“The site got slow around 2pm” is the most common support ticket on a shared cPanel host. The customer cannot point to a code change, the server load looks fine, and nothing in the access log is unusual. The signal is almost always in lveinfo and lvestats — CloudLinux’s built-in per-account resource history, going back days. If you are reaching for top first on this kind of ticket, you are looking at the wrong layer.
What’s happening
CloudLinux records per-LVE resource usage on a one-second sample, aggregates it into one-minute buckets, and stores roughly the last week. lveinfo queries this history with filters by time window, user, and which limit. lvestats is the older daemon-style equivalent that some legacy panels still use. Both surface the same underlying data: “in this minute, this user used this much CPU and hit this limit this many times.”
The shape of an incident matters. A user pinned at 100% CPU for ten minutes is usually a runaway plugin or cron. A user hitting EP limit for thirty seconds every five minutes is a cron herd. A user spiking PMEM at midnight is almost always a backup or a wp-cron heavy task.
How to verify
Top users by CPU over the last hour:
lveinfo --period=1h --by=cpu --limit=10
Same view but for entry processes — the limit that produces 508 errors:
lveinfo --period=1h --by=ep --limit=10 --show-incidents
Drill into one specific user for the last 24 hours:
lveinfo --period=24h --user=<username> --show-incidents
lveinfo --period=24h --user=<username> --by=hour
The --show-incidents column shows how many times each limit was actually hit, not just touched. A user with 20 EP incidents in an hour is being throttled hard; a user with 1 incident is fine.
The fix
The diagnosis flow we follow:
- Identify the user.
lveinfo --period=2h --by=cpuand--by=ep— usually one user stands out. - Identify the limit. Which column has the most incidents? CPU, EP, PMEM, IOPS — each implies a different next step.
- Identify the time pattern.
lveinfo --period=24h --user=<u> --by=hour. Periodic spikes = cron. Continuous = a runaway worker. Sudden onset = a deploy or a traffic shift. - Look at the application.
top -u <u>,ps auxf | grep <u>, the slow query log, the access log for the same window. The LVE data tells you which limit; the app logs tell you why. - Decide: raise the limit or fix the app. If a customer’s traffic legitimately outgrew their package, raise the limit and notify them. If the app is misbehaving, fix the app — raising the limit will just delay the next incident.
For postmortems on a one-off incident that already resolved, the same commands with --from= and --to= reconstruct the timeline. Stack Harbor uses this flow for environment management.