add sled-agent endpoints to control destroying orphaned datasets

jgallagher · jgallagher · commit f39206da4f1b · 2025-06-11T20:50:35.000-04:00
diff --git a/openapi/sled-agent.json b/openapi/sled-agent.json
@@ -334,6 +334,57 @@
         }
       }
     },
+    "/chicken-switch/destroy-orphaned-datasets": {
+      "get": {
+        "summary": "This endpoint reports the status of the `destroy_orphaned_datasets`",
+        "description": "chicken switch. It will be removed with omicron#6177.",
+        "operationId": "chicken_switch_destroy_orphaned_datasets_get",
+        "responses": {
+          "200": {
+            "description": "successful operation",
+            "content": {
+              "application/json": {
+                "schema": {
+                  "$ref": "#/components/schemas/ChickenSwitchDestroyOrphanedDatasets"
+                }
+              }
+            }
+          },
+          "4XX": {
+            "$ref": "#/components/responses/Error"
+          },
+          "5XX": {
+            "$ref": "#/components/responses/Error"
+          }
+        }
+      },
+      "put": {
+        "summary": "This endpoint sets the `destroy_orphaned_datasets` chicken switch",
+        "description": "(allowing sled-agent to delete datasets it believes are orphaned). It will be removed with omicron#6177.",
+        "operationId": "chicken_switch_destroy_orphaned_datasets_put",
+        "requestBody": {
+          "content": {
+            "application/json": {
+              "schema": {
+                "$ref": "#/components/schemas/ChickenSwitchDestroyOrphanedDatasets"
+              }
+            }
+          },
+          "required": true
+        },
+        "responses": {
+          "204": {
+            "description": "resource updated"
+          },
+          "4XX": {
+            "$ref": "#/components/responses/Error"
+          },
+          "5XX": {
+            "$ref": "#/components/responses/Error"
+          }
+        }
+      }
+    },
     "/disks/{disk_id}": {
       "put": {
         "operationId": "disk_put",
@@ -2926,6 +2977,18 @@
         "format": "uint64",
         "minimum": 0
       },
+      "ChickenSwitchDestroyOrphanedDatasets": {
+        "type": "object",
+        "properties": {
+          "destroy_orphans": {
+            "description": "If true, sled-agent will attempt to destroy durable ZFS datasets that it believes were associated with now-expunged Omicron zones.",
+            "type": "boolean"
+          }
+        },
+        "required": [
+          "destroy_orphans"
+        ]
+      },
       "Chipset": {
         "description": "A kind of virtual chipset.",
         "oneOf": [
diff --git a/sled-agent/api/src/lib.rs b/sled-agent/api/src/lib.rs
@@ -682,6 +682,35 @@ pub trait SledAgentApi {
         path_params: Path<SledDiagnosticsLogsDownloadPathParm>,
         query_params: Query<SledDiagnosticsLogsDownloadQueryParam>,
     ) -> Result<http::Response<Body>, HttpError>;
+
+    /// This endpoint reports the status of the `destroy_orphaned_datasets`
+    /// chicken switch. It will be removed with omicron#6177.
+    #[endpoint {
+        method = GET,
+        path = "/chicken-switch/destroy-orphaned-datasets",
+    }]
+    async fn chicken_switch_destroy_orphaned_datasets_get(
+        request_context: RequestContext<Self::Context>,
+    ) -> Result<HttpResponseOk<ChickenSwitchDestroyOrphanedDatasets>, HttpError>;
+
+    /// This endpoint sets the `destroy_orphaned_datasets` chicken switch
+    /// (allowing sled-agent to delete datasets it believes are orphaned). It
+    /// will be removed with omicron#6177.
+    #[endpoint {
+        method = PUT,
+        path = "/chicken-switch/destroy-orphaned-datasets",
+    }]
+    async fn chicken_switch_destroy_orphaned_datasets_put(
+        request_context: RequestContext<Self::Context>,
+        body: TypedBody<ChickenSwitchDestroyOrphanedDatasets>,
+    ) -> Result<HttpResponseUpdatedNoContent, HttpError>;
+}
+
+#[derive(Clone, Debug, Deserialize, JsonSchema, Serialize)]
+pub struct ChickenSwitchDestroyOrphanedDatasets {
+    /// If true, sled-agent will attempt to destroy durable ZFS datasets that it
+    /// believes were associated with now-expunged Omicron zones.
+    pub destroy_orphans: bool,
 }
 
