import sys, getopt, types
import string, os, stat, popen2, socket, time, random, fcntl, select
-import re, exceptions
+import re, exceptions, signal
import xml.dom.minidom
if sys.version[0] == '1':
else:
from fcntl import F_GETFL, F_SETFL
+PYMOD_DIR = "/usr/lib/lustre/python"
+
+def development_mode():
+ base = os.path.dirname(sys.argv[0])
+ if os.access(base+"/Makefile.am", os.R_OK):
+ return 1
+ return 0
+
+if not development_mode():
+ sys.path.append(PYMOD_DIR)
+
+import Lustre
+
# Global parameters
MAXTCPBUF = 1048576
DEFAULT_TCPBUF = 1048576
if not first_cleanup_error:
first_cleanup_error = rc
-
-def usage():
- print """usage: lconf config.xml
-
-config.xml Lustre configuration in xml format.
---ldapurl LDAP server URL, eg. ldap://localhost
---config Cluster config name used for LDAP query
---node <nodename> Load config for <nodename>
---select service=nodeA,service2=nodeB U
--d | --cleanup Cleans up config. (Shutdown)
--f | --force Forced unmounting and/or obd detach during cleanup
--v | --verbose Print system commands as they are run
--h | --help Print this help
---gdb Prints message after creating gdb module script
- and sleeps for 5 seconds.
--n | --noexec Prints the commands and steps that will be run for a
- config without executing them. This can used to check if a
- config file is doing what it should be doing. (Implies -v)
---nomod Skip load/unload module step.
---nosetup Skip device setup/cleanup step.
---reformat Reformat all devices (without question)
---dump <file> Dump the kernel debug log before portals is unloaded
---minlevel <num> Specify the minimum level of services to configure/cleanup (default 0)
---maxlevel <num> Specify the maximum level of services to configure/cleanup (default 100)
- Levels are aproximatly like:
- 10 - network
- 20 - device, ldlm
- 30 - osd, mdd
- 40 - mds, ost
- 50 - mdc, osc
- 60 - lov
- 70 - mountpoint, echo_client
---lustre=src_dir Base directory of lustre sources. This parameter will cause lconf
- to load modules from a source tree.
---portals=src_dir Portals source directory. If this is a relative path, then it is
- assumed to be relative to lustre.
-
-"""
- TODO = """
---ldap server LDAP server with lustre config database
---makeldiff Translate xml source to LDIFF
-This are perhaps not needed:
-"""
- sys.exit()
-
-# ============================================================
-# Config parameters, encapsulated in a class
-class Config:
- def __init__(self):
- # flags
- self._noexec = 0
- self._verbose = 0
- self._reformat = 0
- self._cleanup = 0
- self._gdb = 0
- self._nomod = 0
- self._nosetup = 0
- self._force = 0
- # parameters
- self._modules = None
- self._node = None
- self._url = None
- self._gdb_script = '/tmp/ogdb'
- self._debug_path = '/tmp/lustre-log'
- self._dump_file = None
- self._lustre_dir = ''
- self._portals_dir = ''
- self._minlevel = 0
- self._maxlevel = 100
- self._timeout = 0
- self._recovery_upcall = ''
- self._ldapurl = ''
- self._config_name = ''
- self._select = {}
- self._lctl_dump = ''
-
- def verbose(self, flag = None):
- if flag: self._verbose = flag
- return self._verbose
-
- def noexec(self, flag = None):
- if flag: self._noexec = flag
- return self._noexec
-
- def reformat(self, flag = None):
- if flag: self._reformat = flag
- return self._reformat
-
- def cleanup(self, flag = None):
- if flag: self._cleanup = flag
- return self._cleanup
-
- def gdb(self, flag = None):
- if flag: self._gdb = flag
- return self._gdb
-
- def nomod(self, flag = None):
- if flag: self._nomod = flag
- return self._nomod
-
- def nosetup(self, flag = None):
- if flag: self._nosetup = flag
- return self._nosetup
-
- def force(self, flag = None):
- if flag: self._force = flag
- return self._force
-
- def node(self, val = None):
- if val: self._node = val
- return self._node
-
- def gdb_script(self):
- if os.path.isdir('/r'):
- return '/r' + self._gdb_script
- else:
- return self._gdb_script
-
- def debug_path(self):
- if os.path.isdir('/r'):
- return '/r' + self._debug_path
- else:
- return self._debug_path
-
- def dump_file(self, val = None):
- if val: self._dump_file = val
- return self._dump_file
- def minlevel(self, val = None):
- if val: self._minlevel = int(val)
- return self._minlevel
-
- def maxlevel(self, val = None):
- if val: self._maxlevel = int(val)
- return self._maxlevel
-
- def portals_dir(self, val = None):
- if val: self._portals_dir = val
- return self._portals_dir
-
- def lustre_dir(self, val = None):
- if val: self._lustre_dir = val
- return self._lustre_dir
-
- def timeout(self, val = None):
- if val: self._timeout = val
- return self._timeout
-
- def recovery_upcall(self, val = None):
- if val: self._recovery_upcall = val
- return self._recovery_upcall
-
- def ldapurl(self, val = None):
- if val: self._ldapurl = val
- return self._ldapurl
-
- def config_name(self, val = None):
- if val: self._config_name = val
- return self._config_name
-
- def init_select(self, arg):
- # arg = "service=nodeA,service2=nodeB"
- list = string.split(arg, ',')
- for entry in list:
- srv, node = string.split(entry, '=')
- self._select[srv] = node
-
- def select(self, srv):
- if self._select.has_key(srv):
- return self._select[srv]
- return None
-
- def lctl_dump(self, val = None):
- if val: self._lctl_dump = val
- return self._lctl_dump
-
-
-config = Config()
-
# ============================================================
# debugging and error funcs
def fixme(msg = "this feature"):
- raise LconfError, msg + ' not implmemented yet.'
