func (a *apiServer) Version(version int64) error {
protolog.Printf("apiServer.Version %d", version)
a.versionLock.Lock()
defer a.versionLock.Unlock()
protolog.Printf("got lock")
a.version = version
return nil
}
func (a *sharder) announceServer(
address string,
server Server,
versionChan chan int64,
cancel chan bool,
) error {
serverState := &ServerState{
Address: address,
Version: InvalidVersion,
}
for {
shards, err := server.LocalShards()
if err != nil {
return err
}
serverState.Shards = shards
encodedServerState, err := marshaler.MarshalToString(serverState)
if err != nil {
return err
}
if err := a.discoveryClient.Set(a.serverStateKey(address), encodedServerState, holdTTL); err != nil {
protolog.Printf("Error setting server state: %s", err.Error())
}
protolog.Debug(&SetServerState{serverState})
select {
case <-cancel:
return nil
case version := <-versionChan:
serverState.Version = version
case <-time.After(time.Second * time.Duration(holdTTL/2)):
}
}
}
func getFinalPipelineRunStatus(apiClient pps.ApiClient, pipelineRunID string) (*pps.PipelineRunStatus, error) {
// TODO(pedge): not good
ticker := time.NewTicker(time.Second)
for i := 0; i < 20; i++ {
<-ticker.C
pipelineRunStatuses, err := apiClient.GetPipelineRunStatus(
context.Background(),
&pps.GetPipelineRunStatusRequest{
PipelineRunId: pipelineRunID,
},
)
if err != nil {
return nil, err
}
pipelineRunStatus := pipelineRunStatuses.PipelineRunStatus[0]
protolog.Printf("status at tick %d: %v\n", i, pipelineRunStatus)
switch pipelineRunStatus.PipelineRunStatusType {
case pps.PipelineRunStatusType_PIPELINE_RUN_STATUS_TYPE_ERROR:
return pipelineRunStatus, nil
case pps.PipelineRunStatusType_PIPELINE_RUN_STATUS_TYPE_SUCCESS:
return pipelineRunStatus, nil
}
}
return nil, fmt.Errorf("did not get final pipeline status for %s", pipelineRunID)
}
func (a *apiServer) CreatePipeline(ctx context.Context, request *pps.CreatePipelineRequest) (response *google_protobuf.Empty, err error) {
defer func(start time.Time) { a.Log(request, response, err, time.Since(start)) }(time.Now())
if request.Pipeline == nil {
return nil, fmt.Errorf("pachyderm.pps.pipelineserver: request.Pipeline cannot be nil")
}
repoSet := make(map[string]bool)
for _, input := range request.Inputs {
repoSet[input.Repo.Name] = true
}
if len(repoSet) < len(request.Inputs) {
return nil, fmt.Errorf("pachyderm.pps.pipelineserver: duplicate input repos")
}
repo := pps.PipelineRepo(request.Pipeline)
persistPipelineInfo := &persist.PipelineInfo{
PipelineName: request.Pipeline.Name,
Transform: request.Transform,
Shards: request.Shards,
Inputs: request.Inputs,
OutputRepo: repo,
}
if _, err := a.persistAPIServer.CreatePipelineInfo(ctx, persistPipelineInfo); err != nil {
return nil, err
}
if _, err := a.pfsAPIClient.CreateRepo(ctx, &pfs.CreateRepoRequest{Repo: repo}); err != nil {
return nil, err
}
go func() {
if err := a.runPipeline(newPipelineInfo(persistPipelineInfo)); err != nil {
protolog.Printf("pipeline errored: %s", err.Error())
}
}()
return google_protobuf.EmptyInstance, nil
}
func (a *apiServer) CreatePipeline(ctx context.Context, request *pps.CreatePipelineRequest) (response *google_protobuf.Empty, err error) {
defer func(start time.Time) { a.Log(request, response, err, time.Since(start)) }(time.Now())
if request.Pipeline == nil {
return nil, fmt.Errorf("pachyderm.pps.pipelineserver: request.Pipeline cannot be nil")
}
persistPipelineInfo := &persist.PipelineInfo{
PipelineName: request.Pipeline.Name,
Transform: request.Transform,
Shards: request.Shards,
InputRepo: request.InputRepo,
}
if _, err := a.persistAPIServer.CreatePipelineInfo(ctx, persistPipelineInfo); err != nil {
return nil, err
}
repo := pps.PipelineRepo(request.Pipeline)
if _, err := a.pfsAPIClient.CreateRepo(ctx, &pfs.CreateRepoRequest{Repo: repo}); err != nil {
return nil, err
}
go func() {
if err := a.runPipeline(persistPipelineInfoToPipelineInfo(persistPipelineInfo)); err != nil {
