func waitForFilteredReplication(t *testing.T, expectedErr string, initialStats *pbq.RealtimeStats, broadcastStatsFunc func() *pbq.RealtimeStats) {
ts := zktopo.NewTestServer(t, []string{"cell1", "cell2"})
wr := wrangler.New(logutil.NewConsoleLogger(), ts, tmclient.NewTabletManagerClient(), time.Second)
vp := NewVtctlPipe(t, ts)
defer vp.Close()
// source of the filtered replication. We don't start its loop because we don't connect to it.
source := NewFakeTablet(t, wr, "cell1", 0, pbt.TabletType_MASTER,
TabletKeyspaceShard(t, keyspace, "0"))
// dest is the master of the dest shard which receives filtered replication events.
dest := NewFakeTablet(t, wr, "cell1", 1, pbt.TabletType_MASTER,
TabletKeyspaceShard(t, keyspace, destShard))
dest.StartActionLoop(t, wr)
defer dest.StopActionLoop(t)
// Build topology state as we would expect it when filtered replication is enabled.
ctx := context.Background()
wr.SetSourceShards(ctx, keyspace, destShard, []*pbt.TabletAlias{source.Tablet.GetAlias()}, nil)
// Set a BinlogPlayerMap to avoid a nil panic when the explicit RunHealthCheck
// is called by WaitForFilteredReplication.
// Note that for this test we don't mock the BinlogPlayerMap i.e. although
// its state says no filtered replication is running, the code under test will
// observe otherwise because we call TabletServer.BroadcastHealth() directly and
// skip going through the tabletmanager's agent.
dest.Agent.BinlogPlayerMap = tabletmanager.NewBinlogPlayerMap(ts, nil, nil)
// Use real, but trimmed down QueryService.
testConfig := tabletserver.DefaultQsConfig
testConfig.EnablePublishStats = false
testConfig.DebugURLPrefix = fmt.Sprintf("TestWaitForFilteredReplication-%d-", rand.Int63())
qs := tabletserver.NewTabletServer(testConfig)
grpcqueryservice.RegisterForTest(dest.RPCServer, qs)
qs.BroadcastHealth(42, initialStats)
// run vtctl WaitForFilteredReplication
stopBroadcasting := make(chan struct{})
go func() {
defer close(stopBroadcasting)
err := vp.Run([]string{"WaitForFilteredReplication", "-max_delay", "10s", dest.Tablet.Keyspace + "/" + dest.Tablet.Shard})
if expectedErr == "" {
if err != nil {
t.Fatalf("WaitForFilteredReplication must not fail: %v", err)
}
} else {
if err == nil || !strings.Contains(err.Error(), expectedErr) {
t.Fatalf("WaitForFilteredReplication wrong error. got: %v want substring: %v", err, expectedErr)
}
}
}()
// Broadcast health record as long as vtctl is running.
for {
// Give vtctl a head start to consume the initial stats.
// (We do this because there's unfortunately no way to explicitly
// synchronize with the point where conn.StreamHealth() has started.)
// (Tests won't break if vtctl misses the initial stats. Only coverage
// will be impacted.)
timer := time.NewTimer(1 * time.Millisecond)
select {
case <-stopBroadcasting:
timer.Stop()
return
case <-timer.C:
qs.BroadcastHealth(42, broadcastStatsFunc())
// Pace the flooding broadcasting to waste less CPU.
timer.Reset(1 * time.Millisecond)
}
}
// vtctl WaitForFilteredReplication returned.
}
// InitAgent initializes the agent within vttablet.
func InitAgent(
tabletAlias topo.TabletAlias,
dbcfgs *dbconfigs.DBConfigs,
mycnf *mysqlctl.Mycnf,
port, securePort int,
overridesFile string,
) (agent *tabletmanager.ActionAgent, err error) {
schemaOverrides := loadSchemaOverrides(overridesFile)
topoServer := topo.GetServer()
mysqld := mysqlctl.NewMysqld(mycnf, &dbcfgs.Dba, &dbcfgs.Repl)
statsType := stats.NewString("TabletType")
statsKeyspace := stats.NewString("TabletKeyspace")
statsShard := stats.NewString("TabletShard")
statsKeyRangeStart := stats.NewString("TabletKeyRangeStart")
statsKeyRangeEnd := stats.NewString("TabletKeyRangeEnd")
agent, err = tabletmanager.NewActionAgent(topoServer, tabletAlias, mysqld)
if err != nil {
return nil, err
}
// Start the binlog player services, not playing at start.
agent.BinlogPlayerMap = tabletmanager.NewBinlogPlayerMap(topoServer, &dbcfgs.App.ConnectionParams, mysqld)
tabletmanager.RegisterBinlogPlayerMap(agent.BinlogPlayerMap)
// Action agent listens to changes in zookeeper and makes
// modifications to this tablet.
agent.AddChangeCallback(func(oldTablet, newTablet topo.Tablet) {
allowQuery := true
var shardInfo *topo.ShardInfo
var keyspaceInfo *topo.KeyspaceInfo
if newTablet.Type == topo.TYPE_MASTER {
// read the shard to get SourceShards
shardInfo, err = topoServer.GetShard(newTablet.Keyspace, newTablet.Shard)
if err != nil {
log.Errorf("Cannot read shard for this tablet %v: %v", newTablet.Alias, err)
} else {
allowQuery = len(shardInfo.SourceShards) == 0
}
// read the keyspace to get ShardingColumnType
keyspaceInfo, err = topoServer.GetKeyspace(newTablet.Keyspace)
switch err {
case nil:
// continue
case topo.ErrNoNode:
// backward compatible mode
keyspaceInfo = topo.NewKeyspaceInfo(newTablet.Keyspace, &topo.Keyspace{})
default:
log.Errorf("Cannot read keyspace for this tablet %v: %v", newTablet.Alias, err)
keyspaceInfo = nil
}
}
if newTablet.IsRunningQueryService() && allowQuery {
if dbcfgs.App.DbName == "" {
dbcfgs.App.DbName = newTablet.DbName()
}
dbcfgs.App.Keyspace = newTablet.Keyspace
dbcfgs.App.Shard = newTablet.Shard
if newTablet.Type != topo.TYPE_MASTER {
dbcfgs.App.EnableInvalidator = true
} else {
dbcfgs.App.EnableInvalidator = false
}
// Transitioning from replica to master, first disconnect
// existing connections. "false" indicateds that clients must
// re-resolve their endpoint before reconnecting.
if newTablet.Type == topo.TYPE_MASTER && oldTablet.Type != topo.TYPE_MASTER {
ts.DisallowQueries()
}
qrs := ts.LoadCustomRules()
if newTablet.KeyRange.IsPartial() {
qr := ts.NewQueryRule("enforce keyspace_id range", "keyspace_id_not_in_range", ts.QR_FAIL_QUERY)
qr.AddPlanCond(sqlparser.PLAN_INSERT_PK)
err = qr.AddBindVarCond("keyspace_id", true, true, ts.QR_NOTIN, newTablet.KeyRange)
if err != nil {
log.Warningf("Unable to add keyspace rule: %v", err)
} else {
qrs.Add(qr)
}
}
ts.AllowQueries(&dbcfgs.App, schemaOverrides, qrs, mysqld)
// Disable before enabling to force existing streams to stop.
binlog.DisableUpdateStreamService()
binlog.EnableUpdateStreamService(dbcfgs)
} else {
ts.DisallowQueries()
binlog.DisableUpdateStreamService()
}
statsType.Set(string(newTablet.Type))
statsKeyspace.Set(newTablet.Keyspace)
statsShard.Set(newTablet.Shard)
statsKeyRangeStart.Set(string(newTablet.KeyRange.Start.Hex()))
statsKeyRangeEnd.Set(string(newTablet.KeyRange.End.Hex()))
// See if we need to start or stop any binlog player
if newTablet.Type == topo.TYPE_MASTER {
agent.BinlogPlayerMap.RefreshMap(newTablet, keyspaceInfo, shardInfo)
} else {
agent.BinlogPlayerMap.StopAllPlayersAndReset()
}
})
if err := agent.Start(mysqld.Port(), port, securePort); err != nil {
return nil, err
}
// register the RPC services from the agent
agent.RegisterQueryService()
return agent, nil
}
// InitAgent initializes the agent within vttablet.
func InitAgent(
tabletAlias topo.TabletAlias,
dbcfgs *dbconfigs.DBConfigs,
mycnf *mysqlctl.Mycnf,
port, securePort int,
overridesFile string,
) (agent *tabletmanager.ActionAgent, err error) {
schemaOverrides := loadSchemaOverrides(overridesFile)
topoServer := topo.GetServer()
mysqld := mysqlctl.NewMysqld(mycnf, &dbcfgs.Dba, &dbcfgs.Repl)
statsType := stats.NewString("TabletType")
statsKeyspace := stats.NewString("TabletKeyspace")
statsShard := stats.NewString("TabletShard")
statsKeyRangeStart := stats.NewString("TabletKeyRangeStart")
statsKeyRangeEnd := stats.NewString("TabletKeyRangeEnd")
agent, err = tabletmanager.NewActionAgent(topoServer, tabletAlias, mysqld)
if err != nil {
return nil, err
}
// Start the binlog player services, not playing at start.
agent.BinlogPlayerMap = tabletmanager.NewBinlogPlayerMap(topoServer, &dbcfgs.App.ConnectionParams, mysqld)
tabletmanager.RegisterBinlogPlayerMap(agent.BinlogPlayerMap)
// Action agent listens to changes in zookeeper and makes
// modifications to this tablet.
agent.AddChangeCallback(func(oldTablet, newTablet topo.Tablet) {
allowQuery := true
var shardInfo *topo.ShardInfo
var keyspaceInfo *topo.KeyspaceInfo
if newTablet.Type == topo.TYPE_MASTER {
// read the shard to get SourceShards
shardInfo, err = topoServer.GetShard(newTablet.Keyspace, newTablet.Shard)
if err != nil {
log.Errorf("Cannot read shard for this tablet %v: %v", newTablet.Alias, err)
} else {
allowQuery = len(shardInfo.SourceShards) == 0
}
// read the keyspace to get ShardingColumnType
keyspaceInfo, err = topoServer.GetKeyspace(newTablet.Keyspace)
switch err {
case nil:
// continue
case topo.ErrNoNode:
// backward compatible mode
keyspaceInfo = topo.NewKeyspaceInfo(newTablet.Keyspace, &topo.Keyspace{})
default:
log.Errorf("Cannot read keyspace for this tablet %v: %v", newTablet.Alias, err)
keyspaceInfo = nil
}
}
if newTablet.IsRunningQueryService() && allowQuery {
if dbcfgs.App.DbName == "" {
dbcfgs.App.DbName = newTablet.DbName()
}
dbcfgs.App.Keyspace = newTablet.Keyspace
dbcfgs.App.Shard = newTablet.Shard
if newTablet.Type != topo.TYPE_MASTER {
dbcfgs.App.EnableInvalidator = true
} else {
dbcfgs.App.EnableInvalidator = false
}
// There are a few transitions when we're
// going to need to restart the query service:
// - transitioning from replica to master, so clients
// that were already connected don't keep on using
// the master as replica or rdonly.
// - having different parameters for the query
// service. It needs to stop and restart with the
// new parameters. That includes:
// - changing KeyRange
// - changing the BlacklistedTables list
if (newTablet.Type == topo.TYPE_MASTER &&
oldTablet.Type != topo.TYPE_MASTER) ||
(newTablet.KeyRange != oldTablet.KeyRange) ||
!reflect.DeepEqual(newTablet.BlacklistedTables, oldTablet.BlacklistedTables) {
ts.DisallowQueries()
}
qrs := ts.LoadCustomRules()
if newTablet.KeyRange.IsPartial() {
qr := ts.NewQueryRule("enforce keyspace_id range", "keyspace_id_not_in_range", ts.QR_FAIL_QUERY)
qr.AddPlanCond(sqlparser.PLAN_INSERT_PK)
err = qr.AddBindVarCond("keyspace_id", true, true, ts.QR_NOTIN, newTablet.KeyRange)
if err != nil {
log.Warningf("Unable to add keyspace rule: %v", err)
} else {
qrs.Add(qr)
}
}
if len(newTablet.BlacklistedTables) > 0 {
log.Infof("Blacklisting tables %v", strings.Join(newTablet.BlacklistedTables, ", "))
qr := ts.NewQueryRule("enforce blacklisted tables", "blacklisted_table", ts.QR_FAIL_QUERY)
for _, t := range newTablet.BlacklistedTables {
qr.AddTableCond(t)
}
qrs.Add(qr)
}
ts.AllowQueries(&dbcfgs.App, schemaOverrides, qrs, mysqld)
// Disable before enabling to force existing streams to stop.
binlog.DisableUpdateStreamService()
binlog.EnableUpdateStreamService(dbcfgs)
} else {
ts.DisallowQueries()
binlog.DisableUpdateStreamService()
}
statsType.Set(string(newTablet.Type))
statsKeyspace.Set(newTablet.Keyspace)
statsShard.Set(newTablet.Shard)
statsKeyRangeStart.Set(string(newTablet.KeyRange.Start.Hex()))
statsKeyRangeEnd.Set(string(newTablet.KeyRange.End.Hex()))
// See if we need to start or stop any binlog player
if newTablet.Type == topo.TYPE_MASTER {
agent.BinlogPlayerMap.RefreshMap(newTablet, keyspaceInfo, shardInfo)
} else {
agent.BinlogPlayerMap.StopAllPlayersAndReset()
}
})
if err := agent.Start(mysqld.Port(), port, securePort); err != nil {
return nil, err
}
// register the RPC services from the agent
agent.RegisterQueryService()
// start health check if needed
initHeathCheck(agent)
return agent, nil
}