mirrors/scylladb
1
0
Fork 0
mirror of https://github.com/scylladb/scylladb.git synced 2026-05-01 21:55:50 +00:00
Code Issues Packages Projects Releases Wiki Activity

Merge '[Backport 5.4] cql3/statement/select_statement: do not parallelize single-partition aggregations #19414' from Michał Jadwiszczak

Browse Source 
This patch adds a check if aggregation query is doing single-partition read and if so, makes the query to not use forward_service and do not parallelize the request.

Fixes scylladb/scylladb#19349

(cherry picked from commit e9ace7c203)

(cherry picked from commit 8eb5ca8202)

Refs scylladb/scylladb#19350

Closes scylladb/scylladb#19500

* github.com:scylladb/scylladb:
  test/boost/cql_query_test: add test for single-partition aggregation
  cql3/select_statement: do not parallelize single-partition aggregations
This commit is contained in:
Piotr Dulikowski
2024-07-03 05:57:49 +02:00
parent f07fbcf929 bea1f4891d
commit 65daae0fbe
2 changed files with 72 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

68
test/boost/cql_query_test.cc
View File

@@ -26,6 +26,8 @@
#include <seastar/core/future-util.hh>
#include <seastar/core/sleep.hh>
#include "transport/messages/result_message.hh"
#include "transport/messages/result_message_base.hh"
#include "types/types.hh"
#include "utils/big_decimal.hh"
#include "types/user.hh"
#include "types/map.hh"
@@ -5361,6 +5363,72 @@ SEASTAR_TEST_CASE(test_parallelized_select_counter_type) {
});
}
SEASTAR_TEST_CASE(test_single_partition_aggregation_is_not_parallelized) {
// It's pointless from performance pov to parallelize
// aggregation queries which reads only single partition.
return with_parallelized_aggregation_enabled_thread([](cql_test_env& e) {
auto& qp = e.local_qp();
const auto stat_parallelized = qp.get_cql_stats().select_parallelized;
e.execute_cql("CREATE TABLE tbl (pk int, ck int, col int, PRIMARY KEY (pk, ck));").get();
const int value_count = 10;
for (int pk = 0; pk < 2; pk++) {
for (int c = 0; c < value_count; c++) {
e.execute_cql(format("INSERT INTO tbl (pk, ck, col) VALUES ({:d}, {:d}, {:d});", pk, c, c)).get();
}
}
const auto result1 = e.execute_cql("SELECT COUNT(*) FROM tbl WHERE pk = 1;").get();
assert_that(result1).is_rows().with_rows({
{long_type->decompose(int64_t(value_count))}
});
BOOST_CHECK_EQUAL(stat_parallelized, qp.get_cql_stats().select_parallelized);
const auto result2 = e.execute_cql("SELECT COUNT(*) FROM tbl WHERE pk = 1 AND ck = 1;").get();
assert_that(result2).is_rows().with_rows({
{long_type->decompose(int64_t(1))}
});
BOOST_CHECK_EQUAL(stat_parallelized, qp.get_cql_stats().select_parallelized);
const auto result3 = e.execute_cql("SELECT COUNT(*) FROM tbl WHERE token(pk) = 1;").get();
// We don't check value of count(*) here but only if it wasn't parallelized
BOOST_CHECK_EQUAL(stat_parallelized, qp.get_cql_stats().select_parallelized);
const auto result4 = e.execute_cql("SELECT COUNT(*) FROM tbl WHERE pk = 1 AND pk = 2;").get();
assert_that(result4).is_rows().with_rows({
{long_type->decompose(int64_t(0))}
});
BOOST_CHECK_EQUAL(stat_parallelized, qp.get_cql_stats().select_parallelized);
e.execute_cql("CREATE TABLE tbl2 (pk1 int, pk2 int, ck int, col int, PRIMARY KEY((pk1, pk2), ck));").get();
for (int pk1 = 0; pk1 < 2; pk1++) {
for (int pk2 = 0; pk2 < 2; pk2++) {
for (int c = 0; c < value_count; c++) {
e.execute_cql(format("INSERT INTO tbl2 (pk1, pk2, ck, col) VALUES ({:d}, {:d}, {:d}, {:d});", pk1, pk2, c, c)).get();
}
}
}
const auto result_pk12 = e.execute_cql("SELECT COUNT(*) FROM tbl2 WHERE pk1 = 1 AND pk2 = 0;").get();
assert_that(result_pk12).is_rows().with_rows({
{long_type->decompose(int64_t(value_count))}
});
BOOST_CHECK_EQUAL(stat_parallelized, qp.get_cql_stats().select_parallelized);
// Query with only partly restricted partition key requires `ALLOW FILTERING` clause
// and we doesn't parallelize queries which need filtering.
// See issue #19369.
const auto result_pk1 = e.execute_cql("SELECT COUNT(*) FROM tbl2 WHERE pk1 = 1 ALLOW FILTERING;").get();
// This query contains also column for pk1
assert_that(result_pk1).is_rows().with_rows({
{long_type->decompose(int64_t(value_count * 2)), int32_type->decompose(int32_t(1))}
});
BOOST_CHECK_EQUAL(stat_parallelized, qp.get_cql_stats().select_parallelized);
});
}
static future<> with_udf_and_parallel_aggregation_enabled_thread(std::function<void(cql_test_env&)>&& func) {
auto db_cfg_ptr = make_shared<db::config>();
auto& db_cfg = *db_cfg_ptr;