#include "handle.h"
#include "rsm.h"
-#include "tprintf.h"
+#include "threaded_log.h"
#include "lang/verify.h"
#include "rsm_client.h"
#include "lock.h"
rsmrpc->reg(rsm_protocol::joinreq, &rsm::joinreq, this);
// tester must be on different port, otherwise it may partition itself
- testsvr = new rpcs(atoi(_me.c_str()) + 1);
+ testsvr = new rpcs((uint32_t)std::stoi(_me) + 1);
testsvr->reg(rsm_test_protocol::net_repair, &rsm::test_net_repairreq, this);
testsvr->reg(rsm_test_protocol::breakpoint, &rsm::breakpointreq, this);
}
// The recovery thread runs this function
-void rsm::recovery() {
+void rsm::recovery() [[noreturn]] {
bool r = true;
lock ml(rsm_mutex);
while (!cfg->ismember(cfg->myaddr(), vid_commit)) {
// XXX iannucci 2013/09/15 -- I don't understand whether accessing
// cfg->view_id in this manner involves a race. I suspect not.
- if (join(primary)) {
- tprintf("recovery: joined\n");
- commit_change_wo(cfg->view_id());
+ if (join(primary, ml)) {
+ LOG("recovery: joined");
+ commit_change(cfg->view_id(), ml);
} else {
ml.unlock();
std::this_thread::sleep_for(std::chrono::seconds(30)); // XXX make another node in cfg primary?
}
}
vid_insync = vid_commit;
- tprintf("recovery: sync vid_insync %d\n", vid_insync);
+ LOG("recovery: sync vid_insync " << vid_insync);
if (primary == cfg->myaddr()) {
- r = sync_with_backups();
+ r = sync_with_backups(ml);
} else {
- r = sync_with_primary();
+ r = sync_with_primary(ml);
}
- tprintf("recovery: sync done\n");
+ LOG("recovery: sync done");
// If there was a commited viewchange during the synchronization, restart
// the recovery
myvs.seqno = 1;
inviewchange = false;
}
- tprintf("recovery: go to sleep %d %d\n", insync, inviewchange);
+ LOG("recovery: go to sleep " << insync << " " << inviewchange);
recovery_cond.wait(ml);
}
}
-template <class A>
-std::ostream & operator<<(std::ostream &o, const std::vector<A> &d) {
- o << "[";
- for (typename std::vector<A>::const_iterator i=d.begin(); i!=d.end(); i++) {
- o << *i;
- if (i+1 != d.end())
- o << ", ";
- }
- o << "]";
- return o;
-}
-
-bool rsm::sync_with_backups() {
- adopt_lock ml(rsm_mutex);
- ml.unlock();
+bool rsm::sync_with_backups(lock & rsm_mutex_lock) {
+ rsm_mutex_lock.unlock();
{
// Make sure that the state of lock_server is stable during
// synchronization; otherwise, the primary's state may be more recent
// than replicas after the synchronization.
- lock ml(invoke_mutex);
+ lock invoke_mutex_lock(invoke_mutex);
// By acquiring and releasing the invoke_mutex once, we make sure that
// the state of lock_server will not be changed until all
// replicas are synchronized. The reason is that client_invoke arrives
// after this point of time will see inviewchange == true, and returns
// BUSY.
}
- ml.lock();
+ rsm_mutex_lock.lock();
// Start accepting synchronization request (statetransferreq) now!
insync = true;
cfg->get_view(vid_insync, backups);
backups.erase(find(backups.begin(), backups.end(), cfg->myaddr()));
- LOG("rsm::sync_with_backups " << backups);
- sync_cond.wait(ml);
+ LOG("rsm::sync_with_backups " << make_iterator_pair(backups.begin(), backups.end()));
+ sync_cond.wait(rsm_mutex_lock);
insync = false;
return true;
}
-bool rsm::sync_with_primary() {
+bool rsm::sync_with_primary(lock & rsm_mutex_lock) {
// Remember the primary of vid_insync
std::string m = primary;
while (vid_insync == vid_commit) {
- if (statetransfer(m))
+ if (statetransfer(m, rsm_mutex_lock))
break;
}
- return statetransferdone(m);
+ return statetransferdone(m, rsm_mutex_lock);
}
* Call to transfer state from m to the local node.
