X-Git-Url: https://code.grnet.gr/git/ganeti-local/blobdiff_plain/2899d9de6d928f3c3e3f7292b739c1a2076fa408..da961187f97344fde390140ebb2f10d10d334d51:/lib/bdev.py?ds=sidebyside diff --git a/lib/bdev.py b/lib/bdev.py index 8910f80..9d3f08b 100644 --- a/lib/bdev.py +++ b/lib/bdev.py @@ -33,6 +33,38 @@ from ganeti import errors from ganeti import constants +def _IgnoreError(fn, *args, **kwargs): + """Executes the given function, ignoring BlockDeviceErrors. + + This is used in order to simplify the execution of cleanup or + rollback functions. + + @rtype: boolean + @return: True when fn didn't raise an exception, False otherwise + + """ + try: + fn(*args, **kwargs) + return True + except errors.BlockDeviceError, err: + logging.warning("Caught BlockDeviceError but ignoring: %s" % str(err)) + return False + + +def _ThrowError(msg, *args): + """Log an error to the node daemon and the raise an exception. + + @type msg: string + @param msg: the text of the exception + @raise errors.BlockDeviceError + + """ + if args: + msg = msg % args + logging.error(msg) + raise errors.BlockDeviceError(msg) + + class BlockDev(object): """Block device abstract class. @@ -87,36 +119,16 @@ class BlockDev(object): def Assemble(self): """Assemble the device from its components. - If this is a plain block device (e.g. LVM) than assemble does - nothing, as the LVM has no children and we don't put logical - volumes offline. - - One guarantee is that after the device has been assembled, it - knows its major/minor numbers. This allows other devices (usually - parents) to probe correctly for their children. + Implementations of this method by child classes must ensure that: + - after the device has been assembled, it knows its major/minor + numbers; this allows other devices (usually parents) to probe + correctly for their children + - calling this method on an existing, in-use device is safe + - if the device is already configured (and in an OK state), + this method is idempotent """ - status = True - for child in self._children: - if not isinstance(child, BlockDev): - raise TypeError("Invalid child passed of type '%s'" % type(child)) - if not status: - break - status = status and child.Assemble() - if not status: - break - - try: - child.Open() - except errors.BlockDeviceError: - for child in self._children: - child.Shutdown() - raise - - if not status: - for child in self._children: - child.Shutdown() - return status + pass def Attach(self): """Find a device which matches our config and attach to it. @@ -202,9 +214,6 @@ class BlockDev(object): If this device is a mirroring device, this function returns the status of the mirror. - Returns: - (sync_percent, estimated_time, is_degraded, ldisk) - If sync_percent is None, it means the device is not syncing. If estimated_time is None, it means we can't estimate @@ -218,6 +227,9 @@ class BlockDev(object): data. This is only valid for some devices, the rest will always return False (not degraded). + @rtype: tuple + @return: (sync_percent, estimated_time, is_degraded, ldisk) + """ return None, None, False, False @@ -259,10 +271,7 @@ class BlockDev(object): def Grow(self, amount): """Grow the block device. - Arguments: - amount: the amount (in mebibytes) to grow with - - Returns: None + @param amount: the amount (in mebibytes) to grow with """ raise NotImplementedError @@ -298,12 +307,12 @@ class LogicalVolume(BlockDev): """ if not isinstance(unique_id, (tuple, list)) or len(unique_id) != 2: - raise ValueError("Invalid configuration data %s" % str(unique_id)) + raise errors.ProgrammerError("Invalid configuration data %s" % + str(unique_id)) vg_name, lv_name = unique_id pvs_info = cls.GetPVInfo(vg_name) if not pvs_info: - raise errors.BlockDeviceError("Can't compute PV info for vg %s" % - vg_name) + _ThrowError("Can't compute PV info for vg %s", vg_name) pvs_info.sort() pvs_info.reverse() @@ -313,24 +322,23 @@ class LogicalVolume(BlockDev): # The size constraint should have been checked from the master before # calling the create function. if free_size < size: - raise errors.BlockDeviceError("Not enough free space: required %s," - " available %s" % (size, free_size)) + _ThrowError("Not enough free space: required %s," + " available %s", size, free_size) result = utils.RunCmd(["lvcreate", "-L%dm" % size, "-n%s" % lv_name, vg_name] + pvlist) if result.failed: - raise errors.BlockDeviceError("%s - %s" % (result.fail_reason, - result.output)) + _ThrowError("LV create failed (%s): %s", + result.fail_reason, result.output) return LogicalVolume(unique_id, children) @staticmethod def GetPVInfo(vg_name): """Get the free space info for PVs in a volume group. - Args: - vg_name: the volume group name + @param vg_name: the volume group name - Returns: - list of (free_space, name) with free_space in mebibytes + @rtype: list + @return: list of tuples (free_space, name) with free_space in mebibytes """ command = ["pvs", "--noheadings", "--nosuffix", "--units=m", @@ -360,14 +368,11 @@ class LogicalVolume(BlockDev): """ if