liquidctl/extra/yoda.py

384 lines
14 KiB
Python
Executable File
Raw Permalink Blame History

This file contains ambiguous Unicode characters

This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.

#!/usr/bin/env python3
"""yoda dynamically adjust liquidctl device pump and fan speeds.
Periodically adjusts pump and fan speeds according to user-specified profiles.
Different sensors can be used for each channel. Use show-sensors to view the
sensors available for use with a particular device.
To avoid jerks in pump or fan speeds, an exponential moving average is used as
low-pass filter on sensor data.
Profiles are specified as comma-separated lists of `(temperature,duty)` pairs.
For example: `(20,50),(40,65),(40,65),(50,100)` specifies a duty of 65% at
40°C. The profile will be linearly interpolated between the specified points.
In device controlled mode, sets the device internal control profile and periodically
sends sensor data, but the device will independently control duty cycles to match the temperature.
Named profiles implemented by the device manufacturer, such as "silent", "game", "smart",
are only available in device controlled mode.
Escape sequences or appropriate single or double quotes should be employed to
escape characters that are reserved by the shell in use (e.g. in the case of
bash, the parenthesis and any optional whitespace). In practice, wrapping the
profile in double quotes should be sufficient for most users.
Examples:
yoda --match grid show-sensors
yoda --match grid control fan1 with "(20,20),(35,100)" on nct6793.systin
yoda --match kraken show-sensors
yoda --match kraken control pump with "(20,50),(50,100)" on istats.cpu and fan with "(20,25),(34,100)" on _internal.liquid
yoda --match msi control pump with "smart" on coretemp.package_id_0 and fans with "silent" on coretemp.package_id_0
Usage:
yoda [options] show-sensors
yoda [options] control (<channel> with <profile> on <sensor> [and])...
yoda --help
yoda --version
Options:
--interval <seconds> Update interval in seconds [default: 2]
-m, --match <substring> Filter devices by description substring
-n, --pick <number> Pick among many results for a given filter
--vendor <id> Filter devices by vendor id
--product <id> Filter devices by product id
--release <number> Filter devices by release number
--serial <number> Filter devices by serial number
--bus <bus> Filter devices by bus
--address <address> Filter devices by address in bus
--usb-port <port> Filter devices by USB port in bus
--unsafe <features> Comma-separated bleeding-edge features to enable
-v, --verbose Output additional information
-g, --debug Show debug information on stderr
--legacy-690lc Use Asetek 690LC in legacy mode (old Krakens)
--use-device-controller Use the control loop integrated to the device (MPG coreliquid device)
--version Display the version number
--help Show this message
Requirements:
all platforms liquidctl, including the Python APIs (pip install liquidctl)
Linux/FreeBSD psutil [optional] (pip install psutil)
macOS iStats (gem install iStats)
Windows none, system sensors not yet supported
Changelog:
0.0.6 Add --use-device-controller and make psutil optional.
0.0.5 Document how profiles are specified
0.0.4 Fix casing of log and error messages
0.0.3 Remove duplicate option definition
0.0.2 Add low-pass filter and basic error handling.
0.0.1 Generalization of krakencurve-poc 0.0.2 to multiple devices.
Copyright Jonas Malaco and contributors
SPDX-License-Identifier: GPL-3.0-or-later
"""
import ast
import logging
import math
import sys
import time
from datetime import datetime
from docopt import docopt
import liquidctl.cli as _borrow
from liquidctl.util import normalize_profile, interpolate_profile
import liquidctl.driver
VERSION = "0.0.6"
LOGGER = logging.getLogger(__name__)
INTERNAL_CHIP_NAME = "_internal"
MAX_FAILURES = 3
if sys.platform == "darwin":
import re
import subprocess
elif sys.platform.startswith("linux") or sys.platform.startswith("freebsd"):
try:
import psutil
except ModuleNotFoundError:
psutil = None
def read_sensors(device, **kwargs):
sensors = {}
for k, v, u in device.get_status(**kwargs):
if u == "°C":
sensor_name = k.lower().replace(" ", "_").replace("_temperature", "")
sensors[f"{INTERNAL_CHIP_NAME}.{sensor_name}"] = v
if sys.platform == "darwin":
istats_stdout = subprocess.check_output(["istats"]).decode("utf-8")
for line in istats_stdout.split("\n"):
if line.startswith("CPU"):
cpu_temp = float(re.search(r"\d+\.\d+", line).group(0))
sensors["istats.cpu"] = cpu_temp
break
elif psutil:
for m, li in psutil.sensors_temperatures().items():
for label, current, _, _ in li:
sensor_name = label.lower().replace(" ", "_")
sensors[f"{m}.{sensor_name}"] = current
sensors["cpu_freq"] = psutil.cpu_freq().current
return sensors
def show_sensors(device, **kwargs):
print("{:<60} {:>14}".format("Sensor identifier", "Value"))
print("-" * 80)
sensors = read_sensors(device, **kwargs)
for k, v in sensors.items():
unit = "MHz" if k == "cpu_freq" else "°C"
print(f"{k:<60} {v:>14.1f} {unit}")
def parse_profile(arg, mintemp=0, maxtemp=100, minduty=0, maxduty=100, str_allowed=False):
"""Parse, validate and normalize a temperatureduty profile.
