Usage¶

When using service from oslo-incubator¶

from foo.openstack.common import service

launcher = service.launch(service, workers=2)

When using oslo.service¶

from oslo_config import cfg
from oslo_service import service

CONF = cfg.CONF
launcher = service.launch(CONF, service, workers=2)

Using oslo.service with oslo-config-generator¶

The oslo.service provides several entry points to generate configuration files.

• oslo.service.service

  The options from the service and eventlet_backdoor modules for the [DEFAULT] section.

• oslo.service.periodic_task

  The options from the periodic_task module for the [DEFAULT] section.

• oslo.service.sslutils

  The options from the sslutils module for the ssl section.

• oslo.service.wsgi

  The options from the wsgi module for the [DEFAULT] section.

ATTENTION: The library doesn’t provide an oslo.service entry point.

$ oslo-config-generator --namespace oslo.service.service \
--namespace oslo.service.periodic_task \
--namespace oslo.service.sslutils

Launchers¶

The oslo_service.service module provides two launchers for running services:

• oslo_service.service.ServiceLauncher - used for running one or more service in a parent process.

• oslo_service.service.ProcessLauncher - forks a given number of workers in which service(s) are then started.

It is possible to initialize whatever launcher is needed and then launch a service using it.

from oslo_config import cfg
from oslo_service import service

CONF = cfg.CONF


service_launcher = service.ServiceLauncher(CONF)
service_launcher.launch_service(service.Service())

process_launcher = service.ProcessLauncher(CONF, wait_interval=1.0)
process_launcher.launch_service(service.Service(), workers=2)

Or one can simply call oslo_service.service.launch() which will automatically pick an appropriate launcher based on a number of workers that are passed to it (ServiceLauncher if workers=1 or None and ProcessLauncher in other case).

from oslo_config import cfg
from oslo_service import service

CONF = cfg.CONF

launcher = service.launch(CONF, service.Service(), workers=3)

Signal handling¶

oslo_service.service provides handlers for such signals as SIGTERM, SIGINT, and SIGHUP.

SIGTERM is used for graceful termination of services. This can allow a server to wait for all clients to close connections while rejecting new incoming requests. Config option graceful_shutdown_timeout specifies how many seconds after receiving a SIGTERM signal a server should continue to run, handling the existing connections. Setting graceful_shutdown_timeout to zero means that the server will wait indefinitely until all remaining requests have been fully served.

To force instantaneous termination the SIGINT signal must be sent.

The behavior on receiving SIGHUP varies based on how the service is configured. If the launcher uses restart_method='reload' (the default), then the service will reload its configuration and any threads will be completely restarted. If restart_method='mutate' is used, then only the configuration options marked as mutable will be reloaded and the service threads will not be restarted.

See oslo_service.service.Launcher for more details on the restart_method parameter.

Below is an example of a service with a reset method that allows reloading logging options by sending a SIGHUP.

from oslo_config import cfg
from oslo_log import log as logging
from oslo_service import service

CONF = cfg.CONF

LOG = logging.getLogger(__name__)

class FooService(service.ServiceBase):

    def start(self):
        pass

    def wait(self):
        pass

    def stop(self):
        pass

    def reset(self):
        logging.setup(cfg.CONF, 'foo')

Profiling¶

Processes spawned through oslo_service.service can be profiled (function calltrace) through the eventlet_backdoor module. The service must be configured with the backdoor_port option to enable its workers to listen on TCP ports. The user can then send the prof() command to capture the worker process’s function calltrace.

1. To start profiling send the prof() command on the process’s listening port

2. To stop profiling and capture pstat calltrace to a file, send the prof() command with a file basename as an argument (prof(basename)) to the worker process’s listening port. A stats file (in pstat format) will be generated in the temp directory with the user-provided basename with a .prof suffix .

For example, to profile a neutron server process (which is listening on port 8002 configured through the backdoor_port option):

$ echo "prof()" | nc localhost 8002
$ neutron net-create n1; neutron port-create --name p1 n1;
$ neutron port-delete p1; neutron port-delete p1
$ echo "prof('neutron')" | nc localhost 8002

This will generate a stats file in /tmp/neutron.prof. Stats can be printed from the trace file as follows:

import pstats

stats = pstats.Stats('/tmp/neutron.prof')
stats.print_stats()