Implement switching clusters

kiss/api/src/ansible/mod.rs

@@ -12,7 +12,7 @@ use kube::{
     Api, Client, Error,
 };
 
-use crate::r#box::{BoxPowerSpec, BoxSpec, BoxState};
+use crate::r#box::{BoxPowerSpec, BoxSpec, BoxState, BoxStatus};
 
 pub struct AnsibleClient {
     ansible_image: String,
@@ -23,6 +23,7 @@ impl AnsibleClient {
     pub const LABEL_BOX_ACCESS_ADDRESS: &'static str = "kiss.netai-cloud/box_access_address";
     pub const LABEL_BOX_MACHINE_UUID: &'static str = "kiss.netai-cloud/box_machine_uuid";
     pub const LABEL_COMPLETED_STATE: &'static str = "kiss.netai-cloud/completed_state";
+    pub const LABEL_TARGET_CLUSTER: &'static str = "kiss.netai-cloud/target_cluster";
 
     pub const ANSIBLE_IMAGE: &'static str = "quay.io/kubespray/kubespray:v2.19.1";
 
@@ -36,6 +37,13 @@ impl AnsibleClient {
         let ns = "kiss";
         let box_name = job.spec.machine.uuid.to_string();
         let name = format!("box-{}-{}", &job.task, &box_name);
+        let cluster = job
+            .status
+            .as_ref()
+            .and_then(|status| status.bind_cluster.as_ref())
+            .or_else(|| job.spec.cluster.as_ref())
+            .map(String::as_str)
+            .unwrap_or("default");
 
         // delete all previous cronjobs
         {
@@ -78,6 +86,7 @@ impl AnsibleClient {
                         .as_ref()
                         .map(ToString::to_string)
                         .map(|state| (Self::LABEL_COMPLETED_STATE.into(), state)),
+                    Some((Self::LABEL_TARGET_CLUSTER.into(), cluster.to_string())),
                 ]
                 .into_iter()
                 .flatten()
@@ -190,14 +199,7 @@ impl AnsibleClient {
                         Volume {
                             name: "ansible".into(),
                             config_map: Some(ConfigMapVolumeSource {
-                                name: Some(format!(
-                                    "ansible-control-planes-{}",
-                                    job.spec
-                                        .cluster
-                                        .as_ref()
-                                        .map(String::as_str)
-                                        .unwrap_or("default")
-                                )),
+                                name: Some(format!("ansible-control-planes-{cluster}")),
                                 default_mode: Some(0o400),
                                 ..Default::default()
                             }),
@@ -284,5 +286,6 @@ pub struct AnsibleJob<'a> {
     pub cron: Option<&'static str>,
     pub task: &'static str,
     pub spec: &'a BoxSpec,
+    pub status: Option<&'a BoxStatus>,
     pub completed_state: Option<BoxState>,
 }

kiss/api/src/box.rs

@@ -54,6 +54,7 @@ pub struct BoxSpec {
 #[serde(rename_all = "camelCase")]
 pub struct BoxStatus {
     pub state: BoxState,
+    pub bind_cluster: Option<String>,
     pub last_updated: DateTime<Utc>,
 }
 

kiss/controller/src/ctx.rs

@@ -63,6 +63,7 @@ impl ::kiss_api::manager::Ctx for Ctx {
                             cron: new_state.cron(),
                             task,
                             spec: &data.spec,
+                            status: data.status.as_ref(),
                             completed_state: new_state.complete(),
                         },
                     )
@@ -76,6 +77,7 @@ impl ::kiss_api::manager::Ctx for Ctx {
             "kind": crd.kind,
             "status": BoxStatus {
                 state: new_state,
+                bind_cluster: status.as_ref().and_then(|status| status.bind_cluster.as_ref()).cloned(),
                 last_updated: if old_state == new_state {
                     status
                         .as_ref()

