mirror of
https://github.com/scylladb/scylladb.git
synced 2026-04-27 03:45:11 +00:00
The write paths in storage_proxy pass replica sets as
std::unordered_set<gms::inet_address>. This is a complex type, with
N+1 allocations for N members, so we change it to a small_vector (via
inet_address_vector_replica_set) which requires just one allocation, and
even zero when up to three replicas are used.
This change is more nuanced than the corresponding change to the read path
abe3d7d7 ("Merge 'storage_proxy: use small_vector for vectors of
inet_address' from Avi Kivity"), for two reasons:
- there is a quadratic algorithm in
abstract_write_response_handler::response(): it searches for a replica
and erases it. Since this happens for every replica, it happens N^2/2
times.
- replica sets for writes always include all datacenters, while reads
usually involve just one datacenter.
So, a write to a keyspace that has 5 datacenters will invoke 15*(15-1)/2
=105 compares.
We could remove this by sending the index of the replica in the replica
set to the replica and ask it to include the index in the response, but
I think that this is unnecessary. Those 105 compares need to be only
105/15 = 7 times cheaper than the corresponding unordered_set operation,
which they surely will. Handling a response after a cross-datacenter round
trip surely involves L3 cache misses, and a small_vector reduces these
to a minimum compared to an unordered_set with its bucket table, linked
list walking and managent, and table rehashing.
Tests using perf_simple_query --write --smp 1 --operations-per-shard 1000000
--task-quota-ms show two allocations removed (as expected) and a nice
reduction in instructions executed.
before: median 204842.54 tps ( 54.2 allocs/op, 13.2 tasks/op, 49890 insns/op)
after: median 206077.65 tps ( 52.2 allocs/op, 13.2 tasks/op, 49138 insns/op)
Closes #8847
337 lines
12 KiB
C++
337 lines
12 KiB
C++
/*
|
|
* Licensed to the Apache Software Foundation (ASF) under one
|
|
* or more contributor license agreements. See the NOTICE file
|
|
* distributed with this work for additional information
|
|
* regarding copyright ownership. The ASF licenses this file
|
|
* to you under the Apache License, Version 2.0 (the
|
|
* "License"); you may not use this file except in compliance
|
|
* with the License. You may obtain a copy of the License at
|
|
*
|
|
* http://www.apache.org/licenses/LICENSE-2.0
|
|
*
|
|
* Unless required by applicable law or agreed to in writing, software
|
|
* distributed under the License is distributed on an "AS IS" BASIS,
|
|
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
* See the License for the specific language governing permissions and
|
|
* limitations under the License.
|
|
*/
|
|
|
|
/*
|
|
* Copyright (C) 2016-present ScyllaDB
|
|
*
|
|
* Modified by 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 <chrono>
|
|
#include "cql3/statements/prepared_statement.hh"
|
|
#include "tracing/trace_state.hh"
|
|
#include "tracing/trace_keyspace_helper.hh"
|
|
#include "service/storage_proxy.hh"
|
|
#include "to_string.hh"
|
|
#include "timestamp.hh"
|
|
|
|
#include "cql3/values.hh"
|
|
#include "cql3/query_options.hh"
|
|
#include "utils/UUID_gen.hh"
|
|
|
|
namespace tracing {
|
|
|
|
logging::logger trace_state_logger("trace_state");
|
|
|
|
struct trace_state::params_values {
|
|
std::optional<inet_address_vector_replica_set> batchlog_endpoints;
|
|
std::optional<api::timestamp_type> user_timestamp;
|
|
std::vector<sstring> queries;
|
|
std::optional<db::consistency_level> cl;
|
|
std::optional<db::consistency_level> serial_cl;
|
|
std::optional<int32_t> page_size;
|
|
std::vector<prepared_checked_weak_ptr> prepared_statements;
|
|
std::vector<std::optional<std::vector<sstring_view>>> query_option_names;
|
|
std::vector<std::vector<cql3::raw_value_view>> query_option_values;
|
|
};
|
|
|
|
trace_state::params_values* trace_state::params_ptr::get_ptr_safe() {
|
|
if (!_vals) {
|
|
_vals = std::unique_ptr<params_values, params_values_deleter>(new params_values, params_values_deleter());
|
|
}
|
|
return _vals.get();
|
|
}
|
|
|
|
void trace_state::params_values_deleter::operator()(params_values* pv) {
|
|
delete pv;
|
|
}
|
|
|
|
void trace_state::init_session_records(
|
|
trace_type type,
|
|
std::chrono::seconds slow_query_ttl,
|
|
std::optional<utils::UUID> session_id,
