sdk/cwl/arvados_cwl/pathmapper.py

   1 # Copyright (C) The Arvados Authors. All rights reserved.
   2 #
   3 # SPDX-License-Identifier: Apache-2.0
   4
   5 from future import standard_library
   6 standard_library.install_aliases()
   7 from builtins import str
   8 from past.builtins import basestring
   9 from future.utils import viewitems
  10
  11 import re
  12 import logging
  13 import uuid
  14 import os
  15 import urllib.request, urllib.parse, urllib.error
  16
  17 import arvados_cwl.util
  18 import arvados.commands.run
  19 import arvados.collection
  20
  21 from schema_salad.sourceline import SourceLine
  22
  23 from arvados.errors import ApiError
  24 from cwltool.pathmapper import PathMapper, MapperEnt
  25 from cwltool.utils import adjustFileObjs, adjustDirObjs
  26 from cwltool.stdfsaccess import abspath
  27 from cwltool.workflow import WorkflowException
  28
  29 from .http import http_to_keep
  30
  31 logger = logging.getLogger('arvados.cwl-runner')
  32
  33 def trim_listing(obj):
  34     """Remove 'listing' field from Directory objects that are keep references.
  35
  36     When Directory objects represent Keep references, it is redundant and
  37     potentially very expensive to pass fully enumerated Directory objects
  38     between instances of cwl-runner (e.g. a submitting a job, or using the
  39     RunInSingleContainer feature), so delete the 'listing' field when it is
  40     safe to do so.
  41
  42     """
  43
  44     if obj.get("location", "").startswith("keep:") and "listing" in obj:
  45         del obj["listing"]
  46
  47 collection_pdh_path = re.compile(r'^keep:[0-9a-f]{32}\+\d+/.+$')
  48 collection_pdh_pattern = re.compile(r'^keep:([0-9a-f]{32}\+\d+)(/.*)?')
  49 collection_uuid_pattern = re.compile(r'^keep:([a-z0-9]{5}-4zz18-[a-z0-9]{15})(/.*)?$')
  50
  51 class ArvPathMapper(PathMapper):
  52     """Convert container-local paths to and from Keep collection ids."""
  53
  54     def __init__(self, arvrunner, referenced_files, input_basedir,
  55                  collection_pattern, file_pattern, name=None, single_collection=False,
  56                  optional_deps=None):
  57         self.arvrunner = arvrunner
  58         self.input_basedir = input_basedir
  59         self.collection_pattern = collection_pattern
  60         self.file_pattern = file_pattern
  61         self.name = name
  62         self.referenced_files = [r["location"] for r in referenced_files]
  63         self.single_collection = single_collection
  64         self.pdh_to_uuid = {}
  65         self.optional_deps = optional_deps or []
  66         super(ArvPathMapper, self).__init__(referenced_files, input_basedir, None)
  67
  68     def visit(self, srcobj, uploadfiles):
  69         src = srcobj["location"]
  70         if "#" in src:
  71             src = src[:src.index("#")]
  72
  73         debug = logger.isEnabledFor(logging.DEBUG)
  74
  75         if isinstance(src, basestring) and src.startswith("keep:"):
  76             if collection_pdh_pattern.match(src):
  77                 self._pathmap[src] = MapperEnt(src, self.collection_pattern % urllib.parse.unquote(src[5:]), srcobj["class"], True)
  78
  79                 if arvados_cwl.util.collectionUUID in srcobj:
  80                     self.pdh_to_uuid[src.split("/", 1)[0][5:]] = srcobj[arvados_cwl.util.collectionUUID]
  81             elif not collection_uuid_pattern.match(src):
  82                 with SourceLine(srcobj, "location", WorkflowException, debug):
  83                     raise WorkflowException("Invalid keep reference '%s'" % src)
  84
  85         if src not in self._pathmap:
  86             if src.startswith("file:"):
  87                 # Local FS ref, may need to be uploaded or may be on keep
  88                 # mount.
  89                 ab = abspath(src, self.input_basedir)
  90                 st = arvados.commands.run.statfile("", ab,
  91                                                    fnPattern="keep:%s/%s",
  92                                                    dirPattern="keep:%s/%s",
  93                                                    raiseOSError=True)
  94                 with SourceLine(srcobj, "location", WorkflowException, debug):
  95                     if isinstance(st, arvados.commands.run.UploadFile):
  96                         uploadfiles.add((src, ab, st))
  97                     elif isinstance(st, arvados.commands.run.ArvFile):
  98                         self._pathmap[src] = MapperEnt(st.fn, self.collection_pattern % urllib.parse.unquote(st.fn[5:]), "File", True)
  99                     else:
 100                         raise WorkflowException("Input file path '%s' is invalid" % st)
 101             elif src.startswith("_:"):
 102                 if srcobj["class"] == "File" and "contents" not in srcobj:
 103                     raise WorkflowException("File literal '%s' is missing `contents`" % src)
 104                 if srcobj["class"] == "Directory" and "listing" not in srcobj:
 105                     raise WorkflowException("Directory literal '%s' is missing `listing`" % src)
 106             elif src.startswith("http:") or src.startswith("https:"):
 107                 try:
 108                     if self.arvrunner.defer_downloads:
 109                         # passthrough, we'll download it later.
 110                         self._pathmap[src] = MapperEnt(src, src, srcobj["class"], True)
 111                     else:
 112                         keepref = http_to_keep(self.arvrunner.api, self.arvrunner.project_uuid, src)
 113                         logger.info("%s is %s", src, keepref)
 114                         self._pathmap[src] = MapperEnt(keepref, keepref, srcobj["class"], True)
 115                 except Exception as e:
 116                     logger.warning(str(e))
 117             else:
 118                 self._pathmap[src] = MapperEnt(src, src, srcobj["class"], True)
 119
 120         with SourceLine(srcobj, "secondaryFiles", WorkflowException, debug):
 121             for l in srcobj.get("secondaryFiles", []):
 122                 self.visit(l, uploadfiles)
 123         with SourceLine(srcobj, "listing", WorkflowException, debug):
 124             for l in srcobj.get("listing", []):
 125                 self.visit(l, uploadfiles)
 126
 127     def addentry(self, obj, c, path, remap):
 128         if obj["location"] in self._pathmap:
 129             src, srcpath = self.arvrunner.fs_access.get_collection(self._pathmap[obj["location"]].resolved)
 130             if srcpath == "":
 131                 srcpath = "."
 132             c.copy(srcpath, path + "/" + obj["basename"], source_collection=src, overwrite=True)
 133             remap.append((obj["location"], path + "/" + obj["basename"]))
 134             for l in obj.get("secondaryFiles", []):
 135                 self.addentry(l, c, path, remap)
 136         elif obj["class"] == "Directory":
 137             for l in obj.get("listing", []):
 138                 self.addentry(l, c, path + "/" + obj["basename"], remap)
 139             remap.append((obj["location"], path + "/" + obj["basename"]))
 140         elif obj["location"].startswith("_:") and "contents" in obj:
 141             with c.open(path + "/" + obj["basename"], "w") as f:
 142                 f.write(obj["contents"])
 143             remap.append((obj["location"], path + "/" + obj["basename"]))
 144         else:
 145             for opt in self.optional_deps:
 146                 if obj["location"] == opt["location"]:
 147                     return
 148             raise SourceLine(obj, "location", WorkflowException).makeError("Don't know what to do with '%s'" % obj["location"])
 149
 150     def needs_new_collection(self, srcobj, prefix=""):
 151         """Check if files need to be staged into a new collection.
 152
 153         If all the files are in the same collection and in the same
 154         paths they would be staged to, return False.  Otherwise, a new
 155         collection is needed with files copied/created in the
 156         appropriate places.
 157         """
 158
 159         loc = srcobj["location"]
 160         if loc.startswith("_:"):
 161             return True
 162
 163         i = loc.rfind("/")
 164         if i > -1:
 165             loc_prefix = loc[:i+1]
 166             if not prefix:
 167                 prefix = loc_prefix
 168             # quote/unquote to ensure consistent quoting
 169             suffix = urllib.parse.quote(urllib.parse.unquote(loc[i+1:]), "/+@")
 170         else:
 171             # no '/' found
 172             loc_prefix = loc+"/"
 173             prefix = loc+"/"
 174             suffix = ""
 175
 176         if prefix != loc_prefix:
 177             return True
 178
 179         if "basename" in srcobj and suffix != urllib.parse.quote(srcobj["basename"], "/+@"):
 180             return True
 181
 182         if srcobj["class"] == "File" and loc not in self._pathmap:
 183             return True
 184         for s in srcobj.get("secondaryFiles", []):
 185             if self.needs_new_collection(s, prefix):
 186                 return True
 187         if srcobj.get("listing"):
 188             prefix = "%s%s/" % (prefix, urllib.parse.quote(srcobj.get("basename", suffix), "/+@"))
 189             for l in srcobj["listing"]:
 190                 if self.needs_new_collection(l, prefix):
 191                     return True
 192         return False
 193
 194     def setup(self, referenced_files, basedir):
 195         # type: (List[Any], unicode) -> None
 196         uploadfiles = set()
 197
 198         collection = None
 199         if self.single_collection:
 200             collection = arvados.collection.Collection(api_client=self.arvrunner.api,
 201                                                        keep_client=self.arvrunner.keep_client,
 202                                                        num_retries=self.arvrunner.num_retries)
 203
 204         for srcobj in referenced_files:
 205             self.visit(srcobj, uploadfiles)
 206
 207         arvados.commands.run.uploadfiles([u[2] for u in uploadfiles],
 208                                          self.arvrunner.api,
 209                                          dry_run=False,
 210                                          num_retries=self.arvrunner.num_retries,
 211                                          fnPattern="keep:%s/%s",
 212                                          name=self.name,
 213                                          project=self.arvrunner.project_uuid,
 214                                          collection=collection,
 215                                          packed=False)
 216
 217         for src, ab, st in uploadfiles:
 218             self._pathmap[src] = MapperEnt(urllib.parse.quote(st.fn, "/:+@"), urllib.parse.quote(self.collection_pattern % st.fn[5:], "/:+@"),
 219                                            "Directory" if os.path.isdir(ab) else "File", True)
 220
 221         for srcobj in referenced_files:
 222             remap = []
 223             if srcobj["class"] == "Directory" and srcobj["location"] not in self._pathmap:
 224                 c = arvados.collection.Collection(api_client=self.arvrunner.api,
 225                                                   keep_client=self.arvrunner.keep_client,
 226                                                   num_retries=self.arvrunner.num_retries)
 227                 for l in srcobj.get("listing", []):
 228                     self.addentry(l, c, ".", remap)
 229
 230                 container = arvados_cwl.util.get_current_container(self.arvrunner.api, self.arvrunner.num_retries, logger)
 231                 info = arvados_cwl.util.get_intermediate_collection_info(None, container, self.arvrunner.intermediate_output_ttl)
 232
 233                 c.save_new(name=info["name"],
 234                            owner_uuid=self.arvrunner.project_uuid,
 235                            ensure_unique_name=True,
 236                            trash_at=info["trash_at"],
 237                            properties=info["properties"])
 238
 239                 ab = self.collection_pattern % c.portable_data_hash()
 240                 self._pathmap[srcobj["location"]] = MapperEnt("keep:"+c.portable_data_hash(), ab, "Directory", True)
 241             elif srcobj["class"] == "File" and self.needs_new_collection(srcobj):
 242                 c = arvados.collection.Collection(api_client=self.arvrunner.api,
 243                                                   keep_client=self.arvrunner.keep_client,
 244                                                   num_retries=self.arvrunner.num_retries)
 245                 self.addentry(srcobj, c, ".", remap)
 246
 247                 container = arvados_cwl.util.get_current_container(self.arvrunner.api, self.arvrunner.num_retries, logger)
 248                 info = arvados_cwl.util.get_intermediate_collection_info(None, container, self.arvrunner.intermediate_output_ttl)
 249
 250                 c.save_new(name=info["name"],
 251                            owner_uuid=self.arvrunner.project_uuid,
 252                            ensure_unique_name=True,
 253                            trash_at=info["trash_at"],
 254                            properties=info["properties"])
 255
 256                 ab = self.file_pattern % (c.portable_data_hash(), srcobj["basename"])
 257                 self._pathmap[srcobj["location"]] = MapperEnt("keep:%s/%s" % (c.portable_data_hash(), srcobj["basename"]),
 258                                                               ab, "File", True)
 259                 if srcobj.get("secondaryFiles"):
 260                     ab = self.collection_pattern % c.portable_data_hash()
 261                     self._pathmap["_:" + str(uuid.uuid4())] = MapperEnt("keep:"+c.portable_data_hash(), ab, "Directory", True)
 262
 263             if remap:
 264                 for loc, sub in remap:
 265                     # subdirs start with "./", strip it off
 266                     if sub.startswith("./"):
 267                         ab = self.file_pattern % (c.portable_data_hash(), sub[2:])
 268                     else:
 269                         ab = self.file_pattern % (c.portable_data_hash(), sub)
 270                     self._pathmap[loc] = MapperEnt("keep:%s/%s" % (c.portable_data_hash(), sub[2:]),
 271                                                    ab, "Directory", True)
 272
 273         self.keepdir = None
 274
 275     def reversemap(self, target):
 276         p = super(ArvPathMapper, self).reversemap(target)
 277         if p:
 278             return p
 279         elif target.startswith("keep:"):
 280             return (target, target)
 281         elif self.keepdir and target.startswith(self.keepdir):
 282             kp = "keep:" + target[len(self.keepdir)+1:]
 283             return (kp, kp)
 284         else:
 285             return None
 286
 287
 288 class StagingPathMapper(PathMapper):
 289     # Note that StagingPathMapper internally maps files from target to source.
