def uuid(self):
ids = self.get('id')
if len(ids) != 1:
raise GsyncdError(
"volume info of %s obtained from %s: "
"ambiguous uuid", self.volume, self.host)
return ids[0].text
def gethostbyname(hnam):
"""gethostbyname wrapper"""
try:
return socket.gethostbyname(hnam)
except socket.gaierror:
ex = sys.exc_info()[1]
raise GsyncdError("failed to resolve %s: %s" %
(hnam, ex.strerror))
def _unlink(path):
try:
os.unlink(path)
except (OSError, IOError):
if sys.exc_info()[1].errno == ENOENT:
pass
else:
raise GsyncdError('Unlink error: %s' % path)
def parse_url(ustr):
"""instantiate an url object by scheme-to-class dispatch
The url classes taken into consideration are the ones in
this module whose names are full-caps.
"""
m = UrlRX.match(ustr)
if not m:
ustr = desugar(ustr)
m = UrlRX.match(ustr)
if not m:
raise GsyncdError("malformed url")
sch, path = m.groups()
this = sys.modules[__name__]
if not hasattr(this, sch.upper()):
raise GsyncdError("unknown url scheme " + sch)
return getattr(this, sch.upper())(path)
def checkpt_service(self, chan, chkpt, tgt):
"""checkpoint service loop
monitor and verify checkpoint status for @chkpt, and listen
for incoming requests for whom we serve a pretty-formatted
status report"""
if not chkpt:
# dummy loop for the case when there is no checkpt set
while True:
select([chan], [], [])
conn, _ = chan.accept()
conn.send(self.get_extra_info())
conn.close()
completed = self._checkpt_param(chkpt, 'completed', xtimish=False)
if completed:
completed = tuple(int(x) for x in completed.split('.'))
while True:
s,_,_ = select([chan], [], [], (not completed) and 5 or None)
# either request made and we re-check to not
# give back stale data, or we still hunting for completion
if self.native_xtime(tgt) and self.native_xtime(tgt) < self.volmark:
# indexing has been reset since setting the checkpoint
status = "is invalid"
else:
xtr = self.xtime('.', self.slave)
if isinstance(xtr, int):
raise GsyncdError("slave root directory is unaccessible (%s)",
os.strerror(xtr))
ncompleted = self.xtime_geq(xtr, tgt)
if completed and not ncompleted: # stale data
logging.warn("completion time %s for checkpoint %s became stale" % \
(self.humantime(*completed), chkpt))
completed = None
gconf.confdata.delete('checkpoint-completed')
if ncompleted and not completed: # just reaching completion
completed = "%.6f" % time.time()
self._set_checkpt_param(chkpt, 'completed', completed, xtimish=False)
completed = tuple(int(x) for x in completed.split('.'))
logging.info("checkpoint %s completed" % chkpt)
status = completed and \
"completed at " + self.humantime(completed[0]) or \
"not reached yet"
if s:
conn = None
try:
conn, _ = chan.accept()
try:
conn.send(" | checkpoint %s %s %s" % (chkpt, status,self.get_extra_info()))
except:
exc = sys.exc_info()[1]
if (isinstance(exc, OSError) or isinstance(exc, IOError)) and \
exc.errno == EPIPE:
logging.debug('checkpoint client disconnected')
else:
raise
finally:
if conn:
conn.close()
def makersc(aa, check=True):
if not aa:
return ([], None, None)
ra = [resource.parse_url(u) for u in aa]
local = ra[0]
remote = None
if len(ra) > 1:
remote = ra[1]
if check and not local.can_connect_to(remote):
raise GsyncdError("%s cannot work with %s" % (local.path, remote and remote.path))
return (ra, local, remote)
def rsync(self, files, *args):
"""invoke rsync"""
if not files:
raise GsyncdError("no files to sync")
logging.debug("files: " + ", ".join(files))
argv = gconf.rsync_command.split() + gconf.rsync_extra.split() + [
'-aR'
] + files + list(args)
po = Popen(argv, stderr=subprocess.PIPE)
po.wait()
po.terminate_geterr(fail_on_err=False)
return po
def distribute(*resources):
master, slave = resources
mvol = Volinfo(master.volume, master.host)
logging.debug('master bricks: ' + repr(mvol.bricks))
prelude = []
si = slave
slave_host = None
slave_vol = None
if isinstance(slave, SSH):
prelude = gconf.ssh_command.split() + [slave.remote_addr]
si = slave.inner_rsc
logging.debug('slave SSH gateway: ' + slave.remote_addr)
if isinstance(si, FILE):
sbricks = {'host': 'localhost', 'dir': si.path}
suuid = uuid.uuid5(uuid.NAMESPACE_URL, slave.get_url(canonical=True))
elif isinstance(si, GLUSTER):
svol = Volinfo(si.volume, slave.remote_addr.split('@')[-1])
sbricks = svol.bricks
suuid = svol.uuid
slave_host = slave.remote_addr.split('@')[-1]
slave_vol = si.volume
else:
raise GsyncdError("unknown slave type " + slave.url)
logging.info('slave bricks: ' + repr(sbricks))
if isinstance(si, FILE):
slaves = [slave.url]
else:
slavenodes = set(b['host'] for b in sbricks)
if isinstance(slave, SSH) and not gconf.isolated_slave:
rap = SSH.parse_ssh_address(slave)
slaves = [
'ssh://' + rap['user'] + '@' + h + ':' + si.url
for h in slavenodes
]
else:
slavevols = [h + ':' + si.volume for h in slavenodes]
if isinstance(slave, SSH):
slaves = [
'ssh://' + rap.remote_addr + ':' + v for v in slavevols
]
else:
slaves = slavevols
workerspex = []
for idx, brick in enumerate(mvol.bricks):
if is_host_local(brick['host']):
is_hot = mvol.is_hot(":".join([brick['host'], brick['dir']]))
workerspex.append((brick['dir'], slaves[idx % len(slaves)],
get_subvol_num(idx, mvol, is_hot), is_hot))
logging.info('worker specs: ' + repr(workerspex))
return workerspex, suuid, slave_vol, slave_host, master
def _unlink(path):
import os
from errno import ENOENT
from syncdutils import GsyncdError
import sys
try:
os.unlink(path)
except (OSError, IOError):
if sys.exc_info()[1].errno == ENOENT:
pass
else:
raise GsyncdError('Unlink error: %s' % path)
def get_sys_volinfo(self):
"""query volume marks on fs root
err out on multiple foreign masters
"""
fgn_vis, nat_vi = self.master.server.aggregated.foreign_volume_infos(), \
self.master.server.aggregated.native_volume_info()
fgn_vi = None
if fgn_vis:
if len(fgn_vis) > 1:
raise GsyncdError("cannot work with multiple foreign masters")
fgn_vi = fgn_vis[0]
return fgn_vi, nat_vi
def _volinfo_hook_relax_foreign(self):
volinfo_sys = self.get_sys_volinfo()
fgn_vi = volinfo_sys[self.KFGN]
if fgn_vi:
expiry = fgn_vi['timeout'] - int(time.time()) + 1
logging.info('foreign volume info found, waiting %d sec for expiry' % \
expiry)
time.sleep(expiry)