not self.minor and not self.Attach(): # the LV does not exist - return True + return result = utils.RunCmd(["lvremove", "-f", "%s/%s" % (self._vg_name, self._lv_name)]) if result.failed: - logging.error("Can't lvremove: %s - %s", - result.fail_reason, result.output) - - return not result.failed + _ThrowError("Can't lvremove: %s - %s", result.fail_reason, result.output) def Rename(self, new_id): """Rename this logical volume. @@ -382,8 +387,7 @@ class LogicalVolume(BlockDev): (self._vg_name, new_vg)) result = utils.RunCmd(["lvrename", new_vg, self._lv_name, new_name]) if result.failed: - raise errors.BlockDeviceError("Failed to rename the logical volume: %s" % - result.output) + _ThrowError("Failed to rename the logical volume: %s", result.output) self._lv_name = new_name self.dev_path = "/dev/%s/%s" % (self._vg_name, self._lv_name) @@ -437,9 +441,7 @@ class LogicalVolume(BlockDev): """ result = utils.RunCmd(["lvchange", "-ay", self.dev_path]) if result.failed: - logging.error("Can't activate lv %s: %s", self.dev_path, result.output) - return False - return self.Attach() + _ThrowError("Can't activate lv %s: %s", self.dev_path, result.output) def Shutdown(self): """Shutdown the device. @@ -448,7 +450,7 @@ class LogicalVolume(BlockDev): volumes on shutdown. """ - return True + pass def GetSyncStatus(self): """Returns the sync status of the device. @@ -456,9 +458,6 @@ class LogicalVolume(BlockDev): If this device is a mirroring device, this function returns the status of the mirror. - Returns: - (sync_percent, estimated_time, is_degraded, ldisk) - For logical volumes, sync_percent and estimated_time are always None (no recovery in progress, as we don't handle the mirrored LV case). The is_degraded parameter is the inverse of the ldisk @@ -472,6 +471,9 @@ class LogicalVolume(BlockDev): The status was already read in Attach, so we just return it. + @rtype: tuple + @return: (sync_percent, estimated_time, is_degraded, ldisk) + """ return None, None, self._degraded, self._degraded @@ -499,25 +501,23 @@ class LogicalVolume(BlockDev): # remove existing snapshot if found snap = LogicalVolume((self._vg_name, snap_name), None) - snap.Remove() + _IgnoreError(snap.Remove) pvs_info = self.GetPVInfo(self._vg_name) if not pvs_info: - raise errors.BlockDeviceError("Can't compute PV info for vg %s" % - self._vg_name) + _ThrowError("Can't compute PV info for vg %s", self._vg_name) pvs_info.sort() pvs_info.reverse() free_size, pv_name = pvs_info[0] if free_size < size: - raise errors.BlockDeviceError("Not enough free space: required %s," - " available %s" % (size, free_size)) + _ThrowError("Not enough free space: required %s," + " available %s", size, free_size) result = utils.RunCmd(["lvcreate", "-L%dm" % size, "-s", "-n%s" % snap_name, self.dev_path]) if result.failed: - raise errors.BlockDeviceError("command: %s error: %s - %s" % - (result.cmd, result.fail_reason, - result.output)) + _ThrowError("command: %s error: %s - %s", + result.cmd, result.fail_reason, result.output) return snap_name @@ -537,9 +537,9 @@ class LogicalVolume(BlockDev): result = utils.RunCmd(["lvchange", "--addtag", text, self.dev_path]) if result.failed: - raise errors.BlockDeviceError("Command: %s error: %s - %s" % - (result.cmd, result.fail_reason, - result.output)) + _ThrowError("Command: %s error: %s - %s", result.cmd, result.fail_reason, + result.output) + def Grow(self, amount): """Grow the logical volume. @@ -553,8 +553,7 @@ class LogicalVolume(BlockDev): "-L", "+%dm" % amount, self.dev_path]) if not result.failed: return - raise errors.BlockDeviceError("Can't grow LV %s: %s" % - (self.dev_path, result.output)) + _ThrowError("Can't grow LV %s: %s", self.dev_path, result.output) class DRBD8Status(object): @@ -563,20 +562,28 @@ class DRBD8Status(object): Note that this doesn't support unconfigured devices (cs:Unconfigured). """ - LINE_RE = re.compile(r"\s*[0-9]+:\s*cs:(\S+)\s+st:([^/]+)/(\S+)" + UNCONF_RE = re.compile(r"\s*[0-9]+:\s*cs:Unconfigured$") + LINE_RE = re.compile(r"\s*[0-9]+:\s*cs:(\S+)\s+(?:st|ro):([^/]+)/(\S+)" "\s+ds:([^/]+)/(\S+)\s+.*$") SYNC_RE = re.compile(r"^.*\ssync'ed:\s*([0-9.]+)%.*" "\sfinish: ([0-9]+):([0-9]+):([0-9]+)\s.*$") def __init__(self, procline): - m = self.LINE_RE.match(procline) - if not m: - raise errors.BlockDeviceError("Can't parse input data '%s'" % procline) - self.cstatus = m.group(1) - self.lrole = m.group(2) - self.rrole = m.group(3) - self.ldisk = m.group(4) - self.rdisk = m.group(5) + u = self.UNCONF_RE.match(procline) + if u: + self.cstatus = "Unconfigured" + self.lrole = self.rrole = self.ldisk = self.rdisk = None + else: + m = self.LINE_RE.match(procline) + if not m: + raise errors.BlockDeviceError("Can't parse input data '%s'" % procline) + self.cstatus = m.group(1) + self.lrole = m.group(2) + self.rrole = m.group(3) + self.ldisk = m.group(4) + self.rdisk = m.group(5) + + # end reading of data from the LINE_RE or UNCONF_RE self.is_standalone = self.cstatus == "StandAlone" self.is_wfconn = self.cstatus == "WFConnection" @@ -591,6 +598,9 @@ class DRBD8Status(object): self.is_diskless = self.ldisk == "Diskless" self.is_disk_uptodate = self.ldisk == "UpToDate" + self.is_in_resync = self.cstatus in ("SyncSource", "SyncTarget") + self.is_in_use = self.cstatus != "Unconfigured" + m = self.SYNC_RE.match(procline) if m: self.sync_percent = float(m.group(1)) @@ -629,21 +639,28 @@ class BaseDRBD(BlockDev): """Return data from /proc/drbd. """ - stat = open(filename, "r") try: - data = stat.read().splitlines() - finally: - stat.close() + stat = open(filename, "r") + try: + data = stat.read().splitlines() + finally: + stat.close() + except EnvironmentError, err: + if err.errno == errno.ENOENT: + _ThrowError("The file %s cannot be opened, check if the module" + " is loaded (%s)", filename, str(err)) + else: + _ThrowError("Can't read the DRBD proc file %s: %s", filename, str(err)) if not data: - raise errors.BlockDeviceError("Can't read any data from %s" % filename) + _ThrowError("Can't read any data from %s", filename) return data @staticmethod def _MassageProcData(data): """Transform the output of _GetProdData into a nicer form. - Returns: - a dictionary of minor: joined lines from /proc/drbd for that minor + @return: a dictionary of minor: joined lines from /proc/drbd + for that minor """ lmatch = re.compile("^ *([0-9]+):.*$") @@ -669,12 +686,12 @@ class BaseDRBD(BlockDev): """Return the DRBD version. This will return a dict with keys: - k_major, - k_minor, - k_point, - api, - proto, - proto2 (only on drbd > 8.2.X) + - k_major + - k_minor + - k_point + - api + - proto + - proto2 (only on drbd > 8.2.X) """ proc_data = cls._GetProcData() @@ -704,7 +721,7 @@ class BaseDRBD(BlockDev): return "/dev/drbd%d" % minor @classmethod - def _GetUsedDevs(cls): + def GetUsedDevs(cls): """Compute the list of used DRBD devices. """ @@ -748,22 +765,17 @@ class BaseDRBD(BlockDev): """ result = utils.RunCmd(["blockdev", "--getsize", meta_device]) if result.failed: - logging.error("Failed to get device size: %s - %s", - result.fail_reason, result.output) - return False + _ThrowError("Failed to get device size: %s - %s", + result.fail_reason, result.output) try: sectors = int(result.stdout) except ValueError: - logging.error("Invalid output from blockdev: '%s'", result.stdout) - return False + _ThrowError("Invalid output from blockdev: '%s'", result.stdout) bytes = sectors * 512 if bytes < 128 * 1024 * 1024: # less than 128MiB - logging.error("Meta device too small (%.2fMib)", (bytes / 1024 / 1024)) - return False + _ThrowError("Meta device too small (%.2fMib)", (bytes / 1024 / 1024)) if bytes > (128 + 32) * 1024 * 1024: # account for an extra (big) PE on LVM - logging.error("Meta device too big (%.2fMiB)", (bytes / 1024 / 1024)) - return False - return True + _ThrowError("Meta device too big (%.2fMiB)", (bytes / 1024 / 1024)) def Rename(self, new_id): """Rename a device. @@ -790,6 +802,9 @@ class DRBD8(BaseDRBD): _MAX_MINORS = 255 _PARSE_SHOW = None + # timeout constants + _NET_RECONFIG_TIMEOUT = 60 + def __init__(self, unique_id, children): if children and children.count(None) > 0: children = [] @@ -797,10 +812,9 @@ class DRBD8(BaseDRBD): self.major = self._DRBD_MAJOR version = self._GetVersion() if version['k_major'] != 8 : - raise errors.BlockDeviceError("Mismatch in DRBD kernel version and" - " requested ganeti usage: kernel is" - " %s.%s, ganeti wants 8.x" % - (version['k_major'], version['k_minor'])) + _ThrowError("Mismatch in DRBD kernel version and requested ganeti" + " usage: kernel is %s.%s, ganeti wants 8.x", + version['k_major'], version['k_minor']) if len(children) not in (0, 2): raise ValueError("Invalid configuration data %s" % str(children)) @@ -825,8 +839,7 @@ class DRBD8(BaseDRBD): result = utils.RunCmd(["drbdmeta", "--force", cls._DevPath(minor), "v08", dev_path, "0", "create-md"]) if result.failed: - raise errors.BlockDeviceError("Can't initialize meta device: %s" % - result.output) + _ThrowError("Can't initialize meta device: %s", result.output) @classmethod def _FindUnusedMinor(cls): @@ -857,21 +870,6 @@ class DRBD8(BaseDRBD): return highest + 1 @classmethod - def _IsValidMeta(cls, meta_device): - """Check if the given meta device looks like a valid one. - - """ - minor = cls._FindUnusedMinor() - minor_path = cls._DevPath(minor) - result = utils.RunCmd(["drbdmeta", minor_path, - "v08", meta_device, "0", - "dstate"]) - if result.failed: - logging.error("Invalid meta device %s: %s", meta_device, result.output) - return False - return True - - @classmethod def _GetShowParser(cls): """Return a parser for `drbd show` output. @@ -898,15 +896,20 @@ class DRBD8(BaseDRBD): # value types value = pyp.Word(pyp.alphanums + '_-/.:') quoted = dbl_quote + pyp.CharsNotIn('"') + dbl_quote - addr_port = (pyp.Word(pyp.nums + '.') + pyp.Literal(':').suppress() + - number) + addr_type = (pyp.Optional(pyp.Literal("ipv4")).suppress() + + pyp.Optional(pyp.Literal("ipv6")).suppress()) + addr_port = (addr_type + pyp.Word(pyp.nums + '.') + + pyp.Literal(':').suppress() + number) # meta device, extended syntax meta_value = ((value ^ quoted) + pyp.Literal('[').suppress() + number + pyp.Word(']').suppress()) + # device name, extended syntax + device_value = pyp.Literal("minor").suppress() + number # a statement stmt = (~rbrace + keyword + ~lbrace + - pyp.Optional(addr_port ^ value ^ quoted ^ meta_value) + + pyp.Optional(addr_port ^ value ^ quoted ^ meta_value ^ + device_value) + pyp.Optional(defa) + semi + pyp.Optional(pyp.restOfLine).suppress()) @@ -953,8 +956,7 @@ class DRBD8(BaseDRBD): try: results = bnf.parseString(out) except pyp.ParseException, err: - raise errors.BlockDeviceError("Can't parse drbdsetup show output: %s" % - str(err)) + _ThrowError("Can't parse drbdsetup show output: %s", str(err)) # and massage the results into our desired format for section in results: @@ -1032,18 +1034,12 @@ class DRBD8(BaseDRBD): def _AssembleLocal(cls, minor, backend, meta): """Configure the local part of a DRBD device. - This is the first thing that must be done on an unconfigured DRBD - device. And it must be done only once. - """ - if not cls._IsValidMeta(meta): - return False args = ["drbdsetup", cls._DevPath(minor), "disk", backend, meta, "0", "-e", "detach", "--create-device"] result = utils.RunCmd(args) if result.failed: - logging.error("Can't attach local disk: %s", result.output) - return not result.failed + _ThrowError("drbd%d: can't attach local disk: %s", minor, result.output) @classmethod def _AssembleNet(cls, minor, net_info, protocol, @@ -1055,7 +1051,16 @@ class DRBD8(BaseDRBD): if None in net_info: # we don't want network connection and actually want to make # sure its shutdown - return cls._ShutdownNet(minor) + cls._ShutdownNet(minor) + return + + # Workaround for a race condition. When DRBD is doing its dance to + # establish a connection with its peer, it also sends the + # synchronization speed over the wire. In some cases setting the + # sync speed only after setting up both sides can race with DRBD + # connecting, hence we set it here before telling DRBD anything + # about its peer. + cls._SetMinorSyncSpeed(minor, constants.SYNC_SPEED) args = ["drbdsetup", cls._DevPath(minor), "net", "%s:%s" % (lhost, lport), "%s:%s" % (rhost, rport), protocol, @@ -1069,9 +1074,8 @@ class DRBD8(BaseDRBD): args.extend(["-a", hmac, "-x", secret]) result = utils.RunCmd(args) if result.failed: - logging.error("Can't setup network for dbrd device: %s - %s", - result.fail_reason, result.output) - return False + _ThrowError("drbd%d: can't setup network: %s - %s", + minor, result.fail_reason, result.output) timeout = time.time() + 10 ok = False @@ -1087,34 +1091,29 @@ class DRBD8(BaseDRBD): ok = True break if not ok: - logging.error("Timeout while configuring network") - return False - return True + _ThrowError("drbd%d: timeout while configuring network", minor) def AddChildren(self, devices): """Add a disk to the DRBD device. """ if self.minor is None: - raise errors.BlockDeviceError("Can't attach to dbrd8 during AddChildren") + _ThrowError("drbd%d: can't attach to dbrd8 during AddChildren", + self._aminor) if len(devices) != 2: - raise errors.BlockDeviceError("Need two devices for AddChildren") + _ThrowError("drbd%d: need two devices for AddChildren", self.minor) info = self._GetDevInfo(self._GetShowData(self.minor)) if "local_dev" in info: - raise errors.BlockDeviceError("DRBD8 already attached to a local disk") + _ThrowError("drbd%d: already attached to a local disk", self.minor) backend, meta = devices if backend.dev_path is None or meta.dev_path is None: - raise errors.BlockDeviceError("Children not ready during AddChildren") + _ThrowError("drbd%d: children not ready during AddChildren", self.minor) backend.Open() meta.Open() - if not self._CheckMetaSize(meta.dev_path): - raise errors.BlockDeviceError("Invalid meta device size") + self._CheckMetaSize(meta.dev_path) self._InitMeta(self._FindUnusedMinor(), meta.dev_path) - if not self._IsValidMeta(meta.dev_path): - raise errors.BlockDeviceError("Cannot initalize meta device") - if not self._AssembleLocal(self.minor, backend.dev_path, meta.dev_path): - raise errors.BlockDeviceError("Can't attach to local storage") + self._AssembleLocal(self.minor, backend.dev_path, meta.dev_path) self._children = devices def RemoveChildren(self, devices): @@ -1122,62 +1121,78 @@ class DRBD8(BaseDRBD): """ if self.minor is None: - raise errors.BlockDeviceError("Can't attach to drbd8 during" - " RemoveChildren") + _ThrowError("drbd%d: can't attach to drbd8 during RemoveChildren", + self._aminor) # early return if we don't actually have backing storage info = self._GetDevInfo(self._GetShowData(self.minor)) if "local_dev" not in info: return if len(self._children) != 2: - raise errors.BlockDeviceError("We don't have two children: %s" % - self._children) + _ThrowError("drbd%d: we don't have two children: %s", self.minor, + self._children) if self._children.count(None) == 2: # we don't actually have children :) - logging.error("Requested detach while detached") + logging.warning("drbd%d: requested detach while detached", self.minor) return if len(devices) != 2: - raise errors.BlockDeviceError("We need two children in RemoveChildren") + _ThrowError("drbd%d: we need two children in RemoveChildren", self.minor) for child, dev in zip(self._children, devices): if dev != child.dev_path: - raise errors.BlockDeviceError("Mismatch in local storage" - " (%s != %s) in RemoveChildren" % - (dev, child.dev_path)) + _ThrowError("drbd%d: mismatch in local storage (%s != %s) in" + " RemoveChildren", self.minor, dev, child.dev_path) - if not self._ShutdownLocal(self.minor): - raise errors.BlockDeviceError("Can't detach from local storage") + self._ShutdownLocal(self.minor) self._children = [] - def SetSyncSpeed(self, kbytes): + @classmethod + def _SetMinorSyncSpeed(cls, minor, kbytes): """Set the speed of the DRBD syncer. + This is the low-level implementation. + + @type minor: int + @param minor: the drbd minor whose settings we change + @type kbytes: int + @param kbytes: the speed in kbytes/second + @rtype: boolean + @return: the success of the operation + """ - children_result = super(DRBD8, self).SetSyncSpeed(kbytes) - if self.minor is None: - logging.info("Instance not attached to a device") - return False - result = utils.RunCmd(["drbdsetup", self.dev_path, "syncer", "-r", "%d" % - kbytes]) + result = utils.RunCmd(["drbdsetup", cls._DevPath(minor), "syncer", + "-r", "%d" % kbytes, "--create-device"]) if result.failed: logging.error("Can't change syncer rate: %s - %s", result.fail_reason, result.output) - return not result.failed and children_result + return not result.failed + + def SetSyncSpeed(self, kbytes): + """Set the speed of the DRBD syncer. + + @type kbytes: int + @param kbytes: the speed in kbytes/second + @rtype: boolean + @return: the success of the operation + + """ + if self.minor is None: + logging.info("Not attached during SetSyncSpeed") + return False + children_result = super(DRBD8, self).SetSyncSpeed(kbytes) + return self._SetMinorSyncSpeed(self.minor, kbytes) and children_result def GetProcStatus(self): """Return device data from /proc. """ if self.minor is None: - raise errors.BlockDeviceError("GetStats() called while not attached") + _ThrowError("drbd%d: GetStats() called while not attached", self._aminor) proc_info = self._MassageProcData(self._GetProcData()) if self.minor not in proc_info: - raise errors.BlockDeviceError("Can't find myself in /proc (minor %d)" % - self.minor) + _ThrowError("drbd%d: can't find myself in /proc", self.minor) return DRBD8Status(proc_info[self.minor]) def GetSyncStatus(self): """Returns the sync status of the device. - Returns: - (sync_percent, estimated_time, is_degraded) If sync_percent is None, it means all is ok If estimated_time is None, it means we can't esimate @@ -1190,9 +1205,12 @@ class DRBD8(BaseDRBD): We compute the ldisk parameter based on wheter we have a local disk or not. + @rtype: tuple + @return: (sync_percent, estimated_time, is_degraded, ldisk) + """ if self.minor is None and not self.Attach(): - raise errors.BlockDeviceError("Can't attach to device in GetSyncStatus") + _ThrowError("drbd%d: can't Attach() in GetSyncStatus", self._aminor) stats = self.GetProcStatus() ldisk = not stats.is_disk_uptodate is_degraded = not stats.is_connected @@ -1215,9 +1233,8 @@ class DRBD8(BaseDRBD): cmd.append("-o") result = utils.RunCmd(cmd) if result.failed: - msg = ("Can't make drbd device primary: %s" % result.output) - logging.error(msg) - raise errors.BlockDeviceError(msg) + _ThrowError("drbd%d: can't make drbd device primary: %s", self.minor, + result.output) def Close(self): """Make the local state secondary. @@ -1226,43 +1243,165 @@ class DRBD8(BaseDRBD): """ if self.minor is None and not self.Attach(): - logging.info("Instance not attached to a device") - raise errors.BlockDeviceError("Can't find device") + _ThrowError("drbd%d: can't Attach() in Close()", self._aminor) result = utils.RunCmd(["drbdsetup", self.dev_path, "secondary"]) if result.failed: - msg = ("Can't switch drbd device to" - " secondary: %s" % result.output) - logging.error(msg) - raise errors.BlockDeviceError(msg) + _ThrowError("drbd%d: can't switch drbd device to secondary: %s", + self.minor, result.output) + + def DisconnectNet(self): + """Removes network configuration. + + This method shutdowns the network side of the device. + + The method will wait up to a hardcoded timeout for the device to + go into standalone after the 'disconnect' command before + re-configuring it, as sometimes it takes a while for the + disconnect to actually propagate and thus we might issue a 'net' + command while the device is still connected. If the device will + still be attached to the network and we time out, we raise an + exception. + + """ + if self.minor is None: + _ThrowError("drbd%d: disk not attached in re-attach net", self._aminor) + + if None in (self._lhost, self._lport, self._rhost, self._rport): + _ThrowError("drbd%d: DRBD disk missing