func AddSelfToTopology(cm *ConnectionManager, conns map[common.RMId]paxos.Connection, topology *server.Topology, fInc int, active, passive []common.RMId, lc *client.LocalConnection) (*server.Topology, bool, error) { seg := capn.NewBuffer(nil) txn := msgs.NewTxn(seg) txn.SetSubmitter(uint32(cm.RMId)) txn.SetSubmitterBootCount(cm.BootCount) actions := msgs.NewActionList(seg, 1) txn.SetActions(actions) topologyAction := actions.At(0) topologyAction.SetVarId(server.TopologyVarUUId[:]) topologyAction.SetReadwrite() rw := topologyAction.Readwrite() rw.SetVersion(topology.DBVersion[:]) rw.SetValue(topology.Serialize()) if topology.RootVarUUId == nil { rw.SetReferences(msgs.NewVarIdPosList(seg, 0)) } else { refs := msgs.NewVarIdPosList(seg, 1) rw.SetReferences(refs) varIdPos := refs.At(0) varIdPos.SetId(topology.RootVarUUId[:]) varIdPos.SetPositions((capn.UInt8List)(*topology.RootPositions)) } allocs := msgs.NewAllocationList(seg, len(topology.AllRMs)) txn.SetAllocations(allocs) idx := 0 for listIdx, rmIds := range [][]common.RMId{active, passive} { for _, rmId := range rmIds { alloc := allocs.At(idx) idx++ alloc.SetRmId(uint32(rmId)) if listIdx == 0 { alloc.SetActive(conns[rmId].BootCount()) } else { alloc.SetActive(0) } indices := seg.NewUInt16List(1) alloc.SetActionIndices(indices) indices.Set(0, 0) } } txn.SetFInc(uint8(fInc)) txn.SetTopologyVersion(topology.Version) result, err := lc.RunTransaction(&txn, true, active...) if err != nil || result == nil { return nil, false, err } txnId := common.MakeTxnId(result.Txn().Id()) if result.Which() == msgs.OUTCOME_COMMIT { topology.DBVersion = txnId server.Log("Topology Txn Committed ok with txnId", topology.DBVersion) return topology, false, nil } abort := result.Abort() server.Log("Topology Txn Aborted", txnId) if abort.Which() == msgs.OUTCOMEABORT_RESUBMIT { return nil, true, nil } abortUpdates := abort.Rerun() if abortUpdates.Len() != 1 { return nil, false, fmt.Errorf("Internal error: readwrite of topology gave %v updates (1 expected)", abortUpdates.Len()) } update := abortUpdates.At(0) dbversion := common.MakeTxnId(update.TxnId()) updateActions := update.Actions() if updateActions.Len() != 1 && updateActions.Len() != 2 { return nil, false, fmt.Errorf("Internal error: readwrite of topology gave multiple actions: %v", updateActions.Len()) } var updateAction *msgs.Action for idx, l := 0, updateActions.Len(); idx < l; idx++ { action := updateActions.At(idx) if bytes.Equal(action.VarId(), server.TopologyVarUUId[:]) { updateAction = &action break } } if updateAction == nil { return nil, false, fmt.Errorf("Internal error: unable to find action for topology from readwrite") } if updateAction.Which() != msgs.ACTION_WRITE { return nil, false, fmt.Errorf("Internal error: readwrite of topology gave non-write action") } write := updateAction.Write() var rootVarPos *msgs.VarIdPos if refs := write.References(); refs.Len() == 1 { root := refs.At(0) rootVarPos = &root } else if refs.Len() > 1 { return nil, false, fmt.Errorf("Internal error: readwrite of topology had wrong references: %v", refs.Len()) } topology, err = server.TopologyDeserialize(dbversion, rootVarPos, write.Value()) if err != nil { return nil, false, err } found := false for _, rmId := range topology.AllRMs { if found = rmId == cm.RMId; found { server.Log("Topology Txn Aborted, but found self in topology.") return topology, false, nil } } return topology, true, nil }
