Merge

Author: smklein

Cargo.lock

@@ -653,6 +653,7 @@ rand_distr = "0.5.1"
 rand_seeder = "0.4.0"
 range-requests = { path = "range-requests" }
 ratatui = "0.29.0"
+raw-cpuid = { git = "https://github.com/oxidecomputer/rust-cpuid.git", rev = "0a8dbd2311263f6a59ea58089e33c8331436ff3a" }
 rayon = "1.10"
 rcgen = "0.12.1"
 reconfigurator-cli = { path = "dev-tools/reconfigurator-cli" }

Cargo.toml

@@ -1197,6 +1197,10 @@ pub struct Instance {
 
     #[serde(flatten)]
     pub auto_restart_status: InstanceAutoRestartStatus,
+
+    /// The CPU platform for this instance. If this is `null`, the instance
+    /// requires no particular CPU platform.
+    pub cpu_platform: Option<InstanceCpuPlatform>,
 }
 
 /// Status of control-plane driven automatic failure recovery for this instance.
@@ -1261,6 +1265,51 @@ pub enum InstanceAutoRestartPolicy {
     BestEffort,
 }
 
+/// A required CPU platform for an instance.
+///
+/// When an instance specifies a required CPU platform:
+///
+/// - The system may expose (to the VM) new CPU features that are only present
+///   on that platform (or on newer platforms of the same lineage that also
+///   support those features).
+/// - The instance must run on hosts that have CPUs that support all the
+///   features of the supplied platform.
+///
+/// That is, the instance is restricted to hosts that have the CPUs which
+/// support all features of the required platform, but in exchange the CPU
+/// features exposed by the platform are available for the guest to use. Note
+/// that this may prevent an instance from starting (if the hosts that could run
+/// it are full but there is capacity on other incompatible hosts).
+///
+/// If an instance does not specify a required CPU platform, then when
+/// it starts, the control plane selects a host for the instance and then
+/// supplies the guest with the "minimum" CPU platform supported by that host.
+/// This maximizes the number of hosts that can run the VM if it later needs to
+/// migrate to another host.
+///
+/// In all cases, the CPU features presented by a given CPU platform are a
+/// subset of what the corresponding hardware may actually support; features
+/// which cannot be used from a virtual environment or do not have full
+/// hypervisor support may be masked off. See RFD 314 for specific CPU features
+/// in a CPU platform.
+#[derive(
+    Copy, Clone, Debug, Deserialize, Serialize, JsonSchema, Eq, PartialEq,
+)]
+#[serde(rename_all = "snake_case")]
+pub enum InstanceCpuPlatform {
+    /// An AMD Milan-like CPU platform.
+    AmdMilan,
+
+    /// An AMD Turin-like CPU platform.
+    // Note that there is only Turin, not Turin Dense - feature-wise there are
+    // collapsed together as the guest-visible platform is the same.
+    // If the two must be distinguished for instance placement, we'll want to
+    // track whatever the motivating constraint is more explicitly. CPU
+    // families, and especially the vendor code names, don't necessarily promise
+    // details about specific processor packaging choices.
+    AmdTurin,
+}
+
 // AFFINITY GROUPS
 
 /// Affinity policy used to describe "what to do when a request cannot be satisfied"

common/src/api/external/mod.rs

@@ -4918,6 +4918,7 @@ async fn cmd_db_instance_info(
                     propolis_ip: _,
                     propolis_port: _,
                     instance_id: _,
+                    cpu_platform: _,
                     time_created,
                     time_deleted,
                     runtime:
@@ -7520,6 +7521,7 @@ fn prettyprint_vmm(
     const INSTANCE_ID: &'static str = "instance ID";
     const SLED_ID: &'static str = "sled ID";
     const SLED_SERIAL: &'static str = "sled serial";
+    const CPU_PLATFORM: &'static str = "CPU platform";
     const ADDRESS: &'static str = "propolis address";
     const STATE: &'static str = "state";
     const WIDTH: usize = const_max_len(&[
@@ -7530,6 +7532,7 @@ fn prettyprint_vmm(
         INSTANCE_ID,
         SLED_ID,
         SLED_SERIAL,
+        CPU_PLATFORM,
         STATE,
         ADDRESS,
     ]);
@@ -7543,6 +7546,7 @@ fn prettyprint_vmm(
         sled_id,
         propolis_ip,
         propolis_port,
+        cpu_platform,
         runtime: db::model::VmmRuntimeState { state, r#gen, time_state_updated },
     } = vmm;
 
