diff --git a/src/facade/dragonfly_connection.cc b/src/facade/dragonfly_connection.cc
index 2909cf6bd..c5286c61b 100644
--- a/src/facade/dragonfly_connection.cc
+++ b/src/facade/dragonfly_connection.cc
@@ -43,18 +43,13 @@ using nonstd::make_unexpected;
 namespace facade {
 namespace {
 
-void SendProtocolError(RedisParser::Result pres, FiberSocketBase* peer) {
+void SendProtocolError(RedisParser::Result pres, SinkReplyBuilder* builder) {
   string res("-ERR Protocol error: ");
   if (pres == RedisParser::BAD_BULKLEN) {
-    res.append("invalid bulk length\r\n");
+    builder->SendProtocolError("invalid bulk length");
   } else {
     CHECK_EQ(RedisParser::BAD_ARRAYLEN, pres);
-    res.append("invalid multibulk length\r\n");
-  }
-
-  error_code ec = peer->Write(::io::Buffer(res));
-  if (ec) {
-    LOG(WARNING) << "Error " << ec;
+    builder->SendProtocolError("invalid multibulk length");
   }
 }
 
@@ -503,20 +498,17 @@ void Connection::ConnectionFlow(FiberSocketBase* peer) {
   // offending request.
   if (parse_status == ERROR) {
     VLOG(1) << "Error parser status " << parser_error_;
-    ++stats_->parser_err_cnt;
 
     if (redis_parser_) {
-      SendProtocolError(RedisParser::Result(parser_error_), peer);
+      SendProtocolError(RedisParser::Result(parser_error_), orig_builder);
     } else {
-      string_view sv{"CLIENT_ERROR bad command line format\r\n"};
-      error_code ec2 = peer->Write(::io::Buffer(sv));
-      if (ec2) {
-        LOG(WARNING) << "Error " << ec2;
-        ec = ec2;
-      }
+      DCHECK(memcache_parser_);
+      orig_builder->SendProtocolError("bad command line format");
     }
     error_code ec2 = peer->Shutdown(SHUT_RDWR);
     LOG_IF(WARNING, ec2) << "Could not shutdown socket " << ec2;
+
+    FetchBuilderStats(stats_, orig_builder);
   }
 
   if (ec && !FiberSocketBase::IsConnClosed(ec)) {
diff --git a/src/facade/facade.cc b/src/facade/facade.cc
index 69873f4ac..8bc26ca99 100644
--- a/src/facade/facade.cc
+++ b/src/facade/facade.cc
@@ -21,7 +21,7 @@ constexpr size_t kSizeConnStats = sizeof(ConnectionStats);
 
 ConnectionStats& ConnectionStats::operator+=(const ConnectionStats& o) {
   // To break this code deliberately if we add/remove a field to this struct.
-  static_assert(kSizeConnStats == 184);
+  static_assert(kSizeConnStats == 176);
 
   ADD(read_buf_capacity);
   ADD(pipeline_cache_capacity);
@@ -31,7 +31,6 @@ ConnectionStats& ConnectionStats::operator+=(const ConnectionStats& o) {
   ADD(io_write_bytes);
   ADD(command_cnt);
   ADD(pipelined_cmd_cnt);
-  ADD(parser_err_cnt);
   ADD(async_writes_cnt);
   ADD(conn_received_cnt);
   ADD(num_conns);
diff --git a/src/facade/facade_types.h b/src/facade/facade_types.h
index 42a531a10..b25420876 100644
--- a/src/facade/facade_types.h
+++ b/src/facade/facade_types.h
@@ -44,7 +44,6 @@ struct ConnectionStats {
   size_t io_write_bytes = 0;
   uint64_t command_cnt = 0;
   uint64_t pipelined_cmd_cnt = 0;
-  uint64_t parser_err_cnt = 0;
 
   // Writes count that happened via DispatchOperations call.
   uint64_t async_writes_cnt = 0;
diff --git a/src/facade/redis_parser.cc b/src/facade/redis_parser.cc
index c3754c234..5949094fb 100644
--- a/src/facade/redis_parser.cc
+++ b/src/facade/redis_parser.cc
@@ -13,6 +13,7 @@ using namespace std;
 
 namespace {
 
+// When changing this constant, also update `test_large_cmd` test in connection_test.py.
 constexpr int kMaxArrayLen = 65536;
 constexpr int64_t kMaxBulkLen = 64 * (1ul << 20);  // 64MB.
 
diff --git a/src/facade/reply_builder.cc b/src/facade/reply_builder.cc
index 517199f40..5bfb3be49 100644
--- a/src/facade/reply_builder.cc
+++ b/src/facade/reply_builder.cc
@@ -54,7 +54,7 @@ void SinkReplyBuilder::Send(const iovec* v, uint32_t len) {
   }
 
