chore: default batch size to 32, refine code style

src/main/java/io/lettuce/core/AutoBatchFlushOptions.java

@@ -24,7 +24,7 @@ public class AutoBatchFlushOptions implements Serializable {
 
     public static final int DEFAULT_WRITE_SPIN_COUNT = 16;
 
-    public static final int DEFAULT_BATCH_SIZE = 20;
+    public static final int DEFAULT_BATCH_SIZE = 32;
 
     public static final boolean DEFAULT_USE_MPSC_QUEUE = true;
 

src/main/java/io/lettuce/core/protocol/DefaultAutoBatchFlushEndpoint.java

@@ -650,13 +650,22 @@ private void scheduleSendJobIfNeeded(final ContextualChannel chan) {
         }
 
         if (chan.context.autoBatchFlushEndPointContext.hasOngoingSendLoop.tryEnter()) {
-            // Benchmark result of using tryEnterSafeGetVolatile() or not (1 thread, async get):
-            // 1. uses tryEnterSafeGetVolatile() to avoid unnecessary eventLoop.execute() calls
-            // Avg latency: 3.2956217278663s
-            // Avg QPS: 495238.50056392356/s
-            // 2. uses eventLoop.execute() directly
-            // Avg latency: 3.2677197021496998s
-            // Avg QPS: 476925.0751855796/s
+            // Benchmark result:
+            // Redis:
+            // engine: 7.1.0
+            // server: AWS elasticcache cache.r7g.large
+            // Client: EC2-c5n.2xlarge
+            // Test Model:
+            // multi-thread sync exists (./bench-multi-thread-exists.sh -b 32 -s 10 -n 80000 -t 64)
+            // Test Parameter:
+            // thread num: 64, loop num: 80000, batch size: 32, write spin count: 10
+            //
+            // With tryEnter():
+            // Avg latency: 0.64917373203125ms
+            // Avg QPS: 196037.67991971457/s
+            // Without tryEnter():
+            // Avg latency: 0.6618976359375001ms
+            // Avg QPS: 192240.1301551348/s
             eventLoop.execute(() -> loopSend(chan, true));
         }
 
@@ -681,31 +690,33 @@ private void loopSend(final ContextualChannel chan, boolean entered) {
     }
 
     private void loopSend0(final AutoBatchFlushEndPointContext autoBatchFlushEndPointContext, final ContextualChannel chan,
-            int remainingSpinnCount, boolean entered) {
+            int remainingSpinnCount, final boolean entered) {
         do {
-            final int count = pollBatch(autoBatchFlushEndPointContext, chan);
-            if (count < 0) {
-                return;
-            }
+            final int count = DefaultAutoBatchFlushEndpoint.this.pollBatch(autoBatchFlushEndPointContext, chan);
             if (count < batchSize) {
-                if (!entered) {
-                    return;
-                }
-                // queue was empty
-                // The send loop will be triggered later when a new task is added,
-                // // Don't setUnsafe here because loopSend0() may result in a delayed loopSend() call.
-                autoBatchFlushEndPointContext.hasOngoingSendLoop.exit();
-                if (taskQueue.isEmpty()) {
+                if (count < 0) {
                     return;
+                } else {
+                    break;
                 }
-                entered = false;
-                // // Guarantee thread-safety: no dangling tasks in the queue.
             }
         } while (--remainingSpinnCount > 0);
 
-        final boolean finalEntered = entered;
-        // Don't need to exitUnsafe since we still have an ongoing consume tasks in this thread.
-        chan.eventLoop().execute(() -> loopSend(chan, finalEntered));
+        if (remainingSpinnCount <= 0) {
+            // Don't need to exitUnsafe since we still have an ongoing consume tasks in this thread.
+            chan.eventLoop().execute(() -> loopSend(chan, entered));
+            return;
+        }
+
+        if (entered) {
+            // queue was empty
+            // The send loop will be triggered later when a new task is added,
+            autoBatchFlushEndPointContext.hasOngoingSendLoop.exit();
+            // Guarantee thread-safety: no dangling tasks in the queue, see scheduleSendJobIfNeeded()
+            if (!taskQueue.isEmpty()) {
+                loopSend0(autoBatchFlushEndPointContext, chan, remainingSpinnCount, false);
+            }
+        }
     }
 
     private int pollBatch(final AutoBatchFlushEndPointContext autoBatchFlushEndPointContext, ContextualChannel chan) {
@@ -771,24 +782,6 @@ private void trySetEndpointQuiescence(ContextualChannel chan) {
         }
     }
 
-    private void onEndpointQuiescence() {
-        if (channel.context.initialState == ConnectionContext.State.ENDPOINT_CLOSED) {
-            return;
-        }
-
-        this.logPrefix = null;
-        // Create happens-before with channelActive()
-        if (!CHANNEL.compareAndSet(this, DummyContextualChannelInstances.CHANNEL_WILL_RECONNECT,
-                DummyContextualChannelInstances.CHANNEL_CONNECTING)) {
-            onUnexpectedState("onEndpointQuiescence", ConnectionContext.State.WILL_RECONNECT);
-            return;
-        }
-
-        // neither connectionWatchdog nor doReconnectOnEndpointQuiescence could be null
-        // noinspection DataFlowIssue
-        connectionWatchdog.reconnectOnAutoBatchFlushEndpointQuiescence();
-    }
-
     private void onWillReconnect(@Nonnull final ConnectionContext.CloseStatus closeStatus,
             final AutoBatchFlushEndPointContext autoBatchFlushEndPointContext) {
         final @Nullable Deque<RedisCommand<?, ?, ?>> retryableFailedToSendTasks = autoBatchFlushEndPointContext
@@ -831,6 +824,25 @@ private void onWontReconnect(@Nonnull final ConnectionContext.CloseStatus closeS
         }
     }
 
+    private void onEndpointQuiescence() {
+        if (channel.context.initialState == ConnectionContext.State.ENDPOINT_CLOSED) {
+            return;
+        }
+
+        this.logPrefix = null;
+        // Create happens-before with channelActive()
+        if (!CHANNEL.compareAndSet(this, DummyContextualChannelInstances.CHANNEL_WILL_RECONNECT,
+                DummyContextualChannelInstances.CHANNEL_CONNECTING)) {
+            onUnexpectedState("onEndpointQuiescence", ConnectionContext.State.WILL_RECONNECT);
+            return;
+        }
+
+        // notify connectionWatchDog that it is safe to reconnect now.
+        // neither connectionWatchdog nor doReconnectOnEndpointQuiescence could be null
+        // noinspection DataFlowIssue
+        connectionWatchdog.reconnectOnAutoBatchFlushEndpointQuiescence();
+    }
+
     private void offerFirstAll(Deque<RedisCommand<?, ?, ?>> commands) {
         commands.forEach(cmd -> {
             if (cmd instanceof DemandAware.Sink) {