func (agent *ActionAgent) initHealthCheck() {
if !agent.IsRunningHealthCheck() {
log.Infof("No target_tablet_type specified, disabling any health check")
return
}
tt, err := topoproto.ParseTabletType(*targetTabletType)
if err != nil {
log.Fatalf("Invalid target tablet type %v: %v", *targetTabletType, err)
}
log.Infof("Starting periodic health check every %v with target_tablet_type=%v", *healthCheckInterval, *targetTabletType)
t := timer.NewTimer(*healthCheckInterval)
servenv.OnTermSync(func() {
// When we enter lameduck mode, we want to not call
// the health check any more. After this returns, we
// are guaranteed to not call it.
log.Info("Stopping periodic health check timer")
t.Stop()
// Now we can finish up and force ourselves to not healthy.
agent.terminateHealthChecks(tt)
})
t.Start(func() {
agent.runHealthCheck(tt)
})
t.Trigger()
}
func main() {
defer exit.Recover()
flag.Parse()
servenv.Init()
ts := topo.GetServer()
defer topo.CloseServers()
resilientSrvTopoServer = vtgate.NewResilientSrvTopoServer(ts, "ResilientSrvTopoServer")
healthCheck = discovery.NewHealthCheck(*healthCheckConnTimeout, *healthCheckRetryDelay, *healthCheckTimeout)
healthCheck.RegisterStats()
tabletTypes := make([]topodatapb.TabletType, 0, 1)
if len(*tabletTypesToWait) != 0 {
for _, ttStr := range strings.Split(*tabletTypesToWait, ",") {
tt, err := topoproto.ParseTabletType(ttStr)
if err != nil {
log.Errorf("unknown tablet type: %v", ttStr)
continue
}
tabletTypes = append(tabletTypes, tt)
}
}
l2vtg := l2vtgate.Init(healthCheck, ts, resilientSrvTopoServer, *cell, *retryCount, tabletTypes)
servenv.OnRun(func() {
addStatusParts(l2vtg)
})
servenv.RunDefault()
}
// TabletTypeToProto turns a TabletType into a proto
func TabletTypeToProto(t TabletType) pb.TabletType {
if result, err := topoproto.ParseTabletType(string(t)); err != nil {
panic(fmt.Errorf("unknown tablet type: %v", t))
} else {
return result
}
}
func main() {
defer exit.Recover()
flag.Parse()
servenv.Init()
if initFakeZK != nil {
initFakeZK()
}
ts := topo.GetServer()
defer topo.CloseServers()
resilientSrvTopoServer = vtgate.NewResilientSrvTopoServer(ts, "ResilientSrvTopoServer")
healthCheck = discovery.NewHealthCheck(*connTimeoutTotal, *healthCheckRetryDelay, *healthCheckTimeout, "" /* statsSuffix */)
tabletTypes := make([]topodatapb.TabletType, 0, 1)
if len(*tabletTypesToWait) != 0 {
for _, ttStr := range strings.Split(*tabletTypesToWait, ",") {
tt, err := topoproto.ParseTabletType(ttStr)
if err != nil {
log.Errorf("unknown tablet type: %v", ttStr)
continue
}
tabletTypes = append(tabletTypes, tt)
}
}
vtg := vtgate.Init(context.Background(), healthCheck, ts, resilientSrvTopoServer, *cell, *retryDelay, *retryCount, *connTimeoutTotal, *connTimeoutPerConn, *connLife, tabletTypes, *maxInFlight, *testGateway)
servenv.OnRun(func() {
addStatusParts(vtg)
})
servenv.RunDefault()
}
// tabletTypesLocked returns the tablet types needed to be displayed in the heatmap based on the dropdown filters.
// It returns tablet type if a specific one was chosen or returns all of them if 'all' is chosen for keyspace and/or cell.
// This method is used by heatmapData to traverse over the desired tablet types.
func (c *tabletStatsCache) tabletTypesLocked(keyspace, cell, tabletType string) []topodata.TabletType {
if tabletType != "all" {
tabletTypeObj, _ := topoproto.ParseTabletType(tabletType)
return []topodata.TabletType{tabletTypeObj}
}
return c.typesInTopology(keyspace, cell)
}
func main() {
defer exit.Recover()
flag.Parse()
servenv.Init()
if initFakeZK != nil {
initFakeZK()
}
ts := topo.GetServer()
defer topo.CloseServers()
var schema *planbuilder.Schema
if *schemaFile != "" {
var err error
if schema, err = planbuilder.LoadFile(*schemaFile); err != nil {
log.Error(err)
exit.Return(1)
}
log.Infof("v3 is enabled: loaded schema from file: %v", *schemaFile)
} else {
ctx := context.Background()
schemaJSON, err := ts.GetVSchema(ctx)
if err != nil {
log.Warningf("Skipping v3 initialization: GetVSchema failed: %v", err)
goto startServer
}
schema, err = planbuilder.NewSchema([]byte(schemaJSON))
if err != nil {
log.Warningf("Skipping v3 initialization: GetVSchema failed: %v", err)
goto startServer
}
log.Infof("v3 is enabled: loaded schema from topo")
}
startServer:
resilientSrvTopoServer = vtgate.NewResilientSrvTopoServer(ts, "ResilientSrvTopoServer")
healthCheck = discovery.NewHealthCheck(*connTimeoutTotal, *healthCheckRetryDelay, *healthCheckTimeout, "" /* statsSuffix */)
tabletTypes := make([]topodatapb.TabletType, 0, 1)
if len(*tabletTypesToWait) != 0 {
for _, ttStr := range strings.Split(*tabletTypesToWait, ",") {
tt, err := topoproto.ParseTabletType(ttStr)
if err != nil {
log.Errorf("unknown tablet type: %v", ttStr)
continue
}
tabletTypes = append(tabletTypes, tt)
}
}
vtg := vtgate.Init(healthCheck, ts, resilientSrvTopoServer, schema, *cell, *retryDelay, *retryCount, *connTimeoutTotal, *connTimeoutPerConn, *connLife, tabletTypes, *maxInFlight, *testGateway)
servenv.OnRun(func() {
addStatusParts(vtg)
})
servenv.RunDefault()
}
// handleExplorerRedirect returns the redirect target URL.
func handleExplorerRedirect(r *http.Request) (string, error) {
keyspace := r.FormValue("keyspace")
shard := r.FormValue("shard")
cell := r.FormValue("cell")
switch r.FormValue("type") {
case "keyspace":
if keyspace == "" {
return "", errors.New("keyspace is required for this redirect")
}
return explorer.GetKeyspacePath(keyspace), nil
case "shard":
if keyspace == "" || shard == "" {
return "", errors.New("keyspace and shard are required for this redirect")
}
return explorer.GetShardPath(keyspace, shard), nil
case "srv_keyspace":
if keyspace == "" || cell == "" {
return "", errors.New("keyspace and cell are required for this redirect")
}
return explorer.GetSrvKeyspacePath(cell, keyspace), nil
case "srv_shard":
if keyspace == "" || shard == "" || cell == "" {
return "", errors.New("keyspace, shard, and cell are required for this redirect")
}
return explorer.GetSrvShardPath(cell, keyspace, shard), nil
case "srv_type":
tabletType := r.FormValue("tablet_type")
if keyspace == "" || shard == "" || cell == "" || tabletType == "" {
return "", errors.New("keyspace, shard, cell, and tablet_type are required for this redirect")
}
tt, err := topoproto.ParseTabletType(tabletType)
if err != nil {
return "", fmt.Errorf("cannot parse tablet type %v: %v", tabletType, err)
}
return explorer.GetSrvTypePath(cell, keyspace, shard, tt), nil
case "tablet":
alias := r.FormValue("alias")
if alias == "" {
return "", errors.New("alias is required for this redirect")
}
tabletAlias, err := topoproto.ParseTabletAlias(alias)
if err != nil {
return "", fmt.Errorf("bad tablet alias %q: %v", alias, err)
}
return explorer.GetTabletPath(tabletAlias), nil
case "replication":
if keyspace == "" || shard == "" || cell == "" {
return "", errors.New("keyspace, shard, and cell are required for this redirect")
}
return explorer.GetReplicationSlaves(cell, keyspace, shard), nil
default:
return "", errors.New("bad redirect type")
}
}
func commandVtTabletExecute(ctx context.Context, wr *wrangler.Wrangler, subFlags *flag.FlagSet, args []string) error {
transactionID := subFlags.Int("transaction_id", 0, "transaction id to use, if inside a transaction.")
bindVariables := newBindvars(subFlags)
keyspace := subFlags.String("keyspace", "", "keyspace the tablet belongs to")
shard := subFlags.String("shard", "", "shard the tablet belongs to")
tabletType := subFlags.String("tablet_type", "unknown", "tablet type we expect from the tablet (use unknown to use sessionId)")
connectTimeout := subFlags.Duration("connect_timeout", 30*time.Second, "Connection timeout for vttablet client")
json := subFlags.Bool("json", false, "Output JSON instead of human-readable table")
if err := subFlags.Parse(args); err != nil {
return err
}
if subFlags.NArg() != 2 {
return fmt.Errorf("the <tablet_alias> and <sql> arguments are required for the VtTabletExecute command")
}
tt, err := topoproto.ParseTabletType(*tabletType)
if err != nil {
return err
}
tabletAlias, err := topoproto.ParseTabletAlias(subFlags.Arg(0))
if err != nil {
return err
}
tabletInfo, err := wr.TopoServer().GetTablet(ctx, tabletAlias)
if err != nil {
return err
}
ep, err := topo.TabletEndPoint(tabletInfo.Tablet)
if err != nil {
return fmt.Errorf("cannot get EndPoint from tablet record: %v", err)
}
conn, err := tabletconn.GetDialer()(ctx, ep, *keyspace, *shard, tt, *connectTimeout)
if err != nil {
return fmt.Errorf("cannot connect to tablet %v: %v", tabletAlias, err)
}
defer conn.Close()
qr, err := conn.Execute(ctx, subFlags.Arg(1), *bindVariables, int64(*transactionID))
if err != nil {
return fmt.Errorf("Execute failed: %v", err)
}
if *json {
return printJSON(wr.Logger(), qr)
}
printQueryResult(loggerWriter{wr.Logger()}, qr)
return nil
}
// Open must be called with a JSON string that looks like this:
// {"protocol": "gorpc", "address": "localhost:1111", "tablet_type": "master", "timeout": 1000000000}
// protocol specifies the rpc protocol to use.
// address specifies the address for the VTGate to connect to.
// tablet_type represents the consistency level of your operations.
// For example "replica" means eventually consistent reads, while
// "master" supports transactions and gives you read-after-write consistency.
// timeout is specified in nanoseconds. It applies for all operations.
func (d drv) Open(name string) (driver.Conn, error) {
c := &conn{TabletType: "master"}
err := json.Unmarshal([]byte(name), c)
if err != nil {
return nil, err
}
c.tabletType, err = topoproto.ParseTabletType(c.TabletType)
if err != nil {
return nil, err
}
err = c.dial()
if err != nil {
return nil, err
}
return c, nil
}
func (agent *ActionAgent) isMasterEligible() (bool, error) {
switch agent.Tablet().Type {
case topodatapb.TabletType_MASTER, topodatapb.TabletType_REPLICA:
return true, nil
case topodatapb.TabletType_SPARE:
// If we're SPARE, it could be because healthcheck is enabled.
tt, err := topoproto.ParseTabletType(*targetTabletType)
if err != nil {
return false, fmt.Errorf("can't determine if tablet is master-eligible: currently SPARE and no -target_tablet_type flag specified")
}
if tt == topodatapb.TabletType_REPLICA {
return true, nil
}
}
return false, nil
}
func commandVtTabletBegin(ctx context.Context, wr *wrangler.Wrangler, subFlags *flag.FlagSet, args []string) error {
keyspace := subFlags.String("keyspace", "", "keyspace the tablet belongs to")
shard := subFlags.String("shard", "", "shard the tablet belongs to")
tabletType := subFlags.String("tablet_type", "unknown", "tablet type we expect from the tablet (use unknown to use sessionId)")
connectTimeout := subFlags.Duration("connect_timeout", 30*time.Second, "Connection timeout for vttablet client")
if err := subFlags.Parse(args); err != nil {
return err
}
if subFlags.NArg() != 1 {
return fmt.Errorf("the <tablet_alias> argument is required for the VtTabletBegin command")
}
tt, err := topoproto.ParseTabletType(*tabletType)
if err != nil {
return err
}
tabletAlias, err := topoproto.ParseTabletAlias(subFlags.Arg(0))
if err != nil {
return err
}
tabletInfo, err := wr.TopoServer().GetTablet(ctx, tabletAlias)
if err != nil {
return err
}
ep, err := topo.TabletEndPoint(tabletInfo.Tablet)
if err != nil {
return fmt.Errorf("cannot get EndPoint from tablet record: %v", err)
}
conn, err := tabletconn.GetDialer()(ctx, ep, *keyspace, *shard, tt, *connectTimeout)
if err != nil {
return fmt.Errorf("cannot connect to tablet %v: %v", tabletAlias, err)
}
defer conn.Close()
transactionID, err := conn.Begin(ctx)
if err != nil {
return fmt.Errorf("Begin failed: %v", err)
}
result := map[string]int64{
"transaction_id": transactionID,
}
return printJSON(wr, result)
}
// GetSrvTabletTypesPerShard is part of the topo.Server interface
func (zkts *Server) GetSrvTabletTypesPerShard(ctx context.Context, cell, keyspace, shard string) ([]topodatapb.TabletType, error) {
zkSgShardPath := zkPathForVtShard(cell, keyspace, shard)
children, _, err := zkts.zconn.Children(zkSgShardPath)
if err != nil {
if zookeeper.IsError(err, zookeeper.ZNONODE) {
err = topo.ErrNoNode
}
return nil, err
}
result := make([]topodatapb.TabletType, 0, len(children))
for _, tt := range children {
// these two are used for locking
if tt == "action" || tt == "actionlog" {
continue
}
if ptt, err := topoproto.ParseTabletType(tt); err == nil {
result = append(result, ptt)
}
}
return result, nil
}
// Open implements the database/sql/driver.Driver interface.
//
// For "name", the Vitess driver requires that a JSON object is passed in.
//
// Instead of using this call and passing in a hand-crafted JSON string, it's
// recommended to use the public Vitess helper functions like
// Open(), OpenShard() or OpenWithConfiguration() instead. These will generate
// the required JSON string behind the scenes for you.
//
// Example for a JSON string:
//
// {"protocol": "grpc", "address": "localhost:1111", "tablet_type": "master", "timeout": 1000000000}
//
// For a description of the available fields, see the Configuration struct.
// Note: In the JSON string, timeout has to be specified in nanoseconds.
func (d drv) Open(name string) (driver.Conn, error) {
c := &conn{Configuration: newDefaultConfiguration()}
err := json.Unmarshal([]byte(name), c)
if err != nil {
return nil, err
}
if c.Keyspace == "" && c.Shard != "" {
return nil, fmt.Errorf("the shard parameter requires a keyspace parameter. shard: %v", c.Shard)
}
if c.useExecuteShards() {
log.Infof("Sending queries only to keyspace/shard: %v/%v", c.Keyspace, c.Shard)
}
c.tabletTypeProto, err = topoproto.ParseTabletType(c.TabletType)
if err != nil {
return nil, err
}
err = c.dial()
if err != nil {
return nil, err
}
return c, nil
}
// Open implements the database/sql/driver.Driver interface.
//
// For "name", the Vitess driver requires that a JSON object is passed in.
//
// Instead of using this call and passing in a hand-crafted JSON string, it's
// recommended to use the public Vitess helper functions like
// Open(), OpenShard() or OpenWithConfiguration() instead. These will generate
// the required JSON string behind the scenes for you.
//
// Example for a JSON string:
//
// {"protocol": "gorpc", "address": "localhost:1111", "tablet_type": "master", "timeout": 1000000000}
//
// For a description of the available fields, see the Configuration struct.
// Note: In the JSON string, timeout has to be specified in nanoseconds.
func (d drv) Open(name string) (driver.Conn, error) {
c := &conn{Configuration: newDefaultConfiguration()}
err := json.Unmarshal([]byte(name), c)
if err != nil {
return nil, err
}
if (c.Keyspace != "" && c.Shard == "") || (c.Keyspace == "" && c.Shard != "") {
return nil, fmt.Errorf("Always set both keyspace and shard or leave both empty. keyspace: %v shard: %v", c.Keyspace, c.Shard)
}
if c.useExecuteShards() {
log.Infof("Sending queries only to keyspace/shard: %v/%v", c.Keyspace, c.Shard)
}
c.tabletTypeProto, err = topoproto.ParseTabletType(c.TabletType)
if err != nil {
return nil, err
}
err = c.dial()
if err != nil {
return nil, err
}
return c, nil
}
// GetSrvTabletTypesPerShard implements topo.Server.
func (s *Server) GetSrvTabletTypesPerShard(ctx context.Context, cellName, keyspace, shard string) ([]topodatapb.TabletType, error) {
cell, err := s.getCell(cellName)
if err != nil {
return nil, err
}
resp, err := cell.Get(srvShardDirPath(keyspace, shard), false /* sort */, false /* recursive */)
if err != nil {
return nil, convertError(err)
}
if resp.Node == nil {
return nil, ErrBadResponse
}
tabletTypes := make([]topodatapb.TabletType, 0, len(resp.Node.Nodes))
for _, n := range resp.Node.Nodes {
strType := path.Base(n.Key)
if tt, err := topoproto.ParseTabletType(strType); err == nil {
tabletTypes = append(tabletTypes, tt)
}
}
return tabletTypes, nil
}
// InitTablet initializes the tablet record if necessary.
func (agent *ActionAgent) InitTablet(port, gRPCPort int32) error {
// only enabled if one of init_tablet_type (when healthcheck
// is disabled) or init_keyspace (when healthcheck is enabled)
// is passed in, then check other parameters
if *initTabletType == "" && *initKeyspace == "" {
return nil
}
// figure out our default target type
var tabletType topodatapb.TabletType
if *initTabletType != "" {
if *targetTabletType != "" {
log.Fatalf("cannot specify both target_tablet_type and init_tablet_type parameters (as they might conflict)")
}
// use the type specified on the command line
var err error
tabletType, err = topoproto.ParseTabletType(*initTabletType)
if err != nil {
log.Fatalf("Invalid init tablet type %v: %v", *initTabletType, err)
}
if tabletType == topodatapb.TabletType_MASTER {
// We disallow MASTER, so we don't have to change
// shard.MasterAlias, and deal with the corner cases.
log.Fatalf("init_tablet_type cannot be %v", tabletType)
}
} else if *targetTabletType != "" {
if strings.ToUpper(*targetTabletType) == topodatapb.TabletType_name[int32(topodatapb.TabletType_MASTER)] {
log.Fatalf("target_tablet_type cannot be '%v'. Use '%v' instead.", tabletType, topodatapb.TabletType_REPLICA)
}
// use spare, the healthcheck will turn us into what
// we need to be eventually
tabletType = topodatapb.TabletType_SPARE
} else {
log.Fatalf("if init tablet is enabled, one of init_tablet_type or target_tablet_type needs to be specified")
}
// create a context for this whole operation
ctx, cancel := context.WithTimeout(agent.batchCtx, *initTimeout)
defer cancel()
// since we're assigned to a shard, make sure it exists, see if
// we are its master, and update its cells list if necessary
if *initKeyspace == "" || *initShard == "" {
log.Fatalf("if init tablet is enabled and the target type is not idle, init_keyspace and init_shard also need to be specified")
}
shard, _, err := topo.ValidateShardName(*initShard)
if err != nil {
log.Fatalf("cannot validate shard name: %v", err)
}
log.Infof("Reading shard record %v/%v", *initKeyspace, shard)
// read the shard, create it if necessary
si, err := topotools.GetOrCreateShard(ctx, agent.TopoServer, *initKeyspace, shard)
if err != nil {
return fmt.Errorf("InitTablet cannot GetOrCreateShard shard: %v", err)
}
if si.MasterAlias != nil && topoproto.TabletAliasEqual(si.MasterAlias, agent.TabletAlias) {
// we are the current master for this shard (probably
// means the master tablet process was just restarted),
// so InitTablet as master.
tabletType = topodatapb.TabletType_MASTER
}
// See if we need to add the tablet's cell to the shard's cell
// list. If we do, it has to be under the shard lock.
if !si.HasCell(agent.TabletAlias.Cell) {
actionNode := actionnode.UpdateShard()
lockPath, err := actionNode.LockShard(ctx, agent.TopoServer, *initKeyspace, shard)
if err != nil {
return fmt.Errorf("LockShard(%v/%v) failed: %v", *initKeyspace, shard, err)
}
// re-read the shard with the lock
si, err = agent.TopoServer.GetShard(ctx, *initKeyspace, shard)
if err != nil {
return actionNode.UnlockShard(ctx, agent.TopoServer, *initKeyspace, shard, lockPath, err)
}
// see if we really need to update it now
if !si.HasCell(agent.TabletAlias.Cell) {
si.Cells = append(si.Cells, agent.TabletAlias.Cell)
// write it back
if err := agent.TopoServer.UpdateShard(ctx, si); err != nil {
return actionNode.UnlockShard(ctx, agent.TopoServer, *initKeyspace, shard, lockPath, err)
}
}
// and unlock
if err := actionNode.UnlockShard(ctx, agent.TopoServer, *initKeyspace, shard, lockPath, nil); err != nil {
return err
}
}
log.Infof("Initializing the tablet for type %v", tabletType)
// figure out the hostname
hostname := *tabletHostname
if hostname != "" {
log.Infof("Using hostname: %v from -tablet_hostname flag.", hostname)
} else {
hostname, err := netutil.FullyQualifiedHostname()
if err != nil {
return err
}
log.Infof("Using detected machine hostname: %v To change this, fix your machine network configuration or override it with -tablet_hostname.", hostname)
}
// create and populate tablet record
tablet := &topodatapb.Tablet{
Alias: agent.TabletAlias,
Hostname: hostname,
PortMap: make(map[string]int32),
Keyspace: *initKeyspace,
Shard: *initShard,
Type: tabletType,
DbNameOverride: *initDbNameOverride,
Tags: initTags,
}
if port != 0 {
tablet.PortMap["vt"] = port
}
if gRPCPort != 0 {
tablet.PortMap["grpc"] = gRPCPort
}
if err := topo.TabletComplete(tablet); err != nil {
return fmt.Errorf("InitTablet TabletComplete failed: %v", err)
}
// now try to create the record
err = agent.TopoServer.CreateTablet(ctx, tablet)
switch err {
case nil:
// it worked, we're good, can update the replication graph
if err := topo.UpdateTabletReplicationData(ctx, agent.TopoServer, tablet); err != nil {
return fmt.Errorf("UpdateTabletReplicationData failed: %v", err)
}
case topo.ErrNodeExists:
// The node already exists, will just try to update
// it. So we read it first.
oldTablet, err := agent.TopoServer.GetTablet(ctx, tablet.Alias)
if err != nil {
return fmt.Errorf("InitTablet failed to read existing tablet record: %v", err)
}
// Sanity check the keyspace and shard
if oldTablet.Keyspace != tablet.Keyspace || oldTablet.Shard != tablet.Shard {
return fmt.Errorf("InitTablet failed because existing tablet keyspace and shard %v/%v differ from the provided ones %v/%v", oldTablet.Keyspace, oldTablet.Shard, tablet.Keyspace, tablet.Shard)
}
// And overwrite the rest
*(oldTablet.Tablet) = *tablet
if err := agent.TopoServer.UpdateTablet(ctx, oldTablet); err != nil {
return fmt.Errorf("UpdateTablet failed: %v", err)
}
// Note we don't need to UpdateTabletReplicationData
// as the tablet already existed with the right data
// in the replication graph
default:
return fmt.Errorf("CreateTablet failed: %v", err)
}
// and now update the serving graph. Note we do that in any case,
// to clean any inaccurate record from any part of the serving graph.
if err := topotools.UpdateTabletEndpoints(ctx, agent.TopoServer, tablet); err != nil {
return fmt.Errorf("UpdateTabletEndpoints failed: %v", err)
}
return nil
}
// InitTablet initializes the tablet record if necessary.
func (agent *ActionAgent) InitTablet(port, gRPCPort int32) error {
// it should be either we have all three of init_keyspace,
// init_shard and init_tablet_type, or none.
if *initKeyspace == "" && *initShard == "" && *initTabletType == "" {
// not initializing the record
return nil
}
if *initKeyspace == "" || *initShard == "" || *initTabletType == "" {
return fmt.Errorf("either need all of init_keyspace, init_shard and init_tablet_type, or none")
}
// parse init_tablet_type
tabletType, err := topoproto.ParseTabletType(*initTabletType)
if err != nil {
return fmt.Errorf("invalid init_tablet_type %v: %v", *initTabletType, err)
}
if tabletType == topodatapb.TabletType_MASTER {
// We disallow MASTER, so we don't have to change
// shard.MasterAlias, and deal with the corner cases.
return fmt.Errorf("init_tablet_type cannot be master, use replica instead")
}
// parse and validate shard name
shard, _, err := topo.ValidateShardName(*initShard)
if err != nil {
return fmt.Errorf("cannot validate shard name %v: %v", *initShard, err)
}
// create a context for this whole operation
ctx, cancel := context.WithTimeout(agent.batchCtx, *initTimeout)
defer cancel()
// read the shard, create it if necessary
log.Infof("Reading shard record %v/%v", *initKeyspace, shard)
si, err := agent.TopoServer.GetOrCreateShard(ctx, *initKeyspace, shard)
if err != nil {
return fmt.Errorf("InitTablet cannot GetOrCreateShard shard: %v", err)
}
if si.MasterAlias != nil && topoproto.TabletAliasEqual(si.MasterAlias, agent.TabletAlias) {
// We're marked as master in the shard record, which could mean the master
// tablet process was just restarted. However, we need to check if a new
// master is in the process of taking over. In that case, it will let us
// know by forcibly updating the old master's tablet record.
oldTablet, err := agent.TopoServer.GetTablet(ctx, agent.TabletAlias)
switch err {
case topo.ErrNoNode:
// There's no existing tablet record, so we can assume
// no one has left us a message to step down.
tabletType = topodatapb.TabletType_MASTER
case nil:
if oldTablet.Type == topodatapb.TabletType_MASTER {
// We're marked as master in the shard record,
// and our existing tablet record agrees.
tabletType = topodatapb.TabletType_MASTER
}
default:
return fmt.Errorf("InitTablet failed to read existing tablet record: %v", err)
}
}
// See if we need to add the tablet's cell to the shard's cell list.
if !si.HasCell(agent.TabletAlias.Cell) {
si, err = agent.TopoServer.UpdateShardFields(ctx, *initKeyspace, shard, func(si *topo.ShardInfo) error {
if si.HasCell(agent.TabletAlias.Cell) {
// Someone else already did it.
return topo.ErrNoUpdateNeeded
}
si.Cells = append(si.Cells, agent.TabletAlias.Cell)
return nil
})
if err != nil {
return fmt.Errorf("couldn't add tablet's cell to shard record: %v", err)
}
}
log.Infof("Initializing the tablet for type %v", tabletType)
// figure out the hostname
hostname := *tabletHostname
if hostname != "" {
log.Infof("Using hostname: %v from -tablet_hostname flag.", hostname)
} else {
hostname, err := netutil.FullyQualifiedHostname()
if err != nil {
return err
}
log.Infof("Using detected machine hostname: %v To change this, fix your machine network configuration or override it with -tablet_hostname.", hostname)
}
// create and populate tablet record
tablet := &topodatapb.Tablet{
Alias: agent.TabletAlias,
Hostname: hostname,
PortMap: make(map[string]int32),
Keyspace: *initKeyspace,
Shard: *initShard,
Type: tabletType,
DbNameOverride: *initDbNameOverride,
Tags: initTags,
}
if port != 0 {
tablet.PortMap["vt"] = port
}
if gRPCPort != 0 {
tablet.PortMap["grpc"] = gRPCPort
}
if err := topo.TabletComplete(tablet); err != nil {
return fmt.Errorf("InitTablet TabletComplete failed: %v", err)
}
// Now try to create the record (it will also fix up the
// ShardReplication record if necessary).
err = agent.TopoServer.CreateTablet(ctx, tablet)
switch err {
case nil:
// It worked, we're good.
case topo.ErrNodeExists:
// The node already exists, will just try to update
// it. So we read it first.
oldTablet, err := agent.TopoServer.GetTablet(ctx, tablet.Alias)
if err != nil {
return fmt.Errorf("InitTablet failed to read existing tablet record: %v", err)
}
// Sanity check the keyspace and shard
if oldTablet.Keyspace != tablet.Keyspace || oldTablet.Shard != tablet.Shard {
return fmt.Errorf("InitTablet failed because existing tablet keyspace and shard %v/%v differ from the provided ones %v/%v", oldTablet.Keyspace, oldTablet.Shard, tablet.Keyspace, tablet.Shard)
}
// Then overwrite everything, ignoring version mismatch.
if err := agent.TopoServer.UpdateTablet(ctx, topo.NewTabletInfo(tablet, -1)); err != nil {
return fmt.Errorf("UpdateTablet failed: %v", err)
}
default:
return fmt.Errorf("CreateTablet failed: %v", err)
}
return nil
}
func (wr *Wrangler) exportVtnsToZkns(ctx context.Context, zconn zk.Conn, vtnsAddrPath, zknsAddrPath string) ([]string, error) {
zknsPaths := make([]string, 0, 32)
parts := strings.Split(vtnsAddrPath, "/")
if len(parts) != 8 && len(parts) != 9 {
return nil, fmt.Errorf("Invalid leaf zk path: %v", vtnsAddrPath)
}
cell := parts[2]
keyspace := parts[5]
shard := parts[6]
tabletTypeStr := parts[7]
if tabletTypeStr == "action" || tabletTypeStr == "actionlog" {
return nil, nil
}
tabletType, err := topoproto.ParseTabletType(tabletTypeStr)
if err != nil {
return nil, err
}
addrs, _, err := wr.ts.GetEndPoints(ctx, cell, keyspace, shard, tabletType)
if err != nil {
return nil, err
}
// Write the individual endpoints and compute the SRV entries.
