[gateway] ingest sensor measurements from SPs into oximeter

clients/nexus-client/src/lib.rs

@@ -213,6 +213,7 @@ impl From<omicron_common::api::internal::nexus::ProducerKind>
     fn from(kind: omicron_common::api::internal::nexus::ProducerKind) -> Self {
         use omicron_common::api::internal::nexus::ProducerKind;
         match kind {
+            ProducerKind::ManagementGateway => Self::ManagementGateway,
             ProducerKind::SledAgent => Self::SledAgent,
             ProducerKind::Service => Self::Service,
             ProducerKind::Instance => Self::Instance,
@@ -390,6 +391,9 @@ impl From<types::ProducerKind>
     fn from(kind: types::ProducerKind) -> Self {
         use omicron_common::api::internal::nexus::ProducerKind;
         match kind {
+            types::ProducerKind::ManagementGateway => {
+                ProducerKind::ManagementGateway
+            }
             types::ProducerKind::SledAgent => ProducerKind::SledAgent,
             types::ProducerKind::Instance => ProducerKind::Instance,
             types::ProducerKind::Service => ProducerKind::Service,

clients/oximeter-client/src/lib.rs

@@ -26,6 +26,7 @@ impl From<omicron_common::api::internal::nexus::ProducerKind>
     fn from(kind: omicron_common::api::internal::nexus::ProducerKind) -> Self {
         use omicron_common::api::internal::nexus;
         match kind {
+            nexus::ProducerKind::ManagementGateway => Self::ManagementGateway,
             nexus::ProducerKind::Service => Self::Service,
             nexus::ProducerKind::SledAgent => Self::SledAgent,
             nexus::ProducerKind::Instance => Self::Instance,

common/src/api/internal/nexus.rs

@@ -223,6 +223,8 @@ pub enum ProducerKind {
     Service,
     /// The producer is a Propolis VMM managing a guest instance.
     Instance,
+    /// The producer is a management gateway service.
+    ManagementGateway,
 }
 
 /// Information announced by a metric server, used so that clients can contact it and collect

dev-tools/mgs-dev/Cargo.toml

@@ -14,6 +14,7 @@ futures.workspace = true
 gateway-messages.workspace = true
 gateway-test-utils.workspace = true
 libc.workspace = true
+omicron-gateway.workspace = true
 omicron-workspace-hack.workspace = true
 signal-hook-tokio.workspace = true
 tokio.workspace = true

dev-tools/mgs-dev/src/main.rs

@@ -8,6 +8,7 @@ use clap::{Args, Parser, Subcommand};
 use futures::StreamExt;
 use libc::SIGINT;
 use signal_hook_tokio::Signals;
+use std::net::SocketAddr;
 
 #[tokio::main]
 async fn main() -> anyhow::Result<()> {
@@ -36,7 +37,12 @@ enum MgsDevCmd {
 }
 
 #[derive(Clone, Debug, Args)]
-struct MgsRunArgs {}
+struct MgsRunArgs {
+    /// Override the address of the Nexus instance to use when registering the
+    /// Oximeter producer.
+    #[clap(long)]
+    nexus_address: Option<SocketAddr>,
+}
 
 impl MgsRunArgs {
     async fn exec(&self) -> Result<(), anyhow::Error> {
@@ -46,9 +52,22 @@ impl MgsRunArgs {
         let mut signal_stream = signals.fuse();
 
         println!("mgs-dev: setting up MGS ... ");
-        let gwtestctx = gateway_test_utils::setup::test_setup(
+        let (mut mgs_config, sp_sim_config) =
+            gateway_test_utils::setup::load_test_config();
+        if let Some(addr) = self.nexus_address {
+            mgs_config.metrics.dev =
+                Some(omicron_gateway::metrics::DevConfig {
+                    bind_loopback: true,
+                    nexus_address: Some(addr),
+                });
+        }
+
+        let gwtestctx = gateway_test_utils::setup::test_setup_with_config(
             "mgs-dev",
             gateway_messages::SpPort::One,
+            mgs_config,
+            &sp_sim_config,
+            None,
         )
         .await;
         println!("mgs-dev: MGS is running.");

