another attempt...

This commit is contained in:
Fredrik Eriksson 2019-06-22 14:59:26 +02:00
parent 7833254c33
commit 8bd3148a66
No known key found for this signature in database
GPG Key ID: 8825C73A0FD1502A

View File

@ -74,7 +74,7 @@ def restart_services():
# perform a second check to remove potential false positives # perform a second check to remove potential false positives
service_procs = set() service_procs = set()
retest_procs = {} retest_procs = set()
for proc in check_procs: for proc in check_procs:
files = _get_deleted_open_files(proc) files = _get_deleted_open_files(proc)
if not files: if not files:
@ -82,63 +82,32 @@ def restart_services():
continue continue
try: try:
exe = proc.exe() exe = proc.exe()
parents = proc.parents()
except (psutil.NoSuchProcess, psutil.ZombieProcess, psutil.AccessDenied): except (psutil.NoSuchProcess, psutil.ZombieProcess, psutil.AccessDenied):
# either of the above exceptions means the process has quit # either of the above exceptions means the process has quit
continue continue
parent = _get_top_parent(proc)
log.debug('will attempt to restart parent of {}'.format(proc))
if len(parents) < 2:
log.debug('{} is its own top parent'.format(proc))
parent = proc
else:
log.debug('{} has top parent {}'.format(proc, parents[-2]))
parent = parents[-2]
service_procs.add(parent) service_procs.add(parent)
retest_procs[proc] = parent retest_procs.add(proc)
processes = {} processes = {}
services = {} services = {}
for proc in service_procs: for proc in service_procs:
try: try:
service_exe = proc.exe() service_exe = proc.exe()
proc_name = proc.name()
except (psutil.NoSuchProcess, psutil.ZombieProcess, psutil.AccessDenied): except (psutil.NoSuchProcess, psutil.ZombieProcess, psutil.AccessDenied):
log.debug('{} died before it could be restarted'.format(proc)) log.debug('{} died before it could be restarted'.format(proc))
continue continue
service_name = _get_service_from_proc(proc)
if proc_name in services:
processes[services[proc_name]].append(proc)
# we have already checked a process with this name
continue
service_name = conf.get('processes', proc_name, fallback=None)
if service_name == '':
log.debug('Ignoring process {}'.format(proc))
del retest_procs[proc]
continue
if not service_name: if not service_name:
# if the exe file has been deleted since started, service_exe will be empty log.debug('no service for process {}'.format(proc))
# and we'll have to guess
if not service_exe:
log.debug('Could not get full path to executable for process {}, will attempt to guess'.format(proc))
service_exe = get_exe_file(service_name)
if not service_exe:
log.error('Failed to find executable for process {}'.format(proc))
continue
try:
service_name = platform.identify_service_from_bin(service_exe)
except sau.errors.UnknownServiceError:
log.warning('Could not find service for process {}'.format(proc))
continue continue
services[proc_name] = service_name services[proc_name] = service_name
processes[service_name] = [proc] processes[service_name] = [proc]
for service_all in [x for x in services.values() if x]: for service in set([x for x in services.values() if x]):
for service in service_all.split():
policy = _get_service_restart_policy(service) policy = _get_service_restart_policy(service)
if policy == 'ignore': if policy == 'ignore':
log.info('Service "{}" ignored by configuration'.format(service)) log.info('Service "{}" ignored by configuration'.format(service))
@ -152,20 +121,32 @@ def restart_services():
platform.restart_service(service) platform.restart_service(service)
recommend_restart = False recommend_restart = False
for proc, parent in retest_procs.items(): tested_parents = set()
for proc in retest_procs:
try: try:
name = proc.name() name = proc.name()
except (psutil.NoSuchProcess, psutil.ZombieProcess, psutil.AccessDenied): except (psutil.NoSuchProcess, psutil.ZombieProcess, psutil.AccessDenied):
log.debug('{} was successfully killed'.format(proc)) log.debug('{} was successfully killed'.format(proc))
continue continue
parent = _get_top_parent(proc)
if parent in tested_parents:
log.debug('{} belongs to already tested parent {}'.format(proc, parent))
continue
if _get_deleted_open_files(proc): if _get_deleted_open_files(proc):
if name in services and not services[name]: tested_parents.add(parent)
service = _get_service_from_proc(parent)
if not service:
log.warning('could not re-check process {} - failed to identify service'.format(proc))
recommend_restart = True
continue
if service in services and not services[service]:
log.warning('{} (parent {}) does not belong to a service and could not be restarted'.format(proc, parent)) log.warning('{} (parent {}) does not belong to a service and could not be restarted'.format(proc, parent))
recommend_restart = True recommend_restart = True
continue continue
elif name in services: elif service in services:
policy = _get_service_restart_policy(services[parent.name()]) policy = _get_service_restart_policy(service)
if policy in ('ignore', 'warn'): if policy in ('ignore', 'warn'):
continue continue
log.warning('{} (parent {}) still has deleted files open'.format(proc, parent)) log.warning('{} (parent {}) still has deleted files open'.format(proc, parent))
@ -186,3 +167,54 @@ def _get_service_restart_policy(service):
return default_policy.lower() return default_policy.lower()
log.warning('default service policy {} is invalid'.format(default_policy)) log.warning('default service policy {} is invalid'.format(default_policy))
return 'warn' return 'warn'
def _get_service_from_proc(proc):
conf = sau.config
platform = sau.platforms.get_platform()
proc = _get_top_parent(proc)
log = logging.getLogger(sau.LOGNAME)
try:
proc_name = proc.name()
service_exe = proc.exe()
except (psutil.NoSuchProcess, psutil.ZombieProcess, psutil.AccessDenied):
log.debug('{} died'.format(proc))
return None
service_name = conf.get('processes', proc_name, fallback=None)
if service_name == '':
log.debug('Ignoring process {}'.format(proc))
return None
if not service_name:
# if the exe file has been deleted since started, service_exe will be empty
# and we'll have to guess
if not service_exe:
log.debug('Could not get full path to executable for process {}, will attempt to guess'.format(proc))
service_exe = get_exe_file(service_name)
if not service_exe:
log.error('Failed to find executable for process {}'.format(proc))
return None
try:
service_name = platform.identify_service_from_bin(service_exe)
except sau.errors.UnknownServiceError:
log.warning('Could not find service for process {}'.format(proc))
return None
return service_name
def _get_top_parent(proc):
log = logging.getLogger(sau.LOGNAME)
try:
parents = proc.parents()
except (psutil.NoSuchProcess, psutil.ZombieProcess, psutil.AccessDenied):
# either of the above exceptions means the process has quit
return None
if len(parents) < 2:
log.debug('{} is its own top parent'.format(proc))
parent = proc
else:
log.debug('{} has top parent {}'.format(proc, parents[-2]))
parent = parents[-2]
return parent