GitHub employs MySQL as its primary database system, utilizing master-replica setups to handle traffic, with a preference for reading data from replicas to scale efficiently. Challenges arise from replication lag, causing stale data on replicas, which GitHub mitigates through batching large operations into smaller tasks and throttling. This involves pausing between batches to ensure replication lag is within acceptable margins. Traditionally, GitHub used various throttling mechanisms, including a Ruby-based throttler and tools like pt-archiver and gh-ost, each with limitations in handling the scaling infrastructure. To address these, GitHub developed freno, a centralized throttling service that monitors replication lag continuously and recommends appropriate actions to applications, allowing for more efficient traffic management and reduced master reads. Freno operates independently of apps, providing recommendations rather than direct control, and is highly available through a raft cluster. It has significantly reduced read load on masters by enabling safer reads from replicas and improved overall performance by integrating with GitHub's infrastructure and being open-source for community use.