d22fdf4261
Cache imposes requirements on how updates to the on-disk mutation source
are made:
1) each change to the on-disk muation source must be followed
by cache synchronization reflecting that change
2) The two must be serialized with other synchronizations
3) must have strong failure guarantees (atomicity)
Because of that, sstable list update and cache synchronization must be
done under a lock, and cache synchronization cannot fail to synchronize.
Normally cache synchronization achieves no-failure thing by wiping the
cache (which is noexcept) in case failure is detect. There are some
setup steps hoever which cannot be skipped, e.g. taking a lock
followed by switching cache to use the new snapshot. That truly cannot
fail. The lock inside cache synchronizers is redundant, since the
user needs to take it anyway around the combined operation.
In order to make ensuring strong exception guarantees easier, and
making the cache interface easier to use correctly, this patch moves
the control of the combined update into the cache. This is done by
having cache::update() et al accept a callback (external_updater)
which is supposed to perform modiciation of the underlying mutation
source when invoked.
This is in-line with the layering. Cache is layered on top of the
on-disk mutation source (it wraps it) and reading has to go through
cache. After the patch, modification also goes through cache. This way
more of cache's requirements can be confined to its implementation.
The failure semantics of update() and other synchronizers needed to
change due to strong exception guaratnees. Now if it fails, it means
the update was not performed, neither to the cache nor to the
underlying mutation source.
The database::_cache_update_sem goes away, serialization is done
internally by the cache.
The external_updater needs to have strong exception guarantees. This
requirement is not new. It is however currently violated in some
places. This patch marks those callbacks as noexcept and leaves a
FIXME. Those should be fixed, but that's not in the scope of this
patch. Aborting is still better than corrupting the state.
Fixes #2754.
Also fixes the following test failure:
tests/row_cache_test.cc(949): fatal error: in "test_update_failure": critical check it->second.equal(*s, mopt->partition()) has failed
which started to trigger after commit 318423d50b. Thread stack
allocation may fail, in which case we did not do the necessary
invalidation.
258 lines
10 KiB
C++
258 lines
10 KiB
C++
/*
|
|
* Copyright (C) 2015 ScyllaDB
|
|
*/
|
|
|
|
/*
|
|
* This file is part of Scylla.
|
|
*
|
|
* Scylla is free software: you can redistribute it and/or modify
|
|
* it under the terms of the GNU Affero General Public License as published by
|
|
* the Free Software Foundation, either version 3 of the License, or
|
|
* (at your option) any later version.
|
|
*
|
|
* Scylla is distributed in the hope that it will be useful,
|
|
* but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
|
* GNU General Public License for more details.
|
|
*
|
|
* You should have received a copy of the GNU General Public License
|
|
* along with Scylla. If not, see <http://www.gnu.org/licenses/>.
|
|
*/
|
|
|
|
#include <core/distributed.hh>
|
|
#include <core/app-template.hh>
|
|
#include <core/sstring.hh>
|
|
#include <core/thread.hh>
|
|
|
|
#include "utils/managed_bytes.hh"
|
|
#include "utils/logalloc.hh"
|
|
#include "row_cache.hh"
|
|
#include "log.hh"
|
|
#include "schema_builder.hh"
|
|
#include "memtable.hh"
|
|
|
|
#include "disk-error-handler.hh"
|
|
|
|
thread_local disk_error_signal_type commit_error;
|
|
thread_local disk_error_signal_type general_disk_error;
|
|
|
|
static
|
|
partition_key new_key(schema_ptr s) {
|
|
static thread_local int next = 0;
|
|
return partition_key::from_single_value(*s, to_bytes(sprint("key%d", next++)));
|
|
}
|
|
|
|
static
|
|
clustering_key new_ckey(schema_ptr s) {
|
|
static thread_local int next = 0;
|
|
return clustering_key::from_single_value(*s, to_bytes(sprint("ckey%d", next++)));
|
|
}
|
|
|
|
int main(int argc, char** argv) {
|
|
namespace bpo = boost::program_options;
|
|
app_template app;
|
|
app.add_options()
|
|
("debug", "enable debug logging");
|
|
|
|
return app.run(argc, argv, [&app] {
|
|
if (app.configuration().count("debug")) {
|
|
logging::logger_registry().set_all_loggers_level(logging::log_level::debug);
|
|
}
|
|
|
|
// This test is supposed to verify that when we're low on memory but
|
|
// we still have plenty of evictable memory in cache, we should be
|
|
// able to populate cache with large mutations This test works only
|
|
// with seastar's allocator.
