# lconf is the main driver script for starting and stopping
# lustre filesystem services.
#
-# Based in part on the XML obdctl modifications done by Brian Behlendorf
+# Based in part on the XML obdctl modifications done by Brian Behlendorf
import sys, getopt, types
import string, os, stat, popen2, socket, time, random, fcntl, select
def development_mode():
base = os.path.dirname(sys.argv[0])
- if os.access(base+"/Makefile.am", os.R_OK):
+ if os.access(base+"/Makefile", os.R_OK):
return 1
return 0
-if not development_mode():
+if development_mode():
+ sys.path.append('../utils')
+else:
sys.path.append(PYMOD_DIR)
import Lustre
# Global parameters
-MAXTCPBUF = 1048576
-DEFAULT_TCPBUF = 1048576
+MAXTCPBUF = 16777216
+DEFAULT_TCPBUF = 8388608
+DEFAULT_PORT = 988
#
# Maximum number of devices to search for.
# (the /dev/loop* nodes need to be created beforehand)
MAX_LOOP_DEVICES = 256
PORTALS_DIR = 'portals'
+# Needed to call lconf --record
+CONFIG_FILE = ""
# Please keep these in sync with the values in portals/kp30.h
-ptldebug_names = {
+ptldebug_names = {
"trace" : (1 << 0),
"inode" : (1 << 1),
"super" : (1 << 2),
"ha" : (1 << 19),
"rpctrace" : (1 << 20),
"vfstrace" : (1 << 21),
+ "reada" : (1 << 22),
+ "config" : (1 << 23),
}
subsystem_names = {
- "undefined" : (0 << 24),
- "mdc" : (1 << 24),
- "mds" : (2 << 24),
- "osc" : (3 << 24),
- "ost" : (4 << 24),
- "class" : (5 << 24),
- "obdfs" : (6 << 24),
- "llite" : (7 << 24),
- "rpc" : (8 << 24),
- "ext2obd" : (9 << 24),
- "portals" : (10 << 24),
- "socknal" : (11 << 24),
- "qswnal" : (12 << 24),
- "pinger" : (13 << 24),
- "filter" : (14 << 24),
- "trace" : (15 << 24),
- "echo" : (16 << 24),
- "ldlm" : (17 << 24),
- "lov" : (18 << 24),
- "gmnal" : (19 << 24),
- "ptlrouter" : (20 << 24),
- "cobd" : (21 << 24),
- "ptlbd" : (22 << 24),
- "log" : (23 << 24),
- "mgmt" : (24 << 24),
+ "undefined" : (1 << 0),
+ "mdc" : (1 << 1),
+ "mds" : (1 << 2),
+ "osc" : (1 << 3),
+ "ost" : (1 << 4),
+ "class" : (1 << 5),
+ "log" : (1 << 6),
+ "llite" : (1 << 7),
+ "rpc" : (1 << 8),
+ "mgmt" : (1 << 9),
+ "portals" : (1 << 10),
+ "socknal" : (1 << 11),
+ "qswnal" : (1 << 12),
+ "pinger" : (1 << 13),
+ "filter" : (1 << 14),
+ "ptlbd" : (1 << 15),
+ "echo" : (1 << 16),
+ "ldlm" : (1 << 17),
+ "lov" : (1 << 18),
+ "gmnal" : (1 << 19),
+ "ptlrouter" : (1 << 20),
+ "cobd" : (1 << 21),
+ "ibnal" : (1 << 22),
+ "cmobd" : (1 << 23),
}
if not first_cleanup_error:
first_cleanup_error = rc
-# ============================================================
+# ============================================================
# debugging and error funcs
def fixme(msg = "this feature"):
- raise Lustre.LconfError, msg + ' not implmemented yet.'
+ raise Lustre.LconfError, msg + ' not implemented yet.'
def panic(*args):
msg = string.join(map(str,args))
msg = string.join(map(str,args))
print msg
-
# ack, python's builtin int() does not support '0x123' syntax.
# eval can do it, although what a hack!
def my_int(s):
return pid
except IOError:
return 0
-
+
def clean_pidfile(self):
""" Remove a stale pidfile """
log("removing stale pidfile:", self.pidfile())
os.unlink(self.pidfile())
except OSError, e:
log(self.pidfile(), e)
-
+
class AcceptorHandler(DaemonHandler):
- def __init__(self, port, net_type, send_mem, recv_mem, irq_aff, nid_xchg):
+ def __init__(self, port, net_type, send_mem, recv_mem, irq_aff):
DaemonHandler.__init__(self, "acceptor")
self.port = port
self.flags = ''
self.send_mem = send_mem
self.recv_mem = recv_mem
- if net_type == 'toe':
- self.flags = self.flags + ' -N 4'
if irq_aff:
self.flags = self.flags + ' -i'
- if nid_xchg:
- self.flags = self.flags + ' -x'
def pidfile(self):
return "/var/run/%s-%d.pid" % (self.command, self.port)
def command_line(self):
return string.join(map(str,('-s', self.send_mem, '-r', self.recv_mem, self.flags, self.port)))
-
+
acceptors = {}
# start the acceptors
def run_acceptors():
+ if config.lctl_dump or config.record:
+ return
for port in acceptors.keys():
daemon = acceptors[port]
if not daemon.running():
daemon.start()
def run_one_acceptor(port):
+ if config.lctl_dump or config.record:
+ return
if acceptors.has_key(port):
daemon = acceptors[port]
if not daemon.running():
daemon.start()
else:
- panic("run_one_acceptor: No acceptor defined for port:", port)
-
+ panic("run_one_acceptor: No acceptor defined for port:", port)
+
def stop_acceptor(port):
if acceptors.has_key(port):
daemon = acceptors[port]
if daemon.running():
daemon.stop()
-
+
# ============================================================
# handle lctl interface
"""
self.lctl = find_prog(cmd)
self.save_file = ''
+ self.record_device = ''
if not self.lctl:
if config.noexec:
debug('! lctl not found')
def use_save_file(self, file):
self.save_file = file
-
+
+ def record(self, dev_name, logname):
+ log("Recording log", logname, "on", dev_name)
+ self.record_device = dev_name
+ self.record_log = logname
+
+ def end_record(self):
+ log("End recording log", self.record_log, "on", self.record_device)
+ self.record_device = None
+ self.record_log = None
+
def set_nonblock(self, fd):
fl = fcntl.fcntl(fd, F_GETFL)
fcntl.fcntl(fd, F_SETFL, fl | os.O_NDELAY)
"""
cmd_line = self.lctl
if self.save_file:
- cmds = '\n dump ' + self.save_file + cmds
+ cmds = '\n dump ' + self.save_file + '\n' + cmds
+ elif self.record_device:
+ cmds = """
+ device $%s
+ record %s
+ %s""" % (self.record_device, self.record_log, cmds)
debug("+", cmd_line, cmds)
if config.noexec: return (0, [])
child = popen2.Popen3(cmd_line, 1) # Capture stdout and stderr from command
child.tochild.write(cmds + "\n")
child.tochild.close()
+# print "LCTL:", cmds
# From "Python Cookbook" from O'Reilly
outfile = child.fromchild
raise CommandError(self.lctl, out, rc)
return rc, out
-
+
+ def clear_log(self, dev, log):
+ """ clear an existing log """
+ cmds = """
+ device $%s
+ probe
+ clear_log %s
+ quit """ % (dev, log)
+ self.run(cmds)
+
def network(self, net, nid):
- """ initialized network and add "self" """
+ """ set mynid """
cmds = """
network %s
mynid %s
self.run(cmds)
# 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:
- flags = 's'
- if srv.irq_affinity:
- flags = flags + 'i'
- if srv.nid_exchange:
- flags = flags + 'x'
- cmds = """%s
+ def add_uuid(self, net_type, uuid, nid):
+ cmds = "\n add_uuid %s %s %s" %(uuid, nid, net_type)
+ self.run(cmds)
+
+ def add_autoconn(self, net_type, send_mem, recv_mem, nid, hostaddr,
+ port, flags):
+ if net_type in ('tcp',) and not config.lctl_dump:
+ cmds = """
network %s
send_mem %d
recv_mem %d
- add_autoconn %s %s %d %s""" % (cmds, srv.net_type,
- srv.send_mem,
- srv.recv_mem,
- srv.nid, srv.hostaddr, srv.port, flags )
+ add_autoconn %s %s %d %s
+ quit""" % (net_type,
+ send_mem,
+ recv_mem,
+ nid, hostaddr, port, flags )
+ self.run(cmds)
- cmds = cmds + "\n quit"
- self.run(cmds)
+ def connect(self, srv):
+ self.add_uuid(srv.net_type, srv.nid_uuid, srv.nid)
+ if srv.net_type in ('tcp',) and not config.lctl_dump:
+ flags = 's'
+ if srv.irq_affinity:
+ flags = flags + 'i'
+ self.add_autoconn(srv.net_type, srv.send_mem, srv.recv_mem,
+ srv.nid, srv.hostaddr, srv.port, flags)
# Recover a device
def recover(self, dev_name, new_conn):
cmds = """
device $%s
- probe
recover %s""" %(dev_name, new_conn)
self.run(cmds)
-
+
# add a route to a range
def add_route(self, net, gw, lo, hi):
cmds = """
except CommandError, e:
log ("ignore: ")
e.dump()
-
+
def del_route(self, net, gw, lo, hi):
cmds = """
ignore_errors
network %s
- del_route %s
- quit """ % (net, lo)
+ del_route %s %s %s
+ quit """ % (net, gw, lo, hi)
self.run(cmds)
# add a route to a host
def add_route_host(self, net, uuid, gw, tgt):
+ self.add_uuid(net, uuid, tgt)
cmds = """
network %s
- add_uuid %s %s %s
add_route %s %s
quit """ % (net,
- uuid, tgt, net,
gw, tgt)
try:
self.run(cmds)
# add a route to a range
def del_route_host(self, net, uuid, gw, tgt):
+ self.del_uuid(uuid)
cmds = """
ignore_errors
network %s
- del_uuid %s
- del_route %s
- quit """ % (net, uuid, tgt)
+ del_route %s %s
+ quit """ % (net, gw, tgt)
self.run(cmds)
- # disconnect one connection
- def disconnect(self, srv):
- cmds = " ignore_errors\n del_uuid %s" % (srv.uuid)
- if srv.net_type in ('tcp', 'toe') and not config.lctl_dump:
- cmds = """%s
+
+ def del_autoconn(self, net_type, nid, hostaddr):
+ if net_type in ('tcp',) and not config.lctl_dump:
+ cmds = """
+ ignore_errors
network %s
- del_autoconn %s %s s""" % (cmds,
- srv.net_type,
- srv.nid, srv.hostaddr)
- cmds = cmds + "\n quit"
+ del_autoconn %s %s s
+ quit""" % (net_type,
+ nid, hostaddr)
+ self.run(cmds)
- self.run(cmds)
+ # disconnect one connection
+ def disconnect(self, srv):
+ self.del_uuid(srv.nid_uuid)
+ if srv.net_type in ('tcp',) and not config.lctl_dump:
+ self.del_autoconn(srv.net_type, srv.nid, srv.hostaddr)
- def del_uuid(self, servuuid):
+ def del_uuid(self, uuid):
cmds = """
ignore_errors
del_uuid %s
- quit""" % (servuuid,)
+ quit""" % (uuid,)
self.run(cmds)
# disconnect all
quit""" % (net)
self.run(cmds)
- # create a new device with lctl
- def newdev(self, attach, setup = ""):
+ def attach(self, type, name, uuid):
+ cmds = """
+ attach %s %s %s
+ quit""" % (type, name, uuid)
+ self.run(cmds)
+
+ def setup(self, name, setup = ""):
cmds = """
- newdev
- attach %s
+ cfg_device %s
setup %s
- quit""" % (attach, setup)
+ quit""" % (name, setup)
self.run(cmds)
+
+ # create a new device with lctl
+ def newdev(self, type, name, uuid, setup = ""):
+ self.attach(type, name, uuid);
+ try:
+ self.setup(name, setup)
+ except CommandError, e:
+ self.cleanup(name, uuid, 0)
+ raise e
+
+
# cleanup a device
def cleanup(self, name, uuid, force, failover = 0):
if failover: force = 1
cmds = """
ignore_errors
- device $%s
+ cfg_device $%s
cleanup %s %s
detach
quit""" % (name, ('', 'force')[force],
self.run(cmds)
# create an lov
- def lov_setconfig(self, uuid, mdsuuid, stripe_cnt, stripe_sz, stripe_off,
- pattern, devlist):
+ def lov_setup(self, name, uuid, desc_uuid, stripe_cnt,
+ stripe_sz, stripe_off, pattern):
+ cmds = """
+ attach lov %s %s
+ lov_setup %s %d %d %d %s
+ quit""" % (name, uuid, desc_uuid, stripe_cnt, stripe_sz, stripe_off, pattern)
+ self.run(cmds)
+
+ # add an OBD to a LOV
+ def lov_add_obd(self, name, uuid, obd_uuid, index, gen):
+ cmds = """
+ lov_modify_tgts add %s %s %s %s
+ quit""" % (name, obd_uuid, index, gen)
+ self.run(cmds)
+
+ # create an lmv
+ def lmv_setup(self, name, uuid, desc_uuid, devlist):
+ cmds = """
+ attach lmv %s %s
+ lmv_setup %s %s
+ quit""" % (name, uuid, desc_uuid, devlist)
+ self.run(cmds)
+
+ # delete an OBD from a LOV
+ def lov_del_obd(self, name, uuid, obd_uuid, index, gen):
+ cmds = """
+ lov_modify_tgts del %s %s %s %s
+ quit""" % (name, obd_uuid, index, gen)
+ self.run(cmds)
+
+ # deactivate an OBD
+ def deactivate(self, name):
cmds = """
device $%s
- probe
- lov_setconfig %s %d %d %d %s %s
- quit""" % (mdsuuid, uuid, stripe_cnt, stripe_sz, stripe_off, pattern, devlist)
+ deactivate
+ quit""" % (name)
self.run(cmds)
# dump the log file
# get list of devices
def device_list(self):
- try:
- rc, out = self.runcmd('device_list')
- except CommandError, e:
- if config.cleanup:
- out = []
- else:
- raise e
- return out
+ devices = '/proc/fs/lustre/devices'
+ ret = []
+ if os.access(devices, os.R_OK):
+ try:
+ fp = open(devices, 'r')
+ ret = fp.readlines()
+ fp.close()
+ except IOError, e:
+ log(e)
+ return ret
# get lustre version
def lustre_version(self):
return out
# dump mount options
- def mount_option(self, option):
+ def mount_option(self, profile, osc, mdc):
+ cmds = """
+ mount_option %s %s %s
+ quit""" % (profile, osc, mdc)
+ self.run(cmds)
+
+ # delete mount options
+ def del_mount_option(self, profile):
+ cmds = """
+ del_mount_option %s
+ quit""" % (profile,)
+ self.run(cmds)
+
+ def set_timeout(self, timeout):
cmds = """
- mount_option %s
- quit""" % (option)
+ set_timeout %s
+ quit""" % (timeout,)
+ self.run(cmds)
+
+ def set_lustre_upcall(self, upcall):
+ cmds = """
+ set_lustre_upcall %s
+ quit""" % (upcall,)
self.run(cmds)
# ============================================================
# Various system-level functions
return module
def find_module(src_dir, dev_dir, modname):
- mod = '%s.o' % (modname)
- module = src_dir +'/'+ dev_dir +'/'+ mod
- try:
- if os.access(module, os.R_OK):
- return module
- except OSError:
- pass
+ modbase = src_dir +'/'+ dev_dir +'/'+ modname
+ for modext in '.ko', '.o':
+ module = modbase + modext
+ try:
+ if os.access(module, os.R_OK):
+ return module
+ except OSError:
