lib/crunchrun/singularity.go

   1 // Copyright (C) The Arvados Authors. All rights reserved.
   2 //
   3 // SPDX-License-Identifier: AGPL-3.0
   4
   5 package crunchrun
   6
   7 import (
   8         "fmt"
   9         "io/ioutil"
  10         "os"
  11         "os/exec"
  12         "sort"
  13         "syscall"
  14         "time"
  15
  16         "git.arvados.org/arvados.git/sdk/go/arvados"
  17         "golang.org/x/net/context"
  18 )
  19
  20 type singularityExecutor struct {
  21         logf          func(string, ...interface{})
  22         spec          containerSpec
  23         tmpdir        string
  24         child         *exec.Cmd
  25         imageFilename string // "sif" image
  26 }
  27
  28 func newSingularityExecutor(logf func(string, ...interface{})) (*singularityExecutor, error) {
  29         tmpdir, err := ioutil.TempDir("", "crunch-run-singularity-")
  30         if err != nil {
  31                 return nil, err
  32         }
  33         return &singularityExecutor{
  34                 logf:   logf,
  35                 tmpdir: tmpdir,
  36         }, nil
  37 }
  38
  39 func (e *singularityExecutor) Runtime() string { return "singularity" }
  40
  41 func (e *singularityExecutor) getOrCreateProject(ownerUuid string, name string, containerClient *arvados.Client) (*arvados.Group, error) {
  42         var gp arvados.GroupList
  43         err := containerClient.RequestAndDecode(&gp,
  44                 arvados.EndpointGroupList.Method,
  45                 arvados.EndpointGroupList.Path,
  46                 nil, arvados.ListOptions{Filters: []arvados.Filter{
  47                         arvados.Filter{"owner_uuid", "=", ownerUuid},
  48                         arvados.Filter{"name", "=", name},
  49                         arvados.Filter{"group_class", "=", "project"},
  50                 },
  51                         Limit: 1})
  52         if err != nil {
  53                 return nil, err
  54         }
  55         if len(gp.Items) == 1 {
  56                 return &gp.Items[0], nil
  57         }
  58
  59         var rgroup arvados.Group
  60         err = containerClient.RequestAndDecode(&rgroup,
  61                 arvados.EndpointGroupCreate.Method,
  62                 arvados.EndpointGroupCreate.Path,
  63                 nil, map[string]interface{}{
  64                         "group": map[string]string{
  65                                 "owner_uuid":  ownerUuid,
  66                                 "name":        name,
  67                                 "group_class": "project",
  68                         },
  69                 })
  70         if err != nil {
  71                 return nil, err
  72         }
  73         return &rgroup, nil
  74 }
  75
  76 func (e *singularityExecutor) checkImageCache(dockerImageID string, container arvados.Container, arvMountPoint string,
  77         containerClient *arvados.Client) (collection *arvados.Collection, err error) {
  78
  79         // Cache the image to keep
  80         cacheGroup, err := e.getOrCreateProject(container.RuntimeUserUUID, ".cache", containerClient)
  81         if err != nil {
  82                 return nil, fmt.Errorf("error getting '.cache' project: %v", err)
  83         }
  84         imageGroup, err := e.getOrCreateProject(cacheGroup.UUID, "auto-generated singularity images", containerClient)
  85         if err != nil {
  86                 return nil, fmt.Errorf("error getting 'auto-generated singularity images' project: %s", err)
  87         }
  88
  89         collectionName := fmt.Sprintf("singularity image for %v", dockerImageID)
  90         var cl arvados.CollectionList
  91         err = containerClient.RequestAndDecode(&cl,
  92                 arvados.EndpointCollectionList.Method,
  93                 arvados.EndpointCollectionList.Path,
  94                 nil, arvados.ListOptions{Filters: []arvados.Filter{
  95                         arvados.Filter{"owner_uuid", "=", imageGroup.UUID},
  96                         arvados.Filter{"name", "=", collectionName},
  97                 },
  98                         Limit: 1})
  99         if err != nil {
 100                 return nil, fmt.Errorf("error querying for collection '%v': %v", collectionName, err)
 101         }
 102         var imageCollection arvados.Collection
 103         if len(cl.Items) == 1 {
 104                 imageCollection = cl.Items[0]
 105         } else {
 106                 collectionName := "converting " + collectionName
 107                 exp := time.Now().Add(24 * 7 * 2 * time.Hour)
 108                 err = containerClient.RequestAndDecode(&imageCollection,
 109                         arvados.EndpointCollectionCreate.Method,
 110                         arvados.EndpointCollectionCreate.Path,
 111                         nil, map[string]interface{}{
 112                                 "collection": map[string]string{
 113                                         "owner_uuid": imageGroup.UUID,
 114                                         "name":       collectionName,
 115                                         "trash_at":   exp.UTC().Format(time.RFC3339),
 116                                 },
 117                                 "ensure_unique_name": true,
 118                         })
 119                 if err != nil {
 120                         return nil, fmt.Errorf("error creating '%v' collection: %s", collectionName, err)
 121                 }
 122
 123         }
 124
 125         return &imageCollection, nil
 126 }
 127
 128 // LoadImage will satisfy ContainerExecuter interface transforming
 129 // containerImage into a sif file for later use.
 130 func (e *singularityExecutor) LoadImage(dockerImageID string, imageTarballPath string, container arvados.Container, arvMountPoint string,
 131         containerClient *arvados.Client) error {
 132
 133         var imageFilename string
 134         var sifCollection *arvados.Collection
 135         var err error
 136         if containerClient != nil {
 137                 sifCollection, err = e.checkImageCache(dockerImageID, container, arvMountPoint, containerClient)
 138                 if err != nil {
 139                         return err
 140                 }
 141                 imageFilename = fmt.Sprintf("%s/by_uuid/%s/image.sif", arvMountPoint, sifCollection.UUID)
 142         } else {
 143                 imageFilename = e.tmpdir + "/image.sif"
 144         }
 145
 146         if _, err := os.Stat(imageFilename); os.IsNotExist(err) {
 147                 // Make sure the docker image is readable, and error
 148                 // out if not.
 149                 if _, err := os.Stat(imageTarballPath); err != nil {
 150                         return err
 151                 }
 152
 153                 e.logf("building singularity image")
 154                 // "singularity build" does not accept a
 155                 // docker-archive://... filename containing a ":" character,
 156                 // as in "/path/to/sha256:abcd...1234.tar". Workaround: make a
 157                 // symlink that doesn't have ":" chars.
 158                 err := os.Symlink(imageTarballPath, e.tmpdir+"/image.tar")
 159                 if err != nil {
 160                         return err
 161                 }
 162
 163                 err = os.Mkdir(e.tmpdir+"/cache", 0700)
 164                 if err != nil {
 165                         return err
 166                 }
 167                 defer os.RemoveAll(e.tmpdir + "/cache")
 168
 169                 build := exec.Command("singularity", "build", imageFilename, "docker-archive://"+e.tmpdir+"/image.tar")
 170                 build.Env = os.Environ()
 171                 build.Env = append(build.Env, "SINGULARITY_CACHEDIR="+e.tmpdir+"/cache")
 172                 e.logf("%v", build.Args)
 173                 out, err := build.CombinedOutput()
 174                 // INFO:    Starting build...
 175                 // Getting image source signatures
 176                 // Copying blob ab15617702de done
 177                 // Copying config 651e02b8a2 done
 178                 // Writing manifest to image destination
 179                 // Storing signatures
 180                 // 2021/04/22 14:42:14  info unpack layer: sha256:21cbfd3a344c52b197b9fa36091e66d9cbe52232703ff78d44734f85abb7ccd3
 181                 // INFO:    Creating SIF file...
 182                 // INFO:    Build complete: arvados-jobs.latest.sif
 183                 e.logf("%s", out)
 184                 if err != nil {
 185                         return err
 186                 }
 187         }
 188
 189         if containerClient == nil {
 190                 e.imageFilename = imageFilename
 191                 return nil
 192         }
 193
 194         // update TTL to now + two weeks
 195         exp := time.Now().Add(24 * 7 * 2 * time.Hour)
 196
 197         uuidPath, err := containerClient.PathForUUID("update", sifCollection.UUID)
 198         if err != nil {
 199                 e.logf("error PathForUUID: %v", err)
 200                 return nil
 201         }
 202         var imageCollection arvados.Collection
 203         err = containerClient.RequestAndDecode(&imageCollection,
 204                 arvados.EndpointCollectionUpdate.Method,
 205                 uuidPath,
 206                 nil, map[string]interface{}{
 207                         "collection": map[string]string{
 208                                 "name":     fmt.Sprintf("singularity image for %v", dockerImageID),
 209                                 "trash_at": exp.UTC().Format(time.RFC3339),
 210                         },
 211                 })
 212         if err == nil {
 213                 // If we just wrote the image to the cache, the
 214                 // response also returns the updated PDH
 215                 e.imageFilename = fmt.Sprintf("%s/by_id/%s/image.sif", arvMountPoint, imageCollection.PortableDataHash)
 216                 return nil
 217         }
 218
 219         e.logf("error updating/renaming collection for cached sif image: %v", err)
 220         // Failed to update but maybe it lost a race and there is
 221         // another cached collection in the same place, so check the cache
 222         // again
 223         sifCollection, err = e.checkImageCache(dockerImageID, container, arvMountPoint, containerClient)
 224         if err != nil {
 225                 return err
 226         }
 227         e.imageFilename = fmt.Sprintf("%s/by_id/%s/image.sif", arvMountPoint, sifCollection.PortableDataHash)
 228
 229         return nil
 230 }
 231
 232 func (e *singularityExecutor) Create(spec containerSpec) error {
 233         e.spec = spec
 234         return nil
 235 }
 236
 237 func (e *singularityExecutor) Start() error {
 238         args := []string{"singularity", "exec", "--containall", "--cleanenv", "--pwd", e.spec.WorkingDir}
 239         if !e.spec.EnableNetwork {
 240                 args = append(args, "--net", "--network=none")
 241         }
 242         readonlyflag := map[bool]string{
 243                 false: "rw",
 244                 true:  "ro",
 245         }
 246         var binds []string
 247         for path, _ := range e.spec.BindMounts {
 248                 binds = append(binds, path)
 249         }
 250         sort.Strings(binds)
 251         for _, path := range binds {
 252                 mount := e.spec.BindMounts[path]
 253                 if path == e.spec.Env["HOME"] {
 254                         // Singularity treates $HOME as special case
 255                         args = append(args, "--home", mount.HostPath+":"+path)
 256                 } else {
 257                         args = append(args, "--bind", mount.HostPath+":"+path+":"+readonlyflag[mount.ReadOnly])
 258                 }
 259         }
 260
 261         // This is for singularity 3.5.2. There are some behaviors
 262         // that will change in singularity 3.6, please see:
 263         // https://sylabs.io/guides/3.7/user-guide/environment_and_metadata.html
 264         // https://sylabs.io/guides/3.5/user-guide/environment_and_metadata.html
 265         env := make([]string, 0, len(e.spec.Env))
 266         for k, v := range e.spec.Env {
 267                 if k == "HOME" {
 268                         // Singularity treates $HOME as special case, this is handled
 269                         // with --home above
 270                         continue
 271                 }
 272                 env = append(env, "SINGULARITYENV_"+k+"="+v)
 273         }
 274
 275         args = append(args, e.imageFilename)
 276         args = append(args, e.spec.Command...)
 277
 278         path, err := exec.LookPath(args[0])
 279         if err != nil {
 280                 return err
 281         }
 282         child := &exec.Cmd{
 283                 Path:   path,
 284                 Args:   args,
 285                 Env:    env,
 286                 Stdin:  e.spec.Stdin,
 287                 Stdout: e.spec.Stdout,
 288                 Stderr: e.spec.Stderr,
 289         }
 290         err = child.Start()
 291         if err != nil {
 292                 return err
 293         }
 294         e.child = child
 295         return nil
 296 }
 297
 298 func (e *singularityExecutor) CgroupID() string {
 299         return ""
 300 }
 301
 302 func (e *singularityExecutor) Stop() error {
 303         if err := e.child.Process.Signal(syscall.Signal(0)); err != nil {
 304                 // process already exited
 305                 return nil
 306         }
 307         return e.child.Process.Signal(syscall.SIGKILL)
 308 }
 309
 310 func (e *singularityExecutor) Wait(context.Context) (int, error) {
 311         err := e.child.Wait()
 312         if err, ok := err.(*exec.ExitError); ok {
 313                 return err.ProcessState.ExitCode(), nil
 314         }
 315         if err != nil {
 316                 return 0, err
 317         }
 318         return e.child.ProcessState.ExitCode(), nil
 319 }
 320
 321 func (e *singularityExecutor) Close() {
 322         err := os.RemoveAll(e.tmpdir)
 323         if err != nil {
 324                 e.logf("error removing temp dir: %s", err)
 325         }
 326 }