Draft RFC for moving background jobs to a dedicated worker pool:

• Problem: report exports share CPU with request handlers.
• Proposal: a separate deployment that autoscales on queue depth.
• Open question: whether exports need priority lanes.

Added a priority section:

• Lane 1: interactive exports, start within 30 seconds.
• Lane 2: scheduled reports, start within 10 minutes.
• Lane 3: bulk backfills, best effort.

Lanes share one deployment with weighted queue consumers.

Rough cost: a 3 pod worker pool adds about $210 a month, offset by shrinking API pods 31%, which saves about $180.

Rollout in three steps:

1. Deploy the worker pool with the queue consumers off.
2. Move scheduled reports first, then interactive exports.
3. Remove job handlers from API pods after one clean week.