Merge pull request #18289 from taosdata/szhou/fixbugs26

feature: add memory rss threshold for creating new query

src/common/inc/tglobal.h

@@ -69,6 +69,7 @@ extern char     tsTempDir[];
 extern int32_t  tsShortcutFlag;
 extern int32_t  tsMaxSqlGroups;
 extern int8_t   tsSortWhenGroupBy;
+extern int32_t  tsQueryRssThreshold;
 
 // query buffer management
 extern int32_t tsQueryBufferSize;  // maximum allowed usage buffer size in MB for each data node during query processing

src/common/src/tglobal.c

@@ -124,6 +124,9 @@ int32_t tsMaxSqlGroups = 1000000;
 // order by first group by column when group by
 int8_t tsSortWhenGroupBy = 1;
 
+// memory rss thresold for creating new query in MB. 0 means no threshold limitation
+int32_t tsQueryRssThreshold = 0;
+
 int32_t tsProjectExecInterval = 10000;   // every 10sec, the projection will be executed once
 int64_t tsMaxRetentWindow = 24 * 3600L;  // maximum time window tolerance
 
@@ -1828,6 +1831,16 @@ static void doInitGlobalConfig(void) {
   cfg.unitType = TAOS_CFG_UTYPE_NONE;
   taosInitConfigOption(cfg);
 
+  cfg.option = "queryRssThreshold";
+  cfg.ptr = &tsQueryRssThreshold;
+  cfg.valType = TAOS_CFG_VTYPE_INT32;
+  cfg.cfgType = TSDB_CFG_CTYPE_B_CONFIG;
+  cfg.minValue = 0;
+  cfg.maxValue = 2048*1024;
+  cfg.ptrLength = 0;
+  cfg.unitType = TAOS_CFG_UTYPE_NONE;
+  taosInitConfigOption(cfg);
+
 #ifdef TD_TSZ
   // lossy compress
   cfg.option = "lossyColumns";

src/inc/taoserror.h

@@ -298,6 +298,7 @@ int32_t* taosGetErrno();
 #define TSDB_CODE_QRY_INVALID_TIME_CONDITION    TAOS_DEF_ERROR_CODE(0, 0x070E)  //"invalid time condition")
 #define TSDB_CODE_QRY_INVALID_SCHEMA_VERSION    TAOS_DEF_ERROR_CODE(0, 0x0710)  //"invalid schema version")
 #define TSDB_CODE_QRY_RESULT_TOO_LARGE          TAOS_DEF_ERROR_CODE(0, 0x0711)  //"result num is too large")
+#define TSDB_CODE_QRY_RSS_THRESHOLD             TAOS_DEF_ERROR_CODE(0, 0x0712)  //"query memory rss threshold")
 
 // grant
 #define TSDB_CODE_GRANT_EXPIRED                 TAOS_DEF_ERROR_CODE(0, 0x0800)  //"License expired"

src/query/src/queryMain.c

@@ -77,6 +77,15 @@ int32_t qCreateQueryInfo(void* tsdb, int32_t vgId, SQueryTableMsg* pQueryMsg, qi
   if (code != TSDB_CODE_SUCCESS) {
     goto _over;
   }
+  
+  float procMemory = 0;
+  if (taosGetProcMemory(&procMemory)) {
+    if (tsQueryRssThreshold > 0 && procMemory >= tsQueryRssThreshold) {
+      qError("Exceeds query memory RSS threshold. RSS: %f, threshold: %d", procMemory, tsQueryRssThreshold);
+      code = TSDB_CODE_QRY_RSS_THRESHOLD;
+      goto _over;
+    }
+  } 
 
   if (pQueryMsg->numOfTables <= 0) {
     qError("Invalid number of tables to query, numOfTables:%d", pQueryMsg->numOfTables);

src/util/inc/tconfig.h

@@ -20,7 +20,7 @@
 extern "C" {
 #endif
 
-#define TSDB_CFG_MAX_NUM    141
+#define TSDB_CFG_MAX_NUM    142
 #define TSDB_CFG_PRINT_LEN  23
 #define TSDB_CFG_OPTION_LEN 24
 #define TSDB_CFG_VALUE_LEN  41

src/util/src/terror.c

@@ -304,6 +304,7 @@ TAOS_DEFINE_ERROR(TSDB_CODE_QRY_INCONSISTAN,              "File inconsistance in
 TAOS_DEFINE_ERROR(TSDB_CODE_QRY_INVALID_TIME_CONDITION,   "One valid time range condition expected")
 TAOS_DEFINE_ERROR(TSDB_CODE_QRY_SYS_ERROR,                "System error")
 TAOS_DEFINE_ERROR(TSDB_CODE_QRY_RESULT_TOO_LARGE,         "result num is too large")
+TAOS_DEFINE_ERROR(TSDB_CODE_QRY_RSS_THRESHOLD,            "Exceed Memory RSS thresold when creating query")
 
 // grant
 TAOS_DEFINE_ERROR(TSDB_CODE_GRANT_EXPIRED,                "License expired")