[compute_ctl] Do not initialize last_active on start (#4137)

Our scale-to-zero logic was optimized for short auto-suspend intervals,
e.g. minutes or hours. In this case, if compute was restarted by k8s due
to some reason (OOM, k8s node went down, pod relocation, etc.),
`last_active` got bumped, we start counting auto-suspend timeout again.
It's not a big deal, i.e. we suspend completely idle compute not after 5
minutes, but after 10 minutes or so.

Yet, some clients may want days or even weeks. And chance that compute
could be restarted during this interval is pretty high, but in this case
we could be not able to suspend some computes for weeks.

After this commit, we won't initialize `last_active` on start, so
`/status` could return an unset attribute. This means that there was no
user activity since start. Control-plane should deal with it by taking
`max()` out of all available activity timestamps: `started_at`,
`last_active`, etc.

compute_ctl part of neondatabase/cloud#4853

compute_tools/src/compute.rs

@@ -67,8 +67,9 @@ pub struct ComputeNode {
 pub struct ComputeState {
     pub start_time: DateTime<Utc>,
     pub status: ComputeStatus,
-    /// Timestamp of the last Postgres activity
-    pub last_active: DateTime<Utc>,
+    /// Timestamp of the last Postgres activity. It could be `None` if
+    /// compute wasn't used since start.
+    pub last_active: Option<DateTime<Utc>>,
     pub error: Option<String>,
     pub pspec: Option<ParsedSpec>,
     pub metrics: ComputeMetrics,
@@ -79,7 +80,7 @@ impl ComputeState {
         Self {
             start_time: Utc::now(),
             status: ComputeStatus::Empty,
-            last_active: Utc::now(),
+            last_active: None,
             error: None,
             pspec: None,
             metrics: ComputeMetrics::default(),

compute_tools/src/http/openapi_spec.yaml

@@ -181,8 +181,8 @@ components:
     ComputeState:
       type: object
       required:
+        - start_time
         - status
-        - last_active
       properties:
         start_time:
           type: string
@@ -195,11 +195,13 @@ components:
           $ref: '#/components/schemas/ComputeStatus'
         last_active:
           type: string
-          description: The last detected compute activity timestamp in UTC and RFC3339 format.
+          description: |
+            The last detected compute activity timestamp in UTC and RFC3339 format.
+            It could be empty if compute was never used by user since start.
           example: "2022-10-12T07:20:50.52Z"
         error:
           type: string
-          description: Text of the error during compute startup, if any.
+          description: Text of the error during compute startup or reconfiguration, if any.
           example: ""
         tenant:
           type: string
@@ -222,9 +224,12 @@ components:
     ComputeStatus:
       type: string
       enum:
+        - empty
         - init
         - failed
         - running
+        - configuration_pending
+        - configuration
       example: running
 
     #

compute_tools/src/monitor.rs

@@ -74,7 +74,7 @@ fn watch_compute_activity(compute: &ComputeNode) {
                             // Found non-idle backend, so the last activity is NOW.
                             // Save it and exit the for loop. Also clear the idle backend
                             // `state_change` timestamps array as it doesn't matter now.
-                            last_active = Utc::now();
+                            last_active = Some(Utc::now());
                             idle_backs.clear();
                             break;
                         }
@@ -82,15 +82,16 @@ fn watch_compute_activity(compute: &ComputeNode) {
 
                     // Get idle backend `state_change` with the max timestamp.
                     if let Some(last) = idle_backs.iter().max() {
-                        last_active = *last;
+                        last_active = Some(*last);
                     }
                 }
 
                 // Update the last activity in the shared state if we got a more recent one.
                 let mut state = compute.state.lock().unwrap();
+                // NB: `Some(<DateTime>)` is always greater than `None`.
                 if last_active > state.last_active {
                     state.last_active = last_active;
-                    debug!("set the last compute activity time to: {}", last_active);
+                    debug!("set the last compute activity time to: {:?}", last_active);
                 }
             }
             Err(e) => {

libs/compute_api/src/responses.rs

@@ -19,7 +19,7 @@ pub struct ComputeStatusResponse {
     pub timeline: Option<String>,
     pub status: ComputeStatus,
     #[serde(serialize_with = "rfc3339_serialize")]
-    pub last_active: DateTime<Utc>,
+    pub last_active: Option<DateTime<Utc>>,
     pub error: Option<String>,
 }
 
@@ -29,7 +29,7 @@ pub struct ComputeState {
     pub status: ComputeStatus,
     /// Timestamp of the last Postgres activity
     #[serde(serialize_with = "rfc3339_serialize")]
-    pub last_active: DateTime<Utc>,
+    pub last_active: Option<DateTime<Utc>>,
     pub error: Option<String>,
 }
 
@@ -54,11 +54,15 @@ pub enum ComputeStatus {
     Failed,
 }
 
-fn rfc3339_serialize<S>(x: &DateTime<Utc>, s: S) -> Result<S::Ok, S::Error>
+fn rfc3339_serialize<S>(x: &Option<DateTime<Utc>>, s: S) -> Result<S::Ok, S::Error>
 where
     S: Serializer,
 {
-    x.to_rfc3339().serialize(s)
+    if let Some(x) = x {
+        x.to_rfc3339().serialize(s)
+    } else {
+        s.serialize_none()
+    }
 }
 
 /// Response of the /metrics.json API