The incident is over. The customer is back online. The on-call engineer has a vague sense of what happened. Now the postmortem: what actually triggered this, was it preventable, and what changes would catch it earlier next time. CloudLinux’s lvestats history (queried by lveinfo with date ranges) is the most precise dataset for postmortems on shared cPanel hosts. It captures per-minute resource use by user — sufficient to reconstruct most incidents to within a minute of granularity.
What’s happening
lvestats writes per-LVE resource samples continuously and retains them for up to a week (depends on LVESTATS_RETENTION in /etc/sysconfig/lvestats). lveinfo is the query CLI: it accepts --from, --to, user filters, limit filters, and aggregation modes.
A good postmortem from this data answers four questions:
- What happened? Which limit was hit, by which user, at which minute, and for how long.
- Why? What in the app or traffic caused the limit hit. (lvestats alone does not answer this; combine with access logs, slow query log, etc.)
- Could we have caught it earlier? Was there a warning sign in lvestats data in the hour or day before the incident?
- What to change? Tune limits, fix the app, add monitoring, or accept the incident pattern.
How to verify
The reconstruction queries. For an incident at 14:30 UTC on 2026-05-19 on user “client42”:
# Around the incident
lveinfo --from='2026-05-19 14:00:00' --to='2026-05-19 15:00:00' --user=client42 --show-incidents
# The hour before, to look for warning signs
lveinfo --from='2026-05-19 13:00:00' --to='2026-05-19 14:00:00' --user=client42 --by=minute
# The same user's daily pattern, was this a recurring issue
lveinfo --from='2026-05-12 00:00:00' --to='2026-05-19 23:59:59' --user=client42 --by=hour
Cross-reference with access logs:
zgrep '14:30:' /usr/local/apache/domlogs/client42/example.com_access*.gz | head -50
zgrep '14:30:' /usr/local/apache/domlogs/client42/example.com_access*.gz | awk '{print $7}' | sort | uniq -c | sort -rn | head
The URL distribution during the incident often reveals the trigger (a flood of wp-login.php, an unbounded export endpoint, etc).
The fix
The postmortem template we use:
-
Timeline. “14:25 - lvestats shows CPU climbing on client42. 14:30 - first 508 served. 14:33 - alert fired. 14:38 - engineer raised CPU limit. 14:45 - incident resolved.”
-
Resource impact. “lveinfo shows CPU at 100% (limit) for 13 minutes, 47 EP incidents in the window. Other tenants saw normal performance throughout (lveinfo for top 5 users on the host has no incidents).”
-
Root cause. “Access log shows 18,000 requests to /wp-admin/admin-ajax.php?action=heartbeat during the 13-minute window, all from 4 IPs. Heartbeat was running every 15 seconds against a slow database query (slow log shows 1.2s avg for the heartbeat handler).”
-
What was preventable. “Heartbeat tuning per the heartbeat KB article would have reduced load by ~70%. The slow heartbeat handler is a plugin bug we have reported upstream.”
-
Actions.
- Slow heartbeat for this customer’s site (immediate).
- Add monitoring for sustained EP incidents on any account (within 1 week).
- Document the heartbeat-tuning recommendation in customer-onboarding KB (within 1 month).
-
Filed where customer service can read it. A postmortem nobody else sees does not change behavior. Share with the support and engineering teams.
The data is here for the asking. The discipline is in actually doing the postmortem rather than closing the ticket and moving on. Stack Harbor runs postmortems on every incident as part of environment management.