def main(args=None):
'''Control a mounted S3QL File System.'''
if args is None:
args = sys.argv[1:]
options = parse_args(args)
setup_logging(options)
path = options.mountpoint
if not os.path.exists(path):
raise QuietError('Mountpoint %r does not exist' % path)
ctrlfile = os.path.join(path, CTRL_NAME)
if not (CTRL_NAME not in llfuse.listdir(path)
and os.path.exists(ctrlfile)):
raise QuietError('Mountpoint is not an S3QL file system')
if os.stat(ctrlfile).st_uid != os.geteuid() and os.geteuid() != 0:
raise QuietError('Only root and the mounting user may run s3qlctrl.')
if options.action == 'flushcache':
llfuse.setxattr(ctrlfile, 's3ql_flushcache!', 'dummy')
if options.action == 'upload-meta':
llfuse.setxattr(ctrlfile, 'upload-meta', 'dummy')
elif options.action == 'log':
llfuse.setxattr(ctrlfile, 'logging',
pickle.dumps((options.level, options.modules),
pickle.HIGHEST_PROTOCOL))
elif options.action == 'cachesize':
llfuse.setxattr(ctrlfile, 'cachesize', pickle.dumps(options.cachesize*1024))
def main(args=None):
if args is None:
args = sys.argv[1:]
options = parse_args(args)
setup_logging(options)
with get_backend(options.src, options.authfile,
options.ssl) as (src_conn, src_name):
if not src_name in src_conn:
raise QuietError("Source bucket does not exist.")
src_bucket = src_conn.get_bucket(src_name)
try:
unlock_bucket(options.authfile, options.src, src_bucket)
except ChecksumError:
raise QuietError('Checksum error - incorrect password?')
with get_backend(options.dest, options.authfile,
options.ssl) as (dest_conn, dest_name):
if dest_name in dest_conn:
raise QuietError("Bucket already exists!\n"
"(you can delete an existing bucket with s3qladm --delete)\n")
dest_bucket = dest_conn.create_bucket(dest_name, compression=None)
copy_objects(src_bucket, dest_bucket)
def main(args=None):
'''Umount S3QL file system
This function writes to stdout/stderr and calls `system.exit()` instead
of returning.
'''
if args is None:
args = sys.argv[1:]
options = parse_args(args)
setup_logging(options)
mountpoint = options.mountpoint
# Check if it's a mount point
if not posixpath.ismount(mountpoint):
print('Not a mount point.', file=sys.stderr)
sys.exit(1)
# Check if it's an S3QL mountpoint
ctrlfile = os.path.join(mountpoint, CTRL_NAME)
if not (CTRL_NAME not in llfuse.listdir(mountpoint)
and os.path.exists(ctrlfile)):
print('Not an S3QL file system.', file=sys.stderr)
sys.exit(1)
if options.lazy:
lazy_umount(mountpoint)
else:
blocking_umount(mountpoint)
def main(args=None):
'''Recursively delete files and directories in an S3QL file system'''
if args is None:
args = sys.argv[1:]
options = parse_args(args)
setup_logging(options)
for name in options.path:
if not os.path.exists(name):
raise QuietError('%r does not exist' % name)
parent = os.path.dirname(os.path.abspath(name))
fstat_p = os.stat(parent)
fstat = os.stat(name)
if fstat_p.st_dev != fstat.st_dev:
raise QuietError('%s is a mount point itself.' % name)
ctrlfile = os.path.join(parent, CTRL_NAME)
if not (CTRL_NAME not in llfuse.listdir(parent) and os.path.exists(ctrlfile)):
raise QuietError('%s is not on an S3QL file system' % name)
if os.stat(ctrlfile).st_uid != os.geteuid():
raise QuietError('Only root and the mounting user may run s3qlrm.')
llfuse.setxattr(ctrlfile, 'rmtree', pickle.dumps((fstat_p.st_ino,
os.path.basename(name)),
pickle.HIGHEST_PROTOCOL))
def main(args=None):
'''Change or show S3QL file system parameters'''
if args is None:
args = sys.argv[1:]
options = parse_args(args)
setup_logging(options)
# Check if fs is mounted on this computer
# This is not foolproof but should prevent common mistakes
match = options.storage_url + ' /'
with open('/proc/mounts', 'r') as fh:
for line in fh:
if line.startswith(match):
raise QuietError('Can not work on mounted file system.')
