Incremental backup and point in time recovery for XtraBackup (#13156)

* incremental backup is always using 'builtin' engine Signed-off-by: Shlomi Noach <[email protected]> * restore: use 'builtin' for incremental restore Signed-off-by: Shlomi Noach <[email protected]> * test all backup types Signed-off-by: Shlomi Noach <[email protected]> * format code Signed-off-by: Shlomi Noach <[email protected]> * Populate PurgedPosition Signed-off-by: Shlomi Noach <[email protected]> * cleanup backups at the end of each test case Signed-off-by: Shlomi Noach <[email protected]> * improved cleanup Signed-off-by: Shlomi Noach <[email protected]> * rename variable Signed-off-by: Shlomi Noach <[email protected]> * record all backups Signed-off-by: Shlomi Noach <[email protected]> * no need to cleanup backups in between test cases, since each new case has a completely different UUID Signed-off-by: Shlomi Noach <[email protected]> * install xtrabackup on backup_pitr tests Signed-off-by: Shlomi Noach <[email protected]> * use pgzip for xtrabackup Signed-off-by: Shlomi Noach <[email protected]> * more debug info Signed-off-by: Shlomi Noach <[email protected]> * builtin engine: store gtid_purged in manifest Signed-off-by: Shlomi Noach <[email protected]> * use backupfrom-GTID as incremental-from-GTID if first binary log has empty Previous-GTIDS Signed-off-by: Shlomi Noach <[email protected]> * more unit tests Signed-off-by: Shlomi Noach <[email protected]> * improve error message Signed-off-by: Shlomi Noach <[email protected]> * capturing MySQL's stderr and reading and logging if not empty Signed-off-by: Shlomi Noach <[email protected]> * At the end of Xtrabackup restore, validate that @@gtid_purged (and thereby @@gtid_executed) are set to the backup's @@gtid_purge. Make it so if they aren't Signed-off-by: Shlomi Noach <[email protected]> * add comperssion details into test case. Fix GTID validation of manifest file Signed-off-by: Shlomi Noach <[email protected]> * check manifest Signed-off-by: Shlomi Noach <[email protected]> * Refactor into function Signed-off-by: Shlomi Noach <[email protected]> * check manifest.Position.GTIDSet Signed-off-by: Shlomi Noach <[email protected]> * fix wrangler tests Signed-off-by: Shlomi Noach <[email protected]> * typo Signed-off-by: Shlomi Noach <[email protected]> * Update go/vt/mysqlctl/backup.go Co-authored-by: Matt Lord <[email protected]> Signed-off-by: Shlomi Noach <[email protected]> * Update go/vt/mysqlctl/backup.go Co-authored-by: Matt Lord <[email protected]> Signed-off-by: Shlomi Noach <[email protected]> * Update go/vt/mysqlctl/backup.go Co-authored-by: Matt Lord <[email protected]> Signed-off-by: Shlomi Noach <[email protected]> * typo Signed-off-by: Shlomi Noach <[email protected]> * Update go/vt/mysqlctl/mysqld.go Co-authored-by: Matt Lord <[email protected]> Signed-off-by: Shlomi Noach <[email protected]> * Update go/vt/mysqlctl/mysqld.go Co-authored-by: Matt Lord <[email protected]> Signed-off-by: Shlomi Noach <[email protected]> * Update go/vt/mysqlctl/mysqld.go Co-authored-by: Matt Lord <[email protected]> Signed-off-by: Shlomi Noach <[email protected]> --------- Signed-off-by: Shlomi Noach <[email protected]> Co-authored-by: Matt Lord <[email protected]>
vitessio · Jun 6, 2023 · bd50b93 · bd50b93
1 parent 4ba60af
commit bd50b93
Show file tree

Hide file tree

Showing 13 changed files with 392 additions and 203 deletions.
diff --git a/.github/workflows/cluster_endtoend_backup_pitr.yml b/.github/workflows/cluster_endtoend_backup_pitr.yml
@@ -84,15 +84,16 @@ jobs:
       if: steps.skip-workflow.outputs.skip-workflow == 'false' && steps.changes.outputs.end_to_end == 'true'
       run: |
         
-        # Get key to latest MySQL repo
-        sudo apt-key adv --keyserver keyserver.ubuntu.com --recv-keys 467B942D3A79BD29
-        # Setup MySQL 8.0
-        wget -c https://dev.mysql.com/get/mysql-apt-config_0.8.24-1_all.deb
-        echo mysql-apt-config mysql-apt-config/select-server select mysql-8.0 | sudo debconf-set-selections
-        sudo DEBIAN_FRONTEND="noninteractive" dpkg -i mysql-apt-config*
+        # Setup Percona Server for MySQL 8.0
         sudo apt-get update
+        sudo apt-get install -y lsb-release gnupg2 curl
+        wget https://repo.percona.com/apt/percona-release_latest.$(lsb_release -sc)_all.deb
+        sudo DEBIAN_FRONTEND="noninteractive" dpkg -i percona-release_latest.$(lsb_release -sc)_all.deb
+        sudo percona-release setup ps80
+        sudo apt-get update
+
         # Install everything else we need, and configure
-        sudo apt-get install -y mysql-server mysql-client make unzip g++ etcd curl git wget eatmydata xz-utils libncurses5
+        sudo apt-get install -y percona-server-server percona-server-client make unzip g++ etcd git wget eatmydata xz-utils libncurses5
 
         sudo service mysql stop
         sudo service etcd stop
@@ -103,6 +104,8 @@ jobs:
         # install JUnit report formatter
         go install github.com/vitessio/go-junit-report@HEAD
 
+        sudo apt-get install percona-xtrabackup-80 lz4
+
     - name: Setup launchable dependencies
       if: steps.skip-workflow.outputs.skip-workflow == 'false' && steps.changes.outputs.end_to_end == 'true' && github.base_ref == 'main'
       run: |

diff --git a/.github/workflows/cluster_endtoend_backup_pitr_mysql57.yml b/.github/workflows/cluster_endtoend_backup_pitr_mysql57.yml
@@ -13,6 +13,10 @@ env:
   LAUNCHABLE_WORKSPACE: "vitess-app"
   GITHUB_PR_HEAD_SHA: "${{ github.event.pull_request.head.sha }}"
 
+  # This is used if we need to pin the xtrabackup version used in tests.
+  # If this is NOT set then the latest version available will be used.
+  #XTRABACKUP_VERSION: "2.4.24-1"
+
 jobs:
   build:
     name: Run endtoend tests on Cluster (backup_pitr) mysql57
@@ -114,6 +118,18 @@ jobs:
         # install JUnit report formatter
         go install github.com/vitessio/go-junit-report@HEAD
 
+        wget "https://repo.percona.com/apt/percona-release_latest.$(lsb_release -sc)_all.deb"
+        sudo apt-get install -y gnupg2
+        sudo dpkg -i "percona-release_latest.$(lsb_release -sc)_all.deb"
+        sudo apt-get update
+        if [[ -n $XTRABACKUP_VERSION ]]; then
+          debfile="percona-xtrabackup-24_$XTRABACKUP_VERSION.$(lsb_release -sc)_amd64.deb"
+          wget "https://repo.percona.com/pxb-24/apt/pool/main/p/percona-xtrabackup-24/$debfile"
+          sudo apt install -y "./$debfile"
+        else
+          sudo apt-get install -y percona-xtrabackup-24
+        fi
+
     - name: Setup launchable dependencies
       if: steps.skip-workflow.outputs.skip-workflow == 'false' && steps.changes.outputs.end_to_end == 'true' && github.base_ref == 'main'
       run: |

diff --git a/go/test/endtoend/backup/pitr/backup_mysqlctld_pitr_test.go b/go/test/endtoend/backup/pitr/backup_mysqlctld_pitr_test.go
@@ -52,161 +52,184 @@ func waitForReplica(t *testing.T) {
 }
 
 // TestIncrementalBackupMysqlctld - tests incremental backups using myslctld
-func TestIncrementalBackupMysqlctld(t *testing.T) {
+func TestIncrementalBackup(t *testing.T) {
 	defer cluster.PanicHandler(t)
-	// setup cluster for the testing
-	code, err := backup.LaunchCluster(backup.Mysqlctld, "xbstream", 0, nil)
-	require.NoError(t, err, "setup failed with status code %d", code)
-	defer backup.TearDownCluster()
-
-	backup.InitTestTable(t)
-
-	rowsPerPosition := map[string]int{}
-	backupPositions := []string{}
-
-	recordRowsPerPosition := func(t *testing.T) {
-		pos := backup.GetReplicaPosition(t)
-		msgs := backup.ReadRowsFromReplica(t)
-		if _, ok := rowsPerPosition[pos]; !ok {
-			backupPositions = append(backupPositions, pos)
-			rowsPerPosition[pos] = len(msgs)
-		}
-	}
 
-	var fullBackupPos mysql.Position
-	t.Run("full backup", func(t *testing.T) {
-		backup.InsertRowOnPrimary(t, "before-full-backup")
-		waitForReplica(t)
-		manifest, _ := backup.TestReplicaFullBackup(t)
-		fullBackupPos = manifest.Position
-		require.False(t, fullBackupPos.IsZero())
-		//
-		msgs := backup.ReadRowsFromReplica(t)
-		pos := mysql.EncodePosition(fullBackupPos)
-		backupPositions = append(backupPositions, pos)
-		rowsPerPosition[pos] = len(msgs)
-	})
-
-	lastBackupPos := fullBackupPos
-	backup.InsertRowOnPrimary(t, "before-incremental-backups")
-
-	tt := []struct {
-		name              string
-		writeBeforeBackup bool
-		fromFullPosition  bool
-		autoPosition      bool
-		expectError       string
+	tcases := []struct {
+		name      string
+		setupType int
+		comprss   *backup.CompressionDetails
 	}{
 		{
-			name: "first incremental backup",
+			"BuiltinBackup", backup.BuiltinBackup, nil,
 		},
 		{
-			name:              "make writes, succeed",
-			writeBeforeBackup: true,
+			"XtraBackup", backup.XtraBackup, &backup.CompressionDetails{
+				CompressorEngineName: "pgzip",
+			},
 		},
 		{
-			name:        "fail, no binary logs to backup",
-			expectError: "no binary logs to backup",
-		},
-		{
-			name:              "make writes again, succeed",
-			writeBeforeBackup: true,
-		},
-		{
-			name:              "auto position, succeed",
-			writeBeforeBackup: true,
-			autoPosition:      true,
-		},
-		{
-			name:         "fail auto position, no binary logs to backup",
-			autoPosition: true,
-			expectError:  "no binary logs to backup",
-		},
-		{
-			name:              "auto position, make writes again, succeed",
-			writeBeforeBackup: true,
-			autoPosition:      true,
-		},
-		{
-			name:             "from full backup position",
-			fromFullPosition: true,
+			"Mysqlctld", backup.Mysqlctld, nil,
 		},
 	}
-	var fromFullPositionBackups []string
-	for _, tc := range tt {
-		t.Run(tc.name, func(t *testing.T) {
-			if tc.writeBeforeBackup {
-				backup.InsertRowOnPrimary(t, "")
-			}
-			// we wait for 1 second because backups are written to a directory named after the current timestamp,
-			// in 1 second resolution. We want to avoid two backups that have the same pathname. Realistically this
-			// is only ever a problem in this end-to-end test, not in production.
-			// Also, we gie the replica a chance to catch up.
-			time.Sleep(1100 * time.Millisecond)
-			waitForReplica(t)
-			recordRowsPerPosition(t)
-			// configure --incremental-from-pos to either:
-			// - auto
-			// - explicit last backup pos
-			// - back in history to the original full backup
-			var incrementalFromPos mysql.Position
-			if !tc.autoPosition {
-				incrementalFromPos = lastBackupPos
-				if tc.fromFullPosition {
-					incrementalFromPos = fullBackupPos
+	for _, tcase := range tcases {
+		t.Run(tcase.name, func(t *testing.T) {
+			// setup cluster for the testing
+			code, err := backup.LaunchCluster(tcase.setupType, "xbstream", 0, tcase.comprss)
+			require.NoError(t, err, "setup failed with status code %d", code)
+			defer backup.TearDownCluster()
+
+			backup.InitTestTable(t)
+
+			rowsPerPosition := map[string]int{}
+			backupPositions := []string{}
+
+			recordRowsPerPosition := func(t *testing.T) {
+				pos := backup.GetReplicaPosition(t)
+				msgs := backup.ReadRowsFromReplica(t)
+				if _, ok := rowsPerPosition[pos]; !ok {
+					backupPositions = append(backupPositions, pos)
+					rowsPerPosition[pos] = len(msgs)
 				}
 			}
-			manifest, backupName := backup.TestReplicaIncrementalBackup(t, incrementalFromPos, tc.expectError)
-			if tc.expectError != "" {
-				return
-			}
-			defer func() {
-				lastBackupPos = manifest.Position
-			}()
-			if tc.fromFullPosition {
-				fromFullPositionBackups = append(fromFullPositionBackups, backupName)
-			}
-			require.False(t, manifest.FromPosition.IsZero())
-			require.NotEqual(t, manifest.Position, manifest.FromPosition)
-			require.True(t, manifest.Position.GTIDSet.Contains(manifest.FromPosition.GTIDSet))
 
-			gtidPurgedPos, err := mysql.ParsePosition(mysql.Mysql56FlavorID, backup.GetReplicaGtidPurged(t))
-			require.NoError(t, err)
-			fromPositionIncludingPurged := manifest.FromPosition.GTIDSet.Union(gtidPurgedPos.GTIDSet)
+			var fullBackupPos mysql.Position
+			t.Run("full backup", func(t *testing.T) {
+				backup.InsertRowOnPrimary(t, "before-full-backup")
+				waitForReplica(t)
+
+				manifest, _ := backup.TestReplicaFullBackup(t)
+				fullBackupPos = manifest.Position
+				require.False(t, fullBackupPos.IsZero())
+				//
+				msgs := backup.ReadRowsFromReplica(t)
+				pos := mysql.EncodePosition(fullBackupPos)
+				backupPositions = append(backupPositions, pos)
+				rowsPerPosition[pos] = len(msgs)
+			})
 
-			expectFromPosition := lastBackupPos.GTIDSet.Union(gtidPurgedPos.GTIDSet)
-			if !incrementalFromPos.IsZero() {
-				expectFromPosition = incrementalFromPos.GTIDSet.Union(gtidPurgedPos.GTIDSet)
+			lastBackupPos := fullBackupPos
+			backup.InsertRowOnPrimary(t, "before-incremental-backups")
+
+			tt := []struct {
+				name              string
+				writeBeforeBackup bool
+				fromFullPosition  bool
+				autoPosition      bool
+				expectError       string
+			}{
+				{
+					name: "first incremental backup",
+				},
+				{
+					name:              "make writes, succeed",
+					writeBeforeBackup: true,
+				},
+				{
+					name:        "fail, no binary logs to backup",
+					expectError: "no binary logs to backup",
+				},
+				{
+					name:              "make writes again, succeed",
+					writeBeforeBackup: true,
+				},
+				{
+					name:              "auto position, succeed",
+					writeBeforeBackup: true,
+					autoPosition:      true,
+				},
+				{
+					name:         "fail auto position, no binary logs to backup",
+					autoPosition: true,
+					expectError:  "no binary logs to backup",
+				},
+				{
+					name:              "auto position, make writes again, succeed",
+					writeBeforeBackup: true,
+					autoPosition:      true,
+				},
+				{
+					name:             "from full backup position",
+					fromFullPosition: true,
+				},
+			}
+			var fromFullPositionBackups []string
+			for _, tc := range tt {
+				t.Run(tc.name, func(t *testing.T) {
+					if tc.writeBeforeBackup {
+						backup.InsertRowOnPrimary(t, "")
+					}
+					// we wait for 1 second because backups are written to a directory named after the current timestamp,
+					// in 1 second resolution. We want to avoid two backups that have the same pathname. Realistically this
+					// is only ever a problem in this end-to-end test, not in production.
+					// Also, we gie the replica a chance to catch up.
+					time.Sleep(1100 * time.Millisecond)
+					waitForReplica(t)
+					recordRowsPerPosition(t)
+					// configure --incremental-from-pos to either:
+					// - auto
+					// - explicit last backup pos
+					// - back in history to the original full backup
+					var incrementalFromPos mysql.Position
+					if !tc.autoPosition {
+						incrementalFromPos = lastBackupPos
+						if tc.fromFullPosition {
+							incrementalFromPos = fullBackupPos
+						}
+					}
+					manifest, backupName := backup.TestReplicaIncrementalBackup(t, incrementalFromPos, tc.expectError)
+					if tc.expectError != "" {
+						return
+					}
+					defer func() {
+						lastBackupPos = manifest.Position
+					}()
+					if tc.fromFullPosition {
+						fromFullPositionBackups = append(fromFullPositionBackups, backupName)
+					}
+					require.False(t, manifest.FromPosition.IsZero())
+					require.NotEqual(t, manifest.Position, manifest.FromPosition)
+					require.True(t, manifest.Position.GTIDSet.Union(manifest.PurgedPosition.GTIDSet).Contains(manifest.FromPosition.GTIDSet))
+
+					gtidPurgedPos, err := mysql.ParsePosition(mysql.Mysql56FlavorID, backup.GetReplicaGtidPurged(t))
+					require.NoError(t, err)
+					fromPositionIncludingPurged := manifest.FromPosition.GTIDSet.Union(gtidPurgedPos.GTIDSet)
+
+					expectFromPosition := lastBackupPos.GTIDSet.Union(gtidPurgedPos.GTIDSet)
+					if !incrementalFromPos.IsZero() {
+						expectFromPosition = incrementalFromPos.GTIDSet.Union(gtidPurgedPos.GTIDSet)
+					}
+					require.Equalf(t, expectFromPosition, fromPositionIncludingPurged, "expected: %v, found: %v, gtid_purged: %v,  manifest.Position: %v", expectFromPosition, fromPositionIncludingPurged, gtidPurgedPos, manifest.Position)
+				})
 			}
-			require.Equalf(t, expectFromPosition, fromPositionIncludingPurged, "expected: %v, found: %v", expectFromPosition, fromPositionIncludingPurged)
-		})
-	}
 
