update min user backlog

src/queue.rs

@@ -361,10 +361,10 @@ impl QueueActor {
                 let system_wait = system_backlog
                     .checked_sub(status.system.oldest)
                     .unwrap_or_default();
-                self.logger.debug(&format!("User wait: {:?} due to {:?} for oldest {:?}, system wait: {:?} due to {:?} for oldest {:?}",
-                       user_wait, user_backlog, status.user.oldest,
-                       system_wait, system_backlog, status.system.oldest));
                 let slow = user_wait >= system_wait + sec;
+                self.logger.debug(&format!("User wait: {:?} due to {:?} for oldest {:?}, system wait: {:?} due to {:?} for oldest {:?} -> {}",
+                       user_wait, user_backlog, status.user.oldest,
+                       system_wait, system_backlog, status.system.oldest, if slow { "system" } else { "user" }));
                 (min(user_wait, system_wait), AcquireQuery { slow })
             } else {
                 self.logger

src/stats.rs

@@ -133,14 +133,14 @@ impl StatsRecorder {
     }
 
     pub fn min_user_backlog(&self) -> Duration {
-        // The average batch has 60 positions, analysed with 2_000_000 nodes
-        // each. Top end clients take no longer than 35 seconds.
-        let best_batch_seconds = 35;
+        // The average batch has 60 positions, analysed with 1_500_000 nodes
+        // each. Top end clients take no longer than 60 seconds.
+        let best_batch_seconds = 60;
 
         // Estimate how long this client would take for the next batch,
         // capped at timeout.
         let estimated_batch_seconds =
-            u64::from(min(6 * 60, 60 * 2_000_000 / max(1, self.nnue_nps.nps)));
+            u64::from(min(6 * 60, 60 * 1_500_000 / max(1, self.nnue_nps.nps)));
 
         // Its worth joining if queue wait time + estimated time < top client
         // time on empty queue.