feat(prover): Add sending scale requests for Scaler targets

core/lib/config/src/configs/prover_autoscaler.rs

@@ -63,6 +63,9 @@ pub struct ProverAutoscalerScalerConfig {
     pub long_pending_duration: Duration,
     /// List of simple autoscaler targets.
     pub scaler_targets: Vec<ScalerTarget>,
+    /// If dry-run enabled don't send any scale requests.
+    #[serde(default)]
+    pub dry_run: bool,
 }
 
 #[derive(
@@ -122,7 +125,7 @@ pub enum QueueReportFields {
 #[derive(Debug, Clone, PartialEq, Deserialize, Default)]
 pub struct ScalerTarget {
     pub queue_report_field: QueueReportFields,
-    pub pod_name_prefix: String,
+    pub deployment: String,
     /// Max replicas per cluster.
     pub max_replicas: HashMap<String, usize>,
     /// The queue will be divided by the speed and rounded up to get number of replicas.

core/lib/protobuf_config/src/proto/config/prover_autoscaler.proto

@@ -52,9 +52,10 @@ message MaxReplica {
 
 message ScalerTarget {
   optional string queue_report_field = 1; // required
-  optional string pod_name_prefix = 2; // required
+  optional string deployment = 5; // required
   repeated MaxReplica max_replicas = 3; // required at least one
   optional uint64 speed = 4; // optional
+  reserved 2; reserved "pod_name_prefix";
 }
 
 message ProverAutoscalerScalerConfig {
@@ -69,4 +70,5 @@ message ProverAutoscalerScalerConfig {
   repeated MaxProver max_provers = 9; // optional
   repeated MinProver min_provers = 10; // optional
   repeated ScalerTarget scaler_targets = 11; // optional
+  optional bool dry_run = 12; // optional
 }

core/lib/protobuf_config/src/prover_autoscaler.rs

@@ -118,6 +118,7 @@ impl ProtoRepr for proto::ProverAutoscalerScalerConfig {
                 .enumerate()
                 .map(|(i, x)| x.read().context(i).unwrap())
                 .collect::<Vec<_>>(),
+            dry_run: self.dry_run.unwrap_or_default(),
         })
     }
 
@@ -158,6 +159,7 @@ impl ProtoRepr for proto::ProverAutoscalerScalerConfig {
                 .map(|(k, v)| proto::MinProver::build(&(k.clone(), *v)))
                 .collect(),
             scaler_targets: this.scaler_targets.iter().map(ProtoRepr::build).collect(),
+            dry_run: Some(this.dry_run),
         }
     }
 }
@@ -269,9 +271,7 @@ impl ProtoRepr for proto::ScalerTarget {
             queue_report_field: required(&self.queue_report_field)
                 .and_then(|x| Ok((*x).parse()?))
                 .context("queue_report_field")?,
-            pod_name_prefix: required(&self.pod_name_prefix)
-                .context("pod_name_prefix")?
-                .clone(),
+            deployment: required(&self.deployment).context("deployment")?.clone(),
             max_replicas: self
                 .max_replicas
                 .iter()
@@ -289,7 +289,7 @@ impl ProtoRepr for proto::ScalerTarget {
     fn build(this: &Self::Type) -> Self {
         Self {
             queue_report_field: Some(this.queue_report_field.to_string()),
-            pod_name_prefix: Some(this.pod_name_prefix.clone()),
+            deployment: Some(this.deployment.clone()),
             max_replicas: this
                 .max_replicas
                 .iter()

prover/crates/bin/prover_autoscaler/src/agent.rs

@@ -96,7 +96,7 @@ pub struct ScaleRequest {
     pub deployments: Vec<ScaleDeploymentRequest>,
 }
 
