import subprocess
import sys
import time
+import resource
import arvados.commands._util as arv_cmd
from arvados_fuse import crunchstat
type=str, metavar='PATH', action='append', default=[],
help="Create a new collection, mount it in read/write mode at mountpoint/PATH, and delete it when unmounting.")
+
self.add_argument('--debug', action='store_true', help="""Debug mode""")
self.add_argument('--logfile', help="""Write debug logs and errors to the specified file (default stderr).""")
self.add_argument('--foreground', action='store_true', help="""Run in foreground (default is to daemonize unless --exec specified)""", default=False)
self.add_argument('--encoding', type=str, help="Character encoding to use for filesystem, default is utf-8 (see Python codec registry for list of available encodings)", default="utf-8")
- self.add_argument('--file-cache', type=int, help="File data cache size, in bytes (default 256MiB)", default=256*1024*1024)
- self.add_argument('--directory-cache', type=int, help="Directory data cache size, in bytes (default 128MiB)", default=128*1024*1024)
+ self.add_argument('--file-cache', type=int, help="File data cache size, in bytes (default 8 GiB for disk-based cache or 256 MiB with RAM-only cache)", default=0)
+ self.add_argument('--directory-cache', type=int, help="Directory data cache size, in bytes (default 128 MiB)", default=128*1024*1024)
+
+ cachetype = self.add_mutually_exclusive_group()
+ cachetype.add_argument('--ram-cache', action='store_false', dest='disk_cache', help="Use in-memory caching only", default=True)
+ cachetype.add_argument('--disk-cache', action='store_true', dest='disk_cache', help="Use disk based caching (default)", default=True)
+
+ self.add_argument('--disk-cache-dir', type=str, help="Disk cache location (default ~/.cache/arvados/keep)", default=None)
self.add_argument('--disable-event-listening', action='store_true', help="Don't subscribe to events on the API server", dest="disable_event_listening", default=False)
self.add_argument('--read-only', action='store_false', help="Mount will be read only (default)", dest="enable_write", default=False)
self.add_argument('--read-write', action='store_true', help="Mount will be read-write", dest="enable_write", default=False)
+ self.add_argument('--storage-classes', type=str, metavar='CLASSES', help="Specify comma separated list of storage classes to be used when saving data of new collections", default=None)
self.add_argument('--crunchstat-interval', type=float, help="Write stats to stderr every N seconds (default disabled)", default=0)
self.add_argument('--unmount-timeout',
type=float, default=2.0,
help="Time to wait for graceful shutdown after --exec program exits and filesystem is unmounted")
-
+ self.add_argument(
+ '--filters',
+ type=arv_cmd.JSONArgument(arv_cmd.validate_filters),
+ help="""Filters to apply to all project, shared, and tag directory
+contents. Pass filters as either a JSON string or a path to a JSON file.
+The JSON object should be a list of filters in Arvados API list filter syntax.
+""")
self.add_argument('--exec', type=str, nargs=argparse.REMAINDER,
dest="exec_args", metavar=('command', 'args', '...', '--'),
help="""Mount, run a command, then unmount and exit""")
try:
self._setup_logging()
+ except Exception as e:
+ self.logger.exception("exception during setup: %s", e)
+ exit(1)
+
+ try:
+ nofile_limit = resource.getrlimit(resource.RLIMIT_NOFILE)
+
+ minlimit = 10240
+ if self.args.file_cache:
+ # Adjust the file handle limit so it can meet
+ # the desired cache size. Multiply by 8 because the
+ # number of 64 MiB cache slots that keepclient
+ # allocates is RLIMIT_NOFILE / 8
+ minlimit = int((self.args.file_cache/(64*1024*1024)) * 8)
+
+ if nofile_limit[0] < minlimit:
+ resource.setrlimit(resource.RLIMIT_NOFILE, (min(minlimit, nofile_limit[1]), nofile_limit[1]))
+
+ if minlimit > nofile_limit[1]:
+ self.logger.warning("file handles required to meet --file-cache (%s) exceeds hard file handle limit (%s), cache size will be smaller than requested", minlimit, nofile_limit[1])
+
+ except Exception as e:
+ self.logger.warning("unable to adjust file handle limit: %s", e)
+
+ nofile_limit = resource.getrlimit(resource.RLIMIT_NOFILE)
+ self.logger.info("file cache capped at %s bytes or less based on available disk (RLIMIT_NOFILE is %s)", ((nofile_limit[0]//8)*64*1024*1024), nofile_limit)
+
+ try:
self._setup_api()
self._setup_mount()
except Exception as e:
- self.logger.exception("arv-mount: exception during setup: %s", e)
+ self.logger.exception("exception during setup: %s", e)
exit(1)
def __enter__(self):
def _setup_api(self):
try:
+ # default value of file_cache is 0, this tells KeepBlockCache to
+ # choose a default based on whether disk_cache is enabled or not.
+
+ block_cache = arvados.keep.KeepBlockCache(cache_max=self.args.file_cache,
+ disk_cache=self.args.disk_cache,
+ disk_cache_dir=self.args.disk_cache_dir)
+
+ # If there's too many prefetch threads and you
+ # max out the CPU, delivering data to the FUSE
+ # layer actually ends up being slower.
+ # Experimentally, capping 7 threads seems to
+ # be a sweet spot.
+ prefetch_threads = min(max((block_cache.cache_max // (64 * 1024 * 1024)) - 1, 1), 7)
+
self.api = arvados.safeapi.ThreadSafeApiCache(
apiconfig=arvados.config.settings(),
+ api_params={
+ 'num_retries': self.args.retries,
+ },
keep_params={
- 'block_cache': arvados.keep.KeepBlockCache(self.args.file_cache),
+ 'block_cache': block_cache,
+ 'num_prefetch_threads': prefetch_threads,
'num_retries': self.args.retries,
- })
+ },
+ version='v1',
+ )
except KeyError as e:
self.logger.error("Missing environment: %s", e)
exit(1)
usr = self.api.users().current().execute(num_retries=self.args.retries)
now = time.time()
dir_class = None
- dir_args = [llfuse.ROOT_INODE, self.operations.inodes, self.api, self.args.retries]
+ dir_args = [
+ llfuse.ROOT_INODE,
+ self.operations.inodes,
+ self.api,
+ self.args.retries,
+ self.args.enable_write,
+ self.args.filters,
+ ]
mount_readme = False
+ storage_classes = None
+ if self.args.storage_classes is not None:
+ storage_classes = self.args.storage_classes.replace(' ', '').split(',')
+ self.logger.info("Storage classes requested for new collections: {}".format(', '.join(storage_classes)))
+
if self.args.collection is not None:
# Set up the request handler with the collection at the root
# First check that the collection is readable
mount_readme = True
if dir_class is not None:
- ent = dir_class(*dir_args)
+ if dir_class in [TagsDirectory, CollectionDirectory]:
+ ent = dir_class(*dir_args)
+ else:
+ ent = dir_class(*dir_args, storage_classes=storage_classes)
self.operations.inodes.add_entry(ent)
self.listen_for_events = ent.want_event_subscribe()
return
e = self.operations.inodes.add_entry(Directory(
- llfuse.ROOT_INODE, self.operations.inodes, self.api.config))
+ llfuse.ROOT_INODE,
+ self.operations.inodes,
+ self.api.config,
+ self.args.enable_write,
+ self.args.filters,
+ ))
dir_args[0] = e.inode
for name in self.args.mount_by_id:
- self._add_mount(e, name, MagicDirectory(*dir_args, pdh_only=False))
+ self._add_mount(e, name, MagicDirectory(*dir_args, pdh_only=False, storage_classes=storage_classes))
for name in self.args.mount_by_pdh:
self._add_mount(e, name, MagicDirectory(*dir_args, pdh_only=True))
for name in self.args.mount_by_tag:
self._add_mount(e, name, TagsDirectory(*dir_args))
for name in self.args.mount_home:
- self._add_mount(e, name, ProjectDirectory(*dir_args, project_object=usr, poll=True))
+ self._add_mount(e, name, ProjectDirectory(*dir_args, project_object=usr, poll=True, storage_classes=storage_classes))
for name in self.args.mount_shared:
- self._add_mount(e, name, SharedDirectory(*dir_args, exclude=usr, poll=True))
+ self._add_mount(e, name, SharedDirectory(*dir_args, exclude=usr, poll=True, storage_classes=storage_classes))
for name in self.args.mount_tmp:
- self._add_mount(e, name, TmpCollectionDirectory(*dir_args))
+ self._add_mount(e, name, TmpCollectionDirectory(*dir_args, storage_classes=storage_classes))
if mount_readme:
text = self._readme_text(