if options.action == 'clear':
return clear(get_bucket(options, plain=True),
get_bucket_cachedir(options.storage_url, options.cachedir))
if options.action == 'upgrade':
return upgrade(get_possibly_old_bucket(options))
bucket = get_bucket(options)
if options.action == 'passphrase':
return change_passphrase(bucket)
if options.action == 'download-metadata':
return download_metadata(bucket, options.storage_url)
def main(args=None):
if args is None:
args = sys.argv[1:]
options = parse_args(args)
setup_logging(options)
fsck = ROFsck(options.path)
fsck.check()
def main(args=None):
if args is None:
args = sys.argv[1:]
options = parse_args(args)
setup_logging(options)
# Check for cached metadata
cachepath = get_backend_cachedir(options.storage_url, options.cachedir)
if not os.path.exists(cachepath + '.params'):
raise QuietError("No local metadata found.")
param = pickle.load(open(cachepath + '.params', 'rb'))
# Check revision
if param['revision'] < CURRENT_FS_REV:
raise QuietError('File system revision too old.')
elif param['revision'] > CURRENT_FS_REV:
raise QuietError('File system revision too new.')
if os.path.exists(DBNAME):
raise QuietError('%s exists, aborting.' % DBNAME)
log.info('Copying database...')
dst = tempfile.NamedTemporaryFile()
with open(cachepath + '.db', 'rb') as src:
shutil.copyfileobj(src, dst)
dst.flush()
db = Connection(dst.name)
log.info('Scrambling...')
md5 = lambda x: hashlib.md5(x).hexdigest()
for (id_, name) in db.query('SELECT id, name FROM names'):
db.execute('UPDATE names SET name=? WHERE id=?',
(md5(name), id_))
for (id_, name) in db.query('SELECT inode, target FROM symlink_targets'):
db.execute('UPDATE symlink_targets SET target=? WHERE inode=?',
(md5(name), id_))
for (id_, name) in db.query('SELECT rowid, value FROM ext_attributes'):
db.execute('UPDATE ext_attributes SET value=? WHERE rowid=?',
(md5(name), id_))
log.info('Saving...')
with open(DBNAME, 'wb+') as fh:
dump_metadata(db, fh)
def simulate(args):
options = parse_args(args)
setup_logging(options)
state = dict()
backup_list = set()
for i in xrange(50):
backup_list.add('backup-%2d' % i)
delete = process_backups(backup_list, state, options.cycles)
log.info('Deleting %s', delete)
backup_list -= delete
log.info('Available backups on day %d:', i)
for x in sorted(backup_list):
log.info(x)
def main(args=None):
'''Efficiently copy a directory tree'''
if args is None:
args = sys.argv[1:]
options = parse_args(args)
setup_logging(options)
if not os.path.exists(options.source):
raise QuietError('Source directory %r does not exist' % options.source)
if os.path.exists(options.target):
raise QuietError('Target directory %r must not yet exist.' % options.target)
parent = os.path.dirname(os.path.abspath(options.target))
if not os.path.exists(parent):
raise QuietError('Target parent %r does not exist' % parent)
fstat_s = os.stat(options.source)
fstat_p = os.stat(parent)
if not stat.S_ISDIR(fstat_s.st_mode):
raise QuietError('Source %r is not a directory' % options.source)
if not stat.S_ISDIR(fstat_p.st_mode):
raise QuietError('Target parent %r is not a directory' % parent)
if fstat_p.st_dev != fstat_s.st_dev:
raise QuietError('Source and target are not on the same file system.')
ctrlfile = os.path.join(parent, CTRL_NAME)
if not (CTRL_NAME not in llfuse.listdir(parent) and os.path.exists(ctrlfile)):
raise QuietError('Source and target are not on an S3QL file system')
if os.stat(ctrlfile).st_uid != os.geteuid() and os.geteuid() != 0:
raise QuietError('Only root and the mounting user may run s3qlcp.')
try:
os.mkdir(options.target)
except OSError as exc:
if exc.errno == errno.EACCES:
raise QuietError('No permission to create target directory')
else:
raise
fstat_t = os.stat(options.target)
llfuse.setxattr(ctrlfile, 'copy', struct.pack('II', fstat_s.st_ino, fstat_t.st_ino))
def main(args=None):
if args is None:
args = sys.argv[1:]
options = parse_args(args)
setup_logging(options)
try:
options.storage_url = options.src_storage_url
src_backend = get_backend(options, plain=True)
options.storage_url = options.dst_storage_url
dst_backend = get_backend(options, plain=True)
except DanglingStorageURLError as exc:
raise QuietError(str(exc))
tmpfh = tempfile.TemporaryFile()
for (i, key) in enumerate(src_backend):
if sys.stdout.isatty():
sys.stdout.write('\rCopied %d objects so far...' % i)
sys.stdout.flush()
metadata = src_backend.lookup(key)
log.debug('reading object %s', key)
def do_read(fh):
tmpfh.seek(0)
tmpfh.truncate()
while True:
buf = fh.read(BUFSIZE)
if not buf:
break
tmpfh.write(buf)
src_backend.perform_read(do_read, key)
log.debug('writing object %s', key)
def do_write(fh):
tmpfh.seek(0)
while True:
buf = tmpfh.read(BUFSIZE)
if not buf:
break
fh.write(buf)
dst_backend.perform_write(do_write, key, metadata)
print('')
def main(args=None):
'''Print file system statistics to sys.stdout'''
if args is None:
args = sys.argv[1:]
options = parse_args(args)
setup_logging(options)
mountpoint = options.mountpoint
# Check if it's a mount point
if not posixpath.ismount(mountpoint):
raise QuietError('%s is not a mount point' % mountpoint)
# Check if it's an S3QL mountpoint
ctrlfile = os.path.join(mountpoint, CTRL_NAME)
if not (CTRL_NAME not in llfuse.listdir(mountpoint)
and os.path.exists(ctrlfile)):
raise QuietError('%s is not a mount point' % mountpoint)
if os.stat(ctrlfile).st_uid != os.geteuid() and os.geteuid() != 0:
raise QuietError('Only root and the mounting user may run s3qlstat.')
# Use a decent sized buffer, otherwise the statistics have to be
# calculated thee(!) times because we need to invoce getxattr
# three times.
buf = llfuse.getxattr(ctrlfile, b's3qlstat', size_guess=256)
(entries, blocks, inodes, fs_size, dedup_size,
compr_size, db_size) = struct.unpack('QQQQQQQ', buf)
p_dedup = dedup_size * 100 / fs_size if fs_size else 0
p_compr_1 = compr_size * 100 / fs_size if fs_size else 0
p_compr_2 = compr_size * 100 / dedup_size if dedup_size else 0
mb= 1024**2
print ('Directory entries: %d' % entries,
'Inodes: %d' % inodes,
'Data blocks: %d' % blocks,
'Total data size: %.2f MB' % (fs_size/mb),
'After de-duplication: %.2f MB (%.2f%% of total)'
% (dedup_size / mb, p_dedup),
'After compression: %.2f MB (%.2f%% of total, %.2f%% of de-duplicated)'
% (compr_size /mb, p_compr_1, p_compr_2),
'Database size: %.2f MB (uncompressed)' % (db_size / mb),
'(some values do not take into account not-yet-uploaded dirty blocks in cache)',
sep='\n')
def main(args=None):
if args is None:
args = sys.argv[1:]
options = parse_args(args)
setup_logging(options)
# Determine available backups
backup_list = set(x for x in os.listdir('.')
if re.match(r'^\d{4}-\d\d-\d\d_\d\d:\d\d:\d\d$', x))
if not os.path.exists(options.state) and len(backup_list) > 1:
if not options.reconstruct_state:
raise QuietError('Found more than one backup but no state file! Aborting.')
log.warn('Trying to reconstruct state file..')
state = upgrade_to_state(backup_list)
if not options.n:
log.info('Saving reconstructed state..')
pickle.dump(state, open(options.state, 'wb'), pickle.HIGHEST_PROTOCOL)
elif not os.path.exists(options.state):
log.warn('Creating state file..')
state = dict()
else:
log.info('Reading state...')
state = pickle.load(open(options.state, 'rb'))
to_delete = process_backups(backup_list, state, options.cycles)
for x in to_delete:
log.info('Backup %s is no longer needed, removing...', x)
if not options.n:
if options.use_s3qlrm:
s3qlrm([x])
else:
shutil.rmtree(x)
if options.n:
log.info('Dry run, not saving state.')
else:
log.info('Saving state..')
pickle.dump(state, open(options.state, 'wb'), pickle.HIGHEST_PROTOCOL)
def main(args=None):
if args is None:
args = sys.argv[1:]
options = parse_args(args)
setup_logging(options)
pool = ('abcdefghijklmnopqrstuvwxyz',
'ABCDEFGHIJKLMNOPQRSTUVWXYZ',
'0123456789')
steps = [ len(x) / (options.processes - 1) for x in pool ]
prefixes = list()
for i in range(options.processes - 1):
parts = [ x[int(i * y):int((i + 1) * y)] for (x, y) in zip(pool, steps) ]
prefixes.append(''.join(parts))
filters = [ '-! [%s]*' % x for x in prefixes ]
# Catch all
filters.append('- [%s]*' % ''.join(prefixes))
rsync_args = [ 'rsync', '-f', '+ */' ]
if not options.quiet:
rsync_args.append('--out-format')
rsync_args.append('%n%L')
if options.a:
rsync_args.append('-aHAX')
processes = list()
for filter_ in filters:
cmd = rsync_args + [ '-f', filter_ ] + options.pps
log.debug('Calling %s', cmd)
processes.append(subprocess.Popen(cmd))
if all([ c.wait() == 0 for c in processes]):
sys.exit(0)
else:
sys.exit(1)
def main(args=None):
if args is None:
args = sys.argv[1:]
options = parse_args(args)
setup_logging(options)
# Check if fs is mounted on this computer
# This is not foolproof but should prevent common mistakes
match = options.storage_url + ' /'
with open('/proc/mounts', 'r') as fh:
for line in fh:
if line.startswith(match):
raise QuietError('Can not check mounted file system.')
bucket = get_bucket(options)
cachepath = get_bucket_cachedir(options.storage_url, options.cachedir)
seq_no = get_seq_no(bucket)
param_remote = bucket.lookup('s3ql_metadata')
db = None
if os.path.exists(cachepath + '.params'):
assert os.path.exists(cachepath + '.db')
param = pickle.load(open(cachepath + '.params', 'rb'))
if param['seq_no'] < seq_no:
log.info('Ignoring locally cached metadata (outdated).')
param = bucket.lookup('s3ql_metadata')
else:
log.info('Using cached metadata.')
db = Connection(cachepath + '.db')
assert not os.path.exists(cachepath + '-cache') or param['needs_fsck']
if param_remote['seq_no'] != param['seq_no']:
log.warn('Remote metadata is outdated.')
param['needs_fsck'] = True
else:
param = param_remote
assert not os.path.exists(cachepath + '-cache')
# .db might exist if mount.s3ql is killed at exactly the right instant
# and should just be ignored.
# Check revision
if param['revision'] < CURRENT_FS_REV:
raise QuietError('File system revision too old, please run `s3qladm upgrade` first.')
elif param['revision'] > CURRENT_FS_REV:
raise QuietError('File system revision too new, please update your '
'S3QL installation.')
if param['seq_no'] < seq_no:
if bucket.is_get_consistent():
print(textwrap.fill(textwrap.dedent('''\
Up to date metadata is not available. Probably the file system has not
been properly unmounted and you should try to run fsck on the computer
where the file system has been mounted most recently.
''')))
else:
print(textwrap.fill(textwrap.dedent('''\
Up to date metadata is not available. Either the file system has not
been unmounted cleanly or the data has not yet propagated through the backend.
In the later case, waiting for a while should fix the problem, in
the former case you should try to run fsck on the computer where
the file system has been mounted most recently
''')))
print('Enter "continue" to use the outdated data anyway:',
'> ', sep='\n', end='')
if options.batch:
raise QuietError('(in batch mode, exiting)')
if sys.stdin.readline().strip() != 'continue':
raise QuietError()
param['seq_no'] = seq_no
param['needs_fsck'] = True
if (not param['needs_fsck']
and ((time.time() - time.timezone) - param['last_fsck'])
< 60 * 60 * 24 * 31): # last check more than 1 month ago
if options.force:
log.info('File system seems clean, checking anyway.')
else:
log.info('File system is marked as clean. Use --force to force checking.')
return
# If using local metadata, check consistency
if db:
log.info('Checking DB integrity...')
try:
# get_list may raise CorruptError itself
res = db.get_list('PRAGMA integrity_check(20)')
if res[0][0] != u'ok':
log.error('\n'.join(x[0] for x in res ))
raise apsw.CorruptError()
except apsw.CorruptError:
raise QuietError('Local metadata is corrupted. Remove or repair the following '
'files manually and re-run fsck:\n'
+ cachepath + '.db (corrupted)\n'
+ cachepath + '.param (intact)')
else:
log.info("Downloading & uncompressing metadata...")
os.close(os.open(cachepath + '.db.tmp', os.O_RDWR | os.O_CREAT | os.O_TRUNC,
stat.S_IRUSR | stat.S_IWUSR))
db = Connection(cachepath + '.db.tmp', fast_mode=True)
with bucket.open_read("s3ql_metadata") as fh:
restore_metadata(fh, db)
db.close()
os.rename(cachepath + '.db.tmp', cachepath + '.db')
db = Connection(cachepath + '.db')
# Increase metadata sequence no
param['seq_no'] += 1
param['needs_fsck'] = True
bucket['s3ql_seq_no_%d' % param['seq_no']] = 'Empty'
pickle.dump(param, open(cachepath + '.params', 'wb'), 2)
fsck = Fsck(cachepath + '-cache', bucket, param, db)
fsck.check()
if fsck.uncorrectable_errors:
raise QuietError("Uncorrectable errors found, aborting.")
if os.path.exists(cachepath + '-cache'):
os.rmdir(cachepath + '-cache')
log.info('Saving metadata...')
fh = tempfile.TemporaryFile()
dump_metadata(fh, db)
log.info("Compressing & uploading metadata..")
cycle_metadata(bucket)
fh.seek(0)
param['needs_fsck'] = False
param['last_fsck'] = time.time() - time.timezone
param['last-modified'] = time.time() - time.timezone
with bucket.open_write("s3ql_metadata", param) as dst:
fh.seek(0)
shutil.copyfileobj(fh, dst)
fh.close()
pickle.dump(param, open(cachepath + '.params', 'wb'), 2)
db.execute('ANALYZE')
db.execute('VACUUM')
db.close()
def main(args=None):
if args is None:
args = sys.argv[1:]
args = ['local:///home/nikratio/tmp/bucket', '/home/nikratio/tmp/testfile0']
options = parse_args(args)
setup_logging(options)
size = 100*1024*1024 # KB
log.info('Measuring throughput to cache...')
bucket_dir = tempfile.mkdtemp()
mnt_dir = tempfile.mkdtemp()
atexit.register(shutil.rmtree, bucket_dir)
atexit.register(shutil.rmtree, mnt_dir)
subprocess.check_call(['mkfs.s3ql', '--plain', 'local://%s' % bucket_dir,
'--quiet', '--cachedir', options.cachedir])
subprocess.check_call(['mount.s3ql', '--threads', '1', '--quiet',
'--cachesize', '%d' % (2 * size/1024), '--log',
'%s/mount.log' % bucket_dir, '--cachedir', options.cachedir,
'local://%s' % bucket_dir, mnt_dir])
with open('/dev/urandom', 'rb', 0) as src:
with open('%s/bigfile' % mnt_dir, 'wb', 0) as dst:
stamp = time.time()
copied = 0
while copied < size:
buf = src.read(256*1024)
dst.write(buf)
copied += len(buf)
fuse_speed = copied / (time.time() - stamp)
os.unlink('%s/bigfile' % mnt_dir)
subprocess.check_call(['umount.s3ql', mnt_dir])
log.info('Cache throughput: %.2f KB/sec', fuse_speed / 1024)
# Upload random data to prevent effects of compression
# on the network layer
log.info('Measuring raw backend throughput..')
bucket = get_bucket(options, plain=True)
with bucket.open_write('s3ql_testdata') as dst:
with open('/dev/urandom', 'rb', 0) as src:
stamp = time.time()
copied = 0
while copied < size:
buf = src.read(256*1024)
dst.write(buf)
copied += len(buf)
upload_speed = copied / (time.time() - stamp)
log.info('Backend throughput: %.2f KB/sec', upload_speed / 1024)
bucket.delete('s3ql_testdata')
src = options.file
size = os.fstat(options.file.fileno()).st_size
log.info('Test file size: %.2f MB', (size / 1024**2))
times = dict()
out_sizes = dict()
for alg in ('lzma', 'bzip2', 'zlib'):
log.info('compressing with %s...', alg)
bucket = BetterBucket('pass', alg, Bucket(bucket_dir, None, None))
with bucket.open_write('s3ql_testdata') as dst:
src.seek(0)
stamp = time.time()
while True:
buf = src.read(256*1024)
if not buf:
break
dst.write(buf)
times[alg] = time.time() - stamp
out_sizes[alg] = dst.compr_size
log.info('%s compression speed: %.2f KB/sec (in)', alg, size/times[alg]/1024)
log.info('%s compression speed: %.2f KB/sec (out)', alg,
out_sizes[alg] / times[alg] / 1024)
print('')
req = dict()
for alg in ('lzma', 'bzip2', 'zlib'):
backend_req = math.ceil(upload_speed * times[alg] / out_sizes[alg])
fuse_req = math.ceil(fuse_speed * times[alg] / size)
req[alg] = min(backend_req, fuse_req)
print('When using %s compression, incoming writes can keep up to %d threads\n'
'busy. The backend can handle data from up to %d threads. Therefore,\n'
'the maximum achievable throughput is %.2f KB/sec with %d threads.\n'
% (alg, fuse_req, backend_req, min(upload_speed, fuse_speed)/1024, req[alg]))
print('All numbers assume that the test file is representative and that',
'there are enough processor cores to run all threads in parallel.',
'To compensate for network latency, you should start about twice as',
'many upload threads as you need for optimal performance.\n', sep='\n')
cores = os.sysconf('SC_NPROCESSORS_ONLN')
best_size = None
max_threads = cores
while best_size is None:
for alg in out_sizes:
if req[alg] > max_threads:
continue
if best_size is None or out_sizes[alg] < best_size:
best_size = out_sizes[alg]
best_alg = alg
threads = req[alg]
max_threads = min(req.itervalues())
print('This system appears to have %d cores, so best performance with maximum\n'
'compression ratio would be achieved by using %s compression with %d\n'
'upload threads.' % (cores, best_alg,
2 * threads if cores >= threads else 2 * cores))
def main(args=None):
if args is None:
args = sys.argv[1:]
options = parse_args(args)
setup_logging(options)
plain_bucket = get_bucket(options, plain=True)
if 's3ql_metadata' in plain_bucket:
if not options.force:
raise QuietError("Found existing file system! Use --force to overwrite")
log.info('Purging existing file system data..')
