Merge pull request #18535 from taosdata/enh/addTraceInfo

enh: add trace info

include/os/osSocket.h

@@ -169,6 +169,9 @@ void        taosSetMaskSIGPIPE();
 uint32_t    taosInetAddr(const char *ipAddr);
 const char *taosInetNtoa(struct in_addr ipInt, char *dstStr, int32_t len);
 
+uint64_t taosHton64(uint64_t val);
+uint64_t taosNtoh64(uint64_t val);
+
 #ifdef __cplusplus
 }
 #endif

source/libs/transport/inc/transComm.h

@@ -169,6 +169,7 @@ typedef struct {
   char spi : 2;
   char hasEpSet : 2;  // contain epset or not, 0(default): no epset, 1: contain epset
 
+  uint64_t timestamp;
   char     user[TSDB_UNI_LEN];
   uint32_t magicNum;
   STraceId traceId;

source/libs/transport/src/transCli.c

@@ -758,6 +758,14 @@ static void cliSendCb(uv_write_t* req, int status) {
   SCliConn* pConn = transReqQueueRemove(req);
   if (pConn == NULL) return;
 
+  SCliMsg* pMsg = !transQueueEmpty(&pConn->cliMsgs) ? transQueueGet(&pConn->cliMsgs, 0) : NULL;
+  if (pMsg != NULL) {
+    int64_t cost = taosGetTimestampUs() - pMsg->st;
+    if (cost > 1000) {
+      tWarn("%s conn %p send cost:%dus, send exception", CONN_GET_INST_LABEL(pConn), pConn, (int)cost);
+    }
+  }
+
   if (status == 0) {
     tTrace("%s conn %p data already was written out", CONN_GET_INST_LABEL(pConn), pConn);
   } else {
@@ -806,6 +814,7 @@ void cliSend(SCliConn* pConn) {
     pHead->traceId = pMsg->info.traceId;
     pHead->magicNum = htonl(TRANS_MAGIC_NUM);
   }
+  pHead->timestamp = taosHton64(taosGetTimestampUs());
 
   if (pHead->persist == 1) {
     CONN_SET_PERSIST_BY_APP(pConn);
@@ -1662,6 +1671,7 @@ int transReleaseCliHandle(void* handle) {
 
   SCliMsg* cmsg = taosMemoryCalloc(1, sizeof(SCliMsg));
   cmsg->msg = tmsg;
+  cmsg->st = taosGetTimestampUs();
   cmsg->type = Release;
   cmsg->ctx = pCtx;
 

source/libs/transport/src/transSvr.c

@@ -231,14 +231,29 @@ static bool uvHandleReq(SSvrConn* pConn) {
     }
   }
   STraceId* trace = &pHead->traceId;
+
+  int64_t        cost = taosGetTimestampUs() - taosNtoh64(pHead->timestamp);
+  static int64_t EXCEPTION_LIMIT_US = 100 * 1000;
+
   if (pConn->status == ConnNormal && pHead->noResp == 0) {
     transRefSrvHandle(pConn);
-
+    if (cost >= EXCEPTION_LIMIT_US) {
-    tGDebug("%s conn %p %s received from %s, local info:%s, len:%d", transLabel(pTransInst), pConn,
+      tGWarn("%s conn %p %s received from %s, local info:%s, len:%d, cost:%dus, recv exception", transLabel(pTransInst),
-            TMSG_INFO(transMsg.msgType), pConn->dst, pConn->src, msgLen);
+             pConn, TMSG_INFO(transMsg.msgType), pConn->dst, pConn->src, msgLen, (int)cost);
     } else {
-    tGDebug("%s conn %p %s received from %s, local info:%s, len:%d, resp:%d, code:%d", transLabel(pTransInst), pConn,
+      tGDebug("%s conn %p %s received from %s, local info:%s, len:%d, cost:%dus", transLabel(pTransInst), pConn,
-            TMSG_INFO(transMsg.msgType), pConn->dst, pConn->src, msgLen, pHead->noResp, transMsg.code);
+              TMSG_INFO(transMsg.msgType), pConn->dst, pConn->src, msgLen, (int)cost);
+    }
+  } else {
+    if (cost >= EXCEPTION_LIMIT_US) {
+      tGWarn("%s conn %p %s received from %s, local info:%s, len:%d, resp:%d, code:%d, cost:%dus, recv exception",
+             transLabel(pTransInst), pConn, TMSG_INFO(transMsg.msgType), pConn->dst, pConn->src, msgLen, pHead->noResp,
+             transMsg.code, (int)(cost));
+    } else {
+      tGWarn("%s conn %p %s received from %s, local info:%s, len:%d, resp:%d, code:%d, cost:%dus",
+             transLabel(pTransInst), pConn, TMSG_INFO(transMsg.msgType), pConn->dst, pConn->src, msgLen, pHead->noResp,
+             transMsg.code, (int)(cost));
+    }
   }
 
