def test_aggregator(self):
# do not raise with multiple valid rate limits
rate_limit_params_outer = RateLimitParameters("foo", "bar", None, 5)
rate_limit_params_inner = RateLimitParameters("foo", "bar", None, 5)
with RateLimitAggregator(
[rate_limit_params_outer, rate_limit_params_inner]):
pass
# raise when the inner rate limit should fail
rate_limit_params_outer = RateLimitParameters("foo", "bar", None, 0)
rate_limit_params_inner = RateLimitParameters("foo", "bar", None, 5)
with pytest.raises(RateLimitExceeded):
with RateLimitAggregator(
[rate_limit_params_outer, rate_limit_params_inner]):
pass
# raise when the outer rate limit should fail
rate_limit_params_outer = RateLimitParameters("foo", "bar", None, 5)
rate_limit_params_inner = RateLimitParameters("foo", "bar", None, 0)
with pytest.raises(RateLimitExceeded):
with RateLimitAggregator(
[rate_limit_params_outer, rate_limit_params_inner]):
pass
def execute_query_with_rate_limits(
clickhouse_query: Union[Query, CompositeQuery[Table]],
request_settings: RequestSettings,
formatted_query: FormattedQuery,
reader: Reader,
timer: Timer,
stats: MutableMapping[str, Any],
query_settings: MutableMapping[str, Any],
) -> Result:
# XXX: We should consider moving this that it applies to the logical query,
# not the physical query.
with RateLimitAggregator(request_settings.get_rate_limit_params()
) as rate_limit_stats_container:
stats.update(rate_limit_stats_container.to_dict())
timer.mark("rate_limit")
project_rate_limit_stats = rate_limit_stats_container.get_stats(
PROJECT_RATE_LIMIT_NAME)
if ("max_threads" in query_settings
and project_rate_limit_stats is not None
and project_rate_limit_stats.concurrent > 1):
maxt = query_settings["max_threads"]
query_settings["max_threads"] = max(
1, maxt - project_rate_limit_stats.concurrent + 1)
return execute_query(
clickhouse_query,
request_settings,
formatted_query,
reader,
timer,
stats,
query_settings,
)
def execute_query_with_rate_limits(
clickhouse_query: Union[Query, CompositeQuery[Table]],
query_settings: QuerySettings,
formatted_query: FormattedQuery,
reader: Reader,
timer: Timer,
stats: MutableMapping[str, Any],
clickhouse_query_settings: MutableMapping[str, Any],
robust: bool,
) -> Result:
# Global rate limiter is added at the end of the chain to be
# the last for evaluation.
# This allows us not to borrow capacity from the global quota
# during the evaluation if one of the more specific limiters
# (like the project rate limiter) rejects the query first.
query_settings.add_rate_limit(get_global_rate_limit_params())
# XXX: We should consider moving this that it applies to the logical query,
# not the physical query.
with RateLimitAggregator(
query_settings.get_rate_limit_params()
) as rate_limit_stats_container:
stats.update(rate_limit_stats_container.to_dict())
timer.mark("rate_limit")
project_rate_limit_stats = rate_limit_stats_container.get_stats(
PROJECT_RATE_LIMIT_NAME
)
thread_quota = query_settings.get_resource_quota()
if (
("max_threads" in clickhouse_query_settings or thread_quota is not None)
and project_rate_limit_stats is not None
and project_rate_limit_stats.concurrent > 1
):
maxt = (
clickhouse_query_settings["max_threads"]
if thread_quota is None
else thread_quota.max_threads
)
clickhouse_query_settings["max_threads"] = max(
1, maxt - project_rate_limit_stats.concurrent + 1
)
_record_rate_limit_metrics(rate_limit_stats_container, reader, stats)
return execute_query(
clickhouse_query,
query_settings,
formatted_query,
reader,
timer,
stats,
clickhouse_query_settings,
robust=robust,
)
def test_concurrent_limit(self):
# No concurrent limit should not raise
rate_limit_params = RateLimitParameters("foo", "bar", None, None)
with rate_limit(rate_limit_params) as stats:
assert stats is not None
