2 // Replicated state machine implementation with a primary and several
3 // backups. The primary receives requests, assigns each a view stamp (a
4 // vid, and a sequence number) in the order of reception, and forwards
5 // them to all backups. A backup executes requests in the order that
6 // the primary stamps them and replies with an OK to the primary. The
7 // primary executes the request after it receives OKs from all backups,
8 // and sends the reply back to the client.
10 // The config module will tell the RSM about a new view. If the
11 // primary in the previous view is a member of the new view, then it
12 // will stay the primary. Otherwise, the smallest numbered node of
13 // the previous view will be the new primary. In either case, the new
14 // primary will be a node from the previous view. The configuration
15 // module constructs the sequence of views for the RSM and the RSM
16 // ensures there will be always one primary, who was a member of the
19 // When a new node starts, the recovery thread is in charge of joining
20 // the RSM. It will collect the internal RSM state from the primary;
21 // the primary asks the config module to add the new node and returns
22 // to the joining the internal RSM state (e.g., paxos log). Since
23 // there is only one primary, all joins happen in well-defined total
26 // The recovery thread also runs during a view change (e.g, when a node
27 // has failed). After a failure some of the backups could have
28 // processed a request that the primary has not, but those results are
29 // not visible to clients (since the primary responds). If the
30 // primary of the previous view is in the current view, then it will
31 // be the primary and its state is authoritive: the backups download
32 // from the primary the current state. A primary waits until all
33 // backups have downloaded the state. Once the RSM is in sync, the
34 // primary accepts requests again from clients. If one of the backups
35 // is the new primary, then its state is authoritative. In either
36 // scenario, the next view uses a node as primary that has the state
37 // resulting from processing all acknowledged client requests.
38 // Therefore, if the nodes sync up before processing the next request,
39 // the next view will have the correct state.
41 // While the RSM in a view change (i.e., a node has failed, a new view
42 // has been formed, but the sync hasn't completed), another failure
43 // could happen, which complicates a view change. During syncing the
44 // primary or backups can timeout, and initiate another Paxos round.
45 // There are 2 variables that RSM uses to keep track in what state it
47 // - inviewchange: a node has failed and the RSM is performing a view change
48 // - insync: this node is syncing its state
50 // If inviewchange is false and a node is the primary, then it can
51 // process client requests. If it is true, clients are told to retry
52 // later again. While inviewchange is true, the RSM may go through several
53 // member list changes, one by one. After a member list
54 // change completes, the nodes tries to sync. If the sync complets,
55 // the view change completes (and inviewchange is set to false). If
56 // the sync fails, the node may start another member list change
57 // (inviewchange = true and insync = false).
59 // The implementation should be used only with servers that run all
60 // requests run to completion; in particular, a request shouldn't
61 // block. If a request blocks, the backup won't respond to the
62 // primary, and the primary won't execute the request. A request may
63 // send an RPC to another host, but the RPC should be a one-way
64 // message to that host; the backup shouldn't do anything based on the
65 // response or execute after the response, because it is not
66 // guaranteed that all backup will receive the same response and
67 // execute in the same order.
69 // The implementation can be viewed as a layered system:
70 // RSM module ---- in charge of replication
71 // config module ---- in charge of view management
72 // Paxos module ---- in charge of running Paxos to agree on a value
74 // Each module has threads and internal locks. Furthermore, a thread
75 // may call down through the layers (e.g., to run Paxos's proposer).
76 // When Paxos's acceptor accepts a new value for an instance, a thread
77 // will invoke an upcall to inform higher layers of the new value.
78 // The rule is that a module releases its internal locks before it
79 // upcalls, but can keep its locks when calling down.
83 #include "rsm_client.h"
86 rsm::rsm(const string & _first, const string & _me) : primary(_first)
88 cfg = unique_ptr<config>(new config(_first, _me, this));
91 // Commit the first view here. We can not have acceptor::acceptor
92 // do the commit, since at that time this->cfg is not initialized
95 rsmrpc = cfg->get_rpcs();
96 rsmrpc->reg(rsm_client_protocol::invoke, &rsm::client_invoke, this);
97 rsmrpc->reg(rsm_client_protocol::members, &rsm::client_members, this);
98 rsmrpc->reg(rsm_protocol::invoke, &rsm::invoke, this);
99 rsmrpc->reg(rsm_protocol::transferreq, &rsm::transferreq, this);
100 rsmrpc->reg(rsm_protocol::transferdonereq, &rsm::transferdonereq, this);
101 rsmrpc->reg(rsm_protocol::joinreq, &rsm::joinreq, this);
103 // tester must be on different port, otherwise it may partition itself
104 testsvr.reset(new rpcs((in_port_t)stoi(_me) + 1));
105 testsvr->reg(rsm_test_protocol::net_repair, &rsm::test_net_repairreq, this);
106 testsvr->reg(rsm_test_protocol::breakpoint, &rsm::breakpointreq, this);
113 thread(&rsm::recovery, this).detach();
116 // The recovery thread runs this function
117 void rsm::recovery() {
122 while (!cfg->ismember(cfg->myaddr(), vid_commit)) {
123 // XXX iannucci 2013/09/15 -- I don't understand whether accessing
124 // cfg->view_id in this manner involves a race. I suspect not.
125 if (join(primary, ml)) {
127 commit_change(cfg->view_id(), ml);
130 this_thread::sleep_for(seconds(3)); // XXX make another node in cfg primary?
134 vid_insync = vid_commit;
135 LOG("sync vid_insync " << vid_insync);
136 if (primary == cfg->myaddr()) {
137 r = sync_with_backups(ml);
139 r = sync_with_primary(ml);
143 // If there was a commited viewchange during the synchronization, restart
145 if (vid_insync != vid_commit)
149 myvs.vid = vid_commit;
151 inviewchange = false;
153 LOG("go to sleep " << insync << " " << inviewchange);
154 recovery_cond.wait(ml);
158 bool rsm::sync_with_backups(lock & rsm_mutex_lock) {
159 rsm_mutex_lock.unlock();
161 // Make sure that the state of lock_server is stable during
162 // synchronization; otherwise, the primary's state may be more recent
163 // than replicas after the synchronization.
164 lock invoke_mutex_lock(invoke_mutex);
165 // By acquiring and releasing the invoke_mutex once, we make sure that
166 // the state of lock_server will not be changed until all
167 // replicas are synchronized. The reason is that client_invoke arrives
168 // after this point of time will see inviewchange == true, and returns
171 rsm_mutex_lock.lock();
172 // Start accepting synchronization request (statetransferreq) now!
174 cfg->get_view(vid_insync, backups);
175 backups.erase(find(backups.begin(), backups.end(), cfg->myaddr()));
176 LOG("backups " << backups);
177 sync_cond.wait(rsm_mutex_lock);
183 bool rsm::sync_with_primary(lock & rsm_mutex_lock) {
184 // Remember the primary of vid_insync
186 while (vid_insync == vid_commit) {
187 if (statetransfer(m, rsm_mutex_lock))
190 return statetransferdone(m, rsm_mutex_lock);
195 // Call to transfer state from m to the local node.
196 // Assumes that rsm_mutex is already held.
198 bool rsm::statetransfer(const string & m, lock & rsm_mutex_lock)
200 rsm_protocol::transferres r;
203 LOG("contact " << m << " w. my last_myvs(" << last_myvs.vid << "," << last_myvs.seqno << ")");
206 rsm_mutex_lock.unlock();
209 ret = cl->call_timeout(rsm_protocol::transferreq, milliseconds(100),
210 r, cfg->myaddr(), last_myvs, vid_insync);
212 rsm_mutex_lock.lock();
214 if (cl == 0 || ret != rsm_protocol::OK) {
215 LOG("couldn't reach " << m << " " << hex << cl << " " << dec << ret);
218 if (stf && last_myvs != r.last) {
219 stf->unmarshal_state(r.state);
222 LOG("transfer from " << m << " success, vs(" << last_myvs.vid << "," << last_myvs.seqno << ")");
226 bool rsm::statetransferdone(const string & m, lock & rsm_mutex_lock) {
227 rsm_mutex_lock.unlock();
229 rpcc *cl = h.safebind();
233 auto ret = (rsm_protocol::status)cl->call(rsm_protocol::transferdonereq, r, cfg->myaddr(), vid_insync);
234 done = (ret == rsm_protocol::OK);
236 rsm_mutex_lock.lock();
241 bool rsm::join(const string & m, lock & rsm_mutex_lock) {
246 LOG("contacting " << m << " mylast (" << last_myvs.vid << "," << last_myvs.seqno << ")");
249 rsm_mutex_lock.unlock();
252 ret = cl->call_timeout(rsm_protocol::joinreq, milliseconds(12000), log,
253 cfg->myaddr(), last_myvs);
255 rsm_mutex_lock.lock();
258 if (cl == 0 || ret != rsm_protocol::OK) {
259 LOG("couldn't reach " << m << " " << hex << cl << " " << dec << ret);
262 LOG("succeeded " << log);
268 // Config informs rsm whenever it has successfully
269 // completed a view change
271 void rsm::commit_change(unsigned vid) {
273 commit_change(vid, ml);
274 if (cfg->ismember(cfg->myaddr(), vid_commit))
278 void rsm::commit_change(unsigned vid, lock &) {
279 if (vid <= vid_commit)
281 LOG("new view (" << vid << ") last vs (" << last_myvs.vid << "," <<
282 last_myvs.seqno << ") " << primary << " insync " << insync);
286 recovery_cond.notify_one();
287 sync_cond.notify_one();
288 if (cfg->ismember(cfg->myaddr(), vid_commit))
293 void rsm::execute(rpc_protocol::proc_id_t procno, const string & req, string & r) {
295 handler *h = procs[procno];
298 auto ret = (rsm_protocol::status)(*h)(unmarshall(req, false), rep);
299 r = marshall(ret, rep.content()).content();
303 // Clients call client_invoke to invoke a procedure on the replicated state
304 // machine: the primary receives the request, assigns it a sequence
305 // number, and invokes it on all members of the replicated state
308 rsm_client_protocol::status rsm::client_invoke(string & r, rpc_protocol::proc_id_t procno, const string & req) {
309 LOG("invoke procno 0x" << hex << procno);
310 lock ml(invoke_mutex);
316 LOG("Checking for inviewchange");
318 return rsm_client_protocol::BUSY;
319 LOG("Checking for primacy");
320 myaddr = cfg->myaddr();
321 if (primary != myaddr)
322 return rsm_client_protocol::NOTPRIMARY;
323 LOG("Assigning a viewstamp");
324 cfg->get_view(vid_commit, m);
325 // assign the RPC the next viewstamp number
330 // send an invoke RPC to all slaves in the current view with a timeout of 1 second
331 LOG("Invoking slaves");
332 for (unsigned i = 0; i < m.size(); i++) {
333 if (m[i] != myaddr) {
334 // if invoke on slave fails, return rsm_client_protocol::BUSY
336 LOG("Sending invoke to " << m[i]);
337 rpcc *cl = h.safebind();
339 return rsm_client_protocol::BUSY;
341 auto ret = (rsm_protocol::status)cl->call_timeout(rsm_protocol::invoke, milliseconds(100), ignored_rval, procno, vs, req);
342 LOG("Invoke returned " << ret);
343 if (ret != rsm_protocol::OK)
344 return rsm_client_protocol::BUSY;
346 lock rsm_mutex_lock(rsm_mutex);
347 partition1(rsm_mutex_lock);
350 execute(procno, req, r);
351 for (size_t i=0; i<r.size(); i++) {
352 LOG(hex << setfill('0') << setw(2) << (unsigned int)(unsigned char)r[i]);
355 return rsm_client_protocol::OK;
359 // The primary calls the internal invoke at each member of the
360 // replicated state machine
362 // the replica must execute requests in order (with no gaps)
363 // according to requests' seqno
365 rsm_protocol::status rsm::invoke(int &, rpc_protocol::proc_id_t proc, viewstamp vs, const string & req) {
366 LOG("invoke procno 0x" << hex << proc);
367 lock ml(invoke_mutex);
372 // check if !inviewchange
373 LOG("Checking for view change");
375 return rsm_protocol::ERR;
377 LOG("Checking for slave status");
378 myaddr = cfg->myaddr();
379 if (primary == myaddr)
380 return rsm_protocol::ERR;
381 cfg->get_view(vid_commit, m);
382 if (find(m.begin(), m.end(), myaddr) == m.end())
383 return rsm_protocol::ERR;
384 // check sequence number
385 LOG("Checking sequence number");
387 return rsm_protocol::ERR;
391 execute(proc, req, r);
394 return rsm_protocol::OK;
398 // RPC handler: Send back the local node's state to the caller
400 rsm_protocol::status rsm::transferreq(rsm_protocol::transferres & r, const string & src,
401 viewstamp last, unsigned vid) {
403 LOG("transferreq from " << src << " (" << last.vid << "," << last.seqno << ") vs (" <<
404 last_myvs.vid << "," << last_myvs.seqno << ")");
405 if (!insync || vid != vid_insync)
406 return rsm_protocol::BUSY;
407 if (stf && last != last_myvs)
408 r.state = stf->marshal_state();
410 return rsm_protocol::OK;
414 // RPC handler: Inform the local node (the primary) that node m has synchronized
417 rsm_protocol::status rsm::transferdonereq(int &, const string & m, unsigned vid) {
419 if (!insync || vid != vid_insync)
420 return rsm_protocol::BUSY;
421 backups.erase(find(backups.begin(), backups.end(), m));
423 sync_cond.notify_one();
424 return rsm_protocol::OK;
427 // a node that wants to join an RSM as a server sends a
428 // joinreq to the RSM's current primary; this is the
429 // handler for that RPC.
430 rsm_protocol::status rsm::joinreq(string & log, const string & m, viewstamp last) {
431 auto ret = rsm_protocol::OK;
434 LOG("join request from " << m << "; last=(" << last.vid << "," << last.seqno << "), mylast=(" <<
435 last_myvs.vid << "," << last_myvs.seqno << ")");
436 if (cfg->ismember(m, vid_commit)) {
437 LOG(m << " is still a member -- nothing to do");
439 } else if (cfg->myaddr() != primary) {
440 LOG("but I, " << cfg->myaddr() << ", am not the primary, " << primary << "!");
441 ret = rsm_protocol::BUSY;
443 // We cache vid_commit to avoid adding m to a view which already contains
444 // m due to race condition
445 LOG("calling down to config layer");
446 unsigned vid_cache = vid_commit;
450 succ = cfg->add(m, vid_cache);
453 if (cfg->ismember(m, cfg->view_id())) {
455 LOG("ret " << ret << " log " << log);
457 LOG("failed; proposer couldn't add " << succ);
458 ret = rsm_protocol::BUSY;
465 // RPC handler: Responds with the list of known nodes for fall-back on a
468 rsm_client_protocol::status rsm::client_members(vector<string> & r, int) {
471 cfg->get_view(vid_commit, m);
472 m.push_back(primary);
474 LOG("return " << m << " m " << primary);
475 return rsm_client_protocol::OK;
478 // if primary is member of new view, that node is primary
479 // otherwise, the lowest number node of the previous view.
480 // caller should hold rsm_mutex
481 void rsm::set_primary(unsigned vid) {
483 cfg->get_view(vid, c);
484 cfg->get_view(vid - 1, p);
485 VERIFY (c.size() > 0);
487 if (isamember(primary,c)) {
488 LOG("primary stays " << primary);
492 VERIFY(p.size() > 0);
493 for (unsigned i = 0; i < p.size(); i++) {
494 if (isamember(p[i], c)) {
496 LOG("primary is " << primary);
503 bool rsm::amiprimary() {
505 return primary == cfg->myaddr() && !inviewchange;
509 // Test RPCs -- simulate partitions and failures
511 void rsm::net_repair(bool heal, lock & rsm_mutex_lock) {
512 VERIFY(rsm_mutex_lock);
514 cfg->get_view(vid_commit, m);
515 for (unsigned i = 0; i < m.size(); i++) {
516 if (m[i] != cfg->myaddr()) {
518 LOG("member " << m[i] << " " << heal);
519 if (h.safebind()) h.safebind()->set_reachable(heal);
522 rsmrpc->set_reachable(heal);
525 rsm_test_protocol::status rsm::test_net_repairreq(rsm_test_protocol::status & r, int heal) {
527 LOG("heal " << heal << " (dopartition " <<
528 dopartition << ", partitioned " << partitioned << ")");
530 net_repair(heal, ml);
534 return r = rsm_test_protocol::OK;
537 // simulate failure at breakpoint 1 and 2
539 void rsm::breakpoint(int b) {
540 if (breakpoints[b-1]) {
541 LOG("Dying at breakpoint " << b << " in rsm!");
546 void rsm::partition1(lock & rsm_mutex_lock) {
548 net_repair(false, rsm_mutex_lock);
554 rsm_test_protocol::status rsm::breakpointreq(rsm_test_protocol::status & r, int b) {
555 r = rsm_test_protocol::OK;
557 LOG("breakpoint " << b);
558 if (b == 1) breakpoints[1-1] = true;
559 else if (b == 2) breakpoints[2-1] = true;
560 else if (b == 3 || b == 4) cfg->breakpoint(b);
561 else r = rsm_test_protocol::ERR;