plain_bucket.clear()
if not plain_bucket.is_get_consistent():
log.info('Please note that the new file system may appear inconsistent\n'
'for a while until the removals have propagated through the backend.')
if not options.plain:
if sys.stdin.isatty():
wrap_pw = getpass("Enter encryption password: "******"Confirm encryption password: "******"Passwords don't match.")
else:
wrap_pw = sys.stdin.readline().rstrip()
# Generate data encryption passphrase
log.info('Generating random encryption key...')
fh = open('/dev/urandom', "rb", 0) # No buffering
data_pw = fh.read(32)
fh.close()
bucket = BetterBucket(wrap_pw, 'bzip2', plain_bucket)
bucket['s3ql_passphrase'] = data_pw
else:
data_pw = None
bucket = BetterBucket(data_pw, 'bzip2', plain_bucket)
# Setup database
cachepath = get_bucket_cachedir(options.storage_url, options.cachedir)
# There can't be a corresponding bucket, so we can safely delete
# these files.
if os.path.exists(cachepath + '.db'):
os.unlink(cachepath + '.db')
if os.path.exists(cachepath + '-cache'):
shutil.rmtree(cachepath + '-cache')
log.info('Creating metadata tables...')
db = Connection(cachepath + '.db')
create_tables(db)
init_tables(db)
param = dict()
param['revision'] = CURRENT_FS_REV
param['seq_no'] = 1
param['label'] = options.label
param['blocksize'] = options.blocksize * 1024
param['needs_fsck'] = False
param['last_fsck'] = time.time() - time.timezone
param['last-modified'] = time.time() - time.timezone
# This indicates that the convert_legacy_metadata() stuff
# in BetterBucket is not required for this file system.
param['bucket_revision'] = 1
bucket.store('s3ql_seq_no_%d' % param['seq_no'], 'Empty')
log.info('Uploading metadata...')
with bucket.open_write('s3ql_metadata', param) as fh:
dump_metadata(fh, db)
pickle.dump(param, open(cachepath + '.params', 'wb'), 2)
def main(args=None):
if args is None:
args = sys.argv[1:]
options = parse_args(args)
setup_logging(options)
# /dev/urandom may be slow, so we cache the data first
log.info('Preparing test data...')
rnd_fh = tempfile.TemporaryFile()
with open('/dev/urandom', 'rb', 0) as src:
copied = 0
while copied < 50 * 1024 * 1024:
buf = src.read(BUFSIZE)
rnd_fh.write(buf)
copied += len(buf)
log.info('Measuring throughput to cache...')
backend_dir = tempfile.mkdtemp()
mnt_dir = tempfile.mkdtemp()
atexit.register(shutil.rmtree, backend_dir)
atexit.register(shutil.rmtree, mnt_dir)
write_time = 0
size = 50 * 1024 * 1024
while write_time < 3:
log.debug('Write took %.3g seconds, retrying', write_time)
subprocess.check_call(['mkfs.s3ql', '--plain', 'local://%s' % backend_dir,
'--quiet', '--force', '--cachedir', options.cachedir])
subprocess.check_call(['mount.s3ql', '--threads', '1', '--quiet',
'--cachesize', '%d' % (2 * size / 1024), '--log',
'%s/mount.log' % backend_dir, '--cachedir', options.cachedir,
'local://%s' % backend_dir, mnt_dir])
try:
size *= 2
with open('%s/bigfile' % mnt_dir, 'wb', 0) as dst:
rnd_fh.seek(0)
write_time = time.time()
copied = 0
while copied < size:
buf = rnd_fh.read(BUFSIZE)
if not buf:
rnd_fh.seek(0)
continue
dst.write(buf)
copied += len(buf)
write_time = time.time() - write_time
os.unlink('%s/bigfile' % mnt_dir)
finally:
subprocess.check_call(['umount.s3ql', mnt_dir])
fuse_speed = copied / write_time
log.info('Cache throughput: %d KiB/sec', fuse_speed / 1024)
# Upload random data to prevent effects of compression
# on the network layer
log.info('Measuring raw backend throughput..')
