we hit with the admin dashboard during a traffic spike

We ran into this around the admin dashboard during a traffic spike.

We did not rewrite the service. We moved one endpoint at a time.

Dual writes ran until nightly totals matched for several days.

A flag sent a small percentage of reads to the new path.

The hard part was matching rounding the old code had hidden.

Cutover finished with a small blast radius and an easy rollback.