[rust/rqd] Implement NIMBY logic (#1766)

Implemented NIMBY (Not In My BackYard) logic in the new Rust-based RQD,
replicating the behavior of the original Python implementation.

- Add NIMBY support to the Rust-based RQD system, preventing jobs from
running when a user is actively using the machine.
- Add new `rust/crates/rqd/src/system/nimby.rs` module to detect
mouse/keyboard activity using `device_query`
- NIMBY configurable via `rqd.yaml` (nimby_mode, idle_threshold, etc.)
- Dynamically sets `DISPLAY` and `XAUTHORITY` from a file
- Integrated with `MachineMonitor` to `lock/unlock` cores and update
host state
- Graceful shutdown support with `broadcast::channel`

---------

Signed-off-by: Diego Tavares <[email protected]>
Co-authored-by: Ramon Figueiredo <[email protected]>

Author: DiegoTavares

.github/workflows/rust.yml

@@ -15,6 +15,9 @@ jobs:
 
     steps:
       - uses: actions/checkout@v4
+      - name: Install X11 dev libs
+        run: |
+          sudo apt-get update && sudo apt-get install -y libx11-dev
       - name: Install Protoc
         uses: arduino/setup-protoc@v3
       - name: Build

rust/Cargo.lock

@@ -5,6 +5,8 @@ grpc:
   cuebot_endpoints: ["0.0.0.0:4343", "0.0.0.0:4343"]
   connection_expires_after: 15m
 machine:
+  # nimby_mode: true
+  # nimby_idle_threshold: 60s
   worker_threads: 8
   facility: test
   monitor_interval: 3s

rust/config/rqd.fake_linux.yaml

@@ -30,7 +30,7 @@ grpc:
 
   # Cuebot server endpoints (can specify multiple for failover)
   # Default: ["localhost:4343"]
-  cuebot_endpoints: ["cuebot.example.com:8082"]  # Replace with actual endpoints
+  cuebot_endpoints: ["cuebot.example.com:8082"] # Replace with actual endpoints
 
   # How long gRPC connections should remain active before expiring
   # Default: 1h (3600 seconds)
@@ -69,6 +69,23 @@ machine:
   # Default: false
   # nimby_mode: false
 
+  # How long a system must be idle before triggering NIMBY lock
+  # Default: 15 minutes
+  # nimby_idle_threshold: 900s
+
+  # Path to a file containing the DISPLAY number (e.g., "username:0")
+  # This file should contain a single line with the display number and writing it
+  # is not a responsibility of this module
+  # nimby_display_file_path: "/tmp/DISPLAY"
+
+  # How often to retry starting NIMBY mode if it fails
+  # Default: 5 minutes
+  # nimby_start_retry_interval: 300s
+
+  # Path to the Xauthority file path
+  # This file by default is located at user home directory
+  # nimby_display_xauthority_path: "/home/{username}/Xauthority"
+
   # Custom tags to apply to this render node for job targeting
   # Default: []
   # custom_tags: ["gpu", "high_memory"]

rust/config/rqd.yaml

@@ -61,6 +61,7 @@ tower-service = "0.3.3"
 http-body-util = "0.1.3"
 rand = "0.9.1"
 libc = "0.2"
+device_query = "3.0"
 
 [dev-dependencies]
 tempfile = "3.14.0"

rust/crates/rqd/Cargo.toml

@@ -5,7 +5,7 @@ After=network.target
 
 [Service]
 Environment=OPENCUE_RQD_CONFIG=/etc/openrqd/rqd.yaml
-ExecStart=/usr/bin/openrqd
+ExecStart=/usr/local/bin/openrqd
 LimitNOFILE=500000
 LimitNPROC=500000
 StandardOutput=journal+console

rust/crates/rqd/resources/openrqd.service

@@ -73,6 +73,12 @@ pub struct MachineConfig {
     pub temp_path: String,
     pub core_multiplier: u32,
     pub worker_threads: usize,
+    #[serde(with = "humantime_serde")]
+    pub nimby_idle_threshold: Duration,
+    pub nimby_display_file_path: Option<String>,
+    #[serde(with = "humantime_serde")]
+    pub nimby_start_retry_interval: Duration,
+    pub nimby_display_xauthority_path: String,
 }
 
 impl Default for MachineConfig {
@@ -91,6 +97,10 @@ impl Default for MachineConfig {
             temp_path: "/tmp".to_string(),
             core_multiplier: 100,
             worker_threads: 4,
+            nimby_idle_threshold: Duration::from_secs(60 * 15), // 15 min
+            nimby_display_file_path: None,
+            nimby_start_retry_interval: Duration::from_secs(60 * 5), // 5 min
+            nimby_display_xauthority_path: "/home/{username}/Xauthority".to_string(),
         }
     }
 }
@@ -123,6 +133,7 @@ pub struct RunnerConfig {
 pub enum LoggerType {
     #[serde(rename = "file")]
     File,
+    // This is a placeholder for new logging solutions
     // #[serde(rename = "loki")]
     // Loki,
 }
@@ -247,7 +258,7 @@ impl Config {
     pub fn load_file_and_env<P: AsRef<str>>(path: P) -> Result<Self, RqdConfigError> {
         let config = ConfigBase::builder()
             .add_source(File::with_name(path.as_ref()))
-            .add_source(Environment::with_prefix("VNPM").separator("_"))
+            .add_source(Environment::with_prefix("RQD").separator("_"))
             .build();
 
         config

rust/crates/rqd/src/config/config.rs

@@ -853,11 +853,6 @@ impl SystemManager for LinuxSystem {
         &self.attributes
     }
 
-    fn init_nimby(&self) -> Result<bool> {
-        // TODO: missing implementation, returning dummy val
-        Ok(false)
-    }
-
     fn collect_gpu_stats(&self) -> MachineGpuStats {
         // TODO: missing implementation, returning dummy val
         MachineGpuStats {