-#[derive(Debug, Serialize, Deserialize)]
+#[derive(Debug, Default, Serialize, Deserialize)]
 pub struct ScaleResponse {
     pub scale_result: Vec<String>,
 }

prover/crates/bin/prover_autoscaler/src/global/scaler.rs

@@ -83,7 +83,7 @@ pub struct GpuScaler {
 
 pub struct SimpleScaler {
     queue_report_field: QueueReportFields,
-    pod_name_prefix: String,
+    deployment: String,
     /// Which cluster to use first.
     cluster_priorities: HashMap<String, u32>,
     max_replicas: HashMap<String, usize>,
@@ -365,6 +365,47 @@ impl GpuScaler {
 
         provers
     }
+
+    fn diff(
+        namespace: &str,
+        provers: HashMap<GPUPoolKey, u32>,
+        clusters: &Clusters,
+        requests: &mut HashMap<String, ScaleRequest>,
+    ) {
+        provers
+            .into_iter()
+            .for_each(|(GPUPoolKey { cluster, gpu }, replicas)| {
+                let prover = gpu_to_prover(gpu);
+                clusters
+                    .clusters
+                    .get(&cluster)
+                    .and_then(|c| c.namespaces.get(namespace))
+                    .and_then(|ns| ns.deployments.get(&prover))
+                    .map_or_else(
+                        || {
+                            tracing::error!(
+                                "Wasn't able to find deployment {} in cluster {}, namespace {}",
+                                prover,
+                                cluster,
+                                namespace
+                            )
+                        },
+                        |deployment| {
+                            if deployment.desired != replicas as i32 {
+                                requests
+                                    .entry(cluster.clone())
+                                    .or_default()
+                                    .deployments
+                                    .push(ScaleDeploymentRequest {
+                                        namespace: namespace.into(),
+                                        name: prover.clone(),
+                                        size: replicas as i32,
+                                    });
+                            }
+                        },
+                    );
+            })
+    }
 }
 
 #[derive(Default, Debug, PartialEq, Eq)]
@@ -389,7 +430,7 @@ impl SimpleScaler {
     ) -> Self {
         Self {
             queue_report_field: config.queue_report_field.clone(),
-            pod_name_prefix: config.pod_name_prefix.clone(),
+            deployment: config.deployment.clone(),
             cluster_priorities,
             max_replicas: config.max_replicas.clone(),
             speed: config.speed,
@@ -418,7 +459,7 @@ impl SimpleScaler {
         // Initialize pool only if we have ready deployments.
         pool.pods.insert(PodStatus::Running, 0);
 
-        let pod_re = Regex::new(&format!("^{}-", self.pod_name_prefix)).unwrap();
+        let pod_re = Regex::new(&format!("^{}-", self.deployment)).unwrap();
         for (_, pod) in namespace_value
             .pods
             .iter()
@@ -551,47 +592,46 @@ impl SimpleScaler {
 
         pods
     }
-}
 
-fn diff(
-    namespace: &str,
-    provers: HashMap<GPUPoolKey, u32>,
-    clusters: &Clusters,
-    requests: &mut HashMap<String, ScaleRequest>,
-) {
-    provers
-        .into_iter()
-        .for_each(|(GPUPoolKey { cluster, gpu }, n)| {
-            let prover = gpu_to_prover(gpu);
+    fn diff(
+        &self,
+        namespace: &str,
+        replicas: HashMap<String, usize>,
+        clusters: &Clusters,
+        requests: &mut HashMap<String, ScaleRequest>,
+    ) {
+        let deployment_name = self.deployment.clone();
+        replicas.into_iter().for_each(|(cluster, replicas)| {
             clusters
                 .clusters
                 .get(&cluster)
                 .and_then(|c| c.namespaces.get(namespace))
-                .and_then(|ns| ns.deployments.get(&prover))
+                .and_then(|ns| ns.deployments.get(&deployment_name))
                 .map_or_else(
                     || {
                         tracing::error!(
                             "Wasn't able to find deployment {} in cluster {}, namespace {}",
-                            prover,
+                            deployment_name,
                             cluster,
                             namespace
                         )
                     },
-                    |d| {
-                        if d.desired != n as i32 {
+                    |deployment| {
+                        if deployment.desired != replicas as i32 {
                             requests
                                 .entry(cluster.clone())
                                 .or_default()
                                 .deployments
                                 .push(ScaleDeploymentRequest {
                                     namespace: namespace.into(),
-                                    name: prover.clone(),
-                                    size: n as i32,
+                                    name: deployment_name.clone(),
+                                    size: replicas as i32,
                                 });
                         }
                     },
                 );
         })
+    }
 }
 