volinfo_sys = self.get_sys_volinfo()
self.volinfo_state, state_change = self.volinfo_state_machine(self.volinfo_state,
volinfo_sys)
if self.inter_master:
raise GsyncdError("cannot be intermediate master in special mode")
return (volinfo_sys, state_change)
def select_vi(vi0, vi):
param.index += 1
if vi and (not vi0 or vi0['uuid'] == vi['uuid']):
if not vi0 and not param.relax_mismatch:
param.state_change = param.index
# valid new value found; for the rest, we are graceful about
# uuid mismatch
param.relax_mismatch = True
return vi
if vi0 and vi and vi0['uuid'] != vi['uuid'] and not param.relax_mismatch:
# uuid mismatch for master candidate, bail out
raise GsyncdError("aborting on uuid change from %s to %s" % \
(vi0['uuid'], vi['uuid']))
# fall back to old
return vi0
def crawl(self):
changes = []
try:
self.master.server.changelog_scan()
self.crawls += 1
except OSError:
self.fallback_xsync()
changes = self.master.server.changelog_getchanges()
if changes:
xtl = self.xtime(self.FLAT_DIR_HIERARCHY)
if isinstance(xtl, int):
raise GsyncdError('master is corrupt')
logging.debug('processing changes %s' % repr(changes))
self.process(changes)
self.upd_stime(xtl)
def startup(**kw):
"""set up logging, pidfile grabbing, daemonization"""
if getattr(gconf, 'pid_file', None) and kw.get('go_daemon') != 'postconn':
if not grabpidfile():
sys.stderr.write("pidfile is taken, exiting.\n")
sys.exit(2)
gconf.pid_file_owned = True
if kw.get('go_daemon') == 'should':
x, y = os.pipe()
gconf.cpid = os.fork()
if gconf.cpid:
os.close(x)
sys.exit()
os.close(y)
os.setsid()
dn = os.open(os.devnull, os.O_RDWR)
for f in (sys.stdin, sys.stdout, sys.stderr):
os.dup2(dn, f.fileno())
if getattr(gconf, 'pid_file', None):
if not grabpidfile(gconf.pid_file + '.tmp'):
raise GsyncdError("cannot grab temporary pidfile")
os.rename(gconf.pid_file + '.tmp', gconf.pid_file)
# wait for parent to terminate
# so we can start up with
# no messing from the dirty
# ol' bustard
select((x,), (), ())
os.close(x)
lkw = {}
if gconf.log_level:
lkw['level'] = gconf.log_level
if kw.get('log_file'):
if kw['log_file'] in ('-', '/dev/stderr'):
lkw['stream'] = sys.stderr
elif kw['log_file'] == '/dev/stdout':
lkw['stream'] = sys.stdout
else:
lkw['filename'] = kw['log_file']
GLogger.setup(label=kw.get('label'), **lkw)
lkw.update({'saved_label': kw.get('label')})
gconf.log_metadata = lkw
gconf.log_exit = True
def rsync(self, files, *args):
"""invoke rsync"""
if not files:
raise GsyncdError("no files to sync")
logging.debug("files: " + ", ".join(files))
argv = gconf.rsync_command.split() + \
['-aR0', '--files-from=-', '--super', '--numeric-ids', '--no-implied-dirs'] + \
gconf.rsync_options.split() + (boolify(gconf.use_rsync_xattrs) and ['--xattrs'] or []) + \
['.'] + list(args)
po = Popen(argv, stdin=subprocess.PIPE, stderr=subprocess.PIPE)
for f in files:
po.stdin.write(f)
po.stdin.write('\0')
po.stdin.close()
po.wait()
po.terminate_geterr(fail_on_err=False)
return po
def start_fd_client(self, i, o, **opts):
"""set up RePCe client, handshake with server
It's cut out as a separate method to let
subclasses hook into client startup
"""
self.server = RepceClient(i, o)
rv = self.server.__version__()
exrv = {'proto': repce.repce_version, 'object': Server.version()}
da0 = (rv, exrv)
da1 = ({}, {})
for i in range(2):
for k, v in da0[i].iteritems():
da1[i][k] = int(v)
if da1[0] != da1[1]:
raise GsyncdError(
"RePCe major version mismatch: local %s, remote %s" %
(exrv, rv))
def __init__(self, vol, host='localhost', prelude=[]):
po = Popen(prelude + ['gluster', '--xml', '--remote-host=' + host,
'volume', 'info', vol],
stdout=PIPE, stderr=PIPE)
vix = po.stdout.read()
po.wait()
po.terminate_geterr()
vi = XET.fromstring(vix)
if vi.find('opRet').text != '0':
if prelude:
via = '(via %s) ' % prelude.join(' ')
else:
via = ' '
raise GsyncdError('getting volume info of %s%s '
'failed with errorcode %s' %
(vol, via, vi.find('opErrno').text))
self.tree = vi
self.volume = vol
self.host = host
def update_to(self, dct, allow_unresolved=False):
"""update @dct from key/values of ours.
key/values are collected from .config by filtering the regexp sections
according to match, and from .section. The values are treated as
templates, which are substituted from .auxdicts and (in case of regexp
sections) match groups.
"""
if not self.peers:
raise GsyncdError('no peers given, cannot select matching options')
def update_from_sect(sect, mud):
for k, v in self.config._sections[sect].items():
# Template expects String to be passed
# if any config value is not string then it
# fails with ValueError
v = "{0}".format(v)
if k == '__name__':
continue
if allow_unresolved:
dct[k] = Template(v).safe_substitute(mud)
else:
dct[k] = Template(v).substitute(mud)
for sect in self.ord_sections():
sp = self.parse_section(sect)
if isinstance(sp[0], re_type) and len(sp) == len(self.peers):
match = True
mad = {}
for i in range(len(sp)):
m = sp[i].search(self.peers[i])
if not m:
match = False
break
for j in range(len(m.groups())):
mad['match%d_%d' % (i + 1, j + 1)] = m.groups()[j]
if match:
update_from_sect(sect, MultiDict(dct, mad, *self.auxdicts))
if self.config.has_section(self.section()):
update_from_sect(self.section(), MultiDict(dct, *self.auxdicts))
def __init__(self, args, *a, **kw):
"""customizations for subprocess.Popen instantiation
- 'close_fds' is taken to be the default
- if child's stderr is chosen to be managed,
register it with the error handler thread
"""
self.args = args
if 'close_fds' not in kw:
kw['close_fds'] = True
self.lock = threading.Lock()
try:
sup(self, args, *a, **kw)
except:
ex = sys.exc_info()[1]
if not isinstance(ex, OSError):
raise
raise GsyncdError("""execution of "%s" failed with %s (%s)""" % \
(args[0], errno.errorcode[ex.errno], os.strerror(ex.errno)))
if kw['stderr'] == subprocess.PIPE:
assert (getattr(self, 'errhandler', None))
self.errstore[self] = []
def desugar(ustr):
"""transform sugared url strings to standard <scheme>://<urlbody> form
parsing logic enforces the constraint that sugared forms should contatin
a ':' or a '/', which ensures that sugared urls do not conflict with
gluster volume names.
