def _checkPromises(self, computer_partition):
self.logger.info("Checking promises...")
instance_path = os.path.join(self.instance_root, computer_partition.getId())
uid, gid = None, None
stat_info = os.stat(instance_path)
# stat sys call to get statistics informations
uid = stat_info.st_uid
gid = stat_info.st_gid
promise_present = False
# Get the list of promises
promise_dir = os.path.join(instance_path, "etc", "promise")
if os.path.exists(promise_dir) and os.path.isdir(promise_dir):
cwd = instance_path
promises_list = os.listdir(promise_dir)
# Check whether every promise is kept
for promise in promises_list:
promise_present = True
command = [os.path.join(promise_dir, promise)]
promise = os.path.basename(command[0])
self.logger.info("Checking promise %r.", promise)
kw = dict(stdout=subprocess.PIPE, stderr=subprocess.PIPE)
process_handler = SlapPopen(command, preexec_fn=lambda: dropPrivileges(uid, gid), cwd=cwd, env={}, **kw)
time.sleep(self.promise_timeout)
if process_handler.poll() is None:
process_handler.terminate()
raise Slapgrid.PromiseError("The promise %r timed out" % promise)
elif process_handler.poll() != 0:
stderr = process_handler.communicate()[1]
if stderr is None:
stderr = "No error output from %r." % promise
else:
stderr = "Promise %r:" % promise + stderr
raise Slapgrid.PromiseError(stderr)
if not promise_present:
self.logger.info("No promise.")
def agregateAndSendUsage(self):
"""Will agregate usage from each Computer Partition.
"""
slap_computer_usage = self.slap.registerComputer(self.computer_id)
computer_partition_usage_list = []
logger = logging.getLogger('UsageReporting')
logger.info("Aggregating and sending usage reports...")
#We retrieve XSD models
try:
computer_consumption_model = \
pkg_resources.resource_string(
'slapos.slap',
'doc/computer_consumption.xsd')
except IOError:
computer_consumption_model = \
pkg_resources.resource_string(
__name__,
'../../../../slapos/slap/doc/computer_consumption.xsd')
try:
partition_consumption_model = \
pkg_resources.resource_string(
'slapos.slap',
'doc/partition_consumption.xsd')
except IOError:
partition_consumption_model = \
pkg_resources.resource_string(
__name__,
'../../../../slapos/slap/doc/partition_consumption.xsd')
clean_run = True
#We loop on the different computer partitions
for computer_partition in slap_computer_usage.getComputerPartitionList():
computer_partition_id = computer_partition.getId()
#We want execute all the script in the report folder
instance_path = os.path.join(self.instance_root,
computer_partition.getId())
report_path = os.path.join(instance_path, 'etc', 'report')
if os.path.isdir(report_path):
script_list_to_run = os.listdir(report_path)
else:
script_list_to_run = []
#We now generate the pseudorandom name for the xml file
# and we add it in the invocation_list
f = tempfile.NamedTemporaryFile()
name_xml = '%s.%s' % ('slapreport', os.path.basename(f.name))
path_to_slapreport = os.path.join(instance_path, 'var', 'xml_report',
name_xml)
failed_script_list = []
for script in script_list_to_run:
invocation_list = []
invocation_list.append(os.path.join(instance_path, 'etc', 'report',
script))
#We add the xml_file name in the invocation_list
#f = tempfile.NamedTemporaryFile()
#name_xml = '%s.%s' % ('slapreport', os.path.basename(f.name))
#path_to_slapreport = os.path.join(instance_path, 'var', name_xml)
invocation_list.append(path_to_slapreport)
#Dropping privileges
uid, gid = None, None
stat_info = os.stat(instance_path)
#stat sys call to get statistics informations
uid = stat_info.st_uid
gid = stat_info.st_gid
kw = dict()
if not self.console:
kw.update(stdout=subprocess.PIPE, stderr=subprocess.STDOUT)
process_handler = SlapPopen(invocation_list,
preexec_fn=lambda: dropPrivileges(uid, gid),
cwd=os.path.join(instance_path, 'etc', 'report'),
env=None, **kw)
result = process_handler.communicate()[0]
if self.console:
result = 'Please consult messages above'
if process_handler.returncode is None:
process_handler.kill()
if process_handler.returncode != 0:
clean_run = False
failed_script_list.append("Script %r failed with %s." % (script, result))
logger.warning("Failed to run %r, the result was. \n%s" %
(invocation_list, result))
if len(failed_script_list):
computer_partition.error('\n'.join(failed_script_list))
#Now we loop through the different computer partitions to ggetId()et reports
report_usage_issue_cp_list = []
for computer_partition in slap_computer_usage.getComputerPartitionList():
filename_delete_list = []
computer_partition_id = computer_partition.getId()
instance_path = os.path.join(self.instance_root, computer_partition_id)
dir_reports = os.path.join(instance_path, 'var', 'xml_report')
#The directory xml_report contain a number of files equal
#to the number of software instance running inside the same partition
if os.path.isdir(dir_reports):
filename_list = os.listdir(dir_reports)
else:
filename_list = []
#logger.debug('name List %s' % filename_list)
usage = ''
for filename in filename_list:
file_path = os.path.join(dir_reports, filename)
if os.path.exists(file_path):
usage_file = open(file_path, 'r')
usage = usage_file.read()
usage_file.close()
#We check the validity of xml content of each reports
if not self.validateXML(usage, partition_consumption_model):
logger.info('WARNING: The XML file %s generated by slapreport is not valid - ' \
'This report is left as is at %s where you can inspect what went wrong ' % (filename, dir_reports))
#Warn the SlapOS Master that a partition generates corrupted xml report
else:
computer_partition_usage = self.slap.registerComputerPartition(
self.computer_id, computer_partition_id)
computer_partition_usage.setUsage(usage)
computer_partition_usage_list.append(computer_partition_usage)
filename_delete_list.append(filename)
else:
logger.debug("Usage report %r not found, ignored" % file_path)
#last_push_date = self.computer.getLastUsagePush()
#periodicity_timedelta = datetime.timedelta(
# self.usage_report_periodicity)
#if periodicity_timedelta + last_push_date < datetime.datetime.today():
# Pushes informations, if any
#After sending the aggregated file we remove all the valid xml reports
for filename in filename_delete_list:
os.remove(os.path.join(dir_reports, filename))
for computer_partition_usage in computer_partition_usage_list:
logger.info('computer_partition_usage_list : %s - %s' % \
(computer_partition_usage.usage, computer_partition_usage.getId()))
#If there is, at least, one report
if computer_partition_usage_list != []:
try:
#We generate the final XML report with asXML method
computer_consumption = self.asXML(computer_partition_usage_list)
logger.info('Final xml report : %s' % computer_consumption)
#We test the XML report before sending it
if self.validateXML(computer_consumption, computer_consumption_model):
logger.info('XML file generated by asXML is valid')
slap_computer_usage.reportUsage(computer_consumption)
else:
logger.info('XML file generated by asXML is not valid !')
raise 'XML file generated by asXML is not valid !'
except Exception:
computer_partition_id = computer_partition.getId()
exception = traceback.format_exc()
issue = "Cannot report usage for %r: %s" % (computer_partition_id,
exception)
logger.info(issue)
computer_partition.error(issue)
report_usage_issue_cp_list.append(computer_partition_id)
for computer_partition in slap_computer_usage.getComputerPartitionList():
computer_partition_id = computer_partition.getId()
try:
software_url = computer_partition.getSoftwareRelease().getURI()
except NotFoundError:
software_url = None
software_path = os.path.join(self.software_root,
getSoftwareUrlHash(software_url))
local_partition = Partition(
software_path=software_path,
instance_path=os.path.join(self.instance_root,
computer_partition.getId()),
supervisord_partition_configuration_path=os.path.join(
self.supervisord_configuration_directory, '%s.conf' %
computer_partition_id),
supervisord_socket=self.supervisord_socket,
computer_partition=computer_partition,
computer_id=self.computer_id,
partition_id=computer_partition_id,
server_url=self.master_url,
software_release_url=software_url,
certificate_repository_path=self.certificate_repository_path,
console=self.console, buildout=self.buildout
)
if computer_partition.getState() == "destroyed":
try:
local_partition.stop()
try:
computer_partition.stopped()
except (SystemExit, KeyboardInterrupt):
exception = traceback.format_exc()
computer_partition.error(exception)
raise
except Exception:
pass
except (SystemExit, KeyboardInterrupt):
exception = traceback.format_exc()
computer_partition.error(exception)
raise
except Exception:
clean_run = False
exception = traceback.format_exc()
computer_partition.error(exception)
logger.error(exception)
if computer_partition.getId() in report_usage_issue_cp_list:
logger.info('Ignoring destruction of %r, as not report usage was '
'sent' % computer_partition.getId())
continue
local_partition.destroy()
try:
computer_partition.destroyed()
except slap.NotFoundError:
logger.debug('Ignored slap error while trying to inform about '
'destroying not fully configured Computer Partition %r' %
computer_partition.getId())
logger.info("Finished usage reports...")
return clean_run
def processComputerPartitionList(self):
"""Will start supervisord and process each Computer Partition.
"""
logger = logging.getLogger('ComputerPartitionProcessing')
logger.info("Processing computer partitions...")
# Prepares environment
self.checkEnvironmentAndCreateStructure()
self._launchSupervisord()
# Process Computer Partitions
clean_run = True
for computer_partition in self.getComputerPartitionList():
computer_partition_id = computer_partition.getId()
try:
software_url = computer_partition.getSoftwareRelease().getURI()
except NotFoundError:
software_url = None
software_path = os.path.join(self.software_root,
getSoftwareUrlHash(software_url))
local_partition = Partition(
software_path=software_path,
instance_path=os.path.join(self.instance_root,
computer_partition.getId()),
supervisord_partition_configuration_path=os.path.join(
self.supervisord_configuration_directory, '%s.conf' %
computer_partition_id),
supervisord_socket=self.supervisord_socket,
computer_partition=computer_partition,
computer_id=self.computer_id,
partition_id=computer_partition_id,
server_url=self.master_url,
software_release_url=software_url,
certificate_repository_path=self.certificate_repository_path,
console=self.console, buildout=self.buildout)
# There are no conditions to try to instanciate partition
try:
computer_partition_state = computer_partition.getState()
if computer_partition_state == "started":
local_partition.install()
computer_partition.available()
local_partition.start()
computer_partition.started()
elif computer_partition_state == "stopped":
local_partition.install()
computer_partition.available()
local_partition.stop()
computer_partition.stopped()
elif computer_partition_state == "destroyed":
# Stop, but safely
try:
local_partition.stop()
try:
computer_partition.stopped()
except (SystemExit, KeyboardInterrupt):
exception = traceback.format_exc()
computer_partition.error(exception)
raise
except Exception:
pass
except (SystemExit, KeyboardInterrupt):
exception = traceback.format_exc()
computer_partition.error(exception)
raise
except Exception:
clean_run = False
exception = traceback.format_exc()
logger.error(exception)
computer_partition.error(exception)
else:
error_string = "Computer Partition %r has unsupported state: %s" % \
(computer_partition_id, computer_partition_state)
computer_partition.error(error_string)
raise NotImplementedError(error_string)
except (SystemExit, KeyboardInterrupt):
exception = traceback.format_exc()
computer_partition.error(exception)
raise
except Exception:
clean_run = False
exception = traceback.format_exc()
logger.error(exception)
computer_partition.error(exception)
# Promises
instance_path = os.path.join(self.instance_root,
computer_partition.getId())
uid, gid = None, None
stat_info = os.stat(instance_path)
#stat sys call to get statistics informations
uid = stat_info.st_uid
gid = stat_info.st_gid
# Get the list of promises
promise_dir = os.path.join(instance_path, 'etc', 'promise')
if os.path.exists(promise_dir) and os.path.isdir(promise_dir):
cwd = instance_path
promises_list = os.listdir(promise_dir)
# Check whether every promise is kept
for promise in promises_list:
command = os.path.join(promise_dir, promise)
kw = dict()
if not self.console:
kw.update(stdout=subprocess.PIPE, stderr=subprocess.PIPE)
process_handler = SlapPopen(command,
preexec_fn=lambda: dropPrivileges(uid, gid),
cwd=cwd,
env=None, **kw)
time.sleep(self.promise_timeout)
promise = os.path.basename(command)
if process_handler.poll() is None:
process_handler.kill()
computer_partition.error("The promise %r timed out" % promise)
clean_run = False
elif process_handler.poll() != 0:
stderr = process_handler.communicate()[1]
if stderr is None:
stderr = 'No error output from %r.' % promise
computer_partition.error(stderr)
clean_run = False
logger.info("Finished computer partitions...")
return clean_run
trynum += 1
time.sleep(2 * trynum)
else:
log_message = 'Unknown supervisord state %r. Will try to start.' % status
logger.warning(log_message)
break
logger.info("Launching supervisord with clean environment.")
# Extract python binary to prevent shebang size limit
invocation_list = ["supervisord", '-c']
invocation_list.append("import sys ; sys.path=" + str(sys.path) + " ; import "
"supervisor.supervisord ; sys.argv[1:1]=['-c','" +
configuration_file +
"'] ; supervisor.supervisord.main()")
supervisord_popen = SlapPopen(invocation_list,
env={},
executable=sys.executable, stdout=subprocess.PIPE, stderr=subprocess.STDOUT)
result = supervisord_popen.communicate()[0]
if supervisord_popen.returncode == 0:
log_message = 'Supervisord command invoked with: %s' % result
logger.info(log_message)
try:
default_timeout = socketlib.getdefaulttimeout()
current_timeout = 1
trynum = 1
while trynum < 6:
try:
socketlib.setdefaulttimeout(current_timeout)
status = supervisor.getState()
if status['statename'] == 'RUNNING' and status['statecode'] == 1:
return