ecb6fb00f0
The streamed_mutation_freezer class uses a deque to avoid large allocations, but fails as seen in the referenced issue when the vector backing the deque grows too large. This may be a problem in itself, but the issue doesn't provide enough information to tell. Fix the immediate problem by switching to chunked_vector, which is better in avoiding large allocations. We do lose some early-free in serialize_mutation_fragments(), but since most of the memory should be in the clustering row itself, not in the deque/chunked_vector holding it, it should not be a problem. Fixes #28275 Closes scylladb/scylladb#28281
279 lines
8.8 KiB
C++
279 lines
8.8 KiB
C++
/*
|
|
* Copyright (C) 2015-present ScyllaDB
|
|
*/
|
|
|
|
/*
|
|
* SPDX-License-Identifier: LicenseRef-ScyllaDB-Source-Available-1.0
|
|
*/
|
|
|
|
#include <seastar/core/coroutine.hh>
|
|
#include "frozen_mutation.hh"
|
|
#include "schema/schema_registry.hh"
|
|
#include "mutation_partition.hh"
|
|
#include "mutation.hh"
|
|
#include "counters.hh"
|
|
#include "partition_builder.hh"
|
|
#include "mutation_partition_serializer.hh"
|
|
#include "query/query-result-set.hh"
|
|
#include "idl/mutation.dist.hh"
|
|
#include "idl/mutation.dist.impl.hh"
|
|
#include "readers/mutation_reader.hh"
|
|
#include "converting_mutation_partition_applier.hh"
|
|
#include "mutation_partition_view.hh"
|
|
|
|
//
|
|
// Representation layout:
|
|
//
|
|
// <mutation> ::= <column-family-id> <schema-version> <partition-key> <partition>
|
|
//
|
|
|
|
using namespace db;
|
|
|
|
ser::mutation_view frozen_mutation::mutation_view() const {
|
|
auto in = ser::as_input_stream(_bytes);
|
|
return ser::deserialize(in, std::type_identity<ser::mutation_view>());
|
|
}
|
|
|
|
table_id
|
|
frozen_mutation::column_family_id() const {
|
|
return mutation_view().table_id();
|
|
}
|
|
|
|
table_schema_version
|
|
frozen_mutation::schema_version() const {
|
|
return mutation_view().schema_version();
|
|
}
|
|
|
|
partition_key_view
|
|
frozen_mutation::key() const {
|
|
return _pk;
|
|
}
|
|
|
|
dht::decorated_key
|
|
frozen_mutation::decorated_key(const schema& s) const {
|
|
return dht::decorate_key(s, key());
|
|
}
|
|
|
|
partition_key frozen_mutation::deserialize_key() const {
|
|
return mutation_view().key();
|
|
}
|
|
|
|
frozen_mutation::frozen_mutation(bytes_ostream&& b)
|
|
: _bytes(std::move(b))
|
|
, _pk(deserialize_key())
|
|
{
|
|
_bytes.reduce_chunk_count();
|
|
}
|
|
|
|
frozen_mutation::frozen_mutation(bytes_ostream&& b, partition_key pk)
|
|
: _bytes(std::move(b))
|
|
, _pk(std::move(pk))
|
|
{
|
|
_bytes.reduce_chunk_count();
|
|
}
|
|
|
|
frozen_mutation::frozen_mutation(const mutation& m)
|
|
: _pk(m.key())
|
|
{
|
|
mutation_partition_serializer part_ser(*m.schema(), m.partition());
|
|
|
|
ser::writer_of_mutation<bytes_ostream> wom(_bytes);
|
|
std::move(wom).write_table_id(m.schema()->id())
|
|
.write_schema_version(m.schema()->version())
|
|
.write_key(m.key())
|
|
.partition([&] (auto wr) {
|
|
part_ser.write(std::move(wr));
|
|
}).end_mutation();
|
|
_bytes.reduce_chunk_count();
|
|
}
|
|
|
|
mutation
|
|
frozen_mutation::unfreeze(schema_ptr schema) const {