   // Allow batching with up to 8K of data.
-  if ((should_batch_ || should_aggregate_) && batch_.size() + bsize < kMaxBatchSize) {
+  if ((should_batch_ || should_aggregate_) && (batch_.size() + bsize < kMaxBatchSize)) {
     for (unsigned i = 0; i < len; ++i) {
       std::string_view src((char*)v[i].iov_base, v[i].iov_len);
       DVLOG(2) << "Appending to stream " << src;
@@ -71,7 +71,7 @@ void SinkReplyBuilder::Send(const iovec* v, uint32_t len) {
   if (batch_.empty()) {
     ec = sink_->Write(v, len);
   } else {
-    DVLOG(1) << "Sending batch to stream " << sink_ << "\n" << batch_;
+    DVLOG(2) << "Sending batch to stream " << sink_ << "\n" << batch_;
 
     io_write_bytes_ += batch_.size();
 
@@ -162,6 +162,10 @@ void MCReplyBuilder::SendError(string_view str, std::string_view type) {
   SendSimpleString("ERROR");
 }
 
+void MCReplyBuilder::SendProtocolError(std::string_view str) {
+  SendSimpleString(absl::StrCat("CLIENT_ERROR ", str));
+}
+
 void MCReplyBuilder::SendClientError(string_view str) {
   iovec v[] = {IoVec("CLIENT_ERROR "), IoVec(str), IoVec(kCRLF)};
   Send(v, ABSL_ARRAYSIZE(v));
@@ -197,6 +201,8 @@ void RedisReplyBuilder::SetResp3(bool is_resp3) {
 }
 
 void RedisReplyBuilder::SendError(string_view str, string_view err_type) {
+  VLOG(1) << "Error: " << str;
+
   if (err_type.empty()) {
     err_type = str;
     if (err_type == kSyntaxErr)
@@ -214,6 +220,10 @@ void RedisReplyBuilder::SendError(string_view str, string_view err_type) {
   }
 }
 
+void RedisReplyBuilder::SendProtocolError(std::string_view str) {
+  SendError(absl::StrCat("-ERR Protocol error: ", str), "protocol_error");
+}
+
 void RedisReplyBuilder::SendSimpleString(std::string_view str) {
   iovec v[3] = {IoVec(kSimplePref), IoVec(str), IoVec(kCRLF)};
 
@@ -483,7 +493,7 @@ void RedisReplyBuilder::SendStringArrInternal(WrappedStrSpan arr, CollectionType
 }
 
 void ReqSerializer::SendCommand(std::string_view str) {
-  VLOG(1) << "SendCommand: " << str;
+  VLOG(2) << "SendCommand: " << str;
 
   iovec v[] = {IoVec(str), IoVec(kCRLF)};
   ec_ = sink_->Write(v, ABSL_ARRAYSIZE(v));
diff --git a/src/facade/reply_builder.h b/src/facade/reply_builder.h
index 8652c7d48..4dd51bd57 100644
--- a/src/facade/reply_builder.h
+++ b/src/facade/reply_builder.h
@@ -54,6 +54,8 @@ class SinkReplyBuilder {
     SendSimpleString("OK");
   }
 
+  virtual void SendProtocolError(std::string_view str) = 0;
+
   // In order to reduce interrupt rate we allow coalescing responses together using
   // Batch mode. It is controlled by Connection state machine because it makes sense only
   // when pipelined requests are arriving.
@@ -154,6 +156,7 @@ class MCReplyBuilder : public SinkReplyBuilder {
   void SendClientError(std::string_view str);
   void SendNotFound();
   void SendSimpleString(std::string_view str) final;
+  void SendProtocolError(std::string_view str) final;
 
   void SetNoreply(bool noreply) {
     noreply_ = noreply;
@@ -176,6 +179,7 @@ class RedisReplyBuilder : public SinkReplyBuilder {
   void SendStored() override;
   void SendSetSkipped() override;
   virtual void SendError(OpStatus status);
+  void SendProtocolError(std::string_view str) override;
 
   virtual void SendNullArray();   // Send *-1
   virtual void SendEmptyArray();  // Send *0
diff --git a/src/server/server_family.cc b/src/server/server_family.cc
index 924ba7766..924c9e2b7 100644
--- a/src/server/server_family.cc
+++ b/src/server/server_family.cc
@@ -1594,7 +1594,6 @@ void ServerFamily::Info(CmdArgList args, ConnectionContext* cntx) {
     append("total_reads_processed", m.conn_stats.io_read_cnt);
     append("total_writes_processed", m.conn_stats.io_write_cnt);
     append("async_writes_count", m.conn_stats.async_writes_cnt);
-    append("parser_err_count", m.conn_stats.parser_err_cnt);
     append("defrag_attempt_total", m.shard_stats.defrag_attempt_total);
     append("defrag_realloc_total", m.shard_stats.defrag_realloc_total);
     append("defrag_task_invocation_total", m.shard_stats.defrag_task_invocation_total);
diff --git a/src/server/transaction.cc b/src/server/transaction.cc
index b51867d92..54396be79 100644
--- a/src/server/transaction.cc
+++ b/src/server/transaction.cc
@@ -133,8 +133,6 @@ void Transaction::InitShardData(absl::Span<const PerShardCache> shard_index, siz
     auto& sd = shard_data_[i];
     auto& si = shard_index[i];
 
