X-Git-Url: https://code.grnet.gr/git/ganeti-local/blobdiff_plain/30948aa6bf7d6a5fc214d54d9d050bd63b47496d..2ee88aeb76a2430ec0c7f86629bf66cfd0b6f564:/lib/hypervisor/hv_xen.py diff --git a/lib/hypervisor/hv_xen.py b/lib/hypervisor/hv_xen.py index b34b221..6693c6b 100644 --- a/lib/hypervisor/hv_xen.py +++ b/lib/hypervisor/hv_xen.py @@ -42,9 +42,16 @@ class XenHypervisor(hv_base.BaseHypervisor): all the functionality that is identical for both. """ + REBOOT_RETRY_COUNT = 60 + REBOOT_RETRY_INTERVAL = 10 + + ANCILLARY_FILES = [ + '/etc/xen/xend-config.sxp', + '/etc/xen/scripts/vif-bridge', + ] @classmethod - def _WriteConfigFile(cls, instance, block_devices, extra_args): + def _WriteConfigFile(cls, instance, block_devices): """Write the Xen config file for the instance. """ @@ -98,7 +105,7 @@ class XenHypervisor(hv_base.BaseHypervisor): if result.failed: raise errors.HypervisorError("xm list failed, retries" " exceeded (%s): %s" % - (result.fail_reason, result.stderr)) + (result.fail_reason, result.output)) # skip over the heading lines = result.stdout.splitlines()[1:] @@ -159,11 +166,11 @@ class XenHypervisor(hv_base.BaseHypervisor): xm_list = self._GetXMList(False) return xm_list - def StartInstance(self, instance, block_devices, extra_args): + def StartInstance(self, instance, block_devices): """Start an instance. """ - self._WriteConfigFile(instance, block_devices, extra_args) + self._WriteConfigFile(instance, block_devices) result = utils.RunCmd(["xm", "create", instance.name]) if result.failed: @@ -183,26 +190,48 @@ class XenHypervisor(hv_base.BaseHypervisor): result = utils.RunCmd(command) if result.failed: - raise errors.HypervisorError("Failed to stop instance %s: %s" % - (instance.name, result.fail_reason)) + raise errors.HypervisorError("Failed to stop instance %s: %s, %s" % + (instance.name, result.fail_reason, + result.output)) def RebootInstance(self, instance): """Reboot an instance. """ + ini_info = self.GetInstanceInfo(instance.name) result = utils.RunCmd(["xm", "reboot", instance.name]) if result.failed: - raise errors.HypervisorError("Failed to reboot instance %s: %s" % - (instance.name, result.fail_reason)) + raise errors.HypervisorError("Failed to reboot instance %s: %s, %s" % + (instance.name, result.fail_reason, + result.output)) + done = False + retries = self.REBOOT_RETRY_COUNT + while retries > 0: + new_info = self.GetInstanceInfo(instance.name) + # check if the domain ID has changed or the run time has + # decreased + if new_info[1] != ini_info[1] or new_info[5] < ini_info[5]: + done = True + break + time.sleep(self.REBOOT_RETRY_INTERVAL) + retries -= 1 + + if not done: + raise errors.HypervisorError("Failed to reboot instance %s: instance" + " did not reboot in the expected interval" % + (instance.name, )) def GetNodeInfo(self): """Return information about the node. - @return: a dict with the following keys (values in MiB): + @return: a dict with the following keys (memory values in MiB): - memory_total: the total memory size on the node - memory_free: the available memory on the node for instances - memory_dom0: the memory used by the node itself, if available + - nr_cpus: total number of CPUs + - nr_nodes: in a NUMA system, the number of domains + - nr_sockets: the number of physical CPU sockets in the node """ # note: in xen 3, memory has changed to total_memory @@ -214,6 +243,7 @@ class XenHypervisor(hv_base.BaseHypervisor): xmoutput = result.stdout.splitlines() result = {} + cores_per_socket = threads_per_core = nr_cpus = None for line in xmoutput: splitfields = line.split(":", 1) @@ -225,7 +255,18 @@ class XenHypervisor(hv_base.BaseHypervisor): elif key == 'free_memory': result['memory_free'] = int(val) elif key == 'nr_cpus': - result['cpu_total'] = int(val) + nr_cpus = result['cpu_total'] = int(val) + elif key == 'nr_nodes': + result['cpu_nodes'] = int(val) + elif key == 'cores_per_socket': + cores_per_socket = int(val) + elif