fix(rpc): avoid fd leak

source/libs/transport/src/transCli.c

@@ -21,15 +21,16 @@ typedef struct SCliConn {
   uv_connect_t connReq;
   uv_stream_t* stream;
   uv_write_t   writeReq;
+
   void* hostThrd;
+  int   hThrdIdx;
+
   SConnBuffer readBuf;
-  void*        data;
   STransQueue cliMsgs;
   queue       conn;
   uint64_t    expireTime;
-  int          hThrdIdx;
-  STransCtx    ctx;
 
+  STransCtx  ctx;
   bool       broken;  // link broken or not
   ConnStatus status;  //
 
@@ -157,13 +158,11 @@ static void cliWalkCb(uv_handle_t* handle, void* arg);
       transClearBuffer(&conn->readBuf);                                                  \
       transFreeMsg(transContFromHead((char*)head));                                      \
       tDebug("cli conn %p receive release request, ref: %d", conn, T_REF_VAL_GET(conn)); \
-      while (T_REF_VAL_GET(conn) > 1) {                                                  \
-        transUnrefCliHandle(conn);                                                       \
-      }                                                                                  \
-      if (T_REF_VAL_GET(conn) == 1) {                                                    \
+      if (T_REF_VAL_GET(conn) > 1) {                                                     \
         transUnrefCliHandle(conn);                                                       \
       }                                                                                  \
       destroyCmsg(pMsg);                                                                 \
+      addConnToPool(((SCliThrdObj*)conn->hostThrd)->pool, conn);                         \
       return;                                                                            \
     }                                                                                    \
   } while (0)

source/libs/transport/src/transSrv.c

@@ -35,7 +35,6 @@ typedef struct SSrvConn {
   uv_timer_t pTimer;
 
   queue       queue;
-  int         persist;  // persist connection or not
   SConnBuffer readBuf;  // read buf,
   int         inType;
   void*       pTransInst;  // rpc init
@@ -138,6 +137,7 @@ static void destroySmsg(SSrvMsg* smsg);
 // check whether already read complete packet
 static SSrvConn* createConn(void* hThrd);
 static void      destroyConn(SSrvConn* conn, bool clear /*clear handle or not*/);
+static int       reallocConnRefHandle(SSrvConn* conn);
 
 static void uvHandleQuit(SSrvMsg* msg, SWorkThrdObj* thrd);
 static void uvHandleRelease(SSrvMsg* msg, SWorkThrdObj* thrd);
@@ -517,7 +517,7 @@ void uvWorkerAsyncCb(uv_async_t* handle) {
       int64_t    refId = transMsg.refId;
       SExHandle* exh2 = uvAcquireExHandle(refId);
       if (exh2 == NULL || exh1 != exh2) {
-        tTrace("server handle %p except msg, ignore it", exh1);
+        tTrace("server handle except msg %p, ignore it", exh1);
         uvReleaseExHandle(refId);
         destroySmsg(msg);
         continue;
@@ -581,7 +581,8 @@ void uvOnAcceptCb(uv_stream_t* stream, int status) {
 
   if (uv_accept(stream, (uv_stream_t*)cli) == 0) {
     if (pObj->numOfWorkerReady < pObj->numOfThreads) {
-      tError("worker-threads are not ready for all, need %d instead of %d.", pObj->numOfThreads, pObj->numOfWorkerReady);
+      tError("worker-threads are not ready for all, need %d instead of %d.", pObj->numOfThreads,
+             pObj->numOfWorkerReady);
       uv_close((uv_handle_t*)cli, NULL);
       return;
     }
@@ -787,6 +788,19 @@ static void destroyConn(SSrvConn* conn, bool clear) {
     // uv_shutdown(req, (uv_stream_t*)conn->pTcp, uvShutDownCb);
   }
 }
+static int reallocConnRefHandle(SSrvConn* conn) {
+  uvReleaseExHandle(conn->refId);
+  uvRemoveExHandle(conn->refId);
+  // avoid app continue to send msg on invalid handle
+  SExHandle* exh = taosMemoryMalloc(sizeof(SExHandle));
+  exh->handle = conn;
+  exh->pThrd = conn->hostThrd;
+  exh->refId = uvAddExHandle(exh);
+  uvAcquireExHandle(exh->refId);
+  conn->refId = exh->refId;
+
+  return 0;
+}
 static void uvDestroyConn(uv_handle_t* handle) {
   SSrvConn* conn = handle->data;
   if (conn == NULL) {
@@ -859,7 +873,8 @@ void* transInitServer(uint32_t ip, uint32_t port, char* label, int numOfThreads,
   snprintf(pipeName, sizeof(pipeName), "\\\\?\\pipe\\trans.rpc.%p-%lu", taosSafeRand(), GetCurrentProcessId());
 #else
   char pipeName[PATH_MAX] = {0};
-  snprintf(pipeName, sizeof(pipeName), "%s%spipe.trans.rpc.%08X-%lu", tsTempDir, TD_DIRSEP, taosSafeRand(), taosGetSelfPthreadId());
+  snprintf(pipeName, sizeof(pipeName), "%s%spipe.trans.rpc.%08X-%lu", tsTempDir, TD_DIRSEP, taosSafeRand(),
+           taosGetSelfPthreadId());
 #endif
   assert(0 == uv_pipe_bind(&srv->pipeListen, pipeName));
   assert(0 == uv_listen((uv_stream_t*)&srv->pipeListen, SOMAXCONN, uvPipeListenCb));
@@ -958,6 +973,7 @@ void uvHandleQuit(SSrvMsg* msg, SWorkThrdObj* thrd) {
 }
 void uvHandleRelease(SSrvMsg* msg, SWorkThrdObj* thrd) {
   SSrvConn* conn = msg->pConn;
+  reallocConnRefHandle(conn);
   if (conn->status == ConnAcquire) {
     if (!transQueuePush(&conn->srvMsgs, msg)) {
       return;

tests/pytest/util/dnodes.py

@@ -35,7 +35,7 @@ class TDSimClient:
             "tableIncStepPerVnode": "10000",
             "maxVgroupsPerDb": "1000",
             "sdbDebugFlag": "143",
-            "rpcDebugFlag": "135",
+            "rpcDebugFlag": "143",
             "tmrDebugFlag": "131",
             "cDebugFlag": "135",
             "udebugFlag": "135",
@@ -136,7 +136,7 @@ class TDDnode:
             "tsdbDebugFlag": "135",
             "mDebugFlag": "135",
             "sdbDebugFlag": "135",
-            "rpcDebugFlag": "135",
+            "rpcDebugFlag": "143",
             "tmrDebugFlag": "131",
             "cDebugFlag": "135",
             "httpDebugFlag": "135",