Store only last 2 postgres timeline histories

Co-authored-by: Arunvel Sriram <arunvelsriram@gmail.com>

cmd/keeper/cmd/keeper.go

@@ -57,6 +57,8 @@ var CmdKeeper = &cobra.Command{
 	Version: cmd.Version,
 }
 
+const maxPostgresTimelinesHistory = 2
+
 type KeeperLocalState struct {
 	UID        string
 	ClusterUID string
@@ -676,27 +678,12 @@ func (p *PostgresKeeper) GetPGState(pctx context.Context) (*cluster.PostgresStat
 		pgState.TimelineID = sd.TimelineID
 		pgState.XLogPos = sd.XLogPos
 
-		// if timeline <= 1 then no timeline history file exists.
-		pgState.TimelinesHistory = cluster.PostgresTimelinesHistory{}
-		if pgState.TimelineID > 1 {
-			var tlsh []*postgresql.TimelineHistory
-			tlsh, err = p.pgm.GetTimelinesHistory(pgState.TimelineID)
-			if err != nil {
-				log.Errorw("error getting timeline history", zap.Error(err))
-				return pgState, nil
-			}
-			ctlsh := cluster.PostgresTimelinesHistory{}
-
-			for _, tlh := range tlsh {
-				ctlh := &cluster.PostgresTimelineHistory{
-					TimelineID:  tlh.TimelineID,
-					SwitchPoint: tlh.SwitchPoint,
-					Reason:      tlh.Reason,
-				}
-				ctlsh = append(ctlsh, ctlh)
-			}
-			pgState.TimelinesHistory = ctlsh
+		ctlsh, err := getTimeLinesHistory(pgState, p.pgm, maxPostgresTimelinesHistory)
+		if err != nil {
+			log.Errorw("error getting timeline history", zap.Error(err))
+			return pgState, nil
 		}
+		pgState.TimelinesHistory = ctlsh
 
 		ow, err := p.pgm.OlderWalFile()
 		if err != nil {
@@ -711,6 +698,31 @@ func (p *PostgresKeeper) GetPGState(pctx context.Context) (*cluster.PostgresStat
 	return pgState, nil
 }
 
+func getTimeLinesHistory(pgState *cluster.PostgresState, pgm postgresql.PGManager, maxPostgresTimelinesHistory int) (cluster.PostgresTimelinesHistory, error) {
+	ctlsh := cluster.PostgresTimelinesHistory{}
+	// if timeline <= 1 then no timeline history file exists.
+	if pgState.TimelineID > 1 {
+		var tlsh []*postgresql.TimelineHistory
+		tlsh, err := pgm.GetTimelinesHistory(pgState.TimelineID)
+		if err != nil {
+			log.Errorw("error getting timeline history", zap.Error(err))
+			return ctlsh, err
+		}
+		if len(tlsh) > maxPostgresTimelinesHistory {
+			tlsh = tlsh[len(tlsh)-maxPostgresTimelinesHistory:]
+		}
+		for _, tlh := range tlsh {
+			ctlh := &cluster.PostgresTimelineHistory{
+				TimelineID:  tlh.TimelineID,
+				SwitchPoint: tlh.SwitchPoint,
+				Reason:      tlh.Reason,
+			}
+			ctlsh = append(ctlsh, ctlh)
+		}
+	}
+	return ctlsh, nil
+}
+
 func (p *PostgresKeeper) getLastPGState() *cluster.PostgresState {
 	p.pgStateMutex.Lock()
 	pgState := p.lastPGState.DeepCopy()

cmd/keeper/cmd/keeper_test.go

@@ -18,6 +18,9 @@ import (
 	"bytes"
 	"errors"
 	"fmt"
+	"github.com/golang/mock/gomock"
+	"github.com/sorintlab/stolon/internal/mock/postgresql"
+	"github.com/sorintlab/stolon/internal/postgresql"
 	"reflect"
 	"testing"
 
