Problem

Superset is started sequentially on node-0 then node-1. Each node pulls the large superset-service image (~2 GB), initializes Docker Swarm, starts Redis, and waits for the Superset healthcheck (60s start + up to 14×60s retries = up to 15 min per node). Sequential startup means ~30 min for both nodes.

File: src/initialize.py, start_superset(), lines 249–281

Impact / Caveat

node-0 and node-1 run independent Docker Swarm instances (Docker-in-Docker), so they CAN run in parallel. However, verify there is no dependency on shared MySQL state before parallelizing.

Fix

Same ThreadPoolExecutor pattern as DS-1.

Acceptance Criteria

• start_superset() uses ThreadPoolExecutor to start node-0 and node-1 in parallel
• No shared MySQL state dependency is broken by parallelization (verified)
• Superset startup phase completes in ~15 min instead of ~30 min