sqlalchemy/alembic · error · ValueError

No such constraint: '%s'

Error message

No such constraint: '%s'

What it means

Raised by ApplyBatchImpl.drop_constraint() after a KeyError while looking up the constraint name. The name is not present in named_constraints, col_named_constraints or unnamed_constraints, and the constraint is not type-bound (type-bound drops are silently ignored). This means the constraint being dropped does not exist on the table that batch mode reflected or was copied from.

Source

Thrown at alembic/operations/batch.py:697

            if const.name in self.col_named_constraints:
                col, const = self.col_named_constraints.pop(const.name)

                for col_const in list(self.columns[col.name].constraints):
                    if col_const.name == const.name:
                        self.columns[col.name].constraints.remove(col_const)
            elif constraint_name_string(const.name):
                const = self.named_constraints.pop(const.name)
            elif const in self.unnamed_constraints:
                self.unnamed_constraints.remove(const)

        except KeyError:
            if _is_type_bound(const):
                # type-bound constraints are only included in the new
                # table via their type object in any case, so ignore the
                # drop_constraint() that comes here via the
                # Operations.implementation_for(alter_column)
                return
            raise ValueError("No such constraint: '%s'" % const.name)
        else:
            if isinstance(const, PrimaryKeyConstraint):
                for col in const.columns:
                    self.columns[col.name].primary_key = False

    def create_index(self, idx: Index) -> None:
        self.new_indexes[idx.name] = idx  # type: ignore[index]

    def drop_index(self, idx: Index) -> None:
        try:
            del self.indexes[idx.name]  # type: ignore[arg-type]
        except KeyError:
            raise ValueError("No such index: '%s'" % idx.name)

    def rename_table(self, *arg, **kw):
        raise NotImplementedError("TODO")

View on GitHub (pinned to 44fb345033)

Solutions

  1. Inspect the live table for actual constraint names (e.g. via reflection / PRAGMA foreign_key_list / information_schema) and correct the name in the migration.
  2. Confirm the constraint has not already been dropped by a prior revision or earlier in the same batch.
  3. Guard the drop with an existence check or use if_exists semantics where supported.
  4. If the constraint is type-bound, drop it via alter_column on the column type rather than drop_constraint.

Example fix

// before
with op.batch_alter_table('user') as batch_op:
    batch_op.drop_constraint('fk_user_old')  # not present -> ValueError

// after
# verify against the reflected table first
from sqlalchemy import inspect
constraints = [c['name'] for c in inspect(op.get_bind()).get_foreign_keys('user')]
with op.batch_alter_table('user') as batch_op:
    if 'fk_user_old' in constraints:
        batch_op.drop_constraint('fk_user_old')
Defensive patterns

Strategy: validation

Validate before calling

from sqlalchemy import inspect

def constraint_exists(bind, table, name) -> bool:
    fks = [c['name'] for c in inspect(bind).get_foreign_keys(table)]
    uqs = [c['name'] for c in inspect(bind).get_unique_constraints(table)]
    cks = [c['name'] for c in inspect(bind).get_check_constraints(table)]
    return name in (set(fks) | set(uqs) | set(cks))

Try / catch

try:
    batch_op.drop_constraint(name)
except ValueError as e:
    if 'No such constraint' in str(e):
        pass  # already absent, treat as success or log
    else:
        raise

Prevention

When it happens

Trigger: Inside batch_alter_table, calling batch_op.drop_constraint('some_name') where 'some_name' is not a constraint on the target table; dropping a constraint that was already dropped in the same batch; dropping a constraint whose name was changed/never existed; case/quoting mismatch in the constraint name.

Common situations: Stale migration referencing a constraint that a later/earlier revision already removed; typo in constraint name; mismatch between the name in metadata and the name reflected from the DB (e.g. naming convention differences); running a migration against a DB that is ahead of or behind the expected state.

Related errors

AI-assisted analysis of sqlalchemy/alembic@44fb345033 (2026-08-04). Data as JSON: /data/errors/c6e9e47ef4273c1d.json. Report an issue: GitHub.