Merge branch 'new-script' into 'main'

convert fan script from bash to rust

See merge request c4181/server-fan-controller!1

.gitignore

@@ -0,0 +1,6 @@
+/target
+config.toml
+
+# Jetbrains
+.idea
+*.iml

Cargo.toml

@@ -0,0 +1,15 @@
+[package]
+name = "server-fan-controller"
+version = "0.1.0"
+edition = "2021"
+
+# See more keys and their definitions at https://doc.rust-lang.org/cargo/reference/manifest.html
+
+[dependencies]
+confy = "0.6.1"
+nvml-wrapper = "0.10.0"
+rust-ipmi = "0.1.1"
+sysinfo = "0.31.2"
+serde = { version = "1.0.208", features = ["derive"] }
+log4rs = "1.3.0"
+log = "0.4.22"

log4rs.yaml

@@ -0,0 +1,9 @@
+appenders:
+  stdout:
+    kind: console
+#    encoder:
+#      kind: json
+root:
+  level: info
+  appenders:
+    - stdout

src/main.rs

@@ -0,0 +1,228 @@
+use log::{debug, error, info, trace};
+use nvml_wrapper::enum_wrappers::device::TemperatureSensor;
+use nvml_wrapper::Nvml;
+use rust_ipmi::{IPMIClient, IPMIClientError};
+use serde::{Deserialize, Serialize};
+use std::cmp::max;
+use std::fmt::{Display, Formatter};
+use std::path::Path;
+use std::time::Duration;
+use std::{env, thread};
+use sysinfo::Components;
+
+fn main() {
+    let args: Vec<String> = env::args().collect();
+    let config_path = if args.len() > 1 { Path::new(&args[1] ) } else { Path::new("config.toml") };
+    let cfg: AppConfig = confy::load_path(Path::new(config_path)).expect("Couldn't read config");
+    let sleep_duration = Duration::from_secs(cfg.run_interval);
+    match Path::new("log4rs.yaml").exists() {
+        true => log4rs::init_file("log4rs.yaml", Default::default()).unwrap(),
+        false => println!("No log4rs.yaml file found. Logging will not be enabled")
+    }
+
+    loop {
+        let max_core_temp = get_max_core_temp();
+        let cpu_speed = calculate_cpu_curve_speed(max_core_temp as i32);
+
+        let max_gpu_temp = if cfg.monitor_gpus { get_max_gpu_temp() } else { 0 };
+        let gpu_speed = calculate_gpu_curve_speed(max_gpu_temp);
+
+        let fan_speed = max(cpu_speed, gpu_speed);
+        info!("Setting fan speed to {}", fan_speed);
+        match set_fan_speed(fan_speed, &cfg) {
+            Ok(_) => {
+                debug!("Successfully set fan speed");
+                thread::sleep(sleep_duration);
+            },
+            Err(_) => {
+                thread::sleep(Duration::from_secs(3))
+            }
+        };
+    }
+}
+
+fn calculate_cpu_curve_speed(temp: i32) -> FanSpeed {
+    let fan_speed = match temp {
+        t if t <= 35 => {
+            trace!("Setting fan speeds to 4.3k & 3.0k RPM");
+            FanSpeed::VeryLow
+        }
+        t if t <= 45 => {
+            trace!("Setting fan speeds to 5.3k & 3.4k RPM");
+            FanSpeed::Low
+        }
+        t if t <= 55 => {
+            trace!("Setting fan speeds to 5.6k & 3.8k RPM");
+            FanSpeed::Moderate
+        }
+        t if t <= 65 => {
+            trace!("Setting fan speeds to 5.9k & 4.1k RPM");
+            FanSpeed::High
+        }
+        t if t <= 75 => {
+            trace!("Setting fan speeds to 6.2k & 4.3k RPM");
+            FanSpeed::VeryHigh
+        }
+        t if t <= 85 => {
+            trace!("Setting fan speeds to 6.6k & 4.6k RPM");
+            FanSpeed::BeyondVeryHigh
+        }
+        t if t <= 95 => {
+            trace!("Setting fan speeds to 7.5k & 5.0k RPM");
+            FanSpeed::OneStepFromHell
+        }
+        _ => {
+            trace!("Setting fan speeds to 8.0k & 5.5k RPM");
+            FanSpeed::MaxLoad
+        }
+    };
+
+    fan_speed
+}
+
+fn calculate_gpu_curve_speed(temp: u32) -> FanSpeed {
+    let fan_speed = match temp {
+        t if t <= 35 => {
+            FanSpeed::VeryLow
+        }
+        t if t <= 45 => {
+            FanSpeed::Low
+        }
+        t if t <= 55 => {
+            FanSpeed::Moderate
+        }
+        t if t <= 65 => {
+            FanSpeed::VeryHigh
+        }
+        t if t <= 75 => {
+            FanSpeed::BeyondVeryHigh
+        }
+        t if t <= 80 => {
+            FanSpeed::OneStepFromHell
+        }
+        t if t <= 85 => {
+            FanSpeed::MaxLoad
+        }
+        _ => {
+            FanSpeed::GpuHigh
+        }
+    };
+
+    fan_speed
+}
+
+fn set_fan_speed(speed: FanSpeed, cfg: &AppConfig) -> Result<bool, IPMIClientError> {
+    let mut client: IPMIClient = match IPMIClient::new(&cfg.ipmi_config.ip) {
+        Ok(v) => v,
+        Err(e) => {
+            error!("Failed to construct IPMI Client because: {e}");
+            return Err(e)
+        }
+    };
+
+    // establish a session with the BMC using the credentials specified
+    match client.establish_connection(&cfg.ipmi_config.user, &cfg.ipmi_config.password) {
+        Ok(_) => {}
+        Err(e) => {
+            error!("Failed to establish IPMI connection because: {e}");
+            return Err(e)
+        }
+    };
+
+    // send a command to the BMC using raw values
+    match client.send_raw_request(0x30, 0x30, Some(vec![0x02, 0xff, speed as u8])) {
+        Ok(_) => Ok(true),
+        Err(e) => {
+            error!("Failed to send IPMI request because: {e}");
+            Err(e)
+        }
+    }
+}
+
+fn get_max_core_temp() -> f32 {
+    let components = Components::new_with_refreshed_list();
+
+    let mut max_temp:f32 = 0.0;
+    for component in components.iter() {
+        if component.label().starts_with("coretemp Core") {
+            if component.temperature() > max_temp {
+                max_temp = component.temperature();
+            }
+        } else if component.label().contains("coretemp Package id 0") {
+            info!("CPU Package 0 is {}C", component.temperature())
+        } else if component.label().starts_with("coretemp Package id 1") {
+            info!("CPU Package 1 is {}C", component.temperature())
+        }
+    }
+
+    max_temp
+}
+
+fn get_max_gpu_temp() -> u32 {
+    let nvml = Nvml::init().unwrap();
+
+    let count = nvml.device_count().unwrap();
+    let mut max_temp:u32 = 0;
+    for index in 0..count {
+        let device = nvml.device_by_index(index).unwrap();
+        let name = device.name().unwrap();
+        let temp = device.temperature(TemperatureSensor::Gpu).unwrap();
+
+        if temp > max_temp {
+            max_temp = temp;
+        }
+        info!("GPU {} is {}C", name, temp);
+    }
+
+    max_temp
+}
+
+#[derive(Ord, PartialOrd, Eq, PartialEq, Debug)]
+#[repr(u8)]
+enum FanSpeed {
+    VeryLow = 24,         // 0x18 in decimal
+    Low = 29,             // 0x1d in decimal
+    Moderate = 31,        // 0x1f in decimal
+    High = 32,            // 0x20 in decimal
+    VeryHigh = 34,        // 0x22 in decimal
+    BeyondVeryHigh = 36,  // 0x24 in decimal
+    OneStepFromHell = 38, // 0x26 in decimal
+    MaxLoad = 43,         // 0x2b in decimal
+    GpuHigh = 50,
+}
+
+impl Display for FanSpeed {
+    fn fmt(&self, f: &mut Formatter<'_>) -> std::fmt::Result {
+        write!(f, "{}%, {:?}", self.value(), self)
+    }
+}
+
+impl FanSpeed {
+    fn value(&self) -> u8 {
+        unsafe { *(self as *const Self as *const u8) }
+    }
+
+    fn as_hex(&self) -> String {
+        format!("{:#x}", self.value())
+    }
+}
+
+#[derive(Debug, Serialize, Deserialize)]
+struct AppConfig {
+    monitor_gpus: bool,
+    run_interval: u64,
+    ipmi_config: IpmiConfig,
+}
+
+impl Default for AppConfig {
+    fn default() -> Self {
+        panic!("Could not find config file")
+    }
+}
+
+#[derive(Debug, Serialize, Deserialize)]
+struct IpmiConfig {
+    ip: String,
+    user: String,
+    password: String,
+}