feat(health): add NVLink domain UUID to switch telemetry

crates/health/example/config.example.toml

@@ -49,7 +49,9 @@ port = 443
 mac = "11:22:33:44:55:66"
 username = "admin"
 password = "secret"
-switch = { id = "fsw100htjtiaehv1n5vh67tbmqq4eabcjdng40f7jupsadbedhruh6rag1l0", serial = "SN-SWITCH-BMC-001", endpoint_role = "bmc", slot_number = 7, tray_index = 3 }
+# Configure the same domain UUID on every static endpoint for the switch.
+# Invalid or nil values are omitted from telemetry.
+switch = { id = "fsw100htjtiaehv1n5vh67tbmqq4eabcjdng40f7jupsadbedhruh6rag1l0", serial = "SN-SWITCH-BMC-001", endpoint_role = "bmc", slot_number = 7, tray_index = 3, nvlink_domain_uuid = "9f4b45ec-705a-4af4-89f7-a112bc9c8f4e" }
 
 [[endpoint_sources.static_bmc_endpoints]]
 ip = "10.0.1.2"
@@ -60,7 +62,7 @@ password = "secret"
 # For static switch host endpoints, nmxc_enabled controls direct NMX-C
 # Subscribe eligibility after the endpoint_role="host" and is_primary=true
 # checks. If omitted, it defaults to is_primary.
-switch = { id = "fsw100htjtiaehv1n5vh67tbmqq4eabcjdng40f7jupsadbedhruh6rag1l0", serial = "SN-SWITCH-HOST-001", endpoint_role = "host", is_primary = true, nmxc_enabled = true, slot_number = 7, tray_index = 3 }
+switch = { id = "fsw100htjtiaehv1n5vh67tbmqq4eabcjdng40f7jupsadbedhruh6rag1l0", serial = "SN-SWITCH-HOST-001", endpoint_role = "host", is_primary = true, nmxc_enabled = true, slot_number = 7, tray_index = 3, nvlink_domain_uuid = "9f4b45ec-705a-4af4-89f7-a112bc9c8f4e" }
 
 [[endpoint_sources.static_bmc_endpoints]]
 ip = "10.0.2.1"

crates/health/src/api_client.rs

@@ -25,6 +25,7 @@ use std::net::IpAddr;
 use std::str::FromStr;
 use std::sync::{Arc, Mutex};
 
+use carbide_uuid::nvlink::NvLinkDomainId;
 use carbide_uuid::rack::RackId;
 use carbide_uuid::switch::SwitchId;
 use forge_tls::client_config::ClientCert;
@@ -275,6 +276,9 @@ fn switch_endpoint_metadata(
             .placement_in_rack
             .as_ref()
             .and_then(|placement| placement.tray_index),
+        nvlink_domain_uuid: switch
+            .nvlink_domain_uuid
+            .filter(|domain_uuid| domain_uuid != &NvLinkDomainId::nil()),
         endpoint_role,
         is_primary: switch.is_primary,
         nmxc_enabled: config.enable_nmxc,
@@ -755,7 +759,8 @@ impl From<rpc::forge::bmc_credentials::Type> for BmcCredentials {
 mod tests {
     use std::sync::atomic::{AtomicUsize, Ordering};
 
-    use carbide_test_support::value_scenarios;
+    use carbide_test_support::{Check, check_values, value_scenarios};
+    use carbide_uuid::nvlink::NvLinkDomainId;
     use carbide_uuid::switch::{SwitchId, SwitchIdSource, SwitchType};
     use nv_redfish::bmc_http::reqwest::ClientParams as ReqwestClientParams;
 
@@ -847,6 +852,53 @@ mod tests {
         );
     }
 
+    #[test]
+    fn switch_endpoint_metadata_uses_non_nil_api_domain() {
+        let domain = NvLinkDomainId::from_str("9f4b45ec-705a-4af4-89f7-a112bc9c8f4e")
+            .expect("valid domain UUID");
+
+        check_values(
+            [
+                Check {
+                    scenario: "domain is missing",
+                    input: None,
+                    expect: None,
+                },
+                Check {
+                    scenario: "nil domain is absent",
+                    input: Some(NvLinkDomainId::nil()),
+                    expect: None,
+                },
+                Check {
+                    scenario: "non-nil API switch field",
+                    input: Some(domain),
+                    expect: Some(domain),
+                },
+            ],
+            |nvlink_domain_uuid| {
+                let metadata = switch_endpoint_metadata(
+                    &rpc::forge::Switch {
+                        config: Some(rpc::forge::SwitchConfig {
+                            name: "switch-a".to_string(),
+                            ..Default::default()
+                        }),
+                        nvlink_domain_uuid,
+                        ..Default::default()
+                    },
+                    SwitchEndpointRole::Bmc,
+                    false,
+                )
+                .expect("switch metadata");
+
+                let EndpointMetadata::Switch(switch) = metadata else {
+                    panic!("expected switch metadata");
+                };
+
+                switch.nvlink_domain_uuid
+            },
+        );
+    }
+
     #[tokio::test]
     async fn cache_returns_existing_client_on_matching_kind() {
         let mut cache: HashMap<MacAddress, CachedBmcClient> = HashMap::new();

