crosscluster/physical: always destroy reader tenant on cutover

Previously, a reader tenant would only get automatically destroyed if a user
specified an explicit cutover time. This UX is confusing. Now, the reader
tenant will always get destroyed on cutover.

Fixes #149684

Release note (ops change): the PCR reader tenant is always destroyed on cutover

Author: msbutler

pkg/crosscluster/physical/standby_read_ts_poller_job_test.go

@@ -10,7 +10,6 @@ import (
 	"fmt"
 	"net/url"
 	"testing"
-	"time"
 
 	apd "github.com/cockroachdb/apd/v3"
 	"github.com/cockroachdb/cockroach/pkg/crosscluster/replicationtestutils"
@@ -217,16 +216,9 @@ INSERT INTO a VALUES (1);
 `)
 
 		waitForPollerJobToStartDest(t, c, ingestionJobID)
-		if cutoverToLatest {
-			observeValueInReaderTenant(t, c.ReaderTenantSQL)
-			c.Cutover(ctx, producerJobID, ingestionJobID, time.Time{}, false)
-			jobutils.WaitForJobToSucceed(t, c.DestSysSQL, jobspb.JobID(ingestionJobID))
-			observeValueInReaderTenant(t, c.ReaderTenantSQL)
-		} else {
-			c.Cutover(ctx, producerJobID, ingestionJobID, c.SrcCluster.Server(0).Clock().Now().GoTime(), false)
-			waitToRemoveTenant(t, c.DestSysSQL, readerTenantName)
-			jobutils.WaitForJobToSucceed(t, c.DestSysSQL, jobspb.JobID(ingestionJobID))
-		}
+		c.Cutover(ctx, producerJobID, ingestionJobID, c.SrcCluster.Server(0).Clock().Now().GoTime(), false)
+		waitToRemoveTenant(t, c.DestSysSQL, readerTenantName)
+		jobutils.WaitForJobToSucceed(t, c.DestSysSQL, jobspb.JobID(ingestionJobID))
 	})
 }
 

pkg/crosscluster/physical/stream_ingestion_job.go

@@ -456,12 +456,11 @@ func maybeRevertToCutoverTimestamp(
 	// existed in the record at the point of the update rather the
 	// value that may be in the job record before the update.
 	var (
-		shouldRevertToCutover   bool
-		cutoverTimestamp        hlc.Timestamp
-		originalSpanToRevert    roachpb.Span
-		remainingSpansToRevert  roachpb.Spans
-		replicatedTimeAtCutover hlc.Timestamp
-		readerTenantID          roachpb.TenantID
+		shouldRevertToCutover  bool
+		cutoverTimestamp       hlc.Timestamp
+		originalSpanToRevert   roachpb.Span
+		remainingSpansToRevert roachpb.Spans
+		readerTenantID         roachpb.TenantID
 	)
 	if err := ingestionJob.NoTxn().Update(ctx,
 		func(txn isql.Txn, md jobs.JobMetadata, ju *jobs.JobUpdater) error {
@@ -478,7 +477,6 @@ func maybeRevertToCutoverTimestamp(
 			}
 
 			cutoverTimestamp = streamIngestionProgress.CutoverTime
-			replicatedTimeAtCutover = streamIngestionProgress.ReplicatedTimeAtCutover
 			readerTenantID = streamIngestionDetails.ReadTenantID
 			originalSpanToRevert = streamIngestionDetails.Span
 			remainingSpansToRevert = streamIngestionProgress.RemainingCutoverSpans
@@ -500,9 +498,7 @@ func maybeRevertToCutoverTimestamp(
 	if !shouldRevertToCutover {
 		return cutoverTimestamp, false, nil
 	}
-	// Identical cutoverTimestamp and replicatedTimeAtCutover implies that
-	// CUTOVER TO LATEST command was run. Destroy reader tenant if not CUTOVER TO LATEST.
-	if !cutoverTimestamp.Equal(replicatedTimeAtCutover) && readerTenantID.IsSet() {
+	if readerTenantID.IsSet() {
 		if err := stopTenant(ctx, p.ExecCfg(), readerTenantID); err != nil {
 			return cutoverTimestamp, false, errors.Wrapf(err, "failed to stop reader tenant")
 		}