Switch to use semisync source / replica plugins (#15791)

Signed-off-by: Dirkjan Bussink <d.bussink@gmail.com>
vitessio · Apr 26, 2024 · 8f6cfaa · 8f6cfaa
1 parent ee6b837
commit 8f6cfaa
Show file tree

Hide file tree

Showing 46 changed files with 660 additions and 293 deletions.
diff --git a/.github/workflows/cluster_endtoend_ers_prs_newfeatures_heavy.yml b/.github/workflows/cluster_endtoend_ers_prs_newfeatures_heavy.yml
@@ -137,7 +137,7 @@ jobs:
 
         # Increase our open file descriptor limit as we could hit this
         ulimit -n 65536
-        cat <<-EOF>>./config/mycnf/mysql80.cnf
+        cat <<-EOF>>./config/mycnf/mysql8026.cnf
         innodb_buffer_pool_dump_at_shutdown=OFF
         innodb_buffer_pool_in_core_file=OFF
         innodb_buffer_pool_load_at_startup=OFF

diff --git a/.github/workflows/cluster_endtoend_onlineddl_vrepl.yml b/.github/workflows/cluster_endtoend_onlineddl_vrepl.yml
@@ -136,7 +136,7 @@ jobs:
 
         set -exo pipefail
 
-        cat <<-EOF>>./config/mycnf/mysql80.cnf
+        cat <<-EOF>>./config/mycnf/mysql8026.cnf
         binlog-transaction-compression=ON
         EOF
         

diff --git a/.github/workflows/cluster_endtoend_onlineddl_vrepl_stress.yml b/.github/workflows/cluster_endtoend_onlineddl_vrepl_stress.yml
@@ -136,7 +136,7 @@ jobs:
 
         set -exo pipefail
 
-        cat <<-EOF>>./config/mycnf/mysql80.cnf
+        cat <<-EOF>>./config/mycnf/mysql8026.cnf
         binlog-transaction-compression=ON
         EOF
         

diff --git a/.github/workflows/cluster_endtoend_onlineddl_vrepl_stress_suite.yml b/.github/workflows/cluster_endtoend_onlineddl_vrepl_stress_suite.yml
@@ -136,7 +136,7 @@ jobs:
 
         set -exo pipefail
 
-        cat <<-EOF>>./config/mycnf/mysql80.cnf
+        cat <<-EOF>>./config/mycnf/mysql8026.cnf
         binlog-transaction-compression=ON
         EOF
         

diff --git a/.github/workflows/cluster_endtoend_onlineddl_vrepl_suite.yml b/.github/workflows/cluster_endtoend_onlineddl_vrepl_suite.yml
@@ -136,7 +136,7 @@ jobs:
 
         set -exo pipefail
 
-        cat <<-EOF>>./config/mycnf/mysql80.cnf
+        cat <<-EOF>>./config/mycnf/mysql8026.cnf
         binlog-transaction-compression=ON
         EOF
         

diff --git a/.github/workflows/cluster_endtoend_schemadiff_vrepl.yml b/.github/workflows/cluster_endtoend_schemadiff_vrepl.yml
@@ -136,7 +136,7 @@ jobs:
 
         set -exo pipefail
 
-        cat <<-EOF>>./config/mycnf/mysql80.cnf
+        cat <<-EOF>>./config/mycnf/mysql8026.cnf
         binlog-transaction-compression=ON
         EOF
         

diff --git a/.github/workflows/cluster_endtoend_vreplication_across_db_versions.yml b/.github/workflows/cluster_endtoend_vreplication_across_db_versions.yml
@@ -137,7 +137,7 @@ jobs:
 
         # Increase our open file descriptor limit as we could hit this
         ulimit -n 65536
-        cat <<-EOF>>./config/mycnf/mysql80.cnf
+        cat <<-EOF>>./config/mycnf/mysql8026.cnf
         innodb_buffer_pool_dump_at_shutdown=OFF
         innodb_buffer_pool_in_core_file=OFF
         innodb_buffer_pool_load_at_startup=OFF
@@ -153,7 +153,7 @@ jobs:
         slow-query-log=OFF
         EOF
         
-        cat <<-EOF>>./config/mycnf/mysql80.cnf
+        cat <<-EOF>>./config/mycnf/mysql8026.cnf
         binlog-transaction-compression=ON
         EOF
         

diff --git a/.github/workflows/cluster_endtoend_vreplication_basic.yml b/.github/workflows/cluster_endtoend_vreplication_basic.yml
@@ -137,7 +137,7 @@ jobs:
 
         # Increase our open file descriptor limit as we could hit this
         ulimit -n 65536
-        cat <<-EOF>>./config/mycnf/mysql80.cnf
+        cat <<-EOF>>./config/mycnf/mysql8026.cnf
         innodb_buffer_pool_dump_at_shutdown=OFF
         innodb_buffer_pool_in_core_file=OFF
         innodb_buffer_pool_load_at_startup=OFF
@@ -153,7 +153,7 @@ jobs:
         slow-query-log=OFF
         EOF
         
-        cat <<-EOF>>./config/mycnf/mysql80.cnf
+        cat <<-EOF>>./config/mycnf/mysql8026.cnf
         binlog-transaction-compression=ON
         EOF
         

diff --git a/.github/workflows/cluster_endtoend_vreplication_cellalias.yml b/.github/workflows/cluster_endtoend_vreplication_cellalias.yml
@@ -137,7 +137,7 @@ jobs:
 
         # Increase our open file descriptor limit as we could hit this
         ulimit -n 65536
-        cat <<-EOF>>./config/mycnf/mysql80.cnf
+        cat <<-EOF>>./config/mycnf/mysql8026.cnf
         innodb_buffer_pool_dump_at_shutdown=OFF
         innodb_buffer_pool_in_core_file=OFF
         innodb_buffer_pool_load_at_startup=OFF
@@ -153,7 +153,7 @@ jobs:
         slow-query-log=OFF
         EOF
         
-        cat <<-EOF>>./config/mycnf/mysql80.cnf
+        cat <<-EOF>>./config/mycnf/mysql8026.cnf
         binlog-transaction-compression=ON
         EOF
         

diff --git a/.github/workflows/cluster_endtoend_vreplication_foreign_key_stress.yml b/.github/workflows/cluster_endtoend_vreplication_foreign_key_stress.yml
@@ -137,7 +137,7 @@ jobs:
 
         # Increase our open file descriptor limit as we could hit this
         ulimit -n 65536
-        cat <<-EOF>>./config/mycnf/mysql80.cnf
+        cat <<-EOF>>./config/mycnf/mysql8026.cnf
         innodb_buffer_pool_dump_at_shutdown=OFF
         innodb_buffer_pool_in_core_file=OFF
         innodb_buffer_pool_load_at_startup=OFF
@@ -153,7 +153,7 @@ jobs:
         slow-query-log=OFF
         EOF
         
-        cat <<-EOF>>./config/mycnf/mysql80.cnf
+        cat <<-EOF>>./config/mycnf/mysql8026.cnf
         binlog-transaction-compression=ON
         EOF
         

diff --git a/.github/workflows/cluster_endtoend_vreplication_migrate_vdiff2_convert_tz.yml b/.github/workflows/cluster_endtoend_vreplication_migrate_vdiff2_convert_tz.yml
@@ -137,7 +137,7 @@ jobs:
 
         # Increase our open file descriptor limit as we could hit this
         ulimit -n 65536
-        cat <<-EOF>>./config/mycnf/mysql80.cnf
+        cat <<-EOF>>./config/mycnf/mysql8026.cnf
         innodb_buffer_pool_dump_at_shutdown=OFF
         innodb_buffer_pool_in_core_file=OFF
         innodb_buffer_pool_load_at_startup=OFF
@@ -153,7 +153,7 @@ jobs:
         slow-query-log=OFF
         EOF
         
-        cat <<-EOF>>./config/mycnf/mysql80.cnf
+        cat <<-EOF>>./config/mycnf/mysql8026.cnf
         binlog-transaction-compression=ON
         EOF
         

diff --git a/.github/workflows/cluster_endtoend_vreplication_multi_tenant.yml b/.github/workflows/cluster_endtoend_vreplication_multi_tenant.yml
@@ -137,7 +137,7 @@ jobs:
 
         # Increase our open file descriptor limit as we could hit this
         ulimit -n 65536
-        cat <<-EOF>>./config/mycnf/mysql80.cnf
+        cat <<-EOF>>./config/mycnf/mysql8026.cnf
         innodb_buffer_pool_dump_at_shutdown=OFF
         innodb_buffer_pool_in_core_file=OFF
         innodb_buffer_pool_load_at_startup=OFF
@@ -153,7 +153,7 @@ jobs:
         slow-query-log=OFF
         EOF
         
-        cat <<-EOF>>./config/mycnf/mysql80.cnf
+        cat <<-EOF>>./config/mycnf/mysql8026.cnf
         binlog-transaction-compression=ON
         EOF
         

diff --git a/.github/workflows/cluster_endtoend_vreplication_partial_movetables_and_materialize.yml b/.github/workflows/cluster_endtoend_vreplication_partial_movetables_and_materialize.yml
@@ -137,7 +137,7 @@ jobs:
 
         # Increase our open file descriptor limit as we could hit this
         ulimit -n 65536
-        cat <<-EOF>>./config/mycnf/mysql80.cnf
+        cat <<-EOF>>./config/mycnf/mysql8026.cnf
         innodb_buffer_pool_dump_at_shutdown=OFF
         innodb_buffer_pool_in_core_file=OFF
         innodb_buffer_pool_load_at_startup=OFF
@@ -153,7 +153,7 @@ jobs:
         slow-query-log=OFF
         EOF
         
-        cat <<-EOF>>./config/mycnf/mysql80.cnf
+        cat <<-EOF>>./config/mycnf/mysql8026.cnf
         binlog-transaction-compression=ON
         EOF
         

diff --git a/.github/workflows/cluster_endtoend_vreplication_v2.yml b/.github/workflows/cluster_endtoend_vreplication_v2.yml
@@ -137,7 +137,7 @@ jobs:
 
         # Increase our open file descriptor limit as we could hit this
         ulimit -n 65536
-        cat <<-EOF>>./config/mycnf/mysql80.cnf
+        cat <<-EOF>>./config/mycnf/mysql8026.cnf
         innodb_buffer_pool_dump_at_shutdown=OFF
         innodb_buffer_pool_in_core_file=OFF
         innodb_buffer_pool_load_at_startup=OFF
@@ -153,7 +153,7 @@ jobs:
         slow-query-log=OFF
         EOF
         
-        cat <<-EOF>>./config/mycnf/mysql80.cnf
+        cat <<-EOF>>./config/mycnf/mysql8026.cnf
         binlog-transaction-compression=ON
         EOF
         

diff --git a/.github/workflows/cluster_endtoend_vtctlbackup_sharded_clustertest_heavy.yml b/.github/workflows/cluster_endtoend_vtctlbackup_sharded_clustertest_heavy.yml
@@ -137,7 +137,7 @@ jobs:
 
         # Increase our open file descriptor limit as we could hit this
         ulimit -n 65536
-        cat <<-EOF>>./config/mycnf/mysql80.cnf
+        cat <<-EOF>>./config/mycnf/mysql8026.cnf
         innodb_buffer_pool_dump_at_shutdown=OFF
         innodb_buffer_pool_in_core_file=OFF
         innodb_buffer_pool_load_at_startup=OFF

diff --git a/.github/workflows/cluster_endtoend_vtgate_general_heavy.yml b/.github/workflows/cluster_endtoend_vtgate_general_heavy.yml
@@ -137,7 +137,7 @@ jobs:
 
         # Increase our open file descriptor limit as we could hit this
         ulimit -n 65536
-        cat <<-EOF>>./config/mycnf/mysql80.cnf
+        cat <<-EOF>>./config/mycnf/mysql8026.cnf
         innodb_buffer_pool_dump_at_shutdown=OFF
         innodb_buffer_pool_in_core_file=OFF
         innodb_buffer_pool_load_at_startup=OFF

diff --git a/.github/workflows/cluster_endtoend_vtgate_vindex_heavy.yml b/.github/workflows/cluster_endtoend_vtgate_vindex_heavy.yml
@@ -137,7 +137,7 @@ jobs:
 
