def test_unmerge(self): now = before_now(minutes=5).replace(microsecond=0, tzinfo=pytz.utc) def time_from_now(offset=0): return now + timedelta(seconds=offset) project = self.create_project() sequence = itertools.count(0) tag_values = itertools.cycle(["red", "green", "blue"]) user_values = itertools.cycle([{"id": 1}, {"id": 2}]) def create_message_event(template, parameters, environment, release, fingerprint="group1"): i = next(sequence) event_id = uuid.UUID(fields=(i, 0x0, 0x1000, 0x80, 0x80, 0x808080808080)).hex tags = [["color", next(tag_values)]] if release: tags.append(["sentry:release", release]) event = self.store_event( data={ "event_id": event_id, "message": template % parameters, "type": "default", "user": next(user_values), "tags": tags, "fingerprint": [fingerprint], "timestamp": iso_format(now + timedelta(seconds=i)), "environment": environment, "release": release, }, project_id=project.id, ) UserReport.objects.create( project_id=project.id, group_id=event.group.id, event_id=event_id, name="Log Hat", email="*****@*****.**", comments="Quack", ) features.record([event]) return event events = OrderedDict() for event in ( create_message_event( "This is message #%s.", i, environment="production", release="version" ) for i in xrange(10) ): events.setdefault(get_fingerprint(event), []).append(event) for event in ( create_message_event( "This is message #%s!", i, environment="production", release="version2", fingerprint="group2", ) for i in xrange(10, 16) ): events.setdefault(get_fingerprint(event), []).append(event) event = create_message_event( "This is message #%s!", 17, environment="staging", release="version3", fingerprint="group3", ) events.setdefault(get_fingerprint(event), []).append(event) merge_source, source, destination = list(Group.objects.all()) assert len(events) == 3 assert sum(map(len, events.values())) == 17 production_environment = Environment.objects.get( organization_id=project.organization_id, name="production" ) with self.tasks(): eventstream_state = eventstream.start_merge(project.id, [merge_source.id], source.id) merge_groups.delay([merge_source.id], source.id) eventstream.end_merge(eventstream_state) assert set( [ (gtv.value, gtv.times_seen) for gtv in tagstore.get_group_tag_values( project.id, source.id, production_environment.id, "color" ) ] ) == set([("red", 6), ("green", 5), ("blue", 5)]) similar_items = features.compare(source) assert len(similar_items) == 2 assert similar_items[0][0] == source.id assert similar_items[0][1]["message:message:character-shingles"] == 1.0 assert similar_items[1][0] == destination.id assert similar_items[1][1]["message:message:character-shingles"] < 1.0 with self.tasks(): eventstream_state = eventstream.start_unmerge( project.id, [list(events.keys())[0]], source.id, destination.id ) unmerge.delay( project.id, source.id, destination.id, [list(events.keys())[0]], None, batch_size=5 ) eventstream.end_unmerge(eventstream_state) assert ( list( Group.objects.filter(id=merge_source.id).values_list( "times_seen", "first_seen", "last_seen" ) ) == [] ) assert list( Group.objects.filter(id=source.id).values_list("times_seen", "first_seen", "last_seen") ) == [(6, time_from_now(10), time_from_now(15))] assert list( Group.objects.filter(id=destination.id).values_list( "times_seen", "first_seen", "last_seen" ) ) == [(11, time_from_now(0), time_from_now(16))] assert source.id != destination.id assert source.project == destination.project destination_event_ids = map(lambda event: event.event_id, list(events.values())[1]) assert set( UserReport.objects.filter(group_id=source.id).values_list("event_id", flat=True) ) == set(destination_event_ids) assert set( GroupHash.objects.filter(group_id=source.id).values_list("hash", flat=True) ) == set(itertools.islice(events.keys(), 2)) assert set( GroupRelease.objects.filter(group_id=source.id).values_list( "environment", "first_seen", "last_seen" ) ) == set([(u"production", time_from_now(10), time_from_now(15))]) assert set( [ (gtv.value, gtv.times_seen) for gtv in tagstore.get_group_tag_values( project.id, destination.id, production_environment.id, "color" ) ] ) == set([(u"red", 4), (u"green", 