/* * Copyright (C) 2015-present ScyllaDB */ /* * SPDX-License-Identifier: AGPL-3.0-or-later */ #include #include #include #include "dht/i_partitioner.hh" #include "dht/sharder.hh" #include "dht/murmur3_partitioner.hh" #include "schema.hh" #include "types.hh" #include "schema_builder.hh" #include "utils/div_ceil.hh" #include "test/lib/simple_schema.hh" #include "test/lib/log.hh" #include "test/boost/total_order_check.hh" template static void debug(Args&&... args) { if (false) { print(std::forward(args)...); } } static dht::token token_from_long(uint64_t value) { return dht::token(dht::token::kind::key, value); } static int64_t long_from_token(dht::token token) { return token._data; } SEASTAR_THREAD_TEST_CASE(test_decorated_key_is_compatible_with_origin) { auto s = schema_builder("ks", "cf") .with_column("c1", int32_type, column_kind::partition_key) .with_column("c2", int32_type, column_kind::partition_key) .with_column("v", int32_type) .build(); dht::murmur3_partitioner partitioner; auto key = partition_key::from_deeply_exploded(*s, {143, 234}); auto dk = partitioner.decorate_key(*s, key); // Expected value was taken from Origin BOOST_REQUIRE_EQUAL(dk._token, token_from_long(4958784316840156970)); BOOST_REQUIRE(dk._key.equal(*s, key)); } SEASTAR_THREAD_TEST_CASE(test_token_wraparound_1) { auto t1 = token_from_long(0x7000'0000'0000'0000); auto t2 = token_from_long(0xa000'0000'0000'0000); dht::murmur3_partitioner partitioner; BOOST_REQUIRE(t1 > t2); // Even without knowing what the midpoint is, it needs to be inside the // wrapped range, i.e., between t1 and inf, OR between -inf and t2 auto midpoint = dht::token::midpoint(t1, t2); BOOST_REQUIRE(midpoint > t1 || midpoint < t2); // We can also calculate the actual value the midpoint should have: BOOST_REQUIRE_EQUAL(midpoint, token_from_long(0x8800'0000'0000'0000)); } SEASTAR_THREAD_TEST_CASE(test_token_wraparound_2) { auto t1 = token_from_long(0x6000'0000'0000'0000); auto t2 = token_from_long(0x9000'0000'0000'0000); dht::murmur3_partitioner partitioner; BOOST_REQUIRE(t1 > t2); auto midpoint = dht::token::midpoint(t1, t2); BOOST_REQUIRE(midpoint > t1 || midpoint < t2); BOOST_REQUIRE_EQUAL(midpoint, token_from_long(0x7800'0000'0000'0000)); } SEASTAR_THREAD_TEST_CASE(test_ring_position_is_comparable_with_decorated_key) { auto s = schema_builder("ks", "cf") .with_column("pk", bytes_type, column_kind::partition_key) .with_column("v", int32_type) .build(); std::vector keys = { dht::decorate_key(*s, partition_key::from_single_value(*s, "key1")), dht::decorate_key(*s, partition_key::from_single_value(*s, "key2")), }; std::sort(keys.begin(), keys.end(), dht::decorated_key::less_comparator(s)); auto& k1 = keys[0]; auto& k2 = keys[1]; BOOST_REQUIRE(k1._token != k2._token); // The rest of the test assumes that. BOOST_REQUIRE(k1.tri_compare(*s, dht::ring_position::starting_at(k1._token)) > 0); BOOST_REQUIRE(k1.tri_compare(*s, dht::ring_position::ending_at(k1._token)) < 0); BOOST_REQUIRE(k1.tri_compare(*s, dht::ring_position(k1)) == 0); BOOST_REQUIRE(k1.tri_compare(*s, dht::ring_position::starting_at(k2._token)) < 0); BOOST_REQUIRE(k1.tri_compare(*s, dht::ring_position::ending_at(k2._token)) < 0); BOOST_REQUIRE(k1.tri_compare(*s, dht::ring_position(k2)) < 0); BOOST_REQUIRE(k2.tri_compare(*s, dht::ring_position::starting_at(k1._token)) > 0); BOOST_REQUIRE(k2.tri_compare(*s, dht::ring_position::ending_at(k1._token)) > 0); BOOST_REQUIRE(k2.tri_compare(*s, dht::ring_position(k1)) > 0); } SEASTAR_THREAD_TEST_CASE(test_ring_position_ordering) { simple_schema table; auto cmp = dht::ring_position_comparator(*table.schema()); std::vector keys = table.make_pkeys(6); // Force keys[2-4] to share the same token keys[2]._token = keys[3]._token = keys[4]._token; std::sort(keys.begin() + 2, keys.begin() + 5, dht::ring_position_less_comparator(*table.schema())); testlog.info("Keys: {}", keys); auto positions = boost::copy_range>(keys); auto ext_positions = boost::copy_range>(keys); auto views = boost::copy_range>(positions); total_order_check(cmp) .next(dht::ring_position_view::min()) .equal_to(dht::ring_position_ext::min()) .next(dht::ring_position(keys[0].token(), dht::ring_position::token_bound::start)) .equal_to(dht::ring_position_view(keys[0].token(), nullptr, -1)) .equal_to(dht::ring_position_ext(keys[0].token(), std::nullopt, -1)) .next(views[0]) .equal_to(keys[0]) .equal_to(positions[0]) .equal_to(ext_positions[0]) .next(dht::ring_position_view::for_after_key(keys[0])) .equal_to(dht::ring_position_ext::for_after_key(keys[0])) .next(dht::ring_position(keys[0].token(), dht::ring_position::token_bound::end)) .equal_to(dht::ring_position_view(keys[0].token(), nullptr, 1)) .equal_to(dht::ring_position_ext(keys[0].token(), std::nullopt, 1)) .next(dht::ring_position(keys[1].token(), dht::ring_position::token_bound::start)) .equal_to(dht::ring_position_view(keys[1].token(), nullptr, -1)) .equal_to(dht::ring_position_ext(keys[1].token(), std::nullopt, -1)) .next(views[1]) .equal_to(keys[1]) .equal_to(positions[1]) .equal_to(ext_positions[1]) .next(dht::ring_position_view::for_after_key(keys[1])) .equal_to(dht::ring_position_ext::for_after_key(keys[1])) .next(dht::ring_position(keys[1].token(), dht::ring_position::token_bound::end)) .equal_to(dht::ring_position_view(keys[1].token(), nullptr, 1)) .equal_to(dht::ring_position_ext(keys[1].token(), std::nullopt, 1)) .next(dht::ring_position(keys[2].token(), dht::ring_position::token_bound::start)) .equal_to(dht::ring_position_view(keys[2].token(), nullptr, -1)) .equal_to(dht::ring_position_ext(keys[2].token(), std::nullopt, -1)) .next(views[2]) .equal_to(keys[2]) .equal_to(positions[2]) .equal_to(ext_positions[2]) .next(dht::ring_position_view::for_after_key(keys[2])) .equal_to(dht::ring_position_ext::for_after_key(keys[2])) .next(views[3]) .equal_to(keys[3]) .equal_to(positions[3]) .equal_to(ext_positions[3]) .next(dht::ring_position_view::for_after_key(keys[3])) .equal_to(dht::ring_position_ext::for_after_key(keys[3])) .next(views[4]) .equal_to(keys[4]) .equal_to(positions[4]) .equal_to(ext_positions[4]) .next(dht::ring_position_view::for_after_key(keys[4])) .equal_to(dht::ring_position_ext::for_after_key(keys[4])) .next(dht::ring_position(keys[4].token(), dht::ring_position::token_bound::end)) .equal_to(dht::ring_position_view(keys[4].token(), nullptr, 1)) .equal_to(dht::ring_position_ext(keys[4].token(), std::nullopt, 1)) .next(dht::ring_position(keys[5].token(), dht::ring_position::token_bound::start)) .equal_to(dht::ring_position_view(keys[5].token(), nullptr, -1)) .equal_to(dht::ring_position_ext(keys[5].token(), std::nullopt, -1)) .next(views[5]) .equal_to(keys[5]) .equal_to(positions[5]) .equal_to(ext_positions[5]) .next(dht::ring_position_view::for_after_key(keys[5])) .equal_to(dht::ring_position_ext::for_after_key(keys[5])) .next(dht::ring_position(keys[5].token(), dht::ring_position::token_bound::end)) .equal_to(dht::ring_position_view(keys[5].token(), nullptr, 1)) .equal_to(dht::ring_position_ext(keys[5].token(), std::nullopt, 1)) .next(dht::ring_position_view::max()) .equal_to(dht::ring_position_ext::max()) .check(); } SEASTAR_THREAD_TEST_CASE(test_token_no_wraparound_1) { auto t1 = token_from_long(0x5000'0000'0000'0000); auto t2 = token_from_long(0x7000'0000'0000'0000); BOOST_REQUIRE(t1 < t2); auto midpoint = dht::token::midpoint(t1, t2); BOOST_REQUIRE(midpoint > t1 && midpoint < t2); BOOST_REQUIRE_EQUAL(midpoint, token_from_long(0x6000'0000'0000'0000)); } void test_sharding(const dht::sharder& sharder, unsigned shards, std::vector shard_limits, unsigned ignorebits = 0) { auto prev_token = [] (dht::token token) { return token_from_long(long_from_token(token) - 1); }; auto s = schema_builder("ks", "cf") .with_column("c1", int32_type, column_kind::partition_key) .with_column("c2", int32_type, column_kind::partition_key) .with_column("v", int32_type) .build(); for (unsigned i = 0; i < (shards << ignorebits); ++i) { auto lim = shard_limits[i]; BOOST_REQUIRE_EQUAL(sharder.shard_of(lim), i % shards); if (i != 0) { BOOST_REQUIRE_EQUAL(sharder.shard_of(prev_token(lim)), (i - 1) % shards); BOOST_REQUIRE_EQUAL(lim, sharder.token_for_next_shard(prev_token(lim), i % shards)); } if (i != (shards << ignorebits) - 1) { auto next_shard = (i + 1) % shards; BOOST_REQUIRE_EQUAL(sharder.shard_of(sharder.token_for_next_shard(lim, next_shard)), next_shard); } } } SEASTAR_THREAD_TEST_CASE(test_murmur3_sharding) { auto make_token_vector = [] (std::vector v) { return boost::copy_range>( v | boost::adaptors::transformed(token_from_long)); }; dht::sharder mm3p7s(7); auto mm3p7s_shard_limits = make_token_vector({ -9223372036854775807, -6588122883467697006+1, -3952873730080618204+1, -1317624576693539402+1, 1317624576693539401+1, 3952873730080618203+1, 6588122883467697005+1, }); test_sharding(mm3p7s, 7, mm3p7s_shard_limits); dht::sharder mm3p2s(2); auto mm3p2s_shard_limits = make_token_vector({ -9223372036854775807, 0, }); test_sharding(mm3p2s, 2, mm3p2s_shard_limits); dht::sharder mm3p1s(1); auto mm3p1s_shard_limits = make_token_vector({ -9223372036854775807, }); test_sharding(mm3p1s, 1, mm3p1s_shard_limits); } SEASTAR_THREAD_TEST_CASE(test_murmur3_sharding_with_ignorebits) { auto make_token_vector = [] (std::vector v) { return boost::copy_range>( v | boost::adaptors::transformed(token_from_long)); }; dht::sharder mm3p7s2i(7, 2); auto mm3p7s2i_shard_limits = make_token_vector({ -9223372036854775807, -8564559748508006107, -7905747460161236406, -7246935171814466706, -6588122883467697005, -5929310595120927305, -5270498306774157604, -4611686018427387904, -3952873730080618203, -3294061441733848502, -2635249153387078802, -1976436865040309101, -1317624576693539401, -658812288346769700, 0, 658812288346769701, 1317624576693539402, 1976436865040309102, 2635249153387078803, 3294061441733848503, 3952873730080618204, 4611686018427387904, 5270498306774157605, 5929310595120927306, 6588122883467697006, 7246935171814466707, 7905747460161236407, 8564559748508006108, }); test_sharding(mm3p7s2i, 7, mm3p7s2i_shard_limits, 2); dht::sharder mm3p2s4i(2, 4); auto mm3p2s_shard_limits = make_token_vector({ -9223372036854775807, -8646911284551352320, -8070450532247928832, -7493989779944505344, -6917529027641081856, -6341068275337658368, -5764607523034234880, -5188146770730811392, -4611686018427387904, -4035225266123964416, -3458764513820540928, -2882303761517117440, -2305843009213693952, -1729382256910270464, -1152921504606846976, -576460752303423488, 0, 576460752303423488, 1152921504606846976, 1729382256910270464, 2305843009213693952, 2882303761517117440, 3458764513820540928, 4035225266123964416, 4611686018427387904, 5188146770730811392, 5764607523034234880, 6341068275337658368, 6917529027641081856, 7493989779944505344, 8070450532247928832, 8646911284551352320, }); test_sharding(mm3p2s4i, 2, mm3p2s_shard_limits, 4); } static dht::partition_range normalize(dht::partition_range pr) { auto start = pr.start(); if (start && start->value().token() == dht::minimum_token()) { start = std::nullopt; } auto end = pr.end(); if (end && end->value().token() == dht::maximum_token()) { end = std::nullopt; } return dht::partition_range(start, end); }; static void test_something_with_some_interesting_ranges_and_sharder(std::function func_to_test) { auto s = schema_builder("ks", "cf") .with_column("c1", int32_type, column_kind::partition_key) .with_column("c2", int32_type, column_kind::partition_key) .with_column("v", int32_type) .build(); auto some_sharders = { dht::sharder(1, 0), dht::sharder(7, 4), dht::sharder(4, 0), dht::sharder(32, 8), // More, and we OOM since memory isn't configured }; auto t1 = token_from_long(int64_t(-0x7fff'ffff'ffff'fffe)); auto t2 = token_from_long(int64_t(-1)); auto t3 = token_from_long(int64_t(1)); auto t4 = token_from_long(int64_t(0x7fff'ffff'ffff'fffe)); auto make_bound = [] (dht::ring_position rp) { return std::make_optional(range_bound(std::move(rp))); }; auto some_murmur3_ranges = { dht::partition_range::make_open_ended_both_sides(), dht::partition_range::make_starting_with(dht::ring_position::starting_at(t1)), dht::partition_range::make_starting_with(dht::ring_position::starting_at(t2)), dht::partition_range::make_starting_with(dht::ring_position::ending_at(t3)), dht::partition_range::make_starting_with(dht::ring_position::starting_at(t4)), dht::partition_range::make_ending_with(dht::ring_position::starting_at(t1)), dht::partition_range::make_ending_with(dht::ring_position::starting_at(t2)), dht::partition_range::make_ending_with(dht::ring_position::starting_at(t3)), dht::partition_range::make_ending_with(dht::ring_position::starting_at(t4)), dht::partition_range(make_bound(dht::ring_position::starting_at(t2)), make_bound(dht::ring_position::ending_at(t3))), dht::partition_range(make_bound(dht::ring_position::ending_at(t1)), make_bound(dht::ring_position::starting_at(t4))), }; for (auto&& sharder : some_sharders) { auto schema = schema_builder(s) .with_sharder(sharder.shard_count(), sharder.sharding_ignore_msb()).build(); for (auto&& range : some_murmur3_ranges) { func_to_test(*schema, range); } } } static void do_test_split_range_to_single_shard(const schema& s, const dht::partition_range& pr) { for (auto shard : boost::irange(0u, s.get_sharder().shard_count())) { auto ranges = dht::split_range_to_single_shard(s, pr, shard).get0(); auto sharder = dht::ring_position_range_sharder(s.get_sharder(), pr); auto x = sharder.next(s); auto cmp = dht::ring_position_comparator(s); auto reference_ranges = std::vector(); while (x) { if (x->shard == shard) { reference_ranges.push_back(std::move(x->ring_range)); } x = sharder.next(s); } BOOST_REQUIRE(ranges.size() == reference_ranges.size()); for (auto&& rs : boost::combine(ranges, reference_ranges)) { auto&& r1 = normalize(boost::get<0>(rs)); auto&& r2 = normalize(boost::get<1>(rs)); BOOST_REQUIRE(r1.contains(r2, cmp)); BOOST_REQUIRE(r2.contains(r1, cmp)); } } } SEASTAR_THREAD_TEST_CASE(test_split_range_single_shard) { return test_something_with_some_interesting_ranges_and_sharder(do_test_split_range_to_single_shard); } static void test_something_with_some_interesting_ranges_and_sharder_with_token_range(std::function func_to_test) { auto s = schema_builder("ks", "cf") .with_column("c1", int32_type, column_kind::partition_key) .with_column("c2", int32_type, column_kind::partition_key) .with_column("v", int32_type) .build(); auto some_sharder = { dht::sharder(1, 0), dht::sharder(7, 4), dht::sharder(4, 0), dht::sharder(32, 8), // More, and we OOM since memory isn't configured }; auto t1 = token_from_long(int64_t(-0x7fff'ffff'ffff'fffe)); auto t2 = token_from_long(int64_t(-1)); auto t3 = token_from_long(int64_t(1)); auto t4 = token_from_long(int64_t(0x7fff'ffff'ffff'fffe)); auto make_bound = [] (dht::token t) { return range_bound(std::move(t)); }; auto some_ranges = { dht::token_range::make_open_ended_both_sides(), dht::token_range::make_starting_with(make_bound(t1)), dht::token_range::make_starting_with(make_bound(t2)), dht::token_range::make_starting_with(make_bound(t3)), dht::token_range::make_starting_with(make_bound(t4)), dht::token_range::make_ending_with(make_bound(t1)), dht::token_range::make_ending_with(make_bound(t2)), dht::token_range::make_ending_with(make_bound(t3)), dht::token_range::make_ending_with(make_bound(t4)), dht::token_range(make_bound(t2), make_bound(t3)), dht::token_range(make_bound(t1), make_bound(t4)), }; for (auto&& sharder : some_sharder) { for (auto&& range : some_ranges) { func_to_test(sharder, *s, range); } } } static void do_test_selective_token_range_sharder(const dht::sharder& input_sharder, const schema& s, const dht::token_range& range) { bool debug = false; for (auto shard : boost::irange(0u, input_sharder.shard_count())) { auto sharder = dht::selective_token_range_sharder(input_sharder, range, shard); auto range_shard = sharder.next(); while (range_shard) { if (range_shard->start() && range_shard->start()->is_inclusive()) { auto start_shard = input_sharder.shard_of(range_shard->start()->value()); if (debug) { std::cout << " start_shard " << start_shard << " shard " << shard << " range " << range_shard << "\n"; } BOOST_REQUIRE(start_shard == shard); } if (range_shard->end() && range_shard->end()->is_inclusive()) { auto end_shard = input_sharder.shard_of(range_shard->end()->value()); if (debug) { std::cout << " end_shard " << end_shard << " shard " << shard << " range " << range_shard << "\n"; } BOOST_REQUIRE(end_shard == shard); } auto midpoint = dht::token::midpoint( range_shard->start() ? range_shard->start()->value() : dht::minimum_token(), range_shard->end() ? range_shard->end()->value() : dht::minimum_token()); auto mid_shard = input_sharder.shard_of(midpoint); if (debug) { std::cout << " mid " << mid_shard << " shard " << shard << " range " << range_shard << "\n"; } BOOST_REQUIRE(mid_shard == shard); range_shard = sharder.next(); } } } SEASTAR_THREAD_TEST_CASE(test_selective_token_range_sharder) { return test_something_with_some_interesting_ranges_and_sharder_with_token_range(do_test_selective_token_range_sharder); } SEASTAR_THREAD_TEST_CASE(test_find_first_token_for_shard) { const unsigned cpu_count = 3; const unsigned ignore_msb_bits = 10; dht::sharder sharder(cpu_count, ignore_msb_bits); auto first_boundary = sharder.token_for_next_shard(dht::minimum_token(), 1); auto second_boundary = sharder.token_for_next_shard(dht::minimum_token(), 2); auto third_boundary = sharder.token_for_next_shard(dht::minimum_token(), 0); auto next_token = [] (dht::token t) { assert(dht::token::to_int64(t) < std::numeric_limits::max()); return dht::token::from_int64(dht::token::to_int64(t) + 1); }; auto prev_token = [] (dht::token t) { assert(dht::token::to_int64(t) > std::numeric_limits::min() + 1); return dht::token::from_int64(dht::token::to_int64(t) - 