S3 prefix throttling on burst

Your ingestion pipeline that writes to an S3-compatible object store starts getting a flood of HTTP 429 SlowDown responses during a traffic burst; the dashboard shows request error rate spiking and the per-prefix request graph shows nearly all writes hitting object keys that share the prefix events/2026/06/24/. The pipeline retries aggressively with no backoff, which makes the 429s worse. A new high-volume customer onboarded this morning and the pipeline auto-shards by date, not by entity. How do you triage and stop the throttling, then fix the key design durably?