#include "hotstuff/hotstuff.h"
#include "hotstuff/client.h"
using salticidae::static_pointer_cast;
#define LOG_INFO HOTSTUFF_LOG_INFO
#define LOG_DEBUG HOTSTUFF_LOG_DEBUG
#define LOG_WARN HOTSTUFF_LOG_WARN
namespace hotstuff {
const opcode_t MsgPropose::opcode;
MsgPropose::MsgPropose(const Proposal &proposal) { serialized << proposal; }
void MsgPropose::postponed_parse(HotStuffCore *hsc) {
proposal.hsc = hsc;
serialized >> proposal;
}
const opcode_t MsgVote::opcode;
MsgVote::MsgVote(const Vote &vote) { serialized << vote; }
void MsgVote::postponed_parse(HotStuffCore *hsc) {
vote.hsc = hsc;
serialized >> vote;
}
const opcode_t MsgReqBlock::opcode;
MsgReqBlock::MsgReqBlock(const std::vector<uint256_t> &blk_hashes) {
serialized << (uint32_t)htole(blk_hashes.size());
for (const auto &h: blk_hashes)
serialized << h;
}
MsgReqBlock::MsgReqBlock(DataStream &&s) {
uint32_t size;
s >> size;
size = letoh(size);
blk_hashes.resize(size);
for (auto &h: blk_hashes) s >> h;
}
const opcode_t MsgRespBlock::opcode;
MsgRespBlock::MsgRespBlock(const std::vector<block_t> &blks) {
serialized << htole((uint32_t)blks.size());
for (auto blk: blks) serialized << *blk;
}
void MsgRespBlock::postponed_parse(HotStuffCore *hsc) {
uint32_t size;
serialized >> size;
size = letoh(size);
blks.resize(size);
for (auto &blk: blks)
{
Block _blk;
_blk.unserialize(serialized, hsc);
blk = hsc->storage->add_blk(std::move(_blk), hsc->get_config());
}
}
promise_t HotStuffBase::exec_command(command_t cmd) {
const uint256_t &cmd_hash = cmd->get_hash();
ReplicaID proposer = pmaker->get_proposer();
/* not the proposer */
if (proposer != get_id())
return promise_t([proposer, cmd_hash](promise_t &pm) {
pm.resolve(Finality(proposer, -1, 0, 0,
cmd_hash, uint256_t()));
});
auto it = decision_waiting.find(cmd_hash);
if (it == decision_waiting.end())
{
cmd_pending.push(storage->add_cmd(cmd));
it = decision_waiting.insert(std::make_pair(cmd_hash, promise_t())).first;
}
if (cmd_pending.size() >= blk_size)
{
std::vector<command_t> cmds;
for (uint32_t i = 0; i < blk_size; i++)
{
cmds.push_back(cmd_pending.front());
cmd_pending.pop();
}
pmaker->beat().then([this, cmds = std::move(cmds)](ReplicaID proposer) {
if (proposer != get_id())
{
for (auto &cmd: cmds)
decision_waiting
.at(cmd->get_hash())
.resolve(Finality(proposer, -1, 0, 0,
cmd->get_hash(), uint256_t()));
}
else
on_propose(cmds, pmaker->get_parents());
});
}
return it->second;
}
void HotStuffBase::add_replica(ReplicaID idx, const NetAddr &addr,
pubkey_bt &&pub_key) {
HotStuffCore::add_replica(idx, addr, std::move(pub_key));
if (addr != listen_addr)
pn.add_peer(addr);
}
void HotStuffBase::on_fetch_blk(const block_t &blk) {
#ifdef HOTSTUFF_BLK_PROFILE
blk_profiler.get_tx(blk->get_hash());
#endif
LOG_DEBUG("fetched %.10s", get_hex(blk->get_hash()).c_str());
part_fetched++;
fetched++;
for (auto cmd: blk->get_cmds()) on_fetch_cmd(cmd);
const uint256_t &blk_hash = blk->get_hash();
auto it = blk_fetch_waiting.find(blk_hash);
if (it != blk_fetch_waiting.end())
{
it->second.resolve(blk);
blk_fetch_waiting.erase(it);
}
}
void HotStuffBase::on_fetch_cmd(const command_t &cmd) {
const uint256_t &cmd_hash = cmd->get_hash();
auto it = cmd_fetch_waiting.find(cmd_hash);
if (it != cmd_fetch_waiting.end())
{
it->second.resolve(cmd);
cmd_fetch_waiting.erase(it);
}
}
void HotStuffBase::on_deliver_blk(const block_t &blk) {
const uint256_t &blk_hash = blk->get_hash();
bool valid;
/* sanity check: all parents must be delivered */
for (const auto &p: blk->get_parent_hashes())
assert(storage->is_blk_delivered(p));
if ((valid = HotStuffCore::on_deliver_blk(blk)))
{
LOG_DEBUG("block %.10s delivered",
get_hex(blk_hash).c_str());
part_parent_size += blk->get_parent_hashes().size();
part_delivered++;
delivered++;
}
else
{
LOG_WARN("dropping invalid block");
}
auto it = blk_delivery_waiting.find(blk_hash);
if (it != blk_delivery_waiting.end())
{
auto &pm = it->second;
if (valid)
{
pm.elapsed.stop(false);
auto sec = pm.elapsed.elapsed_sec;
part_delivery_time += sec;
part_delivery_time_min = std::min(part_delivery_time_min, sec);
part_delivery_time_max = std::max(part_delivery_time_max, sec);
pm.resolve(blk);
}
else
{
pm.reject(blk);
// TODO: do we need to also free it from storage?
}
blk_delivery_waiting.erase(it);
}
}
promise_t HotStuffBase::async_fetch_blk(const uint256_t &blk_hash,
const NetAddr *replica_id,
bool fetch_now) {
if (storage->is_blk_fetched(blk_hash))
return promise_t([this, &blk_hash](promise_t pm){
pm.resolve(storage->find_blk(blk_hash));
});
auto it = blk_fetch_waiting.find(blk_hash);
if (it == blk_fetch_waiting.end())
{
#ifdef HOTSTUFF_BLK_PROFILE
blk_profiler.rec_tx(blk_hash, false);
#endif
it = blk_fetch_waiting.insert(
std::make_pair(
blk_hash,
BlockFetchContext(blk_hash, this))).first;
}
if (replica_id != nullptr)
it->second.add_replica(*replica_id, fetch_now);
return static_cast<promise_t &>(it->second);
}
promise_t HotStuffBase::async_fetch_cmd(const uint256_t &cmd_hash,
const NetAddr *replica_id,
bool fetch_now) {
if (storage->is_cmd_fetched(cmd_hash))
return promise_t([this, &cmd_hash](promise_t pm){
pm.resolve(storage->find_cmd(cmd_hash));
});
auto it = cmd_fetch_waiting.find(cmd_hash);
if (it == cmd_fetch_waiting.end())
{
it = cmd_fetch_waiting.insert(
std::make_pair(cmd_hash, CmdFetchContext(cmd_hash, this))).first;
}
if (replica_id != nullptr)
it->second.add_replica(*replica_id, fetch_now);
return static_cast<promise_t &>(it->second);
}
promise_t HotStuffBase::async_deliver_blk(const uint256_t &blk_hash,
const NetAddr &replica_id) {
if (storage->is_blk_delivered(blk_hash))
return promise_t([this, &blk_hash](promise_t pm) {
pm.resolve(storage->find_blk(blk_hash));
});
auto it = blk_delivery_waiting.find(blk_hash);
if (it != blk_delivery_waiting.end())
return static_cast<promise_t &>(it->second);
BlockDeliveryContext pm{[](promise_t){}};
it = blk_delivery_waiting.insert(std::make_pair(blk_hash, pm)).first;
/* otherwise the on_deliver_batch will resolve */
async_fetch_blk(blk_hash, &replica_id).then([this, replica_id](block_t blk) {
/* qc_ref should be fetched */
std::vector<promise_t> pms;
const auto &qc = blk->get_qc();