1); }; BOOST_REQUIRE_EQUAL(dht::token::from_int64(std::numeric_limits::min() + 1), dht::find_first_token_for_shard(sharder, dht::minimum_token(), dht::maximum_token(), 0)); BOOST_REQUIRE_EQUAL(first_boundary, dht::find_first_token_for_shard(sharder, dht::minimum_token(), dht::maximum_token(), 1)); BOOST_REQUIRE_EQUAL(second_boundary, dht::find_first_token_for_shard(sharder, dht::minimum_token(), dht::maximum_token(), 2)); BOOST_REQUIRE_EQUAL(dht::token::from_int64(std::numeric_limits::min() + 1), dht::find_first_token_for_shard(sharder, dht::minimum_token(), dht::token::from_int64(std::numeric_limits::min() + 1), 0)); BOOST_REQUIRE_EQUAL(first_boundary, dht::find_first_token_for_shard(sharder, dht::minimum_token(), first_boundary, 1)); BOOST_REQUIRE_EQUAL(second_boundary, dht::find_first_token_for_shard(sharder, dht::minimum_token(), second_boundary, 2)); BOOST_REQUIRE_EQUAL(third_boundary, dht::find_first_token_for_shard(sharder, prev_token(first_boundary), dht::maximum_token(), 0)); BOOST_REQUIRE_EQUAL(third_boundary, dht::find_first_token_for_shard(sharder, first_boundary, dht::maximum_token(), 0)); BOOST_REQUIRE_EQUAL(third_boundary, dht::find_first_token_for_shard(sharder, second_boundary, dht::maximum_token(), 0)); BOOST_REQUIRE_EQUAL(next_token(first_boundary), dht::find_first_token_for_shard(sharder, first_boundary, dht::maximum_token(), 1)); BOOST_REQUIRE_EQUAL(next_token(second_boundary), dht::find_first_token_for_shard(sharder, second_boundary, dht::maximum_token(), 2)); BOOST_REQUIRE_EQUAL(next_token(third_boundary), dht::find_first_token_for_shard(sharder, third_boundary, dht::maximum_token(), 0)); BOOST_REQUIRE_EQUAL(first_boundary, dht::find_first_token_for_shard(sharder, prev_token(first_boundary), dht::maximum_token(), 1)); BOOST_REQUIRE_EQUAL(second_boundary, dht::find_first_token_for_shard(sharder, prev_token(second_boundary), dht::maximum_token(), 2)); BOOST_REQUIRE_EQUAL(third_boundary, dht::find_first_token_for_shard(sharder, prev_token(third_boundary), dht::maximum_token(), 0)); BOOST_REQUIRE_EQUAL(prev_token(first_boundary), dht::find_first_token_for_shard(sharder, dht::minimum_token(), prev_token(first_boundary), 1)); BOOST_REQUIRE_EQUAL(prev_token(second_boundary), dht::find_first_token_for_shard(sharder, dht::minimum_token(), prev_token(second_boundary), 2)); BOOST_REQUIRE_EQUAL(prev_token(third_boundary), dht::find_first_token_for_shard(sharder, prev_token(first_boundary), prev_token(third_boundary), 0)); auto last_token = dht::token::from_int64(std::numeric_limits::max()); auto last_shard = sharder.shard_of(last_token); BOOST_REQUIRE_EQUAL(last_token, dht::find_first_token_for_shard(sharder, prev_token(last_token), dht::maximum_token(), last_shard)); BOOST_REQUIRE_EQUAL(dht::maximum_token(), dht::find_first_token_for_shard(sharder, prev_token(last_token), dht::maximum_token(), (last_shard + 1) % cpu_count)); BOOST_REQUIRE_EQUAL(dht::maximum_token(), dht::find_first_token_for_shard(sharder, prev_token(last_token), dht::maximum_token(), (last_shard + 2) % cpu_count)); BOOST_REQUIRE_EQUAL(dht::token::from_int64(std::numeric_limits::min() + 1), dht::find_first_token_for_shard(sharder, prev_token(last_token), third_boundary, (last_shard + 1) % cpu_count)); BOOST_REQUIRE_EQUAL(first_boundary, dht::find_first_token_for_shard(sharder, prev_token(last_token), third_boundary, (last_shard + 2) % cpu_count)); BOOST_REQUIRE_EQUAL(prev_token(first_boundary), dht::find_first_token_for_shard(sharder, prev_token(last_token), prev_token(first_boundary), (last_shard + 2) % cpu_count)); }