def _start_collectors(self, master_state, slave_state):
def collect_deltas(source, dest):
for delta in source[0].get_changes():
if delta:
LOGGER.debug(
f"Incremental delta received from {source[0]}:\n{delta}"
)
self._queue.put((delta, source, dest))
LOGGER.trace("Collector thread is terminating.")
watches = [
threading.Thread(
name="Watch-MS",
target=collect_deltas,
args=((self.master_fs, master_state), (self.slave_fs, slave_state)),
),
threading.Thread(
name="Watch-SM",
target=collect_deltas,
args=((self.slave_fs, slave_state), (self.master_fs, master_state)),
),
]
for watch in watches:
watch.daemon = True # Kill with main thread
watch.start()
while True:
LOGGER.info("Watching for FS state changes")
delta, source, dest = self._queue.get()
delta.apply(source, dest)
LOGGER.debug(f"Incremental delta applied to {dest[0]}")
for watch in watches:
watch.join()
def read(self, path):
file = self.search(path)
if not file:
raise FileNotFoundError(path)
LOGGER.info(f"Downloading {file} at {path}")
request = self._drive.files().get_media(fileId=file._id)
stream = io.BytesIO()
self._download(request, stream)
# TODO: This code should be fixed in order to support Google Docs
#
# except HttpError as e:
# LOGGER.error(f"Error reading {file}. Cause: {e}")
# # Export a Google Doc file
# if e.resp.status == 403:
# pass
# try:
# self._download(
# self._drive.files().export_media(
# fileId=file.id, mimeType=file.mime_type
# ),
# stream,
# )
# except HttpError as f:
# if e.resp.status != 403:
# raise
stream.flush()
stream.seek(0)
return stream
def collect_deltas(source, dest):
for delta in source[0].get_changes():
if delta:
LOGGER.debug(
f"Incremental delta received from {source[0]}:\n{delta}"
)
self._queue.put((delta, source, dest))
LOGGER.trace("Collector thread is terminating.")
def remove(self, path):
LOGGER.debug(f"Removing local file at {path}")
abs_path = self._abs_path(path)
try:
os.remove(abs_path)
except IsADirectoryError:
rmtree(abs_path)
except FileNotFoundError:
pass
def __exit__(self, exc_type, exc_value, traceback):
if self._orig_sig_handlers:
LOGGER.debug("Restoring signal handlers")
for h, s in zip(self._orig_sig_handlers, self.SIGNALS):
signal.signal(s, h)
if exc_value and exc_type not in (FSNotReady, ):
LOGGER.critical(
f"Emergency shutdown. Current FS states persisted. Cause: {exc_value}"
)
raise exc_value
def load_fs_states(self, alias=None):
if not alias:
alias, = self._config.keys()
self._master_state_file = os.path.join(STATES_DIR,
f"{alias}_master.pickle")
master_state = State.load(self._master_state_file)
LOGGER.debug(f"Previous state of Master FS loaded")
self._slave_state_file = os.path.join(STATES_DIR,
f"{alias}_slave.pickle")
slave_state = State.load(self._slave_state_file)
LOGGER.debug("Previous state of Slave FS loaded")
return master_state, slave_state
def wrapper(*args, **kwargs):
try:
CONNECTED.wait()
return f(*args, **kwargs)
except HttpError as e:
LOGGER.warning(
f"HTTP error {e.resp.status} suppressed after call to {f} with arguments "
f"{args}, {kwargs}: {e}"
)
except ServerNotFoundError as e:
LOGGER.error(
f"Cannot call {f} with arguments {args}, {kwargs} at this time. "
f"Reason: {e}"
)
CONNECTED.clear()
CONNECTED.wait()
return suppresserror(f)(*args, **kwargs)
def func_wrapper(*args, **kwargs):
backoff = delay
while True:
try:
error = False
return f(*args, **kwargs)
except FSNotReady as e:
LOGGER.error(f"A file system is not ready yet: {e}")
LOGGER.info(
f"A new start attempt will be made in {int(backoff)} seconds"
)
sleep(backoff)
backoff = min(cap, backoff * ratio)
error = True
finally:
if not error:
backoff = delay
def get_changes(self):
backoff = 5
while True:
LOGGER.trace(f"Getting Drive changes (backoff: {backoff})")
try:
with STATE_LOCK:
yield self._get_changes()
backoff = 5
CONNECTED.set()
except ServerNotFoundError:
backoff *= 1.618
LOGGER.error(
f"The Google Drive API is unreachable. Retrying in {int(backoff)} seconds."