-    CHECK_LT(si.args.size(), 1u << 15);
-
     sd.arg_count = si.args.size();
     sd.arg_start = args_.size();
 
@@ -157,7 +155,7 @@ void Transaction::InitShardData(absl::Span<const PerShardCache> shard_index, siz
     }
   }
 
-  CHECK(args_.size() == num_args);
+  CHECK_EQ(args_.size(), num_args);
 }
 
 void Transaction::InitMultiData(KeyIndex key_index) {
diff --git a/src/server/transaction.h b/src/server/transaction.h
index 94ce97e52..4e3bc7d57 100644
--- a/src/server/transaction.h
+++ b/src/server/transaction.h
@@ -329,16 +329,16 @@ class Transaction {
     char pad[46];  // to make sure PerShardData is 64 bytes and takes full cacheline.
 
     uint32_t arg_start = 0;  // Indices into args_ array.
-    uint16_t arg_count = 0;
-
-    // Accessed within shard thread.
-    // Bitmask of LocalState enums.
-    uint16_t local_mask = 0;
+    uint32_t arg_count = 0;
 
     // Needed to rollback inconsistent schedulings or remove OOO transactions from
     // tx queue.
     uint32_t pq_pos = TxQueue::kEnd;
 
+    // Accessed within shard thread.
+    // Bitmask of LocalState enums.
+    uint16_t local_mask = 0;
+
     // Index of key relative to args in shard that the shard was woken up after blocking wait.
     uint16_t wake_key_pos = UINT16_MAX;
   };
diff --git a/tests/dragonfly/conftest.py b/tests/dragonfly/conftest.py
index b7c6d7845..5de3fa3a6 100644
--- a/tests/dragonfly/conftest.py
+++ b/tests/dragonfly/conftest.py
@@ -80,7 +80,7 @@ def df_factory(request, tmp_dir, test_env) -> DflyInstanceFactory:
         args=request.config.getoption("--df"),
         existing_port=int(existing) if existing else None,
         existing_admin_port=int(existing_admin) if existing_admin else None,
-        existing_mc_port=int(existing_mc) if existing else None,
+        existing_mc_port=int(existing_mc) if existing_mc else None,
         env=test_env
     )
 
diff --git a/tests/dragonfly/connection_test.py b/tests/dragonfly/connection_test.py
index 059ee4d17..8b32e3e71 100644
--- a/tests/dragonfly/connection_test.py
+++ b/tests/dragonfly/connection_test.py
@@ -4,7 +4,8 @@ import asyncio
 from redis import asyncio as aioredis
 import async_timeout
 
-from . import DflyInstance
+from . import DflyInstance, dfly_args
+
 
 async def run_monitor_eval(monitor, expected):
     async with monitor as mon:
@@ -346,7 +347,8 @@ async def test_subscribe_in_pipeline(async_client: aioredis.Redis):
     pipe.echo("three")
     res = await pipe.execute()
 
-    assert res == ['one', ['subscribe', 'ch1', 1], 'two', ['subscribe', 'ch2', 2], 'three']
+    assert res == ['one', ['subscribe', 'ch1', 1],
+                   'two', ['subscribe', 'ch2', 2], 'three']
 
 """
 This test makes sure that Dragonfly can receive blocks of pipelined commands even
@@ -376,6 +378,7 @@ PACKET3 = """
 PING
 """ * 500 + "ECHO DONE\n"
 
+
 async def test_parser_while_script_running(async_client: aioredis.Redis, df_server: DflyInstance):
     sha = await async_client.script_load(BUSY_SCRIPT)
 
@@ -399,3 +402,16 @@ async def test_parser_while_script_running(async_client: aioredis.Redis, df_serv
     await reader.readuntil(b"DONE")
     writer.close()
     await writer.wait_closed()
+
+
+@dfly_args({"proactor_threads": 1})
+async def test_large_cmd(async_client: aioredis.Redis):
+    MAX_ARR_SIZE = 65535
+    res = await async_client.hset('foo', mapping={f"key{i}": f"val{i}" for i in range(MAX_ARR_SIZE // 2)})
+    assert res == MAX_ARR_SIZE // 2
+
+    res = await async_client.mset({f"key{i}": f"val{i}" for i in range(MAX_ARR_SIZE // 2)})
+    assert res
+
+    res = await async_client.mget([f"key{i}" for i in range(MAX_ARR_SIZE)])
+    assert len(res) == MAX_ARR_SIZE