Reading the dashboards
Three panels are routinely misread. None of them indicates a fault.
VPP worker CPU is a sum, not a percentage of the box. VPP workers busy adds
up each polling worker's busy percentage. VPP's workers spin at ~100% by design
whether or not traffic is flowing, so a box with five workers reads about 500%
when idle. It is not 500% of a core and nothing is wrong. The Busy worker
equivalents tile divides by 100 to give a figure you can compare against your
worker count — the screenshot above shows 79% and 0.8 on a single-worker
evaluation appliance.
Thermal headroom is a ratio, not a temperature. Each sensor is divided by
the warning threshold the BMC declares for that sensor, so 100% means "at its
own limit". Ranking by raw temperature would mislead: CPUs typically warn at
80 °C and DIMMs at 65 °C on the same board, so a CPU at 59 °C and a DIMM at
46 °C are 74% and 71% — nearly tied on headroom while looking 13 °C apart on a
thermometer.
Discrete IPMI sensors carry no health verdict. These are raw state bitfields
from the BMC, and the masks are model-specific, so there is deliberately no
ok/not-ok series for them — an earlier build reported "not ok" for all of them
on a perfectly healthy chassis. Compare a sensor against itself over time, which
is what the state-change panel does. Do not add a threshold to that table.