DynamoDB hot partition throttling

Your DynamoDB-backed user-events table starts returning ProvisionedThroughputExceededException / throttling at 08:50 and your ingest pipeline's error rate climbs to 15%. CloudWatch shows ConsumedWriteCapacity flat at the provisioned ceiling while ThrottledRequests spikes. Interestingly, total table-level consumed capacity is below the provisioned total, but one partition is hot. A new feature began writing all events for a campaign under a single partition key (campaign_id) this morning. How do you triage and mitigate?