import gflags
+import httplib
import httplib2
import logging
import os
import bz2
import zlib
import fcntl
+import time
+import threading
from apiclient import errors
from apiclient.discovery import build
-class CredentialsFromEnv:
+if 'ARVADOS_DEBUG' in os.environ:
+ logging.basicConfig(level=logging.DEBUG)
+
+class CredentialsFromEnv(object):
@staticmethod
def http_request(self, uri, **kwargs):
from httplib import BadStatusLine
url = ('https://%s/discovery/v1/apis/'
'{api}/{apiVersion}/rest' % os.environ['ARVADOS_API_HOST'])
credentials = CredentialsFromEnv()
-http = httplib2.Http()
+
+# Use system's CA certificates (if we find them) instead of httplib2's
+ca_certs = '/etc/ssl/certs/ca-certificates.crt'
+if not os.path.exists(ca_certs):
+ ca_certs = None # use httplib2 default
+
+http = httplib2.Http(ca_certs=ca_certs)
http = credentials.authorize(http)
-http.disable_ssl_certificate_validation=True
+if re.match(r'(?i)^(true|1|yes)$',
+ os.environ.get('ARVADOS_API_HOST_INSECURE', '')):
+ http.disable_ssl_certificate_validation=True
service = build("arvados", "v1", http=http, discoveryServiceUrl=url)
def task_set_output(self,s):
_current_job = t
return t
+def getjobparam(*args):
+ return current_job()['script_parameters'].get(*args)
+
def api():
return service
-class JobTask:
- def __init__(self, parameters=dict(), resource_limits=dict()):
- print "init jobtask %s %s" % (parameters, resource_limits)
+class JobTask(object):
+ def __init__(self, parameters=dict(), runtime_constraints=dict()):
+ print "init jobtask %s %s" % (parameters, runtime_constraints)
class job_setup:
@staticmethod
).execute()
exit(0)
+ @staticmethod
+ def one_task_per_input_stream(if_sequence=0, and_end_task=True):
+ if if_sequence != current_task()['sequence']:
+ return
+ job_input = current_job()['script_parameters']['input']
+ cr = CollectionReader(job_input)
+ for s in cr.all_streams():
+ task_input = s.tokens()
+ new_task_attrs = {
+ 'job_uuid': current_job()['uuid'],
+ 'created_by_job_task_uuid': current_task()['uuid'],
+ 'sequence': if_sequence + 1,
+ 'parameters': {
+ 'input':task_input
+ }
+ }
+ service.job_tasks().create(job_task=json.dumps(new_task_attrs)).execute()
+ if and_end_task:
+ service.job_tasks().update(uuid=current_task()['uuid'],
+ job_task=json.dumps({'success':True})
+ ).execute()
+ exit(0)
+
class util:
+ @staticmethod
+ def clear_tmpdir(path=None):
+ """
+ Ensure the given directory (or TASK_TMPDIR if none given)
+ exists and is empty.
+ """
+ if path == None:
+ path = current_task().tmpdir
+ if os.path.exists(path):
+ p = subprocess.Popen(['rm', '-rf', path])
+ stdout, stderr = p.communicate(None)
+ if p.returncode != 0:
+ raise Exception('rm -rf %s: %s' % (path, stderr))
+ os.mkdir(path)
+
@staticmethod
def run_command(execargs, **kwargs):
- if 'stdin' not in kwargs:
- kwargs['stdin'] = subprocess.PIPE
- if 'stdout' not in kwargs:
- kwargs['stdout'] = subprocess.PIPE
- if 'stderr' not in kwargs:
- kwargs['stderr'] = subprocess.PIPE
- p = subprocess.Popen(execargs, close_fds=True, shell=False,
- **kwargs)
+ kwargs.setdefault('stdin', subprocess.PIPE)
+ kwargs.setdefault('stdout', subprocess.PIPE)
+ kwargs.setdefault('stderr', sys.stderr)
+ kwargs.setdefault('close_fds', True)
+ kwargs.setdefault('shell', False)
+ p = subprocess.Popen(execargs, **kwargs)
stdoutdata, stderrdata = p.communicate(None)
if p.returncode != 0:
raise Exception("run_command %s exit %d:\n%s" %
for f in CollectionReader(tarball).all_files():
if re.search('\.(tbz|tar.bz2)$', f.name()):
- p = tar_extractor(path, 'j')
+ p = util.tar_extractor(path, 'j')
elif re.search('\.(tgz|tar.gz)$', f.name()):
- p = tar_extractor(path, 'z')
+ p = util.tar_extractor(path, 'z')
elif re.search('\.tar$', f.name()):
- p = tar_extractor(path, '')
+ p = util.tar_extractor(path, '')
else:
raise Exception("tarball_extract cannot handle filename %s"
% f.name())
collection -- collection locator
path -- where to extract: absolute, or relative to job tmp
"""
+ matches = re.search(r'^([0-9a-f]+)(\+[\w@]+)*$', collection)
+ if matches:
+ collection_hash = matches.group(1)
+ else:
+ collection_hash = hashlib.md5(collection).hexdigest()
if not re.search('^/', path):
path = os.path.join(current_job().tmpdir, path)
lockfile = open(path + '.lock', 'w')
os.mkdir(path)
already_have_it = False
try:
- if os.readlink(os.path.join(path, '.locator')) == collection:
+ if os.readlink(os.path.join(path, '.locator')) == collection_hash:
already_have_it = True
except OSError:
pass
# emulate "rm -f" (i.e., if the file does not exist, we win)
- files_got = []
try:
os.unlink(os.path.join(path, '.locator'))
except OSError:
if os.path.exists(os.path.join(path, '.locator')):
os.unlink(os.path.join(path, '.locator'))
- for f in CollectionReader(collection).all_files():
+ files_got = []
+ for s in CollectionReader(collection).all_streams():
+ stream_name = s.name()
+ for f in s.all_files():
+ if (files == [] or
+ ((f.name() not in files_got) and
+ (f.name() in files or
+ (decompress and f.decompressed_name() in files)))):
+ outname = f.decompressed_name() if decompress else f.name()
+ files_got += [outname]
+ if os.path.exists(os.path.join(path, stream_name, outname)):
+ continue
+ util.mkdir_dash_p(os.path.dirname(os.path.join(path, stream_name, outname)))
+ outfile = open(os.path.join(path, stream_name, outname), 'wb')
+ for buf in (f.readall_decompressed() if decompress
+ else f.readall()):
+ outfile.write(buf)
+ outfile.close()
+ if len(files_got) < len(files):
+ raise Exception("Wanted files %s but only got %s from %s" % (files, files_got, map(lambda z: z.name(), list(CollectionReader(collection).all_files()))))
+ os.symlink(collection_hash, os.path.join(path, '.locator'))
+
+ lockfile.close()
+ return path
+
+ @staticmethod
+ def mkdir_dash_p(path):
+ if not os.path.exists(path):
+ util.mkdir_dash_p(os.path.dirname(path))
+ try:
+ os.mkdir(path)
+ except OSError:
+ if not os.path.exists(path):
+ os.mkdir(path)
+
+ @staticmethod
+ def stream_extract(stream, path, files=[], decompress=True):
+ """Retrieve a stream from Keep and extract it to a local
+ directory. Return the absolute path where the stream was
+ extracted.
+
+ stream -- StreamReader object
+ path -- where to extract: absolute, or relative to job tmp
+ """
+ if not re.search('^/', path):
+ path = os.path.join(current_job().tmpdir, path)
+ lockfile = open(path + '.lock', 'w')
+ fcntl.flock(lockfile, fcntl.LOCK_EX)
+ try:
+ os.stat(path)
+ except OSError:
+ os.mkdir(path)
+
+ files_got = []
+ for f in stream.all_files():
if (files == [] or
((f.name() not in files_got) and
(f.name() in files or
outname = f.decompressed_name() if decompress else f.name()
files_got += [outname]
if os.path.exists(os.path.join(path, outname)):
- continue
- outfile = open(os.path.join(path, outname), 'w')
+ os.unlink(os.path.join(path, outname))
+ util.mkdir_dash_p(os.path.dirname(os.path.join(path, outname)))
+ outfile = open(os.path.join(path, outname), 'wb')
for buf in (f.readall_decompressed() if decompress
else f.readall()):
outfile.write(buf)
outfile.close()
if len(files_got) < len(files):
- raise Exception("Wanted files %s but only got %s from %s" % (files, files_got, map(lambda z: z.name(), list(CollectionReader(collection).all_files()))))
- os.symlink(collection, os.path.join(path, '.locator'))
-
+ raise Exception("Wanted files %s but only got %s from %s" %
+ (files, files_got, map(lambda z: z.name(),
+ list(stream.all_files()))))
lockfile.close()
return path
-class DataReader:
- def __init__(self, data_locator):
- self.data_locator = data_locator
- self.p = subprocess.Popen(["whget", "-r", self.data_locator, "-"],
- stdout=subprocess.PIPE,
- stdin=None, stderr=subprocess.PIPE,
- shell=False, close_fds=True)
- def __enter__(self):
- pass
- def __exit__(self):
- self.close()
- def read(self, size, **kwargs):
- return self.p.stdout.read(size, **kwargs)
- def close(self):
- self.p.stdout.close()
- if not self.p.stderr.closed:
- for err in self.p.stderr:
- print >> sys.stderr, err
- self.p.stderr.close()
- self.p.wait()
- if self.p.returncode != 0:
- raise Exception("whget subprocess exited %d" % self.p.returncode)
-
-class StreamFileReader:
+ @staticmethod
+ def listdir_recursive(dirname, base=None):
+ allfiles = []
+ for ent in sorted(os.listdir(dirname)):
+ ent_path = os.path.join(dirname, ent)
+ ent_base = os.path.join(base, ent) if base else ent
+ if os.path.isdir(ent_path):
+ allfiles += util.listdir_recursive(ent_path, ent_base)
+ else:
+ allfiles += [ent_base]
+ return allfiles
+
+class StreamFileReader(object):
def __init__(self, stream, pos, size, name):
self._stream = stream
self._pos = pos
return string.join(self._stream.tokens_for_range(self._pos, self._size),
" ") + "\n"
-class StreamReader:
+class StreamReader(object):
def __init__(self, tokens):
self._tokens = tokens
self._current_datablock_data = None
self.files += [[int(pos), int(size), name]]
else:
raise Exception("Invalid manifest format")
+
+ def tokens(self):
+ return self._tokens
def tokens_for_range(self, range_start, range_size):
resp = [self._stream_name]
return_all_tokens = False
self._pos += len(data)
return data
-class CollectionReader:
+class CollectionReader(object):
def __init__(self, manifest_locator_or_text):
if re.search(r'^\S+( [a-f0-9]{32,}(\+\S+)*)+( \d+:\d+:\S+)+\n', manifest_locator_or_text):
self._manifest_text = manifest_locator_or_text
self._manifest_text = Keep.get(self._manifest_locator)
self._streams = []
for stream_line in self._manifest_text.split("\n"):
- stream_tokens = stream_line.split()
- self._streams += [stream_tokens]
+ if stream_line != '':
+ stream_tokens = stream_line.split()
+ self._streams += [stream_tokens]
def all_streams(self):
self._populate()
resp = []
self._populate()
return self._manifest_text
-class CollectionWriter:
+class CollectionWriter(object):
KEEP_BLOCK_SIZE = 2**26
def __init__(self):
self._data_buffer = []
pass
def __exit__(self):
self.finish()
+ def write_directory_tree(self,
+ path, stream_name='.', max_manifest_depth=-1):
+ self.start_new_stream(stream_name)
+ todo = []
+ if max_manifest_depth == 0:
+ dirents = sorted(util.listdir_recursive(path))
+ else:
+ dirents = sorted(os.listdir(path))
+ for dirent in dirents:
+ target = os.path.join(path, dirent)
+ if os.path.isdir(target):
+ todo += [[target,
+ os.path.join(stream_name, dirent),
+ max_manifest_depth-1]]
+ else:
+ self.start_new_file(dirent)
+ with open(target, 'rb') as f:
+ while True:
+ buf = f.read(2**26)
+ if len(buf) == 0:
+ break
+ self.write(buf)
+ self.finish_current_stream()
+ map(lambda x: self.write_directory_tree(*x), todo)
+
def write(self, newdata):
+ if hasattr(newdata, '__iter__'):
+ for s in newdata:
+ self.write(s)
+ return
self._data_buffer += [newdata]
self._data_buffer_len += len(newdata)
self._current_stream_length += len(newdata)
self.finish_current_file()
self.set_current_file_name(newfilename)
def set_current_file_name(self, newfilename):
+ newfilename = re.sub(r' ', '\\\\040', newfilename)
if re.search(r'[ \t\n]', newfilename):
raise AssertionError("Manifest filenames cannot contain whitespace")
self._current_file_name = newfilename
def set_current_stream_name(self, newstreamname):
if re.search(r'[ \t\n]', newstreamname):
raise AssertionError("Manifest stream names cannot contain whitespace")
- self._current_stream_name = newstreamname
+ self._current_stream_name = '.' if newstreamname=='' else newstreamname
def current_stream_name(self):
return self._current_stream_name
def finish_current_stream(self):
self.finish_current_stream()
manifest = ''
for stream in self._finished_streams:
+ if not re.search(r'^\.(/.*)?$', stream[0]):
+ manifest += './'
manifest += stream[0]
if len(stream[1]) == 0:
manifest += " d41d8cd98f00b204e9800998ecf8427e+0"
manifest += "\n"
return manifest
+global_client_object = None
+
class Keep:
@staticmethod
- def put(data):
- if 'KEEP_LOCAL_STORE' in os.environ:
- return Keep.local_store_put(data)
- p = subprocess.Popen(["whput", "-"],
- stdout=subprocess.PIPE,
- stdin=subprocess.PIPE,
- stderr=subprocess.PIPE,
- shell=False, close_fds=True)
- stdoutdata, stderrdata = p.communicate(data)
- if p.returncode != 0:
- raise Exception("whput subprocess exited %d - stderr:\n%s" % (p.returncode, stderrdata))
- return stdoutdata.rstrip()
+ def global_client_object():
+ global global_client_object
+ if global_client_object == None:
+ global_client_object = KeepClient()
+ return global_client_object
+
@staticmethod
- def get(locator):
+ def get(locator, **kwargs):
+ return Keep.global_client_object().get(locator, **kwargs)
+
+ @staticmethod
+ def put(data, **kwargs):
+ return Keep.global_client_object().put(data, **kwargs)
+
+class KeepClient(object):
+
+ class ThreadLimiter(object):
+ """
+ Limit the number of threads running at a given time to
+ {desired successes} minus {successes reported}. When successes
+ reported == desired, wake up the remaining threads and tell
+ them to quit.
+
+ Should be used in a "with" block.
+ """
+ def __init__(self, todo):
+ self._todo = todo
+ self._done = 0
+ self._todo_lock = threading.Semaphore(todo)
+ self._done_lock = threading.Lock()
+ def __enter__(self):
+ self._todo_lock.acquire()
+ return self
+ def __exit__(self, type, value, traceback):
+ self._todo_lock.release()
+ def shall_i_proceed(self):
+ """
+ Return true if the current thread should do stuff. Return
+ false if the current thread should just stop.
+ """
+ with self._done_lock:
+ return (self._done < self._todo)
+ def increment_done(self):
+ """
+ Report that the current thread was successful.
+ """
+ with self._done_lock:
+ self._done += 1
+ def done(self):
+ """
+ Return how many successes were reported.
+ """
+ with self._done_lock:
+ return self._done
+
+ class KeepWriterThread(threading.Thread):
+ """
+ Write a blob of data to the given Keep server. Call
+ increment_done() of the given ThreadLimiter if the write
+ succeeds.
+ """
+ def __init__(self, **kwargs):
+ super(KeepClient.KeepWriterThread, self).__init__()
+ self.args = kwargs
+ def run(self):
+ with self.args['thread_limiter'] as limiter:
+ if not limiter.shall_i_proceed():
+ # My turn arrived, but the job has been done without
+ # me.
+ return
+ logging.debug("KeepWriterThread %s proceeding %s %s" %
+ (str(threading.current_thread()),
+ self.args['data_hash'],
+ self.args['service_root']))
+ h = httplib2.Http()
+ url = self.args['service_root'] + self.args['data_hash']
+ api_token = os.environ['ARVADOS_API_TOKEN']
+ headers = {'Authorization': "OAuth2 %s" % api_token}
+ try:
+ resp, content = h.request(url.encode('utf-8'), 'PUT',
+ headers=headers,
+ body=self.args['data'])
+ if (resp['status'] == '401' and
+ re.match(r'Timestamp verification failed', content)):
+ body = KeepClient.sign_for_old_server(
+ self.args['data_hash'],
+ self.args['data'])
+ h = httplib2.Http()
+ resp, content = h.request(url.encode('utf-8'), 'PUT',
+ headers=headers,
+ body=body)
+ if re.match(r'^2\d\d$', resp['status']):
+ logging.debug("KeepWriterThread %s succeeded %s %s" %
+ (str(threading.current_thread()),
+ self.args['data_hash'],
