manage fans

ipmitemp.py

from pySMART import Device, DeviceList
import subprocess, datetime, time, logging, sys

#below this triggers an alert and fans spin up
MIN_FAN = 10
MAX_FAN = 64
IPMI_SLEEP_DELAY = 2

SYS_FAN_THRESH = [55, 50, 45, 00]
SYS_FAN_SPEEDS = [64, 48, 30, MIN_FAN]

CPU_FAN_THRESH = [65, 60, 55, 00]
CPU_FAN_SPEEDS = [64, 48, 32, MIN_FAN]

summary_temps = lambda temps: '{:.0f}/{:.0f}'.format(sum(temps) / len(temps), max(temps))

def get_hdd_temps(devs):
    """SMART attribute 194 is temperature"""
    return [int(d.attributes[194].raw) for d in devs]


def get_temp_from_sysctl(sl):
    return [float(_.split(':')[1].strip().replace('C','')) for _ in sl]

def get_system_temps(sysctl_list):
    return filter(lambda _:_.startswith('hw.acpi.thermal.') and 'temperature' in _, sysctl_list)

def get_cpu_temps(sysctl_list):
    return filter(lambda _:_.startswith('dev.cpu.') and 'temperature' in _, sysctl_list)


#ipmitool raw 0x30 0x45 0x00

def ipmi(extra):
    result = subprocess.check_output(['/usr/local/bin/ipmitool', 'raw', '0x30'] + extra)
    time.sleep(IPMI_SLEEP_DELAY)
    return result.decode('ascii')

def thresholds():
    for fan in ['FAN1', 'FAN2', 'FAN3', 'FAN4', 'FANA']:
        'ipmitool sensor thresh {} lower 100 100 100'

def setup_fans():
    ipmi(['0x45','0x01', '0x01'])

def fans_fullspeed():
    result = ipmi([
        '0x45',
        '0x00'
    ])

    return result.strip() == '01'

def set_fan(fan, speed):
    ipmi(['0x70','0x66', '0x01', fan, '0x{}'.format(int(min(MAX_FAN, speed)))])

def set_cpu_fan(speed):
    set_fan('0x01', speed)
    return speed

def set_system_fan(speed):
    set_fan('0x00', speed)
    return speed

def apply_temp_policy(TEMP, SPEEDS, THRESH, FAN, adj=0):

    for i, temp in enumerate(THRESH):

        if TEMP >= temp:
            return FAN(SPEEDS[i]+adj)

    return FAN(SPEEDS[-1]+adj)


def setup_logging(debug=False):

    log = logging.getLogger('ipmitemp')
    log.setLevel(logging.DEBUG if debug else logging.INFO)

    formatter = logging.Formatter('%(asctime)s %(name)s - %(levelname)s - %(message)s')

    sh = logging.StreamHandler(stream=sys.stdout)
    sh.setFormatter(formatter)
    log.addHandler(sh)

    return log


import time
log = setup_logging(False)

def doit():

    summary = []

    hdd_temps = get_hdd_temps(DeviceList().devices)

    sysctl_list = str(subprocess.check_output(['/sbin/sysctl', '-a'])).split('\\n')

    sys_temps = [_ for _ in get_temp_from_sysctl(get_system_temps(sysctl_list))]
    cpu_temps = [_ for _ in get_temp_from_sysctl(get_cpu_temps(sysctl_list))]

    summary.append('HDD:{}'.format(summary_temps(hdd_temps)))
    summary.append('SYS:{}'.format(summary_temps(sys_temps)))
    summary.append('CPU:{}'.format(summary_temps(cpu_temps)))

    if not fans_fullspeed():
        summary.append('FANS-SETUP')
        setup_fans()

    SYS = max(sys_temps + hdd_temps)
    CPU = int(sum(cpu_temps)/len(cpu_temps))

    cpu_speed = apply_temp_policy(CPU, CPU_FAN_SPEEDS, CPU_FAN_THRESH, set_cpu_fan)
    sys_speed = apply_temp_policy(SYS, SYS_FAN_SPEEDS, SYS_FAN_THRESH, set_system_fan, adj=max(0,float(cpu_speed-10)*0.50))
    summary.append('FAN_CPU:{:.0f}'.format(cpu_speed))
    summary.append('FAN_SYS:{:.0f}'.format(sys_speed))

    log.info(' '.join(sorted(summary)))

if __name__ == '__main__':
	while True:
		doit()
		time.sleep(30)