dev-tools/omdb/tests/successes.out

@@ -141,9 +141,16 @@ SP DETAILS: type "Sled" slot 0
 
     COMPONENTS
 
-        NAME         DESCRIPTION             DEVICE          PRESENCE SERIAL 
-        sp3-host-cpu FAKE host cpu           sp3-host-cpu    Present  None   
-        dev-0        FAKE temperature sensor fake-tmp-sensor Failed   None   
+        NAME         DESCRIPTION                              DEVICE          PRESENCE SERIAL 
+        sp3-host-cpu FAKE host cpu                            sp3-host-cpu    Present  None   
+        dev-0        FAKE temperature sensor                  fake-tmp-sensor Failed   None   
+        dev-1        FAKE temperature sensor                  tmp117          Present  None   
+        dev-2        FAKE Southeast temperature sensor        tmp117          Present  None   
+        dev-6        FAKE U.2 Sharkfin A VPD                  at24csw080      Present  None   
+        dev-7        FAKE U.2 Sharkfin A hot swap controller  max5970         Present  None   
+        dev-8        FAKE U.2 A NVMe Basic Management Command nvme_bmc        Present  None   
+        dev-39       FAKE T6 temperature sensor               tmp451          Present  None   
+        dev-53       FAKE Fan controller                      max31790        Present  None   
 
     CABOOSES: none found
 
@@ -167,8 +174,16 @@ SP DETAILS: type "Sled" slot 1
 
     COMPONENTS
 
-        NAME         DESCRIPTION   DEVICE       PRESENCE SERIAL 
-        sp3-host-cpu FAKE host cpu sp3-host-cpu Present  None   
+        NAME         DESCRIPTION                              DEVICE       PRESENCE SERIAL 
+        sp3-host-cpu FAKE host cpu                            sp3-host-cpu Present  None   
+        dev-0        FAKE temperature sensor                  tmp117       Present  None   
+        dev-1        FAKE temperature sensor                  tmp117       Present  None   
+        dev-2        FAKE Southeast temperature sensor        tmp117       Present  None   
+        dev-6        FAKE U.2 Sharkfin A VPD                  at24csw080   Present  None   
+        dev-7        FAKE U.2 Sharkfin A hot swap controller  max5970      Present  None   
+        dev-8        FAKE U.2 A NVMe Basic Management Command nvme_bmc     Present  None   
+        dev-39       FAKE T6 temperature sensor               tmp451       Present  None   
+        dev-53       FAKE Fan controller                      max31790     Present  None   
 
     CABOOSES: none found
 

gateway-test-utils/configs/config.test.toml

@@ -88,6 +88,17 @@ addr = "[::1]:0"
 ignition-target = 3
 location = { switch0 = ["sled", 1], switch1 = ["sled", 1] }
 
+#
+# Configuration for SP sensor metrics polling
+#
+[metrics]
+# Bryan wants to try polling SP sensors at 1Hz.
+sp_poll_interval_ms = 1000
+# Tell Oximeter to collect our metrics every 10 seconds.
+oximeter_collection_interval_secs = 10
+# Allow binding the metrics server on localhost.
+dev = { bind_loopback = true }
+
 #
 # NOTE: for the test suite, if mode = "file", the file path MUST be the sentinel
 # string "UNUSED".  The actual path will be generated by the test suite for each

gateway-test-utils/configs/sp_sim_config.test.toml

@@ -20,13 +20,19 @@ device = "fake-tmp-sensor"
 description = "FAKE temperature sensor 1"
 capabilities = 0x2
 presence = "Present"
+sensors = [
+    {name = "Southwest", kind = "Temperature", last_data.value = 41.7890625, last_data.timestamp = 1234 },
+]
 
 [[simulated_sps.sidecar.components]]
 id = "dev-1"
 device = "fake-tmp-sensor"
 description = "FAKE temperature sensor 2"
 capabilities = 0x2
 presence = "Failed"
+sensors = [
+    { name = "South", kind = "Temperature", last_error.value = "DeviceError", last_error.timestamp = 1234 },
+]
 
 [[simulated_sps.sidecar]]
 multicast_addr = "::1"
@@ -56,6 +62,82 @@ device = "fake-tmp-sensor"
 description = "FAKE temperature sensor"
 capabilities = 0x2
 presence = "Failed"
+sensors = [
+    { name = "Southwest", kind = "Temperature", last_error.value = "DeviceError", last_error.timestamp = 1234 },
+]
+[[simulated_sps.gimlet.components]]
+id = "dev-1"
+device = "tmp117"
+description = "FAKE temperature sensor"
+capabilities = 0x2
+presence = "Present"
+sensors = [
+    { name = "South", kind = "Temperature", last_data.value = 42.5625, last_data.timestamp = 1234 },
+]
+
+[[simulated_sps.gimlet.components]]
+id = "dev-2"
+device = "tmp117"
+description = "FAKE Southeast temperature sensor"
+capabilities = 0x2
+presence = "Present"
+sensors = [
+    { name = "Southeast", kind = "Temperature", last_data.value = 41.570313, last_data.timestamp = 1234 },
+]
+
+[[simulated_sps.gimlet.components]]
+id = "dev-6"
+device = "at24csw080"
+description = "FAKE U.2 Sharkfin A VPD"
+capabilities = 0x0
+presence = "Present"
+
+[[simulated_sps.gimlet.components]]
+id = "dev-7"
+device = "max5970"
+description = "FAKE U.2 Sharkfin A hot swap controller"
+capabilities = 0x2
+presence = "Present"
+sensors = [
+    { name = "V12_U2A_A0", kind = "Current", last_data.value = 0.45898438, last_data.timestamp = 1234 },
+    { name = "V3P3_U2A_A0", kind = "Current", last_data.value = 0.024414063, last_data.timestamp = 1234 },
+    { name = "V12_U2A_A0", kind = "Voltage", last_data.value = 12.03125, last_data.timestamp = 1234 },
+    { name = "V3P3_U2A_A0", kind = "Voltage", last_data.value = 3.328125, last_data.timestamp = 1234 },
+]
+
+[[simulated_sps.gimlet.components]]
+id = "dev-8"
+device = "nvme_bmc"
+description = "FAKE U.2 A NVMe Basic Management Command"
+capabilities = 0x2
+presence = "Present"
+sensors = [
+    { name = "U2_N0", kind = "Temperature", last_data.value = 56.0, last_data.timestamp = 1234 },
+]
+[[simulated_sps.gimlet.components]]
+id = "dev-39"
+device = "tmp451"
+description = "FAKE T6 temperature sensor"
+capabilities = 0x2
+presence = "Present"
+sensors = [
+    { name = "t6", kind = "Temperature", last_data.value = 70.625, last_data.timestamp = 1234 },
+]
+[[simulated_sps.gimlet.components]]
+id = "dev-53"
+device = "max31790"
+description = "FAKE Fan controller"
+capabilities = 0x2
+presence = "Present"
+sensors = [
+    { name = "Southeast", kind = "Speed", last_data.value = 2607.0, last_data.timestamp = 1234 },
+    { name = "Northeast", kind = "Speed", last_data.value = 2476.0, last_data.timestamp = 1234 },
+    { name = "South", kind = "Speed", last_data.value = 2553.0, last_data.timestamp = 1234 },
+    { name = "North", kind = "Speed", last_data.value = 2265.0, last_data.timestamp = 1234 },
+    { name = "Southwest", kind = "Speed", last_data.value = 2649.0, last_data.timestamp = 1234 },
+    { name = "Northwest", kind = "Speed", last_data.value = 2275.0, last_data.timestamp = 1234 },
+]
+
 
