insights: surface suboptimal plans and failed executions

pkg/sql/crdb_internal.go

@@ -6432,7 +6432,7 @@ func populateExecutionInsights(
 		}
 
 		indexRecommendations := tree.NewDArray(types.String)
-		for _, recommendation := range insight.Statement.IndexRecommendation {
+		for _, recommendation := range insight.Statement.IndexRecommendations {
 			if err := indexRecommendations.Append(tree.NewDString(recommendation)); err != nil {
 				return err
 			}
@@ -6446,7 +6446,7 @@ func populateExecutionInsights(
 			tree.NewDBytes(tree.DBytes(sqlstatsutil.EncodeUint64ToBytes(uint64(insight.Statement.FingerprintID)))),
 			problems,
 			tree.NewDString(insight.Statement.Query),
-			tree.NewDString(insight.Statement.Status),
+			tree.NewDString(insight.Statement.Status.String()),
 			startTimestamp,
 			endTimestamp,
 			tree.MakeDBool(tree.DBool(insight.Statement.FullScan)),

pkg/sql/sqlstats/insights/insights.proto

@@ -60,14 +60,19 @@ message Transaction {
 }
 
 message Statement {
+  enum Status {
+    Completed = 0;
+    Failed = 1;
+  }
+
   bytes id = 1 [(gogoproto.customname) = "ID",
     (gogoproto.customtype) = "github.com/cockroachdb/cockroach/pkg/sql/clusterunique.ID",
     (gogoproto.nullable) = false];
   uint64 fingerprint_id = 2 [(gogoproto.customname) = "FingerprintID",
     (gogoproto.casttype) = "github.com/cockroachdb/cockroach/pkg/roachpb.StmtFingerprintID"];
   double latency_in_seconds = 3;
   string query = 4;
-  string status = 5;
+  Status status = 5;
   google.protobuf.Timestamp start_time = 6 [(gogoproto.nullable) = false, (gogoproto.stdtime) = true];
   google.protobuf.Timestamp end_time = 7 [(gogoproto.nullable) = false, (gogoproto.stdtime) = true];
   bool full_scan = 8;
@@ -82,7 +87,7 @@ message Statement {
   // Nodes is the ordered list of nodes ids on which the statement was executed.
   repeated int64 nodes = 17;
   google.protobuf.Duration contention = 18 [(gogoproto.stdduration) = true];
-  repeated string index_recommendation = 19;
+  repeated string index_recommendations = 19;
 
 }
 

pkg/sql/sqlstats/insights/integration/insights_test.go

@@ -110,8 +110,8 @@ func TestInsightsIntegration(t *testing.T) {
 			return err
 		}
 
-		if status != "completed" {
-			return fmt.Errorf("expected 'completed', but was %s", status)
+		if status != "Completed" {
+			return fmt.Errorf("expected 'Completed', but was %s", status)
 		}
 
 		delayFromTable := endInsights.Sub(startInsights).Seconds()

pkg/sql/sqlstats/insights/problems.go

@@ -17,10 +17,18 @@ type problems struct {
 }
 
 func (p *problems) examine(stmt *Statement) (result []Problem) {
+	if len(stmt.IndexRecommendations) > 0 {
+		result = append(result, Problem_SuboptimalPlan)
+	}
+
 	if stmt.Retries >= HighRetryCountThreshold.Get(&p.st.SV) {
 		result = append(result, Problem_HighRetryCount)
 	}
 
+	if stmt.Status == Statement_Failed {
+		result = append(result, Problem_FailedExecution)
+	}
+
 	if len(result) == 0 {
 		result = append(result, Problem_Unknown)
 	}

pkg/sql/sqlstats/insights/problems_test.go

@@ -34,11 +34,21 @@ func TestProblems(t *testing.T) {
 			statement: &Statement{},
 			problems:  []Problem{Problem_Unknown},
 		},
+		{
+			name:      "suboptimal plan",
+			statement: &Statement{IndexRecommendations: []string{"THIS IS AN INDEX RECOMMENDATION"}},
+			problems:  []Problem{Problem_SuboptimalPlan},
+		},
 		{
 			name:      "high retry count",
 			statement: &Statement{Retries: 10},
 			problems:  []Problem{Problem_HighRetryCount},
 		},
+		{
+			name:      "failed execution",
+			statement: &Statement{Status: Statement_Failed},
+			problems:  []Problem{Problem_FailedExecution},
+		},
 	}
 	for _, tc := range testCases {
 		t.Run(tc.name, func(t *testing.T) {

pkg/sql/sqlstats/ssmemstorage/ss_mem_writer.go

@@ -42,12 +42,12 @@ var timestampSize = int64(unsafe.Sizeof(time.Time{}))
 
 var _ sqlstats.Writer = &Container{}
 
-func getStatusString(statementError error) string {
+func getStatus(statementError error) insights.Statement_Status {
 	if statementError == nil {
-		return "completed"
+		return insights.Statement_Completed
 	}
 
-	return "failed"
+	return insights.Statement_Failed
 }
 
 // RecordStatement implements sqlstats.Writer interface.
@@ -182,25 +182,25 @@ func (s *Container) RecordStatement(
 	}
 
 	s.outliersRegistry.ObserveStatement(value.SessionID, &insights.Statement{
-		ID:                  value.StatementID,
-		FingerprintID:       stmtFingerprintID,
-		LatencyInSeconds:    value.ServiceLatency,
-		Query:               value.Query,
-		Status:              getStatusString(value.StatementError),
-		StartTime:           value.StartTime,
-		EndTime:             value.EndTime,
-		FullScan:            value.FullScan,
-		User:                value.SessionData.User().Normalized(),
-		ApplicationName:     value.SessionData.ApplicationName,
-		Database:            value.SessionData.Database,
-		PlanGist:            value.PlanGist,
-		Retries:             int64(value.AutoRetryCount),
-		AutoRetryReason:     autoRetryReason,
-		RowsRead:            value.RowsRead,
-		RowsWritten:         value.RowsWritten,
-		Nodes:               value.Nodes,
-		Contention:          contention,
-		IndexRecommendation: value.IndexRecommendations,
+		ID:                   value.StatementID,
+		FingerprintID:        stmtFingerprintID,
+		LatencyInSeconds:     value.ServiceLatency,
+		Query:                value.Query,
+		Status:               getStatus(value.StatementError),
+		StartTime:            value.StartTime,
+		EndTime:              value.EndTime,
+		FullScan:             value.FullScan,
+		User:                 value.SessionData.User().Normalized(),
+		ApplicationName:      value.SessionData.ApplicationName,
+		Database:             value.SessionData.Database,
+		PlanGist:             value.PlanGist,
+		Retries:              int64(value.AutoRetryCount),
+		AutoRetryReason:      autoRetryReason,
+		RowsRead:             value.RowsRead,
+		RowsWritten:          value.RowsWritten,
+		Nodes:                value.Nodes,
+		Contention:           contention,
+		IndexRecommendations: value.IndexRecommendations,
 	})
 
 	return stats.ID, nil