print_interim_CPU | tee -a $CPU_LOG >/dev/null ?
It seems pretty clear from the log that the BMC reset was triggered by the script intentionally, if a script can have intention.
Thank you, just to understand: the number of fan zones is setted up by the motherboard, so it doesn't matter if in my case (fractal design node 804) I've cpu and hdds separated by a backplate (in practice the case is splitted in two parts, one for the mobo and obviously ram + cpu, the other for the hdds). Or in that case I consider two fan zones?
sysctl instead. Here is my modified CPU_check_adjust function:
##############################################
# function CPU_check_adjust
# Get CPU temp. Calculate a new DUTY_CPU.
# Send to function adjust_fans.
##############################################
function CPU_check_adjust {
DUTY_CPU_LAST=$DUTY_CPU
# CPU_TEMP=$($IPMITOOL sdr | grep "CPU Temp" | grep -Eo '[0-9]{2,5}')
# following may be more efficient:
# CPU_TEMP=$($IPMITOOL sensor get "CPU Temp" | awk '/Sensor Reading/ {print $4}')
# Get number of CPU cores to check
CORES=$(sysctl hw.ncpu | cut -c10)
CORES=$((CORES - 1))
# Find hottest CPU core
MAX_CORE_TEMP=0
for CORE in $(seq 0 $CORES)
do
CORE_TEMP="$(sysctl dev.cpu.${CORE}.temperature | cut -c24-25)"
if [[ $CORE_TEMP -gt $MAX_CORE_TEMP ]]; then MAX_CORE_TEMP=$CORE_TEMP; fi
done
CPU_TEMP=$MAX_CORE_TEMP
# This will break if settings have non-integers
let DUTY_CPU="$(( (CPU_TEMP-CPU_REF)*CPU_SCALE+DUTY_CPU_MIN ))"
# Don't allow duty cycle outside min-max
if [[ $DUTY_CPU -gt $DUTY_CPU_MAX ]]; then DUTY_CPU=$DUTY_CPU_MAX; fi
if [[ $DUTY_CPU -lt $DUTY_CPU_MIN ]]; then DUTY_CPU=$DUTY_CPU_MIN; fi
adjust_fans $ZONE_CPU $DUTY_CPU $DUTY_CPU_LAST
sleep $CPU_T
print_interim_CPU | tee -a $CPU_LOG >/dev/null
}
cut calls and I guess it will have issues with CPUs having more than 10 CPU cores. I should probably have used awk instead, but...
top.PID JID USERNAME THR PRI NICE SIZE RES STATE C TIME CPU COMMAND 933 0 root 1 -16 - 0K 16K ipmire 5 69:30 0.29% ipmi0: kcs
Yeah, in the machines we see around here all the cores typically have close to the same temperature. This could be different for multi socket systems, tho. If you have 2 CPUs and the system is cache and/or NUMA-aware, then it will try to keep thread execution within those boundaries. Then you could see one hot CPU and one cold CPU for workloads small enough to fit on one CPU. Now, if you were to query just the cold CPU for temperatures to asses the cooling requirements...
# Get number of CPU cores to check CORES=$(sysctl hw.ncpu | cut -c10) CORES=$((CORES - 1))
OK, good to know. I set my script to read CPU temps every 2 seconds. In htop (where you do something similar to -S using the settings menu) ipmi0 kcs is normally at 0.9 - 1.4%. Since I have 8 cores, I have 800% to play with, so that doesn't seem like a lot.
I just updated the post, you beat me to it.
CORES=$(sysctl hw.ncpu | cut -c10) cut call with the -n switch of sysctl. This will make sysctl return just the value of the given key. CORES=$(sysctl -n hw.ncpu)
