设计点
- 在redis层使用函数作为模板类型参数,也就是command层定义的函数。
template <typename Cmd, typename ...Args>
auto Redis::command(Cmd cmd, Args &&...args)
-> typename std::enable_if<!std::is_convertible<Cmd, StringView>::value, ReplyUPtr>::type {
if (_connection) {
// Single Connection Mode.
// TODO: In this case, should we reconnect?
auto &connection = _connection->connection();
if (connection.broken()) {
throw Error("Connection is broken");
}
return _command(connection, cmd, std::forward<Args>(args)...);
} else {
assert(_pool);
// Pool Mode, i.e. get connection from pool.
SafeConnection connection(*_pool);
return _command(connection.connection(), cmd, std::forward<Args>(args)...);
}
}
template <typename ...Args>
auto Redis::command(const StringView &cmd_name, Args &&...args)
-> typename std::enable_if<!IsIter<typename LastType<Args...>::type>::value, ReplyUPtr>::type {
auto cmd = [](Connection &connection, const StringView &name, Args &&...params) {
CmdArgs cmd_args;
cmd_args.append(name, std::forward<Args>(params)...);
connection.send(cmd_args);
};
return command(cmd, cmd_name, std::forward<Args>(args)...);
}
template <typename Input>
auto Redis::command(Input first, Input last)
-> typename std::enable_if<IsIter<Input>::value, ReplyUPtr>::type {
range_check("command", first, last);
auto cmd = [](Connection &connection, Input start, Input stop) {
CmdArgs cmd_args;
while (start != stop) {
cmd_args.append(*start);
++start;
}
connection.send(cmd_args);
};
return command(cmd, first, last);
}
- 使用可变参数模板
template <typename Cmd, typename ...Args>
ReplyUPtr _command(Connection &connection, Cmd cmd, Args &&...args);
//command系列
template <typename Cmd, typename ...Args>
auto command(Cmd cmd, Args &&...args)
-> typename std::enable_if<!std::is_convertible<Cmd, StringView>::value, ReplyUPtr>::type;
- 在redis层最终会调用不带连接的command方法,调用带连接参数的_command方法(调用command层具体的函数,然后接收响应)
template <typename Cmd, typename ...Args>
ReplyUPtr Redis::_command(Connection &connection, Cmd cmd, Args &&...args) {
assert(!connection.broken());
cmd(connection, std::forward<Args>(args)...);
auto reply = connection.recv();
return reply;
}
redis与连接
Redis处理命令
connection主要方法及与reply关系
connection只支持移动语义,不支持拷贝和赋值
recv使用ReplyUPtr,即unique_ptr<redisReply, ReplyDeleter>,其中ReplyDeleter定义如下
struct ReplyDeleter {
void operator()(redisReply *reply) const {
if (reply != nullptr) {
freeReplyObject(reply);
}
}
};
其调用redisGetReply获取响应,当中会处理以下几种异常情况
- redisGetReply返回不是REDIS_OK,抛出异常
- 调用broken判断连接是否断了
- 在handle_error_reply标识为true,并且响应的type值为REDIS_REPLY_ERROR,抛出异常
send方法主要是调用redisAppendCommandArgv,会处理下面情况
- redisAppendCommandArgv如果失败,会抛出异常
- 调用broken()看连接是否断了
bool broken() const noexcept {
return !_ctx || _ctx->err != REDIS_OK;
}
连接池及其管理
SafeConnection:包含连接池的引用,构造函数负责从连接池中得到可用的连接,析构函数中将连接还到连接池中
GuardedConnection:使用ConnectionPoolSPtr _pool指向连接池的指针
连接池分配连接是通过ConnectionPool::fetch
- 在池子中没有连接时,创建连接,并且连接使用计数+1
- 池子中有时,从池子中取
Connection ConnectionPool::fetch() {
std::unique_lock<std::mutex> lock(_mutex);
auto connection = _fetch(lock);
auto connection_lifetime = _pool_opts.connection_lifetime;
auto connection_idle_time = _pool_opts.connection_idle_time;
if (_sentinel) {
auto opts = _opts;
auto role_changed = _role_changed(connection.options());
auto sentinel = _sentinel;
lock.unlock();
if (role_changed || _need_reconnect(connection, connection_lifetime, connection_idle_time)) {
try {
connection = _create(sentinel, opts);
} catch (const Error &) {
// Failed to reconnect, return it to the pool, and retry latter.
release(std::move(connection));
throw;
}
}
return connection;
}
lock.unlock();
if (_need_reconnect(connection, connection_lifetime, connection_idle_time)) {
try {
connection.reconnect();
} catch (const Error &) {
// Failed to reconnect, return it to the pool, and retry latter.
release(std::move(connection));
throw;
}
}
return connection;
}
Connection ConnectionPool::_fetch(std::unique_lock<std::mutex> &lock) {
if (_pool.empty()) {
if (_used_connections == _pool_opts.size) {
_wait_for_connection(lock);
} else {
++_used_connections;
// Lazily create a new (broken) connection to avoid connecting with lock.
return Connection(_opts, Connection::Dummy{});
}
}
// _pool is NOT empty.
return _fetch();
}
Connection ConnectionPool::_fetch() {
assert(!_pool.empty());
auto connection = std::move(_pool.front());
_pool.pop_front();
return connection;
}
连接释放到连接池通过ConnectionPool::release
void ConnectionPool::release(Connection connection) {
{
std::lock_guard<std::mutex> lock(_mutex);
_pool.push_back(std::move(connection));
}
_cv.notify_one();
}
解析响应
主要在reply.cpp和reply.h文件中
通过模板函数来作具体类型的的解析转发
template <typename T>
inline T parse(redisReply &reply) {
return parse(ParseTag<T>(), reply);
}
解析tag
template <typename T>
struct ParseTag {};
optional解析
template <typename T>
Optional<T> parse(ParseTag<Optional<T>>, redisReply &reply) {
if (reply::is_nil(reply)) {
// Because of a GCC bug, we cannot return {} for -std=c++17
// Refer to https://gcc.gnu.org/bugzilla/show_bug.cgi?id=86465
#if defined REDIS_PLUS_PLUS_HAS_OPTIONAL
return std::nullopt;
#else
return {};
#endif
}
return Optional<T>(parse<T>(reply));
}
字符串解析
std::string parse(ParseTag<std::string>, redisReply &reply) {
#ifdef REDIS_PLUS_PLUS_RESP_VERSION_3
if (!reply::is_string(reply) && !reply::is_status(reply)
&& !reply::is_verb(reply) && !reply::is_bignum(reply)) {
throw ParseError("STRING or STATUS or VERB or BIGNUM", reply);
}
#else
if (!reply::is_string(reply) && !reply::is_status(reply)) {
throw ParseError("STRING or STATUS", reply);
}
#endif
if (reply.str == nullptr) {
throw ProtoError("A null string reply");
}
// Old version hiredis' *redisReply::len* is of type int.
// So we CANNOT have something like: *return {reply.str, reply.len}*.
return std::string(reply.str, reply.len);
}
解析void
主要是解析set命令类的
- 看type是不是REDIS_REPLY_STATUS
- str是否等于OK
void parse(ParseTag<void>, redisReply &reply) {
if (!reply::is_status(reply)) {
throw ParseError("STATUS", reply);
}
if (reply.str == nullptr) {
throw ProtoError("A null status reply");
}
static const std::string OK = "OK";
// Old version hiredis' *redisReply::len* is of type int.
// So we have to cast it to an unsigned int.
if (static_cast<std::size_t>(reply.len) != OK.size()
|| OK.compare(0, OK.size(), reply.str, reply.len) != 0) {
throw ProtoError("NOT ok status reply: " + reply::to_status(reply));
}
}
long long类型解析
- 检查type是不是REDIS_REPLY_INTEGER
long long parse(ParseTag<long long>, redisReply &reply) {
if (!reply::is_integer(reply)) {
throw ParseError("INTEGER", reply);
}
return reply.integer;
}
double类型解析
- 会区分协议版本,如果是resp3,检查类型是否是REDIS_REPLY_DOUBLE,如果是,则取dval,否则作字符串处理,然后转成double类型
- 如果不是resp3,则字符串处理,转为double类型
double parse(ParseTag<double>, redisReply &reply) {
#ifdef REDIS_PLUS_PLUS_RESP_VERSION_3
if (is_double(reply)) {
return reply.dval;
} else {
// Return by string reply.
#endif
try {
return std::stod(parse<std::string>(reply));
} catch (const std::invalid_argument &) {
throw ProtoError("not a double reply");
} catch (const std::out_of_range &) {
throw ProtoError("double reply out of range");
}
#ifdef REDIS_PLUS_PLUS_RESP_VERSION_3
}
#endif
}
bool类型解析
- resp3协议时,判断类型是否是REDIS_REPLY_BOOL或者REDIS_REPLY_INTEGER
- resp2协议时,当作long long类型来解析
- 根据值是1,0来处理
bool parse(ParseTag<bool>, redisReply &reply) {
#ifdef REDIS_PLUS_PLUS_RESP_VERSION_3
long long ret = 0;
if (is_bool(reply) || is_integer(reply)) {
ret = reply.integer;
} else {
throw ProtoError("BOOL or INTEGER");
}
#else
auto ret = parse<long long>(reply);
#endif
if (ret == 1) {
return true;
} else if (ret == 0) {
return false;
} else {
throw ProtoError("Invalid bool reply: " + std::to_string(ret));
}
}
error体系
异步redis
AsyncEvent
CommandEvent
:主要是由io线程来设置Promise,用户线程通过Future来获取值
CallbackEvent
:Promise的设置以及Future的获取都是由io线程来调用,保证先调用Promise.setXXX,然后调用Future.get