sdk/cwl/arvados_cwl/fsaccess.py

   1 import fnmatch
   2 import os
   3 import errno
   4 import urlparse
   5 import re
   6 import logging
   7
   8 import ruamel.yaml as yaml
   9
  10 import cwltool.stdfsaccess
  11 from cwltool.pathmapper import abspath
  12 import cwltool.resolver
  13
  14 import arvados.util
  15 import arvados.collection
  16 import arvados.arvfile
  17 import arvados.errors
  18
  19 from schema_salad.ref_resolver import DefaultFetcher
  20
  21 logger = logging.getLogger('arvados.cwl-runner')
  22
  23 class CollectionFsAccess(cwltool.stdfsaccess.StdFsAccess):
  24     """Implement the cwltool FsAccess interface for Arvados Collections."""
  25
  26     def __init__(self, basedir, api_client=None, keep_client=None):
  27         super(CollectionFsAccess, self).__init__(basedir)
  28         self.api_client = api_client
  29         self.keep_client = keep_client
  30         self.collections = {}
  31
  32     def get_collection(self, path):
  33         p = path.split("/")
  34         if p[0].startswith("keep:") and arvados.util.keep_locator_pattern.match(p[0][5:]):
  35             pdh = p[0][5:]
  36             if pdh not in self.collections:
  37                 self.collections[pdh] = arvados.collection.CollectionReader(pdh, api_client=self.api_client,
  38                                                                             keep_client=self.keep_client)
  39             return (self.collections[pdh], "/".join(p[1:]))
  40         else:
  41             return (None, path)
  42
  43     def _match(self, collection, patternsegments, parent):
  44         if not patternsegments:
  45             return []
  46
  47         if not isinstance(collection, arvados.collection.RichCollectionBase):
  48             return []
  49
  50         ret = []
  51         # iterate over the files and subcollections in 'collection'
  52         for filename in collection:
  53             if patternsegments[0] == '.':
  54                 # Pattern contains something like "./foo" so just shift
  55                 # past the "./"
  56                 ret.extend(self._match(collection, patternsegments[1:], parent))
  57             elif fnmatch.fnmatch(filename, patternsegments[0]):
  58                 cur = os.path.join(parent, filename)
  59                 if len(patternsegments) == 1:
  60                     ret.append(cur)
  61                 else:
  62                     ret.extend(self._match(collection[filename], patternsegments[1:], cur))
  63         return ret
  64
  65     def glob(self, pattern):
  66         collection, rest = self.get_collection(pattern)
  67         if collection and not rest:
  68             return [pattern]
  69         patternsegments = rest.split("/")
  70         return self._match(collection, patternsegments, "keep:" + collection.manifest_locator())
  71
  72     def open(self, fn, mode):
  73         collection, rest = self.get_collection(fn)
  74         if collection:
  75             return collection.open(rest, mode)
  76         else:
  77             return super(CollectionFsAccess, self).open(self._abs(fn), mode)
  78
  79     def exists(self, fn):
  80         collection, rest = self.get_collection(fn)
  81         if collection:
  82             return collection.exists(rest)
  83         else:
  84             return super(CollectionFsAccess, self).exists(fn)
  85
  86     def isfile(self, fn):  # type: (unicode) -> bool
  87         collection, rest = self.get_collection(fn)
  88         if collection:
  89             if rest:
  90                 return isinstance(collection.find(rest), arvados.arvfile.ArvadosFile)
  91             else:
  92                 return False
  93         else:
  94             return super(CollectionFsAccess, self).isfile(fn)
  95
  96     def isdir(self, fn):  # type: (unicode) -> bool
  97         collection, rest = self.get_collection(fn)
  98         if collection:
  99             if rest:
 100                 return isinstance(collection.find(rest), arvados.collection.RichCollectionBase)
 101             else:
 102                 return True
 103         else:
 104             return super(CollectionFsAccess, self).isdir(fn)
 105
 106     def listdir(self, fn):  # type: (unicode) -> List[unicode]
 107         collection, rest = self.get_collection(fn)
 108         if collection:
 109             if rest:
 110                 dir = collection.find(rest)
 111             else:
 112                 dir = collection
 113             if dir is None:
 114                 raise IOError(errno.ENOENT, "Directory '%s' in '%s' not found" % (rest, collection.portable_data_hash()))
 115             if not isinstance(dir, arvados.collection.RichCollectionBase):
 116                 raise IOError(errno.ENOENT, "Path '%s' in '%s' is not a Directory" % (rest, collection.portable_data_hash()))
 117             return [abspath(l, fn) for l in dir.keys()]
 118         else:
 119             return super(CollectionFsAccess, self).listdir(fn)
 120
 121     def join(self, path, *paths): # type: (unicode, *unicode) -> unicode
 122         if paths and paths[-1].startswith("keep:") and arvados.util.keep_locator_pattern.match(paths[-1][5:]):
 123             return paths[-1]
 124         return os.path.join(path, *paths)
 125
 126     def realpath(self, path):
 127         if path.startswith("$(task.tmpdir)") or path.startswith("$(task.outdir)"):
 128             return path
 129         collection, rest = self.get_collection(path)
 130         if collection:
 131             return path
 132         else:
 133             return os.path.realpath(path)
 134
 135 class CollectionFetcher(DefaultFetcher):
 136     def __init__(self, cache, session, api_client=None, keep_client=None, num_retries=4):
 137         super(CollectionFetcher, self).__init__(cache, session)
 138         self.api_client = api_client
 139         self.fsaccess = CollectionFsAccess("", api_client=api_client, keep_client=keep_client)
 140         self.num_retries = num_retries
 141
 142     def fetch_text(self, url):
 143         if url.startswith("keep:"):
 144             with self.fsaccess.open(url, "r") as f:
 145                 return f.read()
 146         if url.startswith("arvwf:"):
 147             record = self.api_client.workflows().get(uuid=url[6:]).execute(num_retries=self.num_retries)
 148             definition = record["definition"] + ('\nlabel: "%s"\n' % record["name"].replace('"', '\\"'))
 149             return definition
 150         return super(CollectionFetcher, self).fetch_text(url)
 151
 152     def check_exists(self, url):
 153         try:
 154             if url.startswith("keep:"):
 155                 return self.fsaccess.exists(url)
 156             if url.startswith("arvwf:"):
 157                 if self.fetch_text(url):
 158                     return True
 159         except arvados.errors.NotFoundError:
 160             return False
 161         except:
 162             logger.exception("Got unexpected exception checking if file exists:")
 163             return False
 164         return super(CollectionFetcher, self).check_exists(url)
 165
 166     def urljoin(self, base_url, url):
 167         if not url:
 168             return base_url
 169
 170         urlsp = urlparse.urlsplit(url)
 171         if urlsp.scheme or not base_url:
 172             return url
 173
 174         basesp = urlparse.urlsplit(base_url)
 175         if basesp.scheme in ("keep", "arvwf"):
 176             if not basesp.path:
 177                 raise IOError(errno.EINVAL, "Invalid Keep locator", base_url)
 178
 179             baseparts = basesp.path.split("/")
 180             urlparts = urlsp.path.split("/") if urlsp.path else []
 181
 182             pdh = baseparts.pop(0)
 183
 184             if basesp.scheme == "keep" and not arvados.util.keep_locator_pattern.match(pdh):
 185                 raise IOError(errno.EINVAL, "Invalid Keep locator", base_url)
 186
 187             if urlsp.path.startswith("/"):
 188                 baseparts = []
 189                 urlparts.pop(0)
 190
 191             if baseparts and urlsp.path:
 192                 baseparts.pop()
 193
 194             path = "/".join([pdh] + baseparts + urlparts)
 195             return urlparse.urlunsplit((basesp.scheme, "", path, "", urlsp.fragment))
 196
 197         return super(CollectionFetcher, self).urljoin(base_url, url)
 198
 199 workflow_uuid_pattern = re.compile(r'[a-z0-9]{5}-7fd4e-[a-z0-9]{15}')
 200 pipeline_template_uuid_pattern = re.compile(r'[a-z0-9]{5}-p5p6p-[a-z0-9]{15}')
 201
 202 def collectionResolver(api_client, document_loader, uri, num_retries=4):
 203     if workflow_uuid_pattern.match(uri):
 204         return "arvwf:%s#main" % (uri)
 205
 206     if pipeline_template_uuid_pattern.match(uri):
 207         pt = api_client.pipeline_templates().get(uuid=uri).execute(num_retries=num_retries)
 208         return "keep:" + pt["components"].values()[0]["script_parameters"]["cwl:tool"]
 209
 210     p = uri.split("/")
 211     if arvados.util.keep_locator_pattern.match(p[0]):
 212         return "keep:%s" % (uri)
 213
 214     if arvados.util.collection_uuid_pattern.match(p[0]):
 215         return "keep:%s%s" % (api_client.collections().
 216                               get(uuid=p[0]).execute()["portable_data_hash"],
 217                               uri[len(p[0]):])
 218
 219     return cwltool.resolver.tool_resolver(document_loader, uri)