Log slow queries if they pass a configurable threshold

Fixes #6429.

CHANGELOG.md

@@ -16,6 +16,7 @@
 - [#5502](https://github.com/influxdata/influxdb/issues/5502): Add checksum verification to TSM inspect tool
 - [#6444](https://github.com/influxdata/influxdb/pull/6444): Allow setting the config path through an environment variable and default config path.
 - [#3558](https://github.com/influxdata/influxdb/issues/3558): Support field math inside a WHERE clause.
+- [#6429](https://github.com/influxdata/influxdb/issues/6429): Log slow queries if they pass a configurable threshold.
 
 ### Bugfixes
 

cluster/config.go

@@ -43,6 +43,7 @@ type Config struct {
 	ShardMapperTimeout        toml.Duration `toml:"shard-mapper-timeout"`
 	MaxConcurrentQueries      int           `toml:"max-concurrent-queries"`
 	QueryTimeout              toml.Duration `toml:"query-timeout"`
+	LogQueriesAfter           toml.Duration `toml:"log-queries-after"`
 	MaxSelectPointN           int           `toml:"max-select-point"`
 	MaxSelectSeriesN          int           `toml:"max-select-series"`
 	MaxSelectBucketsN         int           `toml:"max-select-buckets"`

cmd/influxd/run/server.go

@@ -184,6 +184,7 @@ func NewServer(c *Config, buildInfo *BuildInfo) (*Server, error) {
 		MaxSelectBucketsN: c.Cluster.MaxSelectBucketsN,
 	}
 	s.QueryExecutor.QueryTimeout = time.Duration(c.Cluster.QueryTimeout)
+	s.QueryExecutor.LogQueriesAfter = time.Duration(c.Cluster.LogQueriesAfter)
 	s.QueryExecutor.MaxConcurrentQueries = c.Cluster.MaxConcurrentQueries
 	if c.Data.QueryLogEnabled {
 		s.QueryExecutor.Logger = log.New(os.Stderr, "[query] ", log.LstdFlags)

influxql/query_executor.go

@@ -105,6 +105,10 @@ type QueryExecutor struct {
 	// Query execution timeout.
 	QueryTimeout time.Duration
 
+	// Log queries if they are slower than this time.
+	// If zero, slow queries will never be logged.
+	LogQueriesAfter time.Duration
+
 	// Maximum number of concurrent queries.
 	MaxConcurrentQueries int
 
@@ -355,6 +359,20 @@ func (e *QueryExecutor) attachQuery(q *Query, database string, interrupt <-chan
 	e.queries[qid] = query
 
 	go e.waitForQuery(qid, query.closing, interrupt, query.monitorCh)
+	if e.LogQueriesAfter != 0 {
+		go query.monitor(func(closing <-chan struct{}) error {
+			t := time.NewTimer(e.LogQueriesAfter)
+			defer t.Stop()
+
+			select {
+			case <-t.C:
+				e.Logger.Printf("Detected slow query: %s (qid: %d, database: %s, threshold: %s)",
+					query.query, qid, query.database, e.LogQueriesAfter)
+			case <-closing:
+			}
+			return nil
+		})
+	}
 	e.nextID++
 	return qid, query, nil
 }