handle conflicts on lowercasing emails (#4255)

Co-authored-by: Richard Kuo (Danswer) <rkuo@onyx.app>

backend/alembic/versions/4d58345da04a_lowercase_user_emails.py

@@ -5,7 +5,10 @@ Revises: f1ca58b2f2ec
 Create Date: 2025-01-29 07:48:46.784041
 
 """
+import logging
+from typing import cast
 from alembic import op
+from sqlalchemy.exc import IntegrityError
 from sqlalchemy.sql import text
 
 
@@ -15,21 +18,45 @@ down_revision = "f1ca58b2f2ec"
 branch_labels = None
 depends_on = None
 
+logger = logging.getLogger("alembic.runtime.migration")
+
 
 def upgrade() -> None:
-    # Get database connection
+    """Conflicts on lowercasing will result in the uppercased email getting a
+    unique integer suffix when converted to lowercase."""
+
     connection = op.get_bind()
 
-    # Update all user emails to lowercase
-    connection.execute(
-        text(
-            """
-            UPDATE "user"
-            SET email = LOWER(email)
-            WHERE email != LOWER(email)
-            """
-        )
-    )
+    # Fetch all user emails that are not already lowercase
+    user_emails = connection.execute(
+        text('SELECT id, email FROM "user" WHERE email != LOWER(email)')
+    ).fetchall()
+
+    for user_id, email in user_emails:
+        email = cast(str, email)
+        username, domain = email.rsplit("@", 1)
+        new_email = f"{username.lower()}@{domain.lower()}"
+        attempt = 1
+
+        while True:
+            try:
+                # Try updating the email
+                connection.execute(
+                    text('UPDATE "user" SET email = :new_email WHERE id = :user_id'),
+                    {"new_email": new_email, "user_id": user_id},
+                )
+                break  # Success, exit loop
+            except IntegrityError:
+                next_email = f"{username.lower()}_{attempt}@{domain.lower()}"
+                # Email conflict occurred, append `_1`, `_2`, etc., to the username
+                logger.warning(
+                    f"Conflict while lowercasing email: "
+                    f"old_email={email} "
+                    f"conflicting_email={new_email} "
+                    f"next_email={next_email}"
+                )
+                new_email = next_email
+                attempt += 1
 
 
 def downgrade() -> None: