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(_LI('Child %(pid)d killed by signal %(sig)d'),
dict(pid=pid, sig=sig))
else:
code = os.WEXITSTATUS(status)
LOG.info(_LI('Child %(pid)s exited with status %(code)d'),
dict(pid=pid, code=code))
if pid not in self.children:
LOG.warning(_LW('pid %d not in child list'), pid)
return None
wrap = self.children.pop(pid)
wrap.children.remove(pid)
return wrap
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)
try:
while True:
self.handle_signal()
self._respawn_children()
if self.sigcaught:
signame = _signo_to_signame(self.sigcaught)
LOG.info(_LI('Caught %s, stopping children'), signame)
if not _is_sighup_and_daemon(self.sigcaught):
break
for pid in self.children:
os.kill(pid, signal.SIGHUP)
self.running = True
self.sigcaught = None
except eventlet.greenlet.GreenletExit:
LOG.info(_LI("Wait called after thread killed. Cleaning up."))
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(_LI('Waiting on %d children to exit'), len(self.children))
while self.children:
self._wait_child()
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(_LI('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_and_daemon(signo):
break
launcher.restart()
os._exit(status)
LOG.info(_LI('Started child %d'), pid)
wrap.children.add(pid)
self.children[pid] = wrap
return pid
def _set_session_sql_mode(dbapi_con, connection_rec,
connection_proxy, sql_mode=None):
"""Set the sql_mode session variable.
MySQL supports several server modes. The default is None, but sessions
may choose to enable server modes like TRADITIONAL, ANSI,
several STRICT_* modes and others.
Note: passing in '' (empty string) for sql_mode clears
the SQL mode for the session, overriding a potentially set
server default. Passing in None (the default) makes this
a no-op, meaning if a server-side SQL mode is set, it still applies.
"""
cursor = dbapi_con.cursor()
if sql_mode is not None:
cursor.execute("SET SESSION sql_mode = %s", [sql_mode])
# Check against the real effective SQL mode. Even when unset by
# our own config, the server may still be operating in a specific
# SQL mode as set by the server configuration
cursor.execute("SHOW VARIABLES LIKE 'sql_mode'")
row = cursor.fetchone()
if row is None:
LOG.warning(_LW('Unable to detect effective SQL mode'))
return
realmode = row[1]
LOG.info(_LI('MySQL server mode set to %s') % realmode)
# 'TRADITIONAL' mode enables several other modes, so
# we need a substring match here
if not ('TRADITIONAL' in realmode.upper() or
'STRICT_ALL_TABLES' in realmode.upper()):
LOG.warning(_LW("MySQL SQL mode is '%s', "
"consider enabling TRADITIONAL or STRICT_ALL_TABLES")
% realmode)
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(_LI('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(_LE('Exception during rpc cleanup.'))
return status, signo
def acquire(self):
basedir = os.path.dirname(self.fname)
if not os.path.exists(basedir):
fileutils.ensure_tree(basedir)
LOG.info(_LI('Created lock path: %s'), basedir)
self.lockfile = open(self.fname, 'w')
while True:
try:
# Using non-blocking locks since green threads are not
# patched to deal with blocking locking calls.
# Also upon reading the MSDN docs for locking(), it seems
# to have a laughable 10 attempts "blocking" mechanism.
self.trylock()
LOG.debug('Got file lock "%s"', self.fname)
return True
except IOError as e:
if e.errno in (errno.EACCES, errno.EAGAIN):
# external locks synchronise things like iptables
# updates - give it some time to prevent busy spinning
time.sleep(0.01)
else:
raise threading.ThreadError(_("Unable to acquire lock on"
" `%(filename)s` due to"
" %(exception)s") %
{
'filename': self.fname,
'exception': e,
})
def acquire(self):
basedir = os.path.dirname(self.fname)
if not os.path.exists(basedir):
fileutils.ensure_tree(basedir)
LOG.info(_LI('Created lock path: %s'), basedir)
self.lockfile = open(self.fname, 'w')
while True:
try:
# Using non-blocking locks since green threads are not
# patched to deal with blocking locking calls.
# Also upon reading the MSDN docs for locking(), it seems
# to have a laughable 10 attempts "blocking" mechanism.
self.trylock()
LOG.debug('Got file lock "%s"', self.fname)
return True
except IOError as e:
if e.errno in (errno.EACCES, errno.EAGAIN):
# external locks synchronise things like iptables
# updates - give it some time to prevent busy spinning
time.sleep(0.01)
else:
raise threading.ThreadError(
_("Unable to acquire lock on"
" `%(filename)s` due to"
" %(exception)s") % {
'filename': self.fname,
'exception': e,
})
def _pipe_watcher(self):
# This will block until the write end is closed when the parent
# dies unexpectedly
self.readpipe.read()
LOG.info(_LI('Parent process has died unexpectedly, exiting'))
sys.exit(1)
def remove_external_lock_file(name, lock_file_prefix=None):
"""Remove a external lock file when it's not used anymore
This will be helpful when we have a lot of lock files
"""
with internal_lock(name):
lock_file_path = _get_lock_path(name, lock_file_prefix)
try:
os.remove(lock_file_path)
except OSError:
LOG.info(_LI('Failed to remove file %(file)s'),
{'file': lock_file_path})
def remove_external_lock_file(name, lock_file_prefix=None):
"""Remove a external lock file when it's not used anymore
This will be helpful when we have a lot of lock files
"""
with internal_lock(name):
lock_file_path = _get_lock_path(name, lock_file_prefix)
try:
os.remove(lock_file_path)
except OSError:
LOG.info(_LI('Failed to remove file %(file)s'),
{'file': lock_file_path})
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(
_LI("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 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(_LI("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 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(
_LI('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 _child_wait_for_exit_or_signal(self, launcher):
status = 0
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(_LI('Caught %s, exiting'), signame)
status = exc.code
signo = exc.signo
except SystemExit as exc:
status = exc.code
except BaseException:
LOG.exception(_LE('Unhandled exception'))
status = 2
finally:
launcher.stop()
return status, signo
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(
_LI('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 launch_service(self, service, workers=1):
wrap = ServiceWrapper(service, workers)
LOG.info(_LI('Starting %d workers'), wrap.workers)
while self.running and len(wrap.children) < wrap.workers:
self._start_child(wrap)