# 0 concurrent limit
rate_limit_params = RateLimitParameters("foo", "bar", None, 0)
with pytest.raises(RateLimitExceeded):
with rate_limit(rate_limit_params):
pass
# Concurrent limit 1 with consecutive queries should not raise
rate_limit_params = RateLimitParameters("foo", "bar", None, 1)
with rate_limit(rate_limit_params):
pass
with rate_limit(rate_limit_params):
pass
# Concurrent limit with concurrent queries
rate_limit_params = RateLimitParameters("foo", "bar", None, 1)
with pytest.raises(RateLimitExceeded):
with rate_limit(rate_limit_params):
with rate_limit(rate_limit_params):
pass
# Concurrent with different buckets should not raise
rate_limit_params1 = RateLimitParameters("foo", "bar", None, 1)
rate_limit_params2 = RateLimitParameters("shoe", "star", None, 1)
with RateLimitAggregator([rate_limit_params1]):
with RateLimitAggregator([rate_limit_params2]):
pass
def test_concurrent_limit(self):
# No concurrent limit should not raise
rate_limit_params = RateLimitParameters('foo', 'bar', None, None)
with rate_limit(rate_limit_params):
pass
# 0 concurrent limit
rate_limit_params = RateLimitParameters('foo', 'bar', None, 0)
with pytest.raises(RateLimitExceeded):
with rate_limit(rate_limit_params):
pass
# Concurrent limit 1 with consecutive queries should not raise
rate_limit_params = RateLimitParameters('foo', 'bar', None, 1)
with rate_limit(rate_limit_params):
pass
with rate_limit(rate_limit_params):
pass
# Concurrent limit with concurrent queries
rate_limit_params = RateLimitParameters('foo', 'bar', None, 1)
with pytest.raises(RateLimitExceeded):
with rate_limit(rate_limit_params):
with rate_limit(rate_limit_params):
pass
# Concurrent with different buckets should not raise
rate_limit_params1 = RateLimitParameters('foo', 'bar', None, 1)
rate_limit_params2 = RateLimitParameters('shoe', 'star', None, 1)
with RateLimitAggregator([rate_limit_params1]):
with RateLimitAggregator([rate_limit_params2]):
pass
def raw_query(
request: Request,
query: DictClickhouseQuery,
reader: Reader[ClickhouseQuery],
timer: Timer,
stats: Optional[MutableMapping[str, Any]] = None,
) -> ClickhouseQueryResult:
"""
Submit a raw SQL query to clickhouse and do some post-processing on it to
fix some of the formatting issues in the result JSON
"""
stats = stats or {}
use_cache, use_deduper, uc_max = state.get_configs(
[("use_cache", 0), ("use_deduper", 1), ("uncompressed_cache_max_cols", 5)]
)
all_confs = state.get_all_configs()
query_settings = {
k.split("/", 1)[1]: v
for k, v in all_confs.items()
if k.startswith("query_settings/")
}
# Experiment, if we are going to grab more than X columns worth of data,
# don't use uncompressed_cache in clickhouse, or result cache in snuba.
if len(request.query.get_all_referenced_columns()) > uc_max:
query_settings["use_uncompressed_cache"] = 0
use_cache = 0
timer.mark("get_configs")
sql = query.format_sql()
query_id = md5(force_bytes(sql)).hexdigest()
with state.deduper(query_id if use_deduper else None) as is_dupe:
timer.mark("dedupe_wait")
result = cache.get(query_id) if use_cache else None
timer.mark("cache_get")
stats.update(
{
"is_duplicate": is_dupe,
"query_id": query_id,
"use_cache": bool(use_cache),
"cache_hit": bool(result),
}
),
if not result:
try:
with RateLimitAggregator(
request.settings.get_rate_limit_params()
) as rate_limit_stats_container:
stats.update(rate_limit_stats_container.to_dict())
timer.mark("rate_limit")
project_rate_limit_stats = rate_limit_stats_container.get_stats(
PROJECT_RATE_LIMIT_NAME
)
if (
"max_threads" in query_settings
and project_rate_limit_stats is not None
and project_rate_limit_stats.concurrent > 1
):
maxt = query_settings["max_threads"]
query_settings["max_threads"] = max(
1, maxt - project_rate_limit_stats.concurrent + 1
)
