Edit: plugin_disk.py
import errno from . import hotplug from .decorators import * import tuned.logs import tuned.consts as consts from tuned.utils.commands import commands import os import re log = tuned.logs.get() class DiskPlugin(hotplug.Plugin): """ `disk`:: Plug-in for tuning various block device options. This plug-in can also dynamically change the advanced power management and spindown timeout setting for a drive according to the current drive utilization. The dynamic tuning is controlled by the [option]`dynamic` and the global [option]`dynamic_tuning` option in `tuned-main.conf`. + The disk plug-in operates on all supported block devices unless a comma separated list of [option]`devices` is passed to it. + .Operate only on the sda block device ==== ---- [disk] # Comma separated list of devices, all devices if commented out. devices=sda ---- ==== + The [option]`elevator` option sets the Linux I/O scheduler. + .Use the bfq I/O scheduler on xvda block device ==== ---- [disk] device=xvda elevator=bfq ---- ==== + The [option]`scheduler_quantum` option only applies to the CFQ I/O scheduler. It defines the number of I/O requests that CFQ sends to one device at one time, essentially limiting queue depth. The default value is 8 requests. The device being used may support greater queue depth, but increasing the value of quantum will also increase latency, especially for large sequential write work loads. + The [option]`apm` option sets the Advanced Power Management feature on drives that support it. It corresponds to using the `-B` option of the `hdparm` utility. The [option]`spindown` option puts the drive into idle (low-power) mode, and also sets the standby (spindown) timeout for the drive. It corresponds to using `-S` option of the `hdparm` utility. + .Use a medium-agressive power management with spindown ==== ---- [disk] apm=128 spindown=6 ---- ==== + The [option]`readahead` option controls how much extra data the operating system reads from disk when performing sequential I/O operations. Increasing the `readahead` value might improve performance in application environments where sequential reading of large files takes place. The default unit for readahead is KiB. This can be adjusted to sectors by specifying the suffix 's'. If the suffix is specified, there must be at least one space between the number and suffix (for example, `readahead=8192 s`). + .Set the `readahead` to 4MB unless already set to a higher value ==== ---- [disk] readahead=>4096 ---- ==== The disk readahead value can be multiplied by the constant specified by the [option]`readahead_multiply` option. """ def __init__(self, *args, **kwargs): super(DiskPlugin, self).__init__(*args, **kwargs) self._power_levels = [254, 225, 195, 165, 145, 125, 105, 85, 70, 55, 30, 20] self._spindown_levels = [0, 250, 230, 210, 190, 170, 150, 130, 110, 90, 70, 60] self._levels = len(self._power_levels) self._level_steps = 6 self._load_smallest = 0.01 self._cmd = commands() def _init_devices(self): super(DiskPlugin, self)._init_devices() self._devices_supported = True self._use_hdparm = True self._free_devices = set() self._hdparm_apm_device_support = dict() for device in self._hardware_inventory.get_devices("block"): if self._device_is_supported(device): self._free_devices.add(device.sys_name) self._assigned_devices = set() def _get_device_objects(self, devices): return [self._hardware_inventory.get_device("block", x) for x in devices] def _is_hdparm_apm_supported(self, device): if not self._use_hdparm: return False if device in self._hdparm_apm_device_support: return self._hdparm_apm_device_support[device] (rc, out, err_msg) = self._cmd.execute(["hdparm", "-C", "/dev/%s" % device], \ no_errors = [errno.ENOENT], return_err=True) if rc == -errno.ENOENT: log.warn("hdparm command not found, ignoring for other devices") self._use_hdparm = False return False elif rc: log.info("Device '%s' not supported by hdparm" % device) log.debug("(rc: %s, msg: '%s')" % (rc, err_msg)) self._hdparm_apm_device_support[device] = False