 290     # Specifically, the 'self._pathmap' dict keys are the target location and the
 291     # values are 'MapperEnt' named tuples from which we use the 'resolved' attribute
 292     # as the file identifier. This makes it possible to map an input file to multiple
 293     # target directories. The exception is for file literals, which store the contents of
 294     # the file in 'MapperEnt.resolved' and are therefore still mapped from source to target.
 295
 296     _follow_dirs = True
 297
 298     def __init__(self, referenced_files, basedir, stagedir, separateDirs=True):
 299         self.targets = set()
 300         super(StagingPathMapper, self).__init__(referenced_files, basedir, stagedir, separateDirs)
 301
 302     def visit(self, obj, stagedir, basedir, copy=False, staged=False):
 303         # type: (Dict[unicode, Any], unicode, unicode, bool) -> None
 304         loc = obj["location"]
 305         stagedir = obj.get("dirname") or stagedir
 306         tgt = os.path.join(stagedir, obj["basename"])
 307         basetgt, baseext = os.path.splitext(tgt)
 308
 309         def targetExists():
 310             return tgt in self.targets and ("contents" not in obj) and (self._pathmap[tgt].resolved != loc)
 311         def literalTargetExists():
 312             return tgt in self.targets and "contents" in obj
 313
 314         n = 1
 315         if targetExists() or literalTargetExists():
 316             while tgt in self.targets:
 317                 n += 1
 318                 tgt = "%s_%i%s" % (basetgt, n, baseext)
 319         self.targets.add(tgt)
 320         if obj["class"] == "Directory":
 321             if obj.get("writable"):
 322                 self._pathmap[tgt] = MapperEnt(loc, tgt, "WritableDirectory", staged)
 323             else:
 324                 self._pathmap[tgt] = MapperEnt(loc, tgt, "Directory", staged)
 325             if loc.startswith("_:") or self._follow_dirs:
 326                 self.visitlisting(obj.get("listing", []), tgt, basedir)
 327         elif obj["class"] == "File":
 328             if tgt in self._pathmap:
 329                 return
 330             if "contents" in obj and loc.startswith("_:"):
 331                 self._pathmap[loc] = MapperEnt(obj["contents"], tgt, "CreateFile", staged)
 332             else:
 333                 if copy or obj.get("writable"):
 334                     self._pathmap[tgt] = MapperEnt(loc, tgt, "WritableFile", staged)
 335                 else:
 336                     self._pathmap[tgt] = MapperEnt(loc, tgt, "File", staged)
 337                 self.visitlisting(obj.get("secondaryFiles", []), stagedir, basedir)
 338
 339     def mapper(self, src):  # type: (Text) -> MapperEnt.
 340         # Overridden to maintain the use case of mapping by source (identifier) to
 341         # target regardless of how the map is structured interally.
 342         def getMapperEnt(src):
 343             for k,v in viewitems(self._pathmap):
 344                 if (v.type != "CreateFile" and v.resolved == src) or (v.type == "CreateFile" and k == src):
 345                     return v
 346
 347         if u"#" in src:
 348             i = src.index(u"#")
 349             v = getMapperEnt(src[i:])
 350             return MapperEnt(v.resolved, v.target + src[i:], v.type, v.staged)
 351         return getMapperEnt(src)
 352
 353
 354 class VwdPathMapper(StagingPathMapper):
 355     def setup(self, referenced_files, basedir):
 356         # type: (List[Any], unicode) -> None
 357
 358         # Go through each file and set the target to its own directory along
 359         # with any secondary files.
 360         self.visitlisting(referenced_files, self.stagedir, basedir)
 361
 362         for path, (ab, tgt, type, staged) in viewitems(self._pathmap):
 363             if type in ("File", "Directory") and ab.startswith("keep:"):
 364                 self._pathmap[path] = MapperEnt("$(task.keep)/%s" % ab[5:], tgt, type, staged)
 365
 366
 367 class NoFollowPathMapper(StagingPathMapper):
 368     _follow_dirs = False
 369     def setup(self, referenced_files, basedir):
 370         # type: (List[Any], unicode) -> None
 371         self.visitlisting(referenced_files, self.stagedir, basedir)