+ raise Lustre.LconfError, msg + ' not implmemented yet.'
def panic(*args):
msg = string.join(map(str,args))
- if not config.noexec():
- raise LconfError(msg)
+ if not config.noexec:
+ raise Lustre.LconfError(msg)
else:
print "! " + msg
print string.strip(s)
def debug(*args):
- if config.verbose():
+ if config.verbose:
msg = string.join(map(str,args))
print msg
+
# ============================================================
# locally defined exceptions
class CommandError (exceptions.Exception):
else:
print self.cmd_err
-class LconfError (exceptions.Exception):
- def __init__(self, args):
- self.args = args
-
# ============================================================
# handle daemons, like the acceptor
self.lctl = find_prog(cmd)
self.save_file = ''
if not self.lctl:
- if config.noexec():
+ if config.noexec:
debug('! lctl not found')
self.lctl = 'lctl'
else:
cmds = '\n dump ' + self.save_file + cmds
debug("+", cmd_line, cmds)
- if config.noexec(): return (0, [])
+ if config.noexec: return (0, [])
child = popen2.Popen3(cmd_line, 1) # Capture stdout and stderr from command
child.tochild.write(cmds + "\n")
# create a new connection
def connect(self, srv):
cmds = "\n add_uuid %s %s %s" % (srv.uuid, srv.nid, srv.net_type)
- if srv.net_type in ('tcp', 'toe') and not config.lctl_dump():
+ if srv.net_type in ('tcp', 'toe') and not config.lctl_dump:
flags = ''
if srv.irq_affinity:
flags = flags + 'i'
cmds = cmds + "\n quit"
self.run(cmds)
+
+ # Recover a device
+ def recover(self, dev_uuid, new_conn):
+ cmds = """
+ device %%%s
+ probe
+ recover %s""" %(dev_uuid, new_conn)
+ self.run(cmds)
# add a route to a range
def add_route(self, net, gw, lo, hi):
device $%s
cleanup %s
detach
- quit""" % (name, ('', 'force')[config.force()])
+ quit""" % (name, ('', 'force')[config.force])
self.run(cmds)
# create an lov
- def lov_setconfig(self, uuid, mdsuuid, stripe_cnt, stripe_sz, stripe_off, pattern, devlist):
+ def lov_setconfig(self, uuid, mdsuuid, stripe_cnt, stripe_sz, stripe_off,
+ pattern, devlist):
cmds = """
device $%s
probe
# save it if necessary
def runcmd(cmd):
debug ("+", cmd)
- if config.noexec(): return (0, [])
+ if config.noexec: return (0, [])
f = os.popen(cmd + ' 2>&1')
out = f.readlines()
ret = f.close()
def run_daemon(*args):
cmd = string.join(map(str,args))
debug ("+", cmd)
- if config.noexec(): return 0
+ if config.noexec: return 0
f = os.popen(cmd + ' 2>&1')
ret = f.close()
if ret:
syspath = string.split(os.environ['PATH'], ':')
cmdpath = os.path.dirname(sys.argv[0])
syspath.insert(0, cmdpath);
- if config.portals_dir():
- syspath.insert(0, os.path.join(config.portals_dir()+'/linux/utils/'))
+ if config.portals:
+ syspath.insert(0, os.path.join(config.portals, 'linux/utils/'))
for d in syspath:
prog = os.path.join(d,cmd)
if os.access(prog, os.X_OK):
# build fs according to type
# fixme: dangerous
-def mkfs(dev, devsize, fstype):
+def mkfs(dev, devsize, fstype,jsize):
block_cnt = ''
+ jopt = ''
if devsize:
# devsize is in 1k, and fs block count is in 4k
block_cnt = devsize/4
if(fstype in ('ext3', 'extN')):
+ # ext3 journal size is in megabytes
+ if jsize: jopt = "-J size=%d" %(jsize,)
mkfs = 'mkfs.ext2 -j -b 4096 -F '
elif (fstype == 'reiserfs'):
+ # reiserfs journal size is in blocks
+ if jsize: jopt = "--journal_size %d" %(jsize,)
mkfs = 'mkreiserfs -ff'
else:
print 'unsupported fs type: ', fstype
- (ret, out) = run (mkfs, dev, block_cnt)
+ (ret, out) = run (mkfs, jopt, dev, block_cnt)
if ret:
panic("Unable to build fs:", dev)
# enable hash tree indexing on fsswe
if dev:
print 'WARNING file:', file, 'already mapped to', dev
return dev
- if config.reformat() or not os.access(file, os.R_OK | os.W_OK):
+ if config.reformat or not os.access(file, os.R_OK | os.W_OK):
if size < 8000:
panic(file, "size must be larger than 8MB, currently set to:", size)
(ret, out) = run("dd if=/dev/zero bs=1k count=0 seek=%d of=%s" %(size,
return 0
# initialize a block device if needed
-def block_dev(dev, size, fstype, format):
- if config.noexec(): return dev
+def block_dev(dev, size, fstype, format, journal_size):
+ if config.noexec: return dev
if not is_block(dev):
dev = init_loop(dev, size, fstype)
- if config.reformat() or (need_format(fstype, dev) and format == 'yes'):
- mkfs(dev, size, fstype)
+ if config.reformat or (need_format(fstype, dev) and format == 'yes'):
+ mkfs(dev, size, fstype, journal_size)
# else:
# panic("device:", dev,
return ip
def get_local_nid(net_type, wildcard):
- """Return the local nid. First look for an elan interface,
- then use the local address. """
+ """Return the local nid."""
local = ""
if os.access('/proc/elan/device0/position', os.R_OK):
local = get_local_address('elan', '*')
def is_prepared(uuid):
"""Return true if a device exists for the uuid"""
- # expect this format:
- # 1 UP ldlm ldlm ldlm_UUID 2
- if config.lctl_dump():
+ if config.lctl_dump:
return 0
+ if config.noexec and config.cleanup:
+ return 1
try:
+ # expect this format:
+ # 1 UP ldlm ldlm ldlm_UUID 2
out = lctl.device_list()
for s in out:
if uuid == string.split(s)[4]:
return 0
def is_network_prepared():
- """If the PTLRPC device exists, then assumet that all networking
+ """If the LDLM device exists, then assume that all networking
has been configured"""
- if config.lctl_dump():
- return 0
- try:
- out = lctl.device_list()
- for s in out:
- if 'RPCDEV_UUID' == string.split(s)[4]:
- return 1
- except CommandError, e:
- e.dump()
- return 0
-
+ return is_prepared('ldlm_UUID')
def fs_is_mounted(path):
"""Return true if path is a mounted lustre filesystem"""
def add_portals_module(self, dev_dir, modname):
"""Append a module to list of modules to load."""
