Add support for user deletion (#2486)

The aim here is to retain the database "User" entry but wholly
anonymised (getting rid of their display name and their provider
username)

backend/oasst_backend/api/v1/users.py

@@ -210,6 +210,7 @@ def delete_user(
 ):
     """
     Delete a user by global user ID. Only trusted clients can delete users.
+    User deletion anonymises the data of the user.
     """
     ur = UserRepository(db, api_client)
     ur.mark_user_deleted(user_id)

backend/oasst_backend/user_repository.py

@@ -4,6 +4,7 @@
 from oasst_backend.config import settings
 from oasst_backend.models import ApiClient, User
 from oasst_backend.utils.database_utils import CommitMode, managed_tx_method
+from oasst_shared import utils as shared_utils
 from oasst_shared.exceptions import OasstError, OasstErrorCode
 from oasst_shared.schemas import protocol as protocol_schema
 from oasst_shared.utils import utcnow
@@ -105,6 +106,7 @@ def update_user(
     def mark_user_deleted(self, id: UUID) -> None:
         """
         Update a user by global user ID to set deleted flag. Only trusted clients may delete users.
+        User deletion anonymises the data of the user.
 
         Raises:
             OasstError: 403 if untrusted client attempts to delete a user. 404 if user with ID not found.
@@ -119,6 +121,12 @@ def mark_user_deleted(self, id: UUID) -> None:
 
         user.deleted = True
 
+        # Anonymise user data
+        user.display_name = shared_utils.DELETED_USER_DISPLAY_NAME
+        # Ensure uniqueness of (username, auth_method, api_client_id) Index
+        user.username = f"{shared_utils.DELETED_USER_ID_PREFIX}{user.id}"
+        user.show_on_leaderboard = False
+
         self.db.add(user)
 
     @managed_tx_method(CommitMode.COMMIT)

inference/server/alembic/versions/2023_04_12_2033-f0e18084aae4_add_deleted_field_to_user.py

@@ -0,0 +1,27 @@
+"""Add deleted field to user
+
+Revision ID: f0e18084aae4
+Revises: 78f16015b904
+Create Date: 2023-04-12 20:33:28.239793
+
+"""
+import sqlalchemy as sa
+from alembic import op
+
+# revision identifiers, used by Alembic.
+revision = "f0e18084aae4"
+down_revision = "78f16015b904"
+branch_labels = None
+depends_on = None
+
+
+def upgrade() -> None:
+    # ### commands auto generated by Alembic - please adjust! ###
+    op.add_column("user", sa.Column("deleted", sa.Boolean(), nullable=False))
+    # ### end Alembic commands ###
+
+
+def downgrade() -> None:
+    # ### commands auto generated by Alembic - please adjust! ###
+    op.drop_column("user", "deleted")
+    # ### end Alembic commands ###

inference/server/oasst_inference_server/models/user.py

@@ -14,6 +14,8 @@ class DbUser(SQLModel, table=True):
 
     display_name: str = Field(nullable=False, max_length=256)
 
+    deleted: bool = Field(default=False)
+
 
 class DbRefreshToken(SQLModel, table=True):
     __tablename__ = "refresh_token"

inference/server/oasst_inference_server/routes/admin.py

@@ -5,6 +5,7 @@
 from oasst_inference_server import auth, database, deps, models
 from oasst_inference_server.schemas import worker as worker_schema
 from oasst_inference_server.settings import settings
+from oasst_shared import utils as shared_utils
 
 router = fastapi.APIRouter(
     prefix="/admin",
@@ -87,3 +88,23 @@ async def revoke_refresh_tokens(
         refresh_token.enabled = False
     await session.commit()
     return fastapi.Response(status_code=200)
+
+
+@router.delete("/users/{user_id}")
+async def delete_user(
+    user_id: str,
+    root_token: str = Depends(get_root_token),
+    session: database.AsyncSession = Depends(deps.create_session),
+):
+    """Deletes a user."""
+    logger.info(f"Deleting user {user_id}")
+    user = await session.get(models.DbUser, user_id)
+    user.deleted = True
+
+    # Anonymise user data
+    user.display_name = shared_utils.DELETED_USER_DISPLAY_NAME
+    # Ensure uniqueness
+    user.provider_account_id = f"{shared_utils.DELETED_USER_ID_PREFIX}{user.id}"
+
+    await session.commit()
+    return fastapi.Response(status_code=200)

oasst-shared/oasst_shared/utils.py

@@ -4,6 +4,9 @@
 
 from loguru import logger
 
+DELETED_USER_DISPLAY_NAME = "Deleted User"
+DELETED_USER_ID_PREFIX = "deleted_"
+
 
 def utcnow() -> datetime:
     """Return the current utc date and time with tzinfo set to UTC."""