func TestTabletExternallyReparented(t *testing.T) {
tabletmanager.SetReparentFlags(time.Minute /* finalizeTimeout */)
ctx := context.Background()
db := fakesqldb.Register()
ts := zktopo.NewTestServer(t, []string{"cell1", "cell2"})
wr := wrangler.New(logutil.NewConsoleLogger(), ts, tmclient.NewTabletManagerClient(), time.Second)
vp := NewVtctlPipe(t, ts)
defer vp.Close()
// Create an old master, a new master, two good slaves, one bad slave
oldMaster := NewFakeTablet(t, wr, "cell1", 0, pb.TabletType_MASTER, db)
newMaster := NewFakeTablet(t, wr, "cell1", 1, pb.TabletType_REPLICA, db)
goodSlave1 := NewFakeTablet(t, wr, "cell1", 2, pb.TabletType_REPLICA, db)
goodSlave2 := NewFakeTablet(t, wr, "cell2", 3, pb.TabletType_REPLICA, db)
badSlave := NewFakeTablet(t, wr, "cell1", 4, pb.TabletType_REPLICA, db)
// Add a new Cell to the Shard, that doesn't map to any read topo cell,
// to simulate a data center being unreachable.
si, err := ts.GetShard(ctx, "test_keyspace", "0")
if err != nil {
t.Fatalf("GetShard failed: %v", err)
}
si.Cells = append(si.Cells, "cell666")
if err := ts.UpdateShard(ctx, si); err != nil {
t.Fatalf("UpdateShard failed: %v", err)
}
// Slightly unrelated test: make sure we can find the tablets
// even with a datacenter being down.
tabletMap, err := ts.GetTabletMapForShardByCell(ctx, "test_keyspace", "0", []string{"cell1"})
if err != nil {
t.Fatalf("GetTabletMapForShardByCell should have worked but got: %v", err)
}
master, err := topotools.FindTabletByIPAddrAndPort(tabletMap, oldMaster.Tablet.Ip, "vt", oldMaster.Tablet.PortMap["vt"])
if err != nil || !topoproto.TabletAliasEqual(&master, oldMaster.Tablet.Alias) {
t.Fatalf("FindTabletByIPAddrAndPort(master) failed: %v %v", err, master)
}
slave1, err := topotools.FindTabletByIPAddrAndPort(tabletMap, goodSlave1.Tablet.Ip, "vt", goodSlave1.Tablet.PortMap["vt"])
if err != nil || !topoproto.TabletAliasEqual(&slave1, goodSlave1.Tablet.Alias) {
t.Fatalf("FindTabletByIPAddrAndPort(slave1) failed: %v %v", err, master)
}
slave2, err := topotools.FindTabletByIPAddrAndPort(tabletMap, goodSlave2.Tablet.Ip, "vt", goodSlave2.Tablet.PortMap["vt"])
if err != topo.ErrNoNode {
t.Fatalf("FindTabletByIPAddrAndPort(slave2) worked: %v %v", err, slave2)
}
// Make sure the master is not exported in other cells
tabletMap, err = ts.GetTabletMapForShardByCell(ctx, "test_keyspace", "0", []string{"cell2"})
master, err = topotools.FindTabletByIPAddrAndPort(tabletMap, oldMaster.Tablet.Ip, "vt", oldMaster.Tablet.PortMap["vt"])
if err != topo.ErrNoNode {
t.Fatalf("FindTabletByIPAddrAndPort(master) worked in cell2: %v %v", err, master)
}
tabletMap, err = ts.GetTabletMapForShard(ctx, "test_keyspace", "0")
if err != topo.ErrPartialResult {
t.Fatalf("GetTabletMapForShard should have returned ErrPartialResult but got: %v", err)
}
master, err = topotools.FindTabletByIPAddrAndPort(tabletMap, oldMaster.Tablet.Ip, "vt", oldMaster.Tablet.PortMap["vt"])
if err != nil || !topoproto.TabletAliasEqual(&master, oldMaster.Tablet.Alias) {
t.Fatalf("FindTabletByIPAddrAndPort(master) failed: %v %v", err, master)
}
// On the elected master, we will respond to
// TabletActionSlaveWasPromoted
newMaster.StartActionLoop(t, wr)
defer newMaster.StopActionLoop(t)
// On the old master, we will only respond to
// TabletActionSlaveWasRestarted.
oldMaster.StartActionLoop(t, wr)
defer oldMaster.StopActionLoop(t)
// On the good slaves, we will respond to
// TabletActionSlaveWasRestarted.
