Background service failures were going unnoticed on both clusters

A sweep of every machine on both clusters found background services in a failed state — up to four per machine, some of them months old. None had been reported, because nothing was watching for them.

Most were services that have no business running on these machines at all: a mail transfer agent, a performance-collection suite, and a news banner, all installed as dependencies and left behind.

Fix

Every failure was cleared and the machines returned to a clean state. The configuration automation was corrected so the packages that caused them are removed rather than merely disabled — a disabled package leaves start-up scripts behind, from which the system silently manufactures the same failing service again. The removals are scoped so they only apply to the cluster machines and cannot strip tools deliberately installed elsewhere.