added biased select!'s, better timestamps

erhant · erhant · commit 6f66d52a8a96 · 2024-12-13T13:51:59.000+03:00
diff --git a/Cargo.lock b/Cargo.lock
diff --git a/Cargo.toml b/Cargo.toml
@@ -9,7 +9,7 @@ default-members = ["compute"]
 
 [workspace.package]
 edition = "2021"
-version = "0.2.29"
+version = "0.2.30"
 license = "Apache-2.0"
 readme = "README.md"
 
@@ -18,14 +18,9 @@ readme = "README.md"
 inherits = "release"
 debug = true
 
-
-# See more keys and their definitions at https://doc.rust-lang.org/cargo/reference/manifest.html
-
 [workspace.dependencies]
 # async stuff
-tokio-util = { version = "0.7.10", features = [
-  "rt",
-] } # tokio-util provides CancellationToken
+tokio-util = { version = "0.7.10", features = ["rt"] }
 tokio = { version = "1", features = ["macros", "rt-multi-thread", "signal"] }
 async-trait = "0.1.81"
 
diff --git a/compute/src/main.rs b/compute/src/main.rs
@@ -118,8 +118,9 @@ async fn main() -> Result<()> {
     let node_token = cancellation.clone();
     task_tracker.spawn(async move {
         if let Err(err) = node.run(node_token).await {
-            log::error!("Node launch error: {}", err);
-            panic!("Node failed.")
+            log::error!("Error within main node loop: {}", err);
+            log::error!("Shutting down node.");
+            node.shutdown().await.expect("could not shutdown node");
         };
         log::info!("Closing node.")
     });
diff --git a/compute/src/node.rs b/compute/src/node.rs
@@ -331,10 +331,9 @@ impl DriaComputeNode {
 
         loop {
             tokio::select! {
-                // check peer count every now and then
-                _ = peer_refresh_interval.tick() => self.handle_diagnostic_refresh().await,
-                // available nodes are refreshed every now and then
-                _ = available_node_refresh_interval.tick() => self.handle_available_nodes_refresh().await,
+                // prioritize the branches in the order below
+                biased;
+
                 // a Workflow message to be published is received from the channel
                 // this is expected to be sent by the workflow worker
                 publish_msg_opt = self.publish_rx.recv() => {
@@ -358,6 +357,11 @@ impl DriaComputeNode {
                         break;
                     };
                 },
+
+                // check peer count every now and then
+                _ = peer_refresh_interval.tick() => self.handle_diagnostic_refresh().await,
+                // available nodes are refreshed every now and then
+                _ = available_node_refresh_interval.tick() => self.handle_available_nodes_refresh().await,
                 // a GossipSub message is received from the channel
                 // this is expected to be sent by the p2p client
                 gossipsub_msg_opt = self.message_rx.recv() => {
diff --git a/compute/src/payloads/stats.rs b/compute/src/payloads/stats.rs
@@ -12,7 +12,12 @@ pub struct TaskStats {
     /// Timestamp at which the task was published back to network.
     pub published_at: u128,
     /// Time taken to execute the task.
+    /// FIXME: will be removed after
     pub execution_time: u128,
+    /// Timestamp at which the task execution had started.
+    pub execution_started_at: u128,
+    /// Timestamp at which the task execution had finished.
+    pub execution_ended_time: u128,
 }
 
 impl TaskStats {
@@ -33,7 +38,20 @@ impl TaskStats {
         self
     }
 
+    /// Records the execution start time within `execution_started_at`.
+    pub fn record_execution_started_at(mut self) -> Self {
+        self.execution_started_at = get_current_time_nanos();
+        self
+    }
+
+    /// Records the execution end time within `execution_ended_time`.
+    pub fn record_execution_ended_at(mut self) -> Self {
+        self.execution_ended_time = get_current_time_nanos();
+        self
+    }
+
     /// Records the execution time of the task.
+    #[deprecated = "will be removed later"]
     pub fn record_execution_time(mut self, started_at: Instant) -> Self {
         self.execution_time = Instant::now().duration_since(started_at).as_nanos();
         self
diff --git a/compute/src/workers/workflow.rs b/compute/src/workers/workflow.rs
@@ -78,7 +78,7 @@ impl WorkflowsWorker {
 
             if let Some(task) = task {
                 log::info!("Processing single workflow for task {}", task.task_id);
-                WorkflowsWorker::execute((task, self.publish_tx.clone())).await
+                WorkflowsWorker::execute((task, &self.publish_tx)).await
             } else {
                 return self.shutdown();
             };
@@ -108,61 +108,53 @@ impl WorkflowsWorker {
 
             // process the batch
             log::info!("Processing {} workflows in batch", num_tasks);
-            let mut batch = task_buffer
-                .into_iter()
-                .map(|b| (b, self.publish_tx.clone()));
+            let mut batch = task_buffer.into_iter().map(|b| (b, &self.publish_tx));
             match num_tasks {
                 1 => {
-                    let r0 = WorkflowsWorker::execute(batch.next().unwrap()).await;
-                    vec![r0]
+                    WorkflowsWorker::execute(batch.next().unwrap()).await;
                 }
                 2 => {
-                    let (r0, r1) = tokio::join!(
+                    tokio::join!(
                         WorkflowsWorker::execute(batch.next().unwrap()),
                         WorkflowsWorker::execute(batch.next().unwrap())
                     );
-                    vec![r0, r1]
                 }
                 3 => {
-                    let (r0, r1, r2) = tokio::join!(
+                    tokio::join!(
                         WorkflowsWorker::execute(batch.next().unwrap()),
                         WorkflowsWorker::execute(batch.next().unwrap()),
                         WorkflowsWorker::execute(batch.next().unwrap())
                     );
-                    vec![r0, r1, r2]
                 }
                 4 => {
-                    let (r0, r1, r2, r3) = tokio::join!(
+                    tokio::join!(
                         WorkflowsWorker::execute(batch.next().unwrap()),
                         WorkflowsWorker::execute(batch.next().unwrap()),
                         WorkflowsWorker::execute(batch.next().unwrap()),
                         WorkflowsWorker::execute(batch.next().unwrap())
                     );
-                    vec![r0, r1, r2, r3]
                 }
                 5 => {
-                    let (r0, r1, r2, r3, r4) = tokio::join!(
+                    tokio::join!(
                         WorkflowsWorker::execute(batch.next().unwrap()),
                         WorkflowsWorker::execute(batch.next().unwrap()),
                         WorkflowsWorker::execute(batch.next().unwrap()),
                         WorkflowsWorker::execute(batch.next().unwrap()),
                         WorkflowsWorker::execute(batch.next().unwrap())
                     );
-                    vec![r0, r1, r2, r3, r4]
                 }
                 6 => {
-                    let (r0, r1, r2, r3, r4, r5) = tokio::join!(
+                    tokio::join!(
                         WorkflowsWorker::execute(batch.next().unwrap()),
                         WorkflowsWorker::execute(batch.next().unwrap()),
                         WorkflowsWorker::execute(batch.next().unwrap()),
                         WorkflowsWorker::execute(batch.next().unwrap()),
                         WorkflowsWorker::execute(batch.next().unwrap()),
                         WorkflowsWorker::execute(batch.next().unwrap())
                     );
-                    vec![r0, r1, r2, r3, r4, r5]
                 }
                 7 => {
-                    let (r0, r1, r2, r3, r4, r5, r6) = tokio::join!(
+                    tokio::join!(
                         WorkflowsWorker::execute(batch.next().unwrap()),
                         WorkflowsWorker::execute(batch.next().unwrap()),
                         WorkflowsWorker::execute(batch.next().unwrap()),
@@ -171,10 +163,9 @@ impl WorkflowsWorker {
                         WorkflowsWorker::execute(batch.next().unwrap()),
                         WorkflowsWorker::execute(batch.next().unwrap())
                     );
-                    vec![r0, r1, r2, r3, r4, r5, r6]
                 }
                 8 => {
-                    let (r0, r1, r2, r3, r4, r5, r6, r7) = tokio::join!(
+                    tokio::join!(
                         WorkflowsWorker::execute(batch.next().unwrap()),
                         WorkflowsWorker::execute(batch.next().unwrap()),
                         WorkflowsWorker::execute(batch.next().unwrap()),
@@ -184,7 +175,6 @@ impl WorkflowsWorker {
                         WorkflowsWorker::execute(batch.next().unwrap()),
                         WorkflowsWorker::execute(batch.next().unwrap())
                     );
-                    vec![r0, r1, r2, r3, r4, r5, r6, r7]
                 }
                 _ => {
                     unreachable!(
@@ -199,23 +189,28 @@ impl WorkflowsWorker {
 
     /// Executes a single task, and publishes the output.
     pub async fn execute(
-        (input, publish_tx): (WorkflowsWorkerInput, mpsc::Sender<WorkflowsWorkerOutput>),
+        (input, publish_tx): (WorkflowsWorkerInput, &mpsc::Sender<WorkflowsWorkerOutput>),
     ) {
+        let mut stats = input.stats;
+
         let mut memory = ProgramMemory::new();
 
+        // TODO: will be removed later
         let started_at = std::time::Instant::now();
+        stats = stats.record_execution_started_at();
         let result = input
             .executor
             .execute(input.entry.as_ref(), &input.workflow, &mut memory)
             .await;
+        stats = stats.record_execution_ended_at();
 
         let output = WorkflowsWorkerOutput {
             result,
             public_key: input.public_key,
             task_id: input.task_id,
             model_name: input.model_name,
             batchable: input.batchable,
-            stats: input.stats.record_execution_time(started_at),
+            stats: stats.record_execution_time(started_at),
         };
 
         if let Err(e) = publish_tx.send(output).await {
diff --git a/p2p/src/client.rs b/p2p/src/client.rs
@@ -129,16 +129,6 @@ impl DriaP2PClient {
             swarm.dial(rpc_addr.clone())?;
         }
 
-        // add rpcs as explicit peers
-        // TODO: may not be necessary
-        // for rpc_peer_id in &nodes.rpc_peerids {
-        //     log::info!("Adding {} as explicit peer.", rpc_peer_id);
-        //     swarm
-        //         .behaviour_mut()
-        //         .gossipsub
-        //         .add_explicit_peer(rpc_peer_id);
-        // }
-
         // create commander
         let (cmd_tx, cmd_rx) = mpsc::channel(COMMAND_CHANNEL_BUFSIZE);
         let commander = DriaP2PCommander::new(cmd_tx, protocol.clone());
@@ -161,7 +151,9 @@ impl DriaP2PClient {
     pub async fn run(mut self) {
         loop {
             tokio::select! {
-                event = self.swarm.select_next_some() => self.handle_event(event).await,
+                // this is a special keyword that changes the polling order from random to linear,
+                // which will effectively prioritize commands over events
+                biased;
                 command = self.cmd_rx.recv() => match command {
                     Some(c) => self.handle_command(c).await,
                     // channel closed, thus shutting down the network event loop
@@ -170,6 +162,7 @@ impl DriaP2PClient {
                         return
                     },
                 },
+                event = self.swarm.select_next_some() => self.handle_event(event).await,
             }
         }
     }