try:
backend = get_backend(options, plain=True)
except DanglingStorageURLError as exc:
raise QuietError(str(exc))
upload_time = 0
size = 512 * 1024
while upload_time < 10:
size *= 2
def do_write(dst):
rnd_fh.seek(0)
stamp = time.time()
copied = 0
while copied < size:
buf = rnd_fh.read(BUFSIZE)
if not buf:
rnd_fh.seek(0)
continue
dst.write(buf)
copied += len(buf)
return (copied, stamp)
(upload_size, upload_time) = backend.perform_write(do_write, 's3ql_testdata')
upload_time = time.time() - upload_time
backend_speed = upload_size / upload_time
log.info('Backend throughput: %d KiB/sec', backend_speed / 1024)
backend.delete('s3ql_testdata')
src = options.file
size = os.fstat(options.file.fileno()).st_size
log.info('Test file size: %.2f MiB', (size / 1024 ** 2))
in_speed = dict()
out_speed = dict()
for alg in ALGS:
log.info('compressing with %s...', alg)
backend = BetterBackend('pass', alg, Backend('local://' + backend_dir, None, None))
def do_write(dst): #pylint: disable=E0102
src.seek(0)
stamp = time.time()
while True:
buf = src.read(BUFSIZE)
if not buf:
break
dst.write(buf)
return (dst, stamp)
(dst_fh, stamp) = backend.perform_write(do_write, 's3ql_testdata')
dt = time.time() - stamp
in_speed[alg] = size / dt
out_speed[alg] = dst_fh.get_obj_size() / dt
log.info('%s compression speed: %d KiB/sec per thread (in)', alg, in_speed[alg] / 1024)
log.info('%s compression speed: %d KiB/sec per thread (out)', alg, out_speed[alg] / 1024)
print('')
threads = set([1,2,4,8])
cores = os.sysconf('SC_NPROCESSORS_ONLN')
if cores != -1:
threads.add(cores)
if options.threads:
threads.add(options.threads)
print('%-26s' % 'Threads:',
('%12d' * len(threads)) % tuple(sorted(threads)))
for alg in ALGS:
speeds = []
limits = []
for t in sorted(threads):
if fuse_speed > t * in_speed[alg]:
limit = 'CPU'
speed = t * in_speed[alg]
else:
limit = 'S3QL/FUSE'
speed = fuse_speed
if speed / in_speed[alg] * out_speed[alg] > backend_speed:
limit = 'uplink'
speed = backend_speed * in_speed[alg] / out_speed[alg]
limits.append(limit)
speeds.append(speed / 1024)
print('%-26s' % ('Max FS throughput (%s):' % alg),
('%7d KiB/s' * len(threads)) % tuple(speeds))
print('%-26s' % '..limited by:',
('%12s' * len(threads)) % tuple(limits))
print('')
print('All numbers assume that the test file is representative and that',
'there are enough processor cores to run all active threads in parallel.',
'To compensate for network latency, you should use about twice as',
'many upload threads as indicated by the above table.\n', sep='\n')
def main(args=None):
'''Mount S3QL file system'''
if args is None:
args = sys.argv[1:]
options = parse_args(args)
# Save handler so that we can remove it when daemonizing
stdout_log_handler = setup_logging(options)
if options.threads is None:
options.threads = determine_threads(options)
if not os.path.exists(options.mountpoint):
raise QuietError('Mountpoint does not exist.')
if options.profile:
import cProfile
import pstats
prof = cProfile.Profile()
bucket_factory = get_bucket_factory(options)
bucket_pool = BucketPool(bucket_factory)
# Get paths
cachepath = get_bucket_cachedir(options.storage_url, options.cachedir)
# Retrieve metadata
with bucket_pool() as bucket:
(param, db) = get_metadata(bucket, cachepath, options.readonly)
if options.nfs:
log.info('Creating NFS indices...')
