// Open runs the invalidation loop.
func (rci *RowcacheInvalidator) Open(dbname string, mysqld mysqlctl.MysqlDaemon) {
// Perform an early check to see if we're already running.
if rci.svm.State() == sync2.SERVICE_RUNNING {
return
}
rp, err := mysqld.MasterPosition()
if err != nil {
panic(NewTabletError(ErrFatal, vtrpc.ErrorCode_INTERNAL_ERROR, "Rowcache invalidator aborting: cannot determine replication position: %v", err))
}
if mysqld.Cnf().BinLogPath == "" {
panic(NewTabletError(ErrFatal, vtrpc.ErrorCode_INTERNAL_ERROR, "Rowcache invalidator aborting: binlog path not specified"))
}
err = rci.qe.ClearRowcache(context.Background())
if err != nil {
panic(NewTabletError(ErrFatal, vtrpc.ErrorCode_INTERNAL_ERROR, "Rowcahe is not reachable"))
}
rci.dbname = dbname
rci.mysqld = mysqld
rci.SetPosition(rp)
ok := rci.svm.Go(rci.run)
if ok {
log.Infof("Rowcache invalidator starting, dbname: %s, path: %s, position: %v", dbname, mysqld.Cnf().BinLogPath, rp)
} else {
log.Infof("Rowcache invalidator already running")
}
}
func SlaveWasPromoted(ts topo.Server, mysqlDaemon mysqlctl.MysqlDaemon, tabletAlias topo.TabletAlias) error {
// We first check we don't have a master any more.
// If we do, it probably means we're not *the* master, and something
// is really wrong.
masterAddr, err := mysqlDaemon.GetMasterAddr()
if err != mysqlctl.ErrNotSlave {
return fmt.Errorf("new master is a slave: %v %v", masterAddr, err)
}
tablet, err := ts.GetTablet(tabletAlias)
if err != nil {
return err
}
return updateReplicationGraphForPromotedSlave(ts, mysqlDaemon, tablet)
}
func SlaveWasRestarted(ts topo.Server, mysqlDaemon mysqlctl.MysqlDaemon, tabletAlias topo.TabletAlias, swrd *actionnode.SlaveWasRestartedArgs) error {
tablet, err := ts.GetTablet(tabletAlias)
if err != nil {
return err
}
// check the reparent actually worked
masterAddr, err := mysqlDaemon.GetMasterAddr()
if err != nil {
return err
}
if masterAddr != swrd.ExpectedMasterAddr && masterAddr != swrd.ExpectedMasterIpAddr {
log.Errorf("SlaveWasRestarted found unexpected master %v for %v (was expecting %v or %v)", masterAddr, tabletAlias, swrd.ExpectedMasterAddr, swrd.ExpectedMasterIpAddr)
// Disabled for now
// if swrd.ContinueOnUnexpectedMaster {
// log.Errorf("ContinueOnUnexpectedMaster is set, we keep going anyway")
// } else
if swrd.ScrapStragglers {
return Scrap(ts, tablet.Alias, false)
} else {
return fmt.Errorf("Unexpected master %v for %v (was expecting %v or %v)", masterAddr, tabletAlias, swrd.ExpectedMasterAddr, swrd.ExpectedMasterIpAddr)
}
}
// Once this action completes, update authoritive tablet node first.
tablet.Parent = swrd.Parent
if tablet.Type == topo.TYPE_MASTER {
tablet.Type = topo.TYPE_SPARE
tablet.State = topo.STATE_READ_ONLY
}
err = topo.UpdateTablet(ts, tablet)
if err != nil {
return err
}
// Update the new tablet location in the replication graph now that
// we've updated the tablet.
err = topo.CreateTabletReplicationData(ts, tablet.Tablet)
if err != nil && err != topo.ErrNodeExists {
return err
}
return nil
}
// Open runs the invalidation loop.
