_MIGRATION_STATUS_RE = re.compile('Migration\s+status:\s+(\w+)',
re.M | re.I)
+ _MIGRATION_INFO_MAX_BAD_ANSWERS = 5
+ _MIGRATION_INFO_RETRY_DELAY = 2
_KVM_NETWORK_SCRIPT = constants.SYSCONFDIR + "/ganeti/kvm-vif-bridge"
dirs = [(dname, constants.RUN_DIRS_MODE) for dname in self._DIRS]
utils.EnsureDirs(dirs)
- def _InstancePidFile(self, instance_name):
+ @classmethod
+ def _InstancePidFile(cls, instance_name):
"""Returns the instance pidfile.
"""
- pidfile = "%s/%s" % (self._PIDS_DIR, instance_name)
- return pidfile
+ return utils.PathJoin(cls._PIDS_DIR, instance_name)
def _InstancePidAlive(self, instance_name):
"""Returns the instance pid and pidfile
return (pidfile, pid, alive)
+ def _CheckDown(self, instance_name):
+ """Raises an error unless the given instance is down.
+
+ """
+ alive = self._InstancePidAlive(instance_name)[2]
+ if alive:
+ raise errors.HypervisorError("Failed to start instance %s: %s" %
+ (instance_name, "already running"))
+
@classmethod
def _InstanceMonitor(cls, instance_name):
"""Returns the instance monitor socket name
"""
- return '%s/%s.monitor' % (cls._CTRL_DIR, instance_name)
+ return utils.PathJoin(cls._CTRL_DIR, "%s.monitor" % instance_name)
@classmethod
def _InstanceSerial(cls, instance_name):
"""Returns the instance serial socket name
"""
- return '%s/%s.serial' % (cls._CTRL_DIR, instance_name)
+ return utils.PathJoin(cls._CTRL_DIR, "%s.serial" % instance_name)
@staticmethod
def _SocatUnixConsoleParams():
"""Returns the instance KVM runtime filename
"""
- return '%s/%s.runtime' % (cls._CONF_DIR, instance_name)
+ return utils.PathJoin(cls._CONF_DIR, "%s.runtime" % instance_name)
@classmethod
def _RemoveInstanceRuntimeFiles(cls, pidfile, instance_name):
script = StringIO()
script.write("#!/bin/sh\n")
script.write("# this is autogenerated by Ganeti, please do not edit\n#\n")
+ script.write("PATH=$PATH:/sbin:/usr/sbin\n")
script.write("export INSTANCE=%s\n" % instance.name)
script.write("export MAC=%s\n" % nic.mac)
if nic.ip:
script.write(" # Execute the user-specific vif file\n")
script.write(" %s\n" % self._KVM_NETWORK_SCRIPT)
script.write("else\n")
- script.write(" /sbin/ifconfig $INTERFACE 0.0.0.0 up\n")
+ script.write(" ifconfig $INTERFACE 0.0.0.0 up\n")
if nic.nicparams[constants.NIC_MODE] == constants.NIC_MODE_BRIDGED:
script.write(" # Connect the interface to the bridge\n")
- script.write(" /usr/sbin/brctl addif $BRIDGE $INTERFACE\n")
+ script.write(" brctl addif $BRIDGE $INTERFACE\n")
elif nic.nicparams[constants.NIC_MODE] == constants.NIC_MODE_ROUTED:
if not nic.ip:
raise errors.HypervisorError("nic/%d is routed, but has no ip." % seq)
script.write(" # Route traffic targeted at the IP to the interface\n")
if nic.nicparams[constants.NIC_LINK]:
- script.write(" while /sbin/ip rule del dev $INTERFACE; do :; done\n")
- script.write(" /sbin/ip rule add dev $INTERFACE table $LINK\n")
- script.write(" /sbin/ip route replace $IP table $LINK proto static"
+ script.write(" while ip rule del dev $INTERFACE; do :; done\n")
+ script.write(" ip rule add dev $INTERFACE table $LINK\n")
+ script.write(" ip route replace $IP table $LINK proto static"
" dev $INTERFACE\n")
else:
- script.write(" /sbin/ip route replace $IP proto static"
+ script.write(" ip route replace $IP proto static"
" dev $INTERFACE\n")
interface_v4_conf = "/proc/sys/net/ipv4/conf/$INTERFACE"
