Commit 9c655172 authored by JINMEI Tatuya's avatar JINMEI Tatuya
Browse files

[1858] stop sending hopeless signal when it once fails with EPERM.

parent 1e0c106c
......@@ -305,3 +305,27 @@ the configuration manager to start up. The total length of time Boss
will wait for the configuration manager before reporting an error is
set with the command line --wait switch, which has a default value of
ten seconds.
% BIND10_SEND_SIGNAL_FAIL sending %1 to %2 (PID %3) failed: %4
The boss module sent a single (either SIGTERM or SIGKILL) to a process,
but it failed due to some system level error. There are two major cases:
the target process has already terminated but the boss module had sent
the signal before it noticed the termination. In this case an error
message should indicate something like "no such process". This can be
safely ignored. The other case is that the boss module doesn't have
the privilege to send a signal to the process. It can typically
happen when the boss module started as a privileged process, spawned a
subprocess, and then dropped the privilege. It includes the case for
the socket creator when the boss process runs with the -u command line
option. In this case, the boss module simply gives up to terminate
the process explicitly because it's unlikely to succeed by keeping
sending the signal. Although the socket creator is implemented so
that it will terminate automatically when the boss process exits
(and that should be the case for any other future process running with
a higher privilege), but it's recommended to check if there's any
remaining BIND 10 process if this message is logged. For all other
cases, the boss module will keep sending the signal until it confirms
all child processes terminate. Although unlikely, this could prevent
the boss module from exiting, just keeping sending the signals. So,
again, it's advisable to check if it really terminates when this
message is logged.
......@@ -712,14 +712,22 @@ class BoB:
'''
logmsg = BIND10_SEND_SIGKILL if forceful else BIND10_SEND_SIGTERM
for component in self.components.values():
# We need to make a copy of values as the components may be modified
# in the loop.
for component in list(self.components.values()):
logger.info(logmsg, component.name(), component.pid())
try:
component.kill(forceful)
except OSError:
# ignore these (usually ESRCH because the child
# finally exited)
pass
except OSError as ex:
# If kill() failed due to EPERM, it doesn't make sense to
# keep trying, so we just log the fact and forget that
# component. Ignore other OSErrors (usually ESRCH because
# the child finally exited)
signame = "SIGKILL" if forceful else "SIGTERM"
logger.info(BIND10_SEND_SIGNAL_FAIL, signame,
component.name(), component.pid(), ex)
if ex.errno == errno.EPERM:
del self.components[component.pid()]
def _get_process_exit_status(self):
return os.waitpid(-1, os.WNOHANG)
......
......@@ -1195,7 +1195,15 @@ class TestBossComponents(unittest.TestCase):
(anyway it is not told so). It does not die if it is killed
the first time. It dies only when killed forcefully.
"""
def __init__(self):
# number of kill() calls, preventing infinite loop.
self.__call_count = 0
def kill(self, forceful=False):
self.__call_count += 1
if self.__call_count > 2:
raise Exception('Too many calls to ImmortalComponent.kill')
killed.append(forceful)
if ex_on_kill is not None:
# If exception is given by the test, raise it here.
......@@ -1247,10 +1255,17 @@ class TestBossComponents(unittest.TestCase):
self.__real_test_kill()
def test_kill_fail(self):
"""Test cases where kill() results in an exception due to OS error."""
"""Test cases where kill() results in an exception due to OS error.
The behavior should be different for EPERM, so we test two cases.
"""
ex = OSError()
ex.errno = errno.ESRCH
ex.errno, ex.strerror = errno.ESRCH, 'No such process'
self.__real_test_kill(ex_on_kill=ex)
ex.errno, ex.strerror = errno.EPERM, 'Operation not permitted'
self.__real_test_kill(ex_on_kill=ex)
def test_nokill(self):
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment