Every Fast Write Moves Work Somewhere Else

A deep dive into storage engine write paths reveals that the fastest acknowledgment points—like copying to memory or syncing a local SSD—shift durability risks elsewhere. The article contrasts local WALs, network volumes, and object storage, showing how batching flushes improves throughput but complicates failure semantics. It argues that unlabeled latency numbers are misleading without knowing what failures the acknowledged write survives, and that every speed gain comes with hidden costs in durability or cleanup.