@@ -248,3 +251,158 @@ func TestGenerateHBA(t *testing.T) {
 		}
 	}
 }
+
+func TestGetTimeLinesHistory(t *testing.T) {
+	t.Run("should return empty if timelineID is not greater than 1", func(t *testing.T) {
+		pgState := &cluster.PostgresState{
+			TimelineID: 1,
+		}
+
+		ctlsh, err := getTimeLinesHistory(pgState, nil, 3)
+
+		if err != nil {
+			t.Errorf("err should be nil, but got %v", err)
+		}
+		if len(ctlsh) != 0 {
+			t.Errorf("expecting empty ctlsh, but got %v", ctlsh)
+		}
+	})
+
+	t.Run("should return error if there is error in getting timeline history", func(t *testing.T) {
+		var timelineID uint64 = 2
+		pgState := &cluster.PostgresState{
+			TimelineID: timelineID,
+		}
+
+		ctrl := gomock.NewController(t)
+		defer ctrl.Finish()
+
+		pgm := mocks.NewMockPGManager(ctrl)
+		pgm.EXPECT().GetTimelinesHistory(timelineID).Return([]*postgresql.TimelineHistory{}, fmt.Errorf("failed to get timeline history"))
+		ctlsh, err := getTimeLinesHistory(pgState, pgm, 3)
+
+		if err == nil {
+			t.Errorf("err should be not be nil")
+		}
+		if err != nil && err.Error() != "failed to get timeline history" {
+			t.Errorf("err should be failed to get timeline history, but got %v", err)
+		}
+		if len(ctlsh) != 0 {
+			t.Errorf("expecting empty ctlsh, but got %v", ctlsh)
+		}
+	})
+
+	t.Run("should return timeline history as is if the given length is less than maxPostgresTimelinesHistory", func(t *testing.T) {
+		var timelineID uint64 = 2
+		pgState := &cluster.PostgresState{
+			TimelineID: timelineID,
+		}
+
+		ctrl := gomock.NewController(t)
+		defer ctrl.Finish()
+
+		pgm := mocks.NewMockPGManager(ctrl)
+		timelineHistories := []*postgresql.TimelineHistory{
+			{
+				TimelineID:  1,
+				SwitchPoint: 1,
+				Reason:      "reason1",
+			},
+			{
+				TimelineID:  2,
+				SwitchPoint: 2,
+				Reason:      "reason2",
+			},
+		}
+		pgm.EXPECT().GetTimelinesHistory(timelineID).Return(timelineHistories, nil)
+		ctlsh, err := getTimeLinesHistory(pgState, pgm, 3)
+
+		if err != nil {
+			t.Errorf("err should be not be nil")
+		}
+		if len(ctlsh) != 2 {
+			t.Errorf("expecting length of ctlsh to be 2, but got %d", len(ctlsh))
+		}
+		expectedTimelineHistories := cluster.PostgresTimelinesHistory{
+			&cluster.PostgresTimelineHistory{
+				TimelineID:  1,
+				SwitchPoint: 1,
+				Reason:      "reason1",
+			},
+			&cluster.PostgresTimelineHistory{
+				TimelineID:  2,
+				SwitchPoint: 2,
+				Reason:      "reason2",
+			},
+		}
+		fmt.Println(ctlsh, expectedTimelineHistories)
+		if *ctlsh[0] != *expectedTimelineHistories[0] {
+			t.Errorf("expected %v, but got %v ", *expectedTimelineHistories[0], *ctlsh[0])
+		}
+		if *ctlsh[1] != *expectedTimelineHistories[1] {
+			t.Errorf("expected %v, but got %v ", *expectedTimelineHistories[1], *ctlsh[1])
+		}
+	})
+
+	t.Run("should return timeline history with last maxPostgresTimelinesHistory elements if timeline history length is greater than maxPostgresTimelinesHistory", func(t *testing.T) {
+		var timelineID uint64 = 4
+		pgState := &cluster.PostgresState{
+			TimelineID: timelineID,
+		}
+
+		ctrl := gomock.NewController(t)
+		defer ctrl.Finish()
+
+		pgm := mocks.NewMockPGManager(ctrl)
+		timelineHistories := []*postgresql.TimelineHistory{
+			{
+				TimelineID:  1,
+				SwitchPoint: 1,
+				Reason:      "reason1",
+			},
+			{
+				TimelineID:  2,
+				SwitchPoint: 2,
+				Reason:      "reason2",
+			},
+			{
+				TimelineID:  3,
+				SwitchPoint: 3,
+				Reason:      "reason3",
+			},
+			{
+				TimelineID:  4,
+				SwitchPoint: 4,
+				Reason:      "reason4",
+			},
+		}
+		pgm.EXPECT().GetTimelinesHistory(timelineID).Return(timelineHistories, nil)
+		ctlsh, err := getTimeLinesHistory(pgState, pgm, 2)
+
+		if err != nil {
+			t.Errorf("err should be not be nil")
+		}
+		if len(ctlsh) != 2 {
+			t.Errorf("expecting length of ctlsh to be 2, but got %d", len(ctlsh))
+		}
+		expectedTimelineHistories := cluster.PostgresTimelinesHistory{
+			&cluster.PostgresTimelineHistory{
+				TimelineID:  3,
+				SwitchPoint: 3,
+				Reason:      "reason3",
+			},
+			&cluster.PostgresTimelineHistory{
+				TimelineID:  4,
+				SwitchPoint: 4,
+				Reason:      "reason4",
+			},
+		}
+		fmt.Println(ctlsh, expectedTimelineHistories)
+		if *ctlsh[0] != *expectedTimelineHistories[0] {
+			t.Errorf("expected %v, but got %v ", *expectedTimelineHistories[0], *ctlsh[0])
+		}
+		if *ctlsh[1] != *expectedTimelineHistories[1] {
+			t.Errorf("expected %v, but got %v ", *expectedTimelineHistories[1], *ctlsh[1])
+		}
+	})
+}

internal/postgresql/postgresql.go

@@ -38,6 +38,8 @@ import (
 	"go.uber.org/zap"
 )
 
+//go:generate mockgen -destination=../mock/postgresql/postgresql.go -package=mocks -source=$GOFILE
+
 const (
 	postgresConf         = "postgresql.conf"
 	postgresRecoveryConf = "recovery.conf"
@@ -53,6 +55,10 @@ var (
 
 var log = slog.S()
 
+type PGManager interface {
+	GetTimelinesHistory(timeline uint64) ([]*TimelineHistory, error)
+}
+
 type Manager struct {
 	pgBinPath             string
 	dataDir               string