#
# Based in part on the XML obdctl modifications done by Brian Behlendorf
-import sys, getopt, types
+import sys, getopt, types, errno
import string, os, stat, popen2, socket, time, random, fcntl, select
import re, exceptions, signal, traceback
import xml.dom.minidom
if n.safe_to_clean_modules():
n.cleanup_module()
+def doMakeServiceScript(services):
+ if config.nosetup:
+ return
+ try:
+ os.makedirs(config.service_scripts)
+ except OSError, e:
+ if e[0] != errno.EEXIST:
+ panic("Couldn't create scripts dir " + config.service_scripts + ": " + e[1])
+
+ for s in services:
+ if s[1].get_class() != 'osd' and s[1].get_class() != 'mdsdev':
+ continue
+
+ target_uuid = s[1].get_first_ref('target')
+ target = toplustreDB.lookup(target_uuid)
+ target_symlink = config.service_scripts + "/" + target.getName()
+ if config.force:
+ try:
+ try:
+ os.unlink(target_symlink)
+ if config.verbose:
+ print "Removed " + target_symlink
+ except OSError, e:
+ if e[0] != errno.EISDIR:
+ raise e
+ os.rmdir(target_symlink)
+ if config.verbose:
+ print "Removed " + target_symlink
+ except OSError, e:
+ if e[0] != errno.ENOENT:
+ panic("Error removing " + target_symlink + ": " + e[1])
+
+ try:
+ os.symlink("/etc/init.d/lustre", target_symlink)
+ if config.verbose:
+ print "Created service link " + target_symlink + " to /etc/init.d/lustre"
+
+ except OSError, e:
+ if e[0] == errno.EEXIST:
+ extra_error = " (use --force option to remove existing files)"
+ else:
+ extra_error = ""
+ panic("Error creating " + target_symlink + ": " + e[1] + extra_error)
+
#
# Load profile for
def doHost(lustreDB, hosts):
# if not cleaning, load modules first.
prof_list = node_db.get_refs('profile')
- if config.write_conf:
+ if config.make_service_scripts:
+ for_each_profile(node_db, prof_list, doMakeServiceScript)
+ return
+
+ elif config.write_conf:
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, doSetup)
lustreDB.close()
+def add_clumanager_node(node_db, nodes, services):
+ new_services = []
+ node_name = node_db.getUUID()
+ nodes[node_name] = []
+
+ for prof_uuid in node_db.get_refs('profile'):
+ prof_db = toplustreDB.lookup(prof_uuid)
+ for ref_class, ref_uuid in prof_db.get_all_refs():
+ if ref_class not in ('osd', 'mdsdev'):
+ continue
+ devdb = toplustreDB.lookup(ref_uuid)
+ tgt_uuid = devdb.get_first_ref('target')
+
+ nodes[node_name].append(ref_uuid)
+
+ if not services.has_key(tgt_uuid):
+ if config.verbose:
+ print "New service: " + tgt_uuid + " (originally found on " + node_name + ")"
+ new_services.append(tgt_uuid)
+ services[tgt_uuid] = []
+ services[tgt_uuid].append(ref_uuid)
+
+ return new_services
+
+def add_clumanager_services(new_services, nodes, dev_list):
+ new_nodes = []
+ for devdb in dev_list:
+ tgt_uuid = devdb.get_first_ref('target')
+ if tgt_uuid in new_services:
+ node_uuid = devdb.get_first_ref('node')
+
+ if not (nodes.has_key(node_uuid) or node_uuid in new_nodes):
+ if config.verbose:
+ print "New node: " + node_uuid + " for service " + tgt_uuid
+ new_nodes.append(node_uuid)
+
+ return new_nodes
+
+def doClumanager(lustreDB, hosts):
+ nodes = {}
+ services = {}
+
+ dev_list = []
+
+ for dev_uuid in toplustreDB.get_refs('osd') + toplustreDB.get_refs('mdsdev'):
+ dev_list.append(lustreDB.lookup(dev_uuid))
+
+ node_db = None
+ for h in hosts:
+ node_db = lustreDB.lookup_name(h, 'node')
+ if node_db:
+ our_host = h
+ new_services = add_clumanager_node(node_db, nodes, services)
+ break
+
+ if not node_db:
+ panic('No host entry found.')
+
+ while 1:
+ if len(new_services) == 0:
+ break
+
+ new_nodes = add_clumanager_services(new_services, nodes, dev_list)
+ if len(new_nodes) == 0:
+ break
+
+ if len(new_nodes) + len(nodes.keys()) > 8:
+ panic("CluManager only supports 8 nodes per failover \"cluster.\"")
+
+ new_services = []
+ for node_uuid in new_nodes:
+ node_db = lustreDB.lookup(node_uuid)
+ if not node_db:
+ panic("No node entry for " + node_uuid + " was found.")