         # Increase our open file descriptor limit as we could hit this
         ulimit -n 65536
-        cat <<-EOF>>./config/mycnf/mysql80.cnf
+        cat <<-EOF>>./config/mycnf/mysql8026.cnf
         innodb_buffer_pool_dump_at_shutdown=OFF
         innodb_buffer_pool_in_core_file=OFF
         innodb_buffer_pool_load_at_startup=OFF

diff --git a/config/embed.go b/config/embed.go
@@ -16,3 +16,9 @@ var MycnfMySQL57 string
 
 //go:embed mycnf/mysql80.cnf
 var MycnfMySQL80 string
+
+//go:embed mycnf/mysql8026.cnf
+var MycnfMySQL8026 string
+
+//go:embed mycnf/mysql84.cnf
+var MycnfMySQL84 string
diff --git a/config/mycnf/mysql8026.cnf b/config/mycnf/mysql8026.cnf
@@ -0,0 +1,37 @@
+# This file is auto-included when MySQL 8.0.26 or later is detected.
+
+# MySQL 8.0 enables binlog by default with sync_binlog and TABLE info repositories
+# It does not enable GTIDs or enforced GTID consistency
+
+gtid_mode = ON
+enforce_gtid_consistency
+relay_log_recovery = 1
+binlog_expire_logs_seconds = 259200
+
+# disable mysqlx
+mysqlx = 0
+
+# 8.0 changes the default auth-plugin to caching_sha2_password
+default_authentication_plugin = mysql_native_password
+
+# Semi-sync replication is required for automated unplanned failover
+# (when the primary goes away). Here we just load the plugin so it's
+# available if desired, but it's disabled at startup.
+#
+# VTTablet will enable semi-sync at the proper time when replication is set up,
+# or when a primary is promoted or demoted based on the durability policy configured.
+plugin-load = rpl_semi_sync_source=semisync_source.so;rpl_semi_sync_replica=semisync_replica.so
+
+# MySQL 8.0.26 and later will not load plugins during --initialize
+# which makes these options unknown. Prefixing with --loose
+# tells the server it's fine if they are not understood.
+loose_rpl_semi_sync_source_timeout = 1000000000000000000
+loose_rpl_semi_sync_source_wait_no_replica = 1
+
+# In order to protect against any errand GTIDs we will start the mysql instance
+# in super-read-only mode.
+super-read-only
+
+# Replication parameters to ensure reparents are fast.
+replica_net_timeout = 8
+
diff --git a/config/mycnf/mysql84.cnf b/config/mycnf/mysql84.cnf
@@ -0,0 +1,39 @@
+# This file is auto-included when MySQL 8.4.0 or later is detected.
+
+# MySQL 8.0 enables binlog by default with sync_binlog and TABLE info repositories
+# It does not enable GTIDs or enforced GTID consistency
+
+gtid_mode = ON
+enforce_gtid_consistency
+relay_log_recovery = 1
+binlog_expire_logs_seconds = 259200
+
+# disable mysqlx
+mysqlx = 0
+
+# 8.4 changes the default auth-plugin to caching_sha2_password and
+# disables mysql_native_password by default.
+mysql_native_password = ON
+default_authentication_plugin = mysql_native_password
+
+# Semi-sync replication is required for automated unplanned failover
+# (when the primary goes away). Here we just load the plugin so it's
+# available if desired, but it's disabled at startup.
+#
+# VTTablet will enable semi-sync at the proper time when replication is set up,
+# or when a primary is promoted or demoted based on the durability policy configured.
+plugin-load = rpl_semi_sync_source=semisync_source.so;rpl_semi_sync_replica=semisync_replica.so
+
+# MySQL 8.0.26 and later will not load plugins during --initialize
+# which makes these options unknown. Prefixing with --loose
+# tells the server it's fine if they are not understood.
+loose_rpl_semi_sync_source_timeout = 1000000000000000000
+loose_rpl_semi_sync_source_wait_no_replica = 1
+
+# In order to protect against any errand GTIDs we will start the mysql instance
+# in super-read-only mode.
+super-read-only
+
+# Replication parameters to ensure reparents are fast.
+replica_net_timeout = 8
+
diff --git a/go/cmd/vtcombo/cli/main.go b/go/cmd/vtcombo/cli/main.go
@@ -31,6 +31,7 @@ import (
 	"github.com/spf13/cobra"
 
