def wait(self):
"""Loop waiting on children to die and respawning as necessary."""
LOG.debug(_('Full set of CONF:'))
CONF.log_opt_values(LOG, std_logging.DEBUG)
while True:
self.handle_signal()
self._respawn_children()
if self.sigcaught:
signame = _signo_to_signame(self.sigcaught)
LOG.info(_('Caught %s, stopping children'), signame)
if not _is_sighup(self.sigcaught):
break
for pid in self.children:
os.kill(pid, signal.SIGHUP)
self.running = True
self.sigcaught = None
for pid in self.children:
try:
os.kill(pid, signal.SIGTERM)
except OSError as exc:
if exc.errno != errno.ESRCH:
raise
# Wait for children to die
if self.children:
LOG.info(_('Waiting on %d children to exit'), len(self.children))
while self.children:
self._wait_child()
def _wait_child(self):
try:
# Don't block if no child processes have exited
pid, status = os.waitpid(0, os.WNOHANG)
if not pid:
return None
except OSError as exc:
if exc.errno not in (errno.EINTR, errno.ECHILD):
raise
return None
if os.WIFSIGNALED(status):
sig = os.WTERMSIG(status)
LOG.info(_('Child %(pid)d killed by signal %(sig)d'),
dict(pid=pid, sig=sig))
else:
code = os.WEXITSTATUS(status)
LOG.info(_('Child %(pid)s exited with status %(code)d'),
dict(pid=pid, code=code))
if pid not in self.children:
LOG.warning(_('pid %d not in child list'), pid)
return None
wrap = self.children.pop(pid)
wrap.children.remove(pid)
return wrap
def _wait_for_exit_or_signal(self, ready_callback=None):
status = None
signo = 0
LOG.debug(_('Full set of CONF:'))
CONF.log_opt_values(LOG, std_logging.DEBUG)
try:
if ready_callback:
ready_callback()
super(ServiceLauncher, self).wait()
except SignalExit as exc:
signame = _signo_to_signame(exc.signo)
LOG.info(_('Caught %s, exiting'), signame)
status = exc.code
signo = exc.signo
except SystemExit as exc:
status = exc.code
finally:
self.stop()
if rpc:
try:
rpc.cleanup()
except Exception:
# We're shutting down, so it doesn't matter at this point.
LOG.exception(_('Exception during rpc cleanup.'))
return status, signo
def _start_child(self, wrap):
if len(wrap.forktimes) > wrap.workers:
# Limit ourselves to one process a second (over the period of
# number of workers * 1 second). This will allow workers to
# start up quickly but ensure we don't fork off children that
# die instantly too quickly.
if time.time() - wrap.forktimes[0] < wrap.workers:
LOG.info(_('Forking too fast, sleeping'))
time.sleep(1)
wrap.forktimes.pop(0)
wrap.forktimes.append(time.time())
pid = os.fork()
if pid == 0:
launcher = self._child_process(wrap.service)
while True:
self._child_process_handle_signal()
status, signo = self._child_wait_for_exit_or_signal(launcher)
if not _is_sighup(signo):
break
launcher.restart()
os._exit(status)
LOG.info(_('Started child %d'), pid)
wrap.children.add(pid)
self.children[pid] = wrap
return pid
def inner(*args, **kwargs):
with lock(name, lock_file_prefix, external, lock_path):
LOG.debug(_('Got semaphore / lock "%(function)s"'),
{'function': f.__name__})
return f(*args, **kwargs)
LOG.debug(_('Semaphore / lock released "%(function)s"'),
{'function': f.__name__})
def _get_not_supported_column(col_name_col_instance, column_name):
try:
column = col_name_col_instance[column_name]
except KeyError:
msg = _("Please specify column %s in col_name_col_instance "
"param. It is required because column has unsupported "
"type by sqlite).")
raise ColumnError(msg % column_name)
if not isinstance(column, Column):
msg = _("col_name_col_instance param has wrong type of "
"column instance for column %s It should be instance "
"of sqlalchemy.Column.")