network info in" + " DisconnectNet()", self.minor) + + ever_disconnected = _IgnoreError(self._ShutdownNet, self.minor) + timeout_limit = time.time() + self._NET_RECONFIG_TIMEOUT + sleep_time = 0.100 # we start the retry time at 100 miliseconds + while time.time() < timeout_limit: + status = self.GetProcStatus() + if status.is_standalone: + break + # retry the disconnect, it seems possible that due to a + # well-time disconnect on the peer, my disconnect command might + # be ingored and forgotten + ever_disconnected = _IgnoreError(self._ShutdownNet, self.minor) or \ + ever_disconnected + time.sleep(sleep_time) + sleep_time = min(2, sleep_time * 1.5) + + if not status.is_standalone: + if ever_disconnected: + msg = ("drbd%d: device did not react to the" + " 'disconnect' command in a timely manner") + else: + msg = "drbd%d: can't shutdown network, even after multiple retries" + _ThrowError(msg, self.minor) + + reconfig_time = time.time() - timeout_limit + self._NET_RECONFIG_TIMEOUT + if reconfig_time > 15: # hardcoded alert limit + logging.info("drbd%d: DisconnectNet: detach took %.3f seconds", + self.minor, reconfig_time) + + def AttachNet(self, multimaster): + """Reconnects the network. + + This method connects the network side of the device with a + specified multi-master flag. The device needs to be 'Standalone' + but have valid network configuration data. + + Args: + - multimaster: init the network in dual-primary mode + + """ + if self.minor is None: + _ThrowError("drbd%d: device not attached in AttachNet", self._aminor) + + if None in (self._lhost, self._lport, self._rhost, self._rport): + _ThrowError("drbd%d: missing network info in AttachNet()", self.minor) + + status = self.GetProcStatus() + + if not status.is_standalone: + _ThrowError("drbd%d: device is not standalone in AttachNet", self.minor) + + self._AssembleNet(self.minor, + (self._lhost, self._lport, self._rhost, self._rport), + constants.DRBD_NET_PROTOCOL, dual_pri=multimaster, + hmac=constants.DRBD_HMAC_ALG, secret=self._secret) def Attach(self): - """Find a DRBD device which matches our config and attach to it. + """Check if our minor is configured. + + This doesn't do any device configurations - it only checks if the + minor is in a state different from Unconfigured. + + Note that this function will not change the state of the system in + any way (except in case of side-effects caused by reading from + /proc). + + """ + used_devs = self.GetUsedDevs() + if self._aminor in used_devs: + minor = self._aminor + else: + minor = None + + self._SetFromMinor(minor) + return minor is not None + + def Assemble(self): + """Assemble the drbd. + + Method: + - if we have a configured device, we try to ensure that it matches + our config + - if not, we create it from zero + + """ + super(DRBD8, self).Assemble() + + self.Attach() + if self.minor is None: + # local device completely unconfigured + self._FastAssemble() + else: + # we have to recheck the local and network status and try to fix + # the device + self._SlowAssemble() + + def _SlowAssemble(self): + """Assembles the DRBD device from a (partially) configured device. In case of partially attached (local device matches but no network setup), we perform the network attach. If successful, we re-test the attach if can return success. """ + net_data = (self._lhost, self._lport, self._rhost, self._rport) for minor in (self._aminor,): info = self._GetDevInfo(self._GetShowData(minor)) match_l = self._MatchesLocal(info) match_r = self._MatchesNet(info) + if match_l and match_r: + # everything matches break + if match_l and not match_r and "local_addr" not in info: - res_r = self._AssembleNet(minor, - (self._lhost, self._lport, - self._rhost, self._rport), - "C", hmac=constants.DRBD_HMAC_ALG, - secret=self._secret - ) - if res_r: - if self._MatchesNet(self._GetDevInfo(self._GetShowData(minor))): - break - # the weakest case: we find something that is only net attached - # even though we were passed some children at init time + # disk matches, but not attached to network, attach and recheck + self._AssembleNet(minor, net_data, constants.DRBD_NET_PROTOCOL, + hmac=constants.DRBD_HMAC_ALG, secret=self._secret) + if self._MatchesNet(self._GetDevInfo(self._GetShowData(minor))): + break + else: + _ThrowError("drbd%d: network attach successful, but 'drbdsetup" + " show' disagrees", minor) + if match_r and "local_dev" not in info: - break + # no local disk, but network attached and it matches + self._AssembleLocal(minor, self._children[0].dev_path, + self._children[1].dev_path) + if self._MatchesNet(self._GetDevInfo(self._GetShowData(minor))): + break + else: + _ThrowError("drbd%d: disk attach successful, but 'drbdsetup" + " show' disagrees", minor) # this case must be considered only if we actually have local # storage, i.e. not in diskless mode, because all diskless @@ -1274,79 +1413,47 @@ class DRBD8(BaseDRBD): # else, even though its local storage is ours; as we own the # drbd space, we try to disconnect from the remote