backup: add checkpointing to backup compactions

This patch adds periodic checkpointing to backup compaction jobs.

Epic: None

Release note: None

Author: kev-cao

Diff for: pkg/backup/compaction_job.go

@@ -40,6 +40,7 @@ import (
 	"github.com/cockroachdb/cockroach/pkg/util/hlc"
 	"github.com/cockroachdb/cockroach/pkg/util/log"
 	"github.com/cockroachdb/cockroach/pkg/util/retry"
+	"github.com/cockroachdb/cockroach/pkg/util/timeutil"
 	"github.com/cockroachdb/cockroach/pkg/util/uuid"
 	"github.com/cockroachdb/errors"
 	"github.com/gogo/protobuf/types"
@@ -231,8 +232,8 @@ func (b *backupResumer) ResumeCompaction(
 
 	var backupManifest *backuppb.BackupManifest
 	updatedDetails := initialDetails
+	testingKnobs := execCtx.ExecCfg().BackupRestoreTestingKnobs
 	if initialDetails.URI == "" {
-		testingKnobs := execCtx.ExecCfg().BackupRestoreTestingKnobs
 		if testingKnobs != nil && testingKnobs.RunBeforeResolvingCompactionDest != nil {
 			if err := testingKnobs.RunBeforeResolvingCompactionDest(); err != nil {
 				return err
@@ -263,7 +264,7 @@ func (b *backupResumer) ResumeCompaction(
 			return err
 		}
 
-		if err := execCtx.ExecCfg().JobRegistry.CheckPausepoint("backup.before.write_first_checkpoint"); err != nil {
+		if err := execCtx.ExecCfg().JobRegistry.CheckPausepoint("backup_compaction.before.write_first_checkpoint"); err != nil {
 			return err
 		}
 
@@ -274,7 +275,7 @@ func (b *backupResumer) ResumeCompaction(
 			return err
 		}
 
-		if err := execCtx.ExecCfg().JobRegistry.CheckPausepoint("backup.after.write_first_checkpoint"); err != nil {
+		if err := execCtx.ExecCfg().JobRegistry.CheckPausepoint("backup_compaction.after.write_first_checkpoint"); err != nil {
 			return err
 		}
 
@@ -297,7 +298,7 @@ func (b *backupResumer) ResumeCompaction(
 			return err
 		}
 
-		if err := execCtx.ExecCfg().JobRegistry.CheckPausepoint("backup.after.details_has_checkpoint"); err != nil {
+		if err := execCtx.ExecCfg().JobRegistry.CheckPausepoint("backup_compaction.after.details_has_checkpoint"); err != nil {
 			return err
 		}
 		// TODO (kev-cao): Add telemetry for backup compactions.
@@ -347,6 +348,10 @@ func (b *backupResumer) ResumeCompaction(
 		}
 	}
 
+	if testingKnobs != nil && testingKnobs.AfterLoadingCompactionManifestOnResume != nil {
+		testingKnobs.AfterLoadingCompactionManifestOnResume(backupManifest)
+	}
+
 	// We retry on pretty generic failures -- any rpc error. If a worker node were
 	// to restart, it would produce this kind of error, but there may be other
 	// errors that are also rpc errors. Don't retry too aggressively.
@@ -355,12 +360,11 @@ func (b *backupResumer) ResumeCompaction(
 		MaxRetries: 5,
 	}
 
-	if execCtx.ExecCfg().BackupRestoreTestingKnobs != nil &&
-		execCtx.ExecCfg().BackupRestoreTestingKnobs.BackupDistSQLRetryPolicy != nil {
-		retryOpts = *execCtx.ExecCfg().BackupRestoreTestingKnobs.BackupDistSQLRetryPolicy
+	if testingKnobs != nil && testingKnobs.BackupDistSQLRetryPolicy != nil {
+		retryOpts = *testingKnobs.BackupDistSQLRetryPolicy
 	}
 
-	if err := execCtx.ExecCfg().JobRegistry.CheckPausepoint("backup.before.flow"); err != nil {
+	if err := execCtx.ExecCfg().JobRegistry.CheckPausepoint("backup_compaction.before.flow"); err != nil {
 		return err
 	}
 
@@ -389,8 +393,6 @@ func (b *backupResumer) ResumeCompaction(
 
 		// Reload the backup manifest to pick up any spans we may have completed on
 		// previous attempts.
-		// TODO (kev-cao): Compactions currently do not create checkpoints, but this
-		// can be used to reload the manifest once we add checkpointing.
 		var reloadBackupErr error
 		mem.Shrink(ctx, memSize)
 		backupManifest, memSize, reloadBackupErr = b.readManifestOnResume(ctx, &mem, execCtx.ExecCfg(),
@@ -765,9 +767,13 @@ func concludeBackupCompaction(
 // the associated manifest.
 func processProgress(
 	ctx context.Context,
+	execCtx sql.JobExecContext,
+	details jobspb.BackupDetails,
 	manifest *backuppb.BackupManifest,
 	progCh <-chan *execinfrapb.RemoteProducerMetadata_BulkProcessorProgress,
+	kmsEnv cloud.KMSEnv,
 ) error {
+	var lastCheckpointTime time.Time
 	// When a processor is done exporting a span, it will send a progress update
 	// to progCh.
 	for progress := range progCh {
@@ -780,11 +786,24 @@ func processProgress(
 			manifest.Files = append(manifest.Files, file)
 			manifest.EntryCounts.Add(file.EntryCounts)
 		}
+
 		// TODO (kev-cao): Add per node progress updates.
+
+		if wroteCheckpoint, err := maybeWriteBackupCheckpoint(
+			ctx, execCtx, details, manifest, lastCheckpointTime, kmsEnv,
+		); err != nil {
+			log.Errorf(ctx, "unable to checkpoint compaction: %+v", err)
+		} else if wroteCheckpoint {
+			lastCheckpointTime = timeutil.Now()
+			if err := execCtx.ExecCfg().JobRegistry.CheckPausepoint("backup_compaction.after.write_checkpoint"); err != nil {
+				return err
+			}
+		}
 	}
 	return nil
 }
 
+// compactionJobDescription generates a redacted description of the job.
 func compactionJobDescription(details jobspb.BackupDetails) (string, error) {
 	fmtCtx := tree.NewFmtCtx(tree.FmtSimple)
 	redactedURIs, err := sanitizeURIList(details.Destination.To)
@@ -835,8 +854,7 @@ func doCompaction(
 		)
 	}
 	checkpointLoop := func(ctx context.Context) error {
-		// TODO (kev-cao): Add logic for checkpointing during loop.
-		return processProgress(ctx, manifest, progCh)
+		return processProgress(ctx, execCtx, details, manifest, progCh, kmsEnv)
 	}
 	// TODO (kev-cao): Add trace aggregator loop.
 
@@ -851,6 +869,34 @@ func doCompaction(
 	)
 }
 
+// maybeWriteBackupCheckpoint writes a checkpoint for the backup if
+// the time since the last checkpoint exceeds the configured interval. If a
+// checkpoint is written, the function returns true.
+func maybeWriteBackupCheckpoint(
+	ctx context.Context,
+	execCtx sql.JobExecContext,
+	details jobspb.BackupDetails,
+	manifest *backuppb.BackupManifest,
+	lastCheckpointTime time.Time,
+	kmsEnv cloud.KMSEnv,
+) (bool, error) {
+	if details.URI == "" {
+		return false, errors.New("backup details does not contain a default URI")
+	}
+	execCfg := execCtx.ExecCfg()
+	interval := BackupCheckpointInterval.Get(&execCfg.Settings.SV)
+	if timeutil.Since(lastCheckpointTime) < interval {
+		return false, nil
+	}
+	if err := backupinfo.WriteBackupManifestCheckpoint(
+		ctx, details.URI, details.EncryptionOptions, kmsEnv,
+		manifest, execCfg, execCtx.User(),
+	); err != nil {
+		return false, err
+	}
+	return true, nil
+}
+
 func init() {
 	builtins.StartCompactionJob = StartCompactionJob
 }

