Problem

The cockpit already groups runs into calendar-day sections, starting with Yesterday and Today, but live updates are currently enabled only for the section identified as Today.

This causes cross-midnight runs to become stale in the cockpit after the date rolls over. A DAG run that started yesterday but is still running, retrying, or finishing today remains bucketed in the Yesterday section, because dag-run listing is based on the run’s start timestamp for that calendar day. Once midnight passes, that section stops revalidating live, so the run still appears in the cockpit but its status no longer updates unless the page is manually refreshed.

This is especially visible for long-running overnight workflows in the configured cockpit timezone.

Expected Behavior

The cockpit should keep live updates enabled for both the Yesterday and Today sections, based on calendar days in the cockpit timezone.

In practice, if a run belongs to either of those two visible day buckets, that section should continue reflecting live status changes without requiring a manual refresh. This should be framed as a Yesterday + Today calendar-day behavior, not a rolling 24-hour window.

Example

schedule: "55 23 * * *"

steps:
  - name: overnight-job
    command: sleep 7200

If this DAG starts at 23:55 and is still running at 00:30 the next day, it should remain live-updating in the Yesterday section while the cockpit is showing Yesterday and Today.

Motivation

This makes the cockpit reliable for overnight and cross-midnight workflows, which are common for batch jobs, scheduled maintenance, and long-running pipelines.

Without this, operators can see stale status for active runs simply because the run belongs to yesterday’s calendar bucket, even though it is still operationally relevant and still visible in the cockpit. Keeping both Yesterday and Today live-updating matches the current cockpit layout and the existing dag-run date-bucketing model.