crates/health/src/collectors/nvue/gnmi/on_change_processor.rs

@@ -786,6 +786,7 @@ mod tests {
                     serial: "SN-SWITCH-001".to_string(),
                     slot_number: Some(7),
                     tray_index: Some(3),
+                    nvlink_domain_uuid: None,
                     endpoint_role: SwitchEndpointRole::Host,
                     is_primary: false,
                     nmxc_enabled: false,

crates/health/src/collectors/nvue/gnmi/sample_processor.rs

@@ -1174,6 +1174,7 @@ mod tests {
                     serial: "SN-SWITCH-001".to_string(),
                     slot_number: Some(7),
                     tray_index: Some(3),
+                    nvlink_domain_uuid: None,
                     endpoint_role: SwitchEndpointRole::Host,
                     is_primary: false,
                     nmxc_enabled: false,

crates/health/src/collectors/runtime.rs

@@ -593,6 +593,12 @@ impl Collector {
         let _ = self.handle.await;
     }
 
+    /// Returns a handle that can force the collector task to stop while its
+    /// normal shutdown future remains available to join the task.
+    pub(crate) fn abort_handle(&self) -> tokio::task::AbortHandle {
+        self.handle.abort_handle()
+    }
+
     pub fn is_finished(&self) -> bool {
         self.handle.is_finished()
     }

crates/health/src/config.rs

@@ -255,6 +255,11 @@ pub struct StaticSwitchEndpoint {
     pub slot_number: Option<i32>,
     #[serde(alias = "compute_tray_index")]
     pub tray_index: Option<i32>,
+
+    /// Optional non-nil NVLink domain UUID associated with this switch.
+    /// Invalid or nil values are omitted from telemetry.
+    pub nvlink_domain_uuid: Option<String>,
+
     #[serde(default = "default_static_switch_endpoint_role")]
     pub endpoint_role: StaticSwitchEndpointRole,
     #[serde(default)]
@@ -1918,6 +1923,7 @@ mod tests {
             serial: Some("switch-serial".to_string()),
             slot_number: None,
             tray_index: None,
+            nvlink_domain_uuid: None,
             endpoint_role: StaticSwitchEndpointRole::Host,
             is_primary: false,
             nmxc_enabled: None,
@@ -3866,7 +3872,7 @@ ip = "10.0.1.2"
 mac = "11:22:33:44:55:77"
 username = "admin"
 password = "pass"
-switch = { id = "fsw100htjtiaehv1n5vh67tbmqq4eabcjdng40f7jupsadbedhruh6rag1l0", serial = "SN-SW-002", endpoint_role = "host", is_primary = false, nmxc_enabled = true, nmxt_enabled = true }
+switch = { id = "fsw100htjtiaehv1n5vh67tbmqq4eabcjdng40f7jupsadbedhruh6rag1l0", serial = "SN-SW-002", endpoint_role = "host", is_primary = false, nmxc_enabled = true, nmxt_enabled = true, nvlink_domain_uuid = "9f4b45ec-705a-4af4-89f7-a112bc9c8f4e" }
 "#;
 
         let config: Config = Figment::new()
@@ -3884,6 +3890,11 @@ switch = { id = "fsw100htjtiaehv1n5vh67tbmqq4eabcjdng40f7jupsadbedhruh6rag1l0",
         assert!(!switch.is_primary);
         assert_eq!(switch.nmxc_enabled, Some(true));
         assert_eq!(switch.nmxt_enabled, Some(true));
+
+        assert_eq!(
+            switch.nvlink_domain_uuid.as_deref(),
+            Some("9f4b45ec-705a-4af4-89f7-a112bc9c8f4e")
+        );
     }
 
     #[test]