Database Write Latency

Database writes can slow down because of storage queues, fsync pressure, compaction, noisy neighbors, or scheduler delay. kprobe connects those causes to the application operation that felt the latency.

Symptom

p99 database write latency increased from 12ms to 900ms

What to inspect

Use the timeline to inspect:

• sys_write
• block_io
• scheduler switches near the write
• page faults in the same service
• colocated workloads doing heavy I/O

Causal pattern

compaction worker block_io
  -> foreground write queued
  -> sys_write duration increased
  -> request timeout

Best practice

Emit query IDs, transaction IDs, or operation names into traces. That makes it possible to search for the slow write directly in kprobe.