@@ -7569,6 +7573,7 @@ fn prettyprint_vmm(
     if let Some(serial) = sled_serial {
         println!("{indent}{SLED_SERIAL:>width$}: {serial}");
     }
+    println!("{indent}{CPU_PLATFORM:>width$}: {cpu_platform}");
 }
 
 async fn cmd_db_vmm_list(
@@ -7644,6 +7649,7 @@ async fn cmd_db_vmm_list(
                 sled_id,
                 propolis_ip: _,
                 propolis_port: _,
+                cpu_platform: _,
                 runtime:
                     db::model::VmmRuntimeState {
                         state,

dev-tools/omdb/src/bin/omdb/db.rs

@@ -4,7 +4,7 @@
 
 //! developer REPL for driving blueprint planning
 
-use anyhow::{Context, anyhow, bail};
+use anyhow::{Context, anyhow, bail, ensure};
 use camino::{Utf8Path, Utf8PathBuf};
 use chrono::{DateTime, Utc};
 use clap::{ArgAction, ValueEnum};
@@ -63,6 +63,7 @@ use omicron_uuid_kinds::VnicUuid;
 use omicron_uuid_kinds::{BlueprintUuid, MupdateOverrideUuid};
 use omicron_uuid_kinds::{CollectionUuid, MupdateUuid};
 use std::borrow::Cow;
+use std::collections::BTreeSet;
 use std::convert::Infallible;
 use std::fmt::{self, Write};
 use std::io::IsTerminal;
@@ -158,6 +159,9 @@ impl ReconfiguratorSim {
         builder.set_internal_dns_version(parent_blueprint.internal_dns_version);
         builder.set_external_dns_version(parent_blueprint.external_dns_version);
 
+        let mut active_nexus_zones = BTreeSet::new();
+        let mut not_yet_nexus_zones = BTreeSet::new();
+
         for (_, zone) in parent_blueprint
             .all_omicron_zones(BlueprintZoneDisposition::is_in_service)
         {
@@ -179,7 +183,26 @@ impl ReconfiguratorSim {
                     .add_omicron_zone_nic(zone.id, nic)
                     .context("adding omicron zone NIC")?;
             }
+
+            match &zone.zone_type {
+                nexus_types::deployment::BlueprintZoneType::Nexus(nexus) => {
+                    if nexus.nexus_generation
+                        == parent_blueprint.nexus_generation
+                    {
+                        active_nexus_zones.insert(zone.id);
+                    } else if nexus.nexus_generation
+                        > parent_blueprint.nexus_generation
+                    {
+                        not_yet_nexus_zones.insert(zone.id);
+                    }
+                }
+                _ => (),
+            }
         }
+
+        builder.set_active_nexus_zones(active_nexus_zones);
+        builder.set_not_yet_nexus_zones(not_yet_nexus_zones);
+
         Ok(builder.build())
     }
 }
@@ -755,7 +778,7 @@ enum BlueprintEditCommands {
         generation: Generation,
     },
     /// expunge a zone
-    ExpungeZone { zone_id: OmicronZoneUuid },
+    ExpungeZones { zone_ids: Vec<OmicronZoneUuid> },
     /// mark an expunged zone ready for cleanup
     MarkForCleanup { zone_id: OmicronZoneUuid },
     /// configure an SP update
@@ -784,6 +807,11 @@ enum BlueprintEditCommands {
         #[command(subcommand)]
         command: BlueprintEditDebugCommands,
     },
+    /// bumps the blueprint's Nexus generation
+    ///
+    /// This initiates a handoff from the current generation of Nexus zones to
+    /// the next generation of Nexus zones.
+    BumpNexusGeneration,
 }
 
 #[derive(Debug, Subcommand)]
@@ -2157,6 +2185,29 @@ fn cmd_blueprint_edit(
                 .context("failed to add CockroachDB zone")?;
             format!("added CockroachDB zone to sled {}", sled_id)
         }
+        BlueprintEditCommands::BumpNexusGeneration => {
+            let current_generation = builder.nexus_generation();
+            let current_max = blueprint
+                .all_nexus_zones(BlueprintZoneDisposition::is_in_service)
+                .fold(
+                    current_generation,
+                    |current_max, (_sled_id, _zone_config, nexus_config)| {
+                        std::cmp::max(
+                            nexus_config.nexus_generation,
+                            current_max,
+                        )
+                    },
+                );
+            ensure!(
+                current_max > current_generation,
+                "cannot bump blueprint generation (currently \
+                 {current_generation}) past highest deployed Nexus \
+                 generation (currently {current_max})",
+            );
+            let next = current_generation.next();
+            builder.set_nexus_generation(next);
+            format!("nexus generation: {current_generation} -> {next}")
+        }
         BlueprintEditCommands::SetRemoveMupdateOverride { sled_id, value } => {
             let sled_id = sled_id.to_sled_id(system.description())?;
             builder
@@ -2212,12 +2263,16 @@ fn cmd_blueprint_edit(
                 .context("failed to set host phase 2 source")?;
             rv
         }
-        BlueprintEditCommands::ExpungeZone { zone_id } => {
-            let sled_id = sled_with_zone(&builder, &zone_id)?;
-            builder
-                .sled_expunge_zone(sled_id, zone_id)
-                .context("failed to expunge zone")?;
-            format!("expunged zone {zone_id} from sled {sled_id}")
+        BlueprintEditCommands::ExpungeZones { zone_ids } => {
+            let mut rv = String::new();
+            for zone_id in zone_ids {
+                let sled_id = sled_with_zone(&builder, &zone_id)?;
+                builder.sled_expunge_zone(sled_id, zone_id).with_context(
+                    || format!("failed to expunge zone {zone_id}"),
+                )?;
+                swriteln!(rv, "expunged zone {zone_id} from sled {sled_id}");
+            }
+            rv
         }
         BlueprintEditCommands::MarkForCleanup { zone_id } => {
             let sled_id = sled_with_zone(&builder, &zone_id)?;

dev-tools/reconfigurator-cli/src/lib.rs

@@ -3,7 +3,7 @@
 load-example --seed test_expunge_newly_added_external_dns
 
 blueprint-show 3f00b694-1b16-4aaa-8f78-e6b3a527b434
-blueprint-edit 3f00b694-1b16-4aaa-8f78-e6b3a527b434 expunge-zone 8429c772-07e8-40a6-acde-2ed47d16cf84
+blueprint-edit 3f00b694-1b16-4aaa-8f78-e6b3a527b434 expunge-zones 8429c772-07e8-40a6-acde-2ed47d16cf84
 
 # Diff DNS to see that the expunged zone is no longer has DNS records.
 blueprint-diff 3f00b694-1b16-4aaa-8f78-e6b3a527b434 366b0b68-d80e-4bc1-abd3-dc69837847e0
@@ -15,5 +15,5 @@ blueprint-diff 366b0b68-d80e-4bc1-abd3-dc69837847e0 9c998c1d-1a7b-440a-ae0c-40f7
 
 blueprint-show 9c998c1d-1a7b-440a-ae0c-40f781dea6e2
 # expunging the new zone should work, then diff again to see the new zone also have its DNS records removed.
-blueprint-edit 9c998c1d-1a7b-440a-ae0c-40f781dea6e2 expunge-zone 8c0a1969-15b6-4165-ba6d-a27c24151037
+blueprint-edit 9c998c1d-1a7b-440a-ae0c-40f781dea6e2 expunge-zones 8c0a1969-15b6-4165-ba6d-a27c24151037
 blueprint-diff 9c998c1d-1a7b-440a-ae0c-40f781dea6e2 2ac8c740-444d-42ff-8d66-9812a7e51288

dev-tools/reconfigurator-cli/tests/input/cmds-expunge-newly-added-external-dns.txt

@@ -2,7 +2,7 @@ load-example
 
 blueprint-show dbcbd3d6-41ff-48ae-ac0b-1becc9b2fd21
 # Expunge an internal DNS zone
-blueprint-edit dbcbd3d6-41ff-48ae-ac0b-1becc9b2fd21 expunge-zone 99e2f30b-3174-40bf-a78a-90da8abba8ca
+blueprint-edit dbcbd3d6-41ff-48ae-ac0b-1becc9b2fd21 expunge-zones 99e2f30b-3174-40bf-a78a-90da8abba8ca
 # Diff against the new blueprint; the zone has been expunged so its records should be removed.
 blueprint-diff dbcbd3d6-41ff-48ae-ac0b-1becc9b2fd21 8da82a8e-bf97-4fbd-8ddd-9f6462732cf1
 

dev-tools/reconfigurator-cli/tests/input/cmds-expunge-newly-added-internal-dns.txt

@@ -0,0 +1,12 @@
+# Load example system
+load-example --nsleds 3 --ndisks-per-sled 3
+blueprint-list
+blueprint-show dbcbd3d6-41ff-48ae-ac0b-1becc9b2fd21
+
+# Expunge one zone.
+blueprint-edit dbcbd3d6-41ff-48ae-ac0b-1becc9b2fd21 expunge-zones 694bd14f-cb24-4be4-bb19-876e79cda2c8
+blueprint-diff 8da82a8e-bf97-4fbd-8ddd-9f6462732cf1
+
+# Expunge multiple zones.
+blueprint-edit 8da82a8e-bf97-4fbd-8ddd-9f6462732cf1 expunge-zones 7c252b64-c5af-4ec1-989e-9a03f3b0f111 dfac80b4-a887-430a-ae87-a4e065dba787
+blueprint-diff 58d5e830-0884-47d8-a7cd-b2b3751adeb4

dev-tools/reconfigurator-cli/tests/input/cmds-expunge-zones.txt

@@ -0,0 +1,18 @@
+# Load example system
+load-example --nsleds 3 --ndisks-per-sled 3
+
+# Print the initial configuration
+show
+sled-list
+
+# Try to bump the Nexus generation.  This should not work because you can't
+# bump it to a generation that has no Nexus instances deployed.
+blueprint-edit dbcbd3d6-41ff-48ae-ac0b-1becc9b2fd21 bump-nexus-generation
+
+# Now, deploy a Nexus instance at the next generation.
+blueprint-edit dbcbd3d6-41ff-48ae-ac0b-1becc9b2fd21 add-nexus serial0 2
+blueprint-diff 8da82a8e-bf97-4fbd-8ddd-9f6462732cf1
+
+# Now we can bump the Nexus generation to trigger a handoff.
+blueprint-edit 8da82a8e-bf97-4fbd-8ddd-9f6462732cf1 bump-nexus-generation
+blueprint-diff 58d5e830-0884-47d8-a7cd-b2b3751adeb4

dev-tools/reconfigurator-cli/tests/input/cmds-nexus-generation.txt

@@ -339,10 +339,11 @@ empty planning report for blueprint 3f00b694-1b16-4aaa-8f78-e6b3a527b434.
 
 
 
-> blueprint-edit 3f00b694-1b16-4aaa-8f78-e6b3a527b434 expunge-zone 8429c772-07e8-40a6-acde-2ed47d16cf84
+> blueprint-edit 3f00b694-1b16-4aaa-8f78-e6b3a527b434 expunge-zones 8429c772-07e8-40a6-acde-2ed47d16cf84
 blueprint 366b0b68-d80e-4bc1-abd3-dc69837847e0 created from blueprint 3f00b694-1b16-4aaa-8f78-e6b3a527b434: expunged zone 8429c772-07e8-40a6-acde-2ed47d16cf84 from sled 711ac7f8-d19e-4572-bdb9-e9b50f6e362a
 
 
+
 > # Diff DNS to see that the expunged zone is no longer has DNS records.
 > blueprint-diff 3f00b694-1b16-4aaa-8f78-e6b3a527b434 366b0b68-d80e-4bc1-abd3-dc69837847e0
 from: blueprint 3f00b694-1b16-4aaa-8f78-e6b3a527b434
@@ -1346,9 +1347,10 @@ planning report for blueprint 9c998c1d-1a7b-440a-ae0c-40f781dea6e2:
 
 
 > # expunging the new zone should work, then diff again to see the new zone also have its DNS records removed.
-> blueprint-edit 9c998c1d-1a7b-440a-ae0c-40f781dea6e2 expunge-zone 8c0a1969-15b6-4165-ba6d-a27c24151037
+> blueprint-edit 9c998c1d-1a7b-440a-ae0c-40f781dea6e2 expunge-zones 8c0a1969-15b6-4165-ba6d-a27c24151037
 blueprint 2ac8c740-444d-42ff-8d66-9812a7e51288 created from blueprint 9c998c1d-1a7b-440a-ae0c-40f781dea6e2: expunged zone 8c0a1969-15b6-4165-ba6d-a27c24151037 from sled 9dc50690-f9bf-4520-bf80-051d0f465c2c
 
+
 > blueprint-diff 9c998c1d-1a7b-440a-ae0c-40f781dea6e2 2ac8c740-444d-42ff-8d66-9812a7e51288
 from: blueprint 9c998c1d-1a7b-440a-ae0c-40f781dea6e2
 to:   blueprint 2ac8c740-444d-42ff-8d66-9812a7e51288