   // pHead->noResp = 1,

source/os/src/osSocket.c

@@ -1103,3 +1103,30 @@ void taosWinSocketInit() {
 #else
 #endif
 }
+
+uint64_t taosHton64(uint64_t val) {
+#if defined(WINDOWS) || defined(DARWIN)
+  return ((val & 0x00000000000000ff) << 7 * 8) | ((val & 0x000000000000ff00) << 5 * 8) |
+         ((val & 0x0000000000ff0000) << 3 * 8) | ((val & 0x00000000ff000000) << 1 * 8) |
+         ((val & 0x000000ff00000000) >> 1 * 8) | ((val & 0x0000ff0000000000) >> 3 * 8) |
+         ((val & 0x00ff000000000000) >> 5 * 8) | ((val & 0xff00000000000000) >> 7 * 8);
+#else
+  if (__BYTE_ORDER == __LITTLE_ENDIAN) {
+    return (((uint64_t)htonl((int)((val << 32) >> 32))) << 32) | (unsigned int)htonl((int)(val >> 32));
+  } else if (__BYTE_ORDER == __BIG_ENDIAN) {
+    return val;
+  }
+#endif
+}
+
+uint64_t taosNtoh64(uint64_t val) {
+#if defined(WINDOWS) || defined(DARWIN)
+  return taosHton64(val);
+#else
+  if (__BYTE_ORDER == __LITTLE_ENDIAN) {
+    return (((uint64_t)htonl((int)((val << 32) >> 32))) << 32) | (unsigned int)htonl((int)(val >> 32));
+  } else if (__BYTE_ORDER == __BIG_ENDIAN) {
+    return val;
+  }
+#endif
+}

tests/system-test/0-others/compatibility.py

@@ -77,6 +77,7 @@ class TDTestCase:
 
 
 	def run(self):
+		print(f"start taosd run")
 		bPath=self.getBuildPath()
 		cPath=self.getCfgPath()
 		dbname = "test"
@@ -86,17 +87,20 @@ class TDTestCase:
 		tableNumbers=100
 		recordNumbers1=100
 		recordNumbers2=1000
-        tdsqlF=tdCom.newTdSql()
+		#tdsqlF=tdCom.newTdSql()
-        print(tdsqlF)
+		#print(tdsqlF)
-        tdsqlF.query(f"SELECT SERVER_VERSION();")
-        print(tdsqlF.query(f"SELECT SERVER_VERSION();"))
-        oldServerVersion=tdsqlF.queryResult[0][0]
-        tdLog.info(f"Base server version is {oldServerVersion}")
-        tdsqlF.query(f"SELECT CLIENT_VERSION();")
 
-        # the oldClientVersion can't be updated in the same python process,so the version is new compiled verison
+		oldServerVersion = '3.0.1.0' 
-        oldClientVersion=tdsqlF.queryResult[0][0]
+		#tdsqlF.query(f"SELECT SERVER_VERSION();")
-        tdLog.info(f"Base client version is {oldClientVersion}")
+		#print(tdsqlF.query(f"SELECT SERVER_VERSION();"))
+		#oldServerVersion=tdsqlF.queryResult[0][0]
+		#tdLog.info(f"Base server version is {oldServerVersion}")
+		#tdsqlF.query(f"SELECT CLIENT_VERSION();")
+		#	
+		## the oldClientVersion can't be updated in the same python process,so the version is new compiled verison
+		#oldClientVersion=tdsqlF.queryResult[0][0]
+
+		#tdLog.info(f"Base client version is {oldClientVersion}")
 
 		tdLog.printNoPrefix(f"==========step1:prepare and check data in old version-{oldServerVersion}")
 		tdLog.info(f" LD_LIBRARY_PATH=/usr/lib  taosBenchmark -t {tableNumbers} -n {recordNumbers1} -y  ")