if (qc)
pms.push_back(async_fetch_blk(qc->get_blk_hash(), &replica_id));
/* the parents should be delivered */
for (const auto &phash: blk->get_parent_hashes())
pms.push_back(async_deliver_blk(phash, replica_id));
promise::all(pms).then([this, blk]() {
on_deliver_blk(blk);
});
});
return static_cast<promise_t &>(pm);
}
void HotStuffBase::propose_handler(MsgPropose &&msg, Conn &conn) {
const NetAddr &peer = conn.get_peer();
msg.postponed_parse(this);
auto &prop = msg.proposal;
block_t blk = prop.blk;
if (!blk) return;
promise::all(std::vector<promise_t>{
async_deliver_blk(prop.bqc_hash, peer),
async_deliver_blk(blk->get_hash(), peer),
}).then([this, prop = std::move(prop)]() {
on_receive_proposal(prop);
});
}
void HotStuffBase::vote_handler(MsgVote &&msg, Conn &conn) {
const NetAddr &peer = conn.get_peer();
msg.postponed_parse(this);
auto &vote = msg.vote;
promise::all(std::vector<promise_t>{
async_deliver_blk(vote.bqc_hash, peer),
async_deliver_blk(vote.blk_hash, peer)
}).then([this, vote = std::move(vote)]() {
on_receive_vote(vote);
});
}
void HotStuffBase::req_blk_handler(MsgReqBlock &&msg, Conn &conn) {
const NetAddr replica = conn.get_peer();
auto &blk_hashes = msg.blk_hashes;
std::vector<promise_t> pms;
for (const auto &h: blk_hashes)
pms.push_back(async_fetch_blk(h, nullptr));
promise::all(pms).then([replica, this](const promise::values_t values) {
std::vector<block_t> blks;
for (auto &v: values)
{
auto blk = promise::any_cast<block_t>(v);
blks.push_back(blk);
}
pn.send_msg(MsgRespBlock(blks), replica);
});
}
void HotStuffBase::resp_blk_handler(MsgRespBlock &&msg, Conn &) {
msg.postponed_parse(this);
for (const auto &blk: msg.blks)
if (blk) on_fetch_blk(blk);
}
void HotStuffBase::print_stat() const {
LOG_INFO("===== begin stats =====");
LOG_INFO("-------- queues -------");
LOG_INFO("blk_fetch_waiting: %lu", blk_fetch_waiting.size());
LOG_INFO("blk_delivery_waiting: %lu", blk_delivery_waiting.size());
LOG_INFO("cmd_fetch_waiting: %lu", cmd_fetch_waiting.size());
LOG_INFO("decision_waiting: %lu", decision_waiting.size());
LOG_INFO("-------- misc ---------");
LOG_INFO("fetched: %lu", fetched);
LOG_INFO("delivered: %lu", delivered);
LOG_INFO("cmd_cache: %lu", storage->get_cmd_cache_size());
LOG_INFO("blk_cache: %lu", storage->get_blk_cache_size());
LOG_INFO("------ misc (10s) -----");
LOG_INFO("fetched: %lu", part_fetched);
LOG_INFO("delivered: %lu", part_delivered);
LOG_INFO("decided: %lu", part_decided);
LOG_INFO("gened: %lu", part_gened);
LOG_INFO("avg. parent_size: %.3f",
part_delivered ? part_parent_size / double(part_delivered) : 0);
LOG_INFO("delivery time: %.3f avg, %.3f min, %.3f max",
part_delivered ? part_delivery_time / double(part_delivered) : 0,
part_delivery_time_min == double_inf ? 0 : part_delivery_time_min,
part_delivery_time_max);
part_parent_size = 0;
part_fetched = 0;
part_delivered = 0;
part_decided = 0;
part_gened = 0;
part_delivery_time = 0;
part_delivery_time_min = double_inf;
part_delivery_time_max = 0;
#ifdef HOTSTUFF_MSG_STAT
LOG_INFO("-- sent opcode (10s) --");