# Force query to use the first shard replica, which
# should have synchronously received any cluster writes
# before this query is run.
consistent = request.settings.get_consistent()
stats["consistent"] = consistent
if consistent:
query_settings["load_balancing"] = "in_order"
query_settings["max_threads"] = 1
try:
result = reader.execute(
query,
query_settings,
# All queries should already be deduplicated at this point
# But the query_id will let us know if they aren't
query_id=query_id if use_deduper else None,
with_totals=request.query.has_totals(),
)
timer.mark("execute")
stats.update(
{
"result_rows": len(result["data"]),
"result_cols": len(result["meta"]),
}
)
if use_cache:
cache.set(query_id, result)
timer.mark("cache_set")
except BaseException as ex:
error = str(ex)
logger.exception("Error running query: %s\n%s", sql, error)
stats = log_query_and_update_stats(
request, sql, timer, stats, "error", query_settings
)
meta = {}
if isinstance(ex, ClickHouseError):
err_type = "clickhouse"
meta["code"] = ex.code
else:
err_type = "unknown"
raise RawQueryException(
err_type=err_type,
message=error,
stats=stats,
sql=sql,
**meta,
)
except RateLimitExceeded as ex:
stats = log_query_and_update_stats(
request, sql, timer, stats, "rate-limited", query_settings
)
raise RawQueryException(
err_type="rate-limited",
message="rate limit exceeded",
stats=stats,
sql=sql,
detail=str(ex),
)
stats = log_query_and_update_stats(
request, sql, timer, stats, "success", query_settings
)
if settings.STATS_IN_RESPONSE or request.settings.get_debug():
result["stats"] = stats
result["sql"] = sql
return result
def raw_query(
request: Request,
query: ClickhouseQuery,
client: ClickhousePool,
timer: Timer,
stats=None,
) -> QueryResult:
"""
Submit a raw SQL query to clickhouse and do some post-processing on it to
fix some of the formatting issues in the result JSON
"""
from snuba.clickhouse.native import NativeDriverReader
stats = stats or {}
use_cache, use_deduper, uc_max = state.get_configs([
('use_cache', 0),
('use_deduper', 1),
('uncompressed_cache_max_cols', 5),
])
all_confs = state.get_all_configs()
query_settings = {
k.split('/', 1)[1]: v
for k, v in all_confs.items() if k.startswith('query_settings/')
}
# Experiment, if we are going to grab more than X columns worth of data,
# don't use uncompressed_cache in clickhouse, or result cache in snuba.
if len(all_referenced_columns(request.query)) > uc_max:
query_settings['use_uncompressed_cache'] = 0
use_cache = 0
timer.mark('get_configs')
sql = query.format_sql()
query_id = md5(force_bytes(sql)).hexdigest()
with state.deduper(query_id if use_deduper else None) as is_dupe:
timer.mark('dedupe_wait')
result = state.get_result(query_id) if use_cache else None
timer.mark('cache_get')
stats.update({
'is_duplicate': is_dupe,
'query_id': query_id,
'use_cache': bool(use_cache),
'cache_hit': bool(result)
}),
if result:
status = 200
else:
try:
with RateLimitAggregator(
request.settings.get_rate_limit_params(
)) as rate_limit_stats_container:
stats.update(rate_limit_stats_container.to_dict())
timer.mark('rate_limit')
project_rate_limit_stats = rate_limit_stats_container.get_stats(
PROJECT_RATE_LIMIT_NAME)
if 'max_threads' in query_settings and \
project_rate_limit_stats is not None and \
project_rate_limit_stats.concurrent > 1:
maxt = query_settings['max_threads']
query_settings['max_threads'] = max(
1, maxt - project_rate_limit_stats.concurrent + 1)
# Force query to use the first shard replica, which
# should have synchronously received any cluster writes
# before this query is run.
consistent = request.settings.get_consistent()
stats['consistent'] = consistent
if consistent:
query_settings['load_balancing'] = 'in_order'
query_settings['max_threads'] = 1
try:
result = NativeDriverReader(client).execute(
query,
query_settings,
# All queries should already be deduplicated at this point
# But the query_id will let us know if they aren't
query_id=query_id if use_deduper else None,
with_totals=request.query.has_totals(),
)
status = 200
logger.debug(sql)
timer.mark('execute')
stats.update({
'result_rows': len(result['data']),
'result_cols': len(result['meta']),
})
if use_cache:
state.set_result(query_id, result)
timer.mark('cache_set')
except BaseException as ex:
error = str(ex)
status = 500
logger.exception("Error running query: %s\n%s", sql,
error)
if isinstance(ex, ClickHouseError):
result = {
'error': {
'type': 'clickhouse',
'code': ex.code,
'message': error,
}
}
else:
result = {
'error': {
'type': 'unknown',
'message': error,
}
}
except RateLimitExceeded as ex:
error = str(ex)
status = 429
result = {
'error': {
'type': 'ratelimit',
'message': 'rate limit exceeded',
'detail': error
}
}
stats.update(query_settings)
if settings.RECORD_QUERIES:
# send to redis
state.record_query({
'request': request.body,
'sql': sql,
'timing': timer,
'stats': stats,
'status': status,
})
timer.send_metrics_to(metrics,
tags={
'status': str(status),
'referrer': stats.get('referrer', 'none'),
'final': str(stats.get('final', False)),
},
mark_tags={
'final': str(stats.get('final', False)),
})
result['timing'] = timer
if settings.STATS_IN_RESPONSE or request.settings.get_debug():
result['stats'] = stats
result['sql'] = sql
return QueryResult(result, status)
def raw_query(
request: Request,
query: ClickhouseQuery,
timer: Timer,
query_metadata: SnubaQueryMetadata,
stats: MutableMapping[str, Any],
trace_id: Optional[str] = None,
) -> RawQueryResult:
"""
Submits a raw SQL query to the DB and does some post-processing on it to
fix some of the formatting issues in the result JSON.
