X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/85bd5f3272cfc8ba8e54f4f47ea9865d7de62779..8de691c25eac0454f8f30cfa35eccff15642e330:/sdk/cwl/arvados_cwl/fsaccess.py diff --git a/sdk/cwl/arvados_cwl/fsaccess.py b/sdk/cwl/arvados_cwl/fsaccess.py index 93e2819084..5e3d685911 100644 --- a/sdk/cwl/arvados_cwl/fsaccess.py +++ b/sdk/cwl/arvados_cwl/fsaccess.py @@ -9,6 +9,7 @@ import urlparse import re import logging import threading +from collections import OrderedDict import ruamel.yaml as yaml @@ -26,19 +27,44 @@ 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): + 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.collections = {} + 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) - self.collections[pdh] = arvados.collection.CollectionReader(pdh, api_client=self.api_client, - keep_client=self.keep_client) - return self.collections[pdh] + cr = arvados.collection.CollectionReader(pdh, api_client=self.api_client, + keep_client=self.keep_client) + 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): @@ -81,14 +107,14 @@ class CollectionFsAccess(cwltool.stdfsaccess.StdFsAccess): 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()) 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) @@ -105,7 +131,7 @@ class CollectionFsAccess(cwltool.stdfsaccess.StdFsAccess): 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: @@ -115,7 +141,7 @@ class CollectionFsAccess(cwltool.stdfsaccess.StdFsAccess): 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: @@ -125,7 +151,7 @@ class CollectionFsAccess(cwltool.stdfsaccess.StdFsAccess): 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: @@ -147,7 +173,7 @@ class CollectionFsAccess(cwltool.stdfsaccess.StdFsAccess): 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) @@ -227,6 +253,9 @@ workflow_uuid_pattern = re.compile(r'[a-z0-9]{5}-7fd4e-[a-z0-9]{15}') 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)