func (agent *ActionAgent) initHealthCheck() {
if !agent.IsRunningHealthCheck() {
log.Infof("No target_tablet_type specified, disabling any health check")
return
}
tt, err := topo.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()
}
// 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 := topo.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 := topo.ParseTabletAliasString(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")
}
}
// 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 = topo.ParseTabletType(c.TabletType)
if err != nil {
return nil, err
}
err = c.dial()
if err != nil {
return nil, err
}
return c, nil
}
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")
if err := subFlags.Parse(args); err != nil {
return err
}
if subFlags.NArg() != 2 {
return fmt.Errorf("the <tablet_alis> and <sql> arguments are required for the VtTabletExecute command")
}
tt, err := topo.ParseTabletType(*tabletType)
if err != nil {
return err
}
tabletAlias, err := topo.ParseTabletAliasString(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)
}
wr.Logger().Printf("%v\n", jscfg.ToJSON(qr))
return 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 := topo.ParseTabletType(*tabletType)
if err != nil {
return err
}
tabletAlias, err := topo.ParseTabletAliasString(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) ([]pb.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([]pb.TabletType, 0, len(children))
for _, tt := range children {
// these two are used for locking
if tt == "action" || tt == "actionlog" {
continue
}
if ptt, err := topo.ParseTabletType(tt); err == nil {
result = append(result, ptt)
}
}
return result, nil
}
// GetSrvTabletTypesPerShard implements topo.Server.
func (s *Server) GetSrvTabletTypesPerShard(ctx context.Context, cellName, keyspace, shard string) ([]pb.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([]pb.TabletType, 0, len(resp.Node.Nodes))
for _, n := range resp.Node.Nodes {
strType := path.Base(n.Key)
if tt, err := topo.ParseTabletType(strType); err == nil {
tabletTypes = append(tabletTypes, tt)
}
}
return tabletTypes, 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 := topo.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
}
// 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 pb.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 = topo.ParseTabletType(*initTabletType)
if err != nil {
log.Fatalf("Invalid init tablet type %v: %v", *initTabletType, err)
}
if tabletType == pb.TabletType_MASTER || tabletType == pb.TabletType_SCRAP {
// We disallow TYPE_MASTER, so we don't have to change
// shard.MasterAlias, and deal with the corner cases.
// We also disallow TYPE_SCRAP, obviously.
log.Fatalf("init_tablet_type cannot be %v", tabletType)
}
} else if *targetTabletType != "" {
if strings.ToUpper(*targetTabletType) == pb.TabletType_name[int32(pb.TabletType_MASTER)] {
log.Fatalf("target_tablet_type cannot be '%v'. Use '%v' instead.", tabletType, pb.TabletType_REPLICA)
}
// use spare, the healthcheck will turn us into what
// we need to be eventually
tabletType = pb.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()
// if we're assigned to a shard, make sure it exists, see if
// we are its master, and update its cells list if necessary
if tabletType != pb.TabletType_IDLE {
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 && topo.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 = pb.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 := topo.UpdateShard(ctx, agent.TopoServer, 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 == "" {
var err error
hostname, err = netutil.FullyQualifiedHostname()
if err != nil {
return err
}
}
// create and populate tablet record
tablet := &pb.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 := topo.CreateTablet(ctx, agent.TopoServer, tablet)
switch err {
case nil:
// it worked, we're good, can update the replication graph
if topo.IsInReplicationGraph(tablet.Type) {
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 {
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 := topo.UpdateTablet(ctx, agent.TopoServer, 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 tabletType != pb.TabletType_IDLE {
if err := topotools.UpdateTabletEndpoints(ctx, agent.TopoServer, tablet); err != nil {
return fmt.Errorf("UpdateTabletEndpoints failed: %v", err)
}
}
return nil
}