"""
m = re.match('([^:]*):(.*)', ustr)
if m:
if not m.groups()[0]:
return "gluster://localhost" + ustr
elif '@' in m.groups()[0] or re.search('[:/]', m.groups()[1]):
return "ssh://" + ustr
else:
return "gluster://" + ustr
else:
if ustr[0] != '/':
raise GsyncdError("cannot resolve sugared url '%s'" % ustr)
ap = os.path.normpath(ustr)
if ap.startswith('//'):
ap = ap[1:]
return "file://" + ap
def crawl(self, path='.', xtr=None, done=0):
""" generate a CHANGELOG file consumable by process_change """
if path == '.':
self.open()
self.crawls += 1
if not xtr:
# get the root stime and use it for all comparisons
xtr = self.xtime('.', self.slave)
if isinstance(xtr, int):
if xtr != ENOENT:
raise GsyncdError('slave is corrupt')
xtr = self.minus_infinity
xtl = self.xtime(path)
if isinstance(xtl, int):
raise GsyncdError('master is corrupt')
if xtr == xtl:
if path == '.':
self.close()
return
self.xtime_reversion_hook(path, xtl, xtr)
logging.debug("entering " + path)
dem = self.master.server.entries(path)
pargfid = self.master.server.gfid(path)
if isinstance(pargfid, int):
logging.warn('skipping directory %s' % (path))
for e in dem:
bname = e
e = os.path.join(path, e)
st = lstat(e)
if isinstance(st, int):
logging.warn('%s got purged in the interim..' % e)
continue
gfid = self.master.server.gfid(e)
if isinstance(gfid, int):
logging.warn('skipping entry %s..' % (e))
continue
xte = self.xtime(e)
if isinstance(xte, int):
raise GsyncdError('master is corrupt')
if not self.need_sync(e, xte, xtr):
continue
mo = st.st_mode
if stat.S_ISDIR(mo):
self.write_entry_change(
"E", [gfid, 'MKDIR',
escape(os.path.join(pargfid, bname))])
self.crawl(e, xtr)
elif stat.S_ISLNK(mo):
rl = errno_wrap(os.readlink, [en], [ENOENT])
if isinstance(rl, int):
continue
self.write_entry_change("E", [
gfid, 'SYMLINK',
escape(os.path.join(pargfid, bname)), rl
])
else:
# if a file has a hardlink, create a Changelog entry as 'LINK' so the slave
# side will decide if to create the new entry, or to create link.
if st.st_nlink == 1:
self.write_entry_change(
"E",
[gfid, 'MKNOD',
escape(os.path.join(pargfid, bname))])
else:
self.write_entry_change(
"E",
[gfid, 'LINK',
escape(os.path.join(pargfid, bname))])
if stat.S_ISREG(mo):
self.write_entry_change("D", [gfid])
if path == '.':
logging.info('processing xsync changelog %s' % self.fname())
self.close()
self.process([self.fname()], done)
self.upd_stime(xtl)
def service_loop(self, *args):
"""enter service loop
- if slave given, instantiate GMaster and
pass control to that instance, which implements
master behavior
- else do that's what's inherited
"""
if args:
slave = args[0]
if gconf.local_path:
class brickserver(FILE.FILEServer):
local_path = gconf.local_path
aggregated = self.server
@classmethod
def entries(cls, path):
e = super(brickserver, cls).entries(path)
# on the brick don't mess with /.glusterfs
if path == '.':
try:
e.remove('.glusterfs')
except ValueError:
pass
return e
@classmethod
def lstat(cls, e):
""" path based backend stat """
return super(brickserver, cls).lstat(e)
@classmethod
def gfid(cls, e):
""" path based backend gfid fetch """
return super(brickserver, cls).gfid(e)
@classmethod
def linkto_check(cls, e):
return super(brickserver, cls).linkto_check(e)
if gconf.slave_id:
# define {,set_}xtime in slave, thus preempting
# the call to remote, so that it takes data from
# the local brick
slave.server.xtime = types.MethodType(
lambda _self, path, uuid: (
brickserver.xtime(path,
uuid + '.' + gconf.slave_id)
),
slave.server)
slave.server.stime = types.MethodType(
lambda _self, path, uuid: (
brickserver.stime(path,
uuid + '.' + gconf.slave_id)
),
slave.server)
slave.server.set_stime = types.MethodType(
lambda _self, path, uuid, mark: (
brickserver.set_stime(path,
uuid + '.' + gconf.slave_id,
mark)
),
slave.server)
(g1, g2, g3) = self.gmaster_instantiate_tuple(slave)
g1.master.server = brickserver
g2.master.server = brickserver
g3.master.server = brickserver
else:
(g1, g2, g3) = self.gmaster_instantiate_tuple(slave)
g1.master.server.aggregated = gmaster.master.server
g2.master.server.aggregated = gmaster.master.server
g3.master.server.aggregated = gmaster.master.server
# bad bad bad: bad way to do things like this
# need to make this elegant
# register the crawlers and start crawling
# g1 ==> Xsync, g2 ==> config.change_detector(changelog by default)
# g3 ==> changelog History
(inf, ouf, ra, wa) = gconf.rpc_fd.split(',')
os.close(int(ra))
os.close(int(wa))
changelog_agent = RepceClient(int(inf), int(ouf))
rv = changelog_agent.version()
if int(rv) != CHANGELOG_AGENT_CLIENT_VERSION:
raise GsyncdError(
"RePCe major version mismatch(changelog agent): "
"local %s, remote %s" %
(CHANGELOG_AGENT_CLIENT_VERSION, rv))
g1.register()
try:
(workdir, logfile) = g2.setup_working_dir()
# register with the changelog library
# 9 == log level (DEBUG)
# 5 == connection retries
changelog_agent.register(gconf.local_path,
workdir, logfile, 9, 5)
g2.register(changelog_agent)
g3.register(changelog_agent)
except ChangelogException as e:
logging.debug("Changelog register failed: %s - %s" %
(e.errno, e.strerror))
# oneshot: Try to use changelog history api, if not
# available switch to FS crawl
# Note: if config.change_detector is xsync then
# it will not use changelog history api
try:
g3.crawlwrap(oneshot=True)
except (ChangelogException, NoPurgeTimeAvailable,
PartialHistoryAvailable) as e:
if isinstance(e, ChangelogException):
logging.debug('Changelog history crawl failed, failback '
'to xsync: %s - %s' % (e.errno, e.strerror))
elif isinstance(e, NoPurgeTimeAvailable):
logging.debug('Using xsync crawl since no purge time '
'available')
elif isinstance(e, PartialHistoryAvailable):
logging.debug('Using xsync crawl after consuming history '
'till %s' % str(e))
g1.crawlwrap(oneshot=True)
# crawl loop: Try changelog crawl, if failed
# switch to FS crawl
try:
g2.crawlwrap()
except ChangelogException as e:
logging.debug('Changelog crawl failed, failback to xsync: '
'%s - %s' % (e.errno, e.strerror))
g1.crawlwrap()
else:
sup(self, *args)
def xtime_reversion_hook(self, path, xtl, xtr):
if xtr > xtl:
raise GsyncdError("timestamp corruption for " + path)
def main_i():
"""internal main routine
parse command line, decide what action will be taken;
we can either:
- query/manipulate configuration
- format gsyncd urls using gsyncd's url parsing engine
- start service in following modes, in given stages:
- agent: startup(), ChangelogAgent()
- monitor: startup(), monitor()
- master: startup(), connect_remote(), connect(), service_loop()
- slave: startup(), connect(), service_loop()
"""
rconf = {'go_daemon': 'should'}
def store_abs(opt, optstr, val, parser):
if val and val != '-':
val = os.path.abspath(val)
setattr(parser.values, opt.dest, val)
def store_local(opt, optstr, val, parser):
rconf[opt.dest] = val
def store_local_curry(val):
return lambda o, oo, vx, p: store_local(o, oo, val, p)
def store_local_obj(op, dmake):
return lambda o, oo, vx, p: store_local(
o, oo, FreeObject(op=op, **dmake(vx)), p)
op = OptionParser(
usage="%prog [options...] <master> <slave>", version="%prog 0.0.1")