 /// is_namespace_running returns true if there are some pods running in it.
@@ -638,7 +678,7 @@ impl Task for Scaler {
                         AUTOSCALER_METRICS.provers[&(k.cluster.clone(), ns.clone(), k.gpu)]
                             .set(*num as u64);
                     }
-                    diff(ns, provers, &guard.clusters, &mut scale_requests);
+                    GpuScaler::diff(ns, provers, &guard.clusters, &mut scale_requests);
                 }
 
                 // Simple Scalers.
@@ -647,15 +687,15 @@ impl Task for Scaler {
                         .get(&(ppv.to_string(), scaler.queue_report_field.clone()))
                         .cloned()
                         .unwrap_or(0);
-                    tracing::debug!("Running eval for namespace {ns}, PPV {ppv}, simple scaler {} found queue {q}", scaler.pod_name_prefix);
+                    tracing::debug!("Running eval for namespace {ns}, PPV {ppv}, simple scaler {} found queue {q}", scaler.deployment);
                     if q > 0 || is_namespace_running(ns, &guard.clusters) {
-                        let pods = scaler.run(ns, q, &guard.clusters);
-                        for (k, num) in &pods {
+                        let replicas = scaler.run(ns, q, &guard.clusters);
+                        for (k, num) in &replicas {
                             AUTOSCALER_METRICS.jobs
-                                [&(scaler.pod_name_prefix.clone(), k.clone(), ns.clone())]
+                                [&(scaler.deployment.clone(), k.clone(), ns.clone())]
                                 .set(*num as u64);
                         }
-                        // TODO: diff and add into scale_requests.
+                        scaler.diff(ns, replicas, &guard.clusters, &mut scale_requests);
                     }
                 }
             }

prover/crates/bin/prover_autoscaler/src/global/watcher.rs

@@ -38,11 +38,12 @@ pub fn check_is_ready(v: &Vec<bool>) -> Result<()> {
 pub struct Watcher {
     /// List of base URLs of all agents.
     pub cluster_agents: Vec<Arc<Url>>,
+    pub dry_run: bool,
     pub data: Arc<Mutex<WatchedData>>,
 }
 
 impl Watcher {
-    pub fn new(agent_urls: Vec<String>) -> Self {
+    pub fn new(agent_urls: Vec<String>, dry_run: bool) -> Self {
         let size = agent_urls.len();
         Self {
             cluster_agents: agent_urls
@@ -54,6 +55,7 @@ impl Watcher {
                     )
                 })
                 .collect(),
+            dry_run,
             data: Arc::new(Mutex::new(WatchedData {
                 clusters: Clusters::default(),
                 is_ready: vec![false; size],
@@ -80,6 +82,7 @@ impl Watcher {
                 .collect();
         }
 
+        let dry_run = self.dry_run;
         let handles: Vec<_> = id_requests
             .into_iter()
             .map(|(id, sr)| {
@@ -92,6 +95,10 @@ impl Watcher {
                 tokio::spawn(async move {
                     let mut headers = HeaderMap::new();
                     headers.insert(CONTENT_TYPE, HeaderValue::from_static("application/json"));
+                    if dry_run {
+                        tracing::info!("Dry-run mode, not sending the request.");
+                        return Ok((id, Ok(ScaleResponse::default())));
+                    }
                     let response = send_request_with_retries(
                         &url,
                         MAX_RETRIES,

prover/crates/bin/prover_autoscaler/src/main.rs

@@ -110,7 +110,8 @@ async fn main() -> anyhow::Result<()> {
             let interval = scaler_config.scaler_run_interval.unsigned_abs();
             let exporter_config = PrometheusExporterConfig::pull(scaler_config.prometheus_port);
             tasks.push(tokio::spawn(exporter_config.run(stop_receiver.clone())));
-            let watcher = global::watcher::Watcher::new(scaler_config.agents.clone());
+            let watcher =
+                global::watcher::Watcher::new(scaler_config.agents.clone(), scaler_config.dry_run);
             let queuer = global::queuer::Queuer::new(scaler_config.prover_job_monitor_url.clone());
             let scaler = global::scaler::Scaler::new(watcher.clone(), queuer, scaler_config);
             tasks.extend(get_tasks(watcher, scaler, interval, stop_receiver)?);