|
|
return seastar::async([] {
|
|
auto s = schema_builder("ks", "cf")
|
|
.with_column("pk", bytes_type, column_kind::partition_key)
|
|
.with_column("ck", bytes_type, column_kind::clustering_key)
|
|
.with_column("v", bytes_type, column_kind::regular_column)
|
|
.build();
|
|
|
|
cache_tracker tracker;
|
|
row_cache cache(s, make_empty_snapshot_source(), tracker);
|
|
|
|
auto mt = make_lw_shared<memtable>(s);
|
|
std::vector<dht::decorated_key> keys;
|
|
|
|
size_t cell_size = 1024;
|
|
size_t row_count = 40 * 1024; // 40M mutations
|
|
size_t large_cell_size = cell_size * row_count;
|
|
|
|
auto make_small_mutation = [&] {
|
|
mutation m(new_key(s), s);
|
|
m.set_clustered_cell(new_ckey(s), "v", data_value(bytes(bytes::initialized_later(), cell_size)), 1);
|
|
return m;
|
|
};
|
|
|
|
auto make_large_mutation = [&] {
|
|
mutation m(new_key(s), s);
|
|
m.set_clustered_cell(new_ckey(s), "v", data_value(bytes(bytes::initialized_later(), large_cell_size)), 2);
|
|
return m;
|
|
};
|
|
|
|
std::random_device random;
|
|
std::default_random_engine random_engine(random());
|
|
|
|
for (int i = 0; i < 10; i++) {
|
|
auto key = dht::global_partitioner().decorate_key(*s, new_key(s));
|
|
|
|
mutation m1(key, s);
|
|
m1.set_clustered_cell(new_ckey(s), "v", data_value(bytes(bytes::initialized_later(), cell_size)), 1);
|
|
cache.populate(m1);
|
|
|
|
// Putting large mutations into the memtable. Should take about row_count*cell_size each.
|
|
mutation m2(key, s);
|
|
for (size_t j = 0; j < row_count; j++) {
|
|
m2.set_clustered_cell(new_ckey(s), "v", data_value(bytes(bytes::initialized_later(), cell_size)), 2);
|
|
}
|
|
|
|
mt->apply(m2);
|
|
keys.push_back(key);
|
|
}
|
|
|
|
auto reclaimable_memory = [] {
|
|
return memory::stats().free_memory() + logalloc::shard_tracker().occupancy().free_space();
|
|
};
|
|
|
|
std::cout << "memtable occupancy: " << mt->occupancy() << "\n";
|
|
std::cout << "Cache occupancy: " << tracker.region().occupancy() << "\n";
|
|
std::cout << "Reclaimable memory: " << reclaimable_memory() << "\n";
|
|
|
|
// We need to have enough Free memory to copy memtable into cache
|
|
// When this assertion fails, increase amount of memory
|
|
assert(mt->occupancy().used_space() < reclaimable_memory());
|
|
|
|
std::deque<dht::decorated_key> cache_stuffing;
|
|
auto fill_cache_to_the_top = [&] {
|
|
std::cout << "Filling up memory with evictable data\n";
|
|
while (true) {
|
|
auto evictions_before = tracker.get_stats().partition_evictions;
|
|
// Ensure that entries matching memtable partitions are evicted
|
|
// last, we want to hit the merge path in row_cache::update()
|
|
for (auto&& key : keys) {
|
|
cache.touch(key);
|
|
}
|
|
auto m = make_small_mutation();
|
|
cache_stuffing.push_back(m.decorated_key());
|
|
cache.populate(m);
|
|
if (tracker.get_stats().partition_evictions > evictions_before) {
|
|
break;
|
|
}
|
|
}
|
|
std::cout << "Shuffling..\n";
|
|
// Evict in random order to create fragmentation.
