+# Copyright (C) The Arvados Authors. All rights reserved.
+#
+# SPDX-License-Identifier: Apache-2.0
+
import fnmatch
import os
import errno
import urlparse
import re
import logging
+import threading
+from collections import OrderedDict
import ruamel.yaml as yaml
import arvados.arvfile
import arvados.errors
+from googleapiclient.errors import HttpError
+
from schema_salad.ref_resolver import DefaultFetcher
logger = logging.getLogger('arvados.cwl-runner')
+class CollectionCache(object):
+ def __init__(self, api_client, keep_client, num_retries,
+ cap=256*1024*1024,
+ min_entries=2):
+ self.api_client = api_client
+ self.keep_client = keep_client
+ self.num_retries = num_retries
+ self.collections = OrderedDict()
+ self.lock = threading.Lock()
+ self.total = 0
+ self.cap = cap
+ self.min_entries = min_entries
+
+ def cap_cache(self):
+ if self.total > self.cap:
+ # ordered list iterates from oldest to newest
+ for pdh, v in self.collections.items():
+ if self.total < self.cap or len(self.collections) < self.min_entries:
+ break
+ # cut it loose
+ logger.debug("Evicting collection reader %s from cache", pdh)
+ del self.collections[pdh]
+ self.total -= v[1]
+
+ def get(self, pdh):
+ with self.lock:
+ if pdh not in self.collections:
+ logger.debug("Creating collection reader for %s", pdh)
+ cr = arvados.collection.CollectionReader(pdh, api_client=self.api_client,
+ keep_client=self.keep_client,
+ num_retries=self.num_retries)
+ sz = len(cr.manifest_text()) * 128
+ self.collections[pdh] = (cr, sz)
+ self.total += sz
+ self.cap_cache()
+ else:
+ cr, sz = self.collections[pdh]
+ # bump it to the back
+ del self.collections[pdh]
+ self.collections[pdh] = (cr, sz)
+ return cr
+
+
class CollectionFsAccess(cwltool.stdfsaccess.StdFsAccess):
"""Implement the cwltool FsAccess interface for Arvados Collections."""
- def __init__(self, basedir, api_client=None, keep_client=None):
+ def __init__(self, basedir, collection_cache=None):
super(CollectionFsAccess, self).__init__(basedir)
- self.api_client = api_client
- self.keep_client = keep_client
- self.collections = {}
+ self.collection_cache = collection_cache
def get_collection(self, path):
sp = path.split("/", 1)
p = sp[0]
if p.startswith("keep:") and arvados.util.keep_locator_pattern.match(p[5:]):
pdh = p[5:]
- if pdh not in self.collections:
- self.collections[pdh] = arvados.collection.CollectionReader(pdh, api_client=self.api_client,
- keep_client=self.keep_client)
- return (self.collections[pdh], sp[1] if len(sp) == 2 else None)
+ return (self.collection_cache.get(pdh), urlparse.unquote(sp[1]) if len(sp) == 2 else None)
else:
return (None, path)
def glob(self, pattern):
collection, rest = self.get_collection(pattern)
- if collection and not rest:
+ if collection is not None and not rest:
return [pattern]
patternsegments = rest.split("/")
- return self._match(collection, patternsegments, "keep:" + collection.manifest_locator())
+ return sorted(self._match(collection, patternsegments, "keep:" + collection.manifest_locator()))
def open(self, fn, mode):
collection, rest = self.get_collection(fn)
- if collection:
+ if collection is not None:
return collection.open(rest, mode)
else:
return super(CollectionFsAccess, self).open(self._abs(fn), mode)
def exists(self, fn):
- collection, rest = self.get_collection(fn)
- if collection:
- return collection.exists(rest)
+ try:
+ collection, rest = self.get_collection(fn)
+ except HttpError as err:
+ if err.resp.status == 404:
+ return False
+ else:
+ raise
+ if collection is not None:
+ if rest:
+ return collection.exists(rest)
+ else:
+ return True
else:
return super(CollectionFsAccess, self).exists(fn)
+ def size(self, fn): # type: (unicode) -> bool
+ collection, rest = self.get_collection(fn)
+ if collection is not None:
+ if rest:
+ arvfile = collection.find(rest)
+ if isinstance(arvfile, arvados.arvfile.ArvadosFile):
+ return arvfile.size()
+ raise IOError(errno.EINVAL, "Not a path to a file %s" % (fn))
+ else:
+ return super(CollectionFsAccess, self).size(fn)
+
def isfile(self, fn): # type: (unicode) -> bool
collection, rest = self.get_collection(fn)
- if collection:
+ if collection is not None:
if rest:
return isinstance(collection.find(rest), arvados.arvfile.ArvadosFile)
else:
def isdir(self, fn): # type: (unicode) -> bool
collection, rest = self.get_collection(fn)
- if collection:
+ if collection is not None:
if rest:
return isinstance(collection.find(rest), arvados.collection.RichCollectionBase)
else:
def listdir(self, fn): # type: (unicode) -> List[unicode]
collection, rest = self.get_collection(fn)
- if collection:
+ if collection is not None:
if rest:
dir = collection.find(rest)
else:
if path.startswith("$(task.tmpdir)") or path.startswith("$(task.outdir)"):
return path
collection, rest = self.get_collection(path)
- if collection:
+ if collection is not None:
return path
else:
return os.path.realpath(path)
class CollectionFetcher(DefaultFetcher):
- def __init__(self, cache, session, api_client=None, keep_client=None, num_retries=4):
+ def __init__(self, cache, session, api_client=None, fs_access=None, num_retries=4):
super(CollectionFetcher, self).__init__(cache, session)
self.api_client = api_client
- self.fsaccess = CollectionFsAccess("", api_client=api_client, keep_client=keep_client)
+ self.fsaccess = fs_access
self.num_retries = num_retries
def fetch_text(self, url):
pipeline_template_uuid_pattern = re.compile(r'[a-z0-9]{5}-p5p6p-[a-z0-9]{15}')
def collectionResolver(api_client, document_loader, uri, num_retries=4):
+ if uri.startswith("keep:") or uri.startswith("arvwf:"):
+ return uri
+
if workflow_uuid_pattern.match(uri):
return "arvwf:%s#main" % (uri)