2 // The rpcc class handles client-side RPC. Each rpcc is bound to a single RPC
3 // server. The jobs of rpcc include maintaining a connection to server, sending
4 // RPC requests and waiting for responses, retransmissions, at-most-once delivery
7 // The rpcs class handles the server side of RPC. Each rpcs handles multiple
8 // connections from different rpcc objects. The jobs of rpcs include accepting
9 // connections, dispatching requests to registered RPC handlers, at-most-once
12 // Both rpcc and rpcs use the connection class as an abstraction for the
13 // underlying communication channel. To send an RPC request/reply, one calls
14 // connection::send() which blocks until data is sent or the connection has
15 // failed (thus the caller can free the buffer when send() returns). When a
16 // request/reply is received, connection makes a callback into the corresponding
17 // rpcc or rpcs (see rpcc::got_pdu() and rpcs::got_pdu()).
19 // Thread organization:
20 // rpcc uses application threads to send RPC requests and blocks to receive the
21 // reply or error. All connections use a single PollMgr object to perform async
22 // socket IO. PollMgr creates a single thread to examine the readiness of socket
23 // file descriptors and informs the corresponding connection whenever a socket is
24 // ready to be read or written. (We use asynchronous socket IO to reduce the
25 // number of threads needed to manage these connections; without async IO, at
26 // least one thread is needed per connection to read data without blocking other
27 // activities.) Each rpcs object creates one thread for listening on the server
28 // port and a pool of threads for executing RPC requests. The thread pool allows
29 // us to control the number of threads spawned at the server (spawning one thread
30 // per request will hurt when the server faces thousands of requests).
32 // In order to delete a connection object, we must maintain a reference count.
33 // For rpcc, multiple client threads might be invoking the rpcc::call() functions
34 // and thus holding multiple references to the underlying connection object. For
35 // rpcs, multiple dispatch threads might be holding references to the same
36 // connection object. A connection object is deleted only when the underlying
37 // connection is dead and the reference count reaches zero.
39 // This version of the RPC library explicitly joins exited threads to make sure
40 // no outstanding references exist before deleting objects.
42 // To delete a rpcc object safely, the users of the library must ensure that
43 // there are no outstanding calls on the rpcc object.
45 // To delete a rpcs object safely, we do the following in sequence: 1. stop
46 // accepting new incoming connections. 2. close existing active connections. 3.
47 // delete the dispatch thread pool which involves waiting for current active RPC
48 // handlers to finish. It is interesting how a thread pool can be deleted
49 // without using thread cancellation. The trick is to inject x "poison pills" for
50 // a thread pool of x threads. Upon getting a poison pill instead of a normal
51 // task, a worker thread will exit (and thread pool destructor waits to join all
52 // x exited worker threads).
57 #include <arpa/inet.h>
58 #include <netinet/tcp.h>
63 inline void set_rand_seed() {
64 auto now = time_point_cast<nanoseconds>(steady_clock::now());
65 srandom((uint32_t)now.time_since_epoch().count()^(uint32_t)getpid());
68 static sockaddr_in make_sockaddr(const string &hostandport);
70 rpcc::rpcc(const string & d, bool retrans) :
71 dst_(make_sockaddr(d)), srv_nonce_(0), bind_done_(false), xid_(1), lossytest_(0),
72 retrans_(retrans), reachable_(true), chan_(), destroy_wait_ (false), xid_rep_done_(-1)
76 clt_nonce_ = (nonce_t)random();
78 // special client nonce 0 means this client does not
79 // require at-most-once logic from the server
80 // because it uses tcp and never retries a failed connection
84 char *loss_env = getenv("RPC_LOSSY");
86 lossytest_ = atoi(loss_env);
88 // xid starts with 1 and latest received reply starts with 0
89 xid_rep_window_.push_back(0);
91 IF_LEVEL(2) LOG("cltn_nonce is " << clt_nonce_ << " lossy " << lossytest_);
94 // IMPORTANT: destruction should happen only when no external threads
95 // are blocked inside rpcc or will use rpcc in the future
98 IF_LEVEL(2) LOG("delete nonce " << clt_nonce_ << " chan " << (chan_?(int)chan_->fd:-1));
100 VERIFY(calls_.size() == 0);
103 int rpcc::bind(milliseconds to) {
105 int ret = call_timeout(rpc_protocol::bind, to, r);
111 IF_LEVEL(2) LOG("bind " << inet_ntoa(dst_.sin_addr) << " failed " << ret);
116 // Cancel all outstanding calls
117 void rpcc::cancel(void) {
120 LOG("force callers to fail");
121 for (auto &p : calls_) {
122 caller *ca = p.second;
124 IF_LEVEL(2) LOG("force caller to fail");
128 ca->intret = rpc_protocol::cancel_failure;
132 destroy_wait_ = true;
133 while (calls_.size () > 0)
134 destroy_wait_c_.wait(ml);
140 int rpcc::call1(proc_id_t proc, marshall &req, string &rep, milliseconds to) {
147 if ((proc != rpc_protocol::bind.id && !bind_done_) || (proc == rpc_protocol::bind.id && bind_done_)) {
148 IF_LEVEL(1) LOG("rpcc has not been bound to dst or binding twice");
149 return rpc_protocol::bind_failure;
153 return rpc_protocol::cancel_failure;
156 calls_[ca.xid] = &ca;
158 req.pack_header(rpc_protocol::request_header{
159 ca.xid, proc, clt_nonce_, srv_nonce_, xid_rep_window_.front()
161 xid_rep = xid_rep_window_.front();
164 milliseconds curr_to = rpc::to_min;
165 auto finaldeadline = steady_clock::now() + to;
167 bool transmit = true;
168 shared_ptr<connection> ch;
178 if (dup_req_.isvalid() && xid_rep_done_ > dup_req_.xid) {
183 if (forgot.isvalid())
184 ch->send(forgot.buf);
187 else IF_LEVEL(1) LOG("not reachable");
188 IF_LEVEL(2) LOG(clt_nonce_ << " just sent req proc " << hex << proc <<
189 " xid " << dec << ca.xid << " clt_nonce " << clt_nonce_);
191 transmit = false; // only send once on a given channel
194 auto nextdeadline = min(steady_clock::now() + curr_to, finaldeadline);
200 IF_LEVEL(2) LOG("wait");
201 if (ca.c.wait_until(cal, nextdeadline) == cv_status::timeout) {
202 IF_LEVEL(2) LOG("timeout");
207 IF_LEVEL(2) LOG("reply received");
212 if (nextdeadline >= finaldeadline)
215 if (retrans_ && (!ch || ch->isdead())) {
216 // since connection is dead, retransmit
217 // on the new connection
223 // no locking of ca.m since only this thread changes ca.xid
225 calls_.erase(ca.xid);
226 // may need to update the xid again here, in case the
227 // packet times out before it's even sent by the channel.
228 // I don't think there's any harm in maybe doing it twice
229 update_xid_rep(ca.xid);
232 destroy_wait_c_.notify_one();
235 if (ca.done && lossytest_)
238 if (!dup_req_.isvalid()) {
240 dup_req_.xid = ca.xid;
242 if (xid_rep > xid_rep_done_)
243 xid_rep_done_ = xid_rep;
248 IF_LEVEL(2) LOG(clt_nonce_ << " call done for req proc " << hex << proc <<
249 " xid " << dec << ca.xid << " " << inet_ntoa(dst_.sin_addr) << ":" <<
250 ntoh(dst_.sin_port) << " done? " << ca.done << " ret " << ca.intret);
252 // destruction of req automatically frees its buffer
253 return (ca.done? ca.intret : rpc_protocol::timeout_failure);
256 void rpcc::get_refconn(shared_ptr<connection> & ch) {
258 if (!chan_ || chan_->isdead())
259 chan_ = connection::to_dst(dst_, this, lossytest_);
265 // PollMgr's thread is being used to
266 // make this upcall from connection object to rpcc.
267 // this funtion must not block.
269 // this function keeps no reference for connection *c
271 rpcc::got_pdu(const shared_ptr<connection> &, const string & b)
273 unmarshall rep(b, true);
274 rpc_protocol::reply_header h;
275 rep.unpack_header(h);
278 IF_LEVEL(1) LOG("unmarshall header failed!!!");
284 update_xid_rep(h.xid);
286 if (calls_.find(h.xid) == calls_.end()) {
287 IF_LEVEL(2) LOG("xid " << h.xid << " no pending request");
290 caller *ca = calls_[h.xid];
296 if (ca->intret < 0) {
297 IF_LEVEL(2) LOG("RPC reply error for xid " << h.xid << " intret " << ca->intret);
305 // assumes thread holds mutex m
307 rpcc::update_xid_rep(int xid)
309 if (xid <= xid_rep_window_.front())
312 for (auto it = xid_rep_window_.begin(); it != xid_rep_window_.end(); it++) {
314 xid_rep_window_.insert(it, xid);
318 xid_rep_window_.push_back(xid);
321 auto it = xid_rep_window_.begin();
322 for (it++; it != xid_rep_window_.end(); it++) {
323 while (xid_rep_window_.front() + 1 == *it)
324 xid_rep_window_.pop_front();
328 rpcs::rpcs(in_port_t p1)
329 : port_(p1), reachable_ (true)
332 nonce_ = (nonce_t)random();
333 IF_LEVEL(2) LOG("created with nonce " << nonce_);
335 reg(rpc_protocol::bind, &rpcs::rpcbind, this);
336 dispatchpool_ = unique_ptr<thread_pool>(new thread_pool(6, false));
340 char *loss_env = getenv("RPC_LOSSY");
341 listener_.reset(new connection_listener(this, port_, loss_env ? atoi(loss_env) : 0));
345 // must delete listener before dispatchpool
347 dispatchpool_ = nullptr;
351 bool rpcs::got_pdu(const shared_ptr<connection> & c, const string & b) {
353 IF_LEVEL(1) LOG("not reachable");
357 return dispatchpool_->addJob(std::bind(&rpcs::dispatch, this, c, b));
360 void rpcs::reg1(proc_id_t proc, handler *h) {
362 VERIFY(procs_.count(proc) == 0);
364 VERIFY(procs_.count(proc) >= 1);
367 void rpcs::dispatch(shared_ptr<connection> c, const string & buf) {
368 unmarshall req(buf, true);
370 rpc_protocol::request_header h;
371 req.unpack_header(h);
372 proc_id_t proc = h.proc;
375 IF_LEVEL(1) LOG("unmarshall header failed");
379 IF_LEVEL(2) LOG("rpc " << h.xid << " (proc " << hex << proc << ", last_rep " <<
380 dec << h.xid_rep << ") from clt " << h.clt_nonce << " for srv instance " << h.srv_nonce);
383 rpc_protocol::reply_header rh{h.xid,0};
385 // is client sending to an old instance of server?
386 if (h.srv_nonce != 0 && h.srv_nonce != nonce_) {
387 IF_LEVEL(2) LOG("rpc for an old server instance " << h.srv_nonce <<
388 " (current " << nonce_ << ") proc " << hex << h.proc);
389 rh.ret = rpc_protocol::oldsrv_failure;
396 // is RPC proc a registered procedure?
399 if (procs_.count(proc) < 1) {
400 LOG("unknown proc 0x" << hex << proc << " with h.srv_nonce=" << h.srv_nonce << ", my srv_nonce=" << nonce_);
408 rpcs::rpcstate_t stat;
412 // have i seen this client before?
414 lock rwl(reply_window_m_);
415 // if we don't know about this clt_nonce, create a cleanup object
416 if (reply_window_.find(h.clt_nonce) == reply_window_.end()) {
417 VERIFY (reply_window_[h.clt_nonce].size() == 0); // create
418 reply_window_[h.clt_nonce].push_back(reply_t(-1)); // store starting reply xid
419 IF_LEVEL(2) LOG("new client " << h.clt_nonce << " xid " << h.xid <<
420 " chan " << c->fd << ", total clients " << (reply_window_.size()-1));
424 // save the latest good connection to the client
427 if (conns_.find(h.clt_nonce) == conns_.end())
428 conns_[h.clt_nonce] = c;
429 else if (conns_[h.clt_nonce]->create_time < c->create_time)
430 conns_[h.clt_nonce] = c;
433 stat = checkduplicate_and_update(h.clt_nonce, h.xid, h.xid_rep, b1);
435 // this client does not require at most once logic
440 case NEW: // new request
441 rh.ret = (*f)(req, rep);
442 if (rh.ret == rpc_protocol::unmarshal_args_failure) {
443 LOG("failed to unmarshall the arguments. You are " <<
444 "probably calling RPC 0x" << hex << proc << " with the wrong " <<
445 "types of arguments.");
453 IF_LEVEL(2) LOG("sending and saving reply of size " << b1.size() << " for rpc " <<
454 h.xid << ", proc " << hex << proc << " ret " << dec << rh.ret << ", clt " << h.clt_nonce);
456 if (h.clt_nonce > 0) {
457 // only record replies for clients that require at-most-once logic
458 add_reply(h.clt_nonce, h.xid, b1);
461 // get the latest connection to the client
465 c = conns_[h.clt_nonce];
470 case INPROGRESS: // server is working on this request
472 case DONE: // duplicate and we still have the response
475 case FORGOTTEN: // very old request and we don't have the response anymore
476 IF_LEVEL(2) LOG("very old request " << h.xid << " from " << h.clt_nonce);
477 rh.ret = rpc_protocol::atmostonce_failure;
484 // rpcs::dispatch calls this when an RPC request arrives.
486 // checks to see if an RPC with xid from clt_nonce has already been received.
487 // if not, remembers the request in reply_window_.
489 // deletes remembered requests with XIDs <= xid_rep; the client
490 // says it has received a reply for every RPC up through xid_rep.
491 // frees the reply_t::buf of each such request.
494 // NEW: never seen this xid before.
495 // INPROGRESS: seen this xid, and still processing it.
496 // DONE: seen this xid, previous reply returned in b.
497 // FORGOTTEN: might have seen this xid, but deleted previous reply.
499 rpcs::checkduplicate_and_update(nonce_t clt_nonce, xid_t xid,
500 xid_t xid_rep, string & b)
502 lock rwl(reply_window_m_);
504 list<reply_t> &l = reply_window_[clt_nonce];
506 VERIFY(l.size() > 0);
507 VERIFY(xid >= xid_rep);
509 xid_t past_xid_rep = l.begin()->xid;
511 list<reply_t>::iterator start = l.begin(), it = ++start;
513 if (past_xid_rep < xid_rep || past_xid_rep == -1) {
514 // scan for deletion candidates
515 while (it != l.end() && it->xid < xid_rep)
518 l.begin()->xid = xid_rep;
521 if (xid < past_xid_rep && past_xid_rep != -1)
524 // skip non-deletion candidates
525 while (it != l.end() && it->xid < xid)
528 // if it's in the list it must be right here
529 if (it != l.end() && it->xid == xid) {
530 if (it->cb_present) {
531 // return information about the remembered reply
537 // remember that a new request has arrived
538 l.insert(it, reply_t(xid));
543 // rpcs::dispatch calls add_reply when it is sending a reply to an RPC,
544 // and passes the return value in b.
545 // add_reply() should remember b.
546 // free_reply_window() and checkduplicate_and_update are responsible for
547 // cleaning up the remembered values.
548 void rpcs::add_reply(nonce_t clt_nonce, xid_t xid, const string & b) {
549 lock rwl(reply_window_m_);
550 // remember the RPC reply value
551 list<reply_t> &l = reply_window_[clt_nonce];
552 list<reply_t>::iterator it = l.begin();
553 // skip to our place in the list
554 for (it++; it != l.end() && it->xid < xid; it++);
555 // there should already be an entry, so whine if there isn't
556 if (it == l.end() || it->xid != xid) {
557 LOG("Could not find reply struct in add_reply");
558 l.insert(it, reply_t(xid, b));
560 *it = reply_t(xid, b);
564 void rpcs::free_reply_window(void) {
565 lock rwl(reply_window_m_);
566 reply_window_.clear();
569 int rpcs::rpcbind(nonce_t &r) {
570 IF_LEVEL(2) LOG("called return nonce " << nonce_);
575 static sockaddr_in make_sockaddr(const string &hostandport) {
576 string host = "127.0.0.1";
577 string port = hostandport;
578 auto colon = hostandport.find(':');
579 if (colon != string::npos) {
580 host = hostandport.substr(0, colon);
581 port = hostandport.substr(colon+1);
584 sockaddr_in dst{}; // zero initialize
585 dst.sin_family = AF_INET;
587 struct in_addr a{inet_addr(host.c_str())};
589 if (a.s_addr != INADDR_NONE)
590 dst.sin_addr.s_addr = a.s_addr;
592 struct hostent *hp = gethostbyname(host.c_str());
594 if (!hp || hp->h_length != 4 || hp->h_addrtype != AF_INET) {
595 LOG_NONMEMBER("cannot find host name " << host);
598 memcpy(&a, hp->h_addr_list[0], sizeof(in_addr_t));
599 dst.sin_addr.s_addr = a.s_addr;
601 dst.sin_port = hton((in_port_t)stoi(port));