+
+ new_services += add_clumanager_node(node_db, nodes, services)
+
+ print """<?xml version="1.0"?>
+<cluconfig version="3.0">
+ <clumembd broadcast="no" interval="750000" loglevel="5" multicast="yes" multicast_ipaddress="225.0.0.11" thread="yes" tko_count="20"/>
+ <cluquorumd loglevel="5" pinginterval="2"/>
+ <clurmtabd loglevel="5" pollinterval="4"/>
+ <clusvcmgrd loglevel="5"/>
+ <clulockd loglevel="5"/>
+ <cluster config_viewnumber="1" name="%s"/>
+ <sharedstate driver="libsharedraw.so" rawprimary="%s" rawshadow="%s" type="raw"/>
+ <members> """ % (our_host, config.rawprimary, config.rawsecondary)
+
+ nodekeys = nodes.keys()
+ nodekeys.sort()
+
+ servicekeys = services.keys()
+ servicekeys.sort()
+
+ i = 0
+ for node in nodekeys:
+ nodedb = lustreDB.lookup(node)
+ print " <member id=\"%d\" name=\"%s\" watchdog=\"yes\"/>" % (i, nodedb.getName())
+ i += 1
+
+ print " </members>\n <failoverdomains>"
+
+ i = 0
+ for service in servicekeys:
+ svcdb = lustreDB.lookup(service)
+ print " <failoverdomain id=\"%d\" name=\"%s\" ordered=\"yes\" restricted=\"yes\">" % (i, svcdb.getName())
+ i += 1
+
+ j = 0
+ active_uuid = get_active_target(svcdb)
+ for svc_uuid in [active_uuid] + services[service]:
+ if svc_uuid == active_uuid and j > 0:
+ continue
+ svcdb = lustreDB.lookup(svc_uuid)
+
+ svc_node_uuid = svcdb.get_first_ref('node')
+ svc_nodedb = lustreDB.lookup(svc_node_uuid)
+
+ print " <failoverdomainnode id=\"%d\" name=\"%s\"/>" % (j, svc_nodedb.getName())
+ j += 1
+
+ print " </failoverdomain>"
+
+ print " </failoverdomains>\n <services>"
+
+ i = 0
+ for service in servicekeys:
+ svcdb = lustreDB.lookup(service)
+ active_uuid = get_active_target(svcdb)
+ activedb = lustreDB.lookup(active_uuid)
+
+ svc_node_uuid = activedb.get_first_ref('node')
+ svc_nodedb = lustreDB.lookup(svc_node_uuid)
+
+ print " <service checkinterval=\"30\" failoverdomain=\"%s\" id=\"%d\" name=\"%s\" userscript=\"%s/%s\">" \
+ % ( svcdb.getName(), i, svcdb.getName(), config.service_scripts, svcdb.getName())
+ print " <service_ipaddresses/>\n </service>"
+ i += 1
+
+ print " </services>\n</cluconfig>"
+
def doRecovery(lustreDB, lctl, tgt_uuid, client_uuid, nid_uuid):
tgt = lustreDB.lookup(tgt_uuid)
if not tgt:
('gdb_script', "Fullname of gdb debug script", PARAM, default_gdb_script()),
('debug_path', "Path to save debug dumps", PARAM, default_debug_path()),
('allow_unprivileged_port', "Allow connections from unprivileged ports"),
+ ('clumanager', "Generate CluManager config file for this node's cluster"),
+ ('rawprimary', "For clumanager, device of the primary quorum", PARAM, "/dev/raw/raw1"),
+ ('rawsecondary', "For clumanager, device of the secondary quorum", PARAM, "/dev/raw/raw2"),
+ ('service_scripts', "For clumanager, directory containing per-service scripts", PARAM, "/etc/lustre/services"),
+ ('make_service_scripts', "Create per-service symlinks for use with clumanager"),
# Client recovery options
('recover', "Recover a device"),
('group', "The group of devices to configure or cleanup", PARAM),
lctl.clear_log(config.record_device, config.record_log)
lctl.record(config.record_device, config.record_log)
- doHost(lustreDB, node_list)
+ if config.clumanager:
+ doClumanager(lustreDB, node_list)
+ else:
+ doHost(lustreDB, node_list)
if config.record:
lctl.end_record()
git://git.whamcloud.com / fs / lustre-release.git / commitdiff