#
# Licensed to the Apache Software Foundation (ASF) under one or more
# contributor license agreements. See the NOTICE file distributed with
# this work for additional information regarding copyright ownership.
# The ASF licenses this file to You under the Apache License, Version 2.0
# (the "License"); you may not use this file except in compliance with
# the License. You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
#
# pytype: skip-file
import contextlib
import dataclasses
import glob
import hashlib
import logging
import os
import re
import shutil
import signal
import socket
import subprocess
import threading
import time
import zipfile
from typing import Any
from typing import Set
from urllib.error import URLError
from urllib.request import urlopen
import grpc
from apache_beam.io.filesystems import FileSystems
from apache_beam.version import __version__ as beam_version
_LOGGER = logging.getLogger(__name__)
@dataclasses.dataclass
class _SharedCacheEntry:
obj: Any
owners: Set[str]
class _SharedCache:
"""A cache that keeps objects alive (and repeatedly returns the same instance)
until the last user indicates that they're done.
The typical usage is as follows::
try:
token = cache.register()
# All objects retrieved from the cache from this point on will be memoized
# and kept alive (including across other threads and callers) at least
# until the purge is called below (and possibly longer, if other calls
# to register were made).
obj = cache.get(...)
another_obj = cache.get(...)
...
finally:
cache.purge(token)
"""
def __init__(self, constructor, destructor):
self._constructor = constructor
self._destructor = destructor
self._live_owners = set()
self._cache = {}
self._lock = threading.Lock()
self._counter = 0
def _next_id(self):
with self._lock:
self._counter += 1
return self._counter
def register(self):
owner = self._next_id()
self._live_owners.add(owner)
return owner
def purge(self, owner):
if owner not in self._live_owners:
raise ValueError(f"{owner} not in {self._live_owners}")
self._live_owners.remove(owner)
to_delete = []
with self._lock:
for key, entry in list(self._cache.items()):
if owner in entry.owners:
entry.owners.remove(owner)
if not entry.owners:
to_delete.append(entry.obj)
del self._cache[key]
# Actually call the destructors outside of the lock.
for value in to_delete:
self._destructor(value)
def get(self, *key):
if not self._live_owners:
raise RuntimeError("At least one owner must be registered.")
with self._lock:
if key not in self._cache:
self._cache[key] = _SharedCacheEntry(self._constructor(*key), set())
for owner in self._live_owners:
self._cache[key].owners.add(owner)
return self._cache[key].obj
[docs]class SubprocessServer(object):
"""An abstract base class for running GRPC Servers as an external process.
This class acts as a context which will start up a server, provides a stub
to connect to it, and then shuts the server down. For example::
with SubprocessServer(GrpcStubClass, [executable, arg, ...]) as stub:
stub.CallService(...)
"""
def __init__(self, stub_class, cmd, port=None):
"""Creates the server object.
:param stub_class: the auto-generated GRPC client stub class used for
connecting to the GRPC service
:param cmd: command (including arguments) for starting up the server,
suitable for passing to `subprocess.POpen`.
:param port: (optional) the port at which the subprocess will serve its
service. If not given, one will be randomly chosen and the special
string "{{PORT}}" will be substituted in the command line arguments
with the chosen port.
"""
self._owner_id = None
self._stub_class = stub_class
self._cmd = [str(arg) for arg in cmd]
self._port = port
self._grpc_channel = None
[docs] @classmethod
@contextlib.contextmanager
def cache_subprocesses(cls):
"""A context that ensures any subprocess created or used in its duration
stay alive for at least the duration of this context.
These subprocesses may be shared with other contexts as well.
"""
try:
unique_id = cls._cache.register()
yield
finally:
cls._cache.purge(unique_id)
def __enter__(self):
return self.start()
def __exit__(self, *unused_args):
self.stop()
[docs] def start(self):
try:
process, endpoint = self.start_process()
wait_secs = .1
channel_options = [("grpc.max_receive_message_length", -1),
("grpc.max_send_message_length", -1)]
self._grpc_channel = grpc.insecure_channel(
endpoint, options=channel_options)
channel_ready = grpc.channel_ready_future(self._grpc_channel)
while True:
if process is not None and process.poll() is not None:
_LOGGER.error("Started job service with %s", process.args)
raise RuntimeError(
'Service failed to start up with error %s' % process.poll())
try:
channel_ready.result(timeout=wait_secs)
break
except (grpc.FutureTimeoutError, grpc.RpcError):
wait_secs *= 1.2
logging.log(
logging.WARNING if wait_secs > 1 else logging.DEBUG,
'Waiting for grpc channel to be ready at %s.',
endpoint)
return self._stub_class(self._grpc_channel)
except: # pylint: disable=bare-except
_LOGGER.exception("Error bringing up service")
self.stop()
raise
[docs] def start_process(self):
if self._owner_id is not None:
self._cache.purge(self._owner_id)
self._owner_id = self._cache.register()
return self._cache.get(tuple(self._cmd), self._port)
def _really_start_process(cmd, port):
if not port:
port, = pick_port(None)
cmd = [arg.replace('{{PORT}}', str(port)) for arg in cmd] # pylint: disable=not-an-iterable
endpoint = 'localhost:%s' % port
_LOGGER.info("Starting service with %s", str(cmd).replace("',", "'"))
process = subprocess.Popen(
cmd, stdout=subprocess.PIPE, stderr=subprocess.STDOUT)
# Emit the output of this command as info level logging.
def log_stdout():
line = process.stdout.readline()
while line:
# The log obtained from stdout is bytes, decode it into string.
# Remove newline via rstrip() to not print an empty line.
_LOGGER.info(line.decode(errors='backslashreplace').rstrip())
line = process.stdout.readline()
t = threading.Thread(target=log_stdout)
t.daemon = True
t.start()
return process, endpoint
[docs] def stop(self):
self.stop_process()
[docs] def stop_process(self):
if self._owner_id is not None:
self._cache.purge(self._owner_id)
self._owner_id = None
if self._grpc_channel:
try:
self._grpc_channel.close()
except: # pylint: disable=bare-except
_LOGGER.error(
"Could not close the gRPC channel started for the "
"expansion service")
finally:
self._grpc_channel = None
def _really_stop_process(process_and_endpoint):
process, _ = process_and_endpoint # pylint: disable=unpacking-non-sequence
if not process:
return
for _ in range(5):
if process.poll() is not None:
break
logging.debug("Sending SIGINT to process")
try:
process.send_signal(signal.SIGINT)
except ValueError:
# process.send_signal raises a ValueError on Windows.
process.terminate()
time.sleep(1)
if process.poll() is None:
process.kill()
_cache = _SharedCache(
constructor=_really_start_process, destructor=_really_stop_process)
[docs]class JavaJarServer(SubprocessServer):
MAVEN_CENTRAL_REPOSITORY = 'https://repo.maven.apache.org/maven2'
BEAM_GROUP_ID = 'org.apache.beam'
JAR_CACHE = os.path.expanduser("~/.apache_beam/cache/jars")
_BEAM_SERVICES = type(
'local', (threading.local, ),
dict(__init__=lambda self: setattr(self, 'replacements', {})))()
def __init__(self, stub_class, path_to_jar, java_arguments, classpath=None):
if classpath:
# java -jar ignores the classpath, so we make a new jar that embeds
# the requested classpath.
path_to_jar = self.make_classpath_jar(path_to_jar, classpath)
super().__init__(
stub_class, ['java', '-jar', path_to_jar] + list(java_arguments))
self._existing_service = path_to_jar if is_service_endpoint(
path_to_jar) else None
[docs] def start_process(self):
if self._existing_service:
return None, self._existing_service
else:
if not shutil.which('java'):
raise RuntimeError(
'Java must be installed on this system to use this '
'transform/runner.')
return super().start_process()
[docs] def stop_process(self):
if self._existing_service:
pass
else:
return super().stop_process()
[docs] @classmethod
def jar_name(cls, artifact_id, version, classifier=None, appendix=None):
return '-'.join(
filter(None, [artifact_id, appendix, version, classifier])) + '.jar'
[docs] @classmethod
def path_to_maven_jar(
cls,
artifact_id,
group_id,
version,
repository=MAVEN_CENTRAL_REPOSITORY,
classifier=None,
appendix=None):
return '/'.join([
repository,
group_id.replace('.', '/'),
artifact_id,
version,
cls.jar_name(artifact_id, version, classifier, appendix)
])
[docs] @classmethod
def path_to_beam_jar(
cls,
gradle_target,
appendix=None,
version=beam_version,
artifact_id=None):
if gradle_target in cls._BEAM_SERVICES.replacements:
return cls._BEAM_SERVICES.replacements[gradle_target]
gradle_package = gradle_target.strip(':').rsplit(':', 1)[0]
if not artifact_id:
artifact_id = 'beam-' + gradle_package.replace(':', '-')
project_root = os.path.sep.join(
os.path.abspath(__file__).split(os.path.sep)[:-5])
local_path = os.path.join(
project_root,
gradle_package.replace(':', os.path.sep),
'build',
'libs',
cls.jar_name(
artifact_id,
version.replace('.dev', ''),
classifier='SNAPSHOT',
appendix=appendix))
if os.path.exists(local_path):
_LOGGER.info('Using pre-built snapshot at %s', local_path)
return local_path
elif '.dev' in version:
# TODO: Attempt to use nightly snapshots?
raise RuntimeError(
(
'%s not found. '
'Please build the server with \n cd %s; ./gradlew %s') %
(local_path, os.path.abspath(project_root), gradle_target))
else:
return cls.path_to_maven_jar(
artifact_id,
cls.BEAM_GROUP_ID,
version,
cls.MAVEN_CENTRAL_REPOSITORY,
appendix=appendix)
[docs] @classmethod
def local_jar(cls, url, cache_dir=None):
if cache_dir is None:
cache_dir = cls.JAR_CACHE
# TODO: Verify checksum?
if is_service_endpoint(url):
return url
elif os.path.exists(url):
return url
else:
cached_jar = os.path.join(cache_dir, os.path.basename(url))
if os.path.exists(cached_jar):
_LOGGER.info('Using cached job server jar from %s' % url)
else:
_LOGGER.info('Downloading job server jar from %s' % url)
if not os.path.exists(cache_dir):
os.makedirs(cache_dir)
# TODO: Clean up this cache according to some policy.
try:
try:
url_read = FileSystems.open(url)
except ValueError:
url_read = urlopen(url)
with open(cached_jar + '.tmp', 'wb') as jar_write:
shutil.copyfileobj(url_read, jar_write, length=1 << 20)
os.rename(cached_jar + '.tmp', cached_jar)
except URLError as e:
raise RuntimeError(
'Unable to fetch remote job server jar at %s: %s' % (url, e))
return cached_jar
[docs] @classmethod
@contextlib.contextmanager
def beam_services(cls, replacements):
try:
old = cls._BEAM_SERVICES.replacements
cls._BEAM_SERVICES.replacements = dict(old, **replacements)
yield
finally:
cls._BEAM_SERVICES.replacements = old
[docs] @classmethod
def make_classpath_jar(cls, main_jar, extra_jars, cache_dir=None):
if cache_dir is None:
cache_dir = cls.JAR_CACHE
composite_jar_dir = os.path.join(cache_dir, 'composite-jars')
os.makedirs(composite_jar_dir, exist_ok=True)
classpath = []
# Class-Path references from a jar must be relative, so we create
# a relatively-addressable subdirectory with symlinks to all the
# required jars.
for pattern in [main_jar] + list(extra_jars):
for path in glob.glob(pattern) or [pattern]:
path = os.path.abspath(path)
rel_path = hashlib.sha256(
path.encode('utf-8')).hexdigest() + os.path.splitext(path)[1]
classpath.append(rel_path)
if not os.path.lexists(os.path.join(composite_jar_dir, rel_path)):
os.symlink(path, os.path.join(composite_jar_dir, rel_path))
# Now create a single jar that simply references the rest and has the same
# main class as main_jar.
composite_jar = os.path.join(
composite_jar_dir,
hashlib.sha256(' '.join(sorted(classpath)).encode('ascii')).hexdigest()
+ '.jar')
if not os.path.exists(composite_jar):
with zipfile.ZipFile(main_jar) as main:
with main.open('META-INF/MANIFEST.MF') as manifest:
main_class = next(
filter(lambda line: line.startswith(b'Main-Class: '), manifest))
with zipfile.ZipFile(composite_jar + '.tmp', 'w') as composite:
with composite.open('META-INF/MANIFEST.MF', 'w') as manifest:
manifest.write(b'Manifest-Version: 1.0\n')
manifest.write(main_class)
manifest.write(
b'Class-Path: ' + '\n '.join(classpath).encode('ascii') + b'\n')
os.rename(composite_jar + '.tmp', composite_jar)
return composite_jar
[docs]def is_service_endpoint(path):
"""Checks whether the path conforms to the 'beam_services' PipelineOption."""
return re.match(r'^[a-zA-Z0-9.-]+:\d+$', path)
[docs]def pick_port(*ports):
"""
Returns a list of ports, same length as input ports list, but replaces
all None or 0 ports with a random free port.
"""
sockets = []
def find_free_port(port):
if port:
return port
else:
try:
s = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
except OSError as e:
# [Errno 97] Address family not supported by protocol
# Likely indicates we are in an IPv6-only environment (BEAM-10618). Try
# again with AF_INET6.
if e.errno == 97:
s = socket.socket(socket.AF_INET6, socket.SOCK_STREAM)
else:
raise e
sockets.append(s)
s.bind(('localhost', 0))
return s.getsockname()[1]
ports = list(map(find_free_port, ports))
# Close sockets only now to avoid the same port to be chosen twice
for s in sockets:
s.close()
return ports