Symptom
Every restart is followed by complaints. The first users in find everything slow for an hour, then it settles.
Because restarts happen after patching and upgrades, the upgrade gets the blame and the next one gets deferred.
What we found
The caches were cold, and the first users were warming them up one page at a time.
The restart runbook ended when the service was up, not when it was ready.
The fix
Add a scripted cache warm-up to every restart runbook, so the service is exercised before the first user arrives.
Then the first user of the morning is not the benchmark, and the upgrade stops getting blamed for the restart.
One scriptin the restart runbook removes the complaint from every restart that follows
What the logs said
timings first hour after restart several times the daily p50, settling by hour two system report restart runbook ends at service start; no warm-up step verdict failure mode THREE: cold caches, first users as benchmark fix scripted warm-up before first login, retest the first hour