+ pass
return None
# is the path a block device?
except OSError:
return 0
return stat.S_ISBLK(s[stat.ST_MODE])
-
+
# build fs according to type
# fixme: dangerous
-def mkfs(dev, devsize, fstype,jsize):
+def mkfs(dev, devsize, fstype, jsize, isize, mkfsoptions, isblock=1):
block_cnt = ''
jopt = ''
+ iopt = ''
if devsize:
if devsize < 8000:
panic("size of filesystem on '%s' must be larger than 8MB, but is set to %s"%
# devsize is in 1k, and fs block count is in 4k
block_cnt = devsize/4
- if fstype in ('ext3', 'extN'):
+ if fstype in ('ext3', 'extN', 'ldiskfs'):
# ext3 journal size is in megabytes
+ if jsize == 0:
+ if devsize == 0:
+ if not is_block(dev):
+ ret, out = runcmd("ls -l %s" %dev)
+ devsize = int(string.split(out[0])[4]) / 1024
+ else:
+ ret, out = runcmd("sfdisk -s %s" %dev)
+ devsize = int(out[0])
+ if devsize > 1024 * 1024:
+ jsize = ((devsize / 102400) * 4)
+ if jsize > 400:
+ jsize = 400
if jsize: jopt = "-J size=%d" %(jsize,)
- mkfs = 'mkfs.ext2 -j -b 4096 -F '
+ if isize: iopt = "-I %d" %(isize,)
+ mkfs = 'mkfs.ext2 -j -b 4096 '
+ if not isblock or config.force:
+ mkfs = mkfs + ' -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
+ panic('unsupported fs type: ', fstype)
- (ret, out) = run (mkfs, jopt, dev, block_cnt)
+ if config.mkfsoptions != None:
+ mkfs = mkfs + ' ' + config.mkfsoptions
+ if mkfsoptions != None:
+ mkfs = mkfs + ' ' + mkfsoptions
+ (ret, out) = run (mkfs, jopt, iopt, dev, block_cnt)
if ret:
panic("Unable to build fs:", dev, string.join(out))
# enable hash tree indexing on fsswe
- if fstype in ('ext3', 'extN'):
+ if fstype in ('ext3', 'extN', 'ldiskfs'):
htree = 'echo "feature FEATURE_C5" | debugfs -w'
(ret, out) = run (htree, dev)
if ret:
panic ("can't access loop devices")
return loop
-# find loop device assigned to thefile
-def find_loop(file):
+# find loop device assigned to the file
+def find_assigned_loop(file):
loop = loop_base()
for n in xrange(0, MAX_LOOP_DEVICES):
dev = loop + str(n)
return ''
# create file if necessary and assign the first free loop device
-def init_loop(file, size, fstype, journal_size):
- dev = find_loop(file)
+def init_loop(file, size, fstype, journal_size, inode_size,
+ mkfsoptions, reformat, autoformat, backfstype, backfile):
+ if fstype == 'smfs':
+ realfile = backfile
+ realfstype = backfstype
+ if is_block(backfile):
+ if reformat or (need_format(realfstype, backfile) and autoformat == 'yes'):
+ mkfs(realfile, size, realfstype, journal_size, inode_size, mkfsoptions, isblock=0)
+ return realfile
+ else:
+ realfile = file
+ realfstype = fstype
+
+ dev = find_assigned_loop(realfile)
if dev:
- print 'WARNING file:', file, 'already mapped to', dev
+ print 'WARNING file:', realfile, 'already mapped to', dev
return dev
- if config.reformat or not os.access(file, os.R_OK | os.W_OK):
+
+ if reformat or not os.access(realfile, os.R_OK | os.W_OK):
if size < 8000:
- panic("size of loopback file '%s' must be larger than 8MB, but is set to %s" % (file,size))
- (ret, out) = run("dd if=/dev/zero bs=1k count=0 seek=%d of=%s" %(size,
- file))
+ panic("size of loopback file '%s' must be larger than 8MB, but is set to %s" % (realfile, size))
+ (ret, out) = run("dd if=/dev/zero bs=1k count=0 seek=%d of=%s" %(size, realfile))
if ret:
- panic("Unable to create backing store:", file)
- mkfs(file, size, fstype, journal_size)
+ panic("Unable to create backing store:", realfile)
+
+ mkfs(realfile, size, realfstype, journal_size, inode_size,
+ mkfsoptions, isblock=0)
loop = loop_base()
# find next free loop
if os.access(dev, os.R_OK):
(stat, out) = run('losetup', dev)
if stat:
- run('losetup', dev, file)
+ run('losetup', dev, realfile)
return dev
else:
print "out of loop devices"
# undo loop assignment
def clean_loop(file):
- dev = find_loop(file)
+ dev = find_assigned_loop(file)
if dev:
ret, out = run('losetup -d', dev)
if ret:
# determine if dev is formatted as a <fstype> filesystem
def need_format(fstype, dev):
- # FIXME don't know how to implement this
+ # FIXME don't know how to implement this
return 0
# initialize a block device if needed
-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, journal_size)
- elif config.reformat or (need_format(fstype, dev) and format == 'yes'):
- mkfs(dev, size, fstype, journal_size)
-
+def block_dev(dev, size, fstype, reformat, autoformat, journal_size,
+ inode_size, mkfsoptions, backfstype, backdev):
+ if config.noexec:
+ return dev
+
+ if fstype == 'smfs' or not is_block(dev):
+ dev = init_loop(dev, size, fstype, journal_size, inode_size,
+ mkfsoptions, reformat, autoformat, backfstype, backdev)
+ elif reformat or (need_format(fstype, dev) and autoformat == 'yes'):
+ mkfs(dev, size, fstype, journal_size, inode_size, mkfsoptions,
+ isblock=0)
# else:
# panic("device:", dev,
# "not prepared, and autoformat is not set.\n",
# "Rerun with --reformat option to format ALL filesystems")
-
+
return dev
def if2addr(iface):
ip = string.split(addr, ':')[1]
return ip
-def get_local_nid(net_type, wildcard):
+def def_mount_options(fstype, target):
+ """returns deafult mount options for passed fstype and target (mds, ost)"""
+ if fstype == 'ext3' or fstype == 'ldiskfs':
+ mountfsoptions = "errors=remount-ro"
+ if target == 'ost' and sys_get_branch() == '2.4':
+ mountfsoptions = "%s,asyncdel" % (mountfsoptions)
+ return mountfsoptions
+ return ""
+
+def sys_get_elan_position_file():
+ procfiles = ["/proc/elan/device0/position",
+ "/proc/qsnet/elan4/device0/position",
+ "/proc/qsnet/elan3/device0/position"]
+ for p in procfiles:
+ if os.access(p, os.R_OK):
+ return p
+ return ""
+
+def sys_get_local_nid(net_type, wildcard, cluster_id):
"""Return the local nid."""
local = ""
- if os.access('/proc/elan/device0/position', os.R_OK):
- local = get_local_address('elan', '*')
+ if sys_get_elan_position_file():
+ local = sys_get_local_address('elan', '*', cluster_id)
else:
- local = get_local_address(net_type, wildcard)
+ local = sys_get_local_address(net_type, wildcard, cluster_id)
return local
-
-def get_local_address(net_type, wildcard):
+
+def sys_get_local_address(net_type, wildcard, cluster_id):
"""Return the local address for the network type."""
local = ""
- if net_type in ('tcp', 'toe'):
+ if net_type in ('tcp',):
if ':' in wildcard:
iface, star = string.split(wildcard, ':')
local = if2addr(iface)
host = socket.gethostname()
local = socket.gethostbyname(host)
elif net_type == 'elan':
- # awk '/NodeId/ { print $2 }' '/proc/elan/device0/position'
+ # awk '/NodeId/ { print $2 }' 'sys_get_elan_position_file()'
+ f = sys_get_elan_position_file()
+ if not f:
+ panic ("unable to determine local Elan ID")
try:
- fp = open('/proc/elan/device0/position', 'r')
+ fp = open(f, 'r')
lines = fp.readlines()
fp.close()
for l in lines:
a = string.split(l)
if a[0] == 'NodeId':
- local = a[1]
+ elan_id = a[1]
break
+ try:
+ nid = my_int(cluster_id) + my_int(elan_id)
+ local = "%d" % (nid)
+ except ValueError, e:
+ local = elan_id
except IOError, e:
log(e)
elif net_type == 'gm':
fixme("automatic local address for GM")
- elif net_type == 'scimac':
- scinode="/opt/scali/sbin/scinode"
- if os.path.exists(scinode):
- (rc,local) = run(scinode)
- else:
- panic (scinode, " not found on node with scimac networking")
- if rc:
- panic (scinode, " failed")
- local=string.rstrip(local[0])
return local
+def sys_get_branch():
+ """Returns kernel release"""
+ try:
+ fp = open('/proc/sys/kernel/osrelease')
+ lines = fp.readlines()
+ fp.close()
+
+ for l in lines:
+ version = string.split(l)
+ a = string.split(version[0], '.')
+ return a[0] + '.' + a[1]
+ except IOError, e:
+ log(e)
+ return ""
+
+
+def mod_loaded(modname):
+ """Check if a module is already loaded. Look in /proc/modules for it."""
+ try:
+ fp = open('/proc/modules')
+ lines = fp.readlines()
+ fp.close()
+ # please forgive my tired fingers for this one
+ ret = filter(lambda word, mod=modname: word == mod,
+ map(lambda line: string.split(line)[0], lines))
+ return ret
+ except Exception, e:
+ return 0
+
# XXX: instead of device_list, ask for $name and see what we get
def is_prepared(name):
"""Return true if a device exists for the name"""
if config.lctl_dump:
return 0
- if config.noexec and config.cleanup:
+ if (config.noexec or config.record) and config.cleanup:
return 1
try:
# expect this format:
return 0
def is_network_prepared():
- """If the LDLM device exists, then assume that all networking
+ """If the any device exists, then assume that all networking
has been configured"""
- return is_prepared('ldlm')
-
+ out = lctl.device_list()
+ return len(out) > 0
+
def fs_is_mounted(path):
"""Return true if path is a mounted lustre filesystem"""
try:
except IOError, e:
log(e)
return 0
-
-# ============================================================
-# Classes to prepare and cleanup the various objects
-#
-class Module:
- """ Base class for the rest of the modules. The default cleanup method is
- defined here, as well as some utilitiy funcs.
- """
- def __init__(self, module_name, db):
- self.db = db
- self.module_name = module_name
- self.name = self.db.getName()
- self.uuid = self.db.getUUID()
+
+class kmod:
+ """Manage kernel modules"""
+ def __init__(self, lustre_dir, portals_dir):
+ self.lustre_dir = lustre_dir
+ self.portals_dir = portals_dir
self.kmodule_list = []
- self._server = None
- self._connected = 0
-
- def info(self, *args):
- msg = string.join(map(str,args))
- print self.module_name + ":", self.name, self.uuid, msg
- def cleanup(self):
- """ default cleanup, used for most modules """
- self.info()
- try:
- lctl.cleanup(self.name, self.uuid, config.force)
- except CommandError, e:
- log(self.module_name, "cleanup failed: ", self.name)
- e.dump()
- cleanup_error(e.rc)
-
def add_portals_module(self, dev_dir, modname):
"""Append a module to list of modules to load."""
- self.kmodule_list.append((config.portals, dev_dir, modname))
+ self.kmodule_list.append((self.portals_dir, 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, dev_dir, modname))
-
- def mod_loaded(self, modname):
- """Check if a module is already loaded. Look in /proc/modules for it."""
- fp = open('/proc/modules')
- lines = fp.readlines()
- fp.close()
- # please forgive my tired fingers for this one
- ret = filter(lambda word, mod=modname: word == mod,
- map(lambda line: string.split(line)[0], lines))
- return ret
+ self.kmodule_list.append((self.lustre_dir, dev_dir, modname))
def load_module(self):
"""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 mod_loaded(mod) and not config.noexec:
continue
log ('loading module:', mod, 'srcdir', src_dir, 'devdir', dev_dir)
if src_dir:
(rc, out) = run('/sbin/modprobe', mod)
if rc:
raise CommandError('modprobe', out, rc)
-
+
def cleanup_module(self):
"""Unload the modules in the list in reverse order."""
- if not self.safe_to_clean():
- return
rev = self.kmodule_list
rev.reverse()
for src_dir, dev_dir, mod in rev:
- if not self.mod_loaded(mod) and not config.noexec:
+ if not mod_loaded(mod) and not config.noexec:
continue
# debug hack
if mod == 'portals' and config.dump:
log('! unable to unload module:', mod)
logall(out)
+# ============================================================
+# Classes to prepare and cleanup the various objects
+#
+class Module:
+ """ Base class for the rest of the modules. The default cleanup method is
+ defined here, as well as some utilitiy funcs.
+ """
+ def __init__(self, module_name, db):
+ self.db = db
+ self.module_name = module_name
+ self.name = self.db.getName()
+ self.uuid = self.db.getUUID()
+ self._server = None
+ self._connected = 0
+ self.kmod = kmod(config.lustre, config.portals)
+
+ def info(self, *args):
+ msg = string.join(map(str,args))
+ print self.module_name + ":", self.name, self.uuid, msg
+
+ def cleanup(self):
+ """ default cleanup, used for most modules """
+ self.info()
+ try:
+ lctl.cleanup(self.name, self.uuid, config.force)
+ except CommandError, e:
+ log(self.module_name, "cleanup failed: ", self.name)
+ e.dump()
+ cleanup_error(e.rc)
+
+ def add_portals_module(self, dev_dir, modname):
+ """Append a module to list of modules to load."""
+ self.kmod.add_portals_module(dev_dir, modname)
+
+ def add_lustre_module(self, dev_dir, modname):
+ """Append a module to list of modules to load."""
+ self.kmod.add_lustre_module(dev_dir, modname)
+
+ def load_module(self):
+ """Load all the modules in the list in the order they appear."""
+ self.kmod.load_module()
+
+ def cleanup_module(self):
+ """Unload the modules in the list in reverse order."""
+ if self.safe_to_clean():
+ self.kmod.cleanup_module()
+
def safe_to_clean(self):
return 1
-
+
def safe_to_clean_modules(self):
return self.safe_to_clean()
-
+
class Network(Module):
def __init__(self,db):
Module.__init__(self, 'NETWORK', db)
self.send_mem = self.db.get_val_int('sendmem', DEFAULT_TCPBUF)
self.recv_mem = self.db.get_val_int('recvmem', DEFAULT_TCPBUF)
self.irq_affinity = self.db.get_val_int('irqaffinity', 0)
- self.nid_exchange = self.db.get_val_int('nidexchange', 0)
if '*' in 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)
+ self.nid = sys_get_local_nid(self.net_type, self.nid, self.cluster_id)
if not self.nid:
- panic("unable to set nid for", self.net_type, self.nid)
+ panic("unable to set nid for", self.net_type, self.nid, cluster_id)
+ self.generic_nid = 1
debug("nid:", self.nid)
+ else:
+ self.generic_nid = 0
+
+ self.nid_uuid = self.nid_to_uuid(self.nid)
self.hostaddr = self.db.get_val('hostaddr', self.nid)
if '*' in self.hostaddr:
- self.hostaddr = get_local_address(self.net_type, self.hostaddr)
- if not self.nid:
- panic("unable to set nid for", self.net_type, self.hostaddr)
+ self.hostaddr = sys_get_local_address(self.net_type, self.hostaddr, self.cluster_id)
+ if not self.hostaddr:
+ panic("unable to set hostaddr for", self.net_type, self.hostaddr, self.cluster_id)
debug("hostaddr:", self.hostaddr)
- self.add_portals_module("libcfs", 'portals')
+ self.add_portals_module("libcfs", 'libcfs')
+ self.add_portals_module("portals", 'portals')
if node_needs_router():
self.add_portals_module("router", 'kptlrouter')
if self.net_type == 'tcp':
self.add_portals_module("knals/socknal", 'ksocknal')
- if self.net_type == 'toe':
- self.add_portals_module("knals/toenal", 'ktoenal')
if self.net_type == 'elan':
self.add_portals_module("knals/qswnal", 'kqswnal')
if self.net_type == 'gm':
self.add_portals_module("knals/gmnal", 'kgmnal')
- if self.net_type == 'scimac':
- self.add_portals_module("knals/scimacnal", 'kscimacnal')
+
+ def nid_to_uuid(self, nid):
+ return "NID_%s_UUID" %(nid,)
def prepare(self):
- if is_network_prepared():
+ if not config.record and is_network_prepared():
return
self.info(self.net_type, self.nid, self.port)
- lctl.network(self.net_type, self.nid)
+ if not (config.record and self.generic_nid):
+ lctl.network(self.net_type, self.nid)
+ if self.net_type == 'tcp':
+ sys_tweak_socknal()
+ if self.net_type == 'elan':
+ sys_optimize_elan()
if self.port and node_is_router():
run_one_acceptor(self.port)
self.connect_peer_gateways()
gw = Network(net)
if (gw.cluster_id == self.cluster_id and
gw.net_type == self.net_type):
- # hack: compare as numbers if possible, this should all
- # go away once autoconnect is done.
- # This also conveniently prevents us from connecting to ourself.
- try:
- gw_nid = my_int(gw.nid)
- self_nid = my_int(self.nid)
- except ValueError, e:
- gw_nid = gw.nid
- self_nid = self.nid
- if gw_nid != self_nid:
+ if gw.nid != self.nid:
lctl.connect(gw)
def disconnect_peer_gateways(self):
gw = Network(net)
if (gw.cluster_id == self.cluster_id and
gw.net_type == self.net_type):
- # hack: compare as numbers if possible, this should all
- # go away once autoconnect is done.
- # This also conveniently prevents us from connecting to ourself.
- try:
- gw_nid = my_int(gw.nid)
- self_nid = my_int(self.nid)
- except ValueError, e:
- gw_nid = gw.nid
- self_nid = self.nid
- if gw_nid != self_nid:
+ if gw.nid != self.nid:
try:
lctl.disconnect(gw)
except CommandError, e:
stop_acceptor(self.port)
if node_is_router():
self.disconnect_peer_gateways()
-#
-# This commented out so connections not created by this
-# config are not disturbed
-#
-# try:
-# lctl.disconnectAll(self.net_type)
-# except CommandError, e:
-# print "disconnectAll failed: ", self.name
-# e.dump()
-# cleanup_error(e.rc)
+
+ def correct_level(self, level, op=None):
+ return level
class RouteTable(Module):
def __init__(self,db):
Module.__init__(self, 'ROUTES', db)
+
+ def server_for_route(self, net_type, gw, gw_cluster_id, tgt_cluster_id,
+ lo, hi):
+ # only setup connections for tcp NALs
+ srvdb = None
+ if not net_type in ('tcp',):
+ return None
+
+ # connect to target if route is to single node and this node is the gw
+ if lo == hi and local_interface(net_type, gw_cluster_id, gw):
+ if not local_cluster(net_type, tgt_cluster_id):
+ panic("target", lo, " not on the local cluster")
+ srvdb = self.db.nid2server(lo, net_type, gw_cluster_id)
+ # connect to gateway if this node is not the gw
+ elif (local_cluster(net_type, gw_cluster_id)
+ and not local_interface(net_type, gw_cluster_id, gw)):
+ srvdb = self.db.nid2server(gw, net_type, gw_cluster_id)
+ else:
+ return None
+
+ if not srvdb:
+ panic("no server for nid", lo)
+ return None
+
+ return Network(srvdb)
+
def prepare(self):
- if is_network_prepared():
+ if not config.record and is_network_prepared():
return
self.info()
for net_type, gw, gw_cluster_id, tgt_cluster_id, lo, hi in self.db.get_route_tbl():
lctl.add_route(net_type, gw, lo, hi)
- if net_type in ('tcp', 'toe') and local_net_type(net_type, tgt_cluster_id) and lo == hi:
- srvdb = self.db.nid2server(lo, net_type)
- if not srvdb:
- panic("no server for nid", lo)
- else:
- srv = Network(srvdb)
- lctl.connect(srv)
+ srv = self.server_for_route(net_type, gw, gw_cluster_id, tgt_cluster_id, lo, hi)
+ if srv:
+ lctl.connect(srv)
def safe_to_clean(self):
return not is_network_prepared()
# the network is still being used, don't clean it up
return
for net_type, gw, gw_cluster_id, tgt_cluster_id, lo, hi in self.db.get_route_tbl():
- if net_type in ('tcp', 'toe') and local_net_type(net_type, tgt_cluster_id) and lo == hi:
- srvdb = self.db.nid2server(lo, net_type)
- if not srvdb:
- panic("no server for nid", lo)
- else:
- srv = Network(srvdb)
- try:
- lctl.disconnect(srv)
- except CommandError, e:
- print "disconnect failed: ", self.name
- e.dump()
- cleanup_error(e.rc)
+ srv = self.server_for_route(net_type, gw, gw_cluster_id, tgt_cluster_id, lo, hi)
+ if srv:
+ try:
+ lctl.disconnect(srv)
+ except CommandError, e:
+ print "disconnect failed: ", self.name
+ e.dump()
+ cleanup_error(e.rc)
+
try:
lctl.del_route(net_type, gw, lo, hi)
except CommandError, e:
class Management(Module):
def __init__(self, db):
Module.__init__(self, 'MGMT', db)
+ self.add_lustre_module('lvfs', 'lvfs')
self.add_lustre_module('obdclass', 'obdclass')
self.add_lustre_module('ptlrpc', 'ptlrpc')
- self.add_lustre_module('ldlm', 'ldlm')
self.add_lustre_module('mgmt', 'mgmt_svc')
def prepare(self):
- if is_prepared(self.name):
+ if not config.record and is_prepared(self.name):
return
self.info()
- lctl.newdev(attach="mgmt %s %s" % (self.name, self.uuid))
+ lctl.newdev("mgmt", self.name, self.uuid)
def safe_to_clean(self):
return 1
if is_prepared(self.name):
Module.cleanup(self)
+ def correct_level(self, level, op=None):
+ return level
+
+# This is only needed to load the modules; the LDLM device
+# is now created automatically.
class LDLM(Module):
def __init__(self,db):
Module.__init__(self, 'LDLM', db)
+ self.add_lustre_module('lvfs', 'lvfs')
self.add_lustre_module('obdclass', 'obdclass')
self.add_lustre_module('ptlrpc', 'ptlrpc')
- self.add_lustre_module('ldlm', 'ldlm')
def prepare(self):
- if is_prepared(self.name):
- return
- self.info()
- lctl.newdev(attach="ldlm %s %s" % ('ldlm', 'ldlm_UUID'))
-
- def safe_to_clean(self):
- out = lctl.device_list()
- return len(out) <= 1
+ return
def cleanup(self):
- if is_prepared(self.name):
- Module.cleanup(self)
+ return
+
+ def correct_level(self, level, op=None):
+ return level
+
class LOV(Module):
- def __init__(self, db, uuid, fs_name):
+ def __init__(self, db, uuid, fs_name, name_override = None, config_only = None):
Module.__init__(self, 'LOV', db)
- self.add_lustre_module('mdc', 'mdc')
+ if name_override != None:
+ self.name = "lov_%s" % name_override
self.add_lustre_module('lov', 'lov')
self.mds_uuid = self.db.get_first_ref('mds')
- mds= self.db.lookup(self.mds_uuid)
- self.mds_name = mds.getName()
- self.stripe_sz = self.db.get_val_int('stripesize', 65536)
+ self.stripe_sz = self.db.get_val_int('stripesize', 1048576)
self.stripe_off = self.db.get_val_int('stripeoffset', 0)
self.pattern = self.db.get_val_int('stripepattern', 0)
- self.devlist = self.db.get_refs('obd')
+ self.devlist = self.db.get_lov_tgts('lov_tgt')
self.stripe_cnt = self.db.get_val_int('stripecount', len(self.devlist))
self.osclist = []
- self.client_uuid = generate_client_uuid(self.name)
+ self.desc_uuid = self.uuid
+ self.uuid = generate_client_uuid(self.name)
self.fs_name = fs_name
- self.mdc_name = ''
- self.mdc = get_mdc(db, self.client_uuid, fs_name, self.mds_uuid)
- for obd_uuid in self.devlist:
+ if config_only:
+ self.config_only = 1
+ return
+ self.config_only = None
+ mds = self.db.lookup(self.mds_uuid)
+ self.mds_name = mds.getName()
+ for (obd_uuid, index, gen, active) in self.devlist:
+ if obd_uuid == '':
+ continue
obd = self.db.lookup(obd_uuid)
- osc = get_osc(obd, self.client_uuid, fs_name)
+ osc = get_osc(obd, self.uuid, fs_name)
if osc:
- self.osclist.append(osc)
+ self.osclist.append((osc, index, gen, active))
else:
panic('osc not found:', obd_uuid)
-
+ def get_uuid(self):
+ return self.uuid
+ def get_name(self):
+ return self.name
def prepare(self):
- if is_prepared(self.name):
+ if not config.record and is_prepared(self.name):
return
- for osc in self.osclist:
+ self.info(self.mds_uuid, self.stripe_cnt, self.stripe_sz,
+ self.stripe_off, self.pattern, self.devlist,
+ self.mds_name)
+ lctl.lov_setup(self.name, self.uuid, self.desc_uuid, self.stripe_cnt,
+ self.stripe_sz, self.stripe_off, self.pattern)
+ for (osc, index, gen, active) in self.osclist:
+ target_uuid = osc.target_uuid
try:
# Only ignore connect failures with --force, which
# isn't implemented here yet.
+ osc.active = active
osc.prepare(ignore_connect_failure=0)
except CommandError, e:
print "Error preparing OSC %s\n" % osc.uuid
raise e
- self.mdc.prepare()
- self.mdc_name = self.mdc.name
- self.info(self.mds_uuid, self.stripe_cnt, self.stripe_sz,
- self.stripe_off, self.pattern, self.devlist, self.mds_name)
- lctl.newdev(attach="lov %s %s" % (self.name, self.uuid),
- setup ="%s" % (self.mdc_name))
+ lctl.lov_add_obd(self.name, self.uuid, target_uuid, index, gen)
def cleanup(self):
+ for (osc, index, gen, active) in self.osclist:
+ target_uuid = osc.target_uuid
+ osc.cleanup()
if is_prepared(self.name):
Module.cleanup(self)
- for osc in self.osclist:
- osc.cleanup()
- mdc = get_mdc(self.db, self.client_uuid, self.fs_name, self.mds_uuid)
- mdc.cleanup()
+ if self.config_only:
+ panic("Can't clean up config_only LOV ", self.name)
def load_module(self):
- for osc in self.osclist:
+ if self.config_only:
+ panic("Can't load modules for config_only LOV ", self.name)
+ for (osc, index, gen, active) in self.osclist:
osc.load_module()
break
Module.load_module(self)
def cleanup_module(self):
+ if self.config_only:
+ panic("Can't cleanup modules for config_only LOV ", self.name)
Module.cleanup_module(self)
- for osc in self.osclist:
- osc.cleanup_module()
+ for (osc, index, gen, active) in self.osclist:
+ if active:
+ osc.cleanup_module()
break
-class LOVConfig(Module):
- def __init__(self, db):
- Module.__init__(self, 'LOVConfig', db)
-
- self.lov_uuid = self.db.get_first_ref('lov')
- l = self.db.lookup(self.lov_uuid)
- self.lov = LOV(l, "YOU_SHOULD_NEVER_SEE_THIS_UUID", '')
-
+ def correct_level(self, level, op=None):
+ return level
+
+class LMV(Module):
+ def __init__(self, db, uuid, fs_name, name_override = None):
+ Module.__init__(self, 'LMV', db)
+ if name_override != None:
+ self.name = "lmv_%s" % name_override
+ self.add_lustre_module('lmv', 'lmv')
+ self.devlist = self.db.get_refs('mds')
+ self.mdclist = []
+ self.desc_uuid = self.uuid
+ self.uuid = uuid
+ self.fs_name = fs_name
+ for mds_uuid in self.devlist:
+ mds = self.db.lookup(mds_uuid)
+ if not mds:
+ panic("MDS not found!")
+ mdc = MDC(mds, self.uuid, fs_name)
+ if mdc:
+ self.mdclist.append(mdc)
+ else:
+ panic('mdc not found:', mds_uuid)
+
def prepare(self):
- lov = self.lov
- self.info(lov.mds_uuid, lov.stripe_cnt, lov.stripe_sz, lov.stripe_off,
- lov.pattern, lov.devlist, lov.mds_name)
- lctl.lov_setconfig(lov.uuid, lov.mds_name, lov.stripe_cnt,
- lov.stripe_sz, lov.stripe_off, lov.pattern,
- string.join(lov.devlist))
+ if is_prepared(self.name):
+ return
+ for mdc in self.mdclist:
+ try:
+ # Only ignore connect failures with --force, which
+ # isn't implemented here yet.
+ mdc.prepare(ignore_connect_failure=0)
+ except CommandError, e:
+ print "Error preparing LMV %s\n" % mdc.uuid
+ raise e
+ lctl.lmv_setup(self.name, self.uuid, self.desc_uuid,
+ string.join(self.devlist))
def cleanup(self):
- #nothing to do here
- pass
+ for mdc in self.mdclist:
+ mdc.cleanup()
+ if is_prepared(self.name):
+ Module.cleanup(self)
+
+ def load_module(self):
+ for mdc in self.mdclist:
+ mdc.load_module()
+ break
+ Module.load_module(self)
+
+ def cleanup_module(self):
+ Module.cleanup_module(self)
+ for mdc in self.mdclist:
+ mdc.cleanup_module()
+ break
+
+ def correct_level(self, level, op=None):
+ return level
class MDSDEV(Module):
def __init__(self,db):
Module.__init__(self, 'MDSDEV', db)
self.devpath = self.db.get_val('devpath','')
+ self.backdevpath = self.db.get_val('backdevpath','')
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', '')
+ self.backfstype = self.db.get_val('backfstype', '')
self.nspath = self.db.get_val('nspath', '')
- # overwrite the orignal MDSDEV name and uuid with the MDS name and uuid
+ self.mkfsoptions = self.db.get_val('mkfsoptions', '')
+ self.mountfsoptions = self.db.get_val('mountfsoptions', '')
+ self.cachetype = self.db.get_val('cachetype', '')
+ # overwrite the orignal MDSDEV name and uuid with the MDS name and uuid
target_uuid = self.db.get_first_ref('target')
mds = self.db.lookup(target_uuid)
self.name = mds.getName()
- self.lovconfig_uuids = mds.get_refs('lovconfig')
self.filesystem_uuids = mds.get_refs('filesystem')
+ self.lmv_uuid = ''
+ self.lmv = ''
+ self.master_mds = ""
+ if not self.filesystem_uuids:
+ self.lmv_uuid = self.db.get_first_ref('lmv')
+ if not self.lmv_uuid:
+ panic("ALERT: can't find lvm uuid")
+ if self.lmv_uuid:
+ self.lmv = self.db.lookup(self.lmv_uuid)
+ if self.lmv:
+ self.filesystem_uuids = self.lmv.get_refs('filesystem')
+ self.master_mds = self.lmv_uuid
# 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.active = 1
else:
self.active = 0
- if self.active and config.group and config.group != ost.get_val('group'):
+ if self.active and config.group and config.group != mds.get_val('group'):
self.active = 0
+ self.inode_size = self.db.get_val_int('inodesize', 0)
+ if self.inode_size == 0:
+ # find the LOV for this MDS
+ lovconfig_uuid = mds.get_first_ref('lovconfig')
+ if not lovconfig_uuid:
+ if not self.lmv_uuid:
+ panic("No LOV found for lovconfig ", lovconfig.name)
+
+ if not self.lmv:
+ panic("No LMV initialized and not lovconfig_uuid found")
+
+ lovconfig_uuid = self.lmv.get_first_ref('lovconfig')
+ lovconfig = self.lmv.lookup(lovconfig_uuid)
+ lov_uuid = lovconfig.get_first_ref('lov')
+ if not lov_uuid:
+ panic("No LOV found for lovconfig ", lovconfig.name)
+ else:
+ lovconfig = mds.lookup(lovconfig_uuid)
+ lov_uuid = lovconfig.get_first_ref('lov')
+ if not lov_uuid:
+ panic("No LOV found for lovconfig ", lovconfig.name)
+
+ if self.lmv:
+ lovconfig_uuid = self.lmv.get_first_ref('lovconfig')
+ lovconfig = self.lmv.lookup(lovconfig_uuid)
+ lov_uuid = lovconfig.get_first_ref('lov')
+
+ lov = LOV(self.db.lookup(lov_uuid), lov_uuid, 'FS_name', config_only = 1)
+
+ # default stripe count controls default inode_size
+ stripe_count = lov.stripe_cnt
+ if stripe_count > 77:
+ self.inode_size = 4096
+ elif stripe_count > 35:
+ self.inode_size = 2048
+ elif stripe_count > 13:
+ self.inode_size = 1024
+ elif stripe_count > 3:
+ self.inode_size = 512
+ else:
+ self.inode_size = 256
+
self.target_dev_uuid = self.uuid
self.uuid = target_uuid
+ # setup LMV
+ if self.master_mds:
+ client_uuid = generate_client_uuid(self.name)
+ client_uuid = self.name + "_lmv_" + "UUID"
+ self.master = LMV(self.db.lookup(self.lmv_uuid), client_uuid, self.name, self.name)
+ self.master_mds = self.master.name
+
# modules
+ self.add_lustre_module('mdc', 'mdc')
+ self.add_lustre_module('osc', 'osc')
+ self.add_lustre_module('lov', 'lov')
+ self.add_lustre_module('lmv', 'lmv')
+ self.add_lustre_module('ost', 'ost')
self.add_lustre_module('mds', 'mds')
- if self.fstype:
- self.add_lustre_module('obdclass', 'fsfilt_%s' % (self.fstype))
+ if self.fstype == 'smfs':
+ self.add_lustre_module('smfs', 'smfs')
+
+ if self.fstype == 'ldiskfs':
+ self.add_lustre_module('ldiskfs', 'ldiskfs')
+
+ if self.fstype:
+ self.add_lustre_module('lvfs', 'fsfilt_%s' % (self.fstype))
+
+ # if fstype is smfs, then we should also take care about backing
+ # store fs.
+ if self.fstype == 'smfs':
+ self.add_lustre_module('lvfs', 'fsfilt_%s' % (self.backfstype))
+
+ for options in string.split(self.mountfsoptions, ','):
+ if options == 'snap':
+ if not self.fstype == 'smfs':
+ panic("mountoptions with snap, but fstype is not smfs\n")
+ self.add_lustre_module('lvfs', 'fsfilt_snap_%s' % (self.fstype))
+ self.add_lustre_module('lvfs', 'fsfilt_snap_%s' % (self.backfstype))
def load_module(self):
if self.active:
Module.load_module(self)
-
+
def prepare(self):
- if is_prepared(self.name):
+ if not config.record and is_prepared(self.name):
return
if not self.active:
debug(self.uuid, "not active")
return
- self.info(self.devpath, self.fstype, self.format)
+ if config.reformat:
+ # run write_conf automatically, if --reformat used
+ self.write_conf()
+ self.info(self.devpath, self.fstype, self.size, self.format)
run_acceptors()
- blkdev = block_dev(self.devpath, self.size, self.fstype, self.format,
- self.journal_size)
+ # prepare LMV
+ if self.master_mds:
+ self.master.prepare()
+ # never reformat here
+ blkdev = block_dev(self.devpath, self.size, self.fstype, 0,
+ self.format, self.journal_size, self.inode_size,
+ self.mkfsoptions, self.backfstype, self.backdevpath)
+
if not is_prepared('MDT'):
- lctl.newdev(attach="mdt %s %s" % ('MDT', 'MDT_UUID'),
- setup ="")
- if self.nspath:
- run ("mkdir", self.nspath)
- lctl.newdev(attach="mds %s %s" % (self.name, self.uuid),
- setup ="%s %s %s" %(blkdev, self.fstype, self.nspath))
- for uuid in self.lovconfig_uuids:
- db = self.db.lookup(uuid)
- lovconfig = LOVConfig(db)
- lovconfig.prepare()
- if config.mds_ost_conn:
- for uuid in self.filesystem_uuids:
- log("open clients for filesystem:", uuid)
- fs = self.db.lookup(uuid)
- obd_uuid = fs.get_first_ref('obd')
- client_uuid = generate_client_uuid(self.name)
- client = VOSC(client_uuid, self.db.lookup(obd_uuid), self.name)
- client.prepare()
+ lctl.newdev("mdt", 'MDT', 'MDT_UUID', setup ="")
+ try:
+ mountfsoptions = def_mount_options(self.fstype, 'mds')
+
+ if config.mountfsoptions:
+ if mountfsoptions:
+ mountfsoptions = mountfsoptions + ',' + config.mountfsoptions
+ else:
+ mountfsoptions = config.mountfsoptions
+ if self.mountfsoptions:
+ mountfsoptions = mountfsoptions + ',' + self.mountfsoptions
+ else:
+ if self.mountfsoptions:
+ if mountfsoptions:
+ mountfsoptions = mountfsoptions + ',' + self.mountfsoptions
+ else:
+ mountfsoptions = self.mountfsoptions
+
+ if self.fstype == 'smfs':
+ realdev = self.fstype
+
+ if mountfsoptions:
+ mountfsoptions = "%s,type=%s,dev=%s" % (mountfsoptions,
+ self.backfstype,
+ blkdev)
+ else:
+ mountfsoptions = "type=%s,dev=%s" % (self.backfstype,
+ blkdev)
+ else:
+ realdev = blkdev
+
+ print 'MDS mount options: ' + mountfsoptions
+
+ if not self.master_mds:
+ self.master_mds = 'dumb'
+ if not self.cachetype:
+ self.cachetype = 'dumb'
+ lctl.newdev("mds", self.name, self.uuid,
+ setup ="%s %s %s %s %s %s" %(realdev, self.fstype,
+ self.name, mountfsoptions,
+ self.master_mds, self.cachetype))
+ except CommandError, e:
+ if e.rc == 2:
+ panic("MDS is missing the config log. Need to run " +
+ "lconf --write_conf.")
+ else:
+ raise e
+
+ def write_conf(self):
+ do_cleanup = 0
+ if not is_prepared(self.name):
+ self.info(self.devpath, self.fstype, self.format)
+
+ blkdev = block_dev(self.devpath, self.size, self.fstype,
+ config.reformat, self.format, self.journal_size,
+ self.inode_size, self.mkfsoptions,
+ self.backfstype, self.backdevpath)
+
+ # Even for writing logs we mount mds with supplied mount options
+ # because it will not mount smfs (if used) otherwise.
+
+ mountfsoptions = def_mount_options(self.fstype, 'mds')
+
+ if config.mountfsoptions:
+ if mountfsoptions:
+ mountfsoptions = mountfsoptions + ',' + config.mountfsoptions
+ else:
+ mountfsoptions = config.mountfsoptions
+ if self.mountfsoptions:
+ mountfsoptions = mountfsoptions + ',' + self.mountfsoptions
+ else:
+ if self.mountfsoptions:
+ if mountfsoptions:
+ mountfsoptions = mountfsoptions + ',' + self.mountfsoptions
+ else:
+ mountfsoptions = self.mountfsoptions
+
+ if self.fstype == 'smfs':
+ realdev = self.fstype
+ if mountfsoptions:
+ mountfsoptions = "%s,type=%s,dev=%s" % (mountfsoptions,
+ self.backfstype,
+ blkdev)
+ else:
+ mountfsoptions = "type=%s,dev=%s" % (self.backfstype,
+ blkdev)
+ else:
+ realdev = blkdev
+
+ print 'MDS mount options: ' + mountfsoptions
+
+ # As mount options are passed by 4th param to config tool, we need
+ # to pass something in 3rd param. But we do not want this 3rd param
+ # be counted as a profile name for reading log on MDS setup, thus,
+ # we pass there some predefined sign like 'dumb', which will be
+ # checked in MDS code and skipped. Probably there is more nice way
+ # like pass empty string and check it in config tool and pass null
+ # as 4th param.
+ lctl.newdev("mds", self.name, self.uuid,
+ setup ="%s %s %s %s" %(realdev, self.fstype,
+ 'dumb', mountfsoptions))
+ do_cleanup = 1
+
+ # record logs for the MDS lov
+ for uuid in self.filesystem_uuids:
+ log("recording clients for filesystem:", uuid)
+ fs = self.db.lookup(uuid)
+
+ # this is ugly, should be organized nice later.
+ target_uuid = self.db.get_first_ref('target')
+ mds = self.db.lookup(target_uuid)
+ lovconfig_uuid = mds.get_first_ref('lovconfig')
+ if lovconfig_uuid:
+ lovconfig = mds.lookup(lovconfig_uuid)
+ obd_uuid = lovconfig.get_first_ref('lov')
+ else:
+ obd_uuid = fs.get_first_ref('obd')
+
+ client_uuid = generate_client_uuid(self.name)
+ client = VOSC(self.db.lookup(obd_uuid), client_uuid, self.name,
+ self.name)
+ config.record = 1
+ lctl.clear_log(self.name, self.name)
+ lctl.record(self.name, self.name)
+ client.prepare()
+ lctl.mount_option(self.name, client.get_name(), "")
+ lctl.end_record()
+ process_updates(self.db, self.name, self.name, client)
+
+ config.cleanup = 1
+ lctl.clear_log(self.name, self.name + '-clean')
+ lctl.record(self.name, self.name + '-clean')
+ client.cleanup()
+ lctl.del_mount_option(self.name)
+ lctl.end_record()
+ process_updates(self.db, self.name, self.name + '-clean', client)
+ config.cleanup = 0
+ config.record = 0
+
+ # record logs for each client
+ if config.noexec:
+ noexec_opt = '-n'
+ else:
+ noexec_opt = ''
+ if config.ldapurl:
+ config_options = "--ldapurl " + config.ldapurl + " --config " + config.config
+ else:
+ config_options = CONFIG_FILE
+
+ for node_db in self.db.lookup_class('node'):
+ client_name = node_db.getName()
+ for prof_uuid in node_db.get_refs('profile'):
+ prof_db = node_db.lookup(prof_uuid)
+ # refactor this into a funtion to test "clientness"
+ # of a node.
+ for ref_class, ref_uuid in prof_db.get_all_refs():
+ if ref_class in ('mountpoint','echoclient'):
+ debug("recording", client_name)
+ old_noexec = config.noexec
+ config.noexec = 0
+ ret, out = run (sys.argv[0], noexec_opt,
+ " -v --record --nomod",
+ "--record_log", client_name,
+ "--record_device", self.name,
+ "--node", client_name,
+ config_options)
+ if config.verbose:
+ for s in out: log("record> ", string.strip(s))
+ ret, out = run (sys.argv[0], noexec_opt,
+ "--cleanup -v --record --nomod",
+ "--record_log", client_name + "-clean",
+ "--record_device", self.name,
+ "--node", client_name,
+ config_options)
+ if config.verbose:
+ for s in out: log("record> ", string.strip(s))
+ config.noexec = old_noexec
+ if do_cleanup:
+ try:
+ lctl.cleanup(self.name, self.uuid, 0, 0)
+ except CommandError, e:
+ log(self.module_name, "cleanup failed: ", self.name)
+ e.dump()
+ cleanup_error(e.rc)
+ Module.cleanup(self)
+
+ if self.fstype == 'smfs':
+ clean_loop(self.backdevpath)
+ else:
+ clean_loop(self.devpath)
+
def msd_remaining(self):
out = lctl.device_list()
for s in out:
if not self.active:
debug(self.uuid, "not active")
return
+ self.info()
if is_prepared(self.name):
- self.info()
try:
lctl.cleanup(self.name, self.uuid, config.force,
config.failover)
e.dump()
cleanup_error(e.rc)
Module.cleanup(self)
- if config.mds_ost_conn:
- for uuid in self.filesystem_uuids:
- log("clean clients for filesystem:", uuid)
- log("open clients for filesystem:", uuid)
- fs = self.db.lookup(uuid)
- obd_uuid = fs.get_first_ref('obd')
- client = VOSC(self.db.lookup(obd_uuid), self.name)
- client.cleanup()
+ # cleanup LMV
+ if self.master_mds:
+ self.master.cleanup()
if not self.msd_remaining() and is_prepared('MDT'):
try:
lctl.cleanup("MDT", "MDT_UUID", config.force,
print "cleanup failed: ", self.name
e.dump()
cleanup_error(e.rc)
- clean_loop(self.devpath)
+
+ if self.fstype == 'smfs':
+ clean_loop(self.backdevpath)
+ else:
+ clean_loop(self.devpath)
+
+ def correct_level(self, level, op=None):
+ #if self.master_mds:
+ # level = level + 2
+ return level
class OSD(Module):
def __init__(self, db):
Module.__init__(self, 'OSD', db)
self.osdtype = self.db.get_val('osdtype')
self.devpath = self.db.get_val('devpath', '')
+ self.backdevpath = self.db.get_val('backdevpath', '')
self.size = self.db.get_val_int('devsize', 0)
self.journal_size = self.db.get_val_int('journalsize', 0)
+ self.inode_size = self.db.get_val_int('inodesize', 0)
+ self.mkfsoptions = self.db.get_val('mkfsoptions', '')
+ self.mountfsoptions = self.db.get_val('mountfsoptions', '')
self.fstype = self.db.get_val('fstype', '')
+ self.backfstype = self.db.get_val('backfstype', '')
self.nspath = self.db.get_val('nspath', '')
target_uuid = self.db.get_first_ref('target')
ost = self.db.lookup(target_uuid)
self.active = 0
if self.active and config.group and config.group != ost.get_val('group'):
self.active = 0
-
+
self.target_dev_uuid = self.uuid
self.uuid = target_uuid
# modules
self.add_lustre_module('ost', 'ost')
- # FIXME: should we default to ext3 here?
+ if self.fstype == 'smfs':
+ self.add_lustre_module('smfs', 'smfs')
+ # FIXME: should we default to ext3 here?
+ if self.fstype == 'ldiskfs':
+ self.add_lustre_module('ldiskfs', 'ldiskfs')
if self.fstype:
- self.add_lustre_module('obdclass' , 'fsfilt_%s' % (self.fstype))
+ self.add_lustre_module('lvfs' , 'fsfilt_%s' % (self.fstype))
+ if self.fstype == 'smfs':
+ self.add_lustre_module('lvfs' , 'fsfilt_%s' % (self.backfstype))
+
+ for options in self.mountfsoptions:
+ if options == 'snap':
+ if not self.fstype == 'smfs':
+ panic("mountoptions with snap, but fstype is not smfs\n")
+ self.add_lustre_module('lvfs', 'fsfilt_snap_%s' % (self.fstype))
+ self.add_lustre_module('lvfs', 'fsfilt_snap_%s' % (self.backfstype))
+
self.add_lustre_module(self.osdtype, self.osdtype)
def load_module(self):
debug(self.uuid, "not active")
return
self.info(self.osdtype, self.devpath, self.size, self.fstype,
- self.format, self.journal_size)
+ self.format, self.journal_size, self.inode_size)
run_acceptors()
if self.osdtype == 'obdecho':
blkdev = ''
else:
blkdev = block_dev(self.devpath, self.size, self.fstype,
- self.format, self.journal_size)
- if self.nspath:
- run ("mkdir", self.nspath)
- lctl.newdev(attach="%s %s %s" % (self.osdtype, self.name, self.uuid),
- setup ="%s %s %s %s" %(blkdev, self.fstype,
- self.failover_ost, self.nspath))
+ config.reformat, self.format, self.journal_size,
+ self.inode_size, self.mkfsoptions, self.backfstype,
+ self.backdevpath)
+
+ mountfsoptions = def_mount_options(self.fstype, 'ost')
+
+ if config.mountfsoptions:
+ if mountfsoptions:
+ mountfsoptions = mountfsoptions + ',' + config.mountfsoptions
+ else:
+ mountfsoptions = config.mountfsoptions
+ if self.mountfsoptions:
+ mountfsoptions = mountfsoptions + ',' + self.mountfsoptions
+ else:
+ if self.mountfsoptions:
+ if mountfsoptions:
+ mountfsoptions = mountfsoptions + ',' + self.mountfsoptions
+ else:
+ mountfsoptions = self.mountfsoptions
+
+ if self.fstype == 'smfs':
+ realdev = self.fstype
+
+ if mountfsoptions:
+ mountfsoptions = "%s,type=%s,dev=%s" % (mountfsoptions,
+ self.backfstype,
+ blkdev)
+ else:
+ mountfsoptions = "type=%s,dev=%s" % (self.backfstype,
+ blkdev)
+ else:
+ realdev = blkdev
+
+ print 'OSD mount options: ' + mountfsoptions
+
+ lctl.newdev(self.osdtype, self.name, self.uuid,
+ setup ="%s %s %s %s" %(realdev, self.fstype,
+ self.failover_ost,
+ mountfsoptions))
if not is_prepared('OSS'):
- lctl.newdev(attach="ost %s %s" % ('OSS', 'OSS_UUID'),
- setup ="")
+ lctl.newdev("ost", 'OSS', 'OSS_UUID', setup ="")
def osd_remaining(self):
out = lctl.device_list()
e.dump()
cleanup_error(e.rc)
if not self.osdtype == 'obdecho':
- clean_loop(self.devpath)
+ if self.fstype == 'smfs':
+ clean_loop(self.backdevpath)
+ else:
+ clean_loop(self.devpath)
+
+ def correct_level(self, level, op=None):
+ return level
def mgmt_uuid_for_fs(mtpt_name):
if not mtpt_name:
return ''
- mtpt_db = toplevel.lookup_name(mtpt_name)
+ mtpt_db = toplustreDB.lookup_name(mtpt_name)
fs_uuid = mtpt_db.get_first_ref('filesystem')
- fs = toplevel.lookup(fs_uuid)
+ fs = toplustreDB.lookup(fs_uuid)
if not fs:
return ''
return fs.get_first_ref('mgmt')
self.target_name = tgtdb.getName()
self.target_uuid = tgtdb.getUUID()
self.db = tgtdb
+ self.active = 1
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.kmodule_list = []
+ panic("No target device found for target(1):", self.target_name)
+
+ self.kmod = kmod(config.lustre, config.portals)
self._server = None
self._connected = 0
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)
-
+ def get_name(self):
+ return self.name
def get_servers(self):
return self._server_nets
def prepare(self, ignore_connect_failure = 0):
self.info(self.target_uuid)
- if is_prepared(self.name):
+ if not config.record and is_prepared(self.name):
self.cleanup()
try:
srv = choose_local_server(self.get_servers())
if srv:
lctl.connect(srv)
else:
- srv, r = find_route(self.get_servers())
- if srv:
- lctl.add_route_host(r[0], srv.uuid, r[1], r[3])
- else:
+ routes = find_route(self.get_servers())
+ if len(routes) == 0:
panic ("no route to", self.target_uuid)
+ for (srv, r) in routes:
+ lctl.add_route_host(r[0], srv.nid_uuid, r[1], r[3])
except CommandError, e:
if not ignore_connect_failure:
raise e
if srv:
- lctl.newdev(attach="%s %s %s" % (self.module, self.name, self.uuid),
- setup ="%s %s %s" % (self.target_uuid, srv.uuid,
- self.mgmt_name))
+ if self.permits_inactive() and (self.target_uuid in config.inactive or self.active == 0):
+ debug("%s inactive" % self.target_uuid)