func (rci *RowcacheInvalidator) Open(dbname string, mysqld mysqlctl.MysqlDaemon) {
rp, err := mysqld.MasterPosition()
if err != nil {
panic(NewTabletError(ErrFatal, "Rowcache invalidator aborting: cannot determine replication position: %v", err))
}
if mysqld.Cnf().BinLogPath == "" {
panic(NewTabletError(ErrFatal, "Rowcache invalidator aborting: binlog path not specified"))
}
rci.dbname = dbname
rci.mysqld = mysqld
rci.SetPosition(rp)
ok := rci.svm.Go(rci.run)
if ok {
log.Infof("Rowcache invalidator starting, dbname: %s, path: %s, position: %v", dbname, mysqld.Cnf().BinLogPath, rp)
} else {
log.Infof("Rowcache invalidator already running")
}
}
// ChecktabletMysqlPort will check the mysql port for the tablet is good,
// and if not will try to update it
func CheckTabletMysqlPort(ts topo.Server, mysqlDaemon mysqlctl.MysqlDaemon, tablet *topo.TabletInfo) *topo.TabletInfo {
mport, err := mysqlDaemon.GetMysqlPort()
if err != nil {
log.Warningf("Cannot get current mysql port, not checking it: %v", err)
return nil
}
if mport == tablet.Portmap["mysql"] {
return nil
}
log.Warningf("MySQL port has changed from %v to %v, updating it in tablet record", tablet.Portmap["mysql"], mport)
tablet.Portmap["mysql"] = mport
if err := topo.UpdateTablet(ts, tablet); err != nil {
log.Warningf("Failed to update tablet record, may use old mysql port")
return nil
}
return tablet
}
// NewActionAgent creates a new ActionAgent and registers all the
// associated services.
//
// batchCtx is the context that the agent will use for any background tasks
// it spawns.
func NewActionAgent(
batchCtx context.Context,
mysqld mysqlctl.MysqlDaemon,
queryServiceControl tabletserver.QueryServiceControl,
tabletAlias topo.TabletAlias,
dbcfgs *dbconfigs.DBConfigs,
mycnf *mysqlctl.Mycnf,
port, securePort, gRPCPort int,
overridesFile string,
lockTimeout time.Duration,
) (agent *ActionAgent, err error) {
schemaOverrides := loadSchemaOverrides(overridesFile)
topoServer := topo.GetServer()
agent = &ActionAgent{
QueryServiceControl: queryServiceControl,
HealthReporter: health.DefaultAggregator,
batchCtx: batchCtx,
TopoServer: topoServer,
TabletAlias: tabletAlias,
MysqlDaemon: mysqld,
DBConfigs: dbcfgs,
SchemaOverrides: schemaOverrides,
LockTimeout: lockTimeout,
History: history.New(historyLength),
lastHealthMapCount: stats.NewInt("LastHealthMapCount"),
_healthy: fmt.Errorf("healthcheck not run yet"),
healthStreamMap: make(map[int]chan<- *actionnode.HealthStreamReply),
}
// try to initialize the tablet if we have to
if err := agent.InitTablet(port, securePort, gRPCPort); err != nil {
return nil, fmt.Errorf("agent.InitTablet failed: %v", err)
}
// Publish and set the TargetTabletType. Not a global var
// since it should never be changed.
statsTabletType := stats.NewString("TargetTabletType")
statsTabletType.Set(*targetTabletType)
// Start the binlog player services, not playing at start.
agent.BinlogPlayerMap = NewBinlogPlayerMap(topoServer, &dbcfgs.Filtered, mysqld)
RegisterBinlogPlayerMap(agent.BinlogPlayerMap)
// try to figure out the mysql port
mysqlPort := mycnf.MysqlPort
if mysqlPort == 0 {
// we don't know the port, try to get it from mysqld
var err error
mysqlPort, err = mysqld.GetMysqlPort()
if err != nil {
log.Warningf("Cannot get current mysql port, will use 0 for now: %v", err)
}
}
if err := agent.Start(batchCtx, mysqlPort, port, securePort, gRPCPort); err != nil {
return nil, err
}
// register the RPC services from the agent
agent.registerQueryService()
// two cases then:
// - restoreFromBackup is set: we restore, then initHealthCheck, all
// in the background
// - restoreFromBackup is not set: we initHealthCheck right away
if *restoreFromBackup {
go func() {
// restoreFromBackup wil just be a regular action
// (same as if it was triggered remotely)
if err := agent.RestoreFromBackup(batchCtx); err != nil {
println(fmt.Sprintf("RestoreFromBackup failed: %v", err))
log.Fatalf("RestoreFromBackup failed: %v", err)
}
// after the restore is done, start health check
agent.initHeathCheck()
}()
} else {
// synchronously start health check if needed
agent.initHeathCheck()
}
return agent, nil
}
// NewActionAgent creates a new ActionAgent and registers all the
// associated services.
//
// batchCtx is the context that the agent will use for any background tasks
// it spawns.
func NewActionAgent(
batchCtx context.Context,
mysqld mysqlctl.MysqlDaemon,
queryServiceControl tabletserver.Controller,
tabletAlias *topodatapb.TabletAlias,
dbcfgs dbconfigs.DBConfigs,
mycnf *mysqlctl.Mycnf,
port, gRPCPort int32,
overridesFile string,
) (agent *ActionAgent, err error) {
schemaOverrides := loadSchemaOverrides(overridesFile)
topoServer := topo.GetServer()
agent = &ActionAgent{
QueryServiceControl: queryServiceControl,
HealthReporter: health.DefaultAggregator,
batchCtx: batchCtx,
TopoServer: topoServer,
TabletAlias: tabletAlias,
MysqlDaemon: mysqld,
DBConfigs: dbcfgs,
SchemaOverrides: schemaOverrides,
History: history.New(historyLength),
lastHealthMapCount: stats.NewInt("LastHealthMapCount"),
_healthy: fmt.Errorf("healthcheck not run yet"),
}
agent.registerQueryRuleSources()
// try to initialize the tablet if we have to
if err := agent.InitTablet(port, gRPCPort); err != nil {
return nil, fmt.Errorf("agent.InitTablet failed: %v", err)
}
// Publish and set the TargetTabletType. Not a global var
// since it should never be changed.
statsTabletType := stats.NewString("TargetTabletType")
statsTabletType.Set(*targetTabletType)
// Create the TabletType stats
agent.exportStats = true
agent.statsTabletType = stats.NewString("TabletType")
// Start the binlog player services, not playing at start.
agent.BinlogPlayerMap = NewBinlogPlayerMap(topoServer, mysqld, func() binlogplayer.VtClient {
return binlogplayer.NewDbClient(&agent.DBConfigs.Filtered)
})
// Stop all binlog players upon entering lameduck.
servenv.OnTerm(agent.BinlogPlayerMap.StopAllPlayersAndReset)
RegisterBinlogPlayerMap(agent.BinlogPlayerMap)
// try to figure out the mysql port
mysqlPort := mycnf.MysqlPort
if mysqlPort == 0 {
// we don't know the port, try to get it from mysqld
var err error
mysqlPort, err = mysqld.GetMysqlPort()
if err != nil {
log.Warningf("Cannot get current mysql port, will use 0 for now: %v", err)
}
}
// Start will get the tablet info, and update our state from it
if err := agent.Start(batchCtx, int32(mysqlPort), port, gRPCPort, true); err != nil {
return nil, err
}
// register the RPC services from the agent
agent.registerQueryService()
// two cases then:
// - restoreFromBackup is set: we restore, then initHealthCheck, all
// in the background
// - restoreFromBackup is not set: we initHealthCheck right away
if *restoreFromBackup {
go func() {
// restoreFromBackup wil just be a regular action
// (same as if it was triggered remotely)
if err := agent.RestoreFromBackup(batchCtx); err != nil {
println(fmt.Sprintf("RestoreFromBackup failed: %v", err))
log.Fatalf("RestoreFromBackup failed: %v", err)
}
// after the restore is done, start health check
agent.initHealthCheck()
}()
} else {
// synchronously start health check if needed
agent.initHealthCheck()
}
return agent, nil
}
// NewActionAgent creates a new ActionAgent and registers all the
// associated services.
