util: record sqls and heap profile when memory usage is higher than 8…

…0% system memory. (#18858) (#20473)

Signed-off-by: ti-srebot <ti-srebot@pingcap.com>

config/config.go

@@ -371,22 +371,24 @@ type Status struct {
 
 // Performance is the performance section of the config.
 type Performance struct {
-	MaxProcs             uint    `toml:"max-procs" json:"max-procs"`
-	MaxMemory            uint64  `toml:"max-memory" json:"max-memory"`
-	StatsLease           string  `toml:"stats-lease" json:"stats-lease"`
-	StmtCountLimit       uint    `toml:"stmt-count-limit" json:"stmt-count-limit"`
-	FeedbackProbability  float64 `toml:"feedback-probability" json:"feedback-probability"`
-	QueryFeedbackLimit   uint    `toml:"query-feedback-limit" json:"query-feedback-limit"`
-	PseudoEstimateRatio  float64 `toml:"pseudo-estimate-ratio" json:"pseudo-estimate-ratio"`
-	ForcePriority        string  `toml:"force-priority" json:"force-priority"`
-	BindInfoLease        string  `toml:"bind-info-lease" json:"bind-info-lease"`
-	TxnTotalSizeLimit    uint64  `toml:"txn-total-size-limit" json:"txn-total-size-limit"`
-	TCPKeepAlive         bool    `toml:"tcp-keep-alive" json:"tcp-keep-alive"`
-	CrossJoin            bool    `toml:"cross-join" json:"cross-join"`
-	RunAutoAnalyze       bool    `toml:"run-auto-analyze" json:"run-auto-analyze"`
-	DistinctAggPushDown  bool    `toml:"distinct-agg-push-down" json:"agg-push-down-join"`
-	CommitterConcurrency int     `toml:"committer-concurrency" json:"committer-concurrency"`
-	MaxTxnTTL            uint64  `toml:"max-txn-ttl" json:"max-txn-ttl"`
+	MaxProcs              uint    `toml:"max-procs" json:"max-procs"`
+	MaxMemory             uint64  `toml:"max-memory" json:"max-memory"`
+	ServerMemoryQuota     uint64  `toml:"server-memory-quota" json:"server-memory-quota"`
+	MemoryUsageAlarmRatio float64 `toml:"memory-usage-alarm-ratio" json:"memory-usage-alarm-ratio"`
+	StatsLease            string  `toml:"stats-lease" json:"stats-lease"`
+	StmtCountLimit        uint    `toml:"stmt-count-limit" json:"stmt-count-limit"`
+	FeedbackProbability   float64 `toml:"feedback-probability" json:"feedback-probability"`
+	QueryFeedbackLimit    uint    `toml:"query-feedback-limit" json:"query-feedback-limit"`
+	PseudoEstimateRatio   float64 `toml:"pseudo-estimate-ratio" json:"pseudo-estimate-ratio"`
+	ForcePriority         string  `toml:"force-priority" json:"force-priority"`
+	BindInfoLease         string  `toml:"bind-info-lease" json:"bind-info-lease"`
+	TxnTotalSizeLimit     uint64  `toml:"txn-total-size-limit" json:"txn-total-size-limit"`
+	TCPKeepAlive          bool    `toml:"tcp-keep-alive" json:"tcp-keep-alive"`
+	CrossJoin             bool    `toml:"cross-join" json:"cross-join"`
+	RunAutoAnalyze        bool    `toml:"run-auto-analyze" json:"run-auto-analyze"`
+	DistinctAggPushDown   bool    `toml:"distinct-agg-push-down" json:"agg-push-down-join"`
+	CommitterConcurrency  int     `toml:"committer-concurrency" json:"committer-concurrency"`
+	MaxTxnTTL             uint64  `toml:"max-txn-ttl" json:"max-txn-ttl"`
 }
 
 // PlanCache is the PlanCache section of the config.
@@ -608,21 +610,23 @@ var defaultConf = Config{
 		RecordQPSbyDB:   false,
 	},
 	Performance: Performance{
-		MaxMemory:            0,
-		TCPKeepAlive:         true,
-		CrossJoin:            true,
-		StatsLease:           "3s",
-		RunAutoAnalyze:       true,
-		StmtCountLimit:       5000,
-		FeedbackProbability:  0.0,
-		QueryFeedbackLimit:   512,
-		PseudoEstimateRatio:  0.8,
-		ForcePriority:        "NO_PRIORITY",
-		BindInfoLease:        "3s",
-		TxnTotalSizeLimit:    DefTxnTotalSizeLimit,
-		DistinctAggPushDown:  false,
-		CommitterConcurrency: 16,
-		MaxTxnTTL:            10 * 60 * 1000, // 10min
+		MaxMemory:             0,
+		ServerMemoryQuota:     0,
+		MemoryUsageAlarmRatio: 0.8,
+		TCPKeepAlive:          true,
+		CrossJoin:             true,
+		StatsLease:            "3s",
+		RunAutoAnalyze:        true,
+		StmtCountLimit:        5000,
+		FeedbackProbability:   0.0,
+		QueryFeedbackLimit:    512,
+		PseudoEstimateRatio:   0.8,
+		ForcePriority:         "NO_PRIORITY",
+		BindInfoLease:         "3s",
+		TxnTotalSizeLimit:     DefTxnTotalSizeLimit,
+		DistinctAggPushDown:   false,
+		CommitterConcurrency:  16,
+		MaxTxnTTL:             10 * 60 * 1000, // 10min
 	},
 	ProxyProtocol: ProxyProtocol{
 		Networks:      "",
@@ -862,6 +866,10 @@ func (c *Config) Valid() error {
 		return fmt.Errorf("txn-total-size-limit should be less than %d", 10<<30)
 	}
 
+	if c.Performance.MemoryUsageAlarmRatio > 1 || c.Performance.MemoryUsageAlarmRatio < 0 {
+		return fmt.Errorf("memory-usage-alarm-ratio in [Performance] must be greater than or equal to 0 and less than or equal to 1")
+	}
+
 	if c.StmtSummary.MaxStmtCount <= 0 {
 		return fmt.Errorf("max-stmt-count in [stmt-summary] should be greater than 0")
 	}

config/config.toml.example

@@ -210,6 +210,17 @@ max-procs = 0
 # Max memory size to use, 0 use the total usable memory in the machine.
 max-memory = 0
 
+# Memory size quota for tidb server, 0 means unlimited
+server-memory-quota = 0
+
+# The alarm threshold when memory usage of the tidb-server exceeds. The valid value range is greater than or equal to 0
+# and less than or equal to 1. The default value is 0.8.
+# If this configuration is set to 0 or 1, it'll disable the alarm.
+# Otherwise, related information will be recorded in the directory `tmp-storage-path/record`.
+# Note: If the configuration `server-memory-quota` is set and larger than 0, the alarm threshold will be
+# `memory-usage-alarm-ratio * server-memory-quota`; otherwise, it'll be `memory-usage-alarm-ratio * system memory size`.
+memory-usage-alarm-ratio = 0.8
+
 # StmtCountLimit limits the max count of statement inside a transaction.
 stmt-count-limit = 5000
 

util/disk/tempDir.go

@@ -32,6 +32,11 @@ var (
 	sf          singleflight.Group
 )
 
+const (
+	lockFile  = "_dir.lock"
+	recordDir = "record"
+)
+
 // CheckAndInitTempDir check whether the temp directory is existed.
 // If not, initializes the temp directory.
 func CheckAndInitTempDir() (err error) {
@@ -64,7 +69,6 @@ func InitializeTempDir() error {
 			return err
 		}
 	}
-	lockFile := "_dir.lock"
 	tempDirLock, err = fslock.Lock(filepath.Join(tempDir, lockFile))
 	if err != nil {
 		switch err {
@@ -77,17 +81,27 @@ func InitializeTempDir() error {
 		return err
 	}
 
+	// Create dir for MemoryUsageAlarmRecord.
+	_, err = os.Stat(filepath.Join(tempDir, "record"))
+	if err != nil && !os.IsExist(err) {
+		err = os.MkdirAll(filepath.Join(tempDir, "record"), 0755)
+		if err != nil {
+			return err
+		}
+	}
+
 	subDirs, err := ioutil.ReadDir(tempDir)
 	if err != nil {
 		return err
 	}
 
 	// If it exists others files except lock file, creates another goroutine to clean them.
-	if len(subDirs) > 1 {
+	if len(subDirs) > 2 {
 		go func() {
 			for _, subDir := range subDirs {
 				// Do not remove the lock file.
-				if subDir.Name() == lockFile {
+				switch subDir.Name() {
+				case lockFile, recordDir:
 					continue
 				}
 				err := os.RemoveAll(filepath.Join(tempDir, subDir.Name()))

util/expensivequery/expensivequery.go

@@ -54,6 +54,7 @@ func (eqh *Handle) Run() {
 	ticker := time.NewTicker(tickInterval)
 	defer ticker.Stop()
 	sm := eqh.sm.Load().(util.SessionManager)
+	record := initMemoryUsageAlarmRecord()
 	for {
 		select {
 		case <-ticker.C:
@@ -73,6 +74,9 @@ func (eqh *Handle) Run() {
 				}
 			}
 			threshold = atomic.LoadUint64(&variable.ExpensiveQueryTimeThreshold)
+			if record.err == nil {
+				record.alarm4ExcessiveMemUsage(sm)
+			}
 		case <-eqh.exitCh:
 			return
 		}