)
finally:
sleep(backoff)
def resolve_conflicts(self, master_deltas, slave_deltas):
mc, sc = master_deltas & slave_deltas
if mc or sc:
LOGGER.debug(
f"Detected possible conflicts since last boot. Master: {mc}; Slave {sc}"
)
def move_conflict(path):
conflict_path = self.slave_fs.conflict(path)
self.slave_fs.copy(path, conflict_path)
LOGGER.info(
f"Conflicting file on slave backed up: {path} -> {conflict_path}"
)
for path in [p for p in master_deltas.removed if p in sc]:
move_conflict(path)
for path in [p for p in master_deltas.added if p in sc]:
master_file = self.slave_fs.search(path)
if not master_file:
raise RuntimeError("Master file is unexpectedly missing.")
if not master_file & self.slave_fs.search(path):
# File is different, so slave file is conflict and we copy
# master file over.
move_conflict(path)
for src, dst in master_deltas.moved:
# src file has been moved/removed
slave_src_file = self.slave_fs.search(src)
if slave_src_file and src in sc:
# Conflict master -> slave
move_conflict(src)
# dst file has been created/modified
master_dst_file = self.master_fs.search(dst)
if not master_dst_file:
raise RuntimeError("Master file is unexpectedly missing.")
slave_dst_file = self.slave_fs.search(dst)
if slave_dst_file and dst in sc and not (master_dst_file & slave_dst_file):
# File is different, so slave file is conflict and we copy
# master file over.
move_conflict(dst)
def _save_states(self, signum=None, frame=None):
if signum:
print("")
LOGGER.warn(
f"Received termination signal ({signum}). Shutting down...")
if self._master_state:
self._master_state.save(self._master_state_file)
LOGGER.info("Master FS state saved")
self._slave_state.save(self._slave_state_file)
LOGGER.info("Slave FS state saved")
if signum:
exit(signum)
def move(self, src: str, dst: str):
try:
LOGGER.debug(f"Moving local file {src} to {dst}")
move(self._abs_path(src), self._abs_path(dst))
except FileNotFoundError:
pass
def read(self, path):
try:
return open(self._abs_path(path), "rb")
except (FileNotFoundError, IOError):
LOGGER.error(f"Cannot read file {self._abs_path(path)} from {self}")
return None
def makedirs(self, path):
LOGGER.debug(f"Creating local directory {path}")
os.makedirs(self._abs_path(path), exist_ok=True)
def move_conflict(path):
conflict_path = self.slave_fs.conflict(path)
self.slave_fs.copy(path, conflict_path)
LOGGER.info(
f"Conflicting file on slave backed up: {path} -> {conflict_path}"
)
def start(self):
with ErwinConfiguration() as config:
LOGGER.info("Erwin configuration loaded successfully.")
# Create master and slave FSs
self.master_fs = GoogleDriveFS(**config.get_master_fs_params())
LOGGER.info("Master FS is online.")
LOGGER.debug(f"Created Master FS of type {type(self.master_fs)}")
self.slave_fs = LocalFS(**config.get_slave_fs_params())
LOGGER.info("Slave FS is online.")
LOGGER.debug(f"Created Slave FS of type {type(self.slave_fs)}")
# Load the previous state
prev_master_state, prev_slave_state = config.load_fs_states()
# Register signal handlers
config.register_state_handler(prev_master_state, prev_slave_state)
LOGGER.info("Previous FS states loaded successfully.")
# Compute deltas since last launch
master_deltas = self.master_fs.state - prev_master_state
LOGGER.debug(f"Master deltas since last state save:\n{master_deltas}")
slave_deltas = self.slave_fs.state - prev_slave_state
LOGGER.debug(f"Slave deltas since last state save:\n{slave_deltas}")
self.resolve_conflicts(master_deltas, slave_deltas)
master_deltas.apply(
(self.master_fs, prev_master_state), (self.slave_fs, prev_slave_state)
)
if self.master_fs.state - prev_master_state:
raise RuntimeError("Not all deltas applied correctly to master!")
# At this point we do not expect to have any conflicts left as we
# have resolved them at master before.
new_slave_deltas = self.slave_fs.state - prev_slave_state
LOGGER.debug(f"New deltas:\n{new_slave_deltas}")
new_slave_deltas.apply(
(self.slave_fs, prev_slave_state), (self.master_fs, prev_master_state)
)
# Start the collectors to watch for changes on both FSs.
self._start_collectors(prev_master_state, prev_slave_state)
