diff options
Diffstat (limited to 'src/pybind/mgr/progress/module.py')
| -rw-r--r-- | src/pybind/mgr/progress/module.py | 693 |
1 files changed, 693 insertions, 0 deletions
diff --git a/src/pybind/mgr/progress/module.py b/src/pybind/mgr/progress/module.py new file mode 100644 index 00000000..f880bb6e --- /dev/null +++ b/src/pybind/mgr/progress/module.py @@ -0,0 +1,693 @@ +from mgr_module import MgrModule +import threading +import datetime +import uuid + +import json + + +ENCODING_VERSION = 1 + +# keep a global reference to the module so we can use it from Event methods +_module = None + + +class Event(object): + """ + A generic "event" that has a start time, completion percentage, + and a list of "refs" that are (type, id) tuples describing which + objects (osds, pools) this relates to. + """ + + def __init__(self, message, refs): + self._message = message + self._refs = refs + + self.started_at = datetime.datetime.utcnow() + + self.id = None + + def _refresh(self): + global _module + _module.log.debug('refreshing mgr for %s (%s) at %f' % (self.id, self._message, + self.progress)) + _module.update_progress_event(self.id, self._message, self.progress) + + @property + def message(self): + return self._message + + @property + def refs(self): + return self._refs + + @property + def progress(self): + raise NotImplementedError() + + @property + def failed(self): + return False + + @property + def failure_message(self): + return None + + def summary(self): + return "{0} {1}".format(self.progress, self.message) + + def _progress_str(self, width): + inner_width = width - 2 + out = "[" + done_chars = int(self.progress * inner_width) + out += done_chars * '=' + out += (inner_width - done_chars) * '.' + out += "]" + + return out + + def twoline_progress(self): + """ + e.g. + + - Eating my delicious strudel + [===============..............] + + """ + return "{0}\n {1}".format( + self._message, self._progress_str(30)) + + def to_json(self): + return { + "id": self.id, + "message": self.message, + "refs": self._refs + } + + +class GhostEvent(Event): + """ + The ghost of a completed event: these are the fields that we persist + after the event is complete. + """ + + def __init__(self, my_id, message, refs, + failed=False, failure_message=None): + super(GhostEvent, self).__init__(message, refs) + self.id = my_id + + if failed: + self._failed = True + self._failure_message = failure_message + else: + self._failed = False + + @property + def progress(self): + return 1.0 + + @property + def failed(self): + return self._failed + + @property + def failure_message(self): + return self._failure_message if self._failed else None + + def to_json(self): + d = { + "id": self.id, + "message": self.message, + "refs": self._refs + } + if self._failed: + d["failed"] = True + d["failure_message"] = self._failure_message + return d + + +class RemoteEvent(Event): + """ + An event that was published by another module: we know nothing about + this, rely on the other module to continuously update us with + progress information as it emerges. + """ + + def __init__(self, my_id, message, refs): + super(RemoteEvent, self).__init__(message, refs) + self.id = my_id + self._progress = 0.0 + self._failed = False + self._refresh() + + def set_progress(self, progress): + self._progress = progress + self._refresh() + + def set_failed(self, message): + self._progress = 1.0 + self._failed = True + self._failure_message = message + self._refresh() + + @property + def progress(self): + return self._progress + + @property + def failed(self): + return self._failed + + @property + def failure_message(self): + return self._failure_message if self._failed else None + + +class PgRecoveryEvent(Event): + """ + An event whose completion is determined by the recovery of a set of + PGs to a healthy state. + + Always call update() immediately after construction. + """ + + def __init__(self, message, refs, which_pgs, evacuate_osds): + super(PgRecoveryEvent, self).