[reconfigurator] Pre-checks and post_update actions for RoT bootloader update

clients/gateway-client/Cargo.toml

@@ -21,5 +21,6 @@ serde.workspace = true
 serde_json.workspace = true
 schemars.workspace = true
 slog.workspace = true
+thiserror.workspace = true
 uuid.workspace = true
 omicron-workspace-hack.workspace = true

clients/gateway-client/src/lib.rs

@@ -63,7 +63,7 @@ progenitor::generate_api!(
         HostPhase2RecoveryImageId = { derives = [PartialEq, Eq, PartialOrd, Ord] },
         ImageVersion = { derives = [PartialEq, Eq, PartialOrd, Ord] },
         RotImageDetails = { derives = [PartialEq, Eq, PartialOrd, Ord] },
-        RotImageError = { derives = [ PartialEq, Eq, PartialOrd, Ord] },
+        RotImageError = { derives = [ thiserror::Error, PartialEq, Eq, PartialOrd, Ord] },
         RotState = { derives = [PartialEq, Eq, PartialOrd, Ord] },
         SpComponentCaboose = { derives = [PartialEq, Eq] },
         SpIdentifier = { derives = [Copy, PartialEq, Hash, Eq] },

nexus/mgs-updates/Cargo.toml

@@ -11,6 +11,7 @@ chrono.workspace = true
 futures.workspace = true
 gateway-client.workspace = true
 gateway-types.workspace = true
+gateway-messages.workspace = true
 id-map.workspace = true
 internal-dns-resolver.workspace = true
 internal-dns-types.workspace = true

nexus/mgs-updates/src/common_sp_update.rs

@@ -267,14 +267,15 @@ pub trait SpComponentUpdateHelper {
         log: &'a slog::Logger,
         mgs_clients: &'a mut MgsClients,
         update: &'a PendingMgsUpdate,
-    ) -> BoxFuture<'a, Result<(), GatewayClientError>>;
+    ) -> BoxFuture<'a, Result<(), PostUpdateError>>;
 }
 
 /// Describes the live state of the component before the update begins
 #[derive(Debug)]
 pub enum PrecheckStatus {
     UpdateComplete,
     ReadyForUpdate,
+    WaitingForOngoingUpdate,
 }
 
 #[derive(Debug, Error)]
@@ -319,6 +320,30 @@ pub enum PrecheckError {
     WrongInactiveVersion { expected: ExpectedVersion, found: FoundVersion },
 }
 
+#[derive(Debug, thiserror::Error)]
+pub enum PostUpdateError {
+    #[error("communicating with MGS")]
+    GatewayClientError(#[from] GatewayClientError),
+
+    #[error("transient error: {message:?}")]
+    TransientError { message: String },
+
+    #[error("fatal error: {error:?}")]
+    FatalError { error: String },
+}
+
+impl PostUpdateError {
+    pub fn is_fatal(&self) -> bool {
+        match self {
+            PostUpdateError::GatewayClientError(error) => {
+                !matches!(error, gateway_client::Error::CommunicationError(_))
+            }
+            PostUpdateError::TransientError { .. } => false,
+            PostUpdateError::FatalError { .. } => true,
+        }
+    }
+}
+
 #[derive(Debug)]
 pub enum FoundVersion {
     MissingVersion,