 #[derive(Clone, Debug, Deserialize, JsonSchema, Serialize)]
diff --git a/sled-agent/config-reconciler/src/handle.rs b/sled-agent/config-reconciler/src/handle.rs
@@ -250,6 +250,11 @@ impl ConfigReconcilerHandle {
         );
     }
 
+    /// Read whether or not we try to destroy orphaned datasets.
+    pub fn will_destroy_orphans(&self) -> bool {
+        self.destroy_orphans.load(Ordering::Relaxed)
+    }
+
     /// Control whether or not we try to destroy orphaned datasets.
     pub fn set_destroy_orphans(&self, destroy_orphans: bool) {
         self.destroy_orphans.store(destroy_orphans, Ordering::Relaxed);
diff --git a/sled-agent/src/http_entrypoints.rs b/sled-agent/src/http_entrypoints.rs
@@ -1053,4 +1053,26 @@ impl SledAgentApi for SledAgentImpl {
             .await
             .map_err(HttpError::from)
     }
+
+    async fn chicken_switch_destroy_orphaned_datasets_get(
+        request_context: RequestContext<Self::Context>,
+    ) -> Result<HttpResponseOk<ChickenSwitchDestroyOrphanedDatasets>, HttpError>
+    {
+        let sa = request_context.context();
+        let destroy_orphans = sa.chicken_switch_destroy_orphaned_datasets();
+        Ok(HttpResponseOk(ChickenSwitchDestroyOrphanedDatasets {
+            destroy_orphans,
+        }))
+    }
+
+    async fn chicken_switch_destroy_orphaned_datasets_put(
+        request_context: RequestContext<Self::Context>,
+        body: TypedBody<ChickenSwitchDestroyOrphanedDatasets>,
+    ) -> Result<HttpResponseUpdatedNoContent, HttpError> {
+        let ChickenSwitchDestroyOrphanedDatasets { destroy_orphans } =
+            body.into_inner();
+        let sa = request_context.context();
+        sa.set_chicken_switch_destroy_orphaned_datasets(destroy_orphans);
+        Ok(HttpResponseUpdatedNoContent())
+    }
 }
diff --git a/sled-agent/src/sim/http_entrypoints.rs b/sled-agent/src/sim/http_entrypoints.rs
@@ -768,6 +768,20 @@ impl SledAgentApi for SledAgentSimImpl {
     ) -> Result<http::Response<dropshot::Body>, HttpError> {
         method_unimplemented()
     }
+
+    async fn chicken_switch_destroy_orphaned_datasets_get(
+        _request_context: RequestContext<Self::Context>,
+    ) -> Result<HttpResponseOk<ChickenSwitchDestroyOrphanedDatasets>, HttpError>
+    {
+        method_unimplemented()
+    }
+
+    async fn chicken_switch_destroy_orphaned_datasets_put(
+        _request_context: RequestContext<Self::Context>,
+        _body: TypedBody<ChickenSwitchDestroyOrphanedDatasets>,
+    ) -> Result<HttpResponseUpdatedNoContent, HttpError> {
+        method_unimplemented()
+    }
 }
 
 fn method_unimplemented<T>() -> Result<T, HttpError> {
diff --git a/sled-agent/src/sled_agent.rs b/sled-agent/src/sled_agent.rs
@@ -828,6 +828,19 @@ impl SledAgent {
         self.inner.config_reconciler.set_sled_config(config).await
     }
 
+    /// Get the status of the "destroy orphaned datasets" chicken switch.
+    pub(crate) fn chicken_switch_destroy_orphaned_datasets(&self) -> bool {
+        self.inner.config_reconciler.will_destroy_orphans()
+    }
+
+    /// Set the status of the "destroy orphaned datasets" chicken switch.
+    pub(crate) fn set_chicken_switch_destroy_orphaned_datasets(
+        &self,
+        destroy_orphans: bool,
+    ) {
+        self.inner.config_reconciler.set_destroy_orphans(destroy_orphans);
+    }
+
     /// Returns whether or not the sled believes itself to be a scrimlet
     pub fn get_role(&self) -> SledRole {
         if self.inner.hardware.is_scrimlet() {