 	"vitess.io/vitess/go/acl"
+	"vitess.io/vitess/go/mysql"
 	"vitess.io/vitess/go/mysql/replication"
 	"vitess.io/vitess/go/stats"
 	"vitess.io/vitess/go/vt/dbconfigs"
@@ -216,7 +217,7 @@ func run(cmd *cobra.Command, args []string) (err error) {
 			return err
 		}
 		servenv.OnClose(func() {
-			ctx, cancel := context.WithTimeout(context.Background(), mysqlctl.DefaultShutdownTimeout+10*time.Second)
+			ctx, cancel := context.WithTimeout(cmd.Context(), mysqlctl.DefaultShutdownTimeout+10*time.Second)
 			defer cancel()
 			mysqld.Shutdown(ctx, cnf, true, mysqlctl.DefaultShutdownTimeout)
 		})
@@ -240,7 +241,7 @@ func run(cmd *cobra.Command, args []string) (err error) {
 	if err != nil {
 		// ensure we start mysql in the event we fail here
 		if startMysql {
-			ctx, cancel := context.WithTimeout(context.Background(), mysqlctl.DefaultShutdownTimeout+10*time.Second)
+			ctx, cancel := context.WithTimeout(cmd.Context(), mysqlctl.DefaultShutdownTimeout+10*time.Second)
 			defer cancel()
 			mysqld.Shutdown(ctx, cnf, true, mysqlctl.DefaultShutdownTimeout)
 		}
@@ -387,11 +388,11 @@ func (mysqld *vtcomboMysqld) StopReplication(hookExtraEnv map[string]string) err
 }
 
 // SetSemiSyncEnabled implements the MysqlDaemon interface
-func (mysqld *vtcomboMysqld) SetSemiSyncEnabled(source, replica bool) error {
+func (mysqld *vtcomboMysqld) SetSemiSyncEnabled(ctx context.Context, source, replica bool) error {
 	return nil
 }
 
 // SemiSyncExtensionLoaded implements the MysqlDaemon interface
-func (mysqld *vtcomboMysqld) SemiSyncExtensionLoaded() (bool, error) {
-	return true, nil
+func (mysqld *vtcomboMysqld) SemiSyncExtensionLoaded(ctx context.Context) (mysql.SemiSyncType, error) {
+	return mysql.SemiSyncTypeSource, nil
 }
diff --git a/go/mysql/flavor_mariadb.go b/go/mysql/flavor_mariadb.go
@@ -144,7 +144,8 @@ func (mariadbFlavor) resetReplicationCommands(c *Conn) []string {
 		"RESET MASTER",
 		"SET GLOBAL gtid_slave_pos = ''",
 	}
-	if c.SemiSyncExtensionLoaded() {
+	semisyncType, _ := c.SemiSyncExtensionLoaded()
+	if semisyncType == SemiSyncTypeMaster {
 		resetCommands = append(resetCommands, "SET GLOBAL rpl_semi_sync_master_enabled = false, GLOBAL rpl_semi_sync_slave_enabled = false") // semi-sync will be enabled if needed when replica is started.
 	}
 	return resetCommands

diff --git a/go/mysql/flavor_mysql.go b/go/mysql/flavor_mysql.go
@@ -154,8 +154,17 @@ func (mysqlFlavor) resetReplicationCommands(c *Conn) []string {
 		"RESET SLAVE ALL", // "ALL" makes it forget source host:port.
 		"RESET MASTER",    // This will also clear gtid_executed and gtid_purged.
 	}
-	if c.SemiSyncExtensionLoaded() {
+	status, err := c.SemiSyncExtensionLoaded()
+	if err != nil {
+		return resetCommands
+	}
+	switch status {
+	case SemiSyncTypeSource:
+		resetCommands = append(resetCommands, "SET GLOBAL rpl_semi_sync_source_enabled = false, GLOBAL rpl_semi_sync_replica_enabled = false") // semi-sync will be enabled if needed when replica is started.
+	case SemiSyncTypeMaster:
 		resetCommands = append(resetCommands, "SET GLOBAL rpl_semi_sync_master_enabled = false, GLOBAL rpl_semi_sync_slave_enabled = false") // semi-sync will be enabled if needed when replica is started.
+	default:
+		// Nothing to do.
 	}
 	return resetCommands
 }