Merge #156149 #156386

156149: restore: fix retry revert stuck in reverting r=jeffswenson a=kev-cao

Previously, if a reverting restore job was paused after dropping its descriptors and then resumed, the restore job would get stuck in a retry loop due to missing descriptors.

This commit adds a field in the restore details that can be set once descriptors have been dropped during restore cleanup. This allows us to avoid attempting to drop descriptors if they have already been dropped.

Fixes: #156019

Release note: Restore no longer gets stuck in a retry loop when reverts are attempted twice.

156386: WORKSPACE: bump go 1.25 to pick up goroutine labels r=dt a=dt



Co-authored-by: Kevin Cao <39608887+kev-cao@users.noreply.github.com>
Co-authored-by: David Taylor <davidt@davidt.io>

Author: 3 people

WORKSPACE

@@ -185,14 +185,14 @@ load(
 go_download_sdk(
     name = "go_sdk",
     sdks = {
-        "darwin_amd64": ("go1.25.3.darwin-amd64.tar.gz", "b229aaaf13ff3bb5f918a31b5e7da27673e173b1578b12a0157be3ca6512bb22"),
-        "darwin_arm64": ("go1.25.3.darwin-arm64.tar.gz", "1601fb97e9f325f9c2addc7193ca79750e20bd8d3e18e5a0c47e406ec0dfc63f"),
-        "linux_amd64": ("go1.25.3.linux-amd64.tar.gz", "fc7d672bafd8a1bfa66ccabbc878e28ecc812e0c2fa53f2ef95c4a04c9f42ca6"),
-        "linux_arm64": ("go1.25.3.linux-arm64.tar.gz", "50e842c343d86baae426a2c872f755ac70f22547d7b7ca85c9f329d331078b31"),
-        "linux_s390x": ("go1.25.3.linux-s390x.tar.gz", "cdec8df65f9f2920b7bd2fa4559ef880f6d90e31367549f014c8f3dd41278ae8"),
-        "windows_amd64": ("go1.25.3.windows-amd64.tar.gz", "78418dab7f8de0c695fffa2c5d3c420df51f511df8cd69745631c73b0023cec9"),
+        "darwin_amd64": ("go1.25.3.darwin-amd64.tar.gz", "2ba409dd10128f529c2b44623b559e0b7890e5f210a311325215a5ae5879474d"),
+        "darwin_arm64": ("go1.25.3.darwin-arm64.tar.gz", "746818c3703980b10279f9afec54145b35b0e3d5801fe6f5bdceeea53bcb6792"),
+        "linux_amd64": ("go1.25.3.linux-amd64.tar.gz", "821f2ede78f535fe10f95c38e2c6c87bbc200649dee20068ae4424de400196c7"),
+        "linux_arm64": ("go1.25.3.linux-arm64.tar.gz", "f6537bed6500b20d3ac1c8fd2e27609cb3c89b5e54579e344c09febb651e9598"),
+        "linux_s390x": ("go1.25.3.linux-s390x.tar.gz", "6d080fdd7333c8c84d679119613b7dc6b8703dc45f7231be78ba13e0523ffd50"),
+        "windows_amd64": ("go1.25.3.windows-amd64.tar.gz", "2f108bdf9956d0e1620d6277ecf761438f4562f7d551ca6258e8579875cf1ce5"),
     },
-    urls = ["https://storage.googleapis.com/public-bazel-artifacts/go/20251016-224003/{}"],
+    urls = ["https://storage.googleapis.com/public-bazel-artifacts/go/20251028-171726/{}"],
     version = "1.25.3",
 )
 