__init__(message, refs) + + self._pgs = which_pgs + + self._evacuate_osds = evacuate_osds + + self._original_pg_count = len(self._pgs) + + self._original_bytes_recovered = None + + self._progress = 0.0 + + self._start_epoch = _module.get_osdmap().get_epoch() + + self.id = str(uuid.uuid4()) + self._refresh() + + @property + def evacuating_osds(self): + return self. _evacuate_osds + + def pg_update(self, raw_pg_stats, pg_ready, log): + # FIXME: O(pg_num) in python + # FIXME: far more fields getting pythonized than we really care about + pg_to_state = dict([(p['pgid'], p) for p in raw_pg_stats['pg_stats']]) + if self._original_bytes_recovered is None: + self._original_bytes_recovered = {} + missing_pgs = [] + for pg in self._pgs: + pg_str = str(pg) + if pg_str in pg_to_state: + self._original_bytes_recovered[pg] = \ + pg_to_state[pg_str]['stat_sum']['num_bytes_recovered'] + else: + missing_pgs.append(pg) + if pg_ready: + for pg in missing_pgs: + self._pgs.remove(pg) + + complete_accumulate = 0.0 + + # Calculating progress as the number of PGs recovered divided by the + # original where partially completed PGs count for something + # between 0.0-1.0. This is perhaps less faithful than looking at the + # total number of bytes recovered, but it does a better job of + # representing the work still to do if there are a number of very + # few-bytes PGs that still need the housekeeping of their recovery + # to be done. This is subjective... + + complete = set() + for pg in self._pgs: + pg_str = str(pg) + try: + info = pg_to_state[pg_str] + except KeyError: + # The PG is gone! Probably a pool was deleted. Drop it. + complete.add(pg) + continue + # Only checks the state of each PGs when it's epoch >= the OSDMap's epoch + if int(info['reported_epoch']) < int(self._start_epoch): + continue + + state = info['state'] + + states = state.split("+") + + if "active" in states and "clean" in states: + complete.add(pg) + else: + if info['stat_sum']['num_bytes'] == 0: + # Empty PGs are considered 0% done until they are + # in the correct state. + pass + else: + recovered = info['stat_sum']['num_bytes_recovered'] + total_bytes = info['stat_sum']['num_bytes'] + if total_bytes > 0: + ratio = float(recovered - + self._original_bytes_recovered[pg]) / \ + total_bytes + # Since the recovered bytes (over time) could perhaps + # exceed the contents of the PG (moment in time), we + # must clamp this + ratio = min(ratio, 1.0) + + else: + # Dataless PGs (e.g. containing only OMAPs) count + # as half done. + ratio = 0.5 + complete_accumulate += ratio + + self._pgs = list(set(self._pgs) ^ complete) + completed_pgs = self._original_pg_count - len(self._pgs) + self._progress = (completed_pgs + complete_accumulate)\ + / self._original_pg_count + self._refresh() + + log.info("Updated progress to {0} ({1})".format( + self._progress, self._message + )) + + @property + def progress(self): + return self._progress + + +class PgId(object): + def __init__(self, pool_id, ps): + self.pool_id = pool_id + self.ps = ps + + def __cmp__(self, other): + return (self.pool_id, self.ps) == (other.pool_id, other.ps) + + def __lt__(self, other): + return (self.pool_id, self.ps) < (other.pool_id, other.ps) + + def __str__(self): + return "{0}.{1:x}".format(self.pool_id, self.ps) + + +class Module(MgrModule): + COMMANDS = [ + {"cmd": "progress", + "desc": "Show progress of recovery operations", + "perm": "r"}, + {"cmd": "progress json", + "desc": "Show machine readable progress information", + "perm": "r"}, + {"cmd": "progress clear", + "desc": "Reset progress tracking", + "perm": "rw"}, + {"cmd": "progress on", + "desc": "Enable progress tracking", + "perm": "rw"}, + {"cmd": "progress off", + "desc": "Disable progress tracking", + "perm": "rw"} + + ] + + MODULE_OPTIONS = [ + { + 'name': 'max_completed_events', + 'default': 50, + 'type': 'int', + 'desc': 'number of past completed events to remember', + 'runtime': True, + }, + { + 'name': 'persist_interval', + 'default': 5, + 'type': 'secs', + 'desc': 'how frequently to persist completed events', + 'runtime': True, + }, + { + 'name': 'enabled', + 'default': True, + 'type': 'bool', + } + ] + + def __init__(self, *args, **kwargs): + super(Module, self).__init__(*args, **kwargs) + + self._events = {} + self._completed_events = [] + + self._old_osd_map = None + + self._ready = threading.Event() + self._shutdown = threading.Event() + + self._latest_osdmap = None + + self._dirty = False + + global _module + _module = self + + def config_notify(self): + for opt in self.MODULE_OPTIONS: + setattr(self, + opt['name'], + self.get_module_option(opt['name'])) + self.log.debug(' %s = %s', opt['name'], getattr(self, opt['name'])) + + def _osd_in_out(self, old_map, old_dump, new_map, osd_id, marked): + # A function that will create or complete an event when an + # OSD is marked in or out according to the affected PGs + affected_pgs = [] + unmoved_pgs = [] + for pool in old_dump['pools']: + pool_id = pool['pool'] + for ps in range(0, pool['pg_num']): + + # Was this OSD affected by the OSD coming in/out? + # Compare old and new osds using + # data from the json dump + old_up_acting = old_map.pg_to_up_acting_osds(pool['pool'], ps) + old_osds = set(old_up_acting['acting']) + + new_up_acting = new_map.pg_to_up_acting_osds(pool['pool'], ps) + new_osds = set(new_up_acting['acting']) + + # Check the osd_id being in the acting set for both old + # and new maps to cover both out and in cases + was_on_out_or_in_osd = osd_id in old_osds or osd_id in new_osds + if not was_on_out_or_in_osd: + continue + + self.log.debug("pool_id, ps = {0}, {1}".format( + pool_id, ps + )) + + self.log.debug( + "old_up_acting: {0}".format(json.dumps(old_up_acting, indent=2))) + + # Has this OSD been assigned a new location? + # (it might not be if there is no suitable place to move + # after an OSD is marked in/out) + if marked == "in": + is_relocated = len(old_osds - new_osds) > 0 + else: + is_relocated = len(new_osds - old_osds) > 0 + + self.log.debug( + "new_up_acting: {0}".format(json.dumps(new_up_acting, + indent=2))) + + if was_on_out_or_in_osd and is_relocated: + # This PG is now in motion, track its progress + affected_pgs.append(PgId(pool_id, ps)) + elif not is_relocated: + # This PG didn't get a new location, we'll log it + unmoved_pgs.append(PgId(pool_id, ps)) + + # In the case that we ignored some PGs, log the reason why (we may + # not end up creating a progress event) + if len(unmoved_pgs): + self.log.warn("{0} PGs were on osd.{1}, but didn't get new locations".format( + len(unmoved_pgs), osd_id)) + + self.log.warn("{0} PGs affected by osd.{1} being marked {2}".format( + len(affected_pgs), osd_id, marked)) + + + # In the case of the osd coming back in, we might need to cancel + # previous recovery event for that osd + if marked == "in": + for ev_id in list(self._events): + ev = self._events[ev_id] + if isinstance(ev, PgRecoveryEvent) and osd_id in ev.evacuating_osds: + self.log.info("osd.{0} came back in, cancelling event".format( + osd_id + )) + self._complete(ev) + + if len(affected_pgs) > 0: + ev = PgRecoveryEvent( + "Rebalancing after osd.{0} marked {1}".format(osd_id, marked), + refs=[("osd", osd_id)], + which_pgs=affected_pgs, + evacuate_osds=[osd_id] + ) + ev.pg_update(self.get("pg_dump"), self.get("pg_ready"), self.log) + self._events[ev.id] = ev + + def _osdmap_changed(self, old_osdmap, new_osdmap): + old_dump = old_osdmap.dump() + new_dump = new_osdmap.dump() + + old_osds = dict([(o['osd'], o) for o in old_dump['osds']]) + + for osd in new_dump['osds']: + osd_id = osd['osd'] + new_weight = osd['in'] + if osd_id in old_osds: + old_weight = old_osds[osd_id]['in'] + + if new_weight == 0.0 and old_weight > new_weight: + self.log.warn("osd.{0} marked out".format(osd_id)) + self._osd_in_out(old_osdmap, old_dump, new_osdmap, osd_id, "out") + elif new_weight >= 1.0 and old_weight == 0.0: + # Only consider weight>=1.0 as "in" to avoid spawning + # individual recovery events on every adjustment + # in a gradual weight-in + self.log.warn("osd.{0} marked in".format(osd_id)) + self._osd_in_out(old_osdmap, old_dump, new_osdmap, osd_id, "in") + + def notify(self, notify_type, notify_data): + self._ready.wait() + if not self.enabled: + return + if notify_type == "osd_map": + old_osdmap = self._latest_osdmap + self._latest_osdmap = self.get_osdmap() + + self.log.info("Processing OSDMap change {0}..{1}".format( + old_osdmap.get_epoch(), self._latest_osdmap.get_epoch() + )) + self._osdmap_changed(old_osdmap, self._latest_osdmap) + elif notify_type == "pg_summary": + # if there are no events we will skip this here to avoid + # expensive get calls + if len(self._events) == 0: + return + data = self.get("pg_stats") + ready = self.get("pg_ready") + for ev_id, ev in self._events.items(): + if isinstance(ev, PgRecoveryEvent): + ev.pg_update(data, ready, self.log) + self.maybe_complete(ev) + + def maybe_complete(self, event): + if event.progress >= 1.0: + self._complete(event) + + def _save(self): + self.log.info("Writing back {0} completed events".format( + len(self._completed_events) + )) + # TODO: bound the number we store. + encoded = json.dumps({ + "events": [ev.to_json() for ev in self._completed_events], + "version": ENCODING_VERSION, + "compat_version": ENCODING_VERSION + }) + self.set_store("completed", encoded) + + def _load(self): + stored = self.get_store("completed") + + if stored is None: + self.log.info("No stored events to load") + return + + decoded = json.loads(stored) + if decoded['compat_version'] > ENCODING_VERSION: + raise RuntimeError("Cannot decode version {0}".format( + decoded['compat_version'])) + + for ev in decoded['events']: + self._completed_events.append(GhostEvent(ev['id'], ev['message'], + ev['refs'], + ev.get('failed', False), + ev.get('failure_message'))) + + self._prune_completed_events() + + def _prune_completed_events(self): + length = len(self._completed_events) + if length > self.max_completed_events: + self._completed_events = self._completed_events[length - self.max_completed_events : length] + self._dirty = True + + def serve(self): + self.config_notify() + self.clear_all_progress_events() + self.log.info("Loading...") + + self._load() + self.log.info("Loaded {0} historic events".format(self._completed_events)) + + self._latest_osdmap = self.get_osdmap() + self.log.info("Loaded OSDMap, ready.") + + self._ready.set() + + while not self._shutdown.is_set(): + # Lazy periodic write back of completed events + if self._dirty: + self._save() + self._dirty = False + + self._shutdown.wait(timeout=self.persist_interval) + + self._shutdown.wait() + + def shutdown(self): + self._shutdown.set() + self.clear_all_progress_events() + + def update(self, ev_id, ev_msg, ev_progress, refs=None): + """ + For calling from other mgr modules + """ + if not self.enabled: + return + + if refs is None: + refs = [] + try: + ev = self._events[ev_id] + except KeyError: + ev = RemoteEvent(ev_id, ev_msg, refs) + self._events[ev_id] = ev + self.log.info("update: starting ev {0} ({1})".format( + ev_id, ev_msg)) + else: + self.log.debug("update: {0} on {1}".format( + ev_progress, ev_msg)) + + ev.set_progress(ev_progress) + ev._refresh() + + def _complete(self, ev): + duration = (datetime.datetime.utcnow() - ev.started_at) + self.log.info("Completed event {0} ({1}) in {2} seconds".format( + ev.id, ev.message, duration.seconds + )) + self.complete_progress_event(ev.id) + + self._completed_events.append( + GhostEvent(ev.id, ev.message, ev.refs, + failed=ev.failed, failure_message=ev.failure_message)) + del self._events[ev.id] + self._prune_completed_events() + self._dirty = True + + def complete(self, ev_id): + """ + For calling from other mgr modules + """ + if not self.enabled: + return + try: + ev = self._events[ev_id] + ev.set_progress(1.0) + self.log.info("complete: finished ev {0} ({1})".format(ev_id, + ev.message)) + self._complete(ev) + except KeyError: + self.log.warn("complete: ev {0} does not exist".format(ev_id)) + pass + + def fail(self, ev_id, message): + """ + For calling from other mgr modules to mark an event as failed (and + complete) + """ + try: + ev = self._events[ev_id] + ev.set_failed(message) + self.log.info("fail: finished ev {0} ({1}): {2}".format(ev_id, + ev.message, + message)) + self._complete(ev) + except KeyError: + self.log.warn("fail: ev {0} does not exist".format(ev_id)) + + def on(self): + self.set_module_option('enabled', True) + + def off(self): + self.set_module_option('enabled', False) + + def _handle_ls(self): + if len(self._events) or len(self._completed_events): + out = "" + chrono_order = sorted(self._events.values(), + key=lambda x: x.started_at, reverse=True) + for ev in chrono_order: + out += ev.twoline_progress() + out += "\n" + + if len(self._completed_events): + # TODO: limit number of completed events to show + out += "\n" + for ev in self._completed_events: + out += "[{0}]: {1}\n".format("Complete" if not ev.failed else "Failed", + ev.twoline_progress()) + + return 0, out, "" + else: + return 0, "", "Nothing in progress" + + def _json(self): + return { + 'events': [ev.to_json() for ev in self._events.values()], + 'completed': [ev.to_json() for ev in self._completed_events] + } + + def clear(self): + self._events = {} + self._completed_events = [] + self._dirty = True + self._save() + self.clear_all_progress_events() + + def _handle_clear(self): + self.clear() + return 0, "", "" + + def handle_command(self, _, cmd): + if cmd['prefix'] == "progress": + return self._handle_ls() + elif cmd['prefix'] == "progress clear": + # The clear command isn't usually needed - it's to enable + # the admin to "kick" this module if it seems to have done + # something wrong (e.g. we have a bug causing a progress event + # that never finishes) + return self._handle_clear() + elif cmd['prefix'] == "progress json": + return 0, json.dumps(self._json(), indent=4, sort_keys=True), "" + elif cmd['prefix'] == "progress on": + if self.enabled: + return 0, "", "progress already enabled!" + self.on() + return 0, "", "progress enabled" + elif cmd['prefix'] == "progress off": + if not self.enabled: + return 0, "", "progress already disabled!" + self.off() + self.clear() + return 0, "", "progress disabled" + else: + raise NotImplementedError(cmd['prefix']) |