- self.kmodule_list.append((config.portals_dir(), dev_dir, modname))
+ self.kmodule_list.append((config.portals, dev_dir, modname))
def add_lustre_module(self, dev_dir, modname):
"""Append a module to list of modules to load."""
- self.kmodule_list.append((config.lustre_dir(), dev_dir, modname))
+ self.kmodule_list.append((config.lustre, dev_dir, modname))
def mod_loaded(self, modname):
"""Check if a module is already loaded. Look in /proc/modules for it."""
"""Load all the modules in the list in the order they appear."""
for src_dir, dev_dir, mod in self.kmodule_list:
# (rc, out) = run ('/sbin/lsmod | grep -s', mod)
- if self.mod_loaded(mod) and not config.noexec():
+ if self.mod_loaded(mod) and not config.noexec:
continue
log ('loading module:', mod)
if src_dir:
rev = self.kmodule_list
rev.reverse()
for src_dir, dev_dir, mod in rev:
- if not self.mod_loaded(mod):
+ if not self.mod_loaded(mod) and not config.noexec:
continue
# debug hack
- if mod == 'portals' and config.dump_file():
- lctl.dump(config.dump_file())
+ if mod == 'portals' and config.dump:
+ lctl.dump(config.dump)
log('unloading module:', mod)
- if config.noexec():
- continue
(rc, out) = run('/sbin/rmmod', mod)
if rc:
log('! unable to unload module:', mod)
self.nid_exchange = self.db.get_val_int('nidexchange', 0)
if '*' in self.nid:
- self.nid = get_local_nid(self.net_type, self.nid)
+ if self.nid_exchange:
+ self.nid = get_local_nid(self.net_type, self.nid)
+ else:
+ self.nid = get_local_address(self.net_type, self.nid)
if not self.nid:
panic("unable to set nid for", self.net_type, self.nid)
debug("nid:", self.nid)
if self.net_type == 'gm':
self.add_portals_module("/linux/gmnal", 'kgmnal')
self.add_lustre_module('obdclass', 'obdclass')
+ self.add_lustre_module('ptlrpc', 'ptlrpc')
def prepare(self):
if is_network_prepared():
if is_prepared(self.uuid):
Module.cleanup(self)
-class PTLRPC(Module):
- def __init__(self,db):
- Module.__init__(self, 'PTLRPC', db)
- self.add_lustre_module('ptlrpc', 'ptlrpc')
- def prepare(self):
- if is_prepared(self.uuid):
- return
- self.info()
- lctl.newdev(attach="ptlrpc %s %s" % (self.name, self.uuid))
- def cleanup(self):
- if is_prepared(self.uuid):
- Module.cleanup(self)
-
class LOV(Module):
def __init__(self,db):
Module.__init__(self, 'LOV', db)
Module.__init__(self, 'MDSDEV', db)
self.devpath = self.db.get_val('devpath','')
self.size = self.db.get_val_int('devsize', 0)
+ self.journal_size = self.db.get_val_int('journalsize', 0)
self.fstype = self.db.get_val('fstype', '')
# overwrite the orignal MDSDEV name and uuid with the MDS name and uuid
target_uuid = self.db.get_first_ref('target')
self.lovconfig_uuids = mds.get_refs('lovconfig')
# FIXME: if fstype not set, then determine based on kernel version
self.format = self.db.get_val('autoformat', "no")
+ if mds.get_val('failover', 0):
+ self.failover_mds = 'f'
+ else:
+ self.failover_mds = ''
- active_uuid = mds.get_active_target()
+ active_uuid = get_active_target(mds)
if not active_uuid:
panic("No target device found:", target_uuid)
if active_uuid == self.uuid:
return
self.info(self.devpath, self.fstype, self.format)
run_acceptors()
- blkdev = block_dev(self.devpath, self.size, self.fstype, self.format)
+ blkdev = block_dev(self.devpath, self.size, self.fstype, self.format,
+ self.journal_size)
if not is_prepared('MDT_UUID'):
lctl.newdev(attach="mdt %s %s" % ('MDT', 'MDT_UUID'),
setup ="")
self.osdtype = self.db.get_val('osdtype')
self.devpath = self.db.get_val('devpath', '')
self.size = self.db.get_val_int('devsize', 0)
+ self.journal_size = self.db.get_val_int('journalsize', 0)
self.fstype = self.db.get_val('fstype', '')
target_uuid = self.db.get_first_ref('target')
ost = self.db.lookup(target_uuid)
self.format = self.db.get_val('autoformat', 'yes')
if self.fstype == 'extN':
self.add_lustre_module('extN', 'extN')
+ if ost.get_val('failover', 0):
+ self.failover_ost = 'f'
+ else:
+ self.failover_ost = ''
- active_uuid = ost.get_active_target()
+ active_uuid = get_active_target(ost)
if not active_uuid:
panic("No target device found:", target_uuid)
if active_uuid == self.uuid:
if not self.active:
debug(self.uuid, "not active")
return
- self.info(self.osdtype, self.devpath, self.size, self.fstype, self.format)
+ self.info(self.osdtype, self.devpath, self.size, self.fstype,
+ self.format, self.journal_size)
run_acceptors()
if self.osdtype == 'obdecho':
blkdev = ''
else:
- blkdev = block_dev(self.devpath, self.size, self.fstype, self.format)
+ blkdev = block_dev(self.devpath, self.size, self.fstype,
+ self.format, self.journal_size)
lctl.newdev(attach="%s %s %s" % (self.osdtype, self.name, self.uuid),
- setup ="%s %s" %(blkdev, self.fstype))
+ setup ="%s %s %s" %(blkdev, self.fstype,
+ self.failover_ost))
if not is_prepared('OSS_UUID'):
lctl.newdev(attach="ost %s %s" % ('OSS', 'OSS_UUID'),
setup ="")
self.target_uuid = tgtdb.getUUID()
self.db = tgtdb
- self.tgt_dev_uuid = tgtdb.get_active_target()
+ self.tgt_dev_uuid = get_active_target(tgtdb)
if not self.tgt_dev_uuid:
panic("No target device found for target:", self.target_name)
self.module_name = string.upper(module)
self.name = '%s_%s_%s' % (self.module_name, owner, self.target_name)
self.uuid = '%05x%05x_%.14s_%05x%05x' % (int(random.random() * 1048576),
- int(random.random() * 1048576),self.name,
+ int(random.random() * 1048576),
+ self.name,
int(random.random() * 1048576),
int(random.random() * 1048576))
self.uuid = self.uuid[0:36]
def lookup_server(self, srv_uuid):
""" Lookup a server's network information """
- self._server_nets = self.db.get_ost_net(srv_uuid)
+ self._server_nets = get_ost_net(self.db, srv_uuid)
if len(self._server_nets) == 0:
panic ("Unable to find a server for:", srv_uuid)
else:
mdc_uuid = self.vosc.get_mdc_uuid()
if not mdc_uuid:
+ self.vosc.cleanup()
panic("Unable to determine MDC UUID. Probably need to cleanup before re-mounting.")
self.info(self.path, self.mds_uuid, self.obd_uuid)
if config.lctl_dump:
run("mkdir", self.path)
ret, val = run(cmd)
if ret:
+ self.vosc.cleanup()
+ if self.vosc.need_mdc():
+ cleanup_mdc(self.db, self.name, self.mds_uuid)
panic("mount failed:", self.path)
def cleanup(self):
self.info(self.path, self.mds_uuid,self.obd_uuid)
if fs_is_mounted(self.path):
- if config.force():
+ if config.force:
(rc, out) = run("umount", "-f", self.path)
else:
(rc, out) = run("umount", self.path)
# ============================================================
# XML processing and query
-class LustreDB:
- def lookup(self, uuid):
- """ lookup returns a new LustreDB instance"""
- return self._lookup_by_uuid(uuid)
-
- def lookup_name(self, name, class_name = ""):
- """ lookup returns a new LustreDB instance"""
- return self._lookup_by_name(name, class_name)
-
- def lookup_class(self, class_name):
- """ lookup returns a new LustreDB instance"""
- return self._lookup_by_class(class_name)
-
- def get_val(self, tag, default=None):
- v = self._get_val(tag)
- if v:
- return v
- if default != None:
- return default
- debug("LustreDB", self.getName(), " no value for:", tag)
- return None
-
- def get_class(self):
- return self._get_class()
-
- def get_val_int(self, tag, default=0):
- str = self._get_val(tag)
- try:
- if str:
- return int(str)
- return default
- except ValueError:
- panic("text value is not integer:", str)
-
- def get_first_ref(self, tag):
- """ Get the first uuidref of the type TAG. Only
- one is expected. Returns the uuid."""
- uuids = self._get_refs(tag)
- if len(uuids) > 0:
- return uuids[0]
- return None
-
- def get_refs(self, tag):
- """ Get all the refs of type TAG. Returns list of uuids. """
- uuids = self._get_refs(tag)
- return uuids
-
- def get_all_refs(self):
- """ Get all the refs. Returns list of uuids. """
- uuids = self._get_all_refs()
- return uuids
-
- def get_ost_net(self, osd_uuid):
- srv_list = []
- if not osd_uuid:
- return srv_list
- osd = self.lookup(osd_uuid)
- node_uuid = osd.get_first_ref('node')
- node = self.lookup(node_uuid)
- if not node:
- panic("unable to find node for osd_uuid:", osd_uuid,
- " node_ref:", node_uuid)
- for net_uuid in node.get_networks():
- db = node.lookup(net_uuid)
- srv_list.append(Network(db))
+def get_ost_net(self, osd_uuid):
+ srv_list = []
+ if not osd_uuid:
return srv_list
+ osd = self.lookup(osd_uuid)
+ node_uuid = osd.get_first_ref('node')
+ node = self.lookup(node_uuid)
+ if not node:
+ panic("unable to find node for osd_uuid:", osd_uuid,
+ " node_ref:", node_uuid)
+ for net_uuid in node.get_networks():
+ db = node.lookup(net_uuid)
+ srv_list.append(Network(db))
+ return srv_list
+
+
+# the order of iniitailization is based on level.
+def getServiceLevel(self):
+ type = self.get_class()
+ ret=0;
+ if type in ('network',):
+ ret = 5
+ elif type in ('routetbl',):
+ ret = 6
+ elif type in ('device', 'ldlm'):
+ ret = 20
+ elif type in ('osd', 'mdd', 'cobd'):
+ ret = 30
+ elif type in ('mdsdev','ost'):
+ ret = 40
+ elif type in ('mdc','osc'):
+ ret = 50
+ elif type in ('lov',):
+ ret = 60
+ elif type in ('mountpoint', 'echoclient'):
+ ret = 70
+
+ if ret < config.minlevel or ret > config.maxlevel:
+ ret = 0
+ return ret
- def nid2server(self, nid, net_type):
- netlist = self.lookup_class('network')
- for net_db in netlist:
- if net_db.get_val('nid') == nid and net_db.get_val('nettype') == net_type:
- return net_db
- return None
-
- # the tag name is the service type
- # fixme: this should do some checks to make sure the dom_node is a service
- #
- # determine what "level" a particular node is at.
-
- # the order of iniitailization is based on level.