vtoccAddrs := LegacyZknsAddrs{make([]string, 0, 8)}
defaultAddrs := LegacyZknsAddrs{make([]string, 0, 8)}
for i, entry := range addrs.Entries {
zknsAddrPath := fmt.Sprintf("%v/%v", zknsAddrPath, i)
zknsPaths = append(zknsPaths, zknsAddrPath)
zknsAddr := zkns.ZknsAddr{
Host: entry.Host,
PortMap: entry.PortMap,
}
err := writeAddr(zconn, zknsAddrPath, &zknsAddr)
if err != nil {
return nil, err
}
defaultAddrs.Endpoints = append(defaultAddrs.Endpoints, zknsAddrPath)
vtoccAddrs.Endpoints = append(vtoccAddrs.Endpoints, zknsAddrPath+":vt")
}
// Prune any zkns entries that are no longer referenced by the
// shard graph.
deleteIdx := len(addrs.Entries)
for {
zknsStaleAddrPath := fmt.Sprintf("%v/%v", zknsAddrPath, deleteIdx)
// A "delete" is a write of sorts - just communicate up that nothing
// needs to be done to this node.
zknsPaths = append(zknsPaths, zknsStaleAddrPath)
err := zconn.Delete(zknsStaleAddrPath, -1)
if zookeeper.IsError(err, zookeeper.ZNONODE) {
break
}
if err != nil {
return nil, err
}
deleteIdx++
}
// Write the VDNS entries for both vtocc and mysql
vtoccVdnsPath := fmt.Sprintf("%v/vt.vdns", zknsAddrPath)
zknsPaths = append(zknsPaths, vtoccVdnsPath)
if err = writeAddrs(zconn, vtoccVdnsPath, &vtoccAddrs); err != nil {
return nil, err
}
defaultVdnsPath := fmt.Sprintf("%v.vdns", zknsAddrPath)
zknsPaths = append(zknsPaths, defaultVdnsPath)
if err = writeAddrs(zconn, defaultVdnsPath, &defaultAddrs); err != nil {
return nil, err
}
return zknsPaths, nil
}
func initAPI(ctx context.Context, ts topo.Server, actions *ActionRepository, realtimeStats *realtimeStats) {
tabletHealthCache := newTabletHealthCache(ts)
tmClient := tmclient.NewTabletManagerClient()
// Cells
handleCollection("cells", func(r *http.Request) (interface{}, error) {
if getItemPath(r.URL.Path) != "" {
return nil, errors.New("cells can only be listed, not retrieved")
}
return ts.GetKnownCells(ctx)
})
// Keyspaces
handleCollection("keyspaces", func(r *http.Request) (interface{}, error) {
keyspace := getItemPath(r.URL.Path)
switch r.Method {
case "GET":
// List all keyspaces.
if keyspace == "" {
return ts.GetKeyspaces(ctx)
}
// Get the keyspace record.
k, err := ts.GetKeyspace(ctx, keyspace)
// Pass the embedded proto directly or jsonpb will panic.
return k.Keyspace, err
// Perform an action on a keyspace.
case "POST":
if keyspace == "" {
return nil, errors.New("A POST request needs a keyspace in the URL")
}
if err := r.ParseForm(); err != nil {
return nil, err
}
action := r.FormValue("action")
if action == "" {
return nil, errors.New("A POST request must specify action")
}
return actions.ApplyKeyspaceAction(ctx, action, keyspace, r), nil
default:
return nil, fmt.Errorf("unsupported HTTP method: %v", r.Method)
}
})
// Shards
handleCollection("shards", func(r *http.Request) (interface{}, error) {
shardPath := getItemPath(r.URL.Path)
if !strings.Contains(shardPath, "/") {
return nil, fmt.Errorf("invalid shard path: %q", shardPath)
}
parts := strings.SplitN(shardPath, "/", 2)
keyspace := parts[0]
shard := parts[1]
// List the shards in a keyspace.
if shard == "" {
return ts.GetShardNames(ctx, keyspace)
}
// Perform an action on a shard.
if r.Method == "POST" {
if err := r.ParseForm(); err != nil {
return nil, err
}
action := r.FormValue("action")
if action == "" {
return nil, errors.New("must specify action")
}
return actions.ApplyShardAction(ctx, action, keyspace, shard, r), nil
}
// Get the shard record.
si, err := ts.GetShard(ctx, keyspace, shard)
// Pass the embedded proto directly or jsonpb will panic.
return si.Shard, err
})
// SrvKeyspace
handleCollection("srv_keyspace", func(r *http.Request) (interface{}, error) {
keyspacePath := getItemPath(r.URL.Path)
parts := strings.SplitN(keyspacePath, "/", 2)
// Request was incorrectly formatted.
if len(parts) != 2 {
return nil, fmt.Errorf("invalid srvkeyspace path: %q expected path: /srv_keyspace/<cell>/<keyspace>", keyspacePath)
}
cell := parts[0]
keyspace := parts[1]
if cell == "local" {
if *localCell == "" {
return nil, fmt.Errorf("local cell requested, but not specified. Please set with -cell flag")
}
cell = *localCell
}
// If a keyspace is provided then return the specified srvkeyspace.
if keyspace != "" {
srvKeyspace, err := ts.GetSrvKeyspace(ctx, cell, keyspace)
if err != nil {
return nil, fmt.Errorf("Can't get server keyspace: %v", err)
}
return srvKeyspace, nil
}
// Else return the srvKeyspace from all keyspaces.
srvKeyspaces := make(map[string]interface{})
keyspaceNamesList, err := ts.GetSrvKeyspaceNames(ctx, cell)
if err != nil {
return nil, fmt.Errorf("can't get list of SrvKeyspaceNames for cell %q: GetSrvKeyspaceNames returned: %v", cell, err)
}
for _, keyspaceName := range keyspaceNamesList {
err := addSrvkeyspace(ctx, ts, cell, keyspaceName, srvKeyspaces)
if err != nil {
return nil, err
}
}
return srvKeyspaces, nil
})
// Tablets
handleCollection("tablets", func(r *http.Request) (interface{}, error) {
tabletPath := getItemPath(r.URL.Path)
// List tablets based on query params.
if tabletPath == "" {
if err := r.ParseForm(); err != nil {
return nil, err
}
shardRef := r.FormValue("shard")
cell := r.FormValue("cell")
if shardRef != "" {
// Look up by keyspace/shard, and optionally cell.
keyspace, shard, err := topoproto.ParseKeyspaceShard(shardRef)
if err != nil {
return nil, err
}
if cell != "" {
result, err := ts.FindAllTabletAliasesInShardByCell(ctx, keyspace, shard, []string{cell})
if err != nil && err != topo.ErrPartialResult {
return result, err
}
return result, nil
}
result, err := ts.FindAllTabletAliasesInShard(ctx, keyspace, shard)
if err != nil && err != topo.ErrPartialResult {
return result, err
}
return result, nil
}
// Get all tablets in a cell.
if cell == "" {
return nil, errors.New("cell param required")
}
return ts.GetTabletsByCell(ctx, cell)
}
// Get tablet health.
if parts := strings.Split(tabletPath, "/"); len(parts) == 2 && parts[1] == "health" {
tabletAlias, err := topoproto.ParseTabletAlias(parts[0])
if err != nil {
return nil, err
}
return tabletHealthCache.Get(ctx, tabletAlias)
}
tabletAlias, err := topoproto.ParseTabletAlias(tabletPath)
if err != nil {
return nil, err
}
// Perform an action on a tablet.
if r.Method == "POST" {
if err := r.ParseForm(); err != nil {
return nil, err
}
action := r.FormValue("action")
if action == "" {
return nil, errors.New("must specify action")
}
return actions.ApplyTabletAction(ctx, action, tabletAlias, r), nil
}
// Get the tablet record.
t, err := ts.GetTablet(ctx, tabletAlias)
// Pass the embedded proto directly or jsonpb will panic.
return t.Tablet, err
})
// Healthcheck real time status per (cell, keyspace, tablet type, metric).
handleCollection("tablet_statuses", func(r *http.Request) (interface{}, error) {
targetPath := getItemPath(r.URL.Path)
// Get the heatmap data based on query parameters.
if targetPath == "" {
if err := r.ParseForm(); err != nil {
return nil, err
}
keyspace := r.FormValue("keyspace")
cell := r.FormValue("cell")
tabletType := r.FormValue("type")
_, err := topoproto.ParseTabletType(tabletType)
// Excluding the case where parse fails because all tabletTypes was chosen.
if err != nil && tabletType != "all" {
return nil, fmt.Errorf("invalid tablet type: %v ", err)
}
metric := r.FormValue("metric")
// Setting default values if none was specified in the query params.
if keyspace == "" {
keyspace = "all"
}
if cell == "" {
cell = "all"
}
if tabletType == "" {
tabletType = "all"
}
if metric == "" {
metric = "health"
}
if realtimeStats == nil {
return nil, fmt.Errorf("realtimeStats not initialized")
}
heatmap, err := realtimeStats.heatmapData(keyspace, cell, tabletType, metric)
if err != nil {
return nil, fmt.Errorf("couldn't get heatmap data: %v", err)
}
return heatmap, nil
}
return nil, fmt.Errorf("invalid target path: %q expected path: ?keyspace=<keyspace>&cell=<cell>&type=<type>&metric=<metric>", targetPath)
})
handleCollection("tablet_health", func(r *http.Request) (interface{}, error) {
tabletPath := getItemPath(r.URL.Path)
parts := strings.SplitN(tabletPath, "/", 2)
// Request was incorrectly formatted.
if len(parts) != 2 {
return nil, fmt.Errorf("invalid tablet_health path: %q expected path: /tablet_health/<cell>/<uid>", tabletPath)
}
if realtimeStats == nil {
return nil, fmt.Errorf("realtimeStats not initialized")
}
cell := parts[0]
uidStr := parts[1]
uid, err := topoproto.ParseUID(uidStr)
if err != nil {
return nil, fmt.Errorf("incorrect uid: %v", err)
}
tabletAlias := topodatapb.TabletAlias{
Cell: cell,
Uid: uid,
}
tabletStat, err := realtimeStats.tabletStats(&tabletAlias)
if err != nil {
return nil, fmt.Errorf("could not get tabletStats: %v", err)
}
return tabletStat, nil
})
handleCollection("topology_info", func(r *http.Request) (interface{}, error) {
targetPath := getItemPath(r.URL.Path)
// Retrieving topology information (keyspaces, cells, and types) based on query params.
if targetPath == "" {
if err := r.ParseForm(); err != nil {
return nil, err
}
keyspace := r.FormValue("keyspace")
cell := r.FormValue("cell")
// Setting default values if none was specified in the query params.
if keyspace == "" {
keyspace = "all"
}
if cell == "" {
cell = "all"
}
if realtimeStats == nil {
return nil, fmt.Errorf("realtimeStats not initialized")
}
return realtimeStats.topologyInfo(keyspace, cell), nil
}
return nil, fmt.Errorf("invalid target path: %q expected path: ?keyspace=<keyspace>&cell=<cell>", targetPath)
})
// Vtctl Command
http.HandleFunc(apiPrefix+"vtctl/", func(w http.ResponseWriter, r *http.Request) {
if err := acl.CheckAccessHTTP(r, acl.ADMIN); err != nil {
httpErrorf(w, r, "Access denied")
return
}
var args []string
resp := struct {
Error string
Output string
}{}
if err := unmarshalRequest(r, &args); err != nil {
httpErrorf(w, r, "can't unmarshal request: %v", err)
return
}
logstream := logutil.NewMemoryLogger()
wr := wrangler.New(logstream, ts, tmClient)
// TODO(enisoc): Context for run command should be request-scoped.