return False elif "unknown" in out: log.info("Driver for device '%s' does not support apm command" % device) self._hdparm_apm_device_support[device] = False return False self._hdparm_apm_device_support[device] = True return True @classmethod def _device_is_supported(cls, device): return device.device_type == "disk" and \ device.attributes.get("removable", None) == b"0" and \ (device.parent is None or \ device.parent.subsystem in ["scsi", "virtio", "xen", "nvme"]) def _hardware_events_init(self): self._hardware_inventory.subscribe(self, "block", self._hardware_events_callback) def _hardware_events_cleanup(self): self._hardware_inventory.unsubscribe(self) def _hardware_events_callback(self, event, device): if self._device_is_supported(device) or event == "remove": super(DiskPlugin, self)._hardware_events_callback(event, device) def _added_device_apply_tuning(self, instance, device_name): if instance._load_monitor is not None: instance._load_monitor.add_device(device_name) super(DiskPlugin, self)._added_device_apply_tuning(instance, device_name) def _removed_device_unapply_tuning(self, instance, device_name): if instance._load_monitor is not None: instance._load_monitor.remove_device(device_name) super(DiskPlugin, self)._removed_device_unapply_tuning(instance, device_name) @classmethod def _get_config_options(cls): return { "dynamic" : True, # FIXME: do we want this default? "elevator" : None, "apm" : None, "spindown" : None, "readahead" : None, "readahead_multiply" : None, "scheduler_quantum" : None, } @classmethod def _get_config_options_used_by_dynamic(cls): return [ "apm", "spindown", ] def _instance_init(self, instance): instance._has_static_tuning = True self._apm_errcnt = 0 self._spindown_errcnt = 0 if self._option_bool(instance.options["dynamic"]): instance._has_dynamic_tuning = True instance._load_monitor = \ self._monitors_repository.create( "disk", instance.assigned_devices) instance._device_idle = {} instance._stats = {} instance._idle = {} instance._spindown_change_delayed = {} else: instance._has_dynamic_tuning = False instance._load_monitor = None def _instance_cleanup(self, instance): if instance._load_monitor is not None: self._monitors_repository.delete(instance._load_monitor) instance._load_monitor = None def _update_errcnt(self, rc, spindown): if spindown: s = "spindown" cnt = self._spindown_errcnt else: s = "apm" cnt = self._apm_errcnt if cnt >= consts.ERROR_THRESHOLD: return if rc == 0: cnt = 0 elif rc == -errno.ENOENT: self._spindown_errcnt = self._apm_errcnt = consts.ERROR_THRESHOLD + 1 log.warn("hdparm command not found, ignoring future set_apm / set_spindown commands") return else: cnt += 1 if cnt == consts.ERROR_THRESHOLD: log.info("disabling set_%s command: too many consecutive errors" % s) if spindown: self._spindown_errcnt = cnt else: self._apm_errcnt = cnt def _change_spindown(self, instance, device, new_spindown_level): log.debug("changing spindown to %d" % new_spindown_level) (rc, out) = self._cmd.execute(["hdparm", "-S%d" % new_spindown_level, "/dev/%s" % device], no_errors = [errno.ENOENT]) self._update_errcnt(rc, True) instance._spindown_change_delayed[device] = False def _drive_spinning(self, device): (rc, out) = self._cmd.execute(["hdparm", "-C", "/dev/%s" % device], no_errors = [errno.ENOENT]) return not "standby" in out and not "sleeping" in out def _instance_update_dynamic(self, instance, device): if not self._is_hdparm_apm_supported(device): return load = instance._load_monitor.get_device_load(device) if load is None: return if not device in instance._stats: self._init_stats_and_idle(instance, device) self._update_stats(instance, device, load) self._update_idle(instance, device) stats = instance._stats[device] idle = instance._idle[device] # level change decision if idle["level"] + 1 < self._levels and idle["read"] >= self._level_steps and idle["write"] >= self._level_steps: level_change = 1 elif idle["level"] > 0 and (idle["read"] == 0 or idle["write"] == 0): level_change = -1 else: level_change = 0 # change level if decided if level_change != 0: idle["level"] += level_change new_power_level = self._power_levels[idle["level"]] new_spindown_level = self._spindown_levels[idle["level"]] log.debug("tuning level changed to %d" % idle["level"]) if self._spindown_errcnt < consts.ERROR_THRESHOLD: if not self._drive_spinning(device) and level_change > 0: log.debug("delaying spindown change to %d, drive has already spun down" % new_spindown_level) instance._spindown_change_delayed[device] = True else: self._change_spindown(instance, device, new_spindown_level) if self._apm_errcnt < consts.ERROR_THRESHOLD: log.debug("changing APM_level to %d" % new_power_level) (rc, out) = self._cmd.execute(["hdparm", "-B%d" % new_power_level, "/dev/%s" % device], no_errors = [errno.ENOENT]) self._update_errcnt(rc, False) elif instance._spindown_change_delayed[device] and self._drive_spinning(device): new_spindown_level = self._spindown_levels[idle["level"]] self._change_spindown(instance, device, new_spindown_level) log.debug("%s load: read %0.2f, write %0.2f" % (device, stats["read"], stats["write"])) log.debug("%s idle: read %d, write %d, level %d" % (device, idle["read"], idle["write"], idle["level"])) def _init_stats_and_idle(self, instance, device): instance._stats[device] = { "new": 11 * [0], "old": 11 * [0], "max": 11 * [1] } instance._idle[device] = { "level": 0, "read": 0, "write": 0 } instance._spindown_change_delayed[device] = False def _update_stats(self, instance, device, new_load): instance._stats[device]["old"] = old_load = instance._stats[device]["new"] instance._stats[device]["new"] = new_load # load difference diff = [new_old[0] - new_old[1] for new_old in zip(new_load, old_load)] instance._stats[device]["diff"] = diff # adapt maximum expected load if the difference is higher old_max_load = instance._stats[device]["max"] max_load = [max(pair) for pair in zip(old_max_load, diff)] instance._stats[device]["max"] = max_load # read/write ratio instance._stats[device]["read"] = float(diff[1]) / float(max_load[1]) instance._stats[device]["write"] = float(diff[5]) / float(max_load[5]) def _update_idle(self, instance, device): # increase counter if there is no load, otherwise reset the counter for operation in ["read", "write"]: if instance._stats[device][operation] < self._load_smallest: instance._idle[device][operation] += 1 else: instance._idle[device][operation] = 0 def _instance_apply_dynamic(self, instance, device): # At the moment we support dynamic tuning just for devices compatible with hdparm apm commands # If in future will be added new functionality not connected to this command, # it is needed to change it here if not self._is_hdparm_apm_supported(device): log.info("There is no dynamic tuning available for device '%s' at time" % device) else: super(DiskPlugin, self)._instance_apply_dynamic(instance, device) def _instance_unapply_dynamic(self, instance, device): pass def _sysfs_path(self, device, suffix, prefix = "/sys/block/"): if "/" in device: dev = os.path.join(prefix, device.replace("/", "!"), suffix) if os.path.exists(dev): return dev return os.path.join(prefix, device, suffix) def _elevator_file(self, device): return self._sysfs_path(device, "queue/scheduler") @command_set("elevator", per_device=True) def _set_elevator(self, value, device, sim, remove): sys_file = self._elevator_file(device) if not sim: self._cmd.write_to_file(sys_file, value, \ no_error = [errno.ENOENT] if remove else False) return value @command_get("elevator") def _get_elevator(self, device, ignore_missing=False): sys_file = self._elevator_file(device) # example of scheduler file content: # noop deadline [cfq] return self._cmd.get_active_option(self._cmd.read_file(sys_file, no_error=ignore_missing)) @command_set("apm", per_device=True) def _set_apm(self, value, device, sim, remove): if not self._is_hdparm_apm_supported(device): if not sim: log.info("apm option is not supported for device '%s'" % device) return None else: return str(value) if self._apm_errcnt < consts.ERROR_THRESHOLD: if not sim: (rc, out) = self._cmd.execute(["hdparm", "-B", str(value), "/dev/" + device], no_errors = [errno.ENOENT]) self._update_errcnt(rc, False) return str(value) else: return None @command_get("apm") def _get_apm(self, device, ignore_missing=False): if not self._is_hdparm_apm_supported(device): if not ignore_missing: log.info("apm option is not supported for device '%s'" % device) return None value = None err = False (rc, out) = self._cmd.execute(["hdparm", "-B", "/dev/" + device], no_errors = [errno.ENOENT]) if rc == -errno.ENOENT: return None elif rc != 0: err = True else: m = re.match(r".*=\s*(\d+).*", out, re.S) if m: try: value = int(m.group(1)) except ValueError: err = True if err: log.error("could not get current APM settings for device '%s'" % device) return value @command_set("spindown", per_device=True) def _set_spindown(self, value, device, sim, remove): if not self._is_hdparm_apm_supported(device): if not sim: log.info("spindown option is not supported for device '%s'" % device) return None else: return str(value) if self._spindown_errcnt < consts.ERROR_THRESHOLD: if not sim: (rc, out) = self._cmd.execute(["hdparm", "-S", str(value), "/dev/" + device], no_errors = [errno.ENOENT]) self._update_errcnt(rc, True) return str(value) else: return None @command_get("spindown") def _get_spindown(self, device, ignore_missing=False): if not self._is_hdparm_apm_supported(device): if not ignore_missing: log.info("spindown option is not supported for device '%s'" % device) return None # There's no way how to get current/old spindown value, hardcoding vendor specific 253 return 253 def _readahead_file(self, device): return self._sysfs_path(device, "queue/read_ahead_kb") def _parse_ra(self, value): val = str(value).split(None, 1) try: v = int(val[0]) except ValueError: return None if len(val) > 1 and val[1][0] == "s": # v *= 512 / 1024 v /= 2 return v @command_set("readahead", per_device=True) def _set_readahead(self, value, device, sim, remove): sys_file = self._readahead_file(device) val = self._parse_ra(value) if val is None: log.error("Invalid readahead value '%s' for device '%s'" % (value, device)) else: if not sim: self._cmd.write_to_file(sys_file, "%d" % val, \ no_error = [errno.ENOENT] if remove else False) return val @command_get("readahead") def _get_readahead(self, device, ignore_missing=False): sys_file = self._readahead_file(device) value = self._cmd.read_file(sys_file, no_error=ignore_missing).strip() if len(value) == 0: return None return int(value) @command_custom("readahead_multiply", per_device=True) def _multiply_readahead(self, enabling, multiplier, device, verify, ignore_missing): if verify: return None storage_key = self._storage_key( command_name = "readahead_multiply", device_name = device) if enabling: old_readahead = self._get_readahead(device) if old_readahead is None: return new_readahead = int(float(multiplier) * old_readahead) self._storage.set(storage_key, old_readahead) self._set_readahead(new_readahead, device, False) else: old_readahead = self._storage.get(storage_key) if old_readahead is None: return self._set_readahead(old_readahead, device, False) self._storage.unset(storage_key) def _scheduler_quantum_file(self, device): return self._sysfs_path(device, "queue/iosched/quantum") @command_set("scheduler_quantum", per_device=True) def _set_scheduler_quantum(self, value, device, sim, remove): sys_file = self._scheduler_quantum_file(device) if not sim: self._cmd.write_to_file(sys_file, "%d" % int(value), \ no_error = [errno.ENOENT] if remove else False) return value @command_get("scheduler_quantum") def _get_scheduler_quantum(self, device, ignore_missing=False): sys_file = self._scheduler_quantum_file(device) value = self._cmd.read_file(sys_file, no_error=ignore_missing).strip() if len(value) == 0: if not ignore_missing: log.info("disk_scheduler_quantum option is not supported for device '%s'" % device) return None return int(value)
Cancel