+ self.args['service_root']))
+ return limiter.increment_done()
+ logging.warning("Request fail: PUT %s => %s %s" %
+ (url, resp['status'], content))
+ except (httplib2.HttpLib2Error, httplib.HTTPException) as e:
+ logging.warning("Request fail: PUT %s => %s: %s" %
+ (url, type(e), str(e)))
+
+ def __init__(self):
+ self.lock = threading.Lock()
+ self.service_roots = None
+
+ def shuffled_service_roots(self, hash):
+ if self.service_roots == None:
+ self.lock.acquire()
+ keep_disks = api().keep_disks().list().execute()['items']
+ roots = (("http%s://%s:%d/" %
+ ('s' if f['service_ssl_flag'] else '',
+ f['service_host'],
+ f['service_port']))
+ for f in keep_disks)
+ self.service_roots = sorted(set(roots))
+ logging.debug(str(self.service_roots))
+ self.lock.release()
+ seed = hash
+ pool = self.service_roots[:]
+ pseq = []
+ while len(pool) > 0:
+ if len(seed) < 8:
+ if len(pseq) < len(hash) / 4: # first time around
+ seed = hash[-4:] + hash
+ else:
+ seed += hash
+ probe = int(seed[0:8], 16) % len(pool)
+ pseq += [pool[probe]]
+ pool = pool[:probe] + pool[probe+1:]
+ seed = seed[8:]
+ logging.debug(str(pseq))
+ return pseq
+
+ def get(self, locator):
if 'KEEP_LOCAL_STORE' in os.environ:
- return Keep.local_store_get(locator)
- p = subprocess.Popen(["whget", locator, "-"],
- stdout=subprocess.PIPE,
- stdin=None,
- stderr=subprocess.PIPE,
- shell=False, close_fds=True)
- stdoutdata, stderrdata = p.communicate(None)
- if p.returncode != 0:
- raise Exception("whget subprocess exited %d - stderr:\n%s" % (p.returncode, stderrdata))
+ return KeepClient.local_store_get(locator)
+ expect_hash = re.sub(r'\+.*', '', locator)
+ for service_root in self.shuffled_service_roots(expect_hash):
+ h = httplib2.Http()
+ url = service_root + expect_hash
+ api_token = os.environ['ARVADOS_API_TOKEN']
+ headers = {'Authorization': "OAuth2 %s" % api_token,
+ 'Accept': 'application/octet-stream'}
+ try:
+ resp, content = h.request(url.encode('utf-8'), 'GET',
+ headers=headers)
+ if re.match(r'^2\d\d$', resp['status']):
+ m = hashlib.new('md5')
+ m.update(content)
+ md5 = m.hexdigest()
+ if md5 == expect_hash:
+ return content
+ logging.warning("Checksum fail: md5(%s) = %s" % (url, md5))
+ except (httplib2.HttpLib2Error, httplib.ResponseNotReady) as e:
+ logging.info("Request fail: GET %s => %s: %s" %
+ (url, type(e), str(e)))
+ raise Exception("Not found: %s" % expect_hash)
+
+ def put(self, data, **kwargs):
+ if 'KEEP_LOCAL_STORE' in os.environ:
+ return KeepClient.local_store_put(data)
m = hashlib.new('md5')
- m.update(stdoutdata)
- try:
- if locator.index(m.hexdigest()) == 0:
- return stdoutdata
- except ValueError:
- pass
- raise Exception("md5 checksum mismatch: md5(get(%s)) == %s" % (locator, m.hexdigest()))
+ m.update(data)
+ data_hash = m.hexdigest()
+ have_copies = 0
+ want_copies = kwargs.get('copies', 2)
+ if not (want_copies > 0):
+ return data_hash
+ threads = []
+ thread_limiter = KeepClient.ThreadLimiter(want_copies)
+ for service_root in self.shuffled_service_roots(data_hash):
+ t = KeepClient.KeepWriterThread(data=data,
+ data_hash=data_hash,
+ service_root=service_root,
+ thread_limiter=thread_limiter)
+ t.start()
+ threads += [t]
+ for t in threads:
+ t.join()
+ have_copies = thread_limiter.done()
+ if have_copies == want_copies:
+ return (data_hash + '+' + str(len(data)))
+ raise Exception("Write fail for %s: wanted %d but wrote %d" %
+ (data_hash, want_copies, have_copies))
+
+ @staticmethod
+ def sign_for_old_server(data_hash, data):
+ return (("-----BEGIN PGP SIGNED MESSAGE-----\n\n\n%d %s\n-----BEGIN PGP SIGNATURE-----\n\n-----END PGP SIGNATURE-----\n" % (int(time.time()), data_hash)) + data)
+
+
@staticmethod
def local_store_put(data):
m = hashlib.new('md5')
projects / arvados.git / blobdiff