diff --git a/CHANGELOG.md b/CHANGELOG.md
index b9a32777..efc4355a 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -6,6 +6,10 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+### Fixed — Connection counter double-decrement (2026-04-12)
+
+- **monoio handler: `maxclients reached` after first disconnect.** `handle_connection_sharded_monoio` called `record_connection_closed()` unconditionally at its exit, while the caller (`conn_accept.rs`) also called it in the non-migration branch. The `AtomicU64` counter wrapped to `u64::MAX` on the second `fetch_sub`, causing every subsequent `try_accept_connection` to reject against `maxclients`. Removed the handler-level decrement to restore symmetry with the `try_accept_connection` increment owned by the caller. Verified: 10 sequential SETs now succeed; `redis-benchmark SET p=16 c=50` reports real throughput (1.25M+ req/s) instead of rejection errors.
+
 ### Added — Graph Engine Integration (v0.1.4, 2026-04-11)
 
 - **Property graph engine** (`src/graph/`, feature-gated under `graph`): segment-aligned CSR storage with SlotMap generational indices, ArcSwap lock-free reads, Roaring validity bitmaps, and Rabbit Order compaction for cache locality. 8,500+ LOC, 319 tests.
diff --git a/src/server/conn/handler_monoio.rs b/src/server/conn/handler_monoio.rs
index 95bfd22a..ef228a27 100644
--- a/src/server/conn/handler_monoio.rs
+++ b/src/server/conn/handler_monoio.rs
@@ -2327,6 +2327,11 @@ pub(crate) async fn handle_connection_sharded_monoio<
         }
     }
 
-    crate::admin::metrics_setup::record_connection_closed();
+    // NOTE: connection close is recorded by the caller (conn_accept.rs) to
+    // preserve symmetry with `try_accept_connection`, which owns the
+    // increment.  Decrementing here too produces a double-decrement on the
+    // AtomicU64 counter — it wraps to u64::MAX on the second subtraction
+    // and all subsequent `try_accept_connection` comparisons against
+    // `maxclients` reject new connections.
     (MonoioHandlerResult::Done, None)
 }
diff --git a/src/shard/conn_accept.rs b/src/shard/conn_accept.rs
index abec51f2..69e3a8ee 100644
--- a/src/shard/conn_accept.rs
+++ b/src/shard/conn_accept.rs
@@ -700,6 +700,11 @@ pub(crate) fn spawn_migrated_monoio_connection(
             fd,
             e
         );
+        // Source shard's `try_accept_connection` already counted this
+        // client; its wrapper skipped the decrement because the migration
+        // succeeded from the source's perspective.  Since we cannot hand
+        // ownership to the handler, we own the balancing decrement here.
+        crate::admin::metrics_setup::record_connection_closed();
         return; // std_stream Drop closes FD
     }
     match monoio::net::TcpStream::from_std(std_stream) {
@@ -765,6 +770,12 @@ pub(crate) fn spawn_migrated_monoio_connection(
                     Some(&state),
                 )
                 .await;
+                // Migrated connection: the source shard's wrapper skipped the
+                // decrement (because `_migrated == true`), so this target-shard
+                // spawn site owns the balancing decrement.  Without this the
+                // AtomicU64 `CONNECTED_CLIENTS` counter would leak upward on
+                // every migration and eventually trip `maxclients`.
+                crate::admin::metrics_setup::record_connection_closed();
             });
         }
         Err(e) => {
@@ -773,6 +784,10 @@ pub(crate) fn spawn_migrated_monoio_connection(
                 shard_id,
                 e
             );
+            // Same rationale as the set_nonblocking failure above:
+            // the source-side increment stands but no handler will run,
+            // so we own the balancing decrement.
+            crate::admin::metrics_setup::record_connection_closed();
         }
     }
 }