raise ColumnError(msg % column_name)
return column
def inner_func(*args, **kwargs):
last_log_time = 0
last_exc_message = None
exc_count = 0
while True:
try:
return infunc(*args, **kwargs)
except Exception as exc:
this_exc_message = six.u(str(exc))
if this_exc_message == last_exc_message:
exc_count += 1
else:
exc_count = 1
# Do not log any more frequently than once a minute unless
# the exception message changes
cur_time = int(time.time())
if (cur_time - last_log_time > 60 or
this_exc_message != last_exc_message):
logging.exception(
_('Unexpected exception occurred %d time(s)... '
'retrying.') % exc_count)
last_log_time = cur_time
last_exc_message = this_exc_message
exc_count = 0
# This should be a very rare event. In case it isn't, do
# a sleep.
time.sleep(1)
def __exit__(self, exc_type, exc_val, exc_tb):
try:
self.unlock()
self.lockfile.close()
except IOError:
LOG.exception(_("Could not release the acquired lock `%s`"),
self.fname)
def deprecated(self, msg, *args, **kwargs):
stdmsg = _("Deprecated: %s") % msg
if CONF.fatal_deprecations:
self.critical(stdmsg, *args, **kwargs)
raise DeprecatedConfig(msg=stdmsg)
else:
self.warn(stdmsg, *args, **kwargs)
def db_sync(abs_path, version=None, init_version=0):
"""Upgrade or downgrade a database.
Function runs the upgrade() or downgrade() functions in change scripts.
:param abs_path: Absolute path to migrate repository.
:param version: Database will upgrade/downgrade until this version.
If None - database will update to the latest
available version.
:param init_version: Initial database version
"""
if version is not None:
try:
version = int(version)
except ValueError:
raise exception.DbMigrationError(
message=_("version should be an integer"))
current_version = db_version(abs_path, init_version)
repository = _find_migrate_repo(abs_path)
if version is None or version > current_version:
return versioning_api.upgrade(get_engine(), repository, version)
else:
return versioning_api.downgrade(get_engine(), repository,
version)
def initialize_if_enabled():
backdoor_locals = {
'exit': _dont_use_this, # So we don't exit the entire process
'quit': _dont_use_this, # So we don't exit the entire process
'fo': _find_objects,
'pgt': _print_greenthreads,
'pnt': _print_nativethreads,
}
if CONF.backdoor_port is None:
return None
start_port, end_port = _parse_port_range(str(CONF.backdoor_port))
# NOTE(johannes): The standard sys.displayhook will print the value of
# the last expression and set it to __builtin__._, which overwrites
# the __builtin__._ that gettext sets. Let's switch to using pprint
# since it won't interact poorly with gettext, and it's easier to
# read the output too.
def displayhook(val):
if val is not None:
pprint.pprint(val)
sys.displayhook = displayhook
sock = _listen('localhost', start_port, end_port, eventlet.listen)
# In the case of backdoor port being zero, a port number is assigned by
# listen(). In any case, pull the port number out here.
port = sock.getsockname()[1]
LOG.info(_('Eventlet backdoor listening on %(port)s for process %(pid)d') %
{'port': port, 'pid': os.getpid()})
eventlet.spawn_n(eventlet.backdoor.backdoor_server, sock,
locals=backdoor_locals)
return port
def _pipe_watcher(self):
# This will block until the write end is closed when the parent
# dies unexpectedly
self.readpipe.read()
LOG.info(_('Parent process has died unexpectedly, exiting'))
sys.exit(1)
def __exit__(self, exc_type, exc_val, exc_tb):
if exc_type is not None:
logging.error(_('Original exception being dropped: %s'),
traceback.format_exception(self.type_,
self.value,
self.tb))
return False
if self.reraise:
six.reraise(self.type_, self.value, self.tb)
def _ping_listener(dbapi_conn, connection_rec, connection_proxy):
"""Ensures that MySQL connections checked out of the pool are alive.