auto &sent_op = pn.get_sent_by_opcode();
for (auto &op: sent_op)
{
auto &val = op.second;
LOG_INFO("%02x: %lu, %.2fBpm", op.first,
val.first, val.first ? val.second / double(val.first) : 0);
val.first = val.second = 0;
}
LOG_INFO("-- recv opcode (10s) --");
auto &recv_op = pn.get_recv_by_opcode();
for (auto &op: recv_op)
{
auto &val = op.second;
LOG_INFO("%02x: %lu, %.2fBpm", op.first,
val.first, val.first ? val.second / double(val.first) : 0);
val.first = val.second = 0;
}
LOG_INFO("--- replica msg. (10s) ---");
size_t _nsent = 0;
size_t _nrecv = 0;
for (const auto &replica: pn.all_peers())
{
auto conn = pn.get_peer_conn(replica);
if (conn == nullptr) continue;
size_t ns = conn->get_nsent();
size_t nr = conn->get_nrecv();
conn->clear_nsent();
conn->clear_nrecv();
LOG_INFO("%s: %u, %u, %u",
std::string(replica).c_str(), ns, nr, part_fetched_replica[replica]);
_nsent += ns;
_nrecv += nr;
part_fetched_replica[replica] = 0;
}
nsent += _nsent;
nrecv += _nrecv;
LOG_INFO("sent: %lu", _nsent);
LOG_INFO("recv: %lu", _nrecv);
LOG_INFO("--- replica msg. total ---");
LOG_INFO("sent: %lu", nsent);
LOG_INFO("recv: %lu", nrecv);
#endif
LOG_INFO("====== end stats ======");
}
HotStuffBase::HotStuffBase(uint32_t blk_size,
ReplicaID rid,
privkey_bt &&priv_key,
NetAddr listen_addr,
pacemaker_bt pmaker,
EventContext eb):
HotStuffCore(rid, std::move(priv_key)),
listen_addr(listen_addr),
blk_size(blk_size),
eb(eb),
pn(eb),
pmaker(std::move(pmaker)),
fetched(0), delivered(0),
nsent(0), nrecv(0),
part_parent_size(0),
part_fetched(0),
part_delivered(0),
part_decided(0),
part_gened(0),
part_delivery_time(0),
part_delivery_time_min(double_inf),
part_delivery_time_max(0)
{
/* register the handlers for msg from replicas */
pn.reg_handler(salticidae::generic_bind(&HotStuffBase::propose_handler, this, _1, _2));
pn.reg_handler(salticidae::generic_bind(&HotStuffBase::vote_handler, this, _1, _2));
pn.reg_handler(salticidae::generic_bind(&HotStuffBase::req_blk_handler, this, _1, _2));
pn.reg_handler(salticidae::generic_bind(&HotStuffBase::resp_blk_handler, this, _1, _2));
pn.listen(listen_addr);
}
void HotStuffBase::do_broadcast_proposal(const Proposal &prop) {
MsgPropose prop_msg(prop);
for (const auto &replica: pn.all_peers())
pn.send_msg(prop_msg, replica);
}
void HotStuffBase::do_vote(ReplicaID last_proposer, const Vote &vote) {
pmaker->next_proposer(last_proposer)
.then([this, vote](ReplicaID proposer) {
if (proposer == get_id())
on_receive_vote(vote);
else
pn.send_msg(MsgVote(vote), get_config().get_addr(proposer));
});
}
void HotStuffBase::do_decide(Finality &&fin) {
state_machine_execute(fin);
auto it = decision_waiting.find(fin.cmd_hash);
if (it != decision_waiting.end())
{
it->second.resolve(std::move(fin));
decision_waiting.erase(it);
}
}
HotStuffBase::~HotStuffBase() {}
void HotStuffBase::start(bool eb_loop) {
/* ((n - 1) + 1 - 1) / 3 */
uint32_t nfaulty = pn.all_peers().size() / 3;
if (nfaulty == 0)
LOG_WARN("too few replicas in the system to tolerate any failure");
pmaker->init(this);
on_init(nfaulty);
if (eb_loop)
eb.dispatch();
}
}