|
|
check_schema_version(schema_version(), *schema);
|
|
mutation m(schema, key());
|
|
partition_builder b(*schema, m.partition());
|
|
try {
|
|
partition().accept(*schema, b);
|
|
} catch (...) {
|
|
std::throw_with_nested(std::runtime_error(format(
|
|
"frozen_mutation::unfreeze(): failed unfreezing mutation {} of {}.{}", key(), schema->ks_name(), schema->cf_name())));
|
|
}
|
|
return m;
|
|
}
|
|
|
|
mutation frozen_mutation::unfreeze_upgrading(schema_ptr schema, const column_mapping& cm) const {
|
|
mutation m(schema, key());
|
|
converting_mutation_partition_applier v(cm, *schema, m.partition());
|
|
try {
|
|
partition().accept(cm, v);
|
|
} catch (...) {
|
|
std::throw_with_nested(std::runtime_error(format(
|
|
"frozen_mutation::unfreeze_upgrading(): failed unfreezing mutation {} of {}.{}", key(), schema->ks_name(), schema->cf_name())));
|
|
}
|
|
return m;
|
|
}
|
|
|
|
frozen_mutation freeze(const mutation& m) {
|
|
return frozen_mutation{ m };
|
|
}
|
|
|
|
utils::chunked_vector<frozen_mutation> freeze(const utils::chunked_vector<mutation>& muts) {
|
|
return muts | std::views::transform([] (const mutation& m) {
|
|
return freeze(m);
|
|
}) | std::ranges::to<utils::chunked_vector<frozen_mutation>>();
|
|
}
|
|
|
|
utils::chunked_vector<mutation> unfreeze(const utils::chunked_vector<frozen_mutation>& muts) {
|
|
return muts | std::views::transform([] (const frozen_mutation& fm) {
|
|
return fm.unfreeze(local_schema_registry().get(fm.schema_version()));
|
|
}) | std::ranges::to<utils::chunked_vector<mutation>>();
|
|
}
|
|
|
|
|
|
mutation_partition_view frozen_mutation::partition() const {
|
|
return mutation_partition_view::from_view(mutation_view().partition());
|
|
}
|
|
|
|
frozen_mutation::printer frozen_mutation::pretty_printer(schema_ptr s) const {
|
|
return { *this, std::move(s) };
|
|
}
|
|
|
|
stop_iteration streamed_mutation_freezer::consume(tombstone pt) {
|
|
_partition_tombstone = pt;
|
|
return stop_iteration::no;
|
|
}
|
|
|
|
stop_iteration streamed_mutation_freezer::consume(static_row&& sr) {
|
|
_sr = std::move(sr);
|
|
return stop_iteration::no;
|
|
}
|
|
|
|
stop_iteration streamed_mutation_freezer::consume(clustering_row&& cr) {
|
|
_crs.emplace_back(std::move(cr));
|
|
return stop_iteration::no;
|
|
}
|
|
|
|
stop_iteration streamed_mutation_freezer::consume(range_tombstone&& rt) {
|
|
_rts.apply(_schema, std::move(rt));
|
|
return stop_iteration::no;
|
|
}
|
|
|
|
frozen_mutation streamed_mutation_freezer::consume_end_of_stream() {
|
|
bytes_ostream out;
|
|
ser::writer_of_mutation<bytes_ostream> wom(out);
|
|
std::move(wom).write_table_id(_schema.id())
|
|
.write_schema_version(_schema.version())
|
|
.write_key(_key)
|
|
.partition([&] (auto wr) {
|
|
serialize_mutation_fragments(_schema, _partition_tombstone,
|
|
std::move(_sr), std::move(_rts),
|
|
std::move(_crs), std::move(wr));
|
|
}).end_mutation();
|
|
return frozen_mutation(std::move(out), std::move(_key));
|
|
}
|
|
|
|
class fragmenting_mutation_freezer {
|
|
const schema& _schema;
|
|
std::optional<partition_key> _key;
|
|
|
|
tombstone _partition_tombstone;
|
|
std::optional<static_row> _sr;
|
|
utils::chunked_vector<clustering_row> _crs;
|
|
range_tombstone_list _rts;
|
|
|
|
frozen_mutation_consumer_fn _consumer;
|
|
|
|
bool _fragmented = false;
|
|
size_t _dirty_size = 0;
|
|
size_t _fragment_size;
|
|
|
|
range_tombstone_change _current_rtc;
|
|
private:
|
|
future<stop_iteration> flush() {
|
|
bytes_ostream out;
|
|
ser::writer_of_mutation<bytes_ostream> wom(out);
|
|
std::move(wom).write_table_id(_schema.id())
|
|
.write_schema_version(_schema.version())
|
|
.write_key(*_key)
|
|
.partition([&] (auto wr) {
|
|
serialize_mutation_fragments(_schema, _partition_tombstone,
|
|
std::move(_sr), std::move(_rts),
|
|
std::move(_crs), std::move(wr));
|
|
}).end_mutation();
|
|
|
|
_sr = { };
|
|
_rts.clear();
|
|
_crs.clear();
|
|
_dirty_size = 0;
|
|
return _consumer(frozen_mutation(std::move(out), *_key), _fragmented);
|
|
}
|
|
|
|
future<stop_iteration> maybe_flush() {
|
|
if (_dirty_size >= _fragment_size) {
|
|
_fragmented = true;
|
|
return flush();
|
|
}
|
|
return make_ready_future<stop_iteration>(stop_iteration::no);
|
|
}
|
|
public:
|
|
fragmenting_mutation_freezer(const schema& s, frozen_mutation_consumer_fn c, size_t fragment_size)
|
|
: _schema(s), _rts(s), _consumer(c), _fragment_size(fragment_size), _current_rtc(position_in_partition::before_all_clustered_rows(), {}) { }
|
|
|
|
future<stop_iteration> consume(partition_start&& ps) {
|
|
_key = std::move(ps.key().key());
|
|
_fragmented = false;
|
|
_dirty_size += sizeof(tombstone);
|
|
_partition_tombstone = ps.partition_tombstone();
|
|
return make_ready_future<stop_iteration>(stop_iteration::no);
|
|
}
|
|
|
|
future<stop_iteration> consume(static_row&& sr) {
|
|
_sr = std::move(sr);
|
|
_dirty_size += _sr->memory_usage(_schema);
|
|
return maybe_flush();
|
|
}
|
|
|
|
future<stop_iteration> consume(clustering_row&& cr) {
|
|
_dirty_size += cr.memory_usage(_schema);
|
|
_crs.emplace_back(std::move(cr));
|
|
return maybe_flush();
|
|
}
|
|
|
|
future<stop_iteration> consume(range_tombstone_change&& rtc) {
|
|
auto ret = make_ready_future<stop_iteration>(stop_iteration::no);
|
|
if (_current_rtc.tombstone()) {
|
|
auto rt = range_tombstone(_current_rtc.position(), rtc.position(), _current_rtc.tombstone());
|
|
_dirty_size += rt.memory_usage(_schema);
|
|
_rts.apply(_schema, std::move(rt));
|
|
ret = maybe_flush();
|
|
}
|
|
_current_rtc = std::move(rtc);
|
|
return ret;
|
|
}
|
|
|
|
future<stop_iteration> consume(partition_end&&) {
|
|
if (_dirty_size) {
|
|
return flush();
|
|
}
|
|
return make_ready_future<stop_iteration>(stop_iteration::no);
|
|
}
|
|
};
|
|
|
|
future<> fragment_and_freeze(mutation_reader mr, frozen_mutation_consumer_fn c, size_t fragment_size)
|
|
{
|
|
std::exception_ptr ex;
|
|
try {
|
|
fragmenting_mutation_freezer freezer(*mr.schema(), c, fragment_size);
|
|
mutation_fragment_v2_opt mfopt;
|
|
while ((mfopt = co_await mr()) && (co_await std::move(*mfopt).consume(freezer) == stop_iteration::no));
|
|
} catch (...) {
|
|
ex = std::current_exception();
|
|
}
|
|
|
|
co_await mr.close();
|
|
|
|
if (ex) {
|
|
std::rethrow_exception(std::move(ex));
|
|
}
|
|
}
|