Sled/instance association for inventory and insights (#3092)

common/src/api/external/mod.rs

@@ -705,6 +705,7 @@ pub enum ResourceType {
     Rack,
     Service,
     Sled,
+    SledInstance,
     Switch,
     SagaDbg,
     Snapshot,

common/src/sql/dbinit.sql

@@ -818,13 +818,14 @@ CREATE TABLE omicron.public.instance (
     time_state_updated TIMESTAMPTZ NOT NULL,
     state_generation INT NOT NULL,
     /*
-     * Server where the VM is currently running, if any.  Note that when we
-     * support live migration, there may be multiple servers associated with
+     * Sled where the VM is currently running, if any.  Note that when we
+     * support live migration, there may be multiple sleds associated with
      * this Instance, but only one will be truly active.  Still, consumers of
      * this information should consider whether they also want to know the other
-     * servers involved in the migration.
+     * sleds involved in the migration.
      */
-    active_server_id UUID,
+    active_sled_id UUID,
+
     /* Identifies the underlying propolis-server backing the instance. */
     active_propolis_id UUID NOT NULL,
     active_propolis_ip INET,
@@ -862,10 +863,38 @@ CREATE UNIQUE INDEX ON omicron.public.instance (
 -- Allow looking up instances by server. This is particularly
 -- useful for resource accounting within a sled.
 CREATE INDEX ON omicron.public.instance (
-    active_server_id
+    active_sled_id
 ) WHERE
     time_deleted IS NULL;
 
+/*
+ * A special view of an instance provided to operators for insights into what's running 
+ * on a sled.
+ */
+
+CREATE VIEW omicron.public.sled_instance 
+AS SELECT
+   instance.id,
+   instance.name,
+   silo.name as silo_name,
+   project.name as project_name,
+   instance.active_sled_id,
+   instance.time_created,
+   instance.time_modified,
+   instance.migration_id,
+   instance.ncpus,
+   instance.memory,
+   instance.state
+FROM
+    omicron.public.instance AS instance
+    JOIN omicron.public.project AS project ON
+            instance.project_id = project.id
+    JOIN omicron.public.silo AS silo ON
+            project.silo_id = silo.id
+WHERE
+    instance.time_deleted IS NULL;
+
+
 /*
  * Guest-Visible, Virtual Disks
  */

nexus/db-model/src/instance.rs

@@ -124,7 +124,7 @@ pub struct InstanceRuntimeState {
     //
     // TODO(#2315): This should be optional so that it can be cleared when the
     // instance is not active.
-    #[diesel(column_name = active_server_id)]
+    #[diesel(column_name = active_sled_id)]
     pub sled_id: Uuid,
     /// The ID of the Propolis server hosting the current incarnation of this
     /// instance.

nexus/db-model/src/lib.rs

@@ -62,6 +62,7 @@ mod silo_group;
 mod silo_user;
 mod silo_user_password_hash;
 mod sled;
+mod sled_instance;
 mod sled_resource;
 mod sled_resource_kind;
 mod snapshot;
@@ -136,6 +137,7 @@ pub use silo_group::*;
 pub use silo_user::*;
 pub use silo_user_password_hash::*;
 pub use sled::*;
+pub use sled_instance::*;
 pub use sled_resource::*;
 pub use sled_resource_kind::*;
 pub use snapshot::*;

nexus/db-model/src/schema.rs

@@ -123,7 +123,7 @@ table! {
         state -> crate::InstanceStateEnum,
         time_state_updated -> Timestamptz,
         state_generation -> Int8,
-        active_server_id -> Uuid,
+        active_sled_id -> Uuid,
         active_propolis_id -> Uuid,
         active_propolis_ip -> Nullable<Inet>,
         target_propolis_id -> Nullable<Uuid>,
@@ -135,6 +135,22 @@ table! {
     }
 }
 
+table! {
+    sled_instance (id) {
+        id -> Uuid,
+        name -> Text,
+        silo_name -> Text,
+        project_name -> Text,
+        time_created -> Timestamptz,
+        time_modified -> Timestamptz,
+        state -> crate::InstanceStateEnum,
+        active_sled_id -> Uuid,
+        migration_id -> Nullable<Uuid>,
+        ncpus -> Int8,
+        memory -> Int8,
+    }
+}
+
 table! {
     metric_producer (id) {
         id -> Uuid,

nexus/db-model/src/sled_instance.rs

@@ -0,0 +1,43 @@
+use crate::schema::sled_instance;
+use crate::InstanceState;
+use crate::Name;
+use db_macros::Asset;
+use nexus_types::external_api::views;
+use nexus_types::identity::Asset;
+use serde::Deserialize;
+use serde::Serialize;
+use uuid::Uuid;
+
+/// An operator view of an instance as exposed by the sled API.
+#[derive(Queryable, Debug, Selectable, Asset, Serialize, Deserialize)]
+#[diesel(table_name = sled_instance)]
+pub struct SledInstance {
+    #[diesel(embed)]
+    identity: SledInstanceIdentity,
+    active_sled_id: Uuid,
+    pub migration_id: Option<Uuid>,
+
+    pub name: Name,
+    pub silo_name: Name,
+    pub project_name: Name,
+
+    pub state: InstanceState,
+    pub ncpus: i64,
+    pub memory: i64,
+}
+
+impl From<SledInstance> for views::SledInstance {
+    fn from(sled_instance: SledInstance) -> Self {
+        Self {
+            identity: sled_instance.identity(),
+            name: sled_instance.name.into(),
+            active_sled_id: sled_instance.active_sled_id,
+            silo_name: sled_instance.silo_name.into(),
+            project_name: sled_instance.project_name.into(),
+            state: *sled_instance.state.state(),
+            migration_id: sled_instance.migration_id,
+            ncpus: sled_instance.ncpus,
+            memory: sled_instance.memory,
+        }
+    }
+}

nexus/db-queries/src/authz/api_resources.rs

@@ -974,6 +974,14 @@ authz_resource! {
     polar_snippet = FleetChild,
 }
 
+authz_resource! {
+    name = "SledInstance",
+    parent = "Fleet",
+    primary_key = Uuid,
+    roles_allowed = false,
+    polar_snippet = FleetChild,
+}
+
 authz_resource! {
     name = "Switch",
     parent = "Fleet",

nexus/db-queries/src/db/datastore/mod.rs

@@ -66,6 +66,7 @@ mod silo;
 mod silo_group;
 mod silo_user;
 mod sled;
+mod sled_instance;
 mod snapshot;
 mod ssh_key;
 mod switch;

nexus/db-queries/src/db/datastore/sled_instance.rs

@@ -0,0 +1,32 @@
+use super::DataStore;
+
+use crate::authz;
+use crate::context::OpContext;
+use crate::db;
+use crate::db::error::public_error_from_diesel_pool;
+use crate::db::error::ErrorHandler;
+use crate::db::pagination::paginated;
+use async_bb8_diesel::AsyncRunQueryDsl;
+use diesel::prelude::*;
+use nexus_db_model::SledInstance;
+use omicron_common::api::external::DataPageParams;
+use omicron_common::api::external::ListResultVec;
+use uuid::Uuid;
+
+impl DataStore {
+    pub async fn sled_instance_list(
+        &self,
+        opctx: &OpContext,
+        authz_sled: &authz::Sled,
+        pagparams: &DataPageParams<'_, Uuid>,
+    ) -> ListResultVec<SledInstance> {
+        opctx.authorize(authz::Action::ListChildren, &authz::FLEET).await?;
+        use db::schema::sled_instance::dsl;
+        paginated(dsl::sled_instance, dsl::id, &pagparams)
+            .filter(dsl::active_sled_id.eq(authz_sled.id()))
+            .select(SledInstance::as_select())
+            .load_async::<SledInstance>(self.pool_authorized(opctx).await?)
+            .await
+            .map_err(|e| public_error_from_diesel_pool(e, ErrorHandler::Server))
+    }
+}

nexus/db-queries/src/db/datastore/vpc.rs

@@ -363,7 +363,7 @@ impl DataStore {
                     .on(instance::id
                         .eq(instance_network_interface::instance_id)),
             )
-            .inner_join(sled::table.on(sled::id.eq(instance::active_server_id)))
+            .inner_join(sled::table.on(sled::id.eq(instance::active_sled_id)))
             .filter(instance_network_interface::vpc_id.eq(vpc_id))
             .filter(instance_network_interface::time_deleted.is_null())
             .filter(instance::time_deleted.is_null())

nexus/db-queries/src/db/lookup.rs

@@ -760,6 +760,15 @@ lookup_resource! {
     primary_key_columns = [ { column_name = "id", rust_type = Uuid } ]
 }
 
+lookup_resource! {
+    name = "SledInstance",
+    ancestors = [],
+    children = [],
+    lookup_by_name = false,
+    soft_deletes = false,
+    primary_key_columns = [ { column_name = "id", rust_type = Uuid } ],
+}
+
 lookup_resource! {
     name = "Switch",
     ancestors = [],

nexus/src/app/mod.rs

@@ -44,6 +44,7 @@ pub mod saga;
 mod session;
 mod silo;
 mod sled;
+mod sled_instance;
 mod snapshot;
 mod switch;
 pub mod test_interfaces;

nexus/src/app/sled.rs

@@ -15,6 +15,7 @@ use crate::internal_api::params::{
     SledRole, ZpoolPutRequest,
 };
 use nexus_db_queries::context::OpContext;
+use nexus_db_queries::db::lookup;
 use omicron_common::api::external::DataPageParams;
 use omicron_common::api::external::Error;
 use omicron_common::api::external::ListResultVec;
@@ -27,6 +28,14 @@ use uuid::Uuid;
 
 impl super::Nexus {
     // Sleds
+    pub fn sled_lookup<'a>(
+        &'a self,
+        opctx: &'a OpContext,
+        sled_id: &Uuid,
+    ) -> LookupResult<lookup::Sled<'a>> {
+        let sled = LookupPath::new(opctx, &self.db_datastore).sled_id(*sled_id);
+        Ok(sled)
+    }
 
     // TODO-robustness we should have a limit on how many sled agents there can
     // be (for graceful degradation at large scale).
@@ -61,26 +70,14 @@ impl super::Nexus {
         Ok(())
     }
 
-    pub async fn sleds_list(
+    pub async fn sled_list(
         &self,
         opctx: &OpContext,
         pagparams: &DataPageParams<'_, Uuid>,
     ) -> ListResultVec<db::model::Sled> {
         self.db_datastore.sled_list(&opctx, pagparams).await
     }
 
-    pub async fn sled_lookup(
-        &self,
-        opctx: &OpContext,
-        sled_id: &Uuid,
-    ) -> LookupResult<db::model::Sled> {
-        let (.., db_sled) = LookupPath::new(opctx, &self.db_datastore)
-            .sled_id(*sled_id)
-            .fetch()
-            .await?;
-        Ok(db_sled)
-    }
-
     pub async fn sled_client(
         &self,
         id: &Uuid,
@@ -92,7 +89,8 @@ impl super::Nexus {
         // Franky, returning an "Arc" here without a connection pool is a little
         // silly; it's not actually used if each client connection exists as a
         // one-shot.
-        let sled = self.sled_lookup(&self.opctx_alloc, id).await?;
+        let (.., sled) =
+            self.sled_lookup(&self.opctx_alloc, id)?.fetch().await?;
 
         let log = self.log.new(o!("SledAgent" => id.clone().to_string()));
         let dur = std::time::Duration::from_secs(60);
@@ -347,7 +345,7 @@ impl super::Nexus {
 
         // Look up the supplied sled's physical host IP.
         let physical_host_ip =
-            *self.sled_lookup(&self.opctx_alloc, &sled_id).await?.ip;
+            *self.sled_lookup(&self.opctx_alloc, &sled_id)?.fetch().await?.1.ip;
 
         let mut last_sled_id: Option<Uuid> = None;
         loop {
@@ -358,7 +356,7 @@ impl super::Nexus {
             };
 
             let sleds_page =
-                self.sleds_list(&self.opctx_alloc, &pagparams).await?;
+                self.sled_list(&self.opctx_alloc, &pagparams).await?;
             let mut join_handles =
                 Vec::with_capacity(sleds_page.len() * instance_nics.len());
 
@@ -444,8 +442,12 @@ impl super::Nexus {
 
         // Lookup the physical host IP of the sled hosting this instance
         let instance_sled_id = db_instance.runtime().sled_id;
-        let physical_host_ip =
-            *self.sled_lookup(&self.opctx_alloc, &instance_sled_id).await?.ip;
+        let physical_host_ip = *self
+            .sled_lookup(&self.opctx_alloc, &instance_sled_id)?
+            .fetch()
+            .await?
+            .1
+            .ip;
 
         let mut last_sled_id: Option<Uuid> = None;
 
@@ -457,7 +459,7 @@ impl super::Nexus {
             };
 
             let sleds_page =
-                self.sleds_list(&self.opctx_alloc, &pagparams).await?;
+                self.sled_list(&self.opctx_alloc, &pagparams).await?;
             let mut join_handles =
                 Vec::with_capacity(sleds_page.len() * instance_nics.len());
 

nexus/src/app/sled_instance.rs

@@ -0,0 +1,23 @@
+use crate::authz;
+use crate::db;
+use nexus_db_queries::context::OpContext;
+use nexus_db_queries::db::lookup;
+use omicron_common::api::external::DataPageParams;
+use omicron_common::api::external::ListResultVec;
+use uuid::Uuid;
+
+impl super::Nexus {
+    pub async fn sled_instance_list(
+        &self,
+        opctx: &OpContext,
+        sled_lookup: &lookup::Sled<'_>,
+        pagparams: &DataPageParams<'_, Uuid>,
+    ) -> ListResultVec<db::model::SledInstance> {
+        let (.., authz_sled) =
+            sled_lookup.lookup_for(authz::Action::Read).await?;
+        opctx.authorize(authz::Action::ListChildren, &authz::FLEET).await?;
+        self.db_datastore
+            .sled_instance_list(opctx, &authz_sled, pagparams)
+            .await
+    }
+}