|
|
span_id parent_id) {
|
|
|
|
_records = make_lw_shared<one_session_records>();
|
|
_records->session_id = session_id ? *session_id : utils::UUID_gen::get_time_UUID();
|
|
|
|
if (full_tracing()) {
|
|
if (!log_slow_query()) {
|
|
_records->ttl = ttl_by_type(type);
|
|
} else {
|
|
_records->ttl = std::max(ttl_by_type(type), slow_query_ttl);
|
|
}
|
|
} else {
|
|
_records->ttl = slow_query_ttl;
|
|
}
|
|
|
|
_records->session_rec.command = type;
|
|
_records->session_rec.slow_query_record_ttl = slow_query_ttl;
|
|
_records->my_span_id = span_id::make_span_id();
|
|
_records->parent_id = parent_id;
|
|
}
|
|
|
|
|
|
void trace_state::set_batchlog_endpoints(const inet_address_vector_replica_set& val) {
|
|
_params_ptr->batchlog_endpoints.emplace(val);
|
|
}
|
|
|
|
void trace_state::set_consistency_level(db::consistency_level val) {
|
|
_params_ptr->cl.emplace(val);
|
|
}
|
|
|
|
void trace_state::set_optional_serial_consistency_level(const std::optional<db::consistency_level>& val) {
|
|
if (val) {
|
|
_params_ptr->serial_cl.emplace(*val);
|
|
}
|
|
}
|
|
|
|
void trace_state::set_page_size(int32_t val) {
|
|
if (val > 0) {
|
|
_params_ptr->page_size.emplace(val);
|
|
}
|
|
}
|
|
|
|
void trace_state::set_request_size(size_t s) noexcept {
|
|
_records->session_rec.request_size = s;
|
|
}
|
|
|
|
void trace_state::set_response_size(size_t s) noexcept {
|
|
_records->session_rec.response_size = s;
|
|
}
|
|
|
|
void trace_state::add_query(sstring_view val) {
|
|
_params_ptr->queries.emplace_back(std::move(val));
|
|
}
|
|
|
|
void trace_state::add_session_param(sstring_view key, sstring_view val) {
|
|
_records->session_rec.parameters.emplace(std::move(key), std::move(val));
|
|
}
|
|
|
|
void trace_state::set_user_timestamp(api::timestamp_type val) {
|
|
_params_ptr->user_timestamp.emplace(val);
|
|
}
|
|
|
|
void trace_state::add_prepared_statement(prepared_checked_weak_ptr& prepared) {
|
|
_params_ptr->prepared_statements.emplace_back(prepared->checked_weak_from_this());
|
|
}
|
|
|
|
void trace_state::add_prepared_query_options(const cql3::query_options& prepared_options_ptr) {
|
|
if (_params_ptr->prepared_statements.empty()) {
|
|
throw std::logic_error("Tracing a prepared statement but no prepared statement is stored");
|
|
}
|
|
|
|
_params_ptr->query_option_names.reserve(_params_ptr->prepared_statements.size());
|
|
_params_ptr->query_option_values.reserve(_params_ptr->prepared_statements.size());
|
|
|
|
for (size_t i = 0; i < _params_ptr->prepared_statements.size(); ++i) {
|
|
_params_ptr->query_option_names.emplace_back(prepared_options_ptr.for_statement(i).get_names());
|
|
_params_ptr->query_option_values.emplace_back(prepared_options_ptr.for_statement(i).get_values());
|
|
}
|
|
}
|
|
|
|
void trace_state::build_parameters_map() {
|
|
if (!_params_ptr) {
|
|
return;
|
|
}
|
|
|
|
auto& params_map = _records->session_rec.parameters;
|
|
params_values& vals = *_params_ptr;
|
|
|
|
if (vals.batchlog_endpoints) {
|
|
params_map.emplace("batch_endpoints", join(sstring(","), *vals.batchlog_endpoints | boost::adaptors::transformed([](gms::inet_address ep) {return seastar::format("/{}", ep);})));
|
|
}
|
|
|
|
if (vals.cl) {
|
|
params_map.emplace("consistency_level", seastar::format("{}", *vals.cl));
|
|
}
|
|
|
|
if (vals.serial_cl) {
|
|
params_map.emplace("serial_consistency_level", seastar::format("{}", *vals.serial_cl));
|
|
}
|
|
|
|
if (vals.page_size) {
|
|
params_map.emplace("page_size", seastar::format("{:d}", *vals.page_size));
|
|
}
|
|
|
|
auto& queries = vals.queries;
|
|
if (!queries.empty()) {
|
|
if (queries.size() == 1) {
|
|
params_map.emplace("query", queries[0]);
|
|
} else {
|
|
// BATCH
|
|
for (size_t i = 0; i < queries.size(); ++i) {
|
|
params_map.emplace(format("query[{:d}]", i), queries[i]);
|
|
}
|
|
}
|
|
}
|
|
|
|
if (vals.user_timestamp) {
|
|
params_map.emplace("user_timestamp", seastar::format("{:d}", *vals.user_timestamp));
|
|
}
|
|
|
|
auto& prepared_statements = vals.prepared_statements;
|
|
|
|
if (!prepared_statements.empty()) {
|
|
// Parameter's key in the map will be "param[X]" for a single query CQL command and "param[Y][X] for a multiple
|
|
// queries CQL command, where X is an index of the parameter in a corresponding query and Y is an index of the
|
|
// corresponding query in the BATCH.
|
|
if (prepared_statements.size() == 1) {
|
|
build_parameters_map_for_one_prepared(prepared_statements[0], vals.query_option_names[0], vals.query_option_values[0], "param");
|
|
} else {
|
|
// BATCH
|
|
for (size_t i = 0; i < prepared_statements.size(); ++i) {
|
|
build_parameters_map_for_one_prepared(prepared_statements[i], vals.query_option_names[i], vals.query_option_values[i], format("param[{:d}]", i));
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
void trace_state::build_parameters_map_for_one_prepared(const prepared_checked_weak_ptr& prepared_ptr,
|
|
std::optional<std::vector<sstring_view>>& names_opt,
|
|
std::vector<cql3::raw_value_view>& values, const sstring& param_name_prefix) {
|
|
auto& params_map = _records->session_rec.parameters;
|
|
size_t i = 0;
|
|
|
|
// Trace parameters native values representations only if the current prepared statement has not been evicted from the cache by the time we got here.
|
|
// Such an eviction is a very unlikely event, however if it happens, since we are unable to recover their types, trace raw representations of the values.
|
|
|
|
if (names_opt) {
|
|
if (names_opt->size() != values.size()) {
|
|
throw std::logic_error(format("Number of \"names\" ({}) doesn't match the number of positional variables ({})", names_opt->size(), values.size()).c_str());
|
|
}
|
|
|
|
auto& names = names_opt.value();
|
|
for (; i < values.size(); ++i) {
|
|
params_map.emplace(format("{}[{:d}]({})", param_name_prefix, i, names[i]), raw_value_to_sstring(values[i], prepared_ptr ? prepared_ptr->bound_names[i]->type : nullptr));
|
|
}
|
|
} else {
|
|
for (; i < values.size(); ++i) {
|
|
params_map.emplace(format("{}[{:d}]", param_name_prefix, i), raw_value_to_sstring(values[i], prepared_ptr ? prepared_ptr->bound_names[i]->type : nullptr));
|
|
}
|
|
}
|
|
}
|
|
|
|
trace_state::~trace_state() {
|
|
if (!is_primary() && is_in_state(state::background)) {
|
|
trace_state_logger.error("Secondary session is in a background state! session_id: {}", session_id());
|
|
}
|
|
|
|
stop_foreground_and_write();
|
|
_local_tracing_ptr->end_session();
|
|
|
|
trace_state_logger.trace("{}: destructing", session_id());
|
|
}
|
|
|
|
void trace_state::stop_foreground_and_write() noexcept {
|
|
// Do nothing if state hasn't been initiated
|
|
if (is_in_state(state::inactive)) {
|
|
return;
|
|
}
|
|
|
|
if (is_in_state(state::foreground)) {
|
|
auto e = elapsed();
|
|
_records->do_log_slow_query = should_log_slow_query(e);
|
|
|
|
if (is_primary()) {
|
|
// We don't account the session_record event when checking a limit
|
|
// of maximum events per session because there may be only one such
|
|
// event and we don't want to cripple the primary session by
|
|
// "stealing" one trace() event from it.
|
|
//
|
|
// We do want to account them however. If for instance there are a
|
|
// lot of tracing sessions that only open itself and then do nothing
|
|
// - they will create a lot of session_record events and we do want
|
|
// to handle this case properly.
|
|
_records->consume_from_budget();
|
|
|
|
_records->session_rec.elapsed = e;
|
|
|
|
// build_parameters_map() may throw. We don't want to record the
|
|
// session's record in this case since its data may be incomplete.
|
|
// These events should be really rare however, therefore we don't
|
|
// want to optimize this flow (e.g. rollback the corresponding
|
|
// events' records that have already been sent to I/O).
|
|
if (should_write_records()) {
|
|
try {
|
|
build_parameters_map();
|
|
} catch (...) {
|
|
// Bump up an error counter, drop any pending records and
|
|
// continue
|
|
++_local_tracing_ptr->stats.trace_errors;
|
|
_records->drop_records();
|
|
}
|
|
}
|
|
}
|
|
|
|
set_state(state::background);
|
|
}
|
|
|
|
trace_state_logger.trace("{}: Current records count is {}", session_id(), _records->size());
|
|
|
|
if (should_write_records()) {
|
|
_local_tracing_ptr->write_session_records(_records, write_on_close());
|
|
} else {
|
|
_records->drop_records();
|
|
}
|
|
}
|
|
|
|
sstring trace_state::raw_value_to_sstring(const cql3::raw_value_view& v, const data_type& t) {
|
|
static constexpr int max_val_bytes = 64;
|
|
|
|
if (v.is_null()) {
|
|
return "null";
|
|
} else if (v.is_unset_value()) {
|
|
return "unset value";
|
|
} else {
|
|
return v.with_linearized([&] (bytes_view val) {
|
|
sstring str_rep;
|
|
|
|
if (t) {
|
|
str_rep = t->to_string(to_bytes(val));
|
|
} else {
|
|
trace_state_logger.trace("{}: data types are unavailable - tracing a raw value", session_id());
|
|
str_rep = to_hex(val);
|
|
}
|
|
|
|
if (str_rep.size() > max_val_bytes) {
|
|
return format("{}...", str_rep.substr(0, max_val_bytes));
|
|
} else {
|
|
return str_rep;
|
|
}
|
|
});
|
|
}
|
|
}
|
|
}
|