goodSlave1.StartActionLoop(t, wr)
defer goodSlave1.StopActionLoop(t)
goodSlave2.StartActionLoop(t, wr)
defer goodSlave2.StopActionLoop(t)
// On the bad slave, we will respond to
// TabletActionSlaveWasRestarted with bad data.
badSlave.StartActionLoop(t, wr)
defer badSlave.StopActionLoop(t)
// First test: reparent to the same master, make sure it works
// as expected.
tmc := tmclient.NewTabletManagerClient()
ti, err := ts.GetTablet(ctx, oldMaster.Tablet.Alias)
if err != nil {
t.Fatalf("GetTablet failed: %v", err)
}
if err := vp.Run([]string{"TabletExternallyReparented", topoproto.TabletAliasString(oldMaster.Tablet.Alias)}); err != nil {
t.Fatalf("TabletExternallyReparented(same master) should have worked: %v", err)
}
// Second test: reparent to a replica, and pretend the old
// master is still good to go.
// This tests a bad case; the new designated master is a slave,
// but we should do what we're told anyway
ti, err = ts.GetTablet(ctx, goodSlave1.Tablet.Alias)
if err != nil {
t.Fatalf("GetTablet failed: %v", err)
}
waitID := makeWaitID()
if err := tmc.TabletExternallyReparented(context.Background(), ti, waitID); err != nil {
t.Fatalf("TabletExternallyReparented(slave) error: %v", err)
}
waitForExternalReparent(t, waitID)
// This tests the good case, where everything works as planned
t.Logf("TabletExternallyReparented(new master) expecting success")
ti, err = ts.GetTablet(ctx, newMaster.Tablet.Alias)
if err != nil {
t.Fatalf("GetTablet failed: %v", err)
}
waitID = makeWaitID()
if err := tmc.TabletExternallyReparented(context.Background(), ti, waitID); err != nil {
t.Fatalf("TabletExternallyReparented(replica) failed: %v", err)
}
waitForExternalReparent(t, waitID)
// Now double-check the serving graph is good.
// Should only have one good replica left.
addrs, _, err := ts.GetEndPoints(ctx, "cell1", "test_keyspace", "0", pb.TabletType_REPLICA)
if err != nil {
t.Fatalf("GetEndPoints failed at the end: %v", err)
}
if len(addrs.Entries) != 1 {
t.Fatalf("GetEndPoints has too many entries: %v", addrs)
}
}
func TestShardExternallyReparented(t *testing.T) {
ts := zktopo.NewTestServer(t, []string{"cell1", "cell2"})
wr := wrangler.New(logutil.NewConsoleLogger(), ts, time.Minute, time.Second)
wr.UseRPCs = false
// Create an old master, a new master, two good slaves, one bad slave
oldMaster := NewFakeTablet(t, wr, "cell1", 0, topo.TYPE_MASTER)
newMaster := NewFakeTablet(t, wr, "cell1", 1, topo.TYPE_REPLICA,
TabletParent(oldMaster.Tablet.Alias))
goodSlave1 := NewFakeTablet(t, wr, "cell1", 2, topo.TYPE_REPLICA,
TabletParent(oldMaster.Tablet.Alias))
goodSlave2 := NewFakeTablet(t, wr, "cell2", 3, topo.TYPE_REPLICA,
TabletParent(oldMaster.Tablet.Alias))
badSlave := NewFakeTablet(t, wr, "cell1", 4, topo.TYPE_REPLICA,
TabletParent(oldMaster.Tablet.Alias))
// Add a new Cell to the Shard, that doesn't map to any read topo cell,
// to simulate a data center being unreachable.
si, err := ts.GetShard("test_keyspace", "0")
if err != nil {
t.Fatalf("GetShard failed: %v", err)
}
si.Cells = append(si.Cells, "cell666")
if err := topo.UpdateShard(ts, si); err != nil {
t.Fatalf("UpdateShard failed: %v", err)
}
// Slightly unrelated test: make sure we can find the tablets
// even with a datacenter being down.
tabletMap, err := topo.GetTabletMapForShardByCell(ts, "test_keyspace", "0", []string{"cell1"})
if err != nil {
t.Fatalf("GetTabletMapForShardByCell should have worked but got: %v", err)
}
master, err := topotools.FindTabletByIPAddrAndPort(tabletMap, oldMaster.Tablet.IPAddr, "vt", oldMaster.Tablet.Portmap["vt"])
if err != nil || master != oldMaster.Tablet.Alias {
t.Fatalf("FindTabletByIPAddrAndPort(master) failed: %v %v", err, master)
}
slave1, err := topotools.FindTabletByIPAddrAndPort(tabletMap, goodSlave1.Tablet.IPAddr, "vt", goodSlave1.Tablet.Portmap["vt"])
if err != nil || slave1 != goodSlave1.Tablet.Alias {
t.Fatalf("FindTabletByIPAddrAndPort(slave1) failed: %v %v", err, master)
}
slave2, err := topotools.FindTabletByIPAddrAndPort(tabletMap, goodSlave2.Tablet.IPAddr, "vt", goodSlave2.Tablet.Portmap["vt"])
if err != topo.ErrNoNode {
t.Fatalf("FindTabletByIPAddrAndPort(slave2) worked: %v %v", err, slave2)
}
// Make sure the master is not exported in other cells
tabletMap, err = topo.GetTabletMapForShardByCell(ts, "test_keyspace", "0", []string{"cell2"})
master, err = topotools.FindTabletByIPAddrAndPort(tabletMap, oldMaster.Tablet.IPAddr, "vt", oldMaster.Tablet.Portmap["vt"])
if err != topo.ErrNoNode {
t.Fatalf("FindTabletByIPAddrAndPort(master) worked in cell2: %v %v", err, master)
}
tabletMap, err = topo.GetTabletMapForShard(ts, "test_keyspace", "0")
if err != topo.ErrPartialResult {
t.Fatalf("GetTabletMapForShard should have returned ErrPartialResult but got: %v", err)
}
master, err = topotools.FindTabletByIPAddrAndPort(tabletMap, oldMaster.Tablet.IPAddr, "vt", oldMaster.Tablet.Portmap["vt"])
if err != nil || master != oldMaster.Tablet.Alias {
t.Fatalf("FindTabletByIPAddrAndPort(master) failed: %v %v", err, master)
}
// First test: reparent to the same master, make sure it works
// as expected.
if err := wr.ShardExternallyReparented("test_keyspace", "0", oldMaster.Tablet.Alias); err == nil {
t.Fatalf("ShardExternallyReparented(same master) should have failed")
} else {
if !strings.Contains(err.Error(), "already master") {
t.Fatalf("ShardExternallyReparented(same master) should have failed with an error that contains 'already master' but got: %v", err)
}
}
// Second test: reparent to the replica, and pretend the old
// master is still good to go.
// On the elected master, we will respond to
// TABLET_ACTION_SLAVE_WAS_PROMOTED
newMaster.FakeMysqlDaemon.MasterAddr = ""
newMaster.StartActionLoop(t, wr)
defer newMaster.StopActionLoop(t)
// On the old master, we will only respond to
// TABLET_ACTION_SLAVE_WAS_RESTARTED.
oldMaster.FakeMysqlDaemon.MasterAddr = newMaster.Tablet.MysqlIpAddr()
oldMaster.StartActionLoop(t, wr)
defer oldMaster.StopActionLoop(t)
// On the good slaves, we will respond to
// TABLET_ACTION_SLAVE_WAS_RESTARTED.
goodSlave1.FakeMysqlDaemon.MasterAddr = newMaster.Tablet.MysqlIpAddr()
goodSlave1.StartActionLoop(t, wr)
defer goodSlave1.StopActionLoop(t)
goodSlave2.FakeMysqlDaemon.MasterAddr = newMaster.Tablet.MysqlIpAddr()
goodSlave2.StartActionLoop(t, wr)
defer goodSlave2.StopActionLoop(t)
// On the bad slave, we will respond to
// TABLET_ACTION_SLAVE_WAS_RESTARTED with bad data.
badSlave.FakeMysqlDaemon.MasterAddr = "234.0.0.1:3301"
badSlave.StartActionLoop(t, wr)
defer badSlave.StopActionLoop(t)
// This tests a bad case; the new designated master is a slave,
// but we should do what we're told anyway
if err := wr.ShardExternallyReparented("test_keyspace", "0", goodSlave1.Tablet.Alias); err != nil {
t.Fatalf("ShardExternallyReparented(slave) error: %v", err)
}
// This tests the good case, where everything works as planned
t.Logf("ShardExternallyReparented(new master) expecting success")
if err := wr.ShardExternallyReparented("test_keyspace", "0", newMaster.Tablet.Alias); err != nil {
t.Fatalf("ShardExternallyReparented(replica) failed: %v", err)
}
// Now double-check the serving graph is good.
// Should only have one good replica left.
addrs, err := ts.GetEndPoints("cell1", "test_keyspace", "0", topo.TYPE_REPLICA)
if err != nil {
t.Fatalf("GetEndPoints failed at the end: %v", err)
}
if len(addrs.Entries) != 1 {
t.Fatalf("GetEndPoints has too many entries: %v", addrs)
}
}