interface_v6_conf = "/proc/sys/net/ipv6/conf/$INTERFACE"
"""
result = []
for name in os.listdir(self._PIDS_DIR):
- filename = "%s/%s" % (self._PIDS_DIR, name)
+ filename = utils.PathJoin(self._PIDS_DIR, name)
if utils.IsProcessAlive(utils.ReadPidFile(filename)):
result.append(name)
return result
if not alive:
return None
- cmdline_file = "/proc/%s/cmdline" % pid
+ cmdline_file = utils.PathJoin("/proc", str(pid), "cmdline")
try:
cmdline = utils.ReadFile(cmdline_file)
except EnvironmentError, err:
"""
data = []
for name in os.listdir(self._PIDS_DIR):
- filename = "%s/%s" % (self._PIDS_DIR, name)
+ filename = utils.PathJoin(self._PIDS_DIR, name)
if utils.IsProcessAlive(utils.ReadPidFile(filename)):
try:
info = self.GetInstanceInfo(name)
@param incoming: (target_host_ip, port)
"""
- pidfile, _, alive = self._InstancePidAlive(instance.name)
hvp = instance.hvparams
- if alive:
- raise errors.HypervisorError("Failed to start instance %s: %s" %
- (instance.name, "already running"))
+ name = instance.name
+ self._CheckDown(name)
temp_files = []
result = utils.RunCmd(kvm_cmd)
if result.failed:
raise errors.HypervisorError("Failed to start instance %s: %s (%s)" %
- (instance.name, result.fail_reason,
- result.output))
+ (name, result.fail_reason, result.output))
- if not utils.IsProcessAlive(utils.ReadPidFile(pidfile)):
- raise errors.HypervisorError("Failed to start instance %s" %
- (instance.name))
+ if not self._InstancePidAlive(name)[2]:
+ raise errors.HypervisorError("Failed to start instance %s" % name)
if vnc_pwd:
change_cmd = 'change vnc password %s' % vnc_pwd
"""Start an instance.
"""
- pidfile, pid, alive = self._InstancePidAlive(instance.name)
- if alive:
- raise errors.HypervisorError("Failed to start instance %s: %s" %
- (instance.name, "already running"))
-
+ self._CheckDown(instance.name)
kvm_runtime = self._GenerateKVMRuntime(instance, block_devices)
self._SaveKVMRuntime(instance, kvm_runtime)
self._ExecuteKVMRuntime(instance, kvm_runtime)
info_command = 'info migrate'
done = False
+ broken_answers = 0
while not done:
result = self._CallMonitorCommand(instance_name, info_command)
match = self._MIGRATION_STATUS_RE.search(result.stdout)
if not match:
- raise errors.HypervisorError("Unknown 'info migrate' result: %s" %
- result.stdout)
+ broken_answers += 1
+ if not result.stdout:
+ logging.info("KVM: empty 'info migrate' result")
+ else:
+ logging.warning("KVM: unknown 'info migrate' result: %s",
+ result.stdout)
+ time.sleep(self._MIGRATION_INFO_RETRY_DELAY)
else:
status = match.group(1)
if status == 'completed':
done = True
elif status == 'active':
- time.sleep(2)
+ # reset the broken answers count
+ broken_answers = 0
+ time.sleep(self._MIGRATION_INFO_RETRY_DELAY)
elif status == 'failed' or status == 'cancelled':
if not live:
self._CallMonitorCommand(instance_name, 'cont')
raise errors.HypervisorError("Migration %s at the kvm level" %
status)
else:
- logging.info("KVM: unknown migration status '%s'", status)
- time.sleep(2)
+ logging.warning("KVM: unknown migration status '%s'", status)
+ broken_answers += 1
+ time.sleep(self._MIGRATION_INFO_RETRY_DELAY)
+ if broken_answers >= self._MIGRATION_INFO_MAX_BAD_ANSWERS:
+ raise errors.HypervisorError("Too many 'info migrate' broken answers")
utils.KillProcess(pid)
self._RemoveInstanceRuntimeFiles(pidfile, instance_name)
projects / ganeti-local / blobdiff