Publish instance vCPU usage statistics to oximeter

Cargo.toml

@@ -301,9 +301,15 @@ prettyplease = "0.2.16"
 proc-macro2 = "1.0"
 progenitor = { git = "https://github.com/oxidecomputer/progenitor", branch = "main" }
 progenitor-client = { git = "https://github.com/oxidecomputer/progenitor", branch = "main" }
-bhyve_api = { git = "https://github.com/oxidecomputer/propolis", rev = "1e25649e8c2ac274bd04adfe0513dd14a482058c" }
-propolis-client = { git = "https://github.com/oxidecomputer/propolis", rev = "1e25649e8c2ac274bd04adfe0513dd14a482058c" }
-propolis-mock-server = { git = "https://github.com/oxidecomputer/propolis", rev = "1e25649e8c2ac274bd04adfe0513dd14a482058c" }
+
+# TODO(ben): Put back to main before merge.
+#bhyve_api = { git = "https://github.com/oxidecomputer/propolis", rev = "1e25649e8c2ac274bd04adfe0513dd14a482058c" }
+#propolis-client = { git = "https://github.com/oxidecomputer/propolis", rev = "1e25649e8c2ac274bd04adfe0513dd14a482058c" }
+#propolis-mock-server = { git = "https://github.com/oxidecomputer/propolis", rev = "1e25649e8c2ac274bd04adfe0513dd14a482058c" }
+bhyve_api = { git = "https://github.com/oxidecomputer/propolis", branch = "vcpu-usage-stats" }
+propolis-client = { git = "https://github.com/oxidecomputer/propolis", branch = "vcpu-usage-stats" }
+propolis-mock-server = { git = "https://github.com/oxidecomputer/propolis", branch = "vcpu-usage-stats" }
+
 proptest = "1.4.0"
 quote = "1.0"
 rand = "0.8.5"

clients/dns-service-client/src/lib.rs

@@ -29,8 +29,10 @@ pub fn is_retryable(error: &DnsConfigError<crate::types::Error>) -> bool {
     let response_value = match error {
         DnsConfigError::CommunicationError(_) => return true,
         DnsConfigError::InvalidRequest(_)
-        | DnsConfigError::InvalidResponsePayload(_)
-        | DnsConfigError::UnexpectedResponse(_) => return false,
+        | DnsConfigError::InvalidResponsePayload(_, _)
+        | DnsConfigError::UnexpectedResponse(_)
+        | DnsConfigError::InvalidUpgrade(_)
+        | DnsConfigError::ResponseBodyError(_) => return false,
         DnsConfigError::ErrorResponse(response_value) => response_value,
     };
 

common/src/api/external/error.rs

@@ -523,6 +523,7 @@ impl<T: ClientError> From<progenitor::progenitor_client::Error<T>> for Error {
             // our version constraints (i.e. that the call was to a newer
             // service with an incompatible response).
             progenitor::progenitor_client::Error::InvalidResponsePayload(
+                _bytes,
                 ee,
             ) => Error::internal_error(&format!(
                 "InvalidResponsePayload: {}",
@@ -539,6 +540,12 @@ impl<T: ClientError> From<progenitor::progenitor_client::Error<T>> for Error {
                     r.status(),
                 ))
             }
+            progenitor::progenitor_client::Error::InvalidUpgrade(e) => {
+                Error::internal_error(&format!("InvalidUpgrade: {e}",))
+            }
+            progenitor::progenitor_client::Error::ResponseBodyError(_) => {
+                Error::internal_error(&format!("ResponseBodyError: {e}",))
+            }
         }
     }
 }

nexus/src/app/instance.rs

@@ -1135,6 +1135,21 @@ impl super::Nexus {
             .map(|ssh_key| ssh_key.public_key)
             .collect::<Vec<String>>();
 
+        // Construct instance metadata used to track its statistics.
+        //
+        // This current means fetching the current silo ID, since we have all
+        // the other metadata already.
+        let silo_id = self
+            .current_silo_lookup(opctx)?
+            .lookup_for(authz::Action::Read)
+            .await?
+            .0
+            .id();
+        let metadata = sled_agent_client::types::InstanceMetadata {
+            silo_id,
+            project_id: db_instance.project_id,
+        };
+
         // Ask the sled agent to begin the state change.  Then update the
         // database to reflect the new intermediate state.  If this update is
         // not the newest one, that's fine.  That might just mean the sled agent
@@ -1178,6 +1193,7 @@ impl super::Nexus {
                         PROPOLIS_PORT,
                     )
                     .to_string(),
+                    metadata,
                 },
             )
             .await

nexus/src/app/sagas/switch_port_settings_apply.rs

@@ -223,7 +223,7 @@ async fn spa_undo_ensure_switch_port_settings(
     let log = sagactx.user_data().log();
 
     let port_id: PortId = PortId::from_str(&params.switch_port_name)
-        .map_err(|e| external::Error::internal_error(e))?;
+        .map_err(|e| external::Error::internal_error(e.to_string().as_str()))?;
 
     let orig_port_settings_id = sagactx
         .lookup::<Option<Uuid>>("original_switch_port_settings_id")

nexus/src/app/sagas/switch_port_settings_clear.rs

@@ -179,7 +179,7 @@ async fn spa_undo_clear_switch_port_settings(
     let log = sagactx.user_data().log();
 
     let port_id: PortId = PortId::from_str(&params.port_name)
-        .map_err(|e| external::Error::internal_error(e))?;
+        .map_err(|e| external::Error::internal_error(e.to_string().as_str()))?;
 
     let orig_port_settings_id = sagactx
         .lookup::<Option<Uuid>>("original_switch_port_settings_id")

openapi/sled-agent.json

@@ -4515,6 +4515,14 @@
               }
             ]
           },
+          "metadata": {
+            "description": "Metadata used to track instance statistics.",
+            "allOf": [
+              {
+                "$ref": "#/components/schemas/InstanceMetadata"
+              }
+            ]
+          },
           "propolis_addr": {
             "description": "The address at which this VMM should serve a Propolis server API.",
             "type": "string"
@@ -4536,6 +4544,7 @@
         "required": [
           "hardware",
           "instance_runtime",
+          "metadata",
           "propolis_addr",
           "propolis_id",
           "vmm_runtime"
@@ -4624,6 +4633,24 @@
           "snapshot_id"
         ]
       },
+      "InstanceMetadata": {
+        "description": "Metadata used to track statistics about an instance.",
+        "type": "object",
+        "properties": {
+          "project_id": {
+            "type": "string",
+            "format": "uuid"
+          },
+          "silo_id": {
+            "type": "string",
+            "format": "uuid"
+          }
+        },
+        "required": [
+          "project_id",
+          "silo_id"
+        ]
+      },
       "InstanceMigrationSourceParams": {
         "description": "Instance runtime state to update for a migration.",
         "type": "object",

oximeter/instruments/Cargo.toml

@@ -18,9 +18,10 @@ uuid.workspace = true
 omicron-workspace-hack.workspace = true
 
 [features]
-default = ["http-instruments", "kstat"]
+default = ["http-instruments", "datalink", "virtual-machine"]
 http-instruments = ["http"]
-kstat = ["kstat-rs"]
+datalink = ["dep:kstat-rs"]
+virtual-machine = ["dep:kstat-rs"]
 
 [dev-dependencies]
 rand.workspace = true

oximeter/instruments/src/kstat/mod.rs

@@ -87,8 +87,11 @@ use std::cmp::Ordering;
 use std::collections::BTreeMap;
 use std::time::Duration;
 
+#[cfg(feature = "datalink")]
 pub mod link;
 mod sampler;
+#[cfg(feature = "virtual-machine")]
+pub mod virtual_machine;
 
 pub use sampler::CollectionDetails;
 pub use sampler::ExpirationBehavior;