fix: fix RPC_CHECK_STATUS incorrect usage in meta_service

[levy5307]

In meta_service, there are some rpc interfaces which are using rpc_holder. And they use RPC_CHECK_STATUS to do some check work. But it will cause the response to be sent repeatedly, for example:

1. A response will replied in RPC_CHECK_STATUS, and another response will be sent in the dstor of rpc_holder. So one response is sent repeatedly here.

2. If this node is not a master, a forward message will send in this macro, And this situation is same as above.

The implementation of RPC_CHECK_STATUS:

#define RPC_CHECK_STATUS(dsn_msg, response_struct)                                                 \
    dinfo("rpc %s called", __FUNCTION__);                                                          \
    int result = check_leader(dsn_msg, nullptr);                                                   \
    if (result == 0)                                                                               \
        return;                                                                                    \
    if (result == -1 || !_started) {                                                               \
        if (result == -1)                                                                          \
            response_struct.err = ERR_FORWARD_TO_OTHERS;                                           \
        else if (_recovering)                                                                      \
            response_struct.err = ERR_UNDER_RECOVERY;                                              \
        else                                                                                       \
            response_struct.err = ERR_SERVICE_NOT_ACTIVE;                                          \
        ddebug("reject request with %s", response_struct.err.to_string());                         \
        reply(dsn_msg, response_struct);                                                           \
        return;                                                                                    \
    }

[neverchanje]

This function name is not self-descriptive enough. Call it try_forward_if_not_leader

I highly recommend you add a unit test here, btw. It seems slightly difficult to mock the case where this meta is not the leader. But since it's necessary, you can add a fail_point to meta_server_failure_detector::get_leader

[levy5307]

I don't think try_forward_if_not_leader is a good name. Because the main purpose for this function is to do some check work, but not forward

[neverchanje]

I checked all the callers of check_status. None is using the parameter forward_address . It's useless I think.

Suggested change

	bool check_status(TRpcHolder rpc, /*out*/ rpc_address *forward_address = nullptr);
	bool check_status(TRpcHolder rpc);

[levy5307]

I will use it in the later pull request, which needs the forward_address.