protolog.Printf("pipeline errored: %s", err.Error())
}
}()
return google_protobuf.EmptyInstance, nil
}
func (a *sharder) announceFrontend(
address string,
frontend Frontend,
versionChan chan int64,
cancel chan bool,
) error {
frontendState := &FrontendState{
Address: address,
Version: InvalidVersion,
}
for {
encodedFrontendState, err := marshaler.MarshalToString(frontendState)
if err != nil {
return err
}
if err := a.discoveryClient.Set(a.frontendStateKey(address), encodedFrontendState, holdTTL); err != nil {
protolog.Printf("Error setting server state: %s", err.Error())
}
protolog.Debug(&SetFrontendState{frontendState})
select {
case <-cancel:
return nil
case version := <-versionChan:
frontendState.Version = version
case <-time.After(time.Second * time.Duration(holdTTL/2)):
}
}
}
func (d *driver) PullDiff(commit *pfs.Commit, shard uint64, diff io.Writer) error {
protolog.Printf("PullDiff: %s", d.readCommitPath(commit, shard))
parent, err := d.getParent(commit, shard)
if err != nil {
return err
}
if parent == nil {
return execSend(d.readCommitPath(commit, shard), "", diff)
}
return execSend(d.readCommitPath(commit, shard), d.readCommitPath(parent, shard), diff)
}
func (d *driver) PushDiff(commit *pfs.Commit, shard uint64, diff io.Reader) error {
protolog.Printf("PushDiff: %s", d.readCommitPath(commit, shard))
if err := execSubvolumeCreate(d.commitPathNoShard(commit)); err != nil && !execSubvolumeExists(d.commitPathNoShard(commit)) {
return err
}
if err := execRecv(d.commitPathNoShard(commit), diff); err != nil {
return err
}
if !execSubvolumeExists(d.readCommitPath(commit, shard)) {
return fmt.Errorf("PushDiff failed %s", d.readCommitPath(commit, shard))
}
return nil
}
func do(appEnvObj interface{}) error {
appEnv := appEnvObj.(*appEnv)
rethinkAPIClient, err := getRethinkAPIClient(appEnv.DatabaseAddress, appEnv.DatabaseName)
if err != nil {
return err
}
pfsdAddress, err := getPfsdAddress()
if err != nil {
return err
}
clientConn, err := grpc.Dial(pfsdAddress, grpc.WithInsecure())
if err != nil {
return err
}
pfsAPIClient := pfs.NewAPIClient(clientConn)
kubeAddr, err := getKubeAddress()
if err != nil {
return err
}
config := &kube.Config{
Host: kubeAddr,
Insecure: true,
}
kubeClient, err := kube.New(config)
if err != nil {
protolog.Printf("Error creating kubernetes client: %s", err.Error())
}
jobAPIServer := jobserver.NewAPIServer(
pfsAPIClient,
rethinkAPIClient,
kubeClient,
)
jobAPIClient := pps.NewLocalJobAPIClient(jobAPIServer)
pipelineAPIServer := pipelineserver.NewAPIServer(pfsAPIClient, jobAPIClient, rethinkAPIClient)
if err := pipelineAPIServer.Start(); err != nil {
return err
}
return protoserver.Serve(
uint16(appEnv.Port),
func(s *grpc.Server) {
pps.RegisterJobAPIServer(s, jobAPIServer)
pps.RegisterInternalJobAPIServer(s, jobAPIServer)
pps.RegisterPipelineAPIServer(s, pipelineAPIServer)
},
protoserver.ServeOptions{
DebugPort: uint16(appEnv.DebugPort),
Version: pachyderm.Version,
},
)
}
func (c *etcdClient) WatchAll(key string, cancel chan bool, callBack func(map[string]string) error) error {
for {
if err := c.watchAllWithoutRetry(key, cancel, callBack); err != nil {
etcdErr, ok := err.(*etcd.EtcdError)
if ok && etcdErr.ErrorCode == 401 {
continue
}
if ok && etcdErr.ErrorCode == 501 {
protolog.Printf("Ignoring: %s", err.Error())
continue
}
return err
}
}
}
func (a *apiServer) Start() error {
pipelineInfos, err := a.ListPipeline(context.Background(), &pps.ListPipelineRequest{})
if err != nil {
return err
}
for _, pipelineInfo := range pipelineInfos.PipelineInfo {
pipelineInfo := pipelineInfo
go func() {
if err := a.runPipeline(pipelineInfo); err != nil {
protolog.Printf("pipeline errored: %s", err.Error())
}
}()
}
return nil
}
func (c *etcdClient) Watch(key string, cancel chan bool, callBack func(string) error) error {
// This retry is needed for when the etcd cluster gets overloaded.
for {
if err := c.watchWithoutRetry(key, cancel, callBack); err != nil {
etcdErr, ok := err.(*etcd.EtcdError)
if ok && etcdErr.ErrorCode == 401 {
continue
}
if ok && etcdErr.ErrorCode == 501 {
protolog.Printf("Ignoring: %s", err.Error())
continue
}
return err
}
}
}
func PutFile(apiClient pfs.ApiClient, repoName string, commitID string, path string, offset int64, reader io.Reader) (_ int, retErr error) {
putFileClient, err := apiClient.PutFile(context.Background())
if err != nil {
return 0, err
}
defer func() {
if _, err := putFileClient.CloseAndRecv(); err != nil && retErr != nil {
retErr = err
}
}()
request := pfs.PutFileRequest{
File: &pfs.File{
Commit: &pfs.Commit{
Repo: &pfs.Repo{
Name: repoName,
},
Id: commitID,
},
Path: path,
},
FileType: pfs.FileType_FILE_TYPE_REGULAR,
OffsetBytes: offset,
}
var size int
for {
value := make([]byte, chunkSize)
iSize, err := reader.Read(value)
request.Value = value[0:iSize]
if err != nil {
if err == io.EOF {
break
}
return 0, err
}
size += iSize
protolog.Printf("pfsutil.PutFile Send(%+v)", request)
if err := putFileClient.Send(&request); err != nil {
return 0, err
}
}
if err != nil && err != io.EOF {
return 0, err
}
return size, err
}
func (a *apiServer) CreatePipeline(ctx context.Context, request *pps.CreatePipelineRequest) (response *google_protobuf.Empty, err error) {
defer func(start time.Time) { a.Log(request, response, err, time.Since(start)) }(time.Now())
persistPipelineInfo := &persist.PipelineInfo{
PipelineName: request.Pipeline.Name,
Transform: request.Transform,
InputRepo: request.InputRepo,
OutputRepo: request.OutputRepo,
}
if _, err := a.persistAPIClient.CreatePipelineInfo(ctx, persistPipelineInfo); err != nil {
return nil, err
}
go func() {
if err := a.runPipeline(persistPipelineInfoToPipelineInfo(persistPipelineInfo)); err != nil {
protolog.Printf("pipeline errored: %s", err.Error())
}
}()
return google_protobuf.EmptyInstance, nil
}
func getKubeClient() (*kube.Client, error) {
kubeAddr, err := getKubeAddress()
if err != nil {
return nil, err
}
config := &kube.Config{
Host: kubeAddr,
Insecure: true,
}
kubeClient, err := kube.New(config)
if err != nil {
protolog.Printf("Error insecure kube client: %s", err.Error())
}
if kubeClient != nil {
return kubeClient, nil
}
return kube.NewInCluster()
}
func getFinalPipelineRunStatus(apiClient pps.ApiClient, pipelineRunID string) (*pps.PipelineRunStatus, error) {
// TODO(pedge): not good
ticker := time.NewTicker(time.Second)
for i := 0; i < 60; i++ {
<-ticker.C
getPipelineRunStatusResponse, err := ppsutil.GetPipelineRunStatus(
apiClient,
pipelineRunID,
)
if err != nil {
return nil, err
}
protolog.Printf("status at tick %d: %v\n", i, getPipelineRunStatusResponse.PipelineRunStatus)
pipelineRunStatus := getPipelineRunStatusResponse.PipelineRunStatus
switch pipelineRunStatus.PipelineRunStatusType {
case pps.PipelineRunStatusType_PIPELINE_RUN_STATUS_TYPE_ERROR:
return pipelineRunStatus, nil
case pps.PipelineRunStatusType_PIPELINE_RUN_STATUS_TYPE_SUCCESS:
return pipelineRunStatus, nil
}
}
return nil, fmt.Errorf("did not get final pipeline status for %s", pipelineRunID)
}
func (a *discoveryAddresser) fillRoles(
id string,
server Server,
versionChan chan int64,
cancel chan bool,
) error {
oldRoles := make(map[int64]proto.ServerRole)
return a.discoveryClient.WatchAll(
a.serverRoleKey(id),
cancel,
func(encodedServerRoles map[string]string) error {
roles := make(map[int64]proto.ServerRole)
var versions int64Slice
// Decode the roles
for _, encodedServerRole := range encodedServerRoles {
var serverRole proto.ServerRole
if err := jsonpb.UnmarshalString(encodedServerRole, &serverRole); err != nil {
return err
}
roles[serverRole.Version] = serverRole
versions = append(versions, serverRole.Version)
}
sort.Sort(versions)
// For each new version bring the server up to date
for _, version := range versions {
if _, ok := oldRoles[version]; ok {
// we've already seen these roles, so nothing to do here
continue
}
serverRole := roles[version]
protolog.Printf("StartAddServerRole %+v", serverRole)
var wg sync.WaitGroup
var addShardErr error
var addShardOnce sync.Once
for _, shard := range shards(serverRole) {
if !containsShard(oldRoles, shard) {
wg.Add(1)
go func(shard uint64) {
defer wg.Done()
if err := server.AddShard(shard); err != nil {
addShardOnce.Do(func() {
addShardErr = err
})
}
}(shard)
}
}
wg.Wait()
if addShardErr != nil {
protolog.Info(&log.AddServerRole{&serverRole, addShardErr.Error()})
return addShardErr
}
protolog.Info(&log.AddServerRole{&serverRole, ""})
oldRoles[version] = serverRole
versionChan <- version
}
// See if there are any old roles that aren't needed
for version, serverRole := range oldRoles {
var wg sync.WaitGroup
var removeShardErr error
var removeShardOnce sync.Once
if _, ok := roles[version]; ok {
// these roles haven't expired yet, so nothing to do
continue
}
for _, shard := range shards(serverRole) {
if !containsShard(roles, shard) {
wg.Add(1)
go func(shard uint64) {
defer wg.Done()
if err := server.RemoveShard(shard); err != nil {
removeShardOnce.Do(func() {
removeShardErr = err
})
}
}(shard)
}
}
wg.Wait()
if removeShardErr != nil {
protolog.Info(&log.RemoveServerRole{&serverRole, removeShardErr.Error()})
return removeShardErr
}
protolog.Info(&log.RemoveServerRole{&serverRole, ""})
}
oldRoles = make(map[int64]proto.ServerRole)
for version, serverRole := range roles {
oldRoles[version] = serverRole
}
return nil
},
)
}
func BenchmarkFreeformfGLog(b *testing.B) {
runBenchmarkGLog(b, func() { protolog.Printf("%s %d\n", s, d) }, false)
}
func BenchmarkThreadFreeformfGLog(b *testing.B) {
runBenchmarkGLog(b, func() { protolog.Printf("%s %d\n", s, d) }, true)
}
func (a *sharder) WaitForAvailability(frontendAddresses []string, serverAddresses []string) error {
version := InvalidVersion
if err := a.discoveryClient.WatchAll(a.serverDir(), nil,
func(encodedServerStatesAndRoles map[string]string) error {
serverStates := make(map[string]*ServerState)
serverRoles := make(map[string]map[int64]*ServerRole)
for key, encodedServerStateOrRole := range encodedServerStatesAndRoles {
if strings.HasPrefix(key, a.serverStateDir()) {
serverState, err := decodeServerState(encodedServerStateOrRole)
if err != nil {
return err
}
serverStates[serverState.Address] = serverState
}
if strings.HasPrefix(key, a.serverRoleDir()) {
serverRole, err := decodeServerRole(encodedServerStateOrRole)
if err != nil {
return err
}
if _, ok := serverRoles[serverRole.Address]; !ok {
serverRoles[serverRole.Address] = make(map[int64]*ServerRole)
}
serverRoles[serverRole.Address][serverRole.Version] = serverRole
}
}
if len(serverStates) != len(serverAddresses) {
return nil
}
if len(serverRoles) != len(serverAddresses) {
return nil
}
for _, address := range serverAddresses {
if _, ok := serverStates[address]; !ok {
return nil
}
if _, ok := serverRoles[address]; !ok {
return nil
}
}
versions := make(map[int64]bool)
for _, serverState := range serverStates {
if serverState.Version == InvalidVersion {
return nil
}
versions[serverState.Version] = true
}
if len(versions) != 1 {
return nil
}
for _, versionToServerRole := range serverRoles {
if len(versionToServerRole) != 1 {
return nil
}
for version := range versionToServerRole {
if !versions[version] {
return nil
}
}
}
// This loop actually does something, it sets the outside
// version variable.
