feat: add verified manual pipeline migration for plugin runners

src/langbot/pkg/persistence/alembic/versions/0027_pipeline_migration_snapshots.py

@@ -0,0 +1,53 @@
+"""Create the manual migration journal only; never convert pipeline JSON."""
+
+from alembic import op
+import sqlalchemy as sa
+
+revision = '0027_pipeline_migration'
+down_revision = '0026_merge_master_beta'
+branch_labels = None
+depends_on = None
+
+
+def upgrade():
+    bind = op.get_bind()
+    table = 'pipeline_migration_snapshots'
+    if not sa.inspect(bind).has_table(table):
+        op.create_table(
+            table,
+            sa.Column('uuid', sa.String(36), primary_key=True),
+            sa.Column(
+                'workspace_uuid', sa.String(36), sa.ForeignKey('workspaces.uuid', ondelete='CASCADE'), nullable=False
+            ),
+            sa.Column('pipeline_uuid', sa.String(255), nullable=False),
+            sa.Column('source_fingerprint', sa.String(64), nullable=False),
+            sa.Column('target_fingerprint', sa.String(64), nullable=False),
+            sa.Column('planner_version', sa.String(64), nullable=False),
+            sa.Column('source_snapshot', sa.JSON(), nullable=False),
+            sa.Column('state', sa.String(32), nullable=False),
+            sa.Column('created_at', sa.DateTime(), nullable=False, server_default=sa.func.now()),
+            sa.ForeignKeyConstraint(
+                ['workspace_uuid', 'pipeline_uuid'],
+                ['legacy_pipelines.workspace_uuid', 'legacy_pipelines.uuid'],
+                name='fk_pipeline_migration_workspace_pipeline',
+                ondelete='CASCADE',
+            ),
+            sa.UniqueConstraint(
+                'workspace_uuid', 'pipeline_uuid', 'source_fingerprint', name='uq_pipeline_migration_source'
+            ),
+            sa.CheckConstraint("state IN ('activation_pending', 'active')", name='ck_pipeline_migration_state'),
+        )
+        op.create_index('ix_pipeline_migration_workspace_pipeline', table, ['workspace_uuid', 'pipeline_uuid'])
+    if bind.dialect.name == 'postgresql':
+        op.execute(f'ALTER TABLE {table} ENABLE ROW LEVEL SECURITY')
+        op.execute(f'ALTER TABLE {table} FORCE ROW LEVEL SECURITY')
+        op.execute(f'DROP POLICY IF EXISTS langbot_workspace_isolation ON {table}')
+        expression = "workspace_uuid = NULLIF(current_setting('langbot.workspace_uuid', true), '')"
+        op.execute(
+            f'CREATE POLICY langbot_workspace_isolation ON {table} FOR ALL TO PUBLIC '
+            f'USING ({expression}) WITH CHECK ({expression})'
+        )
+
+
+def downgrade():
+    op.drop_table('pipeline_migration_snapshots')

src/langbot/pkg/persistence/mgr.py

@@ -71,6 +71,7 @@ _ALEMBIC_TENANT_TABLES = {
     'rerank_models',
     'legacy_pipelines',
     'pipeline_run_records',
+    'pipeline_migration_snapshots',
     'plugin_settings',
     'knowledge_bases',
     'knowledge_base_files',

src/langbot/pkg/persistence/pipeline_admission.py

@@ -0,0 +1,27 @@
+"""Transaction-held, tenant/pipeline-scoped admission for form producers.
+
+Every pending producer must hold this row until its INSERT commits. Migration
+holds the same row for both CAS and activation. No table/process/advisory lock.
+The no-op UPDATE also acquires SQLite's write reservation before any reads.
+"""
+
+import sqlalchemy as sa
+from ..entity.persistence.pipeline import LegacyPipeline
+
+
+async def lock_pipeline_admission(session, workspace_uuid: str, pipeline_uuid: str):
+    if not workspace_uuid or not pipeline_uuid:
+        raise ValueError('Pipeline admission requires exact ownership')
+    t = LegacyPipeline.__table__
+    result = await session.execute(
+        sa.update(t)
+        .where(t.c.workspace_uuid == workspace_uuid, t.c.uuid == pipeline_uuid)
+        .values(updated_at=t.c.updated_at)
+    )
+    if result.rowcount != 1:
+        raise ValueError('Pipeline admission authority unavailable')
+    return (
+        await session.execute(
+            sa.select(t.c.config).where(t.c.workspace_uuid == workspace_uuid, t.c.uuid == pipeline_uuid)
+        )
+    ).scalar_one()

src/langbot/pkg/persistence/tenant_uow.py

@@ -58,6 +58,7 @@ TENANT_TABLE_COLUMNS: dict[str, str] = {
     'rerank_models': 'workspace_uuid',
     'legacy_pipelines': 'workspace_uuid',
     'pipeline_run_records': 'workspace_uuid',
+    'pipeline_migration_snapshots': 'workspace_uuid',
     'plugin_settings': 'workspace_uuid',
     'knowledge_bases': 'workspace_uuid',
     'knowledge_base_files': 'workspace_uuid',
@@ -467,9 +468,21 @@ def _validate_scoped_statement_call(args: tuple[typing.Any, ...], kwargs: dict[s
             raise ScopedSessionTransactionError('TenantUnitOfWork does not allow literal-execute SQL parameters')
 
         if isinstance(element, sqlalchemy.sql.elements.Cast) and type(element.type) not in {Vector, HALFVEC}:
-            raise ScopedSessionTransactionError(
-                'TenantUnitOfWork only allows the trusted pgvector cast used by tenant vector search'
+            # Manual Pipeline migration compares DB-native JSON text, because
+            # PostgreSQL JSON has no equality operator. Permit only these fixed
+            # mapped columns and a built-in Text target, never arbitrary casts.
+            source = element.clause
+            pipeline_json_cas = (
+                type(element.type) is sqlalchemy.Text
+                and isinstance(source, sqlalchemy.Column)
+                and type(source.type) is sqlalchemy.JSON
+                and getattr(getattr(source, 'table', None), 'name', None) == 'legacy_pipelines'
+                and source.name in {'config', 'stages', 'extensions_preferences'}
             )
+            if not pipeline_json_cas:
+                raise ScopedSessionTransactionError(
+                    'TenantUnitOfWork only allows trusted pgvector and Pipeline JSON CAS casts'
+                )
 
         if isinstance(element, sqlalchemy.sql.functions.FunctionElement):
             function_name = str(getattr(element, 'name', '')).casefold()