vmstat: Spot I/O Wait and Swapping in CI

vmstat prints CPU, memory, swap, and I/O counters at an interval, revealing whether a slow runner is CPU-bound, I/O-bound, or swapping itself to death.

A build that is slow without an obvious CPU hog is often swapping or waiting on disk. vmstat’s si/so and wa columns distinguish those causes at a glance.

What it does

vmstat <interval> samples system-wide stats each interval. Key columns: si/so are swap-in/swap-out (KB/s; nonzero means active swapping), wa is the percentage of CPU time waiting on I/O, b is processes blocked on I/O, and us/sy/id are user/system/idle CPU. The first row is an average since boot; ignore it and read the rest.

Common usage

Terminal

vmstat 2                         # sample every 2 seconds
vmstat 2 5                       # 5 samples, then exit
vmstat -s                        # one-shot memory/event totals
vmstat -d                        # per-disk I/O statistics

Options

Field / flag	What it does
<delay> [count]	Sample every delay seconds, count times
si / so	Swap in / out per second (>0 = thrashing)
wa	CPU percent waiting on I/O
b	Processes in uninterruptible sleep (blocked)
-s	Dump memory and event totals
-d	Per-disk statistics

In CI

Run vmstat 2 10 during a slow step. High us with low wa is genuinely CPU-bound (more cores help). High wa with a high b count is I/O-bound (faster disk or fewer parallel writers help). Any sustained si/so means the runner is out of RAM and swapping, which destroys performance and often precedes an OOM kill.

Common errors in CI

Reading the first vmstat row as "current" is a common mistake; it is the since-boot average and is usually misleadingly calm. In a container, vmstat reflects the host, so swap and I/O may be other tenants’ activity, not yours. Persistent wa near 100 with the build stalled points at disk, not CPU.