>>> parse_profile('smart', str_allowed=True)
'smart'
>>> parse_profile('(20,30),(30,50),(34,80),(40,90)', 0, 60, 25, 100)
[(20, 30), (30, 50), (34, 80), (40, 90), (60, 100)]
>>> parse_profile('35', 0, 60, 25, 100)
[(0, 35), (59, 35), (60, 100)]
The profile is validated in structure and acceptable ranges. Duty is
checked against `minduty` and `maxduty`. Temperature must be between
`mintemp` and `maxtemp`.
>>> parse_profile('(20,30),(50,100', 0, 60, 25, 100)
Traceback (most recent call last):
...
ValueError: profile must be comma-separated (temperature, duty) tuples or supported mode name
>>> parse_profile('(20,30),(50,100,2)', 0, 60, 25, 100)
Traceback (most recent call last):
...
ValueError: profile must be comma-separated (temperature, duty) tuples
>>> parse_profile('(20,30),(50,97.6)', 0, 60, 25, 100)
Traceback (most recent call last):
...
ValueError: duty must be integer between 25 and 100
>>> parse_profile('(20,15),(50,100)', 0, 60, 25, 100)
Traceback (most recent call last):
...
ValueError: duty must be integer between 25 and 100
>>> parse_profile('(20,30),(70,100)', 0, 60, 25, 100)
Traceback (most recent call last):
...
ValueError: temperature must be integer between 0 and 60
"""
try:
if str_allowed and arg in liquidctl.driver.msi.MpgCooler.BUILTIN_MODES:
return arg
else:
val = ast.literal_eval("[" + arg + "]")
if len(val) == 1 and isinstance(val[0], int):
# for arg == '<number>' set fixed duty between mintemp and maxtemp - 1
val = [(mintemp, val[0]), (maxtemp - 1, val[0])]
except:
raise ValueError(
"profile must be comma-separated (temperature, duty) tuples or supported mode name"
)
for step in val:
if not isinstance(step, tuple) or len(step) != 2:
raise ValueError("profile must be comma-separated (temperature, duty) tuples")
temp, duty = step
if not isinstance(temp, int) or temp < mintemp or temp > maxtemp:
raise ValueError(
"temperature must be integer between {} and {}".format(mintemp, maxtemp)
)
if not isinstance(duty, int) or duty < minduty or duty > maxduty:
raise ValueError("duty must be integer between {} and {}".format(minduty, maxduty))
return normalize_profile(val, critx=maxtemp)
def auto_control(device, channels, profiles, sensors, update_interval, **kwargs):
"""Communicate sensor data directly with the device.
Implemented for use with the MSI coreliquid AIO.
Allows compatible devices to utilize their internal control loop
to determine appropriate fan speeds for the CPU temperature.
"""
assert getattr(
device, "HAS_AUTOCONTROL", False
), f"No registered control loop capability for device {device}!"
device.set_profiles(channels, profiles, **kwargs)
assert all(
s == sensors[0] for s in sensors
), "Controlling different channels with different sensors not possible with device control"
sensor = sensors[0]
LOGGER.info("starting...")
