arvados.go

   1 package main
   2
   3 import (
   4         "encoding/json"
   5         "errors"
   6         "fmt"
   7         "io/ioutil"
   8         "net/url"
   9         "os"
  10         "regexp"
  11         "strings"
  12         "sync"
  13         "time"
  14
  15         "git.arvados.org/arvados.git/sdk/go/arvados"
  16         "git.arvados.org/arvados.git/sdk/go/arvadosclient"
  17         "git.arvados.org/arvados.git/sdk/go/keepclient"
  18         log "github.com/sirupsen/logrus"
  19         "golang.org/x/crypto/blake2b"
  20         "golang.org/x/net/websocket"
  21 )
  22
  23 type eventMessage struct {
  24         Status     int
  25         ObjectUUID string `json:"object_uuid"`
  26         EventType  string `json:"event_type"`
  27         Properties struct {
  28                 Text string
  29         }
  30 }
  31
  32 type arvadosClient struct {
  33         *arvados.Client
  34         notifying map[string]map[chan<- eventMessage]int
  35         wantClose chan struct{}
  36         wsconn    *websocket.Conn
  37         mtx       sync.Mutex
  38 }
  39
  40 // Listen for events concerning the given uuids. When an event occurs
  41 // (and after connecting/reconnecting to the event stream), send each
  42 // uuid to ch. If a {ch, uuid} pair is subscribed twice, the uuid will
  43 // be sent only once for each update, but two Unsubscribe calls will
  44 // be needed to stop sending them.
  45 func (client *arvadosClient) Subscribe(ch chan<- eventMessage, uuid string) {
  46         client.mtx.Lock()
  47         defer client.mtx.Unlock()
  48         if client.notifying == nil {
  49                 client.notifying = map[string]map[chan<- eventMessage]int{}
  50                 client.wantClose = make(chan struct{})
  51                 go client.runNotifier()
  52         }
  53         chmap := client.notifying[uuid]
  54         if chmap == nil {
  55                 chmap = map[chan<- eventMessage]int{}
  56                 client.notifying[uuid] = chmap
  57         }
  58         needSub := true
  59         for _, nch := range chmap {
  60                 if nch > 0 {
  61                         needSub = false
  62                         break
  63                 }
  64         }
  65         chmap[ch]++
  66         if needSub && client.wsconn != nil {
  67                 go json.NewEncoder(client.wsconn).Encode(map[string]interface{}{
  68                         "method": "subscribe",
  69                         "filters": [][]interface{}{
  70                                 {"object_uuid", "=", uuid},
  71                                 {"event_type", "in", []string{"stderr", "crunch-run", "update"}},
  72                         },
  73                 })
  74         }
  75 }
  76
  77 func (client *arvadosClient) Unsubscribe(ch chan<- eventMessage, uuid string) {
  78         client.mtx.Lock()
  79         defer client.mtx.Unlock()
  80         chmap := client.notifying[uuid]
  81         if n := chmap[ch] - 1; n == 0 {
  82                 delete(chmap, ch)
  83                 if len(chmap) == 0 {
  84                         delete(client.notifying, uuid)
  85                 }
  86                 if client.wsconn != nil {
  87                         go json.NewEncoder(client.wsconn).Encode(map[string]interface{}{
  88                                 "method": "unsubscribe",
  89                                 "filters": [][]interface{}{
  90                                         {"object_uuid", "=", uuid},
  91                                         {"event_type", "in", []string{"stderr", "crunch-run", "update"}},
  92                                 },
  93                         })
  94                 }
  95         } else if n > 0 {
  96                 chmap[ch] = n
  97         }
  98 }
  99
 100 func (client *arvadosClient) Close() {
 101         client.mtx.Lock()
 102         defer client.mtx.Unlock()
 103         if client.notifying != nil {
 104                 client.notifying = nil
 105                 close(client.wantClose)
 106         }
 107 }
 108
 109 func (client *arvadosClient) runNotifier() {
 110 reconnect:
 111         for {
 112                 var cluster arvados.Cluster
 113                 err := client.RequestAndDecode(&cluster, "GET", arvados.EndpointConfigGet.Path, nil, nil)
 114                 if err != nil {
 115                         log.Warnf("error getting cluster config: %s", err)
 116                         time.Sleep(5 * time.Second)
 117                         continue reconnect
 118                 }
 119                 wsURL := cluster.Services.Websocket.ExternalURL
 120                 wsURL.Scheme = strings.Replace(wsURL.Scheme, "http", "ws", 1)
 121                 wsURL.Path = "/websocket"
 122                 wsURL.RawQuery = url.Values{"api_token": []string{client.AuthToken}}.Encode()
 123                 conn, err := websocket.Dial(wsURL.String(), "", cluster.Services.Controller.ExternalURL.String())
 124                 if err != nil {
 125                         log.Warnf("websocket connection error: %s", err)
 126                         time.Sleep(5 * time.Second)
 127                         continue reconnect
 128                 }
 129                 log.Printf("connected to websocket at %s", wsURL)
 130
 131                 client.mtx.Lock()
 132                 client.wsconn = conn
 133                 resubscribe := make([]string, 0, len(client.notifying))
 134                 for uuid := range client.notifying {
 135                         resubscribe = append(resubscribe, uuid)
 136                 }
 137                 client.mtx.Unlock()
 138
 139                 go func() {
 140                         w := json.NewEncoder(conn)
 141                         for _, uuid := range resubscribe {