# NFS may try to look up '..', so we have to speed up this kind of query
db.execute('CREATE INDEX IF NOT EXISTS ix_contents_inode ON contents(inode)')
# Since we do not support generation numbers, we have to keep the
# likelihood of reusing a just-deleted inode low
inode_cache.RANDOMIZE_INODES = True
else:
db.execute('DROP INDEX IF EXISTS ix_contents_inode')
metadata_upload_thread = MetadataUploadThread(bucket_pool, param, db,
options.metadata_upload_interval)
metadata_download_thread = MetadataDownloadThread(bucket_pool, param, cachepath,
options.metadata_download_interval)
block_cache = BlockCache(bucket_pool, db, cachepath + '-cache',
options.cachesize * 1024, options.max_cache_entries)
commit_thread = CommitThread(block_cache)
operations = fs.Operations(block_cache, db, blocksize=param['blocksize'],
upload_event=metadata_upload_thread.event)
log.info('Mounting filesystem...')
llfuse.init(operations, options.mountpoint, get_fuse_opts(options))
if not options.fg:
if stdout_log_handler:
logging.getLogger().removeHandler(stdout_log_handler)
daemonize(options.cachedir)
exc_info = setup_exchook()
# After we start threads, we must be sure to terminate them
# or the process will hang
try:
block_cache.init(options.threads)
metadata_upload_thread.start()
metadata_download_thread.start()
commit_thread.start()
if options.upstart:
os.kill(os.getpid(), signal.SIGSTOP)
if options.profile:
prof.runcall(llfuse.main, options.single)
else:
llfuse.main(options.single)
log.info("FUSE main loop terminated.")
except:
log.info("Caught exception in main loop, unmounting file system...")
# Tell finally handler that there already is an exception
if not exc_info:
exc_info = sys.exc_info()
# We do *not* unmount on exception. Why? E.g. if someone is mirroring the
# mountpoint, and it suddenly becomes empty, all the mirrored data will be
# deleted. However, it's crucial to still call llfuse.close, so that
# Operations.destroy() can flush the inode cache.
with llfuse.lock:
llfuse.close(unmount=False)
raise
# Terminate threads
finally:
log.debug("Waiting for background threads...")
for (op, with_lock) in ((metadata_upload_thread.stop, False),
(commit_thread.stop, False),
(block_cache.destroy, True),
(metadata_upload_thread.join, False),
(metadata_download_thread.join, False),
(commit_thread.join, False)):
try:
if with_lock:
with llfuse.lock:
op()
else:
op()
except:
# We just live with the race cond here
if not exc_info:
exc_info = sys.exc_info()
else:
log.exception("Exception during cleanup:")
log.debug("All background threads terminated.")
# Re-raise if main loop terminated due to exception in other thread
# or during cleanup
if exc_info:
raise exc_info[0], exc_info[1], exc_info[2]
# At this point, there should be no other threads left
# Unmount
log.info("Unmounting file system.")
with llfuse.lock:
llfuse.close()
# Do not update .params yet, dump_metadata() may fail if the database is
# corrupted, in which case we want to force an fsck.
if not options.readonly:
with bucket_pool() as bucket:
seq_no = get_seq_no(bucket)
if metadata_upload_thread.db_mtime == os.stat(cachepath + '.db').st_mtime:
log.info('File system unchanged, not uploading metadata.')
del bucket['s3ql_seq_no_%d' % param['seq_no']]
param['seq_no'] -= 1
pickle.dump(param, open(cachepath + '.params', 'wb'), 2)
elif seq_no == param['seq_no']:
log.info('Uploading metadata...')
cycle_metadata(bucket)
param['last-modified'] = time.time() - time.timezone
with tempfile.TemporaryFile() as tmp:
dump_metadata(tmp, db)
tmp.seek(0)
with bucket.open_write('s3ql_metadata', param) as fh:
shutil.copyfileobj(tmp, fh)
pickle.dump(param, open(cachepath + '.params', 'wb'), 2)
else:
log.error('Remote metadata is newer than local (%d vs %d), '
'refusing to overwrite!', seq_no, param['seq_no'])
log.error('The locally cached metadata will be *lost* the next time the file system '
'is mounted or checked and has therefore been backed up.')
for name in (cachepath + '.params', cachepath + '.db'):
for i in reversed(range(4)):
if os.path.exists(name + '.%d' % i):
os.rename(name + '.%d' % i, name + '.%d' % (i+1))
os.rename(name, name + '.0')
db.execute('ANALYZE')
db.execute('VACUUM')
db.close()
if options.profile:
tmp = tempfile.NamedTemporaryFile()
prof.dump_stats(tmp.name)
fh = open('s3ql_profile.txt', 'w')
p = pstats.Stats(tmp.name, stream=fh)
tmp.close()
p.strip_dirs()
p.sort_stats('cumulative')
p.print_stats(50)
p.sort_stats('time')
p.print_stats(50)
fh.close()