Diff for: pkg/backup/compaction_test.go

@@ -12,10 +12,12 @@ import (
 	"net/url"
 	"strconv"
 	"strings"
+	"sync"
 	"testing"
 	"time"
 
 	"github.com/cockroachdb/cockroach/pkg/backup/backupinfo"
+	"github.com/cockroachdb/cockroach/pkg/backup/backuppb"
 	"github.com/cockroachdb/cockroach/pkg/base"
 	"github.com/cockroachdb/cockroach/pkg/jobs"
 	"github.com/cockroachdb/cockroach/pkg/jobs/jobspb"
@@ -310,7 +312,7 @@ crdb_internal.json_to_pb(
 		db.Exec(t, "INSERT INTO foo VALUES (3, 3)")
 		end := getTime(t)
 		db.Exec(t, fmt.Sprintf(incBackupAostCmd, 7, end))
-		db.Exec(t, "SET CLUSTER SETTING jobs.debug.pausepoints = 'backup.after.details_has_checkpoint'")
+		db.Exec(t, "SET CLUSTER SETTING jobs.debug.pausepoints = 'backup_compaction.after.details_has_checkpoint'")
 
 		var backupPath string
 		db.QueryRow(t, "SHOW BACKUPS IN 'nodelocal://1/backup/7'").Scan(&backupPath)
@@ -324,7 +326,7 @@ crdb_internal.json_to_pb(
 		db.Exec(t, "INSERT INTO foo VALUES (4, 4)")
 		end = getTime(t)
 		db.Exec(t, fmt.Sprintf(incBackupAostCmd, 7, end))
-		db.Exec(t, "SET CLUSTER SETTING jobs.debug.pausepoints = 'backup.after.details_has_checkpoint'")
+		db.Exec(t, "SET CLUSTER SETTING jobs.debug.pausepoints = 'backup_compaction.after.details_has_checkpoint'")
 		jobID = startCompaction(7, backupPath, start, end)
 		jobutils.WaitForJobToPause(t, db, jobID)
 		db.Exec(t, "CANCEL JOB $1", jobID)
@@ -675,6 +677,89 @@ func TestBackupCompactionUnsupportedOptions(t *testing.T) {
 	}
 }
 
+func TestCompactionCheckpointing(t *testing.T) {
+	defer leaktest.AfterTest(t)()
+	defer log.Scope(t).Close(t)
+
+	// Need a large enough backup so that it doesn't finish in one iteration
+	const numAccounts = 10000
+	var mu sync.Mutex
+	manifestNumFiles := -1
+	tc, db, _, cleanup := backupRestoreTestSetupWithParams(
+		t, singleNode, numAccounts, InitManualReplication, base.TestClusterArgs{
+			ServerArgs: base.TestServerArgs{
+				Knobs: base.TestingKnobs{
+					BackupRestore: &sql.BackupRestoreTestingKnobs{
+						AfterLoadingCompactionManifestOnResume: func(m *backuppb.BackupManifest) {
+							mu.Lock()
+							defer mu.Unlock()
+							manifestNumFiles = len(m.Files)
+						},
+					},
+				},
+			},
+		},
+	)
+	defer cleanup()
+	writeQueries := func() {
+		db.Exec(t, "UPDATE data.bank SET balance = balance + 1")
+	}
+	db.Exec(t, "SET CLUSTER SETTING bulkio.backup.checkpoint_interval = '10ms'")
+	start := getTime(t)
+	db.Exec(t, fmt.Sprintf("BACKUP INTO 'nodelocal://1/backup' AS OF SYSTEM TIME %d", start))
+	writeQueries()
+	db.Exec(t, "BACKUP INTO LATEST IN 'nodelocal://1/backup'")
+	writeQueries()
+	end := getTime(t)
+	db.Exec(t, fmt.Sprintf("BACKUP INTO LATEST IN 'nodelocal://1/backup' AS OF SYSTEM TIME %d", end))
+
+	var backupPath string
+	db.QueryRow(t, "SHOW BACKUPS IN 'nodelocal://1/backup'").Scan(&backupPath)
+
+	db.Exec(t, "SET CLUSTER SETTING jobs.debug.pausepoints = 'backup_compaction.after.write_checkpoint'")
+	var jobID jobspb.JobID
+	db.QueryRow(
+		t,
+		"SELECT crdb_internal.backup_compaction(ARRAY['nodelocal://1/backup'], $1, ''::BYTES, $2, $3)",
+		backupPath, start, end,
+	).Scan(&jobID)
+	// Ensure that the very first manifest when the job is initially started has
+	// no files.
+	testutils.SucceedsSoon(t, func() error {
+		mu.Lock()
+		defer mu.Unlock()
+		if manifestNumFiles < 0 {
+			return fmt.Errorf("waiting for manifest to be loaded")
+		}
+		return nil
+	})
+	require.Equal(t, 0, manifestNumFiles, "expected no files in manifest")
+	manifestNumFiles = -1
+
+	// Wait for the job to hit the pausepoint.
+	jobutils.WaitForJobToPause(t, db, jobID)
+	// Don't bother pausing on other checkpoints.
+	db.Exec(t, "SET CLUSTER SETTING jobs.debug.pausepoints = ''")
+	db.Exec(t, "RESUME JOB $1", jobID)
+	jobutils.WaitForJobToRun(t, db, jobID)
+
+	// Now that the job has been paused and resumed after previously hitting a
+	// checkpoint, the initial manifest at the start of the job should have
+	// some files from before the pause.
+	testutils.SucceedsSoon(t, func() error {
+		mu.Lock()
+		defer mu.Unlock()
+		if manifestNumFiles < 0 {
+			return fmt.Errorf("waiting for manifest to be loaded")
+		}
+		return nil
+	})
+	require.Greater(t, manifestNumFiles, 0, "expected non-zero number of files in manifest")
+
+	waitForSuccessfulJob(t, tc, jobID)
+	validateCompactedBackupForTables(t, db, []string{"bank"}, "'nodelocal://1/backup'", start, end)
+}
+
 // Start and end are unix epoch in nanoseconds.
 func validateCompactedBackupForTables(
 	t *testing.T, db *sqlutils.SQLRunner, tables []string, collectionURIs string, start, end int64,

Diff for: pkg/sql/BUILD.bazel

@@ -301,6 +301,7 @@ go_library(
     importpath = "github.com/cockroachdb/cockroach/pkg/sql",
     visibility = ["//visibility:public"],
     deps = [
+        "//pkg/backup/backuppb",
         "//pkg/base",
         "//pkg/build",
         "//pkg/cloud",

Diff for: pkg/sql/exec_util.go

@@ -21,6 +21,7 @@ import (
 	"time"
 
 	apd "github.com/cockroachdb/apd/v3"
+	"github.com/cockroachdb/cockroach/pkg/backup/backuppb"
 	"github.com/cockroachdb/cockroach/pkg/base"
 	"github.com/cockroachdb/cockroach/pkg/cloud/externalconn"
 	"github.com/cockroachdb/cockroach/pkg/clusterversion"
@@ -1873,6 +1874,10 @@ type BackupRestoreTestingKnobs struct {
 	// is written.
 	AfterBackupCheckpoint func()
 
+	// AfterLoadingCompactionManifestOnResume is run once the backup manifest has been
+	// loaded/created on the resumption of a compaction job.
+	AfterLoadingCompactionManifestOnResume func(manifest *backuppb.BackupManifest)
+
 	// CaptureResolvedTableDescSpans allows for intercepting the spans which are
 	// resolved during backup planning, and will eventually be backed up during
 	// execution.