Question
Your LSM-tree key-value store (RocksDB-backed) starts showing climbing write latency and growing disk-write IOPS even though application write QPS is flat. Compaction-pending bytes are piling up, disk-write throughput is 6x the logical write rate (high write amplification), and read latency is rising too as the number of SST files to merge per read grows. A config change two days ago lowered the write-buffer (memtable) size and increased the number of L0 files allowed before compaction triggers, intended to 'reduce write stalls'. The volume is also approaching its IOPS ceiling. How do you triage and stabilize, then fix the storage engine tuning durably?
Stop the bleeding first (mitigate), then form hypotheses from real signals. Separate root cause from symptom, communicate status as you go, and close with what prevents a repeat.