+ inactive_p = "inactive"
+ else:
+ debug("%s active" % self.target_uuid)
+ inactive_p = ""
+ lctl.newdev(self.module, self.name, self.uuid,
+ setup ="%s %s %s %s" % (self.target_uuid, srv.nid_uuid,
+ inactive_p, self.mgmt_name))
def cleanup(self):
if is_prepared(self.name):
if srv:
lctl.disconnect(srv)
else:
- srv, r = find_route(self.get_servers())
- if srv:
- lctl.del_route_host(r[0], srv.uuid, r[1], r[3])
+ for (srv, r) in find_route(self.get_servers()):
+ lctl.del_route_host(r[0], srv.nid_uuid, r[1], r[3])
except CommandError, e:
log(self.module_name, "cleanup failed: ", self.name)
e.dump()
cleanup_error(e.rc)
+ def correct_level(self, level, op=None):
+ return level
+
+ def deactivate(self):
+ try:
+ lctl.deactivate(self.name)
+ except CommandError, e:
+ log(self.module_name, "deactivate failed: ", self.name)
+ e.dump()
+ cleanup_error(e.rc)
class MDC(Client):
def __init__(self, db, uuid, fs_name):
Client.__init__(self, db, uuid, 'mdc', fs_name)
+ def permits_inactive(self):
+ return 0
class OSC(Client):
def __init__(self, db, uuid, fs_name):
Client.__init__(self, db, uuid, 'osc', fs_name)
+ def permits_inactive(self):
+ return 1
+
def mgmtcli_name_for_uuid(uuid):
return 'MGMTCLI_%s' % uuid
Client.__init__(self, db, uuid, 'mgmt_cli', '',
self_name = mgmtcli_name_for_uuid(db.getUUID()),
module_dir = 'mgmt')
-
+class VLOV(Module):
+ def __init__(self, db, uuid, fs_name, name_override = None, config_only = None):
+ Module.__init__(self, 'VLOV', db)
+ if name_override != None:
+ self.name = "lov_%s" % name_override
+ self.add_lustre_module('lov', 'lov')
+ self.stripe_sz = 65536
+ self.stripe_off = 0
+ self.pattern = 0
+ self.stripe_cnt = 1
+ self.desc_uuid = self.uuid
+ self.uuid = generate_client_uuid(self.name)
+ self.fs_name = fs_name
+ self.osc = get_osc(db, self.uuid, fs_name)
+ if not self.osc:
+ panic('osc not found:', self.uuid)
+ if config_only:
+ self.config_only = 1
+ return
+ self.config_only = None
+ def get_uuid(self):
+ return self.uuid
+ def get_name(self):
+ return self.name
+ def prepare(self):
+ if not config.record and is_prepared(self.name):
+ return
+ lctl.lov_setup(self.name, self.uuid, self.desc_uuid, self.stripe_cnt,
+ self.stripe_sz, self.stripe_off, self.pattern)
+ target_uuid = self.osc.target_uuid
+ try:
+ self.osc.active = 1
+ self.osc.prepare(ignore_connect_failure=0)
+ except CommandError, e:
+ print "Error preparing OSC %s\n" % osc.uuid
+ raise e
+ lctl.lov_add_obd(self.name, self.uuid, target_uuid, 0, 1)
+
+ def cleanup(self):
+ target_uuid = self.osc.target_uuid
+ self.osc.cleanup()
+ if is_prepared(self.name):
+ Module.cleanup(self)
+ if self.config_only:
+ panic("Can't clean up config_only LOV ", self.name)
+
+ def load_module(self):
+ if self.config_only:
+ panic("Can't load modules for config_only LOV ", self.name)
+ self.osc.load_module()
+ Module.load_module(self)
+
+ def cleanup_module(self):
+ if self.config_only:
+ panic("Can't cleanup modules for config_only LOV ", self.name)
+ Module.cleanup_module(self)
+ self.osc.cleanup_module()
+
+ def correct_level(self, level, op=None):
+ return level
+
+class CMOBD(Module):
+ def __init__(self,db):
+ Module.__init__(self, 'CMOBD', db)
+ self.name = self.db.getName();
+ self.uuid = generate_client_uuid(self.name)
+ self.master_uuid = self.db.get_first_ref('masterobd')
+ self.cache_uuid = self.db.get_first_ref('cacheobd')
+ self.add_lustre_module('cmobd', 'cmobd')
+ master_obd = self.db.lookup(self.master_uuid)
+ if not master_obd:
+ panic('master obd not found:', self.master_uuid)
+ cache_obd = self.db.lookup(self.cache_uuid)
+ if not cache_obd:
+ panic('cache obd not found:', self.cache_uuid)
+
+ if master_obd.get_class() == 'ost':
+ self.client_uuid = generate_client_uuid(self.name)
+ self.master= VLOV(master_obd, self.client_uuid, self.name,
+ "%s_master" % (self.name))
+ self.master_uuid = self.master.get_uuid()
+ else:
+ self.master = get_mdc(db, self.name, self.master_uuid)
+ # need to check /proc/mounts and /etc/mtab before
+ # formatting anything.
+ # FIXME: check if device is already formatted.
+ def prepare(self):
+ self.master.prepare()
+ if not config.record and is_prepared(self.name):
+ return
+ self.info(self.master_uuid, self.cache_uuid)
+ lctl.newdev("cmobd", self.name, self.uuid,
+ setup ="%s %s" %(self.master_uuid,
+ self.cache_uuid))
+
+ def cleanup(self):
+ if is_prepared(self.name):
+ Module.cleanup(self)
+ self.master.cleanup()
+
+ def load_module(self):
+ self.master.load_module()
+ Module.load_module(self)
+
+ def cleanup_module(self):
+ Module.cleanup_module(self)
+ self.master.cleanup_module()
+
+ def correct_level(self, level, op=None):
+ return level
+
class COBD(Module):
- def __init__(self, db):
+ def __init__(self, db, uuid, name, type, name_override = None):
Module.__init__(self, 'COBD', db)
+ self.name = self.db.getName();
+ self.uuid = generate_client_uuid(self.name)
self.real_uuid = self.db.get_first_ref('realobd')
self.cache_uuid = self.db.get_first_ref('cacheobd')
- self.add_lustre_module('cobd' , 'cobd')
-
+ self.add_lustre_module('cobd', 'cobd')
+ real_obd = self.db.lookup(self.real_uuid)
+ if not real_obd:
+ panic('real obd not found:', self.real_uuid)
+ cache_obd = self.db.lookup(self.cache_uuid)
+ if not cache_obd:
+ panic('cache obd not found:', self.cache_uuid)
+ if type == 'obd':
+ self.real = LOV(real_obd, self.real_uuid, name,
+ "%s_real" % (self.name));
+ self.cache = LOV(cache_obd, self.cache_uuid, name,
+ "%s_cache" % (self.name));
+ else:
+ self.real = get_mdc(db, name, self.real_uuid)
+ self.cache = get_mdc(db, name, self.cache_uuid)
# need to check /proc/mounts and /etc/mtab before
# formatting anything.
# FIXME: check if device is already formatted.
+ def get_uuid(self):
+ return self.uuid
+ def get_name(self):
+ return self.name
+ def get_real_name(self):
+ return self.real.name
+ def get_cache_name(self):
+ return self.cache.name
def prepare(self):
- if is_prepared(self.name):
+ self.real.prepare()
+ self.cache.prepare()
+ if not config.record and is_prepared(self.name):
return
self.info(self.real_uuid, self.cache_uuid)
- lctl.newdev(attach="cobd %s %s" % (self.name, self.uuid),
- setup ="%s %s" %(self.real_uuid, self.cache_uuid))
+ lctl.newdev("cobd", self.name, self.uuid,
+ setup ="%s %s" %(self.real.name,
+ self.cache.name))
+ def cleanup(self):
+ if is_prepared(self.name):
+ Module.cleanup(self)
+ self.real.cleanup()
+ self.cache.cleanup()
+
+ def load_module(self):
+ self.real.load_module()
+ Module.load_module(self)
+
+ def cleanup_module(self):
+ Module.cleanup_module(self)
+ self.real.cleanup_module()
# virtual interface for OSC and LOV
class VOSC(Module):
- def __init__(self, db, uuid, fs_name):
+ def __init__(self, db, client_uuid, name, name_override = None):
Module.__init__(self, 'VOSC', db)
if db.get_class() == 'lov':
- self.osc = LOV(db, uuid, fs_name)
+ self.osc = LOV(db, client_uuid, name, name_override)
+ self.type = 'lov'
+ elif db.get_class() == 'cobd':
+ self.osc = COBD(db, client_uuid, name, 'obd')
+ self.type = 'cobd'
else:
- self.osc = get_osc(db, uuid, fs_name)
+ self.osc = OSC(db, client_uuid, name)
+ self.type = 'osc'
def get_uuid(self):
- return self.osc.uuid
+ return self.osc.get_uuid()
def get_name(self):
- return self.osc.name
+ return self.osc.get_name()
def prepare(self):
self.osc.prepare()
def cleanup(self):
self.osc.load_module()
def cleanup_module(self):
self.osc.cleanup_module()
- def need_mdc(self):
- return self.db.get_class() != 'lov'
- def get_mdc_name(self):
- if self.db.get_class() == 'lov':
- return self.osc.mdc_name
- return ''
-
+ def correct_level(self, level, op=None):
+ return self.osc.correct_level(level, op)
+
+# virtual interface for MDC and LMV
+class VMDC(Module):
+ def __init__(self, db, client_uuid, name, name_override = None):
+ Module.__init__(self, 'VMDC', db)
+ if db.get_class() == 'lmv':
+ self.mdc = LMV(db, client_uuid, name)
+ elif db.get_class() == 'cobd':
+ self.mdc = COBD(db, client_uuid, name, 'mds')
+ else:
+ self.mdc = MDC(db, client_uuid, name)
+ def get_uuid(self):
+ return self.mdc.uuid
+ def get_name(self):
+ return self.mdc.name
+ def prepare(self):
+ self.mdc.prepare()
+ def cleanup(self):
+ self.mdc.cleanup()
+ def load_module(self):
+ self.mdc.load_module()
+ def cleanup_module(self):
+ self.mdc.cleanup_module()
+ def correct_level(self, level, op=None):
+ return self.mdc.correct_level(level, op)
class ECHO_CLIENT(Module):
def __init__(self,db):
self.add_lustre_module('obdecho', 'obdecho')
self.obd_uuid = self.db.get_first_ref('obd')
obd = self.db.lookup(self.obd_uuid)
+ self.uuid = generate_client_uuid(self.name)
self.osc = VOSC(obd, self.uuid, self.name)
def prepare(self):
- if is_prepared(self.name):
+ if not config.record and is_prepared(self.name):
return
+ run_acceptors()
self.osc.prepare() # XXX This is so cheating. -p
self.info(self.obd_uuid)
- lctl.newdev(attach="echo_client %s %s" % (self.name, self.uuid),
+ lctl.newdev("echo_client", self.name, self.uuid,
setup = self.osc.get_name())
def cleanup(self):
Module.cleanup_module(self)
self.osc.cleanup_module()
+ def correct_level(self, level, op=None):
+ return level
def generate_client_uuid(name):
client_uuid = '%05x_%.19s_%05x%05x' % (int(random.random() * 1048576),
int(random.random() * 1048576))
return client_uuid[:36]
-
class Mountpoint(Module):
def __init__(self,db):
Module.__init__(self, 'MTPT', db)
self.path = self.db.get_val('path')
+ self.clientoptions = self.db.get_val('clientoptions', '')
self.fs_uuid = self.db.get_first_ref('filesystem')
fs = self.db.lookup(self.fs_uuid)
- self.mds_uuid = fs.get_first_ref('mds')
+ self.mds_uuid = fs.get_first_ref('lmv')
+ if not self.mds_uuid:
+ self.mds_uuid = fs.get_first_ref('mds')
self.obd_uuid = fs.get_first_ref('obd')
self.mgmt_uuid = fs.get_first_ref('mgmt')
- obd = self.db.lookup(self.obd_uuid)
client_uuid = generate_client_uuid(self.name)
- self.vosc = VOSC(obd, client_uuid, self.name)
- if self.vosc.need_mdc():
- self.add_lustre_module('mdc', 'mdc')
- self.mdc = get_mdc(db, client_uuid, self.name, self.mds_uuid)
+
+ ost = self.db.lookup(self.obd_uuid)
+ if not ost:
+ panic("no ost: ", self.obd_uuid)
+
+ mds = self.db.lookup(self.mds_uuid)
+ if not mds:
+ panic("no mds: ", self.mds_uuid)
+
+ self.add_lustre_module('mdc', 'mdc')
+ self.add_lustre_module('lmv', 'lmv')
self.add_lustre_module('llite', 'llite')
+
+ self.vosc = VOSC(ost, client_uuid, self.name)
+ self.vmdc = VMDC(mds, client_uuid, self.name)
+
if self.mgmt_uuid:
self.mgmtcli = ManagementClient(db.lookup(self.mgmt_uuid),
client_uuid)
self.mgmtcli = None
def prepare(self):
- if fs_is_mounted(self.path):
+ if not config.record and fs_is_mounted(self.path):
log(self.path, "already mounted.")
return
+ run_acceptors()
if self.mgmtcli:
self.mgmtcli.prepare()
self.vosc.prepare()
- if self.vosc.need_mdc():
- self.mdc.prepare()
- mdc_name = self.mdc.name
- else:
- mdc_name = self.vosc.get_mdc_name()
- if not mdc_name:
- self.vosc.cleanup()
- panic("Unable to determine MDC name. Probably need to cleanup before re-mounting.")