//
// batchCtx is the context that the agent will use for any background tasks
// it spawns.
func NewActionAgent(
batchCtx context.Context,
mysqld mysqlctl.MysqlDaemon,
queryServiceControl tabletserver.Controller,
tabletAlias *topodatapb.TabletAlias,
dbcfgs dbconfigs.DBConfigs,
mycnf *mysqlctl.Mycnf,
port, gRPCPort int32,
) (agent *ActionAgent, err error) {
topoServer := topo.GetServer()
orc, err := newOrcClient()
if err != nil {
return nil, err
}
agent = &ActionAgent{
QueryServiceControl: queryServiceControl,
HealthReporter: health.DefaultAggregator,
batchCtx: batchCtx,
TopoServer: topoServer,
TabletAlias: tabletAlias,
MysqlDaemon: mysqld,
DBConfigs: dbcfgs,
History: history.New(historyLength),
_healthy: fmt.Errorf("healthcheck not run yet"),
orc: orc,
}
agent.registerQueryRuleSources()
// try to initialize the tablet if we have to
if err := agent.InitTablet(port, gRPCPort); err != nil {
return nil, fmt.Errorf("agent.InitTablet failed: %v", err)
}
// Create the TabletType stats
agent.exportStats = true
agent.statsTabletType = stats.NewString("TabletType")
// Start the binlog player services, not playing at start.
agent.BinlogPlayerMap = NewBinlogPlayerMap(topoServer, mysqld, func() binlogplayer.VtClient {
return binlogplayer.NewDbClient(&agent.DBConfigs.Filtered)
})
// Stop all binlog players upon entering lameduck.
servenv.OnTerm(agent.BinlogPlayerMap.StopAllPlayersAndReset)
RegisterBinlogPlayerMap(agent.BinlogPlayerMap)
// try to figure out the mysql port
mysqlPort := mycnf.MysqlPort
if mysqlPort == 0 {
// we don't know the port, try to get it from mysqld
var err error
mysqlPort, err = mysqld.GetMysqlPort()
if err != nil {
log.Warningf("Cannot get current mysql port, will use 0 for now: %v", err)
}
}
// Start will get the tablet info, and update our state from it
if err := agent.Start(batchCtx, int32(mysqlPort), port, gRPCPort, true); err != nil {
return nil, err
}
// register the RPC services from the agent
servenv.OnRun(func() {
agent.registerQueryService()
})
// two cases then:
// - restoreFromBackup is set: we restore, then initHealthCheck, all
// in the background
// - restoreFromBackup is not set: we initHealthCheck right away
if *restoreFromBackup {
go func() {
// restoreFromBackup wil just be a regular action
// (same as if it was triggered remotely)
if err := agent.RestoreData(batchCtx, logutil.NewConsoleLogger(), false /* deleteBeforeRestore */); err != nil {
println(fmt.Sprintf("RestoreFromBackup failed: %v", err))
log.Fatalf("RestoreFromBackup failed: %v", err)
}
// after the restore is done, start health check
agent.initHealthCheck()
}()
} else {
// update our state
if err := agent.refreshTablet(batchCtx, "Start"); err != nil {
return nil, err
}
// synchronously start health check if needed
agent.initHealthCheck()
}
// Start periodic Orchestrator self-registration, if configured.
if agent.orc != nil {
go agent.orc.DiscoverLoop(agent)
}
return agent, nil
}