This function is not supposed to depend on anything higher level than the storage
query (ClickhouseQuery as of now). If this function ends up depending on the
dataset, something is wrong.
TODO: As soon as we have a StorageQuery abstraction remove all the references
to the original query from the request.
"""
use_cache, use_deduper, uc_max = state.get_configs([
("use_cache", settings.USE_RESULT_CACHE),
("use_deduper", 1),
("uncompressed_cache_max_cols", 5),
])
all_confs = state.get_all_configs()
query_settings: MutableMapping[str, Any] = {
k.split("/", 1)[1]: v
for k, v in all_confs.items() if k.startswith("query_settings/")
}
# Experiment, if we are going to grab more than X columns worth of data,
# don't use uncompressed_cache in clickhouse, or result cache in snuba.
if len(request.query.get_all_referenced_columns()) > uc_max:
query_settings["use_uncompressed_cache"] = 0
use_cache = 0
timer.mark("get_configs")
sql = query.format_sql()
query_id = md5(force_bytes(sql)).hexdigest()
with state.deduper(query_id if use_deduper else None) as is_dupe:
timer.mark("dedupe_wait")
result = cache.get(query_id) if use_cache else None
timer.mark("cache_get")
stats.update({
"is_duplicate": is_dupe,
"query_id": query_id,
"use_cache": bool(use_cache),
"cache_hit": bool(result),
}),
update_with_status = partial(
update_query_metadata_and_stats,
request,
sql,
timer,
stats,
query_metadata,
query_settings,
trace_id,
)
if not result:
try:
with RateLimitAggregator(
request.settings.get_rate_limit_params(
)) as rate_limit_stats_container:
stats.update(rate_limit_stats_container.to_dict())
timer.mark("rate_limit")
project_rate_limit_stats = rate_limit_stats_container.get_stats(
PROJECT_RATE_LIMIT_NAME)
if ("max_threads" in query_settings
and project_rate_limit_stats is not None
and project_rate_limit_stats.concurrent > 1):
maxt = query_settings["max_threads"]
query_settings["max_threads"] = max(
1, maxt - project_rate_limit_stats.concurrent + 1)
# Force query to use the first shard replica, which
# should have synchronously received any cluster writes
# before this query is run.
consistent = request.settings.get_consistent()
stats["consistent"] = consistent
if consistent:
query_settings["load_balancing"] = "in_order"
query_settings["max_threads"] = 1
try:
result = reader.execute(
query,
query_settings,
# All queries should already be deduplicated at this point
# But the query_id will let us know if they aren't
query_id=query_id if use_deduper else None,
with_totals=request.query.has_totals(),
)
timer.mark("execute")
stats.update({
"result_rows": len(result["data"]),
"result_cols": len(result["meta"]),
})
if use_cache:
cache.set(query_id, result)
timer.mark("cache_set")
except BaseException as ex:
error = str(ex)
logger.exception("Error running query: %s\n%s", sql,
error)
stats = update_with_status("error")
meta = {}
if isinstance(ex, ClickhouseError):
err_type = "clickhouse"
meta["code"] = ex.code
else:
err_type = "unknown"
raise RawQueryException(
err_type=err_type,
message=error,
stats=stats,
sql=sql,
**meta,
)
except RateLimitExceeded as ex:
stats = update_with_status("rate-limited")
raise RawQueryException(
err_type="rate-limited",
message="rate limit exceeded",
stats=stats,
sql=sql,
detail=str(ex),
)
stats = update_with_status("success")
return RawQueryResult(result, {"stats": stats, "sql": sql})