 [[simulated_sps.gimlet]]
 multicast_addr = "::1"
@@ -72,6 +154,90 @@ capabilities = 0
 presence = "Present"
 serial_console = "[::1]:0"
 
+
+[[simulated_sps.gimlet.components]]
+id = "dev-0"
+device = "tmp117"
+description = "FAKE temperature sensor"
+capabilities = 0x2
+presence = "Present"
+sensors = [
+    { name = "Southwest", kind = "Temperature", last_data.value = 41.3629, last_data.timestamp = 1234 },
+]
+[[simulated_sps.gimlet.components]]
+id = "dev-1"
+device = "tmp117"
+description = "FAKE temperature sensor"
+capabilities = 0x2
+presence = "Present"
+sensors = [
+    { name = "South", kind = "Temperature", last_data.value = 42.5625, last_data.timestamp = 1234 },
+]
+
+[[simulated_sps.gimlet.components]]
+id = "dev-2"
+device = "tmp117"
+description = "FAKE Southeast temperature sensor"
+capabilities = 0x2
+presence = "Present"
+sensors = [
+    { name = "Southeast", kind = "Temperature", last_data.value = 41.570313, last_data.timestamp = 1234 },
+]
+
+[[simulated_sps.gimlet.components]]
+id = "dev-6"
+device = "at24csw080"
+description = "FAKE U.2 Sharkfin A VPD"
+capabilities = 0x0
+presence = "Present"
+
+[[simulated_sps.gimlet.components]]
+id = "dev-7"
+device = "max5970"
+description = "FAKE U.2 Sharkfin A hot swap controller"
+capabilities = 0x2
+presence = "Present"
+sensors = [
+    { name = "V12_U2A_A0", kind = "Current", last_data.value = 0.41893438, last_data.timestamp = 1234 },
+    { name = "V3P3_U2A_A0", kind = "Current", last_data.value = 0.025614603, last_data.timestamp = 1234 },
+    { name = "V12_U2A_A0", kind = "Voltage", last_data.value = 12.02914, last_data.timestamp = 1234 },
+    { name = "V3P3_U2A_A0", kind = "Voltage", last_data.value = 3.2618, last_data.timestamp = 1234 },
+]
+
+[[simulated_sps.gimlet.components]]
+id = "dev-8"
+device = "nvme_bmc"
+description = "FAKE U.2 A NVMe Basic Management Command"
+capabilities = 0x2
+presence = "Present"
+sensors = [
+    { name = "U2_N0", kind = "Temperature", last_data.value = 56.0, last_data.timestamp = 1234 },
+]
+[[simulated_sps.gimlet.components]]
+id = "dev-39"
+device = "tmp451"
+description = "FAKE T6 temperature sensor"
+capabilities = 0x2
+presence = "Present"
+sensors = [
+    { name = "t6", kind = "Temperature", last_data.value = 70.625, last_data.timestamp = 1234 },
+]
+[[simulated_sps.gimlet.components]]
+id = "dev-53"
+device = "max31790"
+description = "FAKE Fan controller"
+capabilities = 0x2
+presence = "Present"
+sensors = [
+    { name = "Southeast", kind = "Speed", last_data.value = 2510.0, last_data.timestamp = 1234 },
+    { name = "Northeast", kind = "Speed", last_data.value = 2390.0, last_data.timestamp = 1234 },
+    { name = "South", kind = "Speed", last_data.value = 2467.0, last_data.timestamp = 1234 },
+    { name = "North", kind = "Speed", last_data.value = 2195.0, last_data.timestamp = 1234 },
+    { name = "Southwest", kind = "Speed", last_data.value = 2680.0, last_data.timestamp = 1234 },
+    { name = "Northwest", kind = "Speed", last_data.value = 2212.0, last_data.timestamp = 1234 },
+]
+
+
 #
 # NOTE: for the test suite, the [log] section is ignored; sp-sim logs are rolled
 # into the gateway logfile.