 142                                 w.Encode(map[string]interface{}{
 143                                         "method": "subscribe",
 144                                         "filters": [][]interface{}{
 145                                                 {"object_uuid", "=", uuid},
 146                                                 {"event_type", "in", []string{"stderr", "crunch-run", "update"}},
 147                                         },
 148                                 })
 149                         }
 150                 }()
 151
 152                 r := json.NewDecoder(conn)
 153                 for {
 154                         var msg eventMessage
 155                         err := r.Decode(&msg)
 156                         select {
 157                         case <-client.wantClose:
 158                                 return
 159                         default:
 160                                 if err != nil {
 161                                         log.Printf("error decoding websocket message: %s", err)
 162                                         client.mtx.Lock()
 163                                         client.wsconn = nil
 164                                         client.mtx.Unlock()
 165                                         go conn.Close()
 166                                         continue reconnect
 167                                 }
 168                                 client.mtx.Lock()
 169                                 for ch := range client.notifying[msg.ObjectUUID] {
 170                                         ch <- msg
 171                                 }
 172                                 client.mtx.Unlock()
 173                         }
 174                 }
 175         }
 176 }
 177
 178 type arvadosContainerRunner struct {
 179         Client      *arvados.Client
 180         Name        string
 181         ProjectUUID string
 182         VCPUs       int
 183         RAM         int64
 184         Prog        string // if empty, run /proc/self/exe
 185         Args        []string
 186         Mounts      map[string]map[string]interface{}
 187         Priority    int
 188 }
 189
 190 func (runner *arvadosContainerRunner) Run() (string, error) {
 191         if runner.ProjectUUID == "" {
 192                 return "", errors.New("cannot run arvados container: ProjectUUID not provided")
 193         }
 194
 195         mounts := map[string]map[string]interface{}{
 196                 "/mnt/output": {
 197                         "kind":     "collection",
 198                         "writable": true,
 199                 },
 200         }
 201         for path, mnt := range runner.Mounts {
 202                 mounts[path] = mnt
 203         }
 204
 205         prog := runner.Prog
 206         if prog == "" {
 207                 prog = "/mnt/cmd/lightning"
 208                 cmdUUID, err := runner.makeCommandCollection()
 209                 if err != nil {
 210                         return "", err
 211                 }
 212                 mounts["/mnt/cmd"] = map[string]interface{}{
 213                         "kind": "collection",
 214                         "uuid": cmdUUID,
 215                 }
 216         }
 217         command := append([]string{prog}, runner.Args...)
 218
 219         priority := runner.Priority
 220         if priority < 1 {
 221                 priority = 500
 222         }
 223         rc := arvados.RuntimeConstraints{
 224                 VCPUs:        runner.VCPUs,
 225                 RAM:          runner.RAM,
 226                 KeepCacheRAM: (1 << 26) * 2 * int64(runner.VCPUs),
 227         }
 228         var cr arvados.ContainerRequest
 229         err := runner.Client.RequestAndDecode(&cr, "POST", "arvados/v1/container_requests", nil, map[string]interface{}{
 230                 "container_request": map[string]interface{}{
 231                         "owner_uuid":          runner.ProjectUUID,
 232                         "name":                runner.Name,
 233                         "container_image":     "lightning-runtime",
 234                         "command":             command,
 235                         "mounts":              mounts,
 236                         "use_existing":        true,
 237                         "output_path":         "/mnt/output",
 238                         "runtime_constraints": rc,
 239                         "priority":            runner.Priority,
 240                         "state":               arvados.ContainerRequestStateCommitted,
 241                 },
 242         })
 243         if err != nil {
 244                 return "", err
 245         }
 246         log.Printf("container request UUID: %s", cr.UUID)
 247         log.Printf("container UUID: %s", cr.ContainerUUID)
 248
 249         logch := make(chan eventMessage)
 250         client := arvadosClient{Client: runner.Client}
 251         defer client.Close()
 252         subscribedUUID := ""
 253         defer func() {
 254                 if subscribedUUID != "" {
 255                         client.Unsubscribe(logch, subscribedUUID)
 256                 }
 257         }()
 258
 259         ticker := time.NewTicker(5 * time.Second)
 260         defer ticker.Stop()
 261
 262         lastState := cr.State
 263         refreshCR := func() {
 264                 err = runner.Client.RequestAndDecode(&cr, "GET", "arvados/v1/container_requests/"+cr.UUID, nil, nil)
 265                 if err != nil {
 266                         log.Printf("error getting container request: %s", err)
 267                         return
 268                 }
 269                 if lastState != cr.State {
 270                         log.Printf("container request state: %s", cr.State)
 271                         lastState = cr.State
 272                 }
 273                 if subscribedUUID != cr.ContainerUUID {
 274                         if subscribedUUID != "" {
 275                                 client.Unsubscribe(logch, subscribedUUID)
 276                         }
 277                         client.Subscribe(logch, cr.ContainerUUID)
 278                         subscribedUUID = cr.ContainerUUID
 279                 }
 280         }
 281
 282         for cr.State != arvados.ContainerRequestStateFinal {
 283                 select {
 284                 case <-ticker.C:
 285                         refreshCR()
 286                 case msg := <-logch:
 287                         switch msg.EventType {
 288                         case "update":
 289                                 refreshCR()
 290                         default:
 291                                 for _, line := range strings.Split(msg.Properties.Text, "\n") {
 292                                         if line != "" {
 293                                                 log.Print(line)
 294                                         }
 295                                 }
 296                         }
 297                 }
 298         }
 299
 300         var c arvados.Container
 301         err = runner.Client.RequestAndDecode(&c, "GET", "arvados/v1/containers/"+cr.ContainerUUID, nil, nil)
 302         if err != nil {
 303                 return "", err
 304         } else if c.State != arvados.ContainerStateComplete {
 305                 return "", fmt.Errorf("container did not complete: %s", c.State)
 306         } else if c.ExitCode != 0 {
 307                 return "", fmt.Errorf("container exited %d", c.ExitCode)
 308         }
 309         return cr.OutputUUID, err
 310 }
 311
 312 var collectionInPathRe = regexp.MustCompile(`^(.*/)?([0-9a-f]{32}\+[0-9]+|[0-9a-z]{5}-[0-9a-z]{5}-[0-9a-z]{15})(/.*)?$`)
 313
 314 func (runner *arvadosContainerRunner) TranslatePaths(paths ...*string) error {
 315         if runner.Mounts == nil {
 316                 runner.Mounts = make(map[string]map[string]interface{})
 317         }
 318         for _, path := range paths {
 319                 if *path == "" || *path == "-" {
 320                         continue
 321                 }
 322                 m := collectionInPathRe.FindStringSubmatch(*path)
 323                 if m == nil {
 324                         return fmt.Errorf("cannot find uuid in path: %q", *path)
 325                 }
 326                 uuid := m[2]
 327                 mnt, ok := runner.Mounts["/mnt/"+uuid]
 328                 if !ok {
 329                         mnt = map[string]interface{}{
 330                                 "kind": "collection",
 331                                 "uuid": uuid,
 332                         }
 333                         runner.Mounts["/mnt/"+uuid] = mnt
 334                 }
 335                 *path = "/mnt/" + uuid + m[3]
 336         }
 337         return nil
 338 }
 339
 340 func (runner *arvadosContainerRunner) makeCommandCollection() (string, error) {
 341         exe, err := ioutil.ReadFile("/proc/self/exe")
 342         if err != nil {
 343                 return "", err
 344         }
 345         b2 := blake2b.Sum256(exe)
 346         cname := fmt.Sprintf("lightning-%x", b2)
 347         var existing arvados.CollectionList
 348         err = runner.Client.RequestAndDecode(&existing, "GET", "arvados/v1/collections", nil, arvados.ListOptions{
 349                 Limit: 1,
 350                 Count: "none",
 351                 Filters: []arvados.Filter{
 352                         {Attr: "name", Operator: "=", Operand: cname},
 353                         {Attr: "owner_uuid", Operator: "=", Operand: runner.ProjectUUID},
 354                 },
 355         })
 356         if err != nil {
 357                 return "", err
 358         }
 359         if len(existing.Items) > 0 {
 360                 uuid := existing.Items[0].UUID
 361                 log.Printf("using lightning binary in existing collection %s (name is %q; did not verify whether content matches)", uuid, cname)
 362                 return uuid, nil
 363         }
 364         log.Printf("writing lightning binary to new collection %q", cname)
 365         ac, err := arvadosclient.New(runner.Client)
 366         if err != nil {
 367                 return "", err
 368         }
 369         kc := keepclient.New(ac)
 370         var coll arvados.Collection
 371         fs, err := coll.FileSystem(runner.Client, kc)
 372         if err != nil {
 373                 return "", err
 374         }
 375         f, err := fs.OpenFile("lightning", os.O_CREATE|os.O_WRONLY, 0777)
 376         if err != nil {
 377                 return "", err
 378         }
 379         _, err = f.Write(exe)
 380         if err != nil {
 381                 return "", err
 382         }
 383         err = f.Close()
 384         if err != nil {
 385                 return "", err
 386         }
 387         mtxt, err := fs.MarshalManifest(".")
 388         if err != nil {
 389                 return "", err
 390         }
 391         err = runner.Client.RequestAndDecode(&coll, "POST", "arvados/v1/collections", nil, map[string]interface{}{
 392                 "collection": map[string]interface{}{
 393                         "owner_uuid":    runner.ProjectUUID,
 394                         "manifest_text": mtxt,
 395                         "name":          cname,
 396                 },
 397         })
 398         if err != nil {
 399                 return "", err
 400         }
 401         log.Printf("stored lightning binary in new collection %s", coll.UUID)
 402         return coll.UUID, nil
 403 }