op.add_option('--gluster-command-dir', metavar='DIR', default='')
op.add_option('--gluster-log-file', metavar='LOGF',
default=os.devnull, type=str, action='callback',
callback=store_abs)
op.add_option('--gluster-log-level', metavar='LVL')
op.add_option('--gluster-params', metavar='PRMS', default='')
op.add_option(
'--glusterd-uuid', metavar='UUID', type=str, default='',
help=SUPPRESS_HELP)
op.add_option(
'--gluster-cli-options', metavar='OPTS', default='--log-file=-')
op.add_option('--mountbroker', metavar='LABEL')
op.add_option('-p', '--pid-file', metavar='PIDF', type=str,
action='callback', callback=store_abs)
op.add_option('-l', '--log-file', metavar='LOGF', type=str,
action='callback', callback=store_abs)
op.add_option('--iprefix', metavar='LOGD', type=str,
action='callback', callback=store_abs)
op.add_option('--changelog-log-file', metavar='LOGF', type=str,
action='callback', callback=store_abs)
op.add_option('--log-file-mbr', metavar='LOGF', type=str,
action='callback', callback=store_abs)
op.add_option('--state-file', metavar='STATF', type=str,
action='callback', callback=store_abs)
op.add_option('--state-detail-file', metavar='STATF',
type=str, action='callback', callback=store_abs)
op.add_option('--georep-session-working-dir', metavar='STATF',
type=str, action='callback', callback=store_abs)
op.add_option('--ignore-deletes', default=False, action='store_true')
op.add_option('--isolated-slave', default=False, action='store_true')
op.add_option('--use-rsync-xattrs', default=False, action='store_true')
op.add_option('--pause-on-start', default=False, action='store_true')
op.add_option('-L', '--log-level', metavar='LVL')
op.add_option('-r', '--remote-gsyncd', metavar='CMD',
default=os.path.abspath(sys.argv[0]))
op.add_option('--volume-id', metavar='UUID')
op.add_option('--slave-id', metavar='ID')
op.add_option('--session-owner', metavar='ID')
op.add_option('--local-id', metavar='ID', help=SUPPRESS_HELP, default='')
op.add_option(
'--local-path', metavar='PATH', help=SUPPRESS_HELP, default='')
op.add_option('-s', '--ssh-command', metavar='CMD', default='ssh')
op.add_option('--ssh-command-tar', metavar='CMD', default='ssh')
op.add_option('--rsync-command', metavar='CMD', default='rsync')
op.add_option('--rsync-options', metavar='OPTS', default='')
op.add_option('--rsync-ssh-options', metavar='OPTS', default='--compress')
op.add_option('--timeout', metavar='SEC', type=int, default=120)
op.add_option('--connection-timeout', metavar='SEC',
type=int, default=60, help=SUPPRESS_HELP)
op.add_option('--sync-jobs', metavar='N', type=int, default=3)
op.add_option('--replica-failover-interval', metavar='N',
type=int, default=1)
op.add_option('--changelog-archive-format', metavar='N',
type=str, default="%Y%m")
op.add_option(
'--turns', metavar='N', type=int, default=0, help=SUPPRESS_HELP)
op.add_option('--allow-network', metavar='IPS', default='')
op.add_option('--socketdir', metavar='DIR')
op.add_option('--state-socket-unencoded', metavar='SOCKF',
type=str, action='callback', callback=store_abs)
op.add_option('--checkpoint', metavar='LABEL', default='')
# tunables for failover/failback mechanism:
# None - gsyncd behaves as normal
# blind - gsyncd works with xtime pairs to identify
# candidates for synchronization
# wrapup - same as normal mode but does not assign
# xtimes to orphaned files
# see crawl() for usage of the above tunables
op.add_option('--special-sync-mode', type=str, help=SUPPRESS_HELP)
# changelog or xtime? (TODO: Change the default)
op.add_option(
'--change-detector', metavar='MODE', type=str, default='xtime')
# sleep interval for change detection (xtime crawl uses a hardcoded 1
# second sleep time)
op.add_option('--change-interval', metavar='SEC', type=int, default=3)
# working directory for changelog based mechanism
op.add_option('--working-dir', metavar='DIR', type=str,
action='callback', callback=store_abs)
op.add_option('--use-tarssh', default=False, action='store_true')
op.add_option('-c', '--config-file', metavar='CONF',
type=str, action='callback', callback=store_local)
# duh. need to specify dest or value will be mapped to None :S
op.add_option('--monitor', dest='monitor', action='callback',
callback=store_local_curry(True))
op.add_option('--agent', dest='agent', action='callback',
callback=store_local_curry(True))
op.add_option('--resource-local', dest='resource_local',
type=str, action='callback', callback=store_local)
op.add_option('--resource-remote', dest='resource_remote',
type=str, action='callback', callback=store_local)
op.add_option('--feedback-fd', dest='feedback_fd', type=int,
help=SUPPRESS_HELP, action='callback', callback=store_local)
op.add_option('--rpc-fd', dest='rpc_fd', type=str, help=SUPPRESS_HELP)
op.add_option('--listen', dest='listen', help=SUPPRESS_HELP,
action='callback', callback=store_local_curry(True))
op.add_option('-N', '--no-daemon', dest="go_daemon",
action='callback', callback=store_local_curry('dont'))
op.add_option('--verify', type=str, dest="verify",
action='callback', callback=store_local)
op.add_option('--create', type=str, dest="create",
action='callback', callback=store_local)
op.add_option('--delete', dest='delete', action='callback',
callback=store_local_curry(True))
op.add_option('--debug', dest="go_daemon", action='callback',
callback=lambda *a: (store_local_curry('dont')(*a),
setattr(
a[-1].values, 'log_file', '-'),
setattr(a[-1].values, 'log_level',
'DEBUG'),
setattr(a[-1].values,
'changelog_log_file', '-')))
op.add_option('--path', type=str, action='append')
for a in ('check', 'get'):
op.add_option('--config-' + a, metavar='OPT', type=str, dest='config',
action='callback',
callback=store_local_obj(a, lambda vx: {'opt': vx}))
op.add_option('--config-get-all', dest='config', action='callback',
callback=store_local_obj('get', lambda vx: {'opt': None}))
for m in ('', '-rx', '-glob'):
# call this code 'Pythonic' eh?
# have to define a one-shot local function to be able
# to inject (a value depending on the)
# iteration variable into the inner lambda
def conf_mod_opt_regex_variant(rx):
op.add_option('--config-set' + m, metavar='OPT VAL', type=str,
nargs=2, dest='config', action='callback',
callback=store_local_obj('set', lambda vx: {
'opt': vx[0], 'val': vx[1], 'rx': rx}))
op.add_option('--config-del' + m, metavar='OPT', type=str,
dest='config', action='callback',
callback=store_local_obj('del', lambda vx: {
'opt': vx, 'rx': rx}))
conf_mod_opt_regex_variant(m and m[1:] or False)
op.add_option('--normalize-url', dest='url_print',
action='callback', callback=store_local_curry('normal'))
op.add_option('--canonicalize-url', dest='url_print',
action='callback', callback=store_local_curry('canon'))
op.add_option('--canonicalize-escape-url', dest='url_print',
action='callback', callback=store_local_curry('canon_esc'))
tunables = [norm(o.get_opt_string()[2:])
for o in op.option_list
if (o.callback in (store_abs, 'store_true', None) and
o.get_opt_string() not in ('--version', '--help'))]
remote_tunables = ['listen', 'go_daemon', 'timeout',
'session_owner', 'config_file', 'use_rsync_xattrs']