- def getServiceLevel(self):
- type = self.get_class()
- ret=0;
- if type in ('network',):
- ret = 5
- elif type in ('routetbl',):
- ret = 6
- elif type in ('ptlrpc',):
- ret = 7
- elif type in ('device', 'ldlm'):
- ret = 20
- elif type in ('osd', 'mdd', 'cobd'):
- ret = 30
- elif type in ('mdsdev','ost'):
- ret = 40
- elif type in ('mdc','osc'):
- ret = 50
- elif type in ('lov',):
- ret = 60
- elif type in ('mountpoint', 'echoclient'):
- ret = 70
-
- if ret < config.minlevel() or ret > config.maxlevel():
- ret = 0
- return ret
-
- #
- # return list of services in a profile. list is a list of tuples
- # [(level, db_object),]
- def getServices(self):
- list = []
- for ref_class, ref_uuid in self.get_all_refs():
- servdb = self.lookup(ref_uuid)
- if servdb:
- level = servdb.getServiceLevel()
- if level > 0:
- list.append((level, servdb))
- else:
- panic('service not found: ' + ref_uuid)
-
- list.sort()
- return list
-
- # Find the target_device for target on a node
- # node->profiles->device_refs->target
- def get_target_device(self, target_uuid, node_name):
- node_db = self.lookup_name(node_name)
- if not node_db:
- return None
- prof_list = node_db.get_refs('profile')
- for prof_uuid in prof_list:
- prof_db = node_db.lookup(prof_uuid)
- ref_list = prof_db.get_all_refs()
- for ref in ref_list:
- dev = self.lookup(ref[1])
- if dev and dev.get_first_ref('target') == target_uuid:
- return ref[1]
- return None
-
- def get_active_target(self):
- target_uuid = self.getUUID()
- target_name = self.getName()
- node_name = config.select(target_name)
- if node_name:
- tgt_dev_uuid = self.get_target_device(target_uuid, node_name)
- else:
- tgt_dev_uuid = self.get_first_ref('active')
- return tgt_dev_uuid
-
-
- # get all network uuids for this node
- def get_networks(self):
- ret = []
- prof_list = self.get_refs('profile')
- for prof_uuid in prof_list:
- prof_db = self.lookup(prof_uuid)
- net_list = prof_db.get_refs('network')
- #debug("get_networks():", prof_uuid, net_list)
- for net_uuid in net_list:
- ret.append(net_uuid)
- return ret
-
-class LustreDB_XML(LustreDB):
- def __init__(self, dom, root_node):
- # init xmlfile
- self.dom_node = dom
- self.root_node = root_node
-
- def xmltext(self, dom_node, tag):
- list = dom_node.getElementsByTagName(tag)
- if len(list) > 0:
- dom_node = list[0]
- dom_node.normalize()
- if dom_node.firstChild:
- txt = string.strip(dom_node.firstChild.data)
- if txt:
- return txt
-
- def xmlattr(self, dom_node, attr):
- return dom_node.getAttribute(attr)
-
- def _get_val(self, tag):
- """a value could be an attribute of the current node
- or the text value in a child node"""
- ret = self.xmlattr(self.dom_node, tag)
- if not ret:
- ret = self.xmltext(self.dom_node, tag)
- return ret
-
- def _get_class(self):
- return self.dom_node.nodeName
-
- #
- # [(ref_class, ref_uuid),]
- def _get_all_refs(self):
- list = []
- for n in self.dom_node.childNodes:
- if n.nodeType == n.ELEMENT_NODE:
- ref_uuid = self.xml_get_ref(n)
- ref_class = n.nodeName
- list.append((ref_class, ref_uuid))
-
- list.sort()
- return list
-
- def _get_refs(self, tag):
- """ Get all the refs of type TAG. Returns list of uuids. """
- uuids = []
- refname = '%s_ref' % tag
- reflist = self.dom_node.getElementsByTagName(refname)
- for r in reflist:
- uuids.append(self.xml_get_ref(r))
- return uuids
-
- def xmllookup_by_uuid(self, dom_node, uuid):
- for n in dom_node.childNodes:
- if n.nodeType == n.ELEMENT_NODE:
- if self.xml_get_uuid(n) == uuid:
- return n
- else:
- n = self.xmllookup_by_uuid(n, uuid)
- if n: return n
- return None
-
- def _lookup_by_uuid(self, uuid):
- dom = self. xmllookup_by_uuid(self.root_node, uuid)
- if dom:
- return LustreDB_XML(dom, self.root_node)
-
- def xmllookup_by_name(self, dom_node, name):
- for n in dom_node.childNodes:
- if n.nodeType == n.ELEMENT_NODE:
- if self.xml_get_name(n) == name:
- return n
- else:
- n = self.xmllookup_by_name(n, name)
- if n: return n
- return None
-
- def _lookup_by_name(self, name, class_name):
- dom = self.xmllookup_by_name(self.root_node, name)
- if dom:
- return LustreDB_XML(dom, self.root_node)
-
- def xmllookup_by_class(self, dom_node, class_name):
- return dom_node.getElementsByTagName(class_name)
-
- def _lookup_by_class(self, class_name):
- ret = []
- domlist = self.xmllookup_by_class(self.root_node, class_name)
- for node in domlist:
- ret.append(LustreDB_XML(node, self.root_node))
- return ret
-
- def xml_get_name(self, n):
- return n.getAttribute('name')
-
- def getName(self):
- return self.xml_get_name(self.dom_node)
-
- def xml_get_ref(self, n):
- return n.getAttribute('uuidref')
-
- def xml_get_uuid(self, dom_node):
- return dom_node.getAttribute('uuid')
-
- def getUUID(self):
- return self.xml_get_uuid(self.dom_node)
-
- def get_routes(self, type, gw):
- """ Return the routes as a list of tuples of the form:
- [(type, gw, lo, hi),]"""
- res = []
- tbl = self.dom_node.getElementsByTagName('routetbl')
- for t in tbl:
- routes = t.getElementsByTagName('route')
- for r in routes:
- net_type = self.xmlattr(r, 'type')
- if type != net_type:
- lo = self.xmlattr(r, 'lo')
- hi = self.xmlattr(r, 'hi')
- res.append((type, gw, lo, hi))
- return res
-
- def get_route_tbl(self):
- ret = []
- for r in self.dom_node.getElementsByTagName('route'):
- net_type = self.xmlattr(r, 'type')
- gw = self.xmlattr(r, 'gw')
- lo = self.xmlattr(r, 'lo')
- hi = self.xmlattr(r, 'hi')
- ret.append((net_type, gw, lo, hi))
- return ret
-
-
-# ================================================================
-# LDAP Support
-class LustreDB_LDAP(LustreDB):
- def __init__(self, name, attrs,
- base = "fs=lustre",
- parent = None,
- url = "ldap://localhost",
- user = "cn=Manager, fs=lustre",
- pw = "secret"
- ):
- self._name = name
- self._attrs = attrs
- self._base = base
- self._parent = parent
- self._url = url
- self._user = user
- self._pw = pw
- if parent:
- self.l = parent.l
- self._base = parent._base
- else:
- self.open()
-
- def open(self):
- import ldap
- try:
- self.l = ldap.initialize(self._url)
- # Set LDAP protocol version used
- self.l.protocol_version=ldap.VERSION3
- # user and pw only needed if modifying db
- self.l.bind_s("", "", ldap.AUTH_SIMPLE);
- except ldap.LDAPError, e:
- panic(e)
- # FIXME, do something useful here
-
- def close(self):
- self.l.unbind_s()
-
- def ldap_search(self, filter):
- """Return list of uuids matching the filter."""
