📝 forum-slowdown-diagnosis.mdv4.2 · 2026-09-05

Forum Page-Load Slowdown — Diagnosis Summary

Original Query

I rebooted the VPS after Uninstalling and Installing Termius, defining the 128.199.200.202 session (to Live) and webpage loads have normalised to be quick ... so please furnish data for:

  1. What is your prognosis now?
  2. What is your "best-guess" root cause?
  3. List ALL the remedies that -A. We attempted, -B. All their outcomes, and -C. Rate their contribution towards the final fix(es) ... INCLUDING ALL the active Apache mods. and ALL the other subsystems mods. also.

1. Prognosis

Cautiously good. Page loads are fast now, on a genuinely different tested config than 24 hours ago (higher worker limits, healthier DB buffer pool, better entropy). But we changed multiple things across two reboots without isolating them one at a time, so this is "likely fixed" not "proven fixed." Watch it over the next few days — if it degrades again, the SSL session cache issue is still unresolved and is the most likely next thing to bite.

2. Best-Guess Root Cause

Apache worker exhaustion (MaxClients/ServerLimit capped at 35, actual concurrent load hit 17+) is the strongest candidate for the actual page-load slowness — it's the only issue found with hard, reproducible evidence of real strain (via server-status) and a mechanism that exactly matches the symptoms (random, severe, no MySQL/CPU/disk correlate). Everything else was either ruled out, inconclusive, or a separate red herring (Termius).

3. Full Remedy Log

# Remedy Outcome Contribution to Fix
A1 InnoDB buffer pool 32M → 128M → 256M (my.cnf) Confirmed active, no felt improvement Low — good DB hygiene, not the cause
A2 SSL cipher suite: removed DHE, forced ECDHE-only (ssl.conf) Made it measurably worse (9.2s total vs 4.5s before) Negative — never reverted, currently still active; worth reverting or leaving, low-risk either way
A3 OCSP stapling attempt (SSLUseStapling) Failed — mod_ssl 2.2.15 (final el6 build) doesn't support it, no OS update exists None — dead end, config change was rolled back
A4 haveged installed (entropy 209 → 3421) Confirmed active, no felt improvement at the time Low-moderate — real fix for a real weakness, may help under future load even if not felt yet
A5 IPv4-direct vs IPv6 (domain) test IPv4 was faster than domain (IPv6) at the time Informational only — pointed at possible carrier IPv6 routing flakiness, never fixed server-side (nothing to fix there)
A6 MySQL slow query log + EXPLAIN on smf209_messages join Found 5–16s query times, but EXPLAIN showed a sane plan on small tables — the query itself wasn't badly written, it was a symptom of system-wide stall, not a cause Low — diagnostic value only
A7 MaxClients 35 → 150 + ServerLimit 35 → 150, Apache restarted Confirmed real bottleneck (17/35 busy) before the fix High — primary suspect
A8 SSL session cache (12 hits / 22,868 misses) Identified, config unchanged, still broken None yet — open issue
A9 Termius app (paste-buffer lag, command delays, one hang) Isolated to the app itself via JuiceSSH comparison; fixed by uninstall/reinstall Zero effect on actual page loads (client-side terminal tool only) — but real, and was actively confusing the whole investigation
A10 Two full VPS reboots Clears any transient kernel/network/process state Unknown but plausible — can't rule out something stale got cleared

Active Apache Modules / Subsystems (unchanged all night, per server-status banner)

None of these were modified except mod_ssl's cipher suite line (A2).


Still Open / Worth a Clear-Headed Follow-Up