3), (u"blue", 3)]) destination_event_ids = map( lambda event: event.event_id, list(events.values())[0] + list(events.values())[2] ) assert set( UserReport.objects.filter(group_id=destination.id).values_list("event_id", flat=True) ) == set(destination_event_ids) assert set( GroupHash.objects.filter(group_id=destination.id).values_list("hash", flat=True) ) == set(itertools.islice(events.keys(), 2, 3)) assert set( GroupRelease.objects.filter(group_id=destination.id).values_list( "environment", "first_seen", "last_seen" ) ) == set( [ ("production", time_from_now(0), time_from_now(9)), ("staging", time_from_now(16), time_from_now(16)), ] ) assert set( [ (gtk.value, gtk.times_seen) for gtk in tagstore.get_group_tag_values( project.id, destination.id, production_environment.id, "color" ) ] ) == set([("red", 4), ("blue", 3), ("green", 3)]) rollup_duration = 3600 time_series = tsdb.get_range( tsdb.models.group, [source.id, destination.id], now - timedelta(seconds=rollup_duration), time_from_now(17), rollup_duration, ) environment_time_series = tsdb.get_range( tsdb.models.group, [source.id, destination.id], now - timedelta(seconds=rollup_duration), time_from_now(17), rollup_duration, environment_ids=[production_environment.id], ) def get_expected_series_values(rollup, events, function=None): if function is None: def function(aggregate, event): return (aggregate if aggregate is not None else 0) + 1 expected = {} for event in events: k = float((to_timestamp(event.datetime) // rollup_duration) * rollup_duration) expected[k] = function(expected.get(k), event) return expected def assert_series_contains(expected, actual, default=0): actual = dict(actual) for key, value in expected.items(): assert actual.get(key, 0) == value for key in set(actual.keys()) - set(expected.keys()): assert actual.get(key, 0) == default assert_series_contains( get_expected_series_values(rollup_duration, list(events.values())[1]), time_series[source.id], 0, ) assert_series_contains( get_expected_series_values( rollup_duration, list(events.values())[0] + list(events.values())[2] ), time_series[destination.id], 0, ) assert_series_contains( get_expected_series_values(rollup_duration, list(events.values())[1]), environment_time_series[source.id], 0, ) assert_series_contains( get_expected_series_values( rollup_duration, list(events.values())[0][:-1] + list(events.values())[2] ), environment_time_series[destination.id], 0, ) time_series = tsdb.get_distinct_counts_series( tsdb.models.users_affected_by_group, [source.id, destination.id], now - timedelta(seconds=rollup_duration), time_from_now(17), rollup_duration, ) environment_time_series = tsdb.get_distinct_counts_series( tsdb.models.users_affected_by_group, [source.id, destination.id], now - timedelta(seconds=rollup_duration), time_from_now(17), rollup_duration, environment_id=production_environment.id, ) def collect_by_user_tag(aggregate, event): aggregate = aggregate if aggregate is not None else set() aggregate.add(get_event_user_from_interface(event.data["user"]).tag_value) return aggregate for series in [time_series, environment_time_series]: assert_series_contains( { timestamp: len(values) for timestamp, values in get_expected_series_values( rollup_duration, list(events.values())[1], collect_by_user_tag ).items() }, series[source.id], ) assert_series_contains( { timestamp: len(values) for timestamp, values in get_expected_series_values( rollup_duration, list(events.values())[0] + list(events.values())[2], collect_by_user_tag, ).items() }, time_series[destination.id], ) def strip_zeroes(data): for group_id, series in data.items(): for _, values in series: for key, val in list(values.items()): if val == 0: values.pop(key) return data def collect_by_release(group, aggregate, event): aggregate = aggregate if aggregate is not None else {} release = event.get_tag("sentry:release") if not release: return aggregate release = GroupRelease.objects.get( group_id=group.id, environment=event.data["environment"], release_id=Release.objects.get( organization_id=project.organization_id, version=release ).id, ).id aggregate[release] = aggregate.get(release, 0) + 1 return aggregate items = {} for i in [source.id, destination.id]: items[i] = list(GroupRelease.objects.filter(group_id=i).values_list("id", flat=True)) time_series = strip_zeroes( tsdb.get_frequency_series( tsdb.models.frequent_releases_by_group, items, now - timedelta(seconds=rollup_duration), time_from_now(17), rollup_duration, ) ) assert_series_contains( get_expected_series_values( rollup_duration, list(events.values())[1], functools.partial(collect_by_release, source), ), time_series[source.id], {}, ) assert_series_contains( get_expected_series_values( rollup_duration, list(events.values())[0] + list(events.values())[2], functools.partial(collect_by_release, destination), ), time_series[destination.id], {}, ) items = {} for i in [source.id, destination.id]: items[i] = list(Environment.objects.all().values_list("id", flat=True)) time_series = strip_zeroes( tsdb.get_frequency_series( tsdb.models.frequent_environments_by_group, items, now - timedelta(seconds=rollup_duration), time_from_now(17), rollup_duration, ) ) def collect_by_environment(aggregate, event): aggregate = aggregate if aggregate is not None else {} environment = Environment.objects.get( organization_id=project.organization_id, name=event.data["environment"] ).id aggregate[environment] = aggregate.get(environment, 0) + 1 return aggregate assert_series_contains( get_expected_series_values( rollup_duration, list(events.values())[1], collect_by_environment ), time_series[source.id], {}, ) assert_series_contains( get_expected_series_values( rollup_duration, list(events.values())[0] + list(events.values())[2], collect_by_environment, ), time_series[destination.id], {}, ) source_similar_items = features.compare(source) assert source_similar_items[0] == ( source.id, { "exception:message:character-shingles": None, "exception:stacktrace:application-chunks": None, "exception:stacktrace:pairs": None, "message:message:character-shingles": 1.0, }, ) assert source_similar_items[1][0] == destination.id assert source_similar_items[1][1]["message:message:character-shingles"] < 1.0 destination_similar_items = features.compare(destination) assert destination_similar_items[0] == ( destination.id, { "exception:message:character-shingles": None, "exception:stacktrace:application-chunks": None, "exception:stacktrace:pairs": None, "message:message:character-shingles": 1.0, }, ) assert destination_similar_items[1][0] == source.id assert destination_similar_items[1][1]["message:message:character-shingles"] < 1.0
def unmerge( project_id, source_id, destination_id, fingerprints, actor_id, cursor=None, batch_size=500, source_fields_reset=False, eventstream_state=None, ): # XXX: The queryset chunking logic below is awfully similar to # ``RangeQuerySetWrapper``. Ideally that could be refactored to be able to # be run without iteration by passing around a state object and we could # just use that here instead. source = Group.objects.get(project_id=project_id, id=source_id) # On the first iteration of this loop, we clear out all of the # denormalizations from the source group so that we can have a clean slate # for the new, repaired data. if cursor is None: fingerprints = lock_hashes(project_id, source_id, fingerprints) truncate_denormalizations(source) caches = get_caches() project = caches["Project"](project_id) # We fetch the events in descending order by their primary key to get the # best approximation of the most recently received events. queryset = Event.objects.filter(project_id=project_id, group_id=source_id).order_by("-id") if cursor is not None: queryset = queryset.filter(id__lt=cursor) events = list(queryset[:batch_size]) # If there are no more events to process, we're done with the migration. if not events: tagstore.update_group_tag_key_values_seen(project_id, [source_id, destination_id]) unlock_hashes(project_id, fingerprints) logger.warning("Unmerge complete (eventstream state: %s)", eventstream_state) if eventstream_state: eventstream.end_unmerge(eventstream_state) return destination_id Event.objects.bind_nodes(events, "data") source_events = [] destination_events = [] for event in events: (destination_events if get_fingerprint(event) in fingerprints else