kiss/gateway/src/main.rs

@@ -54,10 +54,11 @@ async fn get_new(client: Data<Arc<Client>>, Query(query): Query<BoxNewQuery>) ->
                     "kind": crd.kind,
                     "status": BoxStatus {
                         state: BoxState::New,
+                        bind_cluster: None,
                         last_updated: Utc::now(),
                     },
                 }));
-                let pp = PatchParams::apply("kiss-controller").force();
+                let pp = PatchParams::apply("kiss-gateway").force();
                 api.patch_status(&name, &pp, &patch).await?;
             }
             Err(_) => {
@@ -74,6 +75,7 @@ async fn get_new(client: Data<Arc<Client>>, Query(query): Query<BoxNewQuery>) ->
                     },
                     status: Some(BoxStatus {
                         state: BoxState::New,
+                        bind_cluster: None,
                         last_updated: Utc::now(),
                     }),
                 };
@@ -112,10 +114,11 @@ async fn get_commission(client: Data<Arc<Client>>, Json(spec): Json<BoxSpec>) ->
                     "spec": spec,
                     "status": BoxStatus {
                         state: BoxState::Ready,
+                        bind_cluster: None,
                         last_updated: Utc::now(),
                     },
                 }));
-                let pp = PatchParams::apply("kiss-controller").force();
+                let pp = PatchParams::apply("kiss-gateway").force();
                 api.patch(&name, &pp, &patch).await?;
                 api.patch_status(&name, &pp, &patch).await?;
             }
@@ -128,6 +131,7 @@ async fn get_commission(client: Data<Arc<Client>>, Json(spec): Json<BoxSpec>) ->
                     spec,
                     status: Some(BoxStatus {
                         state: BoxState::Ready,
+                        bind_cluster: None,
                         last_updated: Utc::now(),
                     }),
                 };

kiss/monitor/src/ctx.rs

@@ -37,6 +37,10 @@ impl ::kiss_api::manager::Ctx for Ctx {
             .labels()
             .get(AnsibleClient::LABEL_COMPLETED_STATE)
             .and_then(|state| state.parse().ok());
+        let cluster = data
+            .labels()
+            .get(AnsibleClient::LABEL_TARGET_CLUSTER)
+            .cloned();
 
         let has_completed = status.and_then(|e| e.succeeded).unwrap_or_default() > 0;
         let has_failed = status.and_then(|e| e.failed).unwrap_or_default() > 0;
@@ -45,7 +49,7 @@ impl ::kiss_api::manager::Ctx for Ctx {
         if has_completed {
             // update the state
             if let Some(completed_state) = completed_state {
-                Self::update_box_state(manager, data, completed_state).await
+                Self::update_box_state(manager, data, completed_state, cluster).await
             }
             // keep the state, scheduled by the controller
             else {
@@ -58,7 +62,9 @@ impl ::kiss_api::manager::Ctx for Ctx {
         else if has_failed {
             let fallback_state = completed_state.unwrap_or(BoxState::Failed).fail();
             match fallback_state {
-                BoxState::Failed => Self::update_box_state(manager, data, fallback_state).await,
+                BoxState::Failed => {
+                    Self::update_box_state(manager, data, fallback_state, cluster).await
+                }
                 // do nothing when the job has no fallback state
                 _ => Ok(Action::requeue(
                     <Self as ::kiss_api::manager::Ctx>::FALLBACK,
@@ -79,6 +85,7 @@ impl Ctx {
         manager: Arc<Manager<Self>>,
         data: Arc<<Self as ::kiss_api::manager::Ctx>::Data>,
         state: BoxState,
+        cluster: Option<String>,
     ) -> Result<Action, Error>
     where
         Self: Sized,
@@ -102,6 +109,7 @@ impl Ctx {
                 "kind": crd.kind,
                 "status": BoxStatus {
                     state,
+                    bind_cluster: cluster,
                     last_updated: Utc::now(),
                 },
             }));