- import ldap
- dn = self._base
- ret = []
- uuids = []
- try:
- for name, attrs in self.l.search_s(dn, ldap.SCOPE_ONELEVEL,
- filter, ["uuid"]):
- for v in attrs['uuid']:
- uuids.append(v)
- except ldap.NO_SUCH_OBJECT, e:
- pass
- except ldap.LDAPError, e:
- print e # FIXME: die here?
- if len(uuids) > 0:
- for uuid in uuids:
- ret.append(self._lookup_by_uuid(uuid))
- return ret
-
- def _lookup_by_name(self, name, class_name):
- list = self.ldap_search("lustreName=%s" %(name))
- if len(list) == 1:
- return list[0]
- return []
-
- def _lookup_by_class(self, class_name):
- return self.ldap_search("objectclass=%s" %(string.upper(class_name)))
-
- def _lookup_by_uuid(self, uuid):
- import ldap
- dn = "uuid=%s,%s" % (uuid, self._base)
- ret = None
- try:
- for name, attrs in self.l.search_s(dn, ldap.SCOPE_BASE,
- "objectclass=*"):
- ret = LustreDB_LDAP(name, attrs, parent = self)
-
- except ldap.NO_SUCH_OBJECT, e:
- debug("NO_SUCH_OBJECT:", uuid)
- pass # just return empty list
- except ldap.LDAPError, e:
- print e # FIXME: die here?
- return ret
-
-
- def _get_val(self, k):
- ret = None
- if self._attrs.has_key(k):
- v = self._attrs[k]
- if type(v) == types.ListType:
- ret = str(v[0])
+#
+# return list of services in a profile. list is a list of tuples
+# [(level, db_object),]
+def getServices(self):
+ list = []
+ for ref_class, ref_uuid in self.get_all_refs():
+ servdb = self.lookup(ref_uuid)
+ if servdb:
+ level = getServiceLevel(servdb)
+ if level > 0:
+ list.append((level, servdb))
else:
- ret = str(v)
- return ret
-
- def _get_class(self):
- return string.lower(self._attrs['objectClass'][0])
-
- #
- # [(ref_class, ref_uuid),]
- def _get_all_refs(self):
- list = []
- for k in self._attrs.keys():
- if re.search('.*Ref', k):
- for uuid in self._attrs[k]:
- list.append((k, uuid))
- return list
+ panic('service not found: ' + ref_uuid)
- def _get_refs(self, tag):
- """ Get all the refs of type TAG. Returns list of uuids. """
- uuids = []
- refname = '%sRef' % tag
- if self._attrs.has_key(refname):
- return self._attrs[refname]
- return []
+ list.sort()
+ return list
- def getName(self):
- return self._get_val('lustreName')
-
- def getUUID(self):
- return self._get_val('uuid')
-
- def get_route_tbl(self):
- return []
############################################################
# MDC UUID hack -
return srv, r
return None,None
+def get_active_target(db):
+ target_uuid = db.getUUID()
+ target_name = db.getName()
+ node_name = get_select(target_name)
+ if node_name:
+ tgt_dev_uuid = db.get_target_device(target_uuid, node_name)
+ else:
+ tgt_dev_uuid = db.get_first_ref('active')
+ return tgt_dev_uuid
+
############################################################
# lconf level logic
n = None
if type == 'ldlm':
n = LDLM(db)
- elif type == 'ptlrpc':
- n = PTLRPC(db)
elif type == 'lov':
n = LOV(db)
elif type == 'network':
prof_db = db.lookup(prof_uuid)
if not prof_db:
panic("profile:", profile, "not found.")
- services = prof_db.getServices()
+ services = getServices(prof_db)
operation(services)
def doSetup(services):
- if config.nosetup():
+ if config.nosetup:
return
for s in services:
n = newService(s[1])
n.prepare()
def doModules(services):
- if config.nomod():
+ if config.nomod:
return
for s in services:
n = newService(s[1])
n.load_module()
def doCleanup(services):
- if config.nosetup():
+ if config.nosetup:
return
services.reverse()
for s in services:
n.cleanup()
def doUnloadModules(services):
- if config.nomod():
+ if config.nomod:
return
services.reverse()
for s in services:
return
router_flag = node_db.get_val_int('router', 0)
- recovery_upcall = node_db.get_val('recovery_upcall', '')
+ recovery_upcall = node_db.get_val('recoveryUpcall', '')
timeout = node_db.get_val_int('timeout', 0)
add_local_interfaces(node_db)