Borrowed from:
http://groups.google.com/group/sqlalchemy/msg/a4ce563d802c929f
"""
try:
dbapi_conn.cursor().execute("select 1")
except dbapi_conn.OperationalError as ex:
if ex.args[0] in (2006, 2013, 2014, 2045, 2055):
LOG.warn(_("Got mysql server has gone away: %s"), ex)
raise sqla_exc.DisconnectionError("Database server went away")
else:
raise
def _child_wait_for_exit_or_signal(self, launcher):
status = None
signo = 0
# NOTE(johannes): All exceptions are caught to ensure this
# doesn't fallback into the loop spawning children. It would
# be bad for a child to spawn more children.
try:
launcher.wait()
except SignalExit as exc:
signame = _signo_to_signame(exc.signo)
LOG.info(_('Caught %s, exiting'), signame)
status = exc.code
signo = exc.signo
except SystemExit as exc:
status = exc.code
except BaseException:
LOG.exception(_('Unhandled exception'))
status = 2
finally:
launcher.stop()
return status, signo
def _inner():
if initial_delay:
greenthread.sleep(initial_delay)
try:
while self._running:
idle = self.f(*self.args, **self.kw)
if not self._running:
break
if periodic_interval_max is not None:
idle = min(idle, periodic_interval_max)
LOG.debug(_("Dynamic looping call sleeping for %.02f " "seconds"), idle)
greenthread.sleep(idle)
except LoopingCallDone as e:
self.stop()
done.send(e.retvalue)
except Exception:
LOG.exception(_("in dynamic looping call"))
done.send_exception(*sys.exc_info())
return
else:
done.send(True)
def drop_old_duplicate_entries_from_table(migrate_engine, table_name,
use_soft_delete, *uc_column_names):
"""Drop all old rows having the same values for columns in uc_columns.
This method drop (or mark ad `deleted` if use_soft_delete is True) old
duplicate rows form table with name `table_name`.
:param migrate_engine: Sqlalchemy engine
:param table_name: Table with duplicates
:param use_soft_delete: If True - values will be marked as `deleted`,
if False - values will be removed from table
:param uc_column_names: Unique constraint columns
"""
meta = MetaData()
meta.bind = migrate_engine
table = Table(table_name, meta, autoload=True)
columns_for_group_by = [table.c[name] for name in uc_column_names]
columns_for_select = [func.max(table.c.id)]
columns_for_select.extend(columns_for_group_by)
duplicated_rows_select = select(columns_for_select,
group_by=columns_for_group_by,
having=func.count(table.c.id) > 1)
for row in migrate_engine.execute(duplicated_rows_select):
# NOTE(boris-42): Do not remove row that has the biggest ID.
delete_condition = table.c.id != row[0]
is_none = None # workaround for pyflakes
delete_condition &= table.c.deleted_at == is_none
for name in uc_column_names:
delete_condition &= table.c[name] == row[name]
rows_to_delete_select = select([table.c.id]).where(delete_condition)
for row in migrate_engine.execute(rows_to_delete_select).fetchall():
LOG.info(_("Deleting duplicated row with id: %(id)s from table: "
"%(table)s") % dict(id=row[0], table=table_name))
if use_soft_delete:
delete_statement = table.update().\
where(delete_condition).\
values({
'deleted': literal_column('id'),
'updated_at': literal_column('updated_at'),
'deleted_at': timeutils.utcnow()
})
else:
delete_statement = table.delete().where(delete_condition)
migrate_engine.execute(delete_statement)
def _inner():
if initial_delay:
greenthread.sleep(initial_delay)
try:
while self._running:
start = timeutils.utcnow()
self.f(*self.args, **self.kw)
end = timeutils.utcnow()
if not self._running:
break
delay = interval - timeutils.delta_seconds(start, end)
if delay <= 0:
LOG.warn(_("task run outlasted interval by %s sec") % -delay)
greenthread.sleep(delay if delay > 0 else 0)
except LoopingCallDone as e:
self.stop()
done.send(e.retvalue)
except Exception:
LOG.exception(_("in fixed duration looping call"))
done.send_exception(*sys.exc_info())
return
else:
done.send(True)
def _find_facility_from_conf():
facility_names = logging.handlers.SysLogHandler.facility_names
facility = getattr(logging.handlers.SysLogHandler,
CONF.syslog_log_facility,
None)
if facility is None and CONF.syslog_log_facility in facility_names:
facility = facility_names.get(CONF.syslog_log_facility)
if facility is None:
valid_facilities = facility_names.keys()
consts = ['LOG_AUTH', 'LOG_AUTHPRIV', 'LOG_CRON', 'LOG_DAEMON',
'LOG_FTP', 'LOG_KERN', 'LOG_LPR', 'LOG_MAIL', 'LOG_NEWS',
'LOG_AUTH', 'LOG_SYSLOG', 'LOG_USER', 'LOG_UUCP',
'LOG_LOCAL0', 'LOG_LOCAL1', 'LOG_LOCAL2', 'LOG_LOCAL3',
'LOG_LOCAL4', 'LOG_LOCAL5', 'LOG_LOCAL6', 'LOG_LOCAL7']
valid_facilities.extend(consts)
raise TypeError(_('syslog facility must be one of: %s') %
', '.join("'%s'" % fac
for fac in valid_facilities))
return facility
def db_version(abs_path, init_version):
"""Show the current version of the repository.
:param abs_path: Absolute path to migrate repository
:param version: Initial database version
"""
repository = _find_migrate_repo(abs_path)
try:
return versioning_api.db_version(get_engine(), repository)
except versioning_exceptions.DatabaseNotControlledError:
meta = sqlalchemy.MetaData()
engine = get_engine()
meta.reflect(bind=engine)
tables = meta.tables
if len(tables) == 0:
db_version_control(abs_path, init_version)
return versioning_api.db_version(get_engine(), repository)
else:
# Some pre-Essex DB's may not be version controlled.
# Require them to upgrade using Essex first.
raise exception.DbMigrationError(
message=_("Upgrade DB using Essex release first."))
def read_cached_file(filename, force_reload=False):
"""Read from a file if it has been modified.
:param force_reload: Whether to reload the file.
:returns: A tuple with a boolean specifying if the data is fresh
or not.
"""
global _FILE_CACHE
if force_reload and filename in _FILE_CACHE:
del _FILE_CACHE[filename]
reloaded = False
mtime = os.path.getmtime(filename)
cache_info = _FILE_CACHE.setdefault(filename, {})
if not cache_info or mtime > cache_info.get('mtime', 0):
LOG.debug(_("Reloading cached file %s") % filename)
with open(filename) as fap:
cache_info['data'] = fap.read()
cache_info['mtime'] = mtime
reloaded = True
return (reloaded, cache_info['data'])
def _wrap(*args, **kwargs):
try:
return f(*args, **kwargs)
except UnicodeEncodeError:
raise exception.DBInvalidUnicodeParameter()
# note(boris-42): We should catch unique constraint violation and
# wrap it by our own DBDuplicateEntry exception. Unique constraint
# violation is wrapped by IntegrityError.
except sqla_exc.OperationalError as e:
_raise_if_deadlock_error(e, get_engine().name)
# NOTE(comstud): A lot of code is checking for OperationalError
# so let's not wrap it for now.
raise
except sqla_exc.IntegrityError as e:
# note(boris-42): SqlAlchemy doesn't unify errors from different
# DBs so we must do this. Also in some tables (for example
# instance_types) there are more than one unique constraint. This
# means we should get names of columns, which values violate
# unique constraint, from error message.
_raise_if_duplicate_entry_error(e, get_engine().name)
raise exception.DBError(e)
except Exception as e:
LOG.exception(_("DB exception wrapped."))
raise exception.DBError(e)
def _get_default_deleted_value(table):
if isinstance(table.c.id.type, Integer):
return 0
if isinstance(table.c.id.type, String):
return ""
raise ColumnError(_("Unsupported id columns type"))
def create_engine(sql_connection, sqlite_fk=False):
"""Return a new SQLAlchemy engine."""
# NOTE(geekinutah): At this point we could be connecting to the normal
# db handle or the slave db handle. Things like
# _wrap_db_error aren't going to work well if their
# backends don't match. Let's check.
_assert_matching_drivers()
connection_dict = sqlalchemy.engine.url.make_url(sql_connection)
engine_args = {"pool_recycle": CONF.database.idle_timeout, "echo": False, "convert_unicode": True}
# Map our SQL debug level to SQLAlchemy's options
if CONF.database.connection_debug >= 100:
engine_args["echo"] = "debug"
elif CONF.database.connection_debug >= 50:
engine_args["echo"] = True
if "sqlite" in connection_dict.drivername:
if sqlite_fk:
engine_args["listeners"] = [SqliteForeignKeysListener()]
engine_args["poolclass"] = NullPool
if CONF.database.connection == "sqlite://":
engine_args["poolclass"] = StaticPool
engine_args["connect_args"] = {"check_same_thread": False}
else:
if CONF.database.max_pool_size is not None:
engine_args["pool_size"] = CONF.database.max_pool_size
if CONF.database.max_overflow is not None:
engine_args["max_overflow"] = CONF.database.max_overflow
if CONF.database.pool_timeout is not None:
engine_args["pool_timeout"] = CONF.database.pool_timeout
engine = sqlalchemy.create_engine(sql_connection, **engine_args)
sqlalchemy.event.listen(engine, "checkin", _thread_yield)
if "mysql" in connection_dict.drivername:
sqlalchemy.event.listen(engine, "checkout", _ping_listener)
elif "sqlite" in connection_dict.drivername:
if not CONF.sqlite_synchronous:
sqlalchemy.event.listen(engine, "connect", _synchronous_switch_listener)
sqlalchemy.event.listen(engine, "connect", _add_regexp_listener)
if CONF.database.connection_trace and engine.dialect.dbapi.__name__ == "MySQLdb":
_patch_mysqldb_with_stacktrace_comments()
try:
engine.connect()
except sqla_exc.OperationalError as e:
if not _is_db_connection_error(e.args[0]):
raise
remaining = CONF.database.max_retries
if remaining == -1:
remaining = "infinite"
while True:
msg = _("SQL connection failed. %s attempts left.")
LOG.warn(msg % remaining)
if remaining != "infinite":
remaining -= 1
time.sleep(CONF.database.retry_interval)
try:
engine.connect()
break
except sqla_exc.OperationalError as e:
if (remaining != "infinite" and remaining == 0) or not _is_db_connection_error(e.args[0]):
raise
return engine
def paginate_query(query, model, limit, sort_keys, marker=None,
sort_dir=None, sort_dirs=None):
"""Returns a query with sorting / pagination criteria added.
Pagination works by requiring a unique sort_key, specified by sort_keys.
(If sort_keys is not unique, then we risk looping through values.)
We use the last row in the previous page as the 'marker' for pagination.
So we must return values that follow the passed marker in the order.
With a single-valued sort_key, this would be easy: sort_key > X.
With a compound-values sort_key, (k1, k2, k3) we must do this to repeat
the lexicographical ordering:
(k1 > X1) or (k1 == X1 && k2 > X2) or (k1 == X1 && k2 == X2 && k3 > X3)
We also have to cope with different sort_directions.
Typically, the id of the last row is used as the client-facing pagination
marker, then the actual marker object must be fetched from the db and
passed in to us as marker.
:param query: the query object to which we should add paging/sorting
:param model: the ORM model class
:param limit: maximum number of items to return
:param sort_keys: array of attributes by which results should be sorted
:param marker: the last item of the previous page; we returns the next
results after this value.
:param sort_dir: direction in which results should be sorted (asc, desc)
:param sort_dirs: per-column array of sort_dirs, corresponding to sort_keys
:rtype: sqlalchemy.orm.query.Query
:return: The query with sorting/pagination added.
"""
if 'id' not in sort_keys:
# TODO(justinsb): If this ever gives a false-positive, check
# the actual primary key, rather than assuming its id
LOG.warn(_('Id not in sort_keys; is sort_keys unique?'))
