One shard CPU at 100%

Your ledger DB is sharded by account_id into 64 shards. At 22:15 about 1.5% of users get 'service unavailable' on every action while everyone else is fine. Dashboards: one shard (shard-41) has CPU at 100% and query latency at 20s; its connection pool is exhausted; a single very-high-volume account (a large marketplace merchant) lives on shard-41 and a payout batch at 22:12 sent it a flood of writes; the other 63 shards are at baseline. No deploy. How do you triage and mitigate?