# if not cleaning, load modules first.
prof_list = node_db.get_refs('profile')
- if config.cleanup():
- if config.force():
+ if config.recover:
+ if not (config.tgt_uuid or config.client_uuid or config.conn_uuid):
+ panic( "--recovery requires the failed uuids")
+ doRecovery(lustreDB, lctl, config.tgt_uuid, config.client_uuid,
+ config.conn_uuid)
+ elif config.cleanup:
+ if config.force:
# the command line can override this value
timeout = 5
# ugly hack, only need to run lctl commands for --dump
- if config.lctl_dump():
+ if config.lctl_dump:
for_each_profile(node_db, prof_list, doCleanup)
return
sys_set_timeout(timeout)
+ sys_set_ptldebug()
sys_set_recovery_upcall(recovery_upcall)
for_each_profile(node_db, prof_list, doCleanup)
else:
# ugly hack, only need to run lctl commands for --dump
- if config.lctl_dump():
+ if config.lctl_dump:
for_each_profile(node_db, prof_list, doSetup)
return
+ sys_make_devices()
+ sys_set_netmem_max('/proc/sys/net/core/rmem_max', MAXTCPBUF)
+ sys_set_netmem_max('/proc/sys/net/core/wmem_max', MAXTCPBUF)
+
for_each_profile(node_db, prof_list, doModules)
sys_set_debug_path()
- script = config.gdb_script()
+ script = config.gdb_script
run(lctl.lctl, ' modules >', script)
- if config.gdb():
+ if config.gdb:
log ("The GDB module script is in", script)
# pause, so user has time to break and
# load the script
for_each_profile(node_db, prof_list, doSetup)
-############################################################
-# Command line processing
-#
-def parse_cmdline(argv):
- short_opts = "hdnvf"
- long_opts = ["ldap", "reformat", "lustre=", "verbose", "gdb",
- "portals=", "makeldiff", "cleanup", "noexec",
- "help", "node=", "nomod", "nosetup",
- "dump=", "force", "minlevel=", "maxlevel=",
- "timeout=", "recovery_upcall=",
- "ldapurl=", "config=", "select=", "lctl_dump="]
- opts = []
- args = []
+def doRecovery(db, lctl, tgt_uuid, client_uuid, conn_uuid):
+ tgt = db.lookup(tgt_uuid)
+ new_uuid = get_active_target(tgt)
+ net = local_net(get_ost_net(db, new_uuid))
+ if not net:
+ panic("Unable to find a connection to:", new_uuid)
+ lctl.connect(net)
+ lctl.recover(client_uuid, net.uuid)
- try:
- opts, args = getopt.getopt(argv, short_opts, long_opts)
- except getopt.error:
- print "invalid opt"
- usage()
-
- for o, a in opts:
- if o in ("-h", "--help"):
- usage()
- if o in ("-d","--cleanup"):
- config.cleanup(1)
- if o in ("-v", "--verbose"):
- config.verbose(1)
- if o in ("-n", "--noexec"):
- config.noexec(1)
- if o == "--portals":
- config.portals_dir(a)
- if o == "--lustre":
- config.lustre_dir(a)
- if o == "--reformat":
- config.reformat(1)
- if o == "--node":
- config.node(a)
- if o == "--gdb":
- config.gdb(1)
- if o == "--nomod":
- config.nomod(1)
- if o == "--nosetup":
- config.nosetup(1)
- if o == "--dump":
- config.dump_file(a)
- if o in ("-f", "--force"):
- config.force(1)
- if o == "--minlevel":
- config.minlevel(a)
- if o == "--maxlevel":
- config.maxlevel(a)
- if o == "--timeout":
- config.timeout(a)
- if o == "--recovery_upcall":
- config.recovery_upcall(a)
- if o == "--ldapurl":
- config.ldapurl(a)
- if o == "--config":
- config.config_name(a)
- if o == "--select":
- config.init_select(a)
- if o == "--lctl_dump":
- config.lctl_dump(a)
-
- return args
-
-def fetch(url):
- import urllib
- data = ""
- try:
- s = urllib.urlopen(url)
- data = s.read()
- except:
- usage()
- return data
def setupModulePath(cmd, portals_dir = PORTALS_DIR):
base = os.path.dirname(cmd)
- if os.access(base+"/Makefile", os.R_OK):
- if not config.lustre_dir():
- config.lustre_dir(os.path.join(base, ".."))
+ if development_mode():
+ if not config.lustre:
+ config.lustre = (os.path.join(base, ".."))
# normalize the portals dir, using command line arg if set
- if config.portals_dir():
- portals_dir = config.portals_dir()
- dir = os.path.join(config.lustre_dir(), portals_dir)
- config.portals_dir(dir)
- elif config.lustre_dir() and config.portals_dir():
+ if config.portals:
+ portals_dir = config.portals
+ dir = os.path.join(config.lustre, portals_dir)
+ config.portals = dir
+ elif config.lustre and config.portals:
# production mode
# if --lustre and --portals, normalize portals
# can ignore POTRALS_DIR here, since it is probly useless here
- dir = config.portals_dir()
- dir = os.path.join(config.lustre_dir(), dir)
- config.portals_dir(dir)
+ config.portals = os.path.join(config.lustre, config.portals)
def sysctl(path, val):
- if config.noexec():
+ if config.noexec:
return
try:
fp = open(os.path.join('/proc/sys', path), 'w')
def sys_set_debug_path():
- debug("debug path: ", config.debug_path())
- sysctl('portals/debug_path', config.debug_path())
+ debug("debug path: ", config.debug_path)
+ sysctl('portals/debug_path', config.debug_path)
def sys_set_recovery_upcall(upcall):
# the command overrides the value in the node config
- if config.recovery_upcall():
- upcall = config.recovery_upcall()
+ if config.recovery_upcall:
+ upcall = config.recovery_upcall
if upcall:
debug("setting recovery_upcall:", upcall)
sysctl('lustre/recovery_upcall', upcall)
def sys_set_timeout(timeout):
# the command overrides the value in the node config
- if config.timeout() > 0:
- timeout = config.timeout()
+ if config.timeout > 0:
+ timeout = config.timeout
if timeout > 0:
debug("setting timeout:", timeout)
sysctl('lustre/timeout', timeout)
-def sys_set_ptldebug(ptldebug):
+def sys_set_ptldebug():
# the command overrides the value in the node config
- if config.ptldebug():
- ptldebug = config.ptldebug()
- sysctl('portals/debug', ptldebug)
+ if config.ptldebug != None:
+ ptldebug = config.ptldebug
+ sysctl('portals/debug', ptldebug)
def sys_set_netmem_max(path, max):
debug("setting", path, "to at least", max)
- if config.noexec():
+ if config.noexec:
return
fp = open(path)
str = fp.readline()
return