failures = 0
while True:
try:
sensor_data = read_sensors(device, **kwargs)
temp = sensor_data[sensor]
freq = sensor_data.get("cpu_freq", 0)
device.set_time(datetime.now(), **kwargs)
device.set_hardware_status(
temp,
cpu_f=freq,
gpu_f=sensor_data.get("gpu_freq", 0),
gpu_U=sensor_data.get("gpu_usage", 0),
**kwargs,
)
failures = 0
except Exception as err:
failures += 1
LOGGER.error(err)
if failures >= MAX_FAILURES:
LOGGER.critical("too many failures in a row: %d", failures)
raise
time.sleep(update_interval)
def control(device, channels, profiles, sensors, update_interval, **kwargs):
LOGGER.info(
"device: %s on bus %s and address %s", device.description, device.bus, device.address
)
for channel, profile, sensor in zip(channels, profiles, sensors):
LOGGER.info("channel: %s following profile %s on %s", channel, str(profile), sensor)
averages = [None] * len(channels)
cutoff_freq = 1 / update_interval / 10
alpha = 1 - math.exp(-2 * math.pi * cutoff_freq)
LOGGER.info(
"update interval: %d s; cutoff frequency (low-pass): %.2f Hz; ema alpha: %.2f",
update_interval,
cutoff_freq,
alpha,
)
try:
# more efficient and safer API, but only supported by very few devices
apply_duty = device.set_instantaneous_speed
except AttributeError:
apply_duty = device.set_fixed_speed
LOGGER.info("starting...")
failures = 0
while True:
try:
sensor_data = read_sensors(device, **kwargs)
for i, (channel, profile, sensor) in enumerate(zip(channels, profiles, sensors)):
# compute the exponential moving average (ema), used as a low-pass filter (lpf)
ema = averages[i]
sample = sensor_data[sensor]
if ema is None:
ema = sample
else:
ema = alpha * sample + (1 - alpha) * ema
averages[i] = ema
# interpolate on sensor ema and apply corresponding duty
duty = interpolate_profile(profile, ema)
LOGGER.info(
"%s control: lpf(%s) = lpf(%.1f°C) = %.1f°C => duty := %d%%",
channel,
sensor,
sample,
ema,
duty,
)
apply_duty(channel, duty, **kwargs)
if getattr(device, "NEEDS_TIME", False):
device.set_time(datetime.now(), **kwargs)
if getattr(device, "NEEDS_HWSTATUS", False):
device.set_hardware_status(
sensor_data[sensors[0]],
cpu_f=sensor_data.get("cpu_freq", 0),
gpu_f=sensor_data.get("gpu_freq", 0),
gpu_U=sensor_data.get("gpu_usage", 0),
**kwargs,
)
failures = 0
except Exception as err:
failures += 1
LOGGER.error(err)
if failures >= MAX_FAILURES:
LOGGER.critical("too many failures in a row: %d", failures)
raise
time.sleep(update_interval)
if __name__ == "__main__":
if len(sys.argv) == 2 and sys.argv[1] == "doctest":
import doctest
doctest.testmod(verbose=True)
sys.exit(0)
args = docopt(__doc__, version="yoda v{}".format(VERSION))
if args["--debug"]:
args["--verbose"] = True
logging.basicConfig(level=logging.DEBUG, format="[%(levelname)s] %(name)s: %(message)s")
import liquidctl.version
LOGGER.debug("yoda v%s", VERSION)
LOGGER.debug("liquidctl v%s", liquidctl.version.__version__)
elif args["--verbose"]:
logging.basicConfig(level=logging.WARNING, format="%(levelname)s: %(message)s")
LOGGER.setLevel(logging.INFO)
else:
logging.basicConfig(level=logging.WARNING, format="%(levelname)s: %(message)s")
sys.tracebacklimit = 0
if args["--unsafe"] is not None:
unsafe = args["--unsafe"].lower().split(",")
else:
unsafe = []
if (sys.platform.startswith("linux") or sys.platform.startswith("freebsd")) and not psutil:
LOGGER.warning("system sensors are not available, psutil not found")
frwd = _borrow._make_opts(args)
selected = list(liquidctl.driver.find_liquidctl_devices(**frwd))
if len(selected) > 1:
raise SystemExit(
"too many devices, filter or select one. See liquidctl --help and yoda --help."
)
elif len(selected) == 0:
raise SystemExit("no devices matches available drivers and selection criteria")
device = selected[0]
device.connect(unsafe=unsafe)
try:
if args["show-sensors"]:
show_sensors(device, unsafe=unsafe)
elif args["control"]:
if args["--use-device-controller"]:
auto_control(
device,
args["<channel>"],
list(map(lambda p: parse_profile(p, str_allowed=True), args["<profile>"])),
args["<sensor>"],
update_interval=int(args["--interval"]),
unsafe=unsafe,
)
else:
control(
device,
args["<channel>"],
list(map(parse_profile, args["<profile>"])),
args["<sensor>"],
update_interval=int(args["--interval"]),
unsafe=unsafe,
)
else:
raise Exception("nothing to do")
except KeyboardInterrupt:
LOGGER.info("stopped by user.")
finally:
device.disconnect()