func GetTopologyFromLocalDatabase(cm *ConnectionManager, varDispatcher *eng.VarDispatcher, lc *client.LocalConnection) (*server.Topology, error) { if paxos.IsDatabaseClean(varDispatcher) { return nil, nil } for { seg := capn.NewBuffer(nil) txn := msgs.NewTxn(seg) txn.SetSubmitter(uint32(cm.RMId)) txn.SetSubmitterBootCount(cm.BootCount) actions := msgs.NewActionList(seg, 1) txn.SetActions(actions) action := actions.At(0) action.SetVarId(server.TopologyVarUUId[:]) action.SetRead() action.Read().SetVersion(common.VersionZero[:]) allocs := msgs.NewAllocationList(seg, 1) txn.SetAllocations(allocs) alloc := allocs.At(0) alloc.SetRmId(uint32(cm.RMId)) alloc.SetActive(cm.BootCount) indices := seg.NewUInt16List(1) alloc.SetActionIndices(indices) indices.Set(0, 0) txn.SetFInc(1) txn.SetTopologyVersion(0) result, err := lc.RunTransaction(&txn, true, cm.RMId) if err != nil { return nil, err } if result == nil { return nil, nil // shutting down } if result.Which() == msgs.OUTCOME_COMMIT { return nil, fmt.Errorf("Internal error: read of topology version 0 failed to abort") } abort := result.Abort() if abort.Which() == msgs.OUTCOMEABORT_RESUBMIT { continue } abortUpdates := abort.Rerun() if abortUpdates.Len() != 1 { return nil, fmt.Errorf("Internal error: read of topology version 0 gave multiple updates") } update := abortUpdates.At(0) dbversion := common.MakeTxnId(update.TxnId()) updateActions := update.Actions() if updateActions.Len() != 1 && updateActions.Len() != 2 { return nil, fmt.Errorf("Internal error: read of topology version 0 gave multiple actions: %v", updateActions.Len()) } var updateAction *msgs.Action for idx, l := 0, updateActions.Len(); idx < l; idx++ { action := updateActions.At(idx) if bytes.Equal(action.VarId(), server.TopologyVarUUId[:]) { updateAction = &action break } } if updateAction == nil { return nil, fmt.Errorf("Internal error: unable to find action for topology from read of topology version 0") } if updateAction.Which() != msgs.ACTION_WRITE { return nil, fmt.Errorf("Internal error: read of topology version 0 gave non-write action") } write := updateAction.Write() var rootPtr *msgs.VarIdPos if refs := write.References(); refs.Len() == 1 { root := refs.At(0) rootPtr = &root } return server.TopologyDeserialize(dbversion, rootPtr, write.Value()) } }
func (s *server) start() { procs := runtime.NumCPU() if procs < 2 { procs = 2 } runtime.GOMAXPROCS(procs) disk, err := mdbs.NewMDBServer(s.dataDir, mdb.WRITEMAP, 0600, goshawk.OneTB, procs/2, time.Millisecond, db.DB) s.maybeShutdown(err) s.addOnShutdown(disk.Shutdown) cm, lc := network.NewConnectionManager(s.rmId, s.bootCount, procs, disk, s.passwordHash) s.connectionManager = cm s.addOnShutdown(cm.Shutdown) s.addOnShutdown(lc.Shutdown) s.Add(1) go s.signalHandler() topologyLocal, err := network.GetTopologyFromLocalDatabase(cm, cm.Dispatchers.VarDispatcher, lc) s.maybeShutdown(err) topology, err := s.chooseTopology(topologyLocal) s.maybeShutdown(err) if topologyLocal == nil { topologyTxnId, err := network.CreateTopologyZero(cm, topology, lc) s.maybeShutdown(err) topology.DBVersion = topologyTxnId } cm.SetTopology(topology) cm.Dispatchers.VarDispatcher.ApplyToVar(func(v *eng.Var, err error) { if err != nil { log.Println("Error trying to subscribe to topology:", err) return } emptyTxnId := common.MakeTxnId([]byte{}) v.AddWriteSubscriber(emptyTxnId, func(v *eng.Var, value []byte, refs *msgs.VarIdPos_List, txn *eng.Txn) { var rootVarPosPtr *msgs.VarIdPos if refs.Len() == 1 { root := refs.At(0) rootVarPosPtr = &root } topology, err := goshawk.TopologyDeserialize(txn.Id, rootVarPosPtr, value) if err != nil { log.Println("Unable to deserialize new topology:", err) } cm.SetTopology(topology) disk.WithEnv(func(env *mdb.Env) (interface{}, error) { return nil, env.SetFlags(mdb.MAPASYNC, topology.AsyncFlush) }) }) }, false, goshawk.TopologyVarUUId) cm.AddSender(network.NewTopologyWriter(topology, lc, cm)) localHost, remoteHosts, err := topology.LocalRemoteHosts(s.port) s.maybeShutdown(err) log.Printf(">==> We are %v (%v) <==<\n", localHost, s.rmId) listener, err := network.NewListener(s.port, cm) s.maybeShutdown(err) s.addOnShutdown(listener.Shutdown) cm.SetDesiredServers(localHost, remoteHosts) defer s.shutdown(nil) s.Wait() }