* Assumes that rsm_mutex is already held.
*/
-bool rsm::statetransfer(std::string m)
+bool rsm::statetransfer(std::string m, lock & rsm_mutex_lock)
{
rsm_protocol::transferres r;
handle h(m);
int ret = 0;
- tprintf("rsm::statetransfer: contact %s w. my last_myvs(%d,%d)\n",
- m.c_str(), last_myvs.vid, last_myvs.seqno);
+ LOG("rsm::statetransfer: contact " << m << " w. my last_myvs(" << last_myvs.vid << "," << last_myvs.seqno << ")");
rpcc *cl;
{
- adopt_lock ml(rsm_mutex);
- ml.unlock();
+ rsm_mutex_lock.unlock();
cl = h.safebind();
if (cl) {
ret = cl->call_timeout(rsm_protocol::transferreq, rpcc::to(1000),
r, cfg->myaddr(), last_myvs, vid_insync);
}
- ml.lock();
+ rsm_mutex_lock.lock();
}
if (cl == 0 || ret != rsm_protocol::OK) {
- tprintf("rsm::statetransfer: couldn't reach %s %lx %d\n", m.c_str(),
- (long unsigned) cl, ret);
+ LOG("rsm::statetransfer: couldn't reach " << m << " " << std::hex << cl << " " << std::dec << ret);
return false;
}
if (stf && last_myvs != r.last) {
stf->unmarshal_state(r.state);
}
last_myvs = r.last;
- tprintf("rsm::statetransfer transfer from %s success, vs(%d,%d)\n",
- m.c_str(), last_myvs.vid, last_myvs.seqno);
+ LOG("rsm::statetransfer transfer from " << m << " success, vs(" << last_myvs.vid << "," << last_myvs.seqno << ")");
return true;
}
-bool rsm::statetransferdone(std::string m) {
- adopt_lock ml(rsm_mutex);
- ml.unlock();
+bool rsm::statetransferdone(std::string m, lock & rsm_mutex_lock) {
+ rsm_mutex_lock.unlock();
handle h(m);
rpcc *cl = h.safebind();
bool done = false;
if (cl) {
int r;
- rsm_protocol::status ret = cl->call(rsm_protocol::transferdonereq, r, cfg->myaddr(), vid_insync);
+ auto ret = (rsm_protocol::status)cl->call(rsm_protocol::transferdonereq, r, cfg->myaddr(), vid_insync);
done = (ret == rsm_protocol::OK);
}
- ml.lock();
+ rsm_mutex_lock.lock();
return done;
}
-bool rsm::join(std::string m) {
+bool rsm::join(std::string m, lock & rsm_mutex_lock) {
handle h(m);
int ret = 0;
rsm_protocol::joinres r;
- tprintf("rsm::join: %s mylast (%d,%d)\n", m.c_str(), last_myvs.vid,
- last_myvs.seqno);
+ LOG("rsm::join: " << m << " mylast (" << last_myvs.vid << "," << last_myvs.seqno << ")");
rpcc *cl;
{
- adopt_lock ml(rsm_mutex);
- ml.unlock();
+ rsm_mutex_lock.unlock();
cl = h.safebind();
if (cl != 0) {
ret = cl->call_timeout(rsm_protocol::joinreq, rpcc::to(120000), r,
cfg->myaddr(), last_myvs);
}
- ml.lock();
+ rsm_mutex_lock.lock();
}
if (cl == 0 || ret != rsm_protocol::OK) {
- tprintf("rsm::join: couldn't reach %s %p %d\n", m.c_str(),
- cl, ret);
+ LOG("rsm::join: couldn't reach " << m << " " << std::hex << cl << " " << std::dec << ret);
return false;
}
- tprintf("rsm::join: succeeded %s\n", r.log.c_str());
