Why MySQL Slaves Lag Behind Masters and How to Speed Up Replication

Generally, a slave has larger latency than a master because the replication thread on the slave cannot run concurrently. The master commits transactions in parallel (especially with InnoDB), while the slave uses a single SQL thread to apply the binlog, causing the slave to fall behind under high concurrency. Since MySQL 5.6, Oracle MySQL supports multi‑threaded replication via the slave_parallel_workers option, but it only allows parallelism across databases within one instance, not true multi‑table parallel replication, so heavy loads can still cause lag. Another key reason is that traditional MySQL replication is asynchronous: the master commits first, then the slave applies the changes later. Even semi‑synchronous replication only guarantees that the transaction reaches the slave, not that it is committed, so some delay is inevitable. Often, slaves are provisioned with lower‑spec hardware or run multiple instances, further widening the performance gap. To help a slave keep up with the master, consider the following methods: Use MariaDB, which offers true parallel replication and usually keeps the slave in sync. If not possible, wait for MySQL 5.7+ which improves parallelism. Ensure every table has an explicit primary key; missing PKs cause full‑table scans in row‑based mode, dramatically increasing latency. Offload work from MySQL by using application‑side caching, write queues, or batch merging of reads/writes to reduce I/O pressure. Apply appropriate sharding or partitioning strategies to reduce the replication load on a single database/table. Improve IOPS performance, for example: Replace spinning disks with SSDs or PCIe SSDs. Increase RAM and enlarge the InnoDB buffer pool. Use XFS or ReiserFS instead of ext3/ext4 for better IOPS under heavy load. Configure RAID 1+0 (or RAID 5/50 with SSDs) for higher throughput and redundancy. Set the RAID write cache policy to WB or FORCE WB. Choose an appropriate I/O scheduler (e.g., deadline or noop for SSDs) instead of the default cfq. Additional methods are welcome from the community.