key == 'threads_per_core': + threads_per_core = int(val) + + if (cores_per_socket is not None and + threads_per_core is not None and nr_cpus is not None): + result['cpu_sockets'] = nr_cpus / (cores_per_socket * threads_per_core) + dom0_info = self.GetInstanceInfo("Domain-0") if dom0_info is not None: result['memory_dom0'] = dom0_info[2] @@ -248,7 +289,7 @@ class XenHypervisor(hv_base.BaseHypervisor): """ result = utils.RunCmd(["xm", "info"]) if result.failed: - return "'xm info' failed: %s" % result.fail_reason + return "'xm info' failed: %s, %s" % (result.fail_reason, result.output) @staticmethod def _GetConfigFileDiskData(disk_template, block_devices): @@ -361,76 +402,49 @@ class XenHypervisor(hv_base.BaseHypervisor): except EnvironmentError: logging.exception("Failure while removing instance config file") - -class XenPvmHypervisor(XenHypervisor): - """Xen PVM hypervisor interface""" - - PARAMETERS = [ - constants.HV_KERNEL_PATH, - constants.HV_INITRD_PATH, - constants.HV_ROOT_PATH, - ] - @classmethod - def CheckParameterSyntax(cls, hvparams): - """Check the given parameters for validity. - - For the PVM hypervisor, this only check the existence of the - kernel. + def PowercycleNode(cls): + """Xen-specific powercycle. - @type hvparams: dict - @param hvparams: dictionary with parameter names/value - @raise errors.HypervisorError: when a parameter is not valid + This first does a Linux reboot (which triggers automatically a Xen + reboot), and if that fails it tries to do a Xen reboot. The reason + we don't try a Xen reboot first is that the xen reboot launches an + external command which connects to the Xen hypervisor, and that + won't work in case the root filesystem is broken and/or the xend + daemon is not working. """ - super(XenPvmHypervisor, cls).CheckParameterSyntax(hvparams) - - if not hvparams[constants.HV_KERNEL_PATH]: - raise errors.HypervisorError("Need a kernel for the instance") - - if not os.path.isabs(hvparams[constants.HV_KERNEL_PATH]): - raise errors.HypervisorError("The kernel path must be an absolute path") - - if not hvparams[constants.HV_ROOT_PATH]: - raise errors.HypervisorError("Need a root partition for the instance") - - if hvparams[constants.HV_INITRD_PATH]: - if not os.path.isabs(hvparams[constants.HV_INITRD_PATH]): - raise errors.HypervisorError("The initrd path must be an absolute path" - ", if defined") - - def ValidateParameters(self, hvparams): - """Check the given parameters for validity. + try: + cls.LinuxPowercycle() + finally: + utils.RunCmd(["xm", "debug", "R"]) - For the PVM hypervisor, this only check the existence of the - kernel. - """ - super(XenPvmHypervisor, self).ValidateParameters(hvparams) +class XenPvmHypervisor(XenHypervisor): + """Xen PVM hypervisor interface""" - kernel_path = hvparams[constants.HV_KERNEL_PATH] - if not os.path.isfile(kernel_path): - raise errors.HypervisorError("Instance kernel '%s' not found or" - " not a file" % kernel_path) - initrd_path = hvparams[constants.HV_INITRD_PATH] - if initrd_path and not os.path.isfile(initrd_path): - raise errors.HypervisorError("Instance initrd '%s' not found or" - " not a file" % initrd_path) + PARAMETERS = { + constants.HV_KERNEL_PATH: hv_base.REQ_FILE_CHECK, + constants.HV_INITRD_PATH: hv_base.OPT_FILE_CHECK, + constants.HV_ROOT_PATH: hv_base.REQUIRED_CHECK, + constants.HV_KERNEL_ARGS: hv_base.NO_CHECK, + } @classmethod - def _WriteConfigFile(cls, instance, block_devices, extra_args): + def _WriteConfigFile(cls, instance, block_devices): """Write the Xen config file for the instance. """ + hvp = instance.hvparams config = StringIO() config.write("# this is autogenerated by Ganeti, please do not edit\n#\n") # kernel handling - kpath = instance.hvparams[constants.HV_KERNEL_PATH] + kpath = hvp[constants.HV_KERNEL_PATH] config.write("kernel = '%s'\n" % kpath) # initrd handling - initrd_path = instance.hvparams[constants.HV_INITRD_PATH] + initrd_path = hvp[constants.HV_INITRD_PATH] if initrd_path: config.write("ramdisk = '%s'\n" % initrd_path) @@ -452,18 +466,15 @@ class XenPvmHypervisor(XenHypervisor): cls._GetConfigFileDiskData(instance.disk_template, block_devices))) - rpath = instance.hvparams[constants.HV_ROOT_PATH] - config.write("root = '%s ro'\n" % rpath) + config.write("root = '%s'\n" % hvp[constants.HV_ROOT_PATH]) config.write("on_poweroff = 'destroy'\n") config.write("on_reboot = 'restart'\n") config.write("on_crash = 'restart'\n") - if extra_args: - config.write("extra = '%s'\n" % extra_args) + config.write("extra = '%s'\n" % hvp[constants.HV_KERNEL_ARGS]) # just in case it exists utils.RemoveFile("/etc/xen/auto/%s" % instance.name) try: - utils.WriteFile("/etc/xen/%s" % instance.name, - data=config.getvalue()) + utils.WriteFile("/etc/xen/%s" % instance.name, data=config.getvalue()) except EnvironmentError, err: raise errors.HypervisorError("Cannot write Xen instance confile" " file /etc/xen/%s: %s" % @@ -475,111 +486,68 @@ class XenPvmHypervisor(XenHypervisor): class XenHvmHypervisor(XenHypervisor): """Xen HVM hypervisor interface""" - PARAMETERS = [ - constants.HV_ACPI, - constants.HV_BOOT_ORDER, - constants.HV_CDROM_IMAGE_PATH, - constants.HV_DISK_TYPE, - constants.HV_NIC_TYPE, - constants.HV_PAE, - constants.HV_VNC_BIND_ADDRESS, - ] - - @classmethod - def CheckParameterSyntax(cls, hvparams): - """Check the given parameter syntax. - - """ - super(XenHvmHypervisor, cls).CheckParameterSyntax(hvparams) - # boot order verification - boot_order = hvparams[constants.HV_BOOT_ORDER] - if not boot_order or len(boot_order.strip("acdn")) != 0: - raise errors.HypervisorError("Invalid boot order '%s' specified," - " must be one or more of [acdn]" % - boot_order) - # device type checks - nic_type = hvparams[constants.HV_NIC_TYPE] - if nic_type not in constants.HT_HVM_VALID_NIC_TYPES: - raise errors.HypervisorError("Invalid NIC type %s specified for Xen HVM" - " hypervisor" % nic_type) - disk_type = hvparams[constants.HV_DISK_TYPE] - if disk_type not in constants.HT_HVM_VALID_DISK_TYPES: - raise errors.HypervisorError("Invalid disk type %s specified for Xen HVM" - " hypervisor" % disk_type) - # vnc_bind_address verification - vnc_bind_address = hvparams[constants.HV_VNC_BIND_ADDRESS] - if vnc_bind_address is not None: - if not utils.IsValidIP(vnc_bind_address): - raise errors.OpPrereqError("given VNC bind address '%s' doesn't look" - " like a valid IP address" % - vnc_bind_address) - - iso_path = hvparams[constants.HV_CDROM_IMAGE_PATH] - if iso_path and not os.path.isabs(iso_path): - raise errors.HypervisorError("The path to the HVM CDROM image must" - " be an absolute path or None, not %s" % - iso_path) - - def ValidateParameters(self, hvparams): - """Check the given parameters for validity. - - For the PVM hypervisor, this only check the existence of the - kernel. - - @type hvparams: dict - @param hvparams: dictionary with parameter names/value - @raise errors.HypervisorError: when a parameter is not valid - - """ - super(XenHvmHypervisor, self).ValidateParameters(hvparams) - - # hvm_cdrom_image_path verification - iso_path = hvparams[constants.HV_CDROM_IMAGE_PATH] - if iso_path and not os.path.isfile(iso_path): - raise errors.HypervisorError("The HVM CDROM image must either be a" - " regular file or a symlink pointing to" - " an existing regular file, not %s" % - iso_path) + ANCILLARY_FILES = XenHypervisor.ANCILLARY_FILES + \ + [constants.VNC_PASSWORD_FILE] + + PARAMETERS = { + constants.HV_ACPI: hv_base.NO_CHECK, + constants.HV_BOOT_ORDER: (True, ) + \ + (lambda x: x and len(x.strip("acdn")) == 0, + "Invalid boot order specified, must be one or more of [acdn]", + None, None), + constants.HV_CDROM_IMAGE_PATH: hv_base.OPT_FILE_CHECK, + constants.HV_DISK_TYPE: \ + hv_base.ParamInSet(True, constants.HT_HVM_VALID_DISK_TYPES), + constants.HV_NIC_TYPE: \ + hv_base.ParamInSet(True, constants.HT_HVM_VALID_NIC_TYPES), + constants.HV_PAE: hv_base.NO_CHECK, + constants.HV_VNC_BIND_ADDRESS: \ + (False, utils.IsValidIP, + "VNC bind address is not a valid IP address", None, None), + constants.HV_KERNEL_PATH: hv_base.REQ_FILE_CHECK, + constants.HV_DEVICE_MODEL: hv_base.REQ_FILE_CHECK, + } @classmethod - def _WriteConfigFile(cls, instance, block_devices, extra_args): + def _WriteConfigFile(cls, instance, block_devices): """Create a Xen 3.1 HVM config file. """ + hvp = instance.hvparams + config = StringIO() config.write("# this is autogenerated by Ganeti, please do not edit\n#\n") - config.write("kernel = '/usr/lib/xen/boot/hvmloader'\n") + + # kernel handling + kpath = hvp[constants.HV_KERNEL_PATH] + config.write("kernel = '%s'\n" % kpath) + config.write("builder = 'hvm'\n") config.write("memory = %d\n" % instance.beparams[constants.BE_MEMORY]) config.write("vcpus = %d\n" % instance.beparams[constants.BE_VCPUS]) config.write("name = '%s'\n" % instance.name) - if instance.hvparams[constants.HV_PAE]: + if hvp[constants.HV_PAE]: config.write("pae = 1\n") else: config.write("pae = 0\n") - if instance.hvparams[constants.HV_ACPI]: + if hvp[constants.HV_ACPI]: config.write("acpi = 1\n") else: config.write("acpi = 0\n") config.write("apic = 1\n") - arch = os.uname()[4] - if '64' in arch: - config.write("device_model = '/usr/lib64/xen/bin/qemu-dm'\n") - else: - config.write("device_model = '/usr/lib/xen/bin/qemu-dm'\n") - config.write("boot = '%s'\n" % instance.hvparams[constants.HV_BOOT_ORDER]) + config.write("device_model = '%s'\n" % hvp[constants.HV_DEVICE_MODEL]) + config.write("boot = '%s'\n" % hvp[constants.HV_BOOT_ORDER]) config.write("sdl = 0\n") config.write("usb = 1\n") config.write("usbdevice = 'tablet'\n") config.write("vnc = 1\n") - if instance.hvparams[constants.HV_VNC_BIND_ADDRESS] is None: + if hvp[constants.HV_VNC_BIND_ADDRESS] is None: config.write("vnclisten = '%s'\n" % constants.VNC_DEFAULT_BIND_ADDRESS) else: - config.write("vnclisten = '%s'\n" % - instance.hvparams["vnc_bind_address"]) + config.write("vnclisten = '%s'\n" % hvp[constants.HV_VNC_BIND_ADDRESS]) - if instance.network_port > constants.HT_HVM_VNC_BASE_PORT: - display = instance.network_port - constants.HT_HVM_VNC_BASE_PORT + if instance.network_port > constants.VNC_BASE_PORT: + display = instance.network_port - constants.VNC_BASE_PORT config.write("vncdisplay = %s\n" % display) config.write("vncunused = 0\n") else: @@ -598,11 +566,11 @@ class XenHvmHypervisor(XenHypervisor): config.write("localtime = 1\n") vif_data = [] - nic_type = instance.hvparams[constants.HV_NIC_TYPE] + nic_type = hvp[constants.HV_NIC_TYPE] if nic_type is None: # ensure old instances don't change nic_type_str = ", type=ioemu" - elif nic_type == constants.HT_HVM_DEV_PARAVIRTUAL: + elif nic_type == constants.HT_NIC_PARAVIRTUAL: nic_type_str = ", type=paravirtualized" else: nic_type_str = ", model=%s, type=ioemu" % nic_type @@ -616,13 +584,13 @@ class XenHvmHypervisor(XenHypervisor): config.write("vif = [%s]\n" % ",".join(vif_data)) disk_data = cls._GetConfigFileDiskData(instance.disk_template, block_devices) - disk_type = instance.hvparams[constants.HV_DISK_TYPE] - if disk_type in (None, constants.HT_HVM_DEV_IOEMU): + disk_type = hvp[constants.HV_DISK_TYPE] + if disk_type in (None, constants.HT_DISK_IOEMU): replacement = ",ioemu:hd" else: replacement = ",hd" disk_data = [line.replace(",sd", replacement) for line in disk_data] - iso_path = instance.hvparams[constants.HV_CDROM_IMAGE_PATH] + iso_path = hvp[constants.HV_CDROM_IMAGE_PATH] if iso_path: iso = "'file:%s,hdc:cdrom,r'" % iso_path disk_data.append(iso) @@ -632,8 +600,6 @@ class XenHvmHypervisor(XenHypervisor): config.write("on_poweroff = 'destroy'\n") config.write("on_reboot = 'restart'\n") config.write("on_crash = 'restart'\n") - if extra_args: - config.write("extra = '%s'\n" % extra_args) # just in case it exists utils.RemoveFile("/etc/xen/auto/%s" % instance.name) try: