From 27564fc055080ac8783ae67b924c62c8888c5b6a Mon Sep 17 00:00:00 2001
From: Eliza Weisman <eliza@elizas.website>
Date: Fri, 27 Sep 2024 12:18:46 -0700
Subject: [PATCH] include active VMM in karmic status check

---
 dev-tools/omdb/src/bin/omdb/db.rs          |  5 ++-
 nexus/db-model/src/instance.rs             | 36 ++++++++++++++++++----
 nexus/src/app/sagas/instance_update/mod.rs |  5 ++-
 3 files changed, 38 insertions(+), 8 deletions(-)

diff --git a/dev-tools/omdb/src/bin/omdb/db.rs b/dev-tools/omdb/src/bin/omdb/db.rs
index 12eec3f783..4a67e5771e 100644
--- a/dev-tools/omdb/src/bin/omdb/db.rs
+++ b/dev-tools/omdb/src/bin/omdb/db.rs
@@ -3027,7 +3027,10 @@ async fn cmd_db_instance_info(
         r#gen.0
     );
     println!("    {LAST_AUTO_RESTART:>WIDTH$}: {time_last_auto_restarted:?}");
-    match instance.auto_restart.status(&instance.runtime_state) {
+    match instance
+        .auto_restart
+        .status(&instance.runtime_state, active_vmm.as_ref())
+    {
         InstanceKarmicStatus::NotFailed => {}
         InstanceKarmicStatus::Ready => {
             println!("(i) {KARMIC_STATUS:>WIDTH$}: ready to reincarnate!");
diff --git a/nexus/db-model/src/instance.rs b/nexus/db-model/src/instance.rs
index 4dbf80fea1..a753e81ae8 100644
--- a/nexus/db-model/src/instance.rs
+++ b/nexus/db-model/src/instance.rs
@@ -4,7 +4,7 @@
 
 use super::{
     ByteCount, Disk, ExternalIp, Generation, InstanceAutoRestartPolicy,
-    InstanceCpuCount, InstanceState,
+    InstanceCpuCount, InstanceState, Vmm,
 };
 use crate::collection::DatastoreAttachTargetConfig;
 use crate::schema::{disk, external_ip, instance};
@@ -294,12 +294,36 @@ impl InstanceAutoRestart {
 
     /// Returns `true` if `self` permits an instance to reincarnate given the
     /// provided `state`.
-    pub fn status(&self, state: &InstanceRuntimeState) -> InstanceKarmicStatus {
+    pub fn status(
+        &self,
+        state: &InstanceRuntimeState,
+        active_vmm: Option<&Vmm>,
+    ) -> InstanceKarmicStatus {
         // Instances only need to be automatically restarted if they are in the
-        // `Failed` state.
-        if state.nexus_state != InstanceState::Failed {
-            return InstanceKarmicStatus::NotFailed;
-        }
+        // `Failed` state, or if their active VMM is in the `SagaUnwound` state.
+        match (state.nexus_state, active_vmm) {
+            (InstanceState::Failed, _vmm) => {
+                debug_assert!(
+                    _vmm.is_none(),
+                    "a Failed instance will never have an active VMM!"
+                );
+            }
+            (InstanceState::Vmm, Some(ref vmm)) => {
+                debug_assert_eq!(
+                    state.propolis_id,
+                    vmm.id(),
+                    "don't call `InstanceAutoRestart::status with a VMM \
+                     that isn't this instance's active VMM!?!?"
+                );
+                // Note that we *don't* reincarnate instances with `Failed`` active
+                // VMMs; in that case, an instance-update saga must first run to
+                // move the *instance* record to the `Failed` state.
+                if vmm.runtime.state != VmmState::SagaUnwound {
+                    return InstanceKarmicStatus::NotFailed;
+                }
+            }
+            _ => return InstanceKarmicStatus::NotFailed,
+        };
 
         // Check if the instance's configured auto-restart policy permits the
         // control plane to automatically restart it.
diff --git a/nexus/src/app/sagas/instance_update/mod.rs b/nexus/src/app/sagas/instance_update/mod.rs
index 9d2fc58512..e55bf33113 100644
--- a/nexus/src/app/sagas/instance_update/mod.rs
+++ b/nexus/src/app/sagas/instance_update/mod.rs
@@ -1289,7 +1289,10 @@ async fn siu_chain_successor_saga(
             // it does, activate the instance-reincarnation background task to
             // automatically restart it.
             let auto_restart = new_state.instance.auto_restart;
-            match auto_restart.status(&new_state.instance.runtime_state) {
+            match auto_restart.status(
+                &new_state.instance.runtime_state,
+                new_state.active_vmm.as_ref(),
+            ) {
                 InstanceKarmicStatus::Ready => {
                     info!(
                         log,