error-troubleshooting

This error means you forgot to call .build() on the WorkflowBuilder before execution. Change runtime.execute(workflow) to runtime.execute(workflow.build()) so the runtime receives a compiled Workflow instead of a builder.

Nexus blocks because auto_discovery=True triggers synchronous workflow execution during DataFlow model registration. Create Nexus with auto_discovery=False, register workflows manually with app.register(), and startup completes in under 2 seconds.

Connection exhaustion happens when each Gunicorn worker creates its own database pool. Create one pool at app startup and pass it via the external_pool parameter, sizing max_pool_size as database max connections divided by worker count.

Use the 4-parameter form: add_connection(from_node, from_output, to_node, to_input). For nested outputs use dot notation like "result.filters". The deprecated 3-parameter syntax causes 'node not found' errors.

Cycles fail to converge when convergence criteria use nested paths or max_iterations is unset. Use flattened field names in converge_when("done == True"), always set max_iterations, and provide initial parameters for the first iteration.

No, DataFlow nodes expect native Python types and reject {{}} templates with errors like "invalid literal for int()". Pass dynamic values through explicit workflow connections instead, which preserves type safety.