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                 client.mtx.Lock()
 130                 client.wsconn = conn
 131                 client.mtx.Unlock()
 132
 133                 w := json.NewEncoder(conn)
 134                 for uuid := range client.notifying {
 135                         w.Encode(map[string]interface{}{
 136                                 "method": "subscribe",
 137                                 "filters": [][]interface{}{
 138                                         {"object_uuid", "=", uuid},
 139                                         {"event_type", "in", []string{"stderr", "crunch-run", "update"}},
 140                                 },
 141                         })
 142                 }
 143
 144                 r := json.NewDecoder(conn)
 145                 for {
 146                         var msg eventMessage
 147                         err := r.Decode(&msg)
 148                         select {
 149                         case <-client.wantClose:
 150                                 return
 151                         default:
 152                                 if err != nil {
 153                                         log.Printf("error decoding websocket message: %s", err)
 154                                         client.mtx.Lock()
 155                                         client.wsconn = nil
 156                                         client.mtx.Unlock()
 157                                         go conn.Close()
 158                                         continue reconnect
 159                                 }
 160                                 for ch := range client.notifying[msg.ObjectUUID] {
 161                                         ch <- msg
 162                                 }
 163                         }
 164                 }
 165         }
 166 }
 167
 168 type arvadosContainerRunner struct {
 169         Client      *arvados.Client
 170         Name        string
 171         ProjectUUID string
 172         VCPUs       int
 173         RAM         int64
 174         Prog        string // if empty, run /proc/self/exe
 175         Args        []string
 176         Mounts      map[string]map[string]interface{}
 177         Priority    int
 178 }
 179
 180 func (runner *arvadosContainerRunner) Run() (string, error) {
 181         if runner.ProjectUUID == "" {
 182                 return "", errors.New("cannot run arvados container: ProjectUUID not provided")
 183         }
 184
 185         mounts := map[string]map[string]interface{}{
 186                 "/mnt/output": {
 187                         "kind":     "tmp",
 188                         "writable": true,
 189                         "capacity": 100000000000,
 190                 },
 191         }
 192         for path, mnt := range runner.Mounts {
 193                 mounts[path] = mnt
 194         }
 195
 196         prog := runner.Prog
 197         if prog == "" {
 198                 prog = "/mnt/cmd/lightning"
 199                 cmdUUID, err := runner.makeCommandCollection()
 200                 if err != nil {
 201                         return "", err
 202                 }
 203                 mounts["/mnt/cmd"] = map[string]interface{}{
 204                         "kind": "collection",
 205                         "uuid": cmdUUID,
 206                 }
 207         }
 208         command := append([]string{prog}, runner.Args...)
 209
 210         priority := runner.Priority
 211         if priority < 1 {
 212                 priority = 500
 213         }
 214         rc := arvados.RuntimeConstraints{
 215                 VCPUs:        runner.VCPUs,
 216                 RAM:          runner.RAM,
 217                 KeepCacheRAM: (1 << 26) * 2 * int64(runner.VCPUs),
 218         }
 219         var cr arvados.ContainerRequest
 220         err := runner.Client.RequestAndDecode(&cr, "POST", "arvados/v1/container_requests", nil, map[string]interface{}{
 221                 "container_request": map[string]interface{}{
 222                         "owner_uuid":          runner.ProjectUUID,
 223                         "name":                runner.Name,
 224                         "container_image":     "lightning-runtime",
 225                         "command":             command,
 226                         "mounts":              mounts,
 227                         "use_existing":        true,
 228                         "output_path":         "/mnt/output",
 229                         "runtime_constraints": rc,
 230                         "priority":            runner.Priority,
 231                         "state":               arvados.ContainerRequestStateCommitted,
 232                 },
 233         })
 234         if err != nil {
 235                 return "", err
 236         }
 237         log.Printf("container request UUID: %s", cr.UUID)
 238         log.Printf("container UUID: %s", cr.ContainerUUID)
 239
 240         logch := make(chan eventMessage)
 241         client := arvadosClient{Client: runner.Client}
 242         defer client.Close()
 243         subscribedUUID := ""
 244         defer func() {
 245                 if subscribedUUID != "" {
 246                         client.Unsubscribe(logch, subscribedUUID)
 247                 }
 248         }()
 249
 250         ticker := time.NewTicker(5 * time.Second)
 251         defer ticker.Stop()
 252
 253         lastState := cr.State
 254         refreshCR := func() {
 255                 err = runner.Client.RequestAndDecode(&cr, "GET", "arvados/v1/container_requests/"+cr.UUID, nil, nil)
 256                 if err != nil {
 257                         log.Printf("error getting container request: %s", err)
 258                         return
