compute_ctl: return LSN in /terminate (#12240)

- Add optional `?mode=fast|immediate` to `/terminate`, `fast` is
default. Immediate avoids waiting 30
  seconds before returning from `terminate`.
- Add `TerminateMode` to `ComputeStatus::TerminationPending`
- Use `/terminate?mode=immediate` in `neon_local` instead of `pg_ctl
stop` for `test_replica_promotes`.
- Change `test_replica_promotes` to check returned LSN
- Annotate `finish_sync_safekeepers` as `noreturn`.

https://github.com/neondatabase/cloud/issues/29807

compute_tools/src/compute.rs

@@ -163,6 +163,10 @@ pub struct ComputeState {
     pub lfc_prewarm_state: LfcPrewarmState,
     pub lfc_offload_state: LfcOffloadState,
 
+    /// WAL flush LSN that is set after terminating Postgres and syncing safekeepers if
+    /// mode == ComputeMode::Primary. None otherwise
+    pub terminate_flush_lsn: Option<Lsn>,
+
     pub metrics: ComputeMetrics,
 }
 
@@ -178,6 +182,7 @@ impl ComputeState {
             metrics: ComputeMetrics::default(),
             lfc_prewarm_state: LfcPrewarmState::default(),
             lfc_offload_state: LfcOffloadState::default(),
+            terminate_flush_lsn: None,
         }
     }
 
@@ -531,12 +536,21 @@ impl ComputeNode {
         // Reap the postgres process
         delay_exit |= this.cleanup_after_postgres_exit()?;
 
+        // /terminate returns LSN. If we don't sleep at all, connection will break and we
+        // won't get result. If we sleep too much, tests will take significantly longer
+        // and Github Action run will error out
+        let sleep_duration = if delay_exit {
+            Duration::from_secs(30)
+        } else {
+            Duration::from_millis(300)
+        };
+
         // If launch failed, keep serving HTTP requests for a while, so the cloud
         // control plane can get the actual error.
         if delay_exit {
             info!("giving control plane 30s to collect the error before shutdown");
-            std::thread::sleep(Duration::from_secs(30));
         }
+        std::thread::sleep(sleep_duration);
         Ok(exit_code)
     }
 
@@ -908,20 +922,25 @@ impl ComputeNode {
         // Maybe sync safekeepers again, to speed up next startup
         let compute_state = self.state.lock().unwrap().clone();
         let pspec = compute_state.pspec.as_ref().expect("spec must be set");
-        if matches!(pspec.spec.mode, compute_api::spec::ComputeMode::Primary) {
+        let lsn = if matches!(pspec.spec.mode, compute_api::spec::ComputeMode::Primary) {
             info!("syncing safekeepers on shutdown");
             let storage_auth_token = pspec.storage_auth_token.clone();
             let lsn = self.sync_safekeepers(storage_auth_token)?;
-            info!("synced safekeepers at lsn {lsn}");
-        }
+            info!(%lsn, "synced safekeepers");
+            Some(lsn)
+        } else {
+            info!("not primary, not syncing safekeepers");
+            None
+        };
 
         let mut delay_exit = false;
         let mut state = self.state.lock().unwrap();
-        if state.status == ComputeStatus::TerminationPending {
+        state.terminate_flush_lsn = lsn;
+        if let ComputeStatus::TerminationPending { mode } = state.status {
             state.status = ComputeStatus::Terminated;
             self.state_changed.notify_all();
             // we were asked to terminate gracefully, don't exit to avoid restart
-            delay_exit = true
+            delay_exit = mode == compute_api::responses::TerminateMode::Fast
         }
         drop(state);
 
@@ -1792,7 +1811,7 @@ impl ComputeNode {
 
                             // exit loop
                             ComputeStatus::Failed
-                            | ComputeStatus::TerminationPending
+                            | ComputeStatus::TerminationPending { .. }
                             | ComputeStatus::Terminated => break 'cert_update,
 
                             // wait

compute_tools/src/http/routes/terminate.rs

@@ -1,29 +1,39 @@
-use std::sync::Arc;
-
+use crate::compute::{ComputeNode, forward_termination_signal};
+use crate::http::JsonResponse;
 use axum::extract::State;
-use axum::response::{IntoResponse, Response};
-use compute_api::responses::ComputeStatus;
+use axum::response::Response;
+use axum_extra::extract::OptionalQuery;
+use compute_api::responses::{ComputeStatus, TerminateResponse};
 use http::StatusCode;
+use serde::Deserialize;
+use std::sync::Arc;
 use tokio::task;
 use tracing::info;
 
-use crate::compute::{ComputeNode, forward_termination_signal};
-use crate::http::JsonResponse;
+#[derive(Deserialize, Default)]
+pub struct TerminateQuery {
+    mode: compute_api::responses::TerminateMode,
+}
 
 /// Terminate the compute.
-pub(in crate::http) async fn terminate(State(compute): State<Arc<ComputeNode>>) -> Response {
+pub(in crate::http) async fn terminate(
+    State(compute): State<Arc<ComputeNode>>,
+    OptionalQuery(terminate): OptionalQuery<TerminateQuery>,
+) -> Response {
+    let mode = terminate.unwrap_or_default().mode;
     {
         let mut state = compute.state.lock().unwrap();
         if state.status == ComputeStatus::Terminated {
-            return StatusCode::CREATED.into_response();
+            return JsonResponse::success(StatusCode::CREATED, state.terminate_flush_lsn);
         }
 
         if !matches!(state.status, ComputeStatus::Empty | ComputeStatus::Running) {
             return JsonResponse::invalid_status(state.status);
         }
-
-        state.set_status(ComputeStatus::TerminationPending, &compute.state_changed);
-        drop(state);
+        state.set_status(
+            ComputeStatus::TerminationPending { mode },
+            &compute.state_changed,
+        );
     }
 
     forward_termination_signal(false);
@@ -34,7 +44,7 @@ pub(in crate::http) async fn terminate(State(compute): State<Arc<ComputeNode>>)
     // be able to serve other requests while some particular request
     // is waiting for compute to finish configuration.
     let c = compute.clone();
-    task::spawn_blocking(move || {
+    let lsn = task::spawn_blocking(move || {
         let mut state = c.state.lock().unwrap();
         while state.status != ComputeStatus::Terminated {
             state = c.state_changed.wait(state).unwrap();
@@ -44,11 +54,10 @@ pub(in crate::http) async fn terminate(State(compute): State<Arc<ComputeNode>>)
                 state.status
             );
         }
+        state.terminate_flush_lsn
     })
     .await
     .unwrap();
-
     info!("terminated Postgres");
-
-    StatusCode::OK.into_response()
+    JsonResponse::success(StatusCode::OK, TerminateResponse { lsn })
 }

compute_tools/src/monitor.rs

@@ -83,7 +83,9 @@ impl ComputeMonitor {
         let compute_status = self.compute.get_status();
         if matches!(
             compute_status,
-            ComputeStatus::Terminated | ComputeStatus::TerminationPending | ComputeStatus::Failed
+            ComputeStatus::Terminated
+                | ComputeStatus::TerminationPending { .. }
+                | ComputeStatus::Failed
         ) {
             info!(
                 "compute is in {} status, stopping compute monitor",