for version = range versions {
}
return errComplete
}); err != errComplete {
return err
}
if err := a.discoveryClient.WatchAll(
a.frontendStateDir(),
nil,
func(encodedFrontendStates map[string]string) error {
frontendStates := make(map[string]*FrontendState)
for _, encodedFrontendState := range encodedFrontendStates {
frontendState, err := decodeFrontendState(encodedFrontendState)
if err != nil {
return err
}
if frontendState.Version != version {
protolog.Printf("Wrong version: %d != %d", frontendState.Version, version)
return nil
}
frontendStates[frontendState.Address] = frontendState
}
protolog.Printf("frontendStates: %+v", frontendStates)
if len(frontendStates) != len(frontendAddresses) {
return nil
}
for _, address := range frontendAddresses {
if _, ok := frontendStates[address]; !ok {
return nil
}
}
return errComplete
}); err != nil && err != errComplete {
return err
}
return nil
}
func do(appEnvObj interface{}) error {
appEnv := appEnvObj.(*appEnv)
discoveryClient, err := getEtcdClient()
if err != nil {
return err
}
address := appEnv.Address
if address == "" {
address, err = netutil.ExternalIP()
if err != nil {
return err
}
}
address = fmt.Sprintf("%s:%d", address, appEnv.Port)
sharder := shard.NewSharder(
discoveryClient,
appEnv.NumShards,
appEnv.NumReplicas,
"namespace",
)
var driver drive.Driver
switch appEnv.DriverType {
case "btrfs":
driver, err = btrfs.NewDriver(appEnv.DriverRoot, "")
if err != nil {
return err
}
default:
return fmt.Errorf("unknown value for PFS_DRIVER_TYPE: %s", appEnv.DriverType)
}
apiServer := server.NewAPIServer(
route.NewSharder(
appEnv.NumShards,
0,
),
route.NewRouter(
sharder,
grpcutil.NewDialer(
grpc.WithInsecure(),
),
address,
),
)
go func() {
if err := sharder.RegisterFrontend(nil, address, apiServer); err != nil {
protolog.Printf("Error from sharder.RegisterFrontend %s", err.Error())
}
}()
internalAPIServer := server.NewInternalAPIServer(
route.NewSharder(
appEnv.NumShards,
0,
),
route.NewRouter(
sharder,
grpcutil.NewDialer(
grpc.WithInsecure(),
),
address,
),
driver,
)
go func() {
if err := sharder.Register(nil, address, internalAPIServer); err != nil {
protolog.Printf("Error from sharder.Register %s", err.Error())
}
}()
return protoserver.Serve(
uint16(appEnv.Port),
func(s *grpc.Server) {
pfs.RegisterAPIServer(s, apiServer)
pfs.RegisterInternalAPIServer(s, internalAPIServer)
},
protoserver.ServeOptions{
HTTPPort: uint16(appEnv.HTTPPort),
DebugPort: uint16(appEnv.DebugPort),
Version: pachyderm.Version,
HTTPRegisterFunc: func(ctx context.Context, mux *runtime.ServeMux, clientConn *grpc.ClientConn) error {
return pfs.RegisterAPIHandler(ctx, mux, clientConn)
},
},
)
}