source_events).append(event) if source_events: if not source_fields_reset: source.update( **get_group_creation_attributes(caches, source_events)) source_fields_reset = True else: source.update( **get_group_backfill_attributes(caches, source, source_events)) (destination_id, eventstream_state) = migrate_events( caches, project, source_id, destination_id, fingerprints, destination_events, actor_id, eventstream_state, ) repair_denormalizations(caches, project, events) unmerge.delay( project_id, source_id, destination_id, fingerprints, actor_id, cursor=events[-1].id, batch_size=batch_size, source_fields_reset=source_fields_reset, eventstream_state=eventstream_state, )
def unmerge( project_id, source_id, destination_id, fingerprints, actor_id, last_event=None, batch_size=500, source_fields_reset=False, eventstream_state=None, ): source = Group.objects.get(project_id=project_id, id=source_id) caches = get_caches() project = caches["Project"](project_id) # On the first iteration of this loop, we clear out all of the # denormalizations from the source group so that we can have a clean slate # for the new, repaired data. if last_event is None: fingerprints = lock_hashes(project_id, source_id, fingerprints) truncate_denormalizations(project, source) last_event, events = celery_run_batch_query( filter=eventstore.Filter(project_ids=[project_id], group_ids=[source.id]), batch_size=batch_size, state=last_event, referrer="unmerge", ) # If there are no more events to process, we're done with the migration. if not events: unlock_hashes(project_id, fingerprints) logger.warning("Unmerge complete (eventstream state: %s)", eventstream_state) if eventstream_state: eventstream.end_unmerge(eventstream_state) return destination_id source_events = [] destination_events = [] for event in events: (destination_events if get_fingerprint(event) in fingerprints else source_events).append(event) if source_events: if not source_fields_reset: source.update( **get_group_creation_attributes(caches, source_events)) source_fields_reset = True else: source.update( **get_group_backfill_attributes(caches, source, source_events)) (destination_id, eventstream_state) = migrate_events( caches, project, source_id, destination_id, fingerprints, destination_events, actor_id, eventstream_state, ) repair_denormalizations(caches, project, events) unmerge.delay( project_id, source_id, destination_id, fingerprints, actor_id, last_event=last_event, batch_size=batch_size, source_fields_reset=source_fields_reset, eventstream_state=eventstream_state, )
def unmerge( project_id, source_id, destination_id, fingerprints, actor_id, last_event=None, batch_size=500, source_fields_reset=False, eventstream_state=None, ): # XXX: The queryset chunking logic below is awfully similar to # ``RangeQuerySetWrapper``. Ideally that could be refactored to be able to # be run without iteration by passing around a state object and we could # just use that here instead. source = Group.objects.get(project_id=project_id, id=source_id) # On the first iteration of this loop, we clear out all of the # denormalizations from the source group so that we can have a clean slate # for the new, repaired data. if last_event is None: fingerprints = lock_hashes(project_id, source_id, fingerprints) truncate_denormalizations(source) caches = get_caches() project = caches["Project"](project_id) # We process events sorted in descending order by -timestamp, -event_id. We need # to include event_id as well as timestamp in the ordering criteria since: # # - Event timestamps are rounded to the second so multiple events are likely # to have the same timestamp. # # - When sorting by timestamp alone, Snuba may not give us a deterministic # order for events with the same timestamp. # # - We need to ensure that we do not skip any events between batches. If we # only sorted by timestamp < last_event.timestamp it would be possible to # have missed an event with the same timestamp as the last item in the # previous batch. conditions = [] if last_event is not None: conditions.extend( [ ["timestamp", "<=", last_event["timestamp"]], [ ["timestamp", "<", last_event["timestamp"]], ["event_id", "<", last_event["event_id"]], ], ] ) events = eventstore.get_events( filter_keys={"project_id": [project_id], "issue": [source.id]}, # We need the text-only "search message" from Snuba, not the raw message # dict field from nodestore. additional_columns=[eventstore.Columns.MESSAGE], conditions=conditions, limit=batch_size, referrer="unmerge", orderby=["-timestamp", "-event_id"], ) # If there are no more events to process, we're done with the migration. if not events: tagstore.update_group_tag_key_values_seen(project_id, [source_id, destination_id]) unlock_hashes(project_id, fingerprints) logger.warning("Unmerge complete (eventstream state: %s)", eventstream_state) if eventstream_state: eventstream.end_unmerge(eventstream_state) return destination_id Event.objects.bind_nodes(events, "data") source_events = [] destination_events = [] for event in events: (destination_events if get_fingerprint(event) in fingerprints else source_events).append( event ) if source_events: if not source_fields_reset: source.update(**get_group_creation_attributes(caches, source_events)) source_fields_reset = True else: source.update(**get_group_backfill_attributes(caches, source, source_events)) (destination_id, eventstream_state) = migrate_events( caches, project, source_id, destination_id, fingerprints, destination_events, actor_id, eventstream_state, ) repair_denormalizations(caches, project, events) unmerge.delay( project_id, source_id, destination_id, fingerprints, actor_id, last_event={"timestamp": events[-1].timestamp, "event_id": events[-1].event_id}, batch_size=batch_size, source_fields_reset=source_fields_reset, eventstream_state=eventstream_state, )
def unmerge( project_id, source_id, destination_id, fingerprints, actor_id, cursor=None, batch_size=500, source_fields_reset=False, eventstream_state=None, ): # XXX: The queryset chunking logic below is awfully similar to # ``RangeQuerySetWrapper``. Ideally that could be refactored to be able to # be run without iteration by passing around a state object and we could # just use that here instead. source = Group.objects.get( project_id=project_id, id=source_id, ) # On the first iteration of this loop, we clear out all of the # denormalizations from the source group so that we can have a clean slate # for the new, repaired data. if cursor is None: fingerprints = lock_hashes(project_id, source_id, fingerprints) truncate_denormalizations(source) caches = get_caches() project = caches['Project'](project_id) # We fetch the events in descending order by their primary key to get the # best approximation of the most recently received events. queryset = Event.objects.filter( project_id=project_id, group_id=source_id, ).order_by('-id') if cursor is not None: queryset = queryset.filter(id__lt=cursor) events = list(queryset[:batch_size]) # If there are no more events to process, we're done with the migration. if not events: tagstore.update_group_tag_key_values_seen(project_id, [source_id, destination_id]) unlock_hashes(project_id, fingerprints) logger.warning('Unmerge complete (eventstream state: %s)', eventstream_state) if eventstream_state: eventstream.end_unmerge(eventstream_state) return destination_id Event.objects.bind_nodes(events, 'data') source_events = [] destination_events = [] for event in events: (destination_events if get_fingerprint(event) in fingerprints else source_events).append(event) if source_events: if not source_fields_reset: source.update(**get_group_creation_attributes( caches, source_events, )) source_fields_reset = True else: source.update(**get_group_backfill_attributes( caches, source, source_events, )) (destination_id, eventstream_state) = migrate_events( caches, project, source_id, destination_id, fingerprints, destination_events, actor_id, eventstream_state, ) repair_denormalizations( caches, project, events, ) unmerge.delay( project_id, source_id, destination_id, fingerprints, actor_id, cursor=events[-1].id, batch_size=batch_size, source_fields_reset=source_fields_reset, eventstream_state=eventstream_state, )
def stop_snuba_replacement(self, eventstream_state: EventstreamState) -> None: eventstream.end_unmerge(eventstream_state)