os.environ['PATH'] = os.environ['PATH'] + ':' + new_dir
+def default_debug_path():
+ path = '/tmp/lustre-log'
+ if os.path.isdir('/r'):
+ return '/r' + path
+ else:
+ return path
+
+def default_gdb_script():
+ script = '/tmp/ogdb'
+ if os.path.isdir('/r'):
+ return '/r' + script
+ else:
+ return script
+
DEFAULT_PATH = ('/sbin', '/usr/sbin', '/bin', '/usr/bin')
# ensure basic elements are in the system path
for dir in DEFAULT_PATH:
add_to_path(dir)
-# Initialize or shutdown lustre according to a configuration file
-# * prepare the system for lustre
-# * configure devices with lctl
-# Shutdown does steps in reverse
-#
+# global hack for the --select handling
+tgt_select = {}
+def init_select(arg):
+ # arg = "service=nodeA,service2=nodeB"
+ global tgt_select
+ list = string.split(arg, ',')
+ for entry in list:
+ srv, node = string.split(entry, '=')
+ tgt_select[srv] = node
+
+def get_select(srv):
+ if tgt_select.has_key(srv):
+ return tgt_select[srv]
+ return None
+
+lconf_options = [
+ ('verbose,v', "Print system commands as they are run"),
+ ('ldapurl',"LDAP server URL, eg. ldap://localhost", Lustre.Options.PARAM),
+ ('config', "Cluster config name used for LDAP query", Lustre.Options.PARAM),
+ ('select', "service=nodeA,service2=nodeB ", Lustre.Options.PARAM),
+ ('node', "Load config for <nodename>", Lustre.Options.PARAM),
+ ('cleanup,d', "Cleans up config. (Shutdown)"),
+ ('force,f', "Forced unmounting and/or obd detach during cleanup",
+ Lustre.Options.FLAG, 0),
+ ('gdb', """Prints message after creating gdb module script
+ and sleeps for 5 seconds."""),
+ ('noexec,n', """Prints the commands and steps that will be run for a
+ config without executing them. This can used to check if a
+ config file is doing what it should be doing"""),
+ ('nomod', "Skip load/unload module step."),
+ ('nosetup', "Skip device setup/cleanup step."),
+ ('reformat', "Reformat all devices (without question)"),
+ ('dump', "Dump the kernel debug log to file before portals is unloaded",
+ Lustre.Options.PARAM),
+ ('minlevel', "Specify the minimum level of services to configure/cleanup",
+ Lustre.Options.PARAM, 0),
+ ('maxlevel', """Specify the maximum level of services to configure/cleanup
+ Levels are aproximatly like:
+ 10 - network
+ 20 - device, ldlm
+ 30 - osd, mdd
+ 40 - mds, ost
+ 70 - mountpoint, echo_client, osc, mdc, lov""",
+ Lustre.Options.PARAM, 100),
+ ('lustre', """Base directory of lustre sources. This parameter will
+ cause lconf to load modules from a source tree.""",
+ Lustre.Options.PARAM),
+ ('portals', """Portals source directory. If this is a relative path,
+ then it is assumed to be relative to lustre. """,
+ Lustre.Options.PARAM),
+ ('timeout', "Set recovery timeout", Lustre.Options.PARAM),
+ ('recovery_upcall', "Set recovery upcall script", Lustre.Options.PARAM),
+ ('lctl_dump', "Save lctl ioctls to the dumfile argument",
+ Lustre.Options.PARAM),
+ ('ptldebug', "Set the portals debug level", Lustre.Options.PARAM),
+ ('gdb_script', "Fullname of gdb debug script", Lustre.Options.PARAM,
+ default_gdb_script()),
+ ('debug_path', "Path to save debug dumps", Lustre.Options.PARAM,
+ default_debug_path()),
+# Client recovery options
+ ('recover', "Recover a device"),
+ ('tgt_uuid', "The failed target (required for recovery)",
+ Lustre.Options.PARAM),
+ ('client_uuid', "The failed client (required for recovery)",
+ Lustre.Options.PARAM),
+ ('conn_uuid', "The failed connection (required for recovery)",
+ Lustre.Options.PARAM),
+ ]
+
def main():
- global lctl, MAXTCPBUF
+ global lctl, config
+
+ # in the upcall this is set to SIG_IGN
+ signal.signal(signal.SIGCHLD, signal.SIG_DFL)
+
+ cl = Lustre.Options("lconf", "config.xml", lconf_options)
+ try:
+ config, args = cl.parse(sys.argv[1:])
+ except Lustre.OptionError, e:
+ print e
+ sys.exit(1)
+
+ setupModulePath(sys.argv[0])
host = socket.gethostname()
sanitise_path()
- args = parse_cmdline(sys.argv[1:])
if len(args) > 0:
if not os.access(args[0], os.R_OK):
print 'File not found or readable:', args[0]
except Exception:
panic("%s does not appear to be a config file." % (args[0]))
sys.exit(1) # make sure to die here, even in debug mode.
- db = LustreDB_XML(dom.documentElement, dom.documentElement)
- elif config.ldapurl():
- if not config.config_name():
+ db = Lustre.LustreDB_XML(dom.documentElement, dom.documentElement)
+ elif config.ldapurl:
+ if not config.config:
panic("--ldapurl requires --config name")
- dn = "config=%s,fs=lustre" % (config.config_name())
- db = LustreDB_LDAP('', {}, base=dn, url = config.ldapurl())
+ dn = "config=%s,fs=lustre" % (config.config)
+ db = Lustre.LustreDB_LDAP('', {}, base=dn, url = config.ldapurl)
else:
- usage()
+ cl.usage()
+ sys.exit(1)
node_list = []
- if config.node():
- node_list.append(config.node())
+ if config.node:
+ node_list.append(config.node)
else:
if len(host) > 0:
node_list.append(host)
node_list.append('localhost')
+
debug("configuring for host: ", node_list)
if len(host) > 0:
- config._debug_path = config._debug_path + '-' + host
- config._gdb_script = config._gdb_script + '-' + host
-
- setupModulePath(sys.argv[0])
+ config.debug_path = config.debug_path + '-' + host
+ config.gdb_script = config.gdb_script + '-' + host
lctl = LCTLInterface('lctl')
- if config.lctl_dump():
- lctl.use_save_file(config.lctl_dump())
- else:
- sys_make_devices()
- sys_set_netmem_max('/proc/sys/net/core/rmem_max', MAXTCPBUF)
- sys_set_netmem_max('/proc/sys/net/core/wmem_max', MAXTCPBUF)
+
+ if config.lctl_dump:
+ lctl.use_save_file(config.lctl_dump)
doHost(db, node_list)
if __name__ == "__main__":
try:
main()
- except LconfError, e:
+ except Lustre.LconfError, e:
print e
except CommandError, e:
e.dump()
if first_cleanup_error:
sys.exit(first_cleanup_error)
-
git://git.whamcloud.com / fs / lustre-release.git / commitdiff