assert(not (sort_dir and sort_dirs))
# Default the sort direction to ascending
if sort_dirs is None and sort_dir is None:
sort_dir = 'asc'
# Ensure a per-column sort direction
if sort_dirs is None:
sort_dirs = [sort_dir for _sort_key in sort_keys]
assert(len(sort_dirs) == len(sort_keys))
# Add sorting
for current_sort_key, current_sort_dir in zip(sort_keys, sort_dirs):
try:
sort_dir_func = {
'asc': sqlalchemy.asc,
'desc': sqlalchemy.desc,
}[current_sort_dir]
except KeyError:
raise ValueError(_("Unknown sort direction, "
"must be 'desc' or 'asc'"))
try:
sort_key_attr = getattr(model, current_sort_key)
except AttributeError:
raise InvalidSortKey()
query = query.order_by(sort_dir_func(sort_key_attr))
# Add pagination
if marker is not None:
marker_values = []
for sort_key in sort_keys:
v = getattr(marker, sort_key)
marker_values.append(v)
# Build up an array of sort criteria as in the docstring
criteria_list = []
for i in range(0, len(sort_keys)):
crit_attrs = []
for j in range(0, i):
model_attr = getattr(model, sort_keys[j])
crit_attrs.append((model_attr == marker_values[j]))
model_attr = getattr(model, sort_keys[i])
if sort_dirs[i] == 'desc':
crit_attrs.append((model_attr < marker_values[i]))
else:
crit_attrs.append((model_attr > marker_values[i]))
criteria = sqlalchemy.sql.and_(*crit_attrs)
criteria_list.append(criteria)
f = sqlalchemy.sql.or_(*criteria_list)
query = query.filter(f)
if limit is not None:
query = query.limit(limit)
return query
def lock(name, lock_file_prefix=None, external=False, lock_path=None):
"""Context based lock
This function yields a `threading.Semaphore` instance (if we don't use
eventlet.monkey_patch(), else `semaphore.Semaphore`) unless external is
True, in which case, it'll yield an InterProcessLock instance.
:param lock_file_prefix: The lock_file_prefix argument is used to provide
lock files on disk with a meaningful prefix.
:param external: The external keyword argument denotes whether this lock
should work across multiple processes. This means that if two different
workers both run a a method decorated with @synchronized('mylock',
external=True), only one of them will execute at a time.
:param lock_path: The lock_path keyword argument is used to specify a
special location for external lock files to live. If nothing is set, then
CONF.lock_path is used as a default.
"""
# NOTE(soren): If we ever go natively threaded, this will be racy.
# See http://stackoverflow.com/questions/5390569/dyn
# amically-allocating-and-destroying-mutexes
sem = _semaphores.get(name, threading.Semaphore())
if name not in _semaphores:
# this check is not racy - we're already holding ref locally
# so GC won't remove the item and there was no IO switch
# (only valid in greenthreads)
_semaphores[name] = sem
with sem:
LOG.debug(_('Got semaphore "%(lock)s"'), {'lock': name})
# NOTE(mikal): I know this looks odd
if not hasattr(local.strong_store, 'locks_held'):
local.strong_store.locks_held = []
local.strong_store.locks_held.append(name)
try:
if external and not CONF.disable_process_locking:
LOG.debug(_('Attempting to grab file lock "%(lock)s"'),
{'lock': name})
# We need a copy of lock_path because it is non-local
local_lock_path = lock_path or CONF.lock_path
if not local_lock_path:
raise cfg.RequiredOptError('lock_path')
if not os.path.exists(local_lock_path):
fileutils.ensure_tree(local_lock_path)
LOG.info(_('Created lock path: %s'), local_lock_path)
def add_prefix(name, prefix):
if not prefix:
return name
sep = '' if prefix.endswith('-') else '-'
return '%s%s%s' % (prefix, sep, name)
# NOTE(mikal): the lock name cannot contain directory
# separators
lock_file_name = add_prefix(name.replace(os.sep, '_'),
lock_file_prefix)
lock_file_path = os.path.join(local_lock_path, lock_file_name)
try:
lock = InterProcessLock(lock_file_path)
with lock as lock:
LOG.debug(_('Got file lock "%(lock)s" at %(path)s'),
{'lock': name, 'path': lock_file_path})
yield lock
finally:
LOG.debug(_('Released file lock "%(lock)s" at %(path)s'),
{'lock': name, 'path': lock_file_path})
else:
yield sem
finally:
local.strong_store.locks_held.remove(name)
def launch_service(self, service, workers=1):
wrap = ServiceWrapper(service, workers)
LOG.info(_('Starting %d workers'), wrap.workers)
while self.running and len(wrap.children) < wrap.workers:
self._start_child(wrap)