build/bazelutil/distdir_files.bzl

@@ -1185,12 +1185,12 @@ DISTDIR_FILES = {
     "https://storage.googleapis.com/public-bazel-artifacts/c-deps/20250801-193032/libproj_foreign.macos.20250801-193032.tar.gz": "8d28434cd175f0a32dfdd8ba8a5fa44c3d04d1e53cccfe9dbb3c6e301a03a47c",
     "https://storage.googleapis.com/public-bazel-artifacts/c-deps/20250801-193032/libproj_foreign.macosarm.20250801-193032.tar.gz": "a4b0bbb056bb462682b49ec34816f02c71047b38733d50d8de78b737c892db61",
     "https://storage.googleapis.com/public-bazel-artifacts/c-deps/20250801-193032/libproj_foreign.windows.20250801-193032.tar.gz": "a61f4faf7a7d017a194c64b453a38c982423ef3678fa049dbf114920759da59c",
-    "https://storage.googleapis.com/public-bazel-artifacts/go/20251016-224003/go1.25.3.darwin-amd64.tar.gz": "b229aaaf13ff3bb5f918a31b5e7da27673e173b1578b12a0157be3ca6512bb22",
-    "https://storage.googleapis.com/public-bazel-artifacts/go/20251016-224003/go1.25.3.darwin-arm64.tar.gz": "1601fb97e9f325f9c2addc7193ca79750e20bd8d3e18e5a0c47e406ec0dfc63f",
-    "https://storage.googleapis.com/public-bazel-artifacts/go/20251016-224003/go1.25.3.linux-amd64.tar.gz": "fc7d672bafd8a1bfa66ccabbc878e28ecc812e0c2fa53f2ef95c4a04c9f42ca6",
-    "https://storage.googleapis.com/public-bazel-artifacts/go/20251016-224003/go1.25.3.linux-arm64.tar.gz": "50e842c343d86baae426a2c872f755ac70f22547d7b7ca85c9f329d331078b31",
-    "https://storage.googleapis.com/public-bazel-artifacts/go/20251016-224003/go1.25.3.linux-s390x.tar.gz": "cdec8df65f9f2920b7bd2fa4559ef880f6d90e31367549f014c8f3dd41278ae8",
-    "https://storage.googleapis.com/public-bazel-artifacts/go/20251016-224003/go1.25.3.windows-amd64.tar.gz": "78418dab7f8de0c695fffa2c5d3c420df51f511df8cd69745631c73b0023cec9",
+    "https://storage.googleapis.com/public-bazel-artifacts/go/20251028-171726/go1.25.3.darwin-amd64.tar.gz": "2ba409dd10128f529c2b44623b559e0b7890e5f210a311325215a5ae5879474d",
+    "https://storage.googleapis.com/public-bazel-artifacts/go/20251028-171726/go1.25.3.darwin-arm64.tar.gz": "746818c3703980b10279f9afec54145b35b0e3d5801fe6f5bdceeea53bcb6792",
+    "https://storage.googleapis.com/public-bazel-artifacts/go/20251028-171726/go1.25.3.linux-amd64.tar.gz": "821f2ede78f535fe10f95c38e2c6c87bbc200649dee20068ae4424de400196c7",
+    "https://storage.googleapis.com/public-bazel-artifacts/go/20251028-171726/go1.25.3.linux-arm64.tar.gz": "f6537bed6500b20d3ac1c8fd2e27609cb3c89b5e54579e344c09febb651e9598",
+    "https://storage.googleapis.com/public-bazel-artifacts/go/20251028-171726/go1.25.3.linux-s390x.tar.gz": "6d080fdd7333c8c84d679119613b7dc6b8703dc45f7231be78ba13e0523ffd50",
+    "https://storage.googleapis.com/public-bazel-artifacts/go/20251028-171726/go1.25.3.windows-amd64.tar.gz": "2f108bdf9956d0e1620d6277ecf761438f4562f7d551ca6258e8579875cf1ce5",
     "https://storage.googleapis.com/public-bazel-artifacts/java/railroad/rr-1.63-java8.zip": "d2791cd7a44ea5be862f33f5a9b3d40aaad9858455828ebade7007ad7113fb41",
     "https://storage.googleapis.com/public-bazel-artifacts/js/rules_jest-v0.18.4.tar.gz": "d3bb833f74b8ad054e6bff5e41606ff10a62880cc99e4d480f4bdfa70add1ba7",
     "https://storage.googleapis.com/public-bazel-artifacts/js/rules_js-v1.42.3.tar.gz": "2cfb3875e1231cefd3fada6774f2c0c5a99db0070e0e48ea398acbff7c6c765b",

build/teamcity/internal/release/build-and-publish-patched-go/commit.txt

@@ -1 +1 @@
-370d2fb9961bf6e8ed50f3e752fdf4ac5e0f284b
+9bec28dcd0f7a372d89ef9292cc72c33bf4c7cba

pkg/backup/restore_job.go

@@ -116,6 +116,10 @@ const (
 	// be _exceeded_ before we no longer fast fail the restore job after hitting the
 	// maxRestoreRetryFastFail threshold.
 	restoreRetryProgressThreshold = 0
+
+	// droppedDescsOnFailKey is an info key that is set for a restore job when it
+	// has finished dropping its descriptors on failure.
+	droppedDescsOnFailKey = "dropped_descs_on_fail"
 )
 
 var restoreStatsInsertionConcurrency = settings.RegisterIntSetting(
@@ -2820,6 +2824,13 @@ func (r *restoreResumer) OnFailOrCancel(
 		return err
 	}
 
+	testingKnobs := execCfg.BackupRestoreTestingKnobs
+	if testingKnobs != nil && testingKnobs.AfterRevertRestoreDropDescriptors != nil {
+		if err := testingKnobs.AfterRevertRestoreDropDescriptors(); err != nil {
+			return err
+		}
+	}
+
 	if details.DescriptorCoverage == tree.AllDescriptors {
 		// The temporary system table descriptors should already have been dropped
 		// in `dropDescriptors` but we still need to drop the temporary system db.
@@ -2865,6 +2876,19 @@ func (r *restoreResumer) dropDescriptors(
 		return nil
 	}
 