|
|
std::shuffle(cache_stuffing.begin(), cache_stuffing.end(), random_engine);
|
|
for (auto&& key : cache_stuffing) {
|
|
cache.touch(key);
|
|
}
|
|
// Ensure that entries matching memtable partitions are evicted
|
|
// last, we want to hit the merge path in row_cache::update()
|
|
for (auto&& key : keys) {
|
|
cache.touch(key);
|
|
}
|
|
std::cout << "Reclaimable memory: " << reclaimable_memory() << "\n";
|
|
std::cout << "Cache occupancy: " << tracker.region().occupancy() << "\n";
|
|
};
|
|
|
|
std::deque<std::unique_ptr<char[]>> stuffing;
|
|
auto fragment_free_space = [&] {
|
|
stuffing.clear();
|
|
std::cout << "Reclaimable memory: " << reclaimable_memory() << "\n";
|
|
std::cout << "Free memory: " << memory::stats().free_memory() << "\n";
|
|
std::cout << "Cache occupancy: " << tracker.region().occupancy() << "\n";
|
|
|
|
// Induce memory fragmentation by taking down cache segments,
|
|
// which should be evicted in random order, and inducing high
|
|
// waste level in them. Should leave around up to 100M free,
|
|
// but no LSA segment should fit.
|
|
for (unsigned i = 0; i < 100 * 1024 * 1024 / (logalloc::segment_size / 2); ++i) {
|
|
stuffing.emplace_back(std::make_unique<char[]>(logalloc::segment_size / 2 + 1));
|
|
}
|
|
|
|
std::cout << "After fragmenting:\n";
|
|
std::cout << "Reclaimable memory: " << reclaimable_memory() << "\n";
|
|
std::cout << "Free memory: " << memory::stats().free_memory() << "\n";
|
|
std::cout << "Cache occupancy: " << tracker.region().occupancy() << "\n";
|
|
};
|
|
|
|
fill_cache_to_the_top();
|
|
|
|
fragment_free_space();
|
|
|
|
cache.update([] {}, *mt).get();
|
|
|
|
stuffing.clear();
|
|
cache_stuffing.clear();
|
|
|
|
// Verify that all mutations from memtable went through
|
|
for (auto&& key : keys) {
|
|
auto range = dht::partition_range::make_singular(key);
|
|
auto reader = cache.make_reader(s, range);
|
|
auto mo = mutation_from_streamed_mutation(reader().get0()).get0();
|
|
assert(mo);
|
|
assert(mo->partition().live_row_count(*s) ==
|
|
row_count + 1 /* one row was already in cache before update()*/);
|
|
}
|
|
|
|
std::cout << "Testing reading from cache.\n";
|
|
|
|
fill_cache_to_the_top();
|
|
|
|
for (auto&& key : keys) {
|
|
cache.touch(key);
|
|
}
|
|
|
|
for (auto&& key : keys) {
|
|
auto range = dht::partition_range::make_singular(key);
|
|
auto reader = cache.make_reader(s, range);
|
|
auto mo = reader().get0();
|
|
assert(mo);
|
|
}
|
|
|
|
std::cout << "Testing reading when memory can't be reclaimed.\n";
|
|
// We want to check that when we really can't reserve memory, allocating_section
|
|
// throws rather than enter infinite loop.
|
|
{
|
|
stuffing.clear();
|
|
cache_stuffing.clear();
|
|
tracker.clear();
|
|
|
|
// eviction victims
|
|
for (unsigned i = 0; i < logalloc::segment_size / cell_size; ++i) {
|
|
cache.populate(make_small_mutation());
|
|
}
|
|
|
|
const mutation& m = make_large_mutation();
|
|
auto range = dht::partition_range::make_singular(m.decorated_key());
|
|
|
|
cache.populate(m);
|
|
|
|
logalloc::shard_tracker().reclaim_all_free_segments();
|
|
|
|
{
|
|
logalloc::reclaim_lock _(tracker.region());
|
|
try {
|
|
while (true) {
|
|
stuffing.emplace_back(std::make_unique<char[]>(logalloc::segment_size));
|
|
}
|
|
} catch (const std::bad_alloc&) {
|
|
//expected
|
|
}
|
|
}
|
|
|
|
try {
|
|
auto reader = cache.make_reader(s, range);
|
|
assert(!reader().get0());
|
|
auto evicted_from_cache = logalloc::segment_size + large_cell_size;
|
|
new char[evicted_from_cache + logalloc::segment_size];
|
|
assert(false); // The test is not invoking the case which it's supposed to test
|
|
} catch (const std::bad_alloc&) {
|
|
// expected
|
|
}
|
|
}
|
|
});
|
|
});
|
|
}
|