void TxnManager::cleanup(RC rc) {
#if CC_ALG == OCC && MODE == NORMAL_MODE
occ_man.finish(rc,this);
#endif
ts_t starttime = get_sys_clock();
uint64_t row_cnt = txn->accesses.get_count();
assert(txn->accesses.get_count() == txn->row_cnt);
//assert((WORKLOAD == YCSB && row_cnt <= g_req_per_query) || (WORKLOAD == TPCC && row_cnt <= g_max_items_per_txn*2 + 3));
DEBUG("Cleanup %ld %ld\n",get_txn_id(),row_cnt);
for (int rid = row_cnt - 1; rid >= 0; rid --) {
cleanup_row(rc,rid);
}
#if CC_ALG == CALVIN
// cleanup locked rows
for (uint64_t i = 0; i < calvin_locked_rows.size(); i++) {
row_t * row = calvin_locked_rows[i];
row->return_row(rc,RD,this,row);
}
#endif
if (rc == Abort) {
txn->release_inserts(get_thd_id());
txn->insert_rows.clear();
INC_STATS(get_thd_id(), abort_time, get_sys_clock() - starttime);
}
}
itemid_t *
txn_man::index_read(INDEX * index, idx_key_t key, int part_id) {
uint64_t starttime = get_sys_clock();
itemid_t * item;
index->index_read(key, item, part_id, get_thd_id());
INC_TMP_STATS(get_thd_id(), time_index, get_sys_clock() - starttime);
return item;
}
itemid_t *
TxnManager::index_read(INDEX * index, idx_key_t key, int part_id) {
uint64_t starttime = get_sys_clock();
itemid_t * item;
index->index_read(key, item, part_id, get_thd_id());
uint64_t t = get_sys_clock() - starttime;
INC_STATS(get_thd_id(), txn_index_time, t);
//txn_time_idx += t;
return item;
}
RC InputThread::server_recv_loop() {
myrand rdm;
rdm.init(get_thd_id());
RC rc = RCOK;
assert (rc == RCOK);
uint64_t starttime;
std::vector<Message*> * msgs;
while (!simulation->is_done()) {
heartbeat();
starttime = get_sys_clock();
msgs = tport_man.recv_msg(get_thd_id());
INC_STATS(_thd_id,mtx[28], get_sys_clock() - starttime);
starttime = get_sys_clock();
if(msgs == NULL)
continue;
while(!msgs->empty()) {
Message * msg = msgs->front();
if(msg->rtype == INIT_DONE) {
msgs->erase(msgs->begin());
continue;
}
#if CC_ALG == CALVIN
if(msg->rtype == CALVIN_ACK ||(msg->rtype == CL_QRY && ISCLIENTN(msg->get_return_id()))) {
work_queue.sequencer_enqueue(get_thd_id(),msg);
msgs->erase(msgs->begin());
continue;
}
if( msg->rtype == RDONE || msg->rtype == CL_QRY) {
assert(ISSERVERN(msg->get_return_id()));
work_queue.sched_enqueue(get_thd_id(),msg);
msgs->erase(msgs->begin());
continue;
}
#endif
work_queue.enqueue(get_thd_id(),msg,false);
msgs->erase(msgs->begin());
}
delete msgs;
INC_STATS(_thd_id,mtx[29], get_sys_clock() - starttime);
}
printf("FINISH %ld:%ld\n",_node_id,_thd_id);
fflush(stdout);
return FINISH;
}
RC txn_man::finish(RC rc) {
if (CC_ALG == HSTORE)
return RCOK;
uint64_t starttime = get_sys_clock();
if (CC_ALG == OCC && rc == RCOK) {
// validation phase.
rc = occ_man.validate(this);
} else
cleanup(rc);
uint64_t timespan = get_sys_clock() - starttime;
INC_TMP_STATS(get_thd_id(), time_man, timespan);
INC_STATS(get_thd_id(), time_cleanup, timespan);
return rc;
}
RC InputThread::client_recv_loop() {
int rsp_cnts[g_servers_per_client];
memset(rsp_cnts, 0, g_servers_per_client * sizeof(int));
run_starttime = get_sys_clock();
uint64_t return_node_offset;
uint64_t inf;
std::vector<Message*> * msgs;
while (!simulation->is_done()) {
heartbeat();
uint64_t starttime = get_sys_clock();
msgs = tport_man.recv_msg(get_thd_id());
INC_STATS(_thd_id,mtx[28], get_sys_clock() - starttime);
starttime = get_sys_clock();
//while((m_query = work_queue.get_next_query(get_thd_id())) != NULL) {
//Message * msg = work_queue.dequeue();
if(msgs == NULL)
continue;
while(!msgs->empty()) {
Message * msg = msgs->front();
assert(msg->rtype == CL_RSP);
return_node_offset = msg->return_node_id - g_server_start_node;
assert(return_node_offset < g_servers_per_client);
rsp_cnts[return_node_offset]++;
INC_STATS(get_thd_id(),txn_cnt,1);
uint64_t timespan = get_sys_clock() - ((ClientResponseMessage*)msg)->client_startts;
INC_STATS(get_thd_id(),txn_run_time, timespan);
if (warmup_done) {
INC_STATS_ARR(get_thd_id(),client_client_latency, timespan);
}
//INC_STATS_ARR(get_thd_id(),all_lat,timespan);
inf = client_man.dec_inflight(return_node_offset);
DEBUG("Recv %ld from %ld, %ld -- %f\n",((ClientResponseMessage*)msg)->txn_id,msg->return_node_id,inf,float(timespan)/BILLION);
assert(inf >=0);
// TODO: delete message
msgs->erase(msgs->begin());
}
delete msgs;
INC_STATS(_thd_id,mtx[29], get_sys_clock() - starttime);
}
printf("FINISH %ld:%ld\n",_node_id,_thd_id);
fflush(stdout);
return FINISH;
}
row_t * txn_man::get_row(row_t * row, access_t type) {
if (CC_ALG == HSTORE)
return row;
uint64_t starttime = get_sys_clock();
RC rc = RCOK;
// assert(row_cnt < MAX_ROW_PER_TXN);
uint64_t part_id = row->get_part_id();
if (accesses[row_cnt] == NULL) {
accesses[row_cnt] = (Access *)
mem_allocator.alloc(sizeof(Access), part_id);
num_accesses_alloc ++;
}
rc = row->get_row(type, this, accesses[ row_cnt ]->data);
if (rc == Abort) {
return NULL;
}
accesses[row_cnt]->type = type;
accesses[row_cnt]->orig_row = row;
#if ROLL_BACK && (CC_ALG == DL_DETECT || CC_ALG == NO_WAIT || CC_ALG == WAIT_DIE)
if (type == WR) {
accesses[row_cnt]->orig_data = (row_t *)
mem_allocator.alloc(sizeof(row_t), part_id);
accesses[row_cnt]->orig_data->init(row->get_table(), part_id, 0);
accesses[row_cnt]->orig_data->copy(row);
}
#endif
row_cnt ++;
if (type == WR)
wr_cnt ++;
uint64_t timespan = get_sys_clock() - starttime;
INC_TMP_STATS(get_thd_id(), time_man, timespan);
return accesses[row_cnt - 1]->data;
}
RC YCSBTxnManager::send_remote_request() {
YCSBQuery* ycsb_query = (YCSBQuery*) query;
uint64_t dest_node_id = GET_NODE_ID(ycsb_query->requests[next_record_id]->key);
ycsb_query->partitions_touched.add_unique(GET_PART_ID(0,dest_node_id));
msg_queue.enqueue(get_thd_id(),Message::create_message(this,RQRY),dest_node_id);
return WAIT_REM;
}
RC TxnManager::commit() {
DEBUG("Commit %ld\n",get_txn_id());
release_locks(RCOK);
#if CC_ALG == MAAT
time_table.release(get_thd_id(),get_txn_id());
#endif
commit_stats();
#if LOGGING
LogRecord * record = logger.createRecord(get_txn_id(),L_NOTIFY,0,0);
if(g_repl_cnt > 0) {
msg_queue.enqueue(get_thd_id(),Message::create_message(record,LOG_MSG),g_node_id + g_node_cnt + g_client_node_cnt);
}
logger.enqueueRecord(record);
return WAIT;
#endif
return Commit;
}
void TxnManager::cleanup_row(RC rc, uint64_t rid) {
access_t type = txn->accesses[rid]->type;
if (type == WR && rc == Abort && CC_ALG != MAAT) {
type = XP;
}
// Handle calvin elsewhere
#if CC_ALG != CALVIN
#if ISOLATION_LEVEL != READ_UNCOMMITTED
row_t * orig_r = txn->accesses[rid]->orig_row;
if (ROLL_BACK && type == XP &&
(CC_ALG == DL_DETECT ||
CC_ALG == NO_WAIT ||
CC_ALG == WAIT_DIE ||
CC_ALG == HSTORE ||
CC_ALG == HSTORE_SPEC
))
{
orig_r->return_row(rc,type, this, txn->accesses[rid]->orig_data);
} else {
#if ISOLATION_LEVEL == READ_COMMITTED
if(type == WR) {
orig_r->return_row(rc,type, this, txn->accesses[rid]->data);
}
#else
orig_r->return_row(rc,type, this, txn->accesses[rid]->data);
#endif
}
#endif
#if ROLL_BACK && (CC_ALG == NO_WAIT || CC_ALG == WAIT_DIE || CC_ALG == HSTORE || CC_ALG == HSTORE_SPEC)
if (type == WR) {
//printf("free 10 %ld\n",get_txn_id());
txn->accesses[rid]->orig_data->free_row();
DEBUG_M("TxnManager::cleanup row_t free\n");
row_pool.put(get_thd_id(),txn->accesses[rid]->orig_data);
if(rc == RCOK) {
INC_STATS(get_thd_id(),record_write_cnt,1);
++txn_stats.write_cnt;
}
}
#endif
#endif
txn->accesses[rid]->data = NULL;
}
void TxnManager::release_locks(RC rc) {
uint64_t starttime = get_sys_clock();
cleanup(rc);
uint64_t timespan = (get_sys_clock() - starttime);
INC_STATS(get_thd_id(), txn_cleanup_time, timespan);
}
void TxnManager::send_prepare_messages() {
rsp_cnt = query->partitions_touched.size() - 1;
DEBUG("%ld Send PREPARE messages to %d\n",get_txn_id(),rsp_cnt);
for(uint64_t i = 0; i < query->partitions_touched.size(); i++) {
if(GET_NODE_ID(query->partitions_touched[i]) == g_node_id) {
continue;
}
msg_queue.enqueue(get_thd_id(),Message::create_message(this,RPREPARE),GET_NODE_ID(query->partitions_touched[i]));
}
}
void TxnManager::send_finish_messages() {
rsp_cnt = query->partitions_touched.size() - 1;
assert(IS_LOCAL(get_txn_id()));
DEBUG("%ld Send FINISH messages to %d\n",get_txn_id(),rsp_cnt);
for(uint64_t i = 0; i < query->partitions_touched.size(); i++) {
if(GET_NODE_ID(query->partitions_touched[i]) == g_node_id) {
continue;
}
msg_queue.enqueue(get_thd_id(),Message::create_message(this,RFIN),GET_NODE_ID(query->partitions_touched[i]));
}
}
RC TxnManager::get_lock(row_t * row, access_t type) {
if (calvin_locked_rows.contains(row)) {
return RCOK;
}
calvin_locked_rows.add(row);
RC rc = row->get_lock(type, this);
if(rc == WAIT) {
INC_STATS(get_thd_id(), txn_wait_cnt, 1);
}
return rc;
}
void
TxnManager::release() {
uint64_t prof_starttime = get_sys_clock();
qry_pool.put(get_thd_id(),query);
INC_STATS(get_thd_id(),mtx[0],get_sys_clock()-prof_starttime);
query = NULL;
prof_starttime = get_sys_clock();
txn_pool.put(get_thd_id(),txn);
INC_STATS(get_thd_id(),mtx[1],get_sys_clock()-prof_starttime);
txn = NULL;
#if CC_ALG == MAAT
delete uncommitted_writes;
delete uncommitted_writes_y;
delete uncommitted_reads;
#endif
#if CC_ALG == CALVIN
calvin_locked_rows.release();
#endif
txn_ready = true;
}
RC TxnManager::get_row_post_wait(row_t *& row_rtn) {
assert(CC_ALG != HSTORE && CC_ALG != HSTORE_SPEC);
uint64_t starttime = get_sys_clock();
row_t * row = this->last_row;
access_t type = this->last_type;
assert(row != NULL);
DEBUG_M("TxnManager::get_row_post_wait access alloc\n")
Access * access;
access_pool.get(get_thd_id(),access);
row->get_row_post_wait(type,this,access->data);
access->type = type;
access->orig_row = row;
#if ROLL_BACK && (CC_ALG == DL_DETECT || CC_ALG == NO_WAIT || CC_ALG == WAIT_DIE)
if (type == WR) {
uint64_t part_id = row->get_part_id();
//printf("alloc 10 %ld\n",get_txn_id());
DEBUG_M("TxnManager::get_row_post_wait row_t alloc\n")
row_pool.get(get_thd_id(),access->orig_data);
access->orig_data->init(row->get_table(), part_id, 0);
access->orig_data->copy(row);
}
#endif
++txn->row_cnt;
if (type == WR)
++txn->write_cnt;
txn->accesses.add(access);
uint64_t timespan = get_sys_clock() - starttime;
INC_STATS(get_thd_id(), txn_manager_time, timespan);
this->last_row_rtn = access->data;
row_rtn = access->data;
return RCOK;
}
void TxnManager::init(uint64_t thd_id, Workload * h_wl) {
uint64_t prof_starttime = get_sys_clock();
if(!txn) {
DEBUG_M("Transaction alloc\n");
txn_pool.get(thd_id,txn);
}
INC_STATS(get_thd_id(),mtx[15],get_sys_clock()-prof_starttime);
prof_starttime = get_sys_clock();
//txn->init();
if(!query) {
DEBUG_M("TxnManager::init Query alloc\n");
qry_pool.get(thd_id,query);
}
INC_STATS(get_thd_id(),mtx[16],get_sys_clock()-prof_starttime);
//query->init();
//reset();
sem_init(&rsp_mutex, 0, 1);
return_id = UINT64_MAX;
this->h_wl = h_wl;
#if CC_ALG == MAAT
uncommitted_writes = new std::set<uint64_t>();
uncommitted_writes_y = new std::set<uint64_t>();
uncommitted_reads = new std::set<uint64_t>();
#endif
#if CC_ALG == CALVIN
phase = CALVIN_RW_ANALYSIS;
locking_done = false;
calvin_locked_rows.init(MAX_ROW_PER_TXN);
#endif
txn_ready = true;
twopl_wait_start = 0;
txn_stats.init();
}
void InputThread::setup() {
std::vector<Message*> * msgs;
while(!simulation->is_setup_done()) {
msgs = tport_man.recv_msg(get_thd_id());
if(msgs == NULL)
continue;
while(!msgs->empty()) {
Message * msg = msgs->front();
if(msg->rtype == INIT_DONE) {
printf("Received INIT_DONE from node %ld\n",msg->return_node_id);
fflush(stdout);
simulation->process_setup_msg();
} else {
assert(ISSERVER || ISREPLICA);
//printf("Received Msg %d from node %ld\n",msg->rtype,msg->return_node_id);
#if CC_ALG == CALVIN
if(msg->rtype == CALVIN_ACK ||(msg->rtype == CL_QRY && ISCLIENTN(msg->get_return_id()))) {
work_queue.sequencer_enqueue(get_thd_id(),msg);
msgs->erase(msgs->begin());
continue;
}
if( msg->rtype == RDONE || msg->rtype == CL_QRY) {
assert(ISSERVERN(msg->get_return_id()));
work_queue.sched_enqueue(get_thd_id(),msg);
msgs->erase(msgs->begin());
continue;
}
#endif
work_queue.enqueue(get_thd_id(),msg,false);
}
msgs->erase(msgs->begin());
}
delete msgs;
}
}
RC
TxnManager::send_remote_reads() {
assert(CC_ALG == CALVIN);
#if !YCSB_ABORT_MODE && WORKLOAD == YCSB
return RCOK;
#endif
assert(query->active_nodes.size() == g_node_cnt);
for(uint64_t i = 0; i < query->active_nodes.size(); i++) {
if(i == g_node_id)
continue;
if(query->active_nodes[i] == 1) {
DEBUG("(%ld,%ld) send_remote_read to %ld\n",get_txn_id(),get_batch_id(),i);
msg_queue.enqueue(get_thd_id(),Message::create_message(this,RFWD),i);
}
}
return RCOK;
}
RC TxnManager::validate() {
#if MODE != NORMAL_MODE
return RCOK;
#endif
if (CC_ALG != OCC && CC_ALG != MAAT) {
return RCOK;
}
RC rc = RCOK;
uint64_t starttime = get_sys_clock();
if(CC_ALG == OCC && rc == RCOK)
rc = occ_man.validate(this);
if(CC_ALG == MAAT && rc == RCOK) {
rc = maat_man.validate(this);
// Note: home node must be last to validate
if(IS_LOCAL(get_txn_id()) && rc == RCOK) {
rc = maat_man.find_bound(this);
}
}
INC_STATS(get_thd_id(),txn_validate_time,get_sys_clock() - starttime);
return rc;
}
RC YCSBTxnManager::acquire_locks() {
uint64_t starttime = get_sys_clock();
assert(CC_ALG == CALVIN);
YCSBQuery* ycsb_query = (YCSBQuery*) query;
locking_done = false;
RC rc = RCOK;
incr_lr();
assert(ycsb_query->requests.size() == g_req_per_query);
assert(phase == CALVIN_RW_ANALYSIS);
for (uint32_t rid = 0; rid < ycsb_query->requests.size(); rid ++) {
ycsb_request * req = ycsb_query->requests[rid];
uint64_t part_id = _wl->key_to_part( req->key );
DEBUG("LK Acquire (%ld,%ld) %d,%ld -> %ld\n",get_txn_id(),get_batch_id(),req->acctype,req->key,GET_NODE_ID(part_id));
if(GET_NODE_ID(part_id) != g_node_id)
continue;
INDEX * index = _wl->the_index;
itemid_t * item;
item = index_read(index, req->key, part_id);
row_t * row = ((row_t *)item->location);
RC rc2 = get_lock(row,req->acctype);
if(rc2 != RCOK) {
rc = rc2;
}
}
if(decr_lr() == 0) {
if(ATOM_CAS(lock_ready,false,true))
rc = RCOK;
}
txn_stats.wait_starttime = get_sys_clock();
/*
if(rc == WAIT && lock_ready_cnt == 0) {
if(ATOM_CAS(lock_ready,false,true))
//lock_ready = true;
rc = RCOK;
}
*/
INC_STATS(get_thd_id(),calvin_sched_time,get_sys_clock() - starttime);
locking_done = true;
return rc;
}
RC TPCCTxnManager::send_remote_request() {
assert(IS_LOCAL(get_txn_id()));
TPCCQuery* tpcc_query = (TPCCQuery*) query;
TPCCRemTxnType next_state = TPCC_FIN;
uint64_t w_id = tpcc_query->w_id;
uint64_t c_w_id = tpcc_query->c_w_id;
uint64_t dest_node_id = UINT64_MAX;
if(state == TPCC_PAYMENT0) {
dest_node_id = GET_NODE_ID(wh_to_part(w_id));
next_state = TPCC_PAYMENT2;
} else if(state == TPCC_PAYMENT4) {
dest_node_id = GET_NODE_ID(wh_to_part(c_w_id));
next_state = TPCC_FIN;
} else if(state == TPCC_NEWORDER0) {
dest_node_id = GET_NODE_ID(wh_to_part(w_id));
next_state = TPCC_NEWORDER6;
} else if(state == TPCC_NEWORDER8) {
dest_node_id = GET_NODE_ID(wh_to_part(tpcc_query->items[next_item_id]->ol_supply_w_id));
/*
while(GET_NODE_ID(wh_to_part(tpcc_query->items[next_item_id]->ol_supply_w_id)) != dest_node_id) {
msg->items.add(tpcc_query->items[next_item_id++]);
}
*/
if(is_done())
next_state = TPCC_FIN;
else
next_state = TPCC_NEWORDER6;
} else {
assert(false);
}
TPCCQueryMessage * msg = (TPCCQueryMessage*)Message::create_message(this,RQRY);
msg->state = state;
query->partitions_touched.add_unique(GET_PART_ID(0,dest_node_id));
msg_queue.enqueue(get_thd_id(),msg,dest_node_id);
state = next_state;
return WAIT_REM;
}
// reset after abort
void TxnManager::reset() {
lock_ready = false;
lock_ready_cnt = 0;
locking_done = true;
ready_part = 0;
rsp_cnt = 0;
aborted = false;
return_id = UINT64_MAX;
twopl_wait_start = 0;
//ready = true;
// MaaT
greatest_write_timestamp = 0;
greatest_read_timestamp = 0;
commit_timestamp = 0;
#if CC_ALG == MAAT
uncommitted_writes->clear();
uncommitted_writes_y->clear();
uncommitted_reads->clear();
#endif
#if CC_ALG == CALVIN
phase = CALVIN_RW_ANALYSIS;
locking_done = false;
calvin_locked_rows.clear();
#endif
assert(txn);
assert(query);
txn->reset(get_thd_id());
// Stats
txn_stats.reset();
}
RC TxnManager::abort() {
if(aborted)
return Abort;
DEBUG("Abort %ld\n",get_txn_id());
txn->rc = Abort;
INC_STATS(get_thd_id(),total_txn_abort_cnt,1);
txn_stats.abort_cnt++;
if(IS_LOCAL(get_txn_id())) {
INC_STATS(get_thd_id(), local_txn_abort_cnt, 1);
} else {
INC_STATS(get_thd_id(), remote_txn_abort_cnt, 1);
txn_stats.abort_stats(get_thd_id());
}
aborted = true;
release_locks(Abort);
#if CC_ALG == MAAT
//assert(time_table.get_state(get_txn_id()) == MAAT_ABORTED);
time_table.release(get_thd_id(),get_txn_id());
#endif
uint64_t timespan = get_sys_clock() - txn_stats.restart_starttime;
if (IS_LOCAL(get_txn_id()) && warmup_done) {
INC_STATS_ARR(get_thd_id(),start_abort_commit_latency, timespan);
}
/*
// latency from most recent start or restart of transaction
PRINT_LATENCY("lat_s %ld %ld 0 %f %f %f %f %f %f 0.0\n"
, get_txn_id()
, txn_stats.work_queue_cnt
, (double) timespan / BILLION
, (double) txn_stats.work_queue_time / BILLION
, (double) txn_stats.msg_queue_time / BILLION
, (double) txn_stats.cc_block_time / BILLION
, (double) txn_stats.cc_time / BILLION
, (double) txn_stats.process_time / BILLION
);
*/
//commit_stats();
return Abort;
}
RC ClientThread::run() {
tsetup();
printf("Running ClientThread %ld\n",_thd_id);
BaseQuery * m_query;
uint64_t iters = 0;
uint32_t num_txns_sent = 0;
int txns_sent[g_servers_per_client];
for (uint32_t i = 0; i < g_servers_per_client; ++i)
txns_sent[i] = 0;
run_starttime = get_sys_clock();
while(!simulation->is_done()) {
heartbeat();
#if SERVER_GENERATE_QUERIES
break;
#endif
//uint32_t next_node = iters++ % g_node_cnt;
progress_stats();
int32_t inf_cnt;
uint32_t next_node = (((iters++) * g_client_thread_cnt) + _thd_id )% g_servers_per_client;
uint32_t next_node_id = next_node + g_server_start_node;
// Just in case...
if (iters == UINT64_MAX)
iters = 0;
#if LOAD_METHOD == LOAD_MAX
if ((inf_cnt = client_man.inc_inflight(next_node)) < 0)
continue;
m_query = client_query_queue.get_next_query(next_node,_thd_id);
if(last_send_time > 0) {
INC_STATS(get_thd_id(),cl_send_intv,get_sys_clock() - last_send_time);
}
last_send_time = get_sys_clock();
#elif LOAD_METHOD == LOAD_RATE
if ((inf_cnt = client_man.inc_inflight(next_node)) < 0)
continue;
uint64_t gate_time;
while((gate_time = get_sys_clock()) - last_send_time < send_interval) { }
if(last_send_time > 0) {
INC_STATS(get_thd_id(),cl_send_intv,gate_time - last_send_time);
}
last_send_time = gate_time;
m_query = client_query_queue.get_next_query(next_node,_thd_id);
#else
assert(false);
#endif
assert(m_query);
DEBUG("Client: thread %lu sending query to node: %u, %d, %f\n",
_thd_id, next_node_id,inf_cnt,simulation->seconds_from_start(get_sys_clock()));
Message * msg = Message::create_message((BaseQuery*)m_query,CL_QRY);
((ClientQueryMessage*)msg)->client_startts = get_sys_clock();
msg_queue.enqueue(get_thd_id(),msg,next_node_id);
num_txns_sent++;
txns_sent[next_node]++;
INC_STATS(get_thd_id(),txn_sent_cnt,1);
}
for (uint64_t l = 0; l < g_servers_per_client; ++l)
printf("Txns sent to node %lu: %d\n", l+g_server_start_node, txns_sent[l]);
//SET_STATS(get_thd_id(), total_runtime, get_sys_clock() - simulation->run_starttime);
printf("FINISH %ld:%ld\n",_node_id,_thd_id);
fflush(stdout);
return FINISH;
}
RC TPCCTxnManager::acquire_locks() {
uint64_t starttime = get_sys_clock();
assert(CC_ALG == CALVIN);
locking_done = false;
RC rc = RCOK;
RC rc2;
INDEX * index;
itemid_t * item;
row_t* row;
uint64_t key;
incr_lr();
TPCCQuery* tpcc_query = (TPCCQuery*) query;
uint64_t w_id = tpcc_query->w_id;
uint64_t d_id = tpcc_query->d_id;
uint64_t c_id = tpcc_query->c_id;
uint64_t d_w_id = tpcc_query->d_w_id;
uint64_t c_w_id = tpcc_query->c_w_id;
uint64_t c_d_id = tpcc_query->c_d_id;
char * c_last = tpcc_query->c_last;
uint64_t part_id_w = wh_to_part(w_id);
uint64_t part_id_c_w = wh_to_part(c_w_id);
switch(tpcc_query->txn_type) {
case TPCC_PAYMENT:
if(GET_NODE_ID(part_id_w) == g_node_id) {
// WH
index = _wl->i_warehouse;
item = index_read(index, w_id, part_id_w);
row_t * row = ((row_t *)item->location);
rc2 = get_lock(row,g_wh_update? WR:RD);
if(rc2 != RCOK)
rc = rc2;
// Dist
key = distKey(d_id, d_w_id);
item = index_read(_wl->i_district, key, part_id_w);
row = ((row_t *)item->location);
rc2 = get_lock(row, WR);
if(rc2 != RCOK)
rc = rc2;
}
if(GET_NODE_ID(part_id_c_w) == g_node_id) {
// Cust
if (tpcc_query->by_last_name) {
key = custNPKey(c_last, c_d_id, c_w_id);
index = _wl->i_customer_last;
item = index_read(index, key, part_id_c_w);
int cnt = 0;
itemid_t * it = item;
itemid_t * mid = item;
while (it != NULL) {
cnt ++;
it = it->next;
if (cnt % 2 == 0)
mid = mid->next;
}
row = ((row_t *)mid->location);
}
else {
key = custKey(c_id, c_d_id, c_w_id);
index = _wl->i_customer_id;
item = index_read(index, key, part_id_c_w);
row = (row_t *) item->location;
}
rc2 = get_lock(row, WR);
if(rc2 != RCOK)
rc = rc2;
}
break;
case TPCC_NEW_ORDER:
if(GET_NODE_ID(part_id_w) == g_node_id) {
// WH
index = _wl->i_warehouse;
item = index_read(index, w_id, part_id_w);
row_t * row = ((row_t *)item->location);
rc2 = get_lock(row,RD);
if(rc2 != RCOK)
rc = rc2;
// Cust
index = _wl->i_customer_id;
key = custKey(c_id, d_id, w_id);
item = index_read(index, key, wh_to_part(w_id));
row = (row_t *) item->location;
rc2 = get_lock(row, RD);
if(rc2 != RCOK)
rc = rc2;
// Dist
key = distKey(d_id, w_id);
item = index_read(_wl->i_district, key, wh_to_part(w_id));
row = ((row_t *)item->location);
rc2 = get_lock(row, WR);
if(rc2 != RCOK)
rc = rc2;
}
// Items
for(uint64_t i = 0; i < tpcc_query->ol_cnt; i++) {
if(GET_NODE_ID(wh_to_part(tpcc_query->items[i]->ol_supply_w_id)) != g_node_id)
continue;
key = tpcc_query->items[i]->ol_i_id;
item = index_read(_wl->i_item, key, 0);
row = ((row_t *)item->location);
rc2 = get_lock(row, RD);
if(rc2 != RCOK)
rc = rc2;
key = stockKey(tpcc_query->items[i]->ol_i_id, tpcc_query->items[i]->ol_supply_w_id);
index = _wl->i_stock;
item = index_read(index, key, wh_to_part(tpcc_query->items[i]->ol_supply_w_id));
row = ((row_t *)item->location);
rc2 = get_lock(row, WR);
if(rc2 != RCOK)
rc = rc2;
}
break;
default: assert(false);
}
if(decr_lr() == 0) {
if(ATOM_CAS(lock_ready,false,true))
rc = RCOK;
}
txn_stats.wait_starttime = get_sys_clock();
locking_done = true;
INC_STATS(get_thd_id(),calvin_sched_time,get_sys_clock() - starttime);
return rc;
}
void TxnManager::commit_stats() {
uint64_t commit_time = get_sys_clock();
uint64_t timespan_short = commit_time - txn_stats.restart_starttime;
uint64_t timespan_long = commit_time - txn_stats.starttime;
INC_STATS(get_thd_id(),total_txn_commit_cnt,1);
if(!IS_LOCAL(get_txn_id()) && CC_ALG != CALVIN) {
INC_STATS(get_thd_id(),remote_txn_commit_cnt,1);
txn_stats.commit_stats(get_thd_id(),get_txn_id(),get_batch_id(), timespan_long, timespan_short);
return;
}
INC_STATS(get_thd_id(),txn_cnt,1);
INC_STATS(get_thd_id(),local_txn_commit_cnt,1);
INC_STATS(get_thd_id(), txn_run_time, timespan_long);
if(query->partitions_touched.size() > 1) {
INC_STATS(get_thd_id(),multi_part_txn_cnt,1);
INC_STATS(get_thd_id(),multi_part_txn_run_time,timespan_long);
} else {
INC_STATS(get_thd_id(),single_part_txn_cnt,1);
INC_STATS(get_thd_id(),single_part_txn_run_time,timespan_long);
}
/*if(cflt) {
INC_STATS(get_thd_id(),cflt_cnt_txn,1);
}*/
txn_stats.commit_stats(get_thd_id(),get_txn_id(),get_batch_id(),timespan_long, timespan_short);
#if CC_ALG == CALVIN
return;
#endif
INC_STATS_ARR(get_thd_id(),start_abort_commit_latency, timespan_short);
INC_STATS_ARR(get_thd_id(),last_start_commit_latency, timespan_short);
INC_STATS_ARR(get_thd_id(),first_start_commit_latency, timespan_long);
assert(query->partitions_touched.size() > 0);
INC_STATS(get_thd_id(),parts_touched,query->partitions_touched.size());
INC_STATS(get_thd_id(),part_cnt[query->partitions_touched.size()-1],1);
for(uint64_t i = 0 ; i < query->partitions_touched.size(); i++) {
INC_STATS(get_thd_id(),part_acc[query->partitions_touched[i]],1);
}
}
void TxnManager::register_thread(Thread * h_thd) {
this->h_thd = h_thd;
#if CC_ALG == HSTORE || CC_ALG == HSTORE_SPEC
this->active_part = GET_PART_ID_FROM_IDX(get_thd_id());
#endif
}
RC tpcc_txn_man::run_payment(tpcc_query * query) {
RC rc = RCOK;
uint64_t key;
itemid_t * item;
uint64_t w_id = query->w_id;
uint64_t d_id = query->d_id;
uint64_t c_w_id = query->c_w_id;
uint64_t c_d_id = query->c_d_id;
uint64_t c_id = query->c_id;
double h_amount = query->h_amount;
row_t * r_cust;
if (query->by_last_name) {
/*==========================================================+
EXEC SQL SELECT count(c_id) INTO :namecnt
FROM customer
WHERE c_last=:c_last AND c_d_id=:c_d_id AND c_w_id=:c_w_id;
+==========================================================*/
/*==========================================================================+
EXEC SQL DECLARE c_byname CURSOR FOR
SELECT c_first, c_middle, c_id, c_street_1, c_street_2, c_city, c_state,
c_zip, c_phone, c_credit, c_credit_lim, c_discount, c_balance, c_since
FROM customer
WHERE c_w_id=:c_w_id AND c_d_id=:c_d_id AND c_last=:c_last
ORDER BY c_first;
EXEC SQL OPEN c_byname;
+===========================================================================*/
uint64_t key = custNPKey(query->c_last, query->c_d_id, query->c_w_id);
// XXX: the list is not sorted. But let's assume it's sorted...
// The performance won't be much different.
INDEX * index = _wl->i_customer_last;
item = index_read(index, key, wh_to_part(c_w_id));
assert(item != NULL);
int cnt = 0;
itemid_t * it = item;
itemid_t * mid = item;
while (it != NULL) {
cnt ++;
it = it->next;
if (cnt % 2 == 0)
mid = mid->next;
}
r_cust = ((row_t *)mid->location);
/*============================================================================+
for (n=0; n<namecnt/2; n++) {
EXEC SQL FETCH c_byname
INTO :c_first, :c_middle, :c_id,
:c_street_1, :c_street_2, :c_city, :c_state, :c_zip,
:c_phone, :c_credit, :c_credit_lim, :c_discount, :c_balance, :c_since;
}
EXEC SQL CLOSE c_byname;
+=============================================================================*/
// XXX: we don't retrieve all the info, just the tuple we are interested in
}
else { // search customers by cust_id
/*=====================================================================+
EXEC SQL SELECT c_first, c_middle, c_last, c_street_1, c_street_2,
c_city, c_state, c_zip, c_phone, c_credit, c_credit_lim,
c_discount, c_balance, c_since
INTO :c_first, :c_middle, :c_last, :c_street_1, :c_street_2,
:c_city, :c_state, :c_zip, :c_phone, :c_credit, :c_credit_lim,
:c_discount, :c_balance, :c_since
FROM customer
WHERE c_w_id=:c_w_id AND c_d_id=:c_d_id AND c_id=:c_id;
+======================================================================*/
key = custKey(query->c_id, query->c_d_id, query->c_w_id);
INDEX * index = _wl->i_customer_id;
item = index_read(index, key, wh_to_part(c_w_id));
assert(item != NULL);
// rc = index->index_read(key, item, wh_to_part(c_w_id));
// assert(rc == RCOK);
r_cust = (row_t *) item->location;
}
/*======================================================================+
EXEC SQL UPDATE customer SET c_balance = :c_balance, c_data = :c_new_data
WHERE c_w_id = :c_w_id AND c_d_id = :c_d_id AND c_id = :c_id;
+======================================================================*/
row_t * r_cust_local = get_row(r_cust, WR);
if (r_cust_local == NULL) {
return finish(Abort);
}
double c_balance;
double c_ytd_payment;
double c_payment_cnt;
r_cust_local->get_value(C_BALANCE, c_balance);
r_cust_local->set_value(C_BALANCE, c_balance - query->h_amount);
r_cust_local->get_value(C_YTD_PAYMENT, c_ytd_payment);
r_cust_local->set_value(C_YTD_PAYMENT, c_ytd_payment + query->h_amount);
r_cust_local->get_value(C_PAYMENT_CNT, c_payment_cnt);
r_cust_local->set_value(C_PAYMENT_CNT, c_payment_cnt + 1);
char * c_credit = r_cust_local->get_value(C_CREDIT);
if ( strstr(c_credit, "BC") ) {
/*=====================================================+
EXEC SQL SELECT c_data
INTO :c_data
FROM customer
WHERE c_w_id=:c_w_id AND c_d_id=:c_d_id AND c_id=:c_id;
+=====================================================*/
// char c_new_data[501];
// sprintf(c_new_data,"| %4d %2d %4d %2d %4d $%7.2f",
// c_id, c_d_id, c_w_id, d_id, w_id, query->h_amount);
// char * c_data = r_cust->get_value("C_DATA");
// strncat(c_new_data, c_data, 500 - strlen(c_new_data));
// r_cust->set_value("C_DATA", c_new_data);
}
/*=====================================================+
EXEC SQL UPDATE district SET d_ytd = d_ytd + :h_amount
WHERE d_w_id=:w_id AND d_id=:d_id;
+=====================================================*/
uint64_t t3 = get_sys_clock();
key = distKey(query->d_id, query->d_w_id);
item = index_read(_wl->i_district, key, wh_to_part(w_id));
assert(item != NULL);
// rc = _wl->i_district->index_read(key, item, wh_to_part(w_id));
// assert(rc == RCOK);
row_t * r_dist = ((row_t *)item->location);
row_t * r_dist_local = get_row(r_dist, WR);
if (r_dist_local == NULL) {
return finish(Abort);
}
uint64_t tt3 = get_sys_clock() - t3;
INC_STATS(get_thd_id(), debug3, tt3);
double d_ytd;
r_dist_local->get_value(D_YTD, d_ytd);
r_dist_local->set_value(D_YTD, d_ytd + query->h_amount);
/*====================================================+
EXEC SQL UPDATE warehouse SET w_ytd = w_ytd + :h_amount
WHERE w_id=:w_id;
+====================================================*/
/*===================================================================+
EXEC SQL SELECT w_street_1, w_street_2, w_city, w_state, w_zip, w_name
INTO :w_street_1, :w_street_2, :w_city, :w_state, :w_zip, :w_name
FROM warehouse
WHERE w_id=:w_id;
+===================================================================*/
key = query->w_id;
INDEX * index = _wl->i_warehouse;
item = index_read(index, key, wh_to_part(w_id));
assert(item != NULL);
// rc = index->index_read(key, item, wh_to_part(w_id));
// assert(rc == RCOK);
row_t * r_wh = ((row_t *)item->location);
row_t * r_wh_local;
uint64_t t5 = get_sys_clock();
if (g_wh_update)
r_wh_local = get_row(r_wh, WR);
else
r_wh_local = get_row(r_wh, RD);
uint64_t tt5 = get_sys_clock() - t5;
INC_STATS(get_thd_id(), debug5, tt5);
if (r_wh_local == NULL) {
return finish(Abort);
}
double w_ytd;
uint64_t t1 = get_sys_clock();
r_wh_local->get_value(W_YTD, w_ytd);
/*=============================================================================+
EXEC SQL INSERT INTO
history (h_c_d_id, h_c_w_id, h_c_id, h_d_id, h_w_id, h_date, h_amount, h_data)
VALUES (:c_d_id, :c_w_id, :c_id, :d_id, :w_id, :datetime, :h_amount, :h_data);
+=============================================================================*/
row_t * r_hist;
uint64_t row_id;
_wl->t_history->get_new_row(r_hist, 0, row_id);
r_hist->set_value(H_C_ID, c_id);
r_hist->set_value(H_C_D_ID, c_d_id);
r_hist->set_value(H_C_W_ID, c_w_id);
r_hist->set_value(H_D_ID, d_id);
r_hist->set_value(H_W_ID, w_id);
int64_t date = 2013;
r_hist->set_value(H_DATE, date);
r_hist->set_value(H_AMOUNT, h_amount);
insert_row(r_hist, _wl->t_history);
if (g_wh_update) {
r_wh_local->set_value(W_YTD, w_ytd + query->h_amount);
}
uint64_t tt1 = get_sys_clock() - t1;
INC_STATS(get_thd_id(), debug1, tt1);
assert( rc == RCOK );
return finish(rc);
}
RC TxnManager::get_row(row_t * row, access_t type, row_t *& row_rtn) {
uint64_t starttime = get_sys_clock();
uint64_t timespan;
RC rc = RCOK;
DEBUG_M("TxnManager::get_row access alloc\n");
Access * access;
access_pool.get(get_thd_id(),access);
//uint64_t row_cnt = txn->row_cnt;
//assert(txn->accesses.get_count() - 1 == row_cnt);
this->last_row = row;
this->last_type = type;
rc = row->get_row(type, this, access->data);
if (rc == Abort || rc == WAIT) {
row_rtn = NULL;
DEBUG_M("TxnManager::get_row(abort) access free\n");
access_pool.put(get_thd_id(),access);
timespan = get_sys_clock() - starttime;
INC_STATS(get_thd_id(), txn_manager_time, timespan);
INC_STATS(get_thd_id(), txn_conflict_cnt, 1);
//cflt = true;
#if DEBUG_TIMELINE
printf("CONFLICT %ld %ld\n",get_txn_id(),get_sys_clock());
#endif
return rc;
}
access->type = type;
access->orig_row = row;
#if ROLL_BACK && (CC_ALG == DL_DETECT || CC_ALG == NO_WAIT || CC_ALG == WAIT_DIE || CC_ALG == HSTORE || CC_ALG == HSTORE_SPEC)
if (type == WR) {
//printf("alloc 10 %ld\n",get_txn_id());
uint64_t part_id = row->get_part_id();
DEBUG_M("TxnManager::get_row row_t alloc\n")
row_pool.get(get_thd_id(),access->orig_data);
access->orig_data->init(row->get_table(), part_id, 0);
access->orig_data->copy(row);
assert(access->orig_data->get_schema() == row->get_schema());
// ARIES-style physiological logging
#if LOGGING
//LogRecord * record = logger.createRecord(LRT_UPDATE,L_UPDATE,get_txn_id(),part_id,row->get_table()->get_table_id(),row->get_primary_key());
LogRecord * record = logger.createRecord(get_txn_id(),L_UPDATE,row->get_table()->get_table_id(),row->get_primary_key());
if(g_repl_cnt > 0) {
msg_queue.enqueue(get_thd_id(),Message::create_message(record,LOG_MSG),g_node_id + g_node_cnt + g_client_node_cnt);
}
logger.enqueueRecord(record);
#endif
}
#endif
++txn->row_cnt;
if (type == WR)
++txn->write_cnt;
txn->accesses.add(access);
timespan = get_sys_clock() - starttime;
INC_STATS(get_thd_id(), txn_manager_time, timespan);
row_rtn = access->data;
if(CC_ALG == HSTORE || CC_ALG == HSTORE_SPEC || CC_ALG == CALVIN)
assert(rc == RCOK);
return rc;
}