+ LOG("rsm::join: succeeded " << r.log);
cfg->restore(r.log);
return true;
}
*/
void rsm::commit_change(unsigned vid) {
lock ml(rsm_mutex);
- commit_change_wo(vid);
+ commit_change(vid, ml);
if (cfg->ismember(cfg->myaddr(), vid_commit))
breakpoint2();
}
-void rsm::commit_change_wo(unsigned vid) {
+void rsm::commit_change(unsigned vid, lock &) {
if (vid <= vid_commit)
return;
- tprintf("commit_change: new view (%d) last vs (%d,%d) %s insync %d\n",
- vid, last_myvs.vid, last_myvs.seqno, primary.c_str(), insync);
+ LOG("commit_change: new view (" << vid << ") last vs (" << last_myvs.vid << "," <<
+ last_myvs.seqno << ") " << primary << " insync " << insync);
vid_commit = vid;
inviewchange = true;
set_primary(vid);
void rsm::execute(int procno, std::string req, std::string &r) {
- tprintf("execute\n");
+ LOG("execute");
handler *h = procs[procno];
VERIFY(h);
unmarshall args(req);
marshall rep;
std::string reps;
- rsm_protocol::status ret = (*h)(args, rep);
+ auto ret = (rsm_protocol::status)(*h)(args, rep);
marshall rep1;
rep1 << ret;
rep1 << rep.str();
std::string myaddr;
viewstamp vs;
{
- lock ml(rsm_mutex);
+ lock ml2(rsm_mutex);
LOG("Checking for inviewchange");
if (inviewchange)
return rsm_client_protocol::BUSY;
rpcc *cl = h.safebind();
if (!cl)
return rsm_client_protocol::BUSY;
- rsm_protocol::status ret;
- int r;
- ret = cl->call_timeout(rsm_protocol::invoke, rpcc::to(1000), r, procno, vs, req);
+ int ignored_rval;
+ auto ret = (rsm_protocol::status)cl->call_timeout(rsm_protocol::invoke, rpcc::to(1000), ignored_rval, procno, vs, req);
LOG("Invoke returned " << ret);
if (ret != rsm_protocol::OK)
return rsm_client_protocol::BUSY;
breakpoint1();
- partition1();
+ lock rsm_mutex_lock(rsm_mutex);
+ partition1(rsm_mutex_lock);
}
}
execute(procno, req, r);
std::vector<std::string> m;
std::string myaddr;
{
- lock ml(rsm_mutex);
+ lock ml2(rsm_mutex);
// check if !inviewchange
LOG("Checking for view change");
if (inviewchange)
rsm_protocol::status rsm::transferreq(rsm_protocol::transferres &r, std::string src,
viewstamp last, unsigned vid) {
lock ml(rsm_mutex);
- int ret = rsm_protocol::OK;
- tprintf("transferreq from %s (%d,%d) vs (%d,%d)\n", src.c_str(),
- last.vid, last.seqno, last_myvs.vid, last_myvs.seqno);
- if (!insync || vid != vid_insync) {
+ LOG("transferreq from " << src << " (" << last.vid << "," << last.seqno << ") vs (" <<
+ last_myvs.vid << "," << last_myvs.seqno << ")");
+ if (!insync || vid != vid_insync)
return rsm_protocol::BUSY;
- }
if (stf && last != last_myvs)
r.state = stf->marshal_state();
r.last = last_myvs;
- return ret;
+ return rsm_protocol::OK;
}
/**
// joinreq to the RSM's current primary; this is the
// handler for that RPC.
rsm_protocol::status rsm::joinreq(rsm_protocol::joinres &r, std::string m, viewstamp last) {
- int ret = rsm_protocol::OK;
+ auto ret = rsm_protocol::OK;
lock ml(rsm_mutex);
- tprintf("joinreq: src %s last (%d,%d) mylast (%d,%d)\n", m.c_str(),
- last.vid, last.seqno, last_myvs.vid, last_myvs.seqno);
+ LOG("joinreq: src " << m << " last (" << last.vid << "," << last.seqno << ") mylast (" <<
+ last_myvs.vid << "," << last_myvs.seqno << ")");
if (cfg->ismember(m, vid_commit)) {
- tprintf("joinreq: is still a member\n");
+ LOG("joinreq: is still a member");
r.log = cfg->dump();
} else if (cfg->myaddr() != primary) {
- tprintf("joinreq: busy\n");
+ LOG("joinreq: busy");
ret = rsm_protocol::BUSY;
} else {
// We cache vid_commit to avoid adding m to a view which already contains
}
if (cfg->ismember(m, cfg->view_id())) {
r.log = cfg->dump();
- tprintf("joinreq: ret %d log %s\n:", ret, r.log.c_str());
+ LOG("joinreq: ret " << ret << " log " << r.log);
} else {
- tprintf("joinreq: failed; proposer couldn't add %d\n", succ);
+ LOG("joinreq: failed; proposer couldn't add " << succ);
ret = rsm_protocol::BUSY;
}
}
* so the client can switch to a different primary
* when it existing primary fails
*/
-rsm_client_protocol::status rsm::client_members(std::vector<std::string> &r, int i) {
+rsm_client_protocol::status rsm::client_members(std::vector<std::string> &r, int) {
std::vector<std::string> m;
lock ml(rsm_mutex);
cfg->get_view(vid_commit, m);
m.push_back(primary);
r = m;
- tprintf("rsm::client_members return %s m %s\n", print_members(m).c_str(),
- primary.c_str());
+ LOG("rsm::client_members return " << print_members(m) << " m " << primary);
return rsm_client_protocol::OK;
}
VERIFY (c.size() > 0);
if (isamember(primary,c)) {
- tprintf("set_primary: primary stays %s\n", primary.c_str());
+ LOG("set_primary: primary stays " << primary);
return;
}
for (unsigned i = 0; i < p.size(); i++) {
if (isamember(p[i], c)) {
primary = p[i];
- tprintf("set_primary: primary is %s\n", primary.c_str());
+ LOG("set_primary: primary is " << primary);
return;
}
}
// Simulate partitions
// assumes caller holds rsm_mutex
-void rsm::net_repair_wo(bool heal) {
+void rsm::net_repair(bool heal, lock &) {
std::vector<std::string> m;
cfg->get_view(vid_commit, m);
for (unsigned i = 0; i < m.size(); i++) {
if (m[i] != cfg->myaddr()) {
handle h(m[i]);
- tprintf("rsm::net_repair_wo: %s %d\n", m[i].c_str(), heal);
+ LOG("rsm::net_repair: " << m[i] << " " << heal);
if (h.safebind()) h.safebind()->set_reachable(heal);
}
}
rsmrpc->set_reachable(heal);
}
-rsm_test_protocol::status rsm::test_net_repairreq(int &r, int heal) {
+rsm_test_protocol::status rsm::test_net_repairreq(rsm_test_protocol::status &r, int heal) {
lock ml(rsm_mutex);
- tprintf("rsm::test_net_repairreq: %d (dopartition %d, partitioned %d)\n",
- heal, dopartition, partitioned);
+ LOG("rsm::test_net_repairreq: " << heal << " (dopartition " <<
+ dopartition << ", partitioned " << partitioned << ")");
if (heal) {
- net_repair_wo(heal);
+ net_repair(heal, ml);
partitioned = false;
} else {
dopartition = true;
void rsm::breakpoint1() {
if (break1) {
- tprintf("Dying at breakpoint 1 in rsm!\n");
+ LOG("Dying at breakpoint 1 in rsm!");
exit(1);
}
}
void rsm::breakpoint2() {
if (break2) {
- tprintf("Dying at breakpoint 2 in rsm!\n");
+ LOG("Dying at breakpoint 2 in rsm!");
exit(1);
}
}
-void rsm::partition1() {
+void rsm::partition1(lock & rsm_mutex_lock) {
if (dopartition) {
- net_repair_wo(false);
+ net_repair(false, rsm_mutex_lock);
dopartition = false;
partitioned = true;
}
}
-rsm_test_protocol::status rsm::breakpointreq(int &r, int b) {
+rsm_test_protocol::status rsm::breakpointreq(rsm_test_protocol::status &r, int b) {
r = rsm_test_protocol::OK;
lock ml(rsm_mutex);
- tprintf("rsm::breakpointreq: %d\n", b);
+ LOG("rsm::breakpointreq: " << b);
if (b == 1) break1 = true;
else if (b == 2) break2 = true;
else if (b == 3 || b == 4) cfg->breakpoint(b);
projects / invirt/third/libt4.git / blobdiff