 259                 }
 260                 if lastState != cr.State {
 261                         log.Printf("container state: %s", cr.State)
 262                         lastState = cr.State
 263                 }
 264                 if subscribedUUID != cr.ContainerUUID {
 265                         if subscribedUUID != "" {
 266                                 client.Unsubscribe(logch, subscribedUUID)
 267                         }
 268                         client.Subscribe(logch, cr.ContainerUUID)
 269                         subscribedUUID = cr.ContainerUUID
 270                 }
 271         }
 272
 273         for cr.State != arvados.ContainerRequestStateFinal {
 274                 select {
 275                 case <-ticker.C:
 276                         refreshCR()
 277                 case msg := <-logch:
 278                         switch msg.EventType {
 279                         case "update":
 280                                 refreshCR()
 281                         default:
 282                                 for _, line := range strings.Split(msg.Properties.Text, "\n") {
 283                                         if line != "" {
 284                                                 log.Print(line)
 285                                         }
 286                                 }
 287                         }
 288                 }
 289         }
 290
 291         var c arvados.Container
 292         err = runner.Client.RequestAndDecode(&c, "GET", "arvados/v1/containers/"+cr.ContainerUUID, nil, nil)
 293         if err != nil {
 294                 return "", err
 295         }
 296         if c.ExitCode != 0 {
 297                 return "", fmt.Errorf("container exited %d", c.ExitCode)
 298         }
 299         return cr.OutputUUID, err
 300 }
 301
 302 var collectionInPathRe = regexp.MustCompile(`^(.*/)?([0-9a-f]{32}\+[0-9]+|[0-9a-z]{5}-[0-9a-z]{5}-[0-9a-z]{15})(/.*)?$`)
 303
 304 func (runner *arvadosContainerRunner) TranslatePaths(paths ...*string) error {
 305         if runner.Mounts == nil {
 306                 runner.Mounts = make(map[string]map[string]interface{})
 307         }
 308         for _, path := range paths {
 309                 if *path == "" || *path == "-" {
 310                         continue
 311                 }
 312                 m := collectionInPathRe.FindStringSubmatch(*path)
 313                 if m == nil {
 314                         return fmt.Errorf("cannot find uuid in path: %q", *path)
 315                 }
 316                 uuid := m[2]
 317                 mnt, ok := runner.Mounts["/mnt/"+uuid]
 318                 if !ok {
 319                         mnt = map[string]interface{}{
 320                                 "kind": "collection",
 321                                 "uuid": uuid,
 322                         }
 323                         runner.Mounts["/mnt/"+uuid] = mnt
 324                 }
 325                 *path = "/mnt/" + uuid + m[3]
 326         }
 327         return nil
 328 }
 329
 330 func (runner *arvadosContainerRunner) makeCommandCollection() (string, error) {
 331         exe, err := ioutil.ReadFile("/proc/self/exe")
 332         if err != nil {
 333                 return "", err
 334         }
 335         b2 := blake2b.Sum256(exe)
 336         cname := fmt.Sprintf("lightning-%x", b2)
 337         var existing arvados.CollectionList
 338         err = runner.Client.RequestAndDecode(&existing, "GET", "arvados/v1/collections", nil, arvados.ListOptions{
 339                 Limit: 1,
 340                 Count: "none",
 341                 Filters: []arvados.Filter{
 342                         {Attr: "name", Operator: "=", Operand: cname},
 343                         {Attr: "owner_uuid", Operator: "=", Operand: runner.ProjectUUID},
 344                 },
 345         })
 346         if err != nil {
 347                 return "", err
 348         }
 349         if len(existing.Items) > 0 {
 350                 uuid := existing.Items[0].UUID
 351                 log.Printf("using lightning binary in existing collection %s (name is %q; did not verify whether content matches)", uuid, cname)
 352                 return uuid, nil
 353         }
 354         log.Printf("writing lightning binary to new collection %q", cname)
 355         ac, err := arvadosclient.New(runner.Client)
 356         if err != nil {
 357                 return "", err
 358         }
 359         kc := keepclient.New(ac)
 360         var coll arvados.Collection
 361         fs, err := coll.FileSystem(runner.Client, kc)
 362         if err != nil {
 363                 return "", err
 364         }
 365         f, err := fs.OpenFile("lightning", os.O_CREATE|os.O_WRONLY, 0777)
 366         if err != nil {
 367                 return "", err
 368         }
 369         _, err = f.Write(exe)
 370         if err != nil {
 371                 return "", err
 372         }
 373         err = f.Close()
 374         if err != nil {
 375                 return "", err
 376         }
 377         mtxt, err := fs.MarshalManifest(".")
 378         if err != nil {
 379                 return "", err
 380         }
 381         err = runner.Client.RequestAndDecode(&coll, "POST", "arvados/v1/collections", nil, map[string]interface{}{
 382                 "collection": map[string]interface{}{
 383                         "owner_uuid":    runner.ProjectUUID,
 384                         "manifest_text": mtxt,
 385                         "name":          cname,
 386                 },
 387         })
 388         if err != nil {
 389                 return "", err
 390         }
 391         log.Printf("stored lightning binary in new collection %s", coll.UUID)
 392         return coll.UUID, nil
 393 }