-	testRestores := func(t *testing.T) {
-		for _, r := range rand.Perm(len(backupPositions)) {
-			pos := backupPositions[r]
-			testName := fmt.Sprintf("%s, %d records", pos, rowsPerPosition[pos])
-			t.Run(testName, func(t *testing.T) {
-				restoreToPos, err := mysql.DecodePosition(pos)
-				require.NoError(t, err)
-				backup.TestReplicaRestoreToPos(t, restoreToPos, "")
-				msgs := backup.ReadRowsFromReplica(t)
-				count, ok := rowsPerPosition[pos]
-				require.True(t, ok)
-				assert.Equalf(t, count, len(msgs), "messages: %v", msgs)
+			testRestores := func(t *testing.T) {
+				for _, r := range rand.Perm(len(backupPositions)) {
+					pos := backupPositions[r]
+					testName := fmt.Sprintf("%s, %d records", pos, rowsPerPosition[pos])
+					t.Run(testName, func(t *testing.T) {
+						restoreToPos, err := mysql.DecodePosition(pos)
+						require.NoError(t, err)
+						backup.TestReplicaRestoreToPos(t, restoreToPos, "")
+						msgs := backup.ReadRowsFromReplica(t)
+						count, ok := rowsPerPosition[pos]
+						require.True(t, ok)
+						assert.Equalf(t, count, len(msgs), "messages: %v", msgs)
+					})
+				}
+			}
+			t.Run("PITR", func(t *testing.T) {
+				testRestores(t)
 			})
-		}
+			t.Run("remove full position backups", func(t *testing.T) {
+				// Delete the fromFullPosition backup(s), which leaves us with less restore options. Try again.
+				for _, backupName := range fromFullPositionBackups {
+					backup.RemoveBackup(t, backupName)
+				}
+			})
+			t.Run("PITR-2", func(t *testing.T) {
+				testRestores(t)
+			})
+		})
 	}
-	t.Run("PITR", func(t *testing.T) {
-		testRestores(t)
-	})
-	t.Run("remove full position backups", func(t *testing.T) {
-		// Delete the fromFullPosition backup(s), which leaves us with less restore options. Try again.
-		for _, backupName := range fromFullPositionBackups {
-			backup.RemoveBackup(t, backupName)
-		}
-	})
-	t.Run("PITR-2", func(t *testing.T) {
-		testRestores(t)
-	})
 }