+ self.vmdc.prepare()
+ vmdc_name = self.vmdc.get_name()
+
self.info(self.path, self.mds_uuid, self.obd_uuid)
- if config.lctl_dump:
- cmd = "osc=%s,mdc=%s" % (self.vosc.get_name(), mdc_name)
- lctl.mount_option(cmd)
+ if config.record or config.lctl_dump:
+ lctl.mount_option(local_node_name, self.vosc.get_name(), vmdc_name)
return
- cmd = "mount -t lustre_lite -o osc=%s,mdc=%s %s %s" % \
- (self.vosc.get_name(), mdc_name, config.config, self.path)
+
+ if config.clientoptions:
+ if self.clientoptions:
+ self.clientoptions = self.clientoptions + ',' + \
+ config.clientoptions
+ else:
+ self.clientoptions = config.clientoptions
+ if self.clientoptions:
+ self.clientoptions = ',' + self.clientoptions
+ # Linux kernel will deal with async and not pass it to ll_fill_super,
+ # so replace it with Lustre async
+ self.clientoptions = string.replace(self.clientoptions, "async",
+ "lasync")
+
+ cmd = "mount -t lustre_lite -o osc=%s,mdc=%s%s %s %s" % \
+ (self.vosc.get_name(), vmdc_name, self.clientoptions,
+ config.config, self.path)
run("mkdir", self.path)
ret, val = run(cmd)
if ret:
+ self.vmdc.cleanup()
self.vosc.cleanup()
- if self.vosc.need_mdc():
- self.mdc.cleanup()
panic("mount failed:", self.path, ":", string.join(val))
def cleanup(self):
self.info(self.path, self.mds_uuid,self.obd_uuid)
- if fs_is_mounted(self.path):
- if config.force:
- (rc, out) = run("umount", "-f", self.path)
- else:
- (rc, out) = run("umount", self.path)
- if rc:
- raise CommandError('umount', out, rc)
- if fs_is_mounted(self.path):
- panic("fs is still mounted:", self.path)
+ if config.record or config.lctl_dump:
+ lctl.del_mount_option(local_node_name)
+ else:
+ if fs_is_mounted(self.path):
+ if config.force:
+ (rc, out) = run("umount", "-f", self.path)
+ else:
+ (rc, out) = run("umount", self.path)
+ if rc:
+ raise CommandError('umount', out, rc)
+
+ if fs_is_mounted(self.path):
+ panic("fs is still mounted:", self.path)
+ self.vmdc.cleanup()
self.vosc.cleanup()
- if self.vosc.need_mdc():
- self.mdc.cleanup()
if self.mgmtcli:
self.mgmtcli.cleanup()
if self.mgmtcli:
self.mgmtcli.cleanup_module()
+ def correct_level(self, level, op=None):
+ return level
# ============================================================
# misc query functions
node = self.lookup(node_uuid)
if not node:
panic("unable to find node for osd_uuid:", osd_uuid,
- " node_ref:", node_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.
+# the order of iniitailization is based on level.
def getServiceLevel(self):
type = self.get_class()
ret=0;
ret = 30
elif type in ('mdsdev',):
ret = 40
+ elif type in ('lmv',):
+ ret = 45
+ elif type in ('cmobd',):
+ ret = 50
elif type in ('mountpoint', 'echoclient'):
ret = 70
else:
panic("Unknown type: ", type)
if ret < config.minlevel or ret > config.maxlevel:
- ret = 0
+ ret = 0
return ret
#
# [(level, db_object),]
def getServices(self):
list = []
- for ref_class, ref_uuid in self.get_all_refs():
+ for ref_class, ref_uuid in self.get_all_refs():
servdb = self.lookup(ref_uuid)
if servdb:
level = getServiceLevel(servdb)
############################################################
-# MDC UUID hack -
+# MDC UUID hack -
# FIXME: clean this mess up!
#
# OSC is no longer in the xml, so we have to fake it.
osc = OSC(ost_db, uuid, fs_name)
return osc
-def get_mdc(db, uuid, fs_name, mds_uuid):
+def get_mdc(db, fs_name, mds_uuid):
mds_db = db.lookup(mds_uuid);
if not mds_db:
- panic("no mds:", mds_uuid)
- mdc = MDC(mds_db, uuid, fs_name)
+ error("no mds:", mds_uuid)
+ mdc = MDC(mds_db, mds_uuid, fs_name)
return mdc
############################################################
# routing ("rooting")
-
-# list of (nettype, cluster_id)
+# list of (nettype, cluster_id, nid)
local_clusters = []
def find_local_clusters(node_db):
net = node_db.lookup(netuuid)
srv = Network(net)
debug("add_local", netuuid)
- local_clusters.append((srv.net_type, srv.cluster_id))
+ local_clusters.append((srv.net_type, srv.cluster_id, srv.nid))
if srv.port > 0:
if acceptors.has_key(srv.port):
panic("duplicate port:", srv.port)
acceptors[srv.port] = AcceptorHandler(srv.port, srv.net_type,
srv.send_mem, srv.recv_mem,
- srv.irq_affinity,
- srv.nid_exchange)
+ srv.irq_affinity)
# This node is a gateway.
is_router = 0
for router in list:
if router.get_val_int('router', 0):
needs_router = 1
- for (local_type, local_cluster_id) in local_clusters:
+ for (local_type, local_cluster_id, local_nid) in local_clusters:
gw = None
for netuuid in router.get_networks():
db = router.lookup(netuuid)
def choose_local_server(srv_list):
for srv in srv_list:
- if local_net_type(srv.net_type, srv.cluster_id):
+ if local_cluster(srv.net_type, srv.cluster_id):
return srv
-def local_net_type(net_type, cluster_id):
+def local_cluster(net_type, cluster_id):
for cluster in local_clusters:
if net_type == cluster[0] and cluster_id == cluster[1]:
return 1
return 0
+def local_interface(net_type, cluster_id, nid):
+ for cluster in local_clusters:
+ if (net_type == cluster[0] and cluster_id == cluster[1]
+ and nid == cluster[2]):
+ return 1
+ return 0
+
def find_route(srv_list):
+ result = []
frm_type = local_clusters[0][0]
for srv in srv_list:
- debug("find_route: srv:", srv.hostaddr, "type: ", srv.net_type)
+ debug("find_route: srv:", srv.nid, "type: ", srv.net_type)
to_type = srv.net_type
- to = srv.hostaddr # XXX should this be hostaddr, or nid?
+ to = srv.nid
cluster_id = srv.cluster_id
debug ('looking for route to', to_type, to)
for r in local_routes:
debug("find_route: ", r)
if (r[3] <= to and to <= r[4]) and cluster_id == r[2]:
- return srv, r
- return None,None
-
+ result.append((srv, r))
+ return result
+
def get_active_target(db):
target_uuid = db.getUUID()
target_name = db.getName()
tgt_dev_uuid = db.get_first_ref('active')
return tgt_dev_uuid
+def get_server_by_nid_uuid(db, nid_uuid):
+ for n in db.lookup_class("network"):
+ net = Network(n)
+ if net.nid_uuid == nid_uuid:
+ return net
+
############################################################
# lconf level logic
elif type == 'osd':
n = OSD(db)
elif type == 'cobd':
- n = COBD(db)
+ n = COBD(db, "YOU_SHOULD_NEVER_SEE_THIS_UUID")
+ elif type == 'cmobd':
+ n = CMOBD(db)
elif type == 'mdsdev':
n = MDSDEV(db)
elif type == 'mountpoint':
n = ECHO_CLIENT(db)
elif type == 'mgmt':
n = Management(db)
+ elif type == 'lmv':
+ n = LMV(db)
else:
panic ("unknown service type:", type)
return n
#
# Prepare the system to run lustre using a particular profile
-# in a the configuration.
+# in a the configuration.
# * load & the modules
# * setup networking for the current node
# * make sure partitions are in place and prepared
panic("profile:", profile, "not found.")
services = getServices(prof_db)
operation(services)
-
+
+def magic_get_osc(db, rec, lov):
+ if lov:
+ lov_uuid = lov.get_uuid()
+ lov_name = lov.osc.fs_name
+ else:
+ lov_uuid = rec.getAttribute('lov_uuidref')
+ # FIXME: better way to find the mountpoint?
+ filesystems = db.root_node.getElementsByTagName('filesystem')
+ fsuuid = None
+ for fs in filesystems:
+ ref = fs.getElementsByTagName('obd_ref')
+ if ref[0].getAttribute('uuidref') == lov_uuid:
+ fsuuid = fs.getAttribute('uuid')
+ break
+
+ if not fsuuid:
+ panic("malformed xml: lov uuid '" + lov_uuid + "' referenced in 'add' record is not used by any filesystems.")
+
+ mtpts = db.root_node.getElementsByTagName('mountpoint')
+ lov_name = None
+ for fs in mtpts:
+ ref = fs.getElementsByTagName('filesystem_ref')
+ if ref[0].getAttribute('uuidref') == fsuuid:
+ lov_name = fs.getAttribute('name')
+ break
+
+ if not lov_name:
+ panic("malformed xml: 'add' record references lov uuid '" + lov_uuid + "', which references filesystem uuid '" + fsuuid + "', which does not reference a mountpoint.")
+
+ print "lov_uuid: " + lov_uuid + "; lov_name: " + lov_name
+
+ ost_uuid = rec.getAttribute('ost_uuidref')
+ obd = db.lookup(ost_uuid)
+
+ if not obd:
+ panic("malformed xml: 'add' record references ost uuid '" + ost_uuid + "' which cannot be found.")
+
+ osc = get_osc(obd, lov_uuid, lov_name)
+ if not osc:
+ panic('osc not found:', obd_uuid)
+ return osc
+
+# write logs for update records. sadly, logs of all types -- and updates in
+# particular -- are something of an afterthought. lconf needs rewritten with
+# these as core concepts. so this is a pretty big hack.
+def process_update_record(db, update, lov):
+ for rec in update.childNodes:
+ if rec.nodeType != rec.ELEMENT_NODE:
+ continue
+
+ log("found "+rec.nodeName+" record in update version " +
+ str(update.getAttribute('version')))
+
+ lov_uuid = rec.getAttribute('lov_uuidref')
+ ost_uuid = rec.getAttribute('ost_uuidref')
+ index = rec.getAttribute('index')
+ gen = rec.getAttribute('generation')
+
+ if not lov_uuid or not ost_uuid or not index or not gen:
+ panic("malformed xml: 'update' record requires lov_uuid, ost_uuid, index, and generation.")
+
+ if not lov:
+ tmplov = db.lookup(lov_uuid)
+ if not tmplov:
+ panic("malformed xml: 'delete' record contains lov UUID '" + lov_uuid + "', which cannot be located.")
+ lov_name = tmplov.getName()
+ else:
+ lov_name = lov.osc.name
+
+ # ------------------------------------------------------------- add
+ if rec.nodeName == 'add':
+ if config.cleanup:
+ lctl.lov_del_obd(lov_name, lov_uuid, ost_uuid, index, gen)
+ continue
+
+ osc = magic_get_osc(db, rec, lov)
+
+ try:
+ # Only ignore connect failures with --force, which
+ # isn't implemented here yet.
+ osc.prepare(ignore_connect_failure=0)
+ except CommandError, e:
+ print "Error preparing OSC %s\n" % osc.uuid
+ raise e
+
+ lctl.lov_add_obd(lov_name, lov_uuid, ost_uuid, index, gen)
+
+ # ------------------------------------------------------ deactivate
+ elif rec.nodeName == 'deactivate':
+ if config.cleanup:
+ continue
+
+ osc = magic_get_osc(db, rec, lov)
+
+ try:
+ osc.deactivate()
+ except CommandError, e:
+ print "Error deactivating OSC %s\n" % osc.uuid
+ raise e
+
+ # ---------------------------------------------------------- delete
+ elif rec.nodeName == 'delete':
+ if config.cleanup:
+ continue
+
+ osc = magic_get_osc(db, rec, lov)
+
+ try:
+ config.cleanup = 1
+ osc.cleanup()
+ config.cleanup = 0
+ except CommandError, e:
+ print "Error cleaning up OSC %s\n" % osc.uuid
+ raise e
+
+ lctl.lov_del_obd(lov_name, lov_uuid, ost_uuid, index, gen)
+
+def process_updates(db, log_device, log_name, lov = None):
+ updates = db.root_node.getElementsByTagName('update')
+ for u in updates:
+ if not u.childNodes:
+ log("ignoring empty update record (version " +
+ str(u.getAttribute('version')) + ")")
+ continue
+
+ version = u.getAttribute('version')
+ real_name = "%s-%s" % (log_name, version)
+ lctl.clear_log(log_device, real_name)
+ lctl.record(log_device, real_name)
+
+ process_update_record(db, u, lov)
+
+ lctl.end_record()
+
+def doWriteconf(services):
+ #if config.nosetup:
+ # return
+ for s in services:
+ if s[1].get_class() == 'mdsdev':
+ n = newService(s[1])
+ n.write_conf()
+
def doSetup(services):
if config.nosetup:
return
+ slist = []
for s in services:
n = newService(s[1])
- n.prepare()
-
+ n.level = s[0]
+ slist.append((n.level, n))
+ nlist = []
+ for n in slist:
+ nl = n[1].correct_level(n[0])
+ nlist.append((nl, n[1]))
+ nlist.sort()
+ for n in nlist:
+ n[1].prepare()
+
def doModules(services):
if config.nomod:
return
def doCleanup(services):
if config.nosetup:
return
- services.reverse()
+ slist = []
for s in services:
n = newService(s[1])
- if n.safe_to_clean():
- n.cleanup()
+ n.level = s[0]
+ slist.append((n.level, n))
+ nlist = []
+ for n in slist:
+ nl = n[1].correct_level(n[0])
+ nlist.append((nl, n[1]))
+ nlist.sort()
+ nlist.reverse()
+ for n in nlist:
+ if n[1].safe_to_clean():
+ n[1].cleanup()
def doUnloadModules(services):
if config.nomod:
n.cleanup_module()
#
-# Load profile for
+# Load profile for
def doHost(lustreDB, hosts):
- global is_router
+ global is_router, local_node_name
node_db = None
for h in hosts:
node_db = lustreDB.lookup_name(h, 'node')
if node_db:
break
if not node_db:
- print 'No host entry found.'
- return
+ panic('No host entry found.')
+ local_node_name = node_db.get_val('name', 0)
is_router = node_db.get_val_int('router', 0)
lustre_upcall = node_db.get_val('lustreUpcall', '')
portals_upcall = node_db.get_val('portalsUpcall', '')
timeout = node_db.get_val_int('timeout', 0)
+ ptldebug = node_db.get_val('ptldebug', '')
+ subsystem = node_db.get_val('subsystem', '')
find_local_clusters(node_db)
if not is_router:
# if not cleaning, load modules first.
prof_list = node_db.get_refs('profile')
- if config.recover:
+ if config.write_conf:
+ lustreDB.close()
+ for_each_profile(node_db, prof_list, doModules)
+ sys_make_devices()
+ for_each_profile(node_db, prof_list, doWriteconf)
+ for_each_profile(node_db, prof_list, doUnloadModules)
+
+ elif config.recover:
if not (config.tgt_uuid and config.client_uuid and config.conn_uuid):
raise Lustre.LconfError( "--recovery requires --tgt_uuid <UUID> " +
"--client_uuid <UUID> --conn_uuid <UUID>")
# 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 or config.record:
for_each_profile(node_db, prof_list, doCleanup)
return
sys_set_timeout(timeout)
- sys_set_ptldebug()
- sys_set_subsystem()
+ sys_set_ptldebug(ptldebug)
+ sys_set_subsystem(subsystem)
sys_set_lustre_upcall(lustre_upcall)
sys_set_portals_upcall(portals_upcall)
for_each_profile(node_db, prof_list, doCleanup)
for_each_profile(node_db, prof_list, doUnloadModules)
+ lustreDB.close()
else:
# ugly hack, only need to run lctl commands for --dump
- if config.lctl_dump:
+ if config.lctl_dump or config.record:
+ sys_set_timeout(timeout)
+ sys_set_lustre_upcall(lustre_upcall)
for_each_profile(node_db, prof_list, doSetup)
return
for_each_profile(node_db, prof_list, doModules)
sys_set_debug_path()
- sys_set_ptldebug()
- sys_set_subsystem()
+ sys_set_ptldebug(ptldebug)
+ sys_set_subsystem(subsystem)
script = config.gdb_script
run(lctl.lctl, ' modules >', script)
if config.gdb:
sys_set_portals_upcall(portals_upcall)
for_each_profile(node_db, prof_list, doSetup)
+ lustreDB.close()
-def doRecovery(db, lctl, tgt_uuid, client_uuid, conn_uuid):
- tgt = db.lookup(tgt_uuid)
+def doRecovery(lustreDB, lctl, tgt_uuid, client_uuid, nid_uuid):
+ tgt = lustreDB.lookup(tgt_uuid)
if not tgt:
raise Lustre.LconfError("doRecovery: "+ tgt_uuid +" not found.")
new_uuid = get_active_target(tgt)
if not new_uuid:
raise Lustre.LconfError("doRecovery: no active target found for: " +
tgt_uuid)
- net = choose_local_server(get_ost_net(db, new_uuid))
+ net = choose_local_server(get_ost_net(lustreDB, new_uuid))
if not net:
raise Lustre.LconfError("Unable to find a connection to:" + new_uuid)
- # XXX, better to do a full disconnect here
- log("Reconnecting", tgt_uuid, " to ", net.uuid);
- lctl.del_uuid(conn_uuid)
- lctl.connect(net)
- lctl.recover(client_uuid, net.uuid)
+
+ log("Reconnecting", tgt_uuid, " to ", net.nid_uuid);
+ try:
+ oldnet = get_server_by_nid_uuid(lustreDB, nid_uuid)
+ lustreDB.close()
+ if oldnet:
+ lctl.disconnect(oldnet)
+ except CommandError, e:
+ log("recover: disconnect", nid_uuid, "failed: ")
+ e.dump()
+
+ try:
+ lctl.connect(net)
+ except CommandError, e:
+ log("recover: connect failed")
+ e.dump()
+
+ lctl.recover(client_uuid, net.nid_uuid)
def setupModulePath(cmd, portals_dir = PORTALS_DIR):
base = os.path.dirname(cmd)
if development_mode():
if not config.lustre:
+ debug('using objdir module paths')
config.lustre = (os.path.join(base, ".."))
# normalize the portals dir, using command line arg if set
if config.portals:
debug('config.portals', config.portals)
elif config.lustre and config.portals:
# production mode
- # if --lustre and --portals, normalize portals
+ # if --lustre and --portals, normalize portals
# can ignore POTRALS_DIR here, since it is probly useless here
config.portals = os.path.join(config.lustre, config.portals)
debug('config.portals B', config.portals)
elif config.upcall:
upcall = config.upcall
if upcall:
- sysctl('lustre/upcall', upcall)
+ lctl.set_lustre_upcall(upcall)
def sys_set_portals_upcall(upcall):
# the command overrides the value in the node config
def sys_set_timeout(timeout):
# the command overrides the value in the node config
- if config.timeout > 0:
+ if config.timeout and config.timeout > 0:
timeout = config.timeout
if timeout != None and timeout > 0:
- sysctl('lustre/timeout', timeout)
-
-def sys_set_ptldebug():
- if config.ptldebug != None:
+ lctl.set_timeout(timeout)
+
+def sys_tweak_socknal ():
+ if config.single_socket:
+ sysctl("socknal/typed", 0)
+
+def sys_optimize_elan ():
+ procfiles = ["/proc/elan/config/eventint_punt_loops",
+ "/proc/qsnet/elan3/config/eventint_punt_loops",
+ "/proc/qsnet/elan4/config/elan4_mainint_punt_loops"]
+ for p in procfiles:
+ if os.access(p, os.R_OK):
+ run ("echo 1 > " + p)
+
+def sys_set_ptldebug(ptldebug):
+ if config.ptldebug:
+ ptldebug = config.ptldebug
+ if ptldebug:
try:
- val = eval(config.ptldebug, ptldebug_names)
- val = "0x%x" % (val,)
+ val = eval(ptldebug, ptldebug_names)
+ val = "0x%x" % (val)
sysctl('portals/debug', val)
except NameError, e:
panic(str(e))
-def sys_set_subsystem():
- if config.subsystem != None:
+def sys_set_subsystem(subsystem):
+ if config.subsystem:
+ subsystem = config.subsystem
+ if subsystem:
try:
- val = eval(config.subsystem, subsystem_names)
- val = "0x%x" % (val,)
+ val = eval(subsystem, subsystem_names)
+ val = "0x%x" % (val)
sysctl('portals/subsystem_debug', val)
except NameError, e:
panic(str(e))
return
fp = open(path)
str = fp.readline()
- fp.close
+ fp.close()
cur = int(str)
if max > cur:
fp = open(path, 'w')
fp.write('%d\n' %(max))
fp.close()
-
-
+
+
def sys_make_devices():
if not os.access('/dev/portals', os.R_OK):
run('mknod /dev/portals c 10 240')
if new_dir in syspath:
return
os.environ['PATH'] = os.environ['PATH'] + ':' + new_dir
-
+
def default_debug_path():
path = '/tmp/lustre-log'
if os.path.isdir('/r'):
# global hack for the --select handling
tgt_select = {}
-def init_select(arg):
- # arg = "service=nodeA,service2=nodeB"
+def init_select(args):
+ # args = [service=nodeA,service2=nodeB service3=nodeC]
global tgt_select
- if arg:
+ for arg in args:
list = string.split(arg, ',')
for entry in list:
srv, node = string.split(entry, '=')
return None
+FLAG = Lustre.Options.FLAG
PARAM = Lustre.Options.PARAM
INTPARAM = Lustre.Options.INTPARAM
+PARAMLIST = Lustre.Options.PARAMLIST
lconf_options = [
('verbose,v', "Print system commands as they are run"),
('ldapurl',"LDAP server URL, eg. ldap://localhost", PARAM),
('config', "Cluster config name used for LDAP query", PARAM),
- ('select', "service=nodeA,service2=nodeB ", PARAM),
+ ('select', "service=nodeA,service2=nodeB ", PARAMLIST),
('node', "Load config for <nodename>", PARAM),
('cleanup,d', "Cleans up config. (Shutdown)"),
('force,f', "Forced unmounting and/or obd detach during cleanup",
- Lustre.Options.FLAG, 0),
- ('mds_ost_conn', "Open connections to OSTs on the MDS"),
+ FLAG, 0),
+ ('single_socket', "socknal option: only use one socket instead of bundle",
+ FLAG, 0),
('failover',"""Used to shut down without saving state.
This will allow this node to "give up" a service to a
another node for failover purposes. This will not
be a clean shutdown.""",
- Lustre.Options.FLAG, 0),
+ 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
('nomod', "Skip load/unload module step."),
('nosetup', "Skip device setup/cleanup step."),
('reformat', "Reformat all devices (without question)"),
+ ('mkfsoptions', "Additional options for the mk*fs command line", PARAM),
+ ('mountfsoptions', "Additional options for mount fs command line", PARAM),
+ ('clientoptions', "Additional options for Lustre", PARAM),
('dump', "Dump the kernel debug log to file before portals is unloaded",
PARAM),
+ ('write_conf', "Save all the client config information on mds."),
+ ('record', "Write config information on mds."),
+ ('record_log', "Name of config record log.", PARAM),
+ ('record_device', "MDS device name that will record the config commands",
+ PARAM),
('minlevel', "Minimum level of services to configure/cleanup",
INTPARAM, 0),
- ('maxlevel', """Maximum level of services to configure/cleanup
+ ('maxlevel', """Maximum level of services to configure/cleanup
Levels are aproximatly like:
- 10 - network
+ 10 - netwrk
20 - device, ldlm
30 - osd, mdd
40 - mds, ost
cause lconf to load modules from a source tree.""", PARAM),
('portals', """Portals source directory. If this is a relative path,
then it is assumed to be relative to lustre. """, PARAM),
- ('timeout', "Set recovery timeout", PARAM),
+ ('timeout', "Set recovery timeout", INTPARAM),
('upcall', "Set both portals and lustre upcall script", PARAM),
('lustre_upcall', "Set lustre upcall script", PARAM),
('portals_upcall', "Set portals upcall script", PARAM),
('tgt_uuid', "The failed target (required for recovery)", PARAM),
('client_uuid', "The failed client (required for recovery)", PARAM),
('conn_uuid', "The failed connection (required for recovery)", PARAM),
- ]
+
+ ('inactive', """The name of an inactive service, to be ignored during
+ mounting (currently OST-only). Can be repeated.""",
+ PARAMLIST),
+ ]
def main():
- global lctl, config, toplevel
+ global lctl, config, toplustreDB, CONFIG_FILE
# 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:])
random.seed(seed)
sanitise_path()
-
+
init_select(config.select)
if len(args) > 0:
- if not os.access(args[0], os.R_OK):
+ # allow config to be fetched via HTTP, but only with python2
+ if sys.version[0] != '1' and args[0].startswith('http://'):
+ import urllib2
+ try:
+ config_file = urllib2.urlopen(args[0])
+ except (urllib2.URLError, socket.error), err:
+ if hasattr(err, 'args'):
+ err = err.args[1]
+ print "Could not access '%s': %s" %(args[0], err)
+ sys.exit(1)
+ elif not os.access(args[0], os.R_OK):
print 'File not found or readable:', args[0]
sys.exit(1)
+ else:
+ # regular file
+ config_file = open(args[0], 'r')
try:
- dom = xml.dom.minidom.parse(args[0])
+ dom = xml.dom.minidom.parse(config_file)
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 = Lustre.LustreDB_XML(dom.documentElement, dom.documentElement)
+ config_file.close()
+ CONFIG_FILE = args[0]
+ lustreDB = Lustre.LustreDB_XML(dom.documentElement, dom.documentElement)
if not config.config:
config.config = os.path.basename(args[0])# use full path?
if config.config[-4:] == '.xml':
if not config.config:
panic("--ldapurl requires --config name")
dn = "config=%s,fs=lustre" % (config.config)
- db = Lustre.LustreDB_LDAP('', {}, base=dn, url = config.ldapurl)
+ lustreDB = Lustre.LustreDB_LDAP('', {}, base=dn, url = config.ldapurl)
+ elif config.ptldebug or config.subsystem:
+ sys_set_ptldebug(None)
+ sys_set_subsystem(None)
+ sys.exit(0)
else:
print 'Missing config file or ldap URL.'
print 'see lconf --help for command summary'
sys.exit(1)
- toplevel = db
+ toplustreDB = lustreDB
- ver = db.get_version()
+ ver = lustreDB.get_version()
if not ver:
panic("No version found in config data, please recreate.")
if ver != Lustre.CONFIG_VERSION:
if config.lctl_dump:
lctl.use_save_file(config.lctl_dump)
- doHost(db, node_list)
+ if config.record:
+ if not (config.record_device and config.record_log):
+ panic("When recording, both --record_log and --record_device must be specified.")
+ lctl.clear_log(config.record_device, config.record_log)
+ lctl.record(config.record_device, config.record_log)
+
+ doHost(lustreDB, node_list)
+
+ if not config.record:
+ return
+
+ lctl.end_record()
+
+ process_updates(lustreDB, config.record_device, config.record_log)
if __name__ == "__main__":
try: