def do_post_processing(
self,
project_ids: Sequence[int],
query: Query,
request_settings: RequestSettings,
) -> None:
if not request_settings.get_turbo():
final, exclude_group_ids = get_projects_query_flags(
project_ids, self.__replacer_state_name)
if not final and exclude_group_ids:
# If the number of groups to exclude exceeds our limit, the query
# should just use final instead of the exclusion set.
max_group_ids_exclude = get_config(
"max_group_ids_exclude",
settings.REPLACER_MAX_GROUP_IDS_TO_EXCLUDE)
if len(exclude_group_ids) > max_group_ids_exclude:
query.set_final(True)
else:
query.add_conditions([(["assumeNotNull", ["group_id"]],
"NOT IN", exclude_group_ids)])
query.add_condition_to_ast(
not_in_condition(
None,
FunctionCall(None, "assumeNotNull",
(Column(None, "group_id", None), )),
[Literal(None, p) for p in exclude_group_ids],
))
else:
query.set_final(final)
def _apply_turbo_sampling_if_needed(
clickhouse_query: Union[Query, CompositeQuery[Table]],
request_settings: RequestSettings,
) -> None:
"""
TODO: Remove this method entirely and move the sampling logic
into a query processor.
"""
if isinstance(clickhouse_query, Query):
if (request_settings.get_turbo()
and not clickhouse_query.get_from_clause().sampling_rate):
clickhouse_query.set_from_clause(
replace(
clickhouse_query.get_from_clause(),
sampling_rate=snuba_settings.TURBO_SAMPLE_RATE,
))
def process_query(self, query: Query,
request_settings: RequestSettings) -> None:
if request_settings.get_turbo():
return
project_ids = get_project_ids_in_query_ast(query,
self.__project_column)
set_final = False
condition_to_add = None
if project_ids:
final, exclude_group_ids = get_projects_query_flags(
list(project_ids),
self.__replacer_state_name,
)
if final:
metrics.increment("final", tags={"cause": "final_flag"})
if not final and exclude_group_ids:
# If the number of groups to exclude exceeds our limit, the query
# should just use final instead of the exclusion set.
max_group_ids_exclude = get_config(
"max_group_ids_exclude",
settings.REPLACER_MAX_GROUP_IDS_TO_EXCLUDE)
if len(exclude_group_ids) > max_group_ids_exclude:
metrics.increment("final", tags={"cause": "max_groups"})
set_final = True
else:
condition_to_add = (
["assumeNotNull", ["group_id"]],
"NOT IN",
exclude_group_ids,
)
query.add_condition_to_ast(
not_in_condition(
None,
FunctionCall(None, "assumeNotNull",
(Column(None, None, "group_id"), )),
[Literal(None, p) for p in exclude_group_ids],
))
else:
set_final = final
query.set_final(set_final)
if condition_to_add:
query.add_conditions([condition_to_add])
def do_post_processing(
self,
project_ids: Sequence[int],
query: Query,
request_settings: RequestSettings,
) -> None:
if not request_settings.get_turbo():
final, exclude_group_ids = get_projects_query_flags(project_ids)
if not final and exclude_group_ids:
# If the number of groups to exclude exceeds our limit, the query
# should just use final instead of the exclusion set.
max_group_ids_exclude = get_config('max_group_ids_exclude', settings.REPLACER_MAX_GROUP_IDS_TO_EXCLUDE)
if len(exclude_group_ids) > max_group_ids_exclude:
query.set_final(True)
else:
query.add_conditions([(['assumeNotNull', ['group_id']], 'NOT IN', exclude_group_ids)])
else:
query.set_final(final)
def format_query(query: FormattableQuery,
settings: RequestSettings) -> FormattedQuery:
"""
Formats a Clickhouse Query from the AST representation into an
intermediate structure that can either be serialized into a string
(for clickhouse) or extracted as a sequence (for logging and tracing).
This is the entry point for any type of query, whether simple or
composite.
TODO: Remove this method entirely and move the sampling logic
into a query processor.
"""
if isinstance(query, Query):
if settings.get_turbo() and not query.get_from_clause().sampling_rate:
query.set_from_clause(
replace(
query.get_from_clause(),
sampling_rate=snuba_settings.TURBO_SAMPLE_RATE,
))
return FormattedQuery(_format_query_content(query))
def __init__(
self,
dataset: Dataset,
query: Query,
settings: RequestSettings,
) -> None:
parsing_context = ParsingContext()
aggregate_exprs = [
column_expr(dataset, col, query, parsing_context, alias, agg)
for (agg, col, alias) in query.get_aggregations()
]
groupby = util.to_list(query.get_groupby())
group_exprs = [
column_expr(dataset, gb, query, parsing_context) for gb in groupby
]
column_names = query.get_selected_columns() or []
selected_cols = [
column_expr(dataset, util.tuplify(colname), query, parsing_context)
for colname in column_names
]
select_clause = u"SELECT {}".format(
", ".join(group_exprs + aggregate_exprs + selected_cols))
from_clause = u"FROM {}".format(query.get_data_source().format_from())
if query.get_final():
from_clause = u"{} FINAL".format(from_clause)
if not query.get_data_source().supports_sample():
sample_rate = None
else:
if query.get_sample():
sample_rate = query.get_sample()
elif settings.get_turbo():
sample_rate = snuba_settings.TURBO_SAMPLE_RATE
else:
sample_rate = None
if sample_rate:
from_clause = u"{} SAMPLE {}".format(from_clause, sample_rate)
join_clause = ""
if query.get_arrayjoin():
join_clause = u"ARRAY JOIN {}".format(query.get_arrayjoin())
where_clause = ""
if query.get_conditions():
where_clause = u"WHERE {}".format(
conditions_expr(dataset, query.get_conditions(), query,
parsing_context))
prewhere_clause = ""
if query.get_prewhere():
prewhere_clause = u"PREWHERE {}".format(
conditions_expr(dataset, query.get_prewhere(), query,
parsing_context))
group_clause = ""
if groupby:
group_clause = "GROUP BY ({})".format(", ".join(
column_expr(dataset, gb, query, parsing_context)
for gb in groupby))
if query.has_totals():
group_clause = "{} WITH TOTALS".format(group_clause)
having_clause = ""
having_conditions = query.get_having()
if having_conditions:
assert groupby, "found HAVING clause with no GROUP BY"
having_clause = u"HAVING {}".format(
conditions_expr(dataset, having_conditions, query,
parsing_context))
order_clause = ""
if query.get_orderby():
orderby = [
column_expr(dataset, util.tuplify(ob), query, parsing_context)
for ob in util.to_list(query.get_orderby())
]
orderby = [
u"{} {}".format(ob.lstrip("-"),
"DESC" if ob.startswith("-") else "ASC")
for ob in orderby
]
order_clause = u"ORDER BY {}".format(", ".join(orderby))
limitby_clause = ""
if query.get_limitby() is not None:
limitby_clause = "LIMIT {} BY {}".format(*query.get_limitby())
limit_clause = ""
if query.get_limit() is not None:
limit_clause = "LIMIT {}, {}".format(query.get_offset(),
query.get_limit())
self.__formatted_query = " ".join([
c for c in [
select_clause,
from_clause,
join_clause,
prewhere_clause,
where_clause,
group_clause,
having_clause,
order_clause,
limitby_clause,
limit_clause,
] if c
])
def __init__(
self,
dataset: Dataset,
query: Query,
settings: RequestSettings,
prewhere_conditions: Sequence[str],
) -> None:
parsing_context = ParsingContext()
aggregate_exprs = [
column_expr(dataset, col, query, parsing_context, alias, agg)
for (agg, col, alias) in query.get_aggregations()
]
groupby = util.to_list(query.get_groupby())
group_exprs = [
column_expr(dataset, gb, query, parsing_context) for gb in groupby
]
column_names = query.get_selected_columns() or []
selected_cols = [
column_expr(dataset, util.tuplify(colname), query, parsing_context)
for colname in column_names
]
select_clause = u'SELECT {}'.format(
', '.join(group_exprs + aggregate_exprs + selected_cols))
from_clause = u'FROM {}'.format(query.get_data_source().format_from())
if query.get_final():
from_clause = u'{} FINAL'.format(from_clause)
if query.get_sample():
sample_rate = query.get_sample()
elif settings.get_turbo():
sample_rate = snuba_settings.TURBO_SAMPLE_RATE
else:
sample_rate = None
if sample_rate:
from_clause = u'{} SAMPLE {}'.format(from_clause, sample_rate)
join_clause = ''
if query.get_arrayjoin():
join_clause = u'ARRAY JOIN {}'.format(query.get_arrayjoin())
where_clause = ''
if query.get_conditions():
where_clause = u'WHERE {}'.format(
conditions_expr(dataset, query.get_conditions(), query,
parsing_context))
prewhere_clause = ''
if prewhere_conditions:
prewhere_clause = u'PREWHERE {}'.format(
conditions_expr(dataset, prewhere_conditions, query,
parsing_context))
group_clause = ''
if groupby:
group_clause = 'GROUP BY ({})'.format(', '.join(
column_expr(dataset, gb, query, parsing_context)
for gb in groupby))
if query.has_totals():
group_clause = '{} WITH TOTALS'.format(group_clause)
having_clause = ''
having_conditions = query.get_having()
if having_conditions:
assert groupby, 'found HAVING clause with no GROUP BY'
having_clause = u'HAVING {}'.format(
conditions_expr(dataset, having_conditions, query,
parsing_context))
order_clause = ''
if query.get_orderby():
orderby = [
column_expr(dataset, util.tuplify(ob), query, parsing_context)
for ob in util.to_list(query.get_orderby())
]
orderby = [
u'{} {}'.format(ob.lstrip('-'),
'DESC' if ob.startswith('-') else 'ASC')
for ob in orderby
]
order_clause = u'ORDER BY {}'.format(', '.join(orderby))
limitby_clause = ''
if query.get_limitby() is not None:
limitby_clause = 'LIMIT {} BY {}'.format(*query.get_limitby())
limit_clause = ''
if query.get_limit() is not None:
limit_clause = 'LIMIT {}, {}'.format(query.get_offset(),
query.get_limit())
self.__formatted_query = ' '.join([
c for c in [
select_clause, from_clause, join_clause, prewhere_clause,
where_clause, group_clause, having_clause, order_clause,
limitby_clause, limit_clause
] if c
])
def callback_func(
storage: str,
query: Query,
request_settings: RequestSettings,
referrer: str,
results: List[Result[QueryResult]],
) -> None:
cache_hit = False
is_duplicate = False
# Captures if any of the queries involved was a cache hit or duplicate, as cache
# hits may a cause of inconsistency between results.
# Doesn't attempt to distinguish between all of the specific scenarios (one or both
# queries, or splits of those queries could have hit the cache).
if any([result.result.extra["stats"].get("cache_hit", 0) for result in results]):
cache_hit = True
elif any(
[result.result.extra["stats"].get("is_duplicate", 0) for result in results]
):
is_duplicate = True
consistent = request_settings.get_consistent()
if not results:
metrics.increment(
"query_result",
tags={"storage": storage, "match": "empty", "referrer": referrer},
)
return
primary_result = results.pop(0)
primary_result_data = primary_result.result.result["data"]
for result in results:
result_data = result.result.result["data"]
metrics.timing(
"diff_ms",
round((result.execution_time - primary_result.execution_time) * 1000),
tags={
"referrer": referrer,
"cache_hit": str(cache_hit),
"is_duplicate": str(is_duplicate),
"consistent": str(consistent),
},
)
# Do not bother diffing the actual results of sampled queries
if request_settings.get_turbo() or query.get_sample() not in [None, 1.0]:
return
if result_data == primary_result_data:
metrics.increment(
"query_result",
tags={
"storage": storage,
"match": "true",
"referrer": referrer,
"cache_hit": str(cache_hit),
"is_duplicate": str(is_duplicate),
"consistent": str(consistent),
},
)
else:
# Do not log cache hits to Sentry as it creates too much noise
if cache_hit:
continue
reason = assign_reason_category(result_data, primary_result_data, referrer)
metrics.increment(
"query_result",
tags={
"storage": storage,
"match": "false",
"referrer": referrer,
"reason": reason,
"cache_hit": str(cache_hit),
"is_duplicate": str(is_duplicate),
"consistent": str(consistent),
},
)
if len(result_data) != len(primary_result_data):
sentry_sdk.capture_message(
f"Non matching {storage} result - different length",
level="warning",
tags={
"referrer": referrer,
"storage": storage,
"reason": reason,
"cache_hit": str(cache_hit),
"is_duplicate": str(is_duplicate),
"consistent": str(consistent),
},
extras={
"query": format_query(query),
"primary_result": len(primary_result_data),
"other_result": len(result_data),
},
)
break
# Avoid sending too much data to Sentry - just one row for now
for idx in range(len(result_data)):
if result_data[idx] != primary_result_data[idx]:
sentry_sdk.capture_message(
"Non matching result - different result",
level="warning",
tags={
"referrer": referrer,
"storage": storage,
"reason": reason,
"cache_hit": str(cache_hit),
"is_duplicate": str(is_duplicate),
"consistent": str(consistent),
},
extras={
"query": format_query(query),
"primary_result": primary_result_data[idx],
"other_result": result_data[idx],
},
)
break