rq_remote_tunables = {'listen': True}
# precedence for sources of values: 1) commandline, 2) cfg file, 3)
# defaults for this to work out we need to tell apart defaults from
# explicitly set options... so churn out the defaults here and call
# the parser with virgin values container.
defaults = op.get_default_values()
opts, args = op.parse_args(values=optparse.Values())
args_orig = args[:]
r = rconf.get('resource_local')
if r:
if len(args) == 0:
args.append(None)
args[0] = r
r = rconf.get('resource_remote')
if r:
if len(args) == 0:
raise GsyncdError('local resource unspecfied')
elif len(args) == 1:
args.append(None)
args[1] = r
confdata = rconf.get('config')
if not (len(args) == 2 or
(len(args) == 1 and rconf.get('listen')) or
(len(args) <= 2 and confdata) or
rconf.get('url_print')):
sys.stderr.write("error: incorrect number of arguments\n\n")
sys.stderr.write(op.get_usage() + "\n")
sys.exit(1)
verify = rconf.get('verify')
if verify:
logging.info(verify)
logging.info("Able to spawn gsyncd.py")
return
restricted = os.getenv('_GSYNCD_RESTRICTED_')
if restricted:
allopts = {}
allopts.update(opts.__dict__)
allopts.update(rconf)
bannedtuns = set(allopts.keys()) - set(remote_tunables)
if bannedtuns:
raise GsyncdError('following tunables cannot be set with '
'restricted SSH invocaton: ' +
', '.join(bannedtuns))
for k, v in rq_remote_tunables.items():
if not k in allopts or allopts[k] != v:
raise GsyncdError('tunable %s is not set to value %s required '
'for restricted SSH invocaton' %
(k, v))
confrx = getattr(confdata, 'rx', None)
def makersc(aa, check=True):
if not aa:
return ([], None, None)
ra = [resource.parse_url(u) for u in aa]
local = ra[0]
remote = None
if len(ra) > 1:
remote = ra[1]
if check and not local.can_connect_to(remote):
raise GsyncdError("%s cannot work with %s" %
(local.path, remote and remote.path))
return (ra, local, remote)
if confrx:
# peers are regexen, don't try to parse them
if confrx == 'glob':
args = ['\A' + fnmatch.translate(a) for a in args]
canon_peers = args
namedict = {}
else:
dc = rconf.get('url_print')
rscs, local, remote = makersc(args_orig, not dc)
if dc:
for r in rscs:
print(r.get_url(**{'normal': {},
'canon': {'canonical': True},
'canon_esc': {'canonical': True,
'escaped': True}}[dc]))
return
pa = ([], [], [])
urlprms = (
{}, {'canonical': True}, {'canonical': True, 'escaped': True})
for x in rscs:
for i in range(len(pa)):
pa[i].append(x.get_url(**urlprms[i]))
_, canon_peers, canon_esc_peers = pa
# creating the namedict, a dict representing various ways of referring
# to / repreenting peers to be fillable in config templates
mods = (lambda x: x, lambda x: x[
0].upper() + x[1:], lambda x: 'e' + x[0].upper() + x[1:])
if remote:
rmap = {local: ('local', 'master'), remote: ('remote', 'slave')}
else:
rmap = {local: ('local', 'slave')}
namedict = {}
for i in range(len(rscs)):
x = rscs[i]
for name in rmap[x]:
for j in range(3):
namedict[mods[j](name)] = pa[j][i]
namedict[name + 'vol'] = x.volume
if name == 'remote':
namedict['remotehost'] = x.remotehost
if not 'config_file' in rconf:
rconf['config_file'] = os.path.join(
os.path.dirname(sys.argv[0]), "conf/gsyncd_template.conf")
upgrade_config_file(rconf['config_file'])
gcnf = GConffile(
rconf['config_file'], canon_peers,
defaults.__dict__, opts.__dict__, namedict)
checkpoint_change = False
if confdata:
opt_ok = norm(confdata.opt) in tunables + [None]
if confdata.op == 'check':
if opt_ok:
sys.exit(0)
else:
sys.exit(1)
elif not opt_ok:
raise GsyncdError("not a valid option: " + confdata.opt)
if confdata.op == 'get':
gcnf.get(confdata.opt)
elif confdata.op == 'set':
gcnf.set(confdata.opt, confdata.val, confdata.rx)
elif confdata.op == 'del':
gcnf.delete(confdata.opt, confdata.rx)
# when modifying checkpoint, it's important to make a log
# of that, so in that case we go on to set up logging even
# if its just config invocation
if confdata.opt == 'checkpoint' and confdata.op in ('set', 'del') and \
not confdata.rx:
checkpoint_change = True
if not checkpoint_change:
return
gconf.__dict__.update(defaults.__dict__)
gcnf.update_to(gconf.__dict__)
gconf.__dict__.update(opts.__dict__)
gconf.configinterface = gcnf
delete = rconf.get('delete')
if delete:
logging.info('geo-replication delete')
# Delete pid file, status file, socket file
cleanup_paths = []
if getattr(gconf, 'pid_file', None):
cleanup_paths.append(gconf.pid_file)
if getattr(gconf, 'state_file', None):
cleanup_paths.append(gconf.state_file)
if getattr(gconf, 'state_detail_file', None):
cleanup_paths.append(gconf.state_detail_file)
if getattr(gconf, 'state_socket_unencoded', None):
cleanup_paths.append(gconf.state_socket_unencoded)
cleanup_paths.append(rconf['config_file'][:-11] + "*")
# Cleanup changelog working dirs
if getattr(gconf, 'working_dir', None):
try:
shutil.rmtree(gconf.working_dir)
except (IOError, OSError):
if sys.exc_info()[1].errno == ENOENT:
pass
else:
raise GsyncdError(
'Error while removing working dir: %s' %
gconf.working_dir)
for path in cleanup_paths:
# To delete temp files
for f in glob.glob(path + "*"):
_unlink(f)
return
if restricted and gconf.allow_network:
ssh_conn = os.getenv('SSH_CONNECTION')
if not ssh_conn:
# legacy env var
ssh_conn = os.getenv('SSH_CLIENT')
if ssh_conn:
allowed_networks = [IPNetwork(a)
for a in gconf.allow_network.split(',')]
client_ip = IPAddress(ssh_conn.split()[0])
allowed = False
for nw in allowed_networks:
if client_ip in nw:
allowed = True
break
if not allowed:
raise GsyncdError("client IP address is not allowed")
ffd = rconf.get('feedback_fd')
if ffd:
fcntl.fcntl(ffd, fcntl.F_SETFD, fcntl.FD_CLOEXEC)
# normalize loglevel
lvl0 = gconf.log_level
if isinstance(lvl0, str):
lvl1 = lvl0.upper()
lvl2 = logging.getLevelName(lvl1)
# I have _never_ _ever_ seen such an utterly braindead
# error condition
if lvl2 == "Level " + lvl1:
raise GsyncdError('cannot recognize log level "%s"' % lvl0)
gconf.log_level = lvl2
if not privileged() and gconf.log_file_mbr:
gconf.log_file = gconf.log_file_mbr
if checkpoint_change:
try:
GLogger._gsyncd_loginit(log_file=gconf.log_file, label='conf')
if confdata.op == 'set':
logging.info('checkpoint %s set' % confdata.val)
gcnf.delete('checkpoint_completed')
gcnf.delete('checkpoint_target')
elif confdata.op == 'del':
logging.info('checkpoint info was reset')
# if it is removing 'checkpoint' then we need
# to remove 'checkpoint_completed' and 'checkpoint_target' too
gcnf.delete('checkpoint_completed')
gcnf.delete('checkpoint_target')
except IOError:
if sys.exc_info()[1].errno == ENOENT:
# directory of log path is not present,
# which happens if we get here from
# a peer-multiplexed "config-set checkpoint"
# (as that directory is created only on the
# original node)
pass
else:
raise
return
create = rconf.get('create')
if create:
if getattr(gconf, 'state_file', None):
update_file(gconf.state_file, lambda f: f.write(create + '\n'))
return
go_daemon = rconf['go_daemon']
be_monitor = rconf.get('monitor')
be_agent = rconf.get('agent')
rscs, local, remote = makersc(args)
if not be_monitor and isinstance(remote, resource.SSH) and \
go_daemon == 'should':
go_daemon = 'postconn'
log_file = None
else:
log_file = gconf.log_file
if be_monitor:
label = 'monitor'
elif be_agent:
label = 'agent'
elif remote:
# master
label = gconf.local_path
else:
label = 'slave'
startup(go_daemon=go_daemon, log_file=log_file, label=label)
resource.Popen.init_errhandler()
if be_agent:
os.setsid()
logging.debug('rpc_fd: %s' % repr(gconf.rpc_fd))
return agent(Changelog(), gconf.rpc_fd)
if be_monitor:
return monitor(*rscs)
logging.info("syncing: %s" % " -> ".join(r.url for r in rscs))
if remote:
go_daemon = remote.connect_remote(go_daemon=go_daemon)
if go_daemon:
startup(go_daemon=go_daemon, log_file=gconf.log_file)
# complete remote connection in child
remote.connect_remote(go_daemon='done')
local.connect()
if ffd:
os.close(ffd)
local.service_loop(*[r for r in [remote] if r])
def subcmd_delete(args):
import logging
import shutil
import glob
import sys
from errno import ENOENT, ENODATA
import struct
from syncdutils import GsyncdError, Xattr, errno_wrap
import gsyncdconfig as gconf
logging.info('geo-replication delete')
# remove the stime xattr from all the brick paths so that
# a re-create of a session will start sync all over again
stime_xattr_prefix = gconf.get('stime-xattr-prefix', None)
# Delete pid file, status file, socket file
cleanup_paths = []
cleanup_paths.append(gconf.get("pid-file"))
# Cleanup Session dir
try:
shutil.rmtree(gconf.get("georep-session-working-dir"))
except (IOError, OSError):
if sys.exc_info()[1].errno == ENOENT:
pass
else:
raise GsyncdError(
'Error while removing working dir: %s' %
gconf.get("georep-session-working-dir"))
# Cleanup changelog working dirs
try:
shutil.rmtree(gconf.get("working-dir"))
except (IOError, OSError):
if sys.exc_info()[1].errno == ENOENT:
pass
else:
raise GsyncdError(
'Error while removing working dir: %s' %
gconf.get("working-dir"))
for path in cleanup_paths:
# To delete temp files
for f in glob.glob(path + "*"):
_unlink(f)
if args.reset_sync_time and stime_xattr_prefix:
for p in args.paths:
if p != "":
# set stime to (0,0) to trigger full volume content resync
# to slave on session recreation
# look at master.py::Xcrawl hint: zero_zero
errno_wrap(Xattr.lsetxattr,
(p, stime_xattr_prefix + ".stime",
struct.pack("!II", 0, 0)),
[ENOENT, ENODATA])
errno_wrap(Xattr.lremovexattr,
(p, stime_xattr_prefix + ".entry_stime"),
[ENOENT, ENODATA])
return
def xtime_reversion_hook(self, path, xtl, xtr):
if not isinstance(xtr[0], int) and \
(isinstance(xtl[0], int) or xtr[0] > xtl[0]):
raise GsyncdError("timestamp corruption for " + path)
def crawl(self, path='.', xtl=None):
"""crawling...
Standing around
All the right people
Crawling
Tennis on Tuesday
The ladder is long
It is your nature
You've gotta suntan
Football on Sunday
Society boy
Recursively walk the master side tree and check if updates are
needed due to xtime differences. One invocation of crawl checks
children of @path and do a recursive enter only on
those directory children where there is an update needed.
Way of updates depend on file type:
- for symlinks, sync them directy and synchronously
- for regular children, register jobs for @path (cf. .add_job) to start
and wait on their rsync
- for directory children, register a job for @path which waits (.wait)
on jobs for the given child
(other kind of filesystem nodes are not considered)
Those slave side children which do not exist on master are simply
purged (see Server.purge).
Behavior is fault tolerant, synchronization is adaptive: if some action fails,
just go on relentlessly, adding a fail job (see .add_failjob) which will prevent
the .sendmark on @path, so when the next crawl will arrive to @path it will not
see it as up-to-date and will try to sync it again. While this semantics can be
supported by funky design principles (http://c2.com/cgi/wiki?LazinessImpatienceHubris),
the ultimate reason which excludes other possibilities is simply transience: we cannot
assert that the file systems (master / slave) underneath do not change and actions
taken upon some condition will not lose their context by the time they are performed.
"""
if path == '.':
if self.start:
self.crawls += 1
logging.debug("... crawl #%d done, took %.6f seconds" % \
(self.crawls, time.time() - self.start))
time.sleep(1)
self.start = time.time()
should_display_info = self.start - self.lastreport['time'] >= 60
if should_display_info:
logging.info("completed %d crawls, %d turns",
self.crawls - self.lastreport['crawls'],
self.turns - self.lastreport['turns'])
self.lastreport.update(crawls = self.crawls,
turns = self.turns,
time = self.start)
volinfo_sys, state_change = self.volinfo_hook()
if self.inter_master:
self.volinfo = volinfo_sys[self.KFGN]
else:
self.volinfo = volinfo_sys[self.KNAT]
if state_change == self.KFGN or (state_change == self.KNAT and not self.inter_master):
logging.info('new master is %s', self.uuid)
if self.volinfo:
logging.info("%s master with volume id %s ..." % \
(self.inter_master and "intermediate" or "primary",
self.uuid))
if state_change == self.KFGN:
gconf.configinterface.set('volume_id', self.uuid)
if self.volinfo:
if self.volinfo['retval']:
raise GsyncdError ("master is corrupt")
self.start_checkpoint_thread()
else:
if should_display_info or self.crawls == 0:
if self.inter_master:
logging.info("waiting for being synced from %s ..." % \
self.volinfo_state[self.KFGN]['uuid'])
else:
logging.info("waiting for volume info ...")
return
logging.debug("entering " + path)
if not xtl:
xtl = self.xtime(path)
if isinstance(xtl, int):
self.add_failjob(path, 'no-local-node')
return
xtr = self.xtime(path, self.slave)
if isinstance(xtr, int):
if xtr != ENOENT:
self.slave.server.purge(path)
try:
self.slave.server.mkdir(path)
except OSError:
self.add_failjob(path, 'no-remote-node')
return
xtr = self.minus_infinity
else:
self.xtime_reversion_hook(path, xtl, xtr)
if xtl == xtr:
if path == '.' and self.change_seen:
self.turns += 1
self.change_seen = False
if self.total_turns:
logging.info("finished turn #%s/%s" % \
(self.turns, self.total_turns))
if self.turns == self.total_turns:
logging.info("reached turn limit")
self.terminate = True
return
if path == '.':
self.change_seen = True
try:
dem = self.master.server.entries(path)
except OSError:
self.add_failjob(path, 'local-entries-fail')
return
random.shuffle(dem)
try:
des = self.slave.server.entries(path)
except OSError:
self.slave.server.purge(path)
try:
self.slave.server.mkdir(path)
des = self.slave.server.entries(path)
except OSError:
self.add_failjob(path, 'remote-entries-fail')
return
dd = set(des) - set(dem)
if dd:
self.purge_missing(path, dd)
chld = []
for e in dem:
e = os.path.join(path, e)
xte = self.xtime(e)
if isinstance(xte, int):
logging.warn("irregular xtime for %s: %s" % (e, errno.errorcode[xte]))
elif self.need_sync(e, xte, xtr):
chld.append((e, xte))
def indulgently(e, fnc, blame=None):
if not blame:
blame = path
try:
return fnc(e)
except (IOError, OSError):
ex = sys.exc_info()[1]
if ex.errno == ENOENT:
logging.warn("salvaged ENOENT for " + e)
self.add_failjob(blame, 'by-indulgently')
return False
else:
raise
for e, xte in chld:
st = indulgently(e, lambda e: os.lstat(e))
if st == False:
continue
mo = st.st_mode
adct = {'own': (st.st_uid, st.st_gid)}
if stat.S_ISLNK(mo):
if indulgently(e, lambda e: self.slave.server.symlink(os.readlink(e), e)) == False:
continue
self.sendmark(e, xte, adct)
elif stat.S_ISREG(mo):
logging.debug("syncing %s ..." % e)
pb = self.syncer.add(e)
timeA=datetime.now()
def regjob(e, xte, pb):
if pb.wait():
logging.debug("synced " + e)
self.sendmark_regular(e, xte)
timeB=datetime.now()
self.lastSyncTime=timeB-timeA
self.syncTime=(self.syncTime+self.lastSyncTime.microseconds)/(10.0**6)
self.filesSynced=self.filesSynced+1
return True
else:
logging.warn("failed to sync " + e)
self.add_job(path, 'reg', regjob, e, xte, pb)
elif stat.S_ISDIR(mo):
adct['mode'] = mo
if indulgently(e, lambda e: (self.add_job(path, 'cwait', self.wait, e, xte, adct),
self.crawl(e, xte),
True)[-1], blame=e) == False:
continue
else:
# ignore fifos, sockets and special files
pass
if path == '.':
self.wait(path, xtl)
def main_i():
"""internal main routine
parse command line, decide what action will be taken;
we can either:
- query/manipulate configuration
- format gsyncd urls using gsyncd's url parsing engine
- start service in following modes, in given stages:
- monitor: startup(), monitor()
- master: startup(), connect_remote(), connect(), service_loop()
- slave: startup(), connect(), service_loop()
"""
rconf = {'go_daemon': 'should'}
def store_abs(opt, optstr, val, parser):
if val and val != '-':
val = os.path.abspath(val)
setattr(parser.values, opt.dest, val)
def store_local(opt, optstr, val, parser):
rconf[opt.dest] = val
def store_local_curry(val):
return lambda o, oo, vx, p: store_local(o, oo, val, p)
def store_local_obj(op, dmake):
return lambda o, oo, vx, p: store_local(o, oo, FreeObject(op=op, **dmake(vx)), p)
op = OptionParser(usage="%prog [options...] <master> <slave>", version="%prog 0.0.1")
op.add_option('--gluster-command-dir', metavar='DIR', default='')
op.add_option('--gluster-log-file', metavar='LOGF', default=os.devnull, type=str, action='callback', callback=store_abs)
op.add_option('--gluster-log-level', metavar='LVL')
op.add_option('--gluster-params', metavar='PRMS', default='')
op.add_option('--gluster-cli-options', metavar='OPTS', default='--log-file=-')
op.add_option('--mountbroker', metavar='LABEL')
op.add_option('-p', '--pid-file', metavar='PIDF', type=str, action='callback', callback=store_abs)
op.add_option('-l', '--log-file', metavar='LOGF', type=str, action='callback', callback=store_abs)
op.add_option('--state-file', metavar='STATF', type=str, action='callback', callback=store_abs)
op.add_option('--ignore-deletes', default=False, action='store_true')
op.add_option('-L', '--log-level', metavar='LVL')
op.add_option('-r', '--remote-gsyncd', metavar='CMD', default=os.path.abspath(sys.argv[0]))
op.add_option('--volume-id', metavar='UUID')
op.add_option('--session-owner', metavar='ID')
op.add_option('-s', '--ssh-command', metavar='CMD', default='ssh')
op.add_option('--rsync-command', metavar='CMD', default='rsync')
op.add_option('--timeout', metavar='SEC', type=int, default=120)
op.add_option('--connection-timeout', metavar='SEC', type=int, default=60, help=SUPPRESS_HELP)
op.add_option('--sync-jobs', metavar='N', type=int, default=3)
op.add_option('--turns', metavar='N', type=int, default=0, help=SUPPRESS_HELP)
op.add_option('--allow-network', metavar='IPS', default='')
op.add_option('-c', '--config-file', metavar='CONF', type=str, action='callback', callback=store_local)
# duh. need to specify dest or value will be mapped to None :S
op.add_option('--monitor', dest='monitor', action='callback', callback=store_local_curry(True))
op.add_option('--feedback-fd', dest='feedback_fd', type=int, help=SUPPRESS_HELP, action='callback', callback=store_local)
op.add_option('--listen', dest='listen', help=SUPPRESS_HELP, action='callback', callback=store_local_curry(True))
op.add_option('-N', '--no-daemon', dest="go_daemon", action='callback', callback=store_local_curry('dont'))
op.add_option('--debug', dest="go_daemon", action='callback', callback=lambda *a: (store_local_curry('dont')(*a),
setattr(a[-1].values, 'log_file', '-'),
setattr(a[-1].values, 'log_level', 'DEBUG'))),
for a in ('check', 'get'):
op.add_option('--config-' + a, metavar='OPT', type=str, dest='config', action='callback',
callback=store_local_obj(a, lambda vx: {'opt': vx}))
op.add_option('--config-get-all', dest='config', action='callback', callback=store_local_obj('get', lambda vx: {'opt': None}))
for m in ('', '-rx', '-glob'):
# call this code 'Pythonic' eh?
# have to define a one-shot local function to be able to inject (a value depending on the)
# iteration variable into the inner lambda
def conf_mod_opt_regex_variant(rx):
op.add_option('--config-set' + m, metavar='OPT VAL', type=str, nargs=2, dest='config', action='callback',
callback=store_local_obj('set', lambda vx: {'opt': vx[0], 'val': vx[1], 'rx': rx}))
op.add_option('--config-del' + m, metavar='OPT', type=str, dest='config', action='callback',
callback=store_local_obj('del', lambda vx: {'opt': vx, 'rx': rx}))
conf_mod_opt_regex_variant(m and m[1:] or False)
op.add_option('--normalize-url', dest='url_print', action='callback', callback=store_local_curry('normal'))
op.add_option('--canonicalize-url', dest='url_print', action='callback', callback=store_local_curry('canon'))
op.add_option('--canonicalize-escape-url', dest='url_print', action='callback', callback=store_local_curry('canon_esc'))
tunables = [ norm(o.get_opt_string()[2:]) for o in op.option_list if o.callback in (store_abs, 'store_true', None) and o.get_opt_string() not in ('--version', '--help') ]
remote_tunables = [ 'listen', 'go_daemon', 'timeout', 'session_owner', 'config_file' ]
rq_remote_tunables = { 'listen': True }
# precedence for sources of values: 1) commandline, 2) cfg file, 3) defaults
# -- for this to work out we need to tell apart defaults from explicitly set
# options... so churn out the defaults here and call the parser with virgin
# values container.
defaults = op.get_default_values()
opts, args = op.parse_args(values=optparse.Values())
confdata = rconf.get('config')
if not (len(args) == 2 or \
(len(args) == 1 and rconf.get('listen')) or \
(len(args) <= 2 and confdata) or \
rconf.get('url_print')):
sys.stderr.write("error: incorrect number of arguments\n\n")
sys.stderr.write(op.get_usage() + "\n")
sys.exit(1)
restricted = os.getenv('_GSYNCD_RESTRICTED_')
if restricted:
allopts = {}
allopts.update(opts.__dict__)
allopts.update(rconf)
bannedtuns = set(allopts.keys()) - set(remote_tunables)
if bannedtuns:
raise GsyncdError('following tunables cannot be set with restricted SSH invocaton: ' + \
', '.join(bannedtuns))
for k, v in rq_remote_tunables.items():
if not k in allopts or allopts[k] != v:
raise GsyncdError('tunable %s is not set to value %s required for restricted SSH invocaton' % \
(k, v))
confrx = getattr(confdata, 'rx', None)
if confrx:
# peers are regexen, don't try to parse them
if confrx == 'glob':
args = [ '\A' + fnmatch.translate(a) for a in args ]
canon_peers = args
namedict = {}
else:
rscs = [resource.parse_url(u) for u in args]
dc = rconf.get('url_print')
if dc:
for r in rscs:
print(r.get_url(**{'normal': {},
'canon': {'canonical': True},
'canon_esc': {'canonical': True, 'escaped': True}}[dc]))
return
local = remote = None
if rscs:
local = rscs[0]
if len(rscs) > 1:
remote = rscs[1]
if not local.can_connect_to(remote):
raise GsyncdError("%s cannot work with %s" % (local.path, remote and remote.path))
pa = ([], [], [])
urlprms = ({}, {'canonical': True}, {'canonical': True, 'escaped': True})
for x in rscs:
for i in range(len(pa)):
pa[i].append(x.get_url(**urlprms[i]))
peers, canon_peers, canon_esc_peers = pa
# creating the namedict, a dict representing various ways of referring to / repreenting
# peers to be fillable in config templates
mods = (lambda x: x, lambda x: x[0].upper() + x[1:], lambda x: 'e' + x[0].upper() + x[1:])
if remote:
rmap = { local: ('local', 'master'), remote: ('remote', 'slave') }
else:
rmap = { local: ('local', 'slave') }
namedict = {}
for i in range(len(rscs)):
x = rscs[i]
for name in rmap[x]:
for j in range(3):
namedict[mods[j](name)] = pa[j][i]
if x.scheme == 'gluster':
namedict[name + 'vol'] = x.volume
if not 'config_file' in rconf:
rconf['config_file'] = os.path.join(os.path.dirname(sys.argv[0]), "conf/gsyncd.conf")
gcnf = GConffile(rconf['config_file'], canon_peers, defaults.__dict__, opts.__dict__, namedict)
if confdata:
opt_ok = norm(confdata.opt) in tunables + [None]
if confdata.op == 'check':
if opt_ok:
sys.exit(0)
else:
sys.exit(1)
elif not opt_ok:
raise GsyncdError("not a valid option: " + confdata.opt)
if confdata.op == 'get':
gcnf.get(confdata.opt)
elif confdata.op == 'set':
gcnf.set(confdata.opt, confdata.val, confdata.rx)
elif confdata.op == 'del':
gcnf.delete(confdata.opt, confdata.rx)
return
gconf.__dict__.update(defaults.__dict__)
gcnf.update_to(gconf.__dict__)
gconf.__dict__.update(opts.__dict__)
gconf.configinterface = gcnf
if restricted and gconf.allow_network:
ssh_conn = os.getenv('SSH_CONNECTION')
if not ssh_conn:
#legacy env var
ssh_conn = os.getenv('SSH_CLIENT')
if ssh_conn:
allowed_networks = [ IPNetwork(a) for a in gconf.allow_network.split(',') ]
client_ip = IPAddress(ssh_conn.split()[0])
allowed = False
for nw in allowed_networks:
if client_ip in nw:
allowed = True
break
if not allowed:
raise GsyncdError("client IP address is not allowed")
ffd = rconf.get('feedback_fd')
if ffd:
fcntl.fcntl(ffd, fcntl.F_SETFD, fcntl.FD_CLOEXEC)
#normalize loglevel
lvl0 = gconf.log_level
if isinstance(lvl0, str):
lvl1 = lvl0.upper()
lvl2 = logging.getLevelName(lvl1)
# I have _never_ _ever_ seen such an utterly braindead
# error condition
if lvl2 == "Level " + lvl1:
raise GsyncdError('cannot recognize log level "%s"' % lvl0)
gconf.log_level = lvl2
go_daemon = rconf['go_daemon']
be_monitor = rconf.get('monitor')
if not be_monitor and isinstance(remote, resource.SSH) and \
go_daemon == 'should':
go_daemon = 'postconn'
log_file = None
else:
log_file = gconf.log_file
if be_monitor:
label = 'monitor'
elif remote:
#master
label = ''
else:
label = 'slave'
startup(go_daemon=go_daemon, log_file=log_file, label=label)
if be_monitor:
return monitor()
logging.info("syncing: %s" % " -> ".join(peers))
resource.Popen.init_errhandler()
if remote:
go_daemon = remote.connect_remote(go_daemon=go_daemon)
if go_daemon:
startup(go_daemon=go_daemon, log_file=gconf.log_file)
# complete remote connection in child
remote.connect_remote(go_daemon='done')
local.connect()
if ffd:
os.close(ffd)
local.service_loop(*[r for r in [remote] if r])
def inhibit(self, *a):
"""inhibit a gluster filesystem
Mount glusterfs over a temporary mountpoint,
change into the mount, and lazy unmount the
filesystem.
"""
mpi, mpo = os.pipe()
mh = Popen.fork()
if mh:
os.close(mpi)
fcntl.fcntl(mpo, fcntl.F_SETFD, fcntl.FD_CLOEXEC)
d = None
margv = self.make_mount_argv(*a)
if self.mntpt:
# mntpt is determined pre-mount
d = self.mntpt
os.write(mpo, d + '\0')
po = Popen(margv, **self.mountkw)
self.handle_mounter(po)
po.terminate_geterr()
logging.debug('auxiliary glusterfs mount in place')
if not d:
# mntpt is determined during mount
d = self.mntpt
os.write(mpo, d + '\0')
os.write(mpo, 'M')
t = syncdutils.Thread(target=lambda: os.chdir(d))
t.start()
tlim = gconf.starttime + int(gconf.connection_timeout)
while True:
if not t.isAlive():
break
if time.time() >= tlim:
syncdutils.finalize(exval=1)
time.sleep(1)
os.close(mpo)
_, rv = syncdutils.waitpid(mh, 0)
if rv:
rv = (os.WIFEXITED(rv) and os.WEXITSTATUS(rv) or 0) - \
(os.WIFSIGNALED(rv) and os.WTERMSIG(rv) or 0)
logging.warn('stale mount possibly left behind on ' + d)
raise GsyncdError("cleaning up temp mountpoint %s failed with status %d" % \
(d, rv))
else:
rv = 0
try:
os.setsid()
os.close(mpo)
mntdata = ''
while True:
c = os.read(mpi, 1)
if not c:
break
mntdata += c
if mntdata:
mounted = False
if mntdata[-1] == 'M':
mntdata = mntdata[:-1]
assert (mntdata)
mounted = True
assert (mntdata[-1] == '\0')
mntpt = mntdata[:-1]
assert (mntpt)
if mounted:
po = self.umount_l(mntpt)
po.terminate_geterr(fail_on_err=False)
if po.returncode != 0:
po.errlog()
rv = po.returncode
self.cleanup_mntpt(mntpt)
except:
logging.exception('mount cleanup failure:')
rv = 200
os._exit(rv)
logging.debug('auxiliary glusterfs mount prepared')
def __init__(self, path, pattern):
m = re.search(pattern, path)
if not m:
raise GsyncdError("malformed path")
self.path = path
return m.groups()
