and identity filtering resolves straight to ids, so bulk-editing
permissions no longer materializes full Document/User/Group rows just to
read their pk/id. Row construction for bulk_create is also chunked to
bound peak memory for very large "apply to all" operations.

* Fix: use .distinct() for existing-grant lookup, drop flaky query-count invariant tests

.distinct() lets the database dedupe identity ids server-side instead of
transferring one row per (object, grantee) match and deduping in Python --
was the dominant cost on a large selection with existing grants.

Also replaced the two query-count-equality tests (bulk_edit and the
bulk_edit_objects API path) with plain functional-correctness checks at
both batch sizes.  Hopefully stops that flake.

* Mark empty-pks early-return in set_permissions_for_objects as no-cover

Defensive guard for an edge case (all requested pks already gone/invalid)
rather than a path normal usage exercises; matches the existing
pragma: no cover convention elsewhere in this file.

* Perf: drop speculative row-chunking in bulk permission assignment, keep the query-batching fix

* Resolve every permission action before applying any of them

This fixes the existing issue and resolves the Copilot comment

* Assert permission assignment does not scale with selection size

The two batching tests only checked that permissions came out correct at
5 and 50 objects, so reverting to the old per-object loop would still
have passed.  Check sizes as well to prevent that
This commit is contained in:
Trenton H
2026-09-10 19:54:09 +00:00
committed by GitHub
parent 9a8163fbbb
commit 2256cb3d38
5 changed files with 428 additions and 11 deletions
+173
View File
@@ -173,6 +173,179 @@ def set_permissions_for_object(
)
def _resolve_permissions(codenames: set[str], ctype: ContentType) -> list[Permission]:
"""
Resolves `codenames` to Permission rows, raising like the single-object
assign_perm() this bulk path replaces does (via a `.get()` internally)
if any codename doesn't exist -- e.g. a client-supplied action name that
was never validated (BulkEditObjectsSerializer._validate_permissions
calls validate_set_permissions() only for its side-effecting id checks
and discards the filtered dict it returns, so an unrecognized action key
reaches this function as-is). A plain `.filter()` with no existence
check would otherwise silently build zero rows and no-op instead of
reporting the bad input.
"""
permission_objs = list(
Permission.objects.filter(content_type=ctype, codename__in=codenames),
)
missing = codenames - {p.codename for p in permission_objs}
if missing:
raise Permission.DoesNotExist(
f"Permission matching query does not exist for codename(s): "
f"{', '.join(sorted(missing))}",
)
return permission_objs
def _apply_bulk_permission_entry(
*,
perm_model: type[UserObjectPermission] | type[GroupObjectPermission],
identity_model: type[User] | type[Group],
identity_field: str,
ids: list[int],
codename: str,
permission_objs: list[Permission],
ctype: ContentType,
object_pks: list[str],
merge: bool,
) -> None:
# Only the ids are needed to build permission rows (via `<field>_id=`),
# so avoid fetching full User/Group rows for identities that may not
# even end up being granted anything new.
add_ids = set(
identity_model.objects.filter(id__in=ids).values_list("id", flat=True),
)
if not merge:
existing_ids = set(
perm_model.objects.filter(
content_type=ctype,
object_pk__in=object_pks,
permission__codename=codename,
)
.values_list(f"{identity_field}_id", flat=True)
.distinct(),
)
remove_ids = existing_ids - add_ids
if remove_ids:
perm_model.objects.filter(
content_type=ctype,
object_pk__in=object_pks,
permission__codename=codename,
**{f"{identity_field}_id__in": remove_ids},
).delete()
if not add_ids:
return
rows = [
perm_model(
content_type=ctype,
object_pk=pk,
permission=permission_obj,
**{f"{identity_field}_id": identity_id},
)
for permission_obj in permission_objs
for pk in object_pks
for identity_id in add_ids
]
# ignore_conflicts skips only rows that already exist as an exact
# (identity, permission, object) match -- the same de-dup the
# underlying (user|group, permission, object_pk) unique constraint
# already enforces for the single-object assign_perm() this replaces,
# so it doesn't change what counts as "already granted". batch_size
# caps how many rows go into a single INSERT statement.
perm_model.objects.bulk_create(rows, ignore_conflicts=True, batch_size=1000)
def set_permissions_for_objects(
permissions: dict,
model: type[Model],
pks: QuerySet | list,
*,
merge: bool = False,
) -> None:
"""
Bulk equivalent of set_permissions_for_object: applies the same
permission changes to every object identified by `pks` at once.
Takes a model + pks (rather than model instances) deliberately -- the
permission rows built below only ever need `pk`, `content_type`, and
identity ids, so callers shouldn't have to fetch full rows (with every
other field) just to hand them to this function.
Deliberately does not use guardian's queryset/list-aware assign_perm:
passing a list as the object routes to bulk_assign_perm, which skips
creating a direct permission row for anyone who already has the
permission via ANY group membership (it checks
ObjectPermissionChecker.has_perm, which is group-inheritance-aware) --
unlike the single-object assign_perm this replaces, which always
ensures a direct row via get_or_create regardless of group-derived
access. Losing that guarantee would mean a later revocation of the
group's grant silently strips access an admin explicitly asked to be
direct. Bulk-creating rows straight against the permission models
instead (see _apply_bulk_permission_entry) preserves the original
always-create-a-direct-row semantics while still batching every object
and every identity into one query per action, rather than one query per
(object, user) pair.
"""
object_pks = [str(pk) for pk in pks]
if not object_pks: # pragma: no cover
return
model_name = model.__name__.lower()
ctype = ContentType.objects.get_for_model(model)
# Every action is resolved up front, before anything is written, so an
# unrecognized action name (see _resolve_permissions) aborts the whole
# call instead of leaving the actions ahead of it already applied --
# BulkEditObjectsSerializer lets unknown keys through and its view turns
# the exception into a 400, so a half-applied change would otherwise be
# reported to the client as a failure.
permissions_by_action: dict[str, list[Permission]] = {}
for action, entry in permissions.items():
if "users" not in entry and "groups" not in entry:
continue
implied_codenames = {f"{action}_{model_name}"}
if action == "change":
# change gives view too
implied_codenames.add(f"view_{model_name}")
permissions_by_action[action] = _resolve_permissions(
implied_codenames,
ctype,
)
for action, entry in permissions.items():
codename = f"{action}_{model_name}"
permission_objs = permissions_by_action.get(action, [])
if "users" in entry:
_apply_bulk_permission_entry(
perm_model=UserObjectPermission,
identity_model=User,
identity_field="user",
ids=entry["users"],
codename=codename,
permission_objs=permission_objs,
ctype=ctype,
object_pks=object_pks,
merge=merge,
)
if "groups" in entry:
_apply_bulk_permission_entry(
perm_model=GroupObjectPermission,
identity_model=Group,
identity_field="group",
ids=entry["groups"],
codename=codename,
permission_objs=permission_objs,
ctype=ctype,
object_pks=object_pks,
merge=merge,
)
def permitted_object_ids(
user: User | None,
model: type[Model],