+	jobInfo := jobs.InfoStorageForJob(txn, r.job.ID())
+	_, hasDropped, err := jobInfo.Get(
+		ctx, "get-restore-dropped-descs-on-fail-key", droppedDescsOnFailKey,
+	)
+	if err != nil {
+		return err
+	}
+	if hasDropped {
+		// Descriptors have already been dropped once before, this is a retry of the
+		// cleanup.
+		return nil
+	}
+
 	b := txn.KV().NewBatch()
 	const kvTrace = false
 	// Collect the tables into mutable versions.
@@ -3186,7 +3210,10 @@ func (r *restoreResumer) dropDescriptors(
 		return errors.Wrap(err, "dropping tables created at the start of restore caused by fail/cancel")
 	}
 
-	return nil
+	return errors.Wrap(
+		jobInfo.Write(ctx, droppedDescsOnFailKey, []byte{}),
+		"checkpointing dropped descs on fail",
+	)
 }
 
 // removeExistingTypeBackReferences removes back references from types that

pkg/backup/restore_test.go

@@ -318,3 +318,55 @@ func TestRestoreDuplicateTempTables(t *testing.T) {
 	result := sqlDB.QueryStr(t, `SELECT table_name FROM [SHOW TABLES] ORDER BY table_name`)
 	require.Equal(t, [][]string{{"permanent_table"}}, result)
 }
+
+func TestRestoreRetryRevert(t *testing.T) {
+	defer leaktest.AfterTest(t)()
+	defer log.Scope(t).Close(t)
+
+	droppedDescs := make(chan struct{})
+	jobPaused := make(chan struct{})
+	testKnobs := &sql.BackupRestoreTestingKnobs{
+		AfterRevertRestoreDropDescriptors: func() error {
+			close(droppedDescs)
+			<-jobPaused
+			return nil
+		},
+	}
+	var params base.TestClusterArgs
+	params.ServerArgs.Knobs.BackupRestore = testKnobs
+
+	_, sqlDB, _, cleanupFn := backuptestutils.StartBackupRestoreTestCluster(
+		t, singleNode, backuptestutils.WithParams(params),
+	)
+	defer cleanupFn()
+
+	// We create a variety of descriptors to ensure that missing descriptors
+	// during drop do not break revert.
+	sqlDB.Exec(t, "CREATE DATABASE foo")
+	sqlDB.Exec(t, "USE foo")
+	sqlDB.Exec(t, "CREATE OR REPLACE FUNCTION bar(a INT) RETURNS INT AS 'SELECT a*a' LANGUAGE SQL;")
+	sqlDB.Exec(t, "CREATE TYPE baz AS ENUM ('a', 'b', 'c')")
+	sqlDB.Exec(t, "CREATE TABLE qux (x INT)")
+	sqlDB.Exec(t, "BACKUP DATABASE foo INTO 'nodelocal://1/backup'")
+
+	// We need restore to publish descriptors so that they will be cleaned up
+	// during restore.
+	sqlDB.Exec(t, "SET CLUSTER SETTING jobs.debug.pausepoints = 'restore.after_publishing_descriptors'")
+
+	var restoreID jobspb.JobID
+	sqlDB.QueryRow(
+		t, "RESTORE DATABASE foo FROM LATEST IN 'nodelocal://1/backup' WITH detached, new_db_name='foo_restored'",
+	).Scan(&restoreID)
+
+	jobutils.WaitForJobToPause(t, sqlDB, restoreID)
+
+	sqlDB.Exec(t, "CANCEL JOB $1", restoreID)
+	<-droppedDescs
+	sqlDB.Exec(t, "PAUSE JOB $1", restoreID)
+	jobutils.WaitForJobToPause(t, sqlDB, restoreID)
+	close(jobPaused)
+	testKnobs.AfterRevertRestoreDropDescriptors = nil
+
+	sqlDB.Exec(t, "RESUME JOB $1", restoreID)
+	jobutils.WaitForJobToCancel(t, sqlDB, restoreID)
+}

pkg/sql/exec_util_backup.go

@@ -90,6 +90,10 @@ type BackupRestoreTestingKnobs struct {
 	// AfterAddRemoteSST is called after a remote SST is linked to pebble during
 	// the link phase of online restore.
 	AfterAddRemoteSST func() error
+
+	// AfterRevertRestoreDropDescriptors is called after a reverting restore
+	// drops its descriptors.
+	AfterRevertRestoreDropDescriptors func() error
 }
 
 var _ base.ModuleTestingKnobs = &BackupRestoreTestingKnobs{}