err := vtctl.RunCommand(ctx, wr, args)
if err != nil {
resp.Error = err.Error()
}
resp.Output = logstream.String()
data, err := json.MarshalIndent(resp, "", " ")
if err != nil {
httpErrorf(w, r, "json error: %v", err)
return
}
w.Header().Set("Content-Type", jsonContentType)
w.Write(data)
})
// Schema Change
http.HandleFunc(apiPrefix+"schema/apply", func(w http.ResponseWriter, r *http.Request) {
if err := acl.CheckAccessHTTP(r, acl.ADMIN); err != nil {
httpErrorf(w, r, "Access denied")
return
}
req := struct {
Keyspace, SQL string
SlaveTimeoutSeconds int
}{}
if err := unmarshalRequest(r, &req); err != nil {
httpErrorf(w, r, "can't unmarshal request: %v", err)
return
}
if req.SlaveTimeoutSeconds <= 0 {
req.SlaveTimeoutSeconds = 10
}
logger := logutil.NewCallbackLogger(func(ev *logutilpb.Event) {
w.Write([]byte(logutil.EventString(ev)))
})
wr := wrangler.New(logger, ts, tmClient)
executor := schemamanager.NewTabletExecutor(
wr, time.Duration(req.SlaveTimeoutSeconds)*time.Second)
schemamanager.Run(ctx,
schemamanager.NewUIController(req.SQL, req.Keyspace, w), executor)
})
}
func initAPI(ctx context.Context, ts topo.Server, actions *ActionRepository) {
tabletHealthCache := newTabletHealthCache(ts)
// Cells
handleCollection("cells", func(r *http.Request) (interface{}, error) {
if getItemPath(r.URL.Path) != "" {
return nil, errors.New("cells can only be listed, not retrieved")
}
return ts.GetKnownCells(ctx)
})
// Keyspaces
handleCollection("keyspaces", func(r *http.Request) (interface{}, error) {
keyspace := getItemPath(r.URL.Path)
// List all keyspaces.
if keyspace == "" {
return ts.GetKeyspaces(ctx)
}
// Perform an action on a keyspace.
if r.Method == "POST" {
if err := r.ParseForm(); err != nil {
return nil, err
}
action := r.FormValue("action")
if action == "" {
return nil, errors.New("must specify action")
}
return actions.ApplyKeyspaceAction(ctx, action, keyspace, r), nil
}
// Get the keyspace record.
return ts.GetKeyspace(ctx, keyspace)
})
// Shards
handleCollection("shards", func(r *http.Request) (interface{}, error) {
shardPath := getItemPath(r.URL.Path)
if !strings.Contains(shardPath, "/") {
return nil, fmt.Errorf("invalid shard path: %q", shardPath)
}
parts := strings.SplitN(shardPath, "/", 2)
keyspace := parts[0]
shard := parts[1]
// List the shards in a keyspace.
if shard == "" {
return ts.GetShardNames(ctx, keyspace)
}
// Perform an action on a shard.
if r.Method == "POST" {
if err := r.ParseForm(); err != nil {
return nil, err
}
action := r.FormValue("action")
if action == "" {
return nil, errors.New("must specify action")
}
return actions.ApplyShardAction(ctx, action, keyspace, shard, r), nil
}
// Get the shard record.
return ts.GetShard(ctx, keyspace, shard)
})
// Tablets
handleCollection("tablets", func(r *http.Request) (interface{}, error) {
tabletPath := getItemPath(r.URL.Path)
// List tablets based on query params.
if tabletPath == "" {
if err := r.ParseForm(); err != nil {
return nil, err
}
shardRef := r.FormValue("shard")
cell := r.FormValue("cell")
if shardRef != "" {
// Look up by keyspace/shard, and optionally cell.
keyspace, shard, err := topoproto.ParseKeyspaceShard(shardRef)
if err != nil {
return nil, err
}
if cell != "" {
return ts.FindAllTabletAliasesInShardByCell(ctx, keyspace, shard, []string{cell})
}
return ts.FindAllTabletAliasesInShard(ctx, keyspace, shard)
}
// Get all tablets in a cell.
if cell == "" {
return nil, errors.New("cell param required")
}
return ts.GetTabletsByCell(ctx, cell)
}
// Get tablet health.
if parts := strings.Split(tabletPath, "/"); len(parts) == 2 && parts[1] == "health" {
tabletAlias, err := topoproto.ParseTabletAlias(parts[0])
if err != nil {
return nil, err
}
return tabletHealthCache.Get(ctx, tabletAlias)
}
tabletAlias, err := topoproto.ParseTabletAlias(tabletPath)
if err != nil {
return nil, err
}
// Perform an action on a tablet.
if r.Method == "POST" {
if err := r.ParseForm(); err != nil {
return nil, err
}
action := r.FormValue("action")
if action == "" {
return nil, errors.New("must specify action")
}
return actions.ApplyTabletAction(ctx, action, tabletAlias, r), nil
}
// Get the tablet record.
return ts.GetTablet(ctx, tabletAlias)
})
// EndPoints
handleCollection("endpoints", func(r *http.Request) (interface{}, error) {
// We expect cell/keyspace/shard/tabletType.
epPath := getItemPath(r.URL.Path)
parts := strings.Split(epPath, "/")
if len(parts) != 4 {
return nil, fmt.Errorf("invalid cell/keyspace/shard/tabletType: %q", epPath)
}
if parts[3] == "" {
// tabletType is empty, so list the tablet types.
return ts.GetSrvTabletTypesPerShard(ctx, parts[0], parts[1], parts[2])
}
tabletType, err := topoproto.ParseTabletType(parts[3])
if err != nil {
return nil, fmt.Errorf("invalid tablet type %v: %v", parts[3], err)
}
// Get the endpoints object for a specific type.
ep, _, err := ts.GetEndPoints(ctx, parts[0], parts[1], parts[2], tabletType)
return ep, err
})
// Schema Change
http.HandleFunc(apiPrefix+"schema/apply", func(w http.ResponseWriter, r *http.Request) {
req := struct{ Keyspace, SQL string }{}
if err := unmarshalRequest(r, &req); err != nil {
httpErrorf(w, r, "can't unmarshal request: %v", err)
return
}
executor := schemamanager.NewTabletExecutor(
tmclient.NewTabletManagerClient(),
ts)
schemamanager.Run(ctx,
schemamanager.NewUIController(req.SQL, req.Keyspace, w), executor)
})
// VSchema
http.HandleFunc(apiPrefix+"vschema/", func(w http.ResponseWriter, r *http.Request) {
// Save VSchema
if r.Method == "POST" {
vschema, err := ioutil.ReadAll(r.Body)
if err != nil {
httpErrorf(w, r, "can't read request body: %v", err)
return
}
if err := ts.SaveVSchema(ctx, string(vschema)); err != nil {
httpErrorf(w, r, "can't save vschema: %v", err)
}
return
}
// Get VSchema
vschema, err := ts.GetVSchema(ctx)
if err != nil {
httpErrorf(w, r, "can't get vschema: %v", err)
return
}
w.Header().Set("Content-Type", jsonContentType)
w.Write([]byte(vschema))
})
}