Client retry loop triggers quota

Your backend calls an internal inventory API that enforces a 2,000 rps per-service quota. After a client-library upgrade shipped this morning, your inventory calls start getting throttled (429s) even though your user traffic is flat. Dashboards show your outbound inventory request rate doubled at the exact deploy time, with no increase in incoming user requests. The new client library changed the default: a failed call now retries up to 5 times with no backoff, and a borderline-slow inventory shard causes a trickle of failures. How do you triage and mitigate?