peer and # reconnect to our correct one - if not self._ShutdownNet(minor): - raise errors.BlockDeviceError("Device has correct local storage," - " wrong remote peer and is unable to" - " disconnect in order to attach to" - " the correct peer") + try: + self._ShutdownNet(minor) + except errors.BlockDeviceError, err: + _ThrowError("drbd%d: device has correct local storage, wrong" + " remote peer and is unable to disconnect in order" + " to attach to the correct peer: %s", minor, str(err)) # note: _AssembleNet also handles the case when we don't want # local storage (i.e. one or more of the _[lr](host|port) is # None) - if (self._AssembleNet(minor, (self._lhost, self._lport, - self._rhost, self._rport), "C", - hmac=constants.DRBD_HMAC_ALG, - secret=self._secret) and - self._MatchesNet(self._GetDevInfo(self._GetShowData(minor)))): + self._AssembleNet(minor, net_data, constants.DRBD_NET_PROTOCOL, + hmac=constants.DRBD_HMAC_ALG, secret=self._secret) + if self._MatchesNet(self._GetDevInfo(self._GetShowData(minor))): break + else: + _ThrowError("drbd%d: network attach successful, but 'drbdsetup" + " show' disagrees", minor) else: minor = None self._SetFromMinor(minor) - return minor is not None + if minor is None: + _ThrowError("drbd%d: cannot activate, unknown or unhandled reason", + self._aminor) - def Assemble(self): - """Assemble the drbd. + def _FastAssemble(self): + """Assemble the drbd device from zero. - Method: - - if we have a local backing device, we bind to it by: - - checking the list of used drbd devices - - check if the local minor use of any of them is our own device - - if yes, abort? - - if not, bind - - if we have a local/remote net info: - - redo the local backing device step for the remote device - - check if any drbd device is using the local port, - if yes abort - - check if any remote drbd device is using the remote - port, if yes abort (for now) - - bind our net port - - bind the remote net port + This is run when in Assemble we detect our minor is unused. """ - self.Attach() - if self.minor is not None: - logging.info("Already assembled") - return True - - result = super(DRBD8, self).Assemble() - if not result: - return result - - # TODO: maybe completely tear-down the minor (drbdsetup ... down) - # before attaching our own? minor = self._aminor - need_localdev_teardown = False if self._children and self._children[0] and self._children[1]: - result = self._AssembleLocal(minor, self._children[0].dev_path, - self._children[1].dev_path) - if not result: - return False - need_localdev_teardown = True + self._AssembleLocal(minor, self._children[0].dev_path, + self._children[1].dev_path) if self._lhost and self._lport and self._rhost and self._rport: - result = self._AssembleNet(minor, - (self._lhost, self._lport, - self._rhost, self._rport), - "C", hmac=constants.DRBD_HMAC_ALG, - secret=self._secret) - if not result: - if need_localdev_teardown: - # we will ignore failures from this - logging.error("net setup failed, tearing down local device") - self._ShutdownAll(minor) - return False + self._AssembleNet(minor, + (self._lhost, self._lport, self._rhost, self._rport), + constants.DRBD_NET_PROTOCOL, + hmac=constants.DRBD_HMAC_ALG, secret=self._secret) self._SetFromMinor(minor) - return True @classmethod def _ShutdownLocal(cls, minor): @@ -1358,8 +1465,7 @@ class DRBD8(BaseDRBD): """ result = utils.RunCmd(["drbdsetup", cls._DevPath(minor), "detach"]) if result.failed: - logging.error("Can't detach local device: %s", result.output) - return not result.failed + _ThrowError("drbd%d: can't detach local disk: %s", minor, result.output) @classmethod def _ShutdownNet(cls, minor): @@ -1370,8 +1476,7 @@ class DRBD8(BaseDRBD): """ result = utils.RunCmd(["drbdsetup", cls._DevPath(minor), "disconnect"]) if result.failed: - logging.error("Can't shutdown network: %s", result.output) - return not result.failed + _ThrowError("drbd%d: can't shutdown network: %s", minor, result.output) @classmethod def _ShutdownAll(cls, minor): @@ -1382,27 +1487,26 @@ class DRBD8(BaseDRBD): """ result = utils.RunCmd(["drbdsetup", cls._DevPath(minor), "down"]) if result.failed: - logging.error("Can't shutdown drbd device: %s", result.output) - return not result.failed + _ThrowError("drbd%d: can't shutdown drbd device: %s", + minor, result.output) def Shutdown(self): """Shutdown the DRBD device. """ if self.minor is None and not self.Attach(): - logging.info("DRBD device not attached to a device during Shutdown") - return True - if not self._ShutdownAll(self.minor): - return False + logging.info("drbd%d: not attached during Shutdown()", self._aminor) + return + minor = self.minor self.minor = None self.dev_path = None - return True + self._ShutdownAll(minor) def Remove(self): """Stub remove for DRBD devices. """ - return self.Shutdown() + self.Shutdown() @classmethod def Create(cls, unique_id, children, size): @@ -1414,15 +1518,23 @@ class DRBD8(BaseDRBD): """ if len(children) != 2: raise errors.ProgrammerError("Invalid setup for the drbd device") + # check that the minor is unused + aminor = unique_id[4] + proc_info = cls._MassageProcData(cls._GetProcData()) + if aminor in proc_info: + status = DRBD8Status(proc_info[aminor]) + in_use = status.is_in_use + else: + in_use = False + if in_use: + _ThrowError("drbd%d: minor is already in use at Create() time", aminor) meta = children[1] meta.Assemble() if not meta.Attach(): - raise errors.BlockDeviceError("Can't attach to meta device") - if not cls._CheckMetaSize(meta.dev_path): - raise errors.BlockDeviceError("Invalid meta device size") - cls._InitMeta(cls._FindUnusedMinor(), meta.dev_path) - if not cls._IsValidMeta(meta.dev_path): - raise errors.BlockDeviceError("Cannot initalize meta device") + _ThrowError("drbd%d: can't attach to meta device '%s'", + aminor, meta) + cls._CheckMetaSize(meta.dev_path) + cls._InitMeta(aminor, meta.dev_path) return cls(unique_id, children) def Grow(self, amount): @@ -1430,15 +1542,13 @@ class DRBD8(BaseDRBD): """ if self.minor is None: - raise errors.ProgrammerError("drbd8: Grow called while not attached") + _ThrowError("drbd%d: Grow called while not attached", self._aminor) if len(self._children) != 2 or None in self._children: - raise errors.BlockDeviceError("Cannot grow diskless DRBD8 device") + _ThrowError("drbd%d: cannot grow diskless device", self.minor) self._children[0].Grow(amount) result = utils.RunCmd(["drbdsetup", self.dev_path, "resize"]) if result.failed: - raise errors.BlockDeviceError("resize failed for %s: %s" % - (self.dev_path, result.output)) - return + _ThrowError("drbd%d: resize failed: %s", self.minor, result.output) class FileStorage(BlockDev): @@ -1460,6 +1570,7 @@ class FileStorage(BlockDev): raise ValueError("Invalid configuration data %s" % str(unique_id)) self.driver = unique_id[0] self.dev_path = unique_id[1] + self.Attach() def Assemble(self): """Assemble the device. @@ -1468,9 +1579,7 @@ class FileStorage(BlockDev): """ if not os.path.exists(self.dev_path): - raise errors.BlockDeviceError("File device '%s' does not exist." % - self.dev_path) - return True + _ThrowError("File device '%s' does not exist" % self.dev_path) def Shutdown(self): """Shutdown the device. @@ -1479,7 +1588,7 @@ class FileStorage(BlockDev): the file on shutdown. """ - return True + pass def Open(self, force=False): """Make the device ready for I/O. @@ -1500,54 +1609,49 @@ class FileStorage(BlockDev): def Remove(self): """Remove the file backing the block device. - Returns: - boolean indicating wheter removal of file was successful or not. + @rtype: boolean + @return: True if the removal was successful """ - if not os.path.exists(self.dev_path): - return True try: os.remove(self.dev_path) - return True except OSError, err: - logging.error("Can't remove file '%s': %s", self.dev_path, err) - return False + if err.errno != errno.ENOENT: + _ThrowError("Can't remove file '%s': %s", self.dev_path, err) def Attach(self): """Attach to an existing file. Check if this file already exists. - Returns: - boolean indicating if file exists or not. + @rtype: boolean + @return: True if file exists """ - if os.path.exists(self.dev_path): - return True - return False + self.attached = os.path.exists(self.dev_path) + return self.attached @classmethod def Create(cls, unique_id, children, size): """Create a new file. - Args: - children: - size: integer size of file in MiB + @param size: the size of file in MiB - Returns: - A ganeti.bdev.FileStorage object. + @rtype: L{bdev.FileStorage} + @return: an instance of FileStorage """ if not isinstance(unique_id, (tuple, list)) or len(unique_id) != 2: raise ValueError("Invalid configuration data %s" % str(unique_id)) dev_path = unique_id[1] + if os.path.exists(dev_path): + _ThrowError("File already existing: %s", dev_path) try: f = open(dev_path, 'w') - except IOError, err: - raise errors.BlockDeviceError("Could not create '%'" % err) - else: f.truncate(size * 1024 * 1024) f.close() + except IOError, err: + _ThrowError("Error in file creation: %", str(err)) return FileStorage(unique_id, children) @@ -1571,25 +1675,20 @@ def FindDevice(dev_type, unique_id, children): device = DEV_MAP[dev_type](unique_id, children) if not device.attached: return None - return device + return device -def AttachOrAssemble(dev_type, unique_id, children): +def Assemble(dev_type, unique_id, children): """Try to attach or assemble an existing device. - This will attach to an existing assembled device or will assemble - the device, as needed, to bring it fully up. + This will attach to assemble the device, as needed, to bring it + fully up. It must be safe to run on already-assembled devices. """ if dev_type not in DEV_MAP: raise errors.ProgrammerError("Invalid block device type '%s'" % dev_type) device = DEV_MAP[dev_type](unique_id, children) - if not device.attached: - device.Assemble() - if not device.attached: - raise errors.BlockDeviceError("Can't find a valid block device for" - " %s/%s/%s" % - (dev_type, unique_id, children)) + device.Assemble() return device