wip: actually check for failed state before telling sled-agent we fai…

…led it.
oxidecomputer · Feb 2, 2024 · b59dace · b59dace
1 parent ac2e7a0
commit b59dace
Showing 1 changed file with 15 additions and 4 deletions.
diff --git a/nexus/src/app/instance.rs b/nexus/src/app/instance.rs
@@ -1084,14 +1084,25 @@ impl super::Nexus {
                     .await?;
                 if !inst_updated && !vmm_updated {
                     // the generation number bumped up by the timeout task.
-                    // TODO check actual state value (need more specificity than these two bools?)
                     let state = self
                         .db_datastore
                         .instance_fetch_with_vmm(opctx, &authz_instance)
                         .await?;
-                    Ok(HandleInstancePutResultResult::TimedOut {
-                        generation: *state.instance().runtime().gen,
-                    })
+                    match state.instance().runtime().nexus_state.0 {
+                        InstanceState::Failed => {
+                            let generation = *state.instance().runtime().gen;
+                            warn!(
+                                self.log,
+                                "Informing sled-agent of timed-out instance";
+                                "instance_id" => %instance_id,
+                                "generation" => %generation
+                            );
+                            Ok(HandleInstancePutResultResult::TimedOut {
+                                generation,
+                            })
+                        }
+                        _ => Ok(HandleInstancePutResultResult::Ok),
+                    }
                 } else {
                     Ok(HandleInstancePutResultResult::Ok)
                 }