mirror of
https://github.com/ceph/ceph
synced 2024-12-21 02:42:48 +00:00
6fe1deadbf
The ceph daemons support being killed at a specific code point with a config option. In some cases, we want to test a kill point only once for a given daemon run (such as replay that only occurs during daemon startup). This task allows running a script or executable and (when the script sends a command to the task) restarting it with a temporary config that has the appropriate kill point set. Once the daemon asserts and gets restarted, the original config is used. Adds a specific restart_with_args() method to the DaemonState in the ceph task. Right now this task follows the workunit task closely, but uses stdout/stdin to specify when to restart a daemon. Signed-off-by: Sam Lang <sam.lang@inktank.com>
282 lines
8.3 KiB
Python
282 lines
8.3 KiB
Python
from cStringIO import StringIO
|
|
|
|
import gevent
|
|
import gevent.event
|
|
import pipes
|
|
import logging
|
|
import shutil
|
|
|
|
log = logging.getLogger(__name__)
|
|
|
|
class RemoteProcess(object):
|
|
__slots__ = [
|
|
'command', 'stdin', 'stdout', 'stderr', 'exitstatus', 'exited',
|
|
# for orchestra.remote.Remote to place a backreference
|
|
'remote',
|
|
]
|
|
def __init__(self, command, stdin, stdout, stderr, exitstatus, exited):
|
|
self.command = command
|
|
self.stdin = stdin
|
|
self.stdout = stdout
|
|
self.stderr = stderr
|
|
self.exitstatus = exitstatus
|
|
self.exited = exited
|
|
|
|
class Raw(object):
|
|
def __init__(self, value):
|
|
self.value = value
|
|
|
|
def __repr__(self):
|
|
return '{cls}({value!r})'.format(
|
|
cls=self.__class__.__name__,
|
|
value=self.value,
|
|
)
|
|
|
|
def quote(args):
|
|
def _quote(args):
|
|
for a in args:
|
|
if isinstance(a, Raw):
|
|
yield a.value
|
|
else:
|
|
yield pipes.quote(a)
|
|
return ' '.join(_quote(args))
|
|
|
|
def execute(client, args):
|
|
"""
|
|
Execute a command remotely.
|
|
|
|
Caller needs to handle stdin etc.
|
|
|
|
:param client: SSHConnection to run the command with
|
|
:param args: command to run
|
|
:type args: list of string
|
|
|
|
Returns a RemoteProcess, where exitstatus is a callable that will
|
|
block until the exit status is available.
|
|
"""
|
|
cmd = quote(args)
|
|
(host,port) = client.get_transport().getpeername()
|
|
log.debug('Running [{h}]: {cmd!r}'.format(h=host, cmd=cmd))
|
|
(in_, out, err) = client.exec_command(cmd)
|
|
|
|
def get_exitstatus():
|
|
status = out.channel.recv_exit_status()
|
|
# -1 on connection loss *and* signals; map to more pythonic None
|
|
if status == -1:
|
|
status = None
|
|
return status
|
|
|
|
def exitstatus_ready():
|
|
return out.channel.exit_status_ready()
|
|
|
|
r = RemoteProcess(
|
|
command=cmd,
|
|
stdin=in_,
|
|
stdout=out,
|
|
stderr=err,
|
|
# this is a callable that will block until the status is
|
|
# available
|
|
exitstatus=get_exitstatus,
|
|
exited=exitstatus_ready,
|
|
)
|
|
return r
|
|
|
|
def copy_to_log(f, logger, loglevel=logging.INFO):
|
|
# i can't seem to get fudge to fake an iterable, so using this old
|
|
# api for now
|
|
for line in f.xreadlines():
|
|
line = line.rstrip()
|
|
logger.log(loglevel, line)
|
|
|
|
def copy_and_close(src, fdst):
|
|
if src is not None:
|
|
if isinstance(src, basestring):
|
|
src = StringIO(src)
|
|
shutil.copyfileobj(src, fdst)
|
|
fdst.close()
|
|
|
|
def copy_file_to(f, dst):
|
|
if hasattr(dst, 'log'):
|
|
# looks like a Logger to me; not using isinstance to make life
|
|
# easier for unit tests
|
|
handler = copy_to_log
|
|
else:
|
|
handler = shutil.copyfileobj
|
|
return handler(f, dst)
|
|
|
|
|
|
class CommandFailedError(Exception):
|
|
def __init__(self, command, exitstatus, node=None):
|
|
self.command = command
|
|
self.exitstatus = exitstatus
|
|
self.node = node
|
|
|
|
def __str__(self):
|
|
return "Command failed on {node} with status {status}: {command!r}".format(
|
|
node=self.node,
|
|
status=self.exitstatus,
|
|
command=self.command,
|
|
)
|
|
|
|
|
|
class CommandCrashedError(Exception):
|
|
def __init__(self, command):
|
|
self.command = command
|
|
|
|
def __str__(self):
|
|
return "Command crashed: {command!r}".format(
|
|
command=self.command,
|
|
)
|
|
|
|
|
|
class ConnectionLostError(Exception):
|
|
def __init__(self, command):
|
|
self.command = command
|
|
|
|
def __str__(self):
|
|
return "SSH connection was lost: {command!r}".format(
|
|
command=self.command,
|
|
)
|
|
|
|
def spawn_asyncresult(fn, *args, **kwargs):
|
|
"""
|
|
Spawn a Greenlet and pass it's results to an AsyncResult.
|
|
|
|
This function is useful to shuffle data from a Greenlet to
|
|
AsyncResult, which then again is useful because any Greenlets that
|
|
raise exceptions will cause tracebacks to be shown on stderr by
|
|
gevent, even when ``.link_exception`` has been called. Using an
|
|
AsyncResult avoids this.
|
|
"""
|
|
r = gevent.event.AsyncResult()
|
|
def wrapper():
|
|
try:
|
|
value = fn(*args, **kwargs)
|
|
except Exception as e:
|
|
r.set_exception(e)
|
|
else:
|
|
r.set(value)
|
|
gevent.spawn(wrapper)
|
|
|
|
return r
|
|
|
|
class Sentinel(object):
|
|
def __init__(self, name):
|
|
self.name = name
|
|
|
|
def __str__(self):
|
|
return self.name
|
|
|
|
PIPE = Sentinel('PIPE')
|
|
|
|
class KludgeFile(object):
|
|
"""
|
|
Wrap Paramiko's ChannelFile in a way that lets ``f.close()``
|
|
actually cause an EOF for the remote command.
|
|
"""
|
|
def __init__(self, wrapped):
|
|
self._wrapped = wrapped
|
|
|
|
def __getattr__(self, name):
|
|
return getattr(self._wrapped, name)
|
|
|
|
def close(self):
|
|
self._wrapped.close()
|
|
self._wrapped.channel.shutdown_write()
|
|
|
|
def run(
|
|
client, args,
|
|
stdin=None, stdout=None, stderr=None,
|
|
logger=None,
|
|
check_status=True,
|
|
wait=True,
|
|
):
|
|
"""
|
|
Run a command remotely.
|
|
|
|
:param client: SSHConnection to run the command with
|
|
:param args: command to run
|
|
:type args: list of string
|
|
:param stdin: Standard input to send; either a string, a file-like object, None, or `PIPE`. `PIPE` means caller is responsible for closing stdin, or command may never exit.
|
|
:param stdout: What to do with standard output. Either a file-like object, a `logging.Logger`, `PIPE`, or `None` for copying to default log. `PIPE` means caller is responsible for reading, or command may never exit.
|
|
:param stderr: What to do with standard error. See `stdout`.
|
|
:param logger: If logging, write stdout/stderr to "out" and "err" children of this logger. Defaults to logger named after this module.
|
|
:param check_status: Whether to raise CalledProcessError on non-zero exit status, and . Defaults to True. All signals and connection loss are made to look like SIGHUP.
|
|
:param wait: Whether to wait for process to exit. If False, returned ``r.exitstatus`` s a `gevent.event.AsyncResult`, and the actual status is available via ``.get()``.
|
|
"""
|
|
r = execute(client, args)
|
|
|
|
r.stdin = KludgeFile(wrapped=r.stdin)
|
|
|
|
g_in = None
|
|
if stdin is not PIPE:
|
|
g_in = gevent.spawn(copy_and_close, stdin, r.stdin)
|
|
r.stdin = None
|
|
else:
|
|
assert not wait, "Using PIPE for stdin without wait=False would deadlock."
|
|
|
|
if logger is None:
|
|
logger = log
|
|
|
|
g_err = None
|
|
if stderr is not PIPE:
|
|
if stderr is None:
|
|
stderr = logger.getChild('err')
|
|
g_err = gevent.spawn(copy_file_to, r.stderr, stderr)
|
|
r.stderr = stderr
|
|
else:
|
|
assert not wait, "Using PIPE for stderr without wait=False would deadlock."
|
|
|
|
g_out = None
|
|
if stdout is not PIPE:
|
|
if stdout is None:
|
|
stdout = logger.getChild('out')
|
|
g_out = gevent.spawn(copy_file_to, r.stdout, stdout)
|
|
r.stdout = stdout
|
|
else:
|
|
assert not wait, "Using PIPE for stdout without wait=False would deadlock."
|
|
|
|
def _check_status(status):
|
|
if g_err is not None:
|
|
g_err.get()
|
|
if g_out is not None:
|
|
g_out.get()
|
|
if g_in is not None:
|
|
g_in.get()
|
|
|
|
status = status()
|
|
if check_status:
|
|
if status is None:
|
|
# command either died due to a signal, or the connection
|
|
# was lost
|
|
transport = client.get_transport()
|
|
if not transport.is_active():
|
|
# look like we lost the connection
|
|
raise ConnectionLostError(command=r.command)
|
|
|
|
# connection seems healthy still, assuming it was a
|
|
# signal; sadly SSH does not tell us which signal
|
|
raise CommandCrashedError(command=r.command)
|
|
if status != 0:
|
|
(host,port) = client.get_transport().getpeername()
|
|
raise CommandFailedError(command=r.command, exitstatus=status, node=host)
|
|
return status
|
|
|
|
if wait:
|
|
r.exitstatus = _check_status(r.exitstatus)
|
|
else:
|
|
r.exitstatus = spawn_asyncresult(_check_status, r.exitstatus)
|
|
|
|
return r
|
|
|
|
|
|
def wait(processes):
|
|
"""
|
|
Wait for all given processes to exit.
|
|
|
|
Raise if any one of them fails.
|
|
"""
|
|
for proc in processes:
|
|
assert isinstance(proc.exitstatus, gevent.event.AsyncResult)
|
|
proc.exitstatus.get()
|