00001 #include "MPIProg.hh"
00002 #include "artdaq/DAQrate/detail/RoutingPacket.hh"
00003 #include "artdaq/DAQdata/TCPConnect.hh"
00004 #include "artdaq/DAQrate/quiet_mpi.hh"
00005 #include "cetlib/filepath_maker.h"
00006 #include "fhiclcpp/ParameterSet.h"
00007 #include "fhiclcpp/make_ParameterSet.h"
00008
00009 #include <boost/program_options.hpp>
00010 #include <boost/filesystem.hpp>
00011 #include "artdaq/Application/RoutingMasterCore.hh"
00012 #include "artdaq/Application/RoutingMasterApp.hh"
00013 #include <netdb.h>
00014 namespace bpo = boost::program_options;
00015
00016 #include <algorithm>
00017 #include <cmath>
00018 #include <cstdio>
00019
00020 extern "C"
00021 {
00022 #include <unistd.h>
00023 }
00024
00025 #include <iostream>
00026 #include <memory>
00027 #include <utility>
00028 #include <arpa/inet.h>
00029 #include <netinet/in.h>
00030 #include <sys/types.h>
00031 #include <sys/socket.h>
00032
00033 extern "C"
00034 {
00035 #include <sys/time.h>
00036 #include <sys/resource.h>
00037 }
00038
00042 class LockFile
00043 {
00044 public:
00049 explicit LockFile(std::string const& path) : fileName_(path)
00050 {
00051 std::ofstream fstream(fileName_);
00052 fstream << "Locked" << std::endl;
00053 }
00054
00058 ~LockFile()
00059 {
00060 if(IsLocked(fileName_)) remove(fileName_.c_str());
00061 }
00067 static bool IsLocked(std::string const& path)
00068 {
00069 return boost::filesystem::exists(path);
00070 }
00071
00072 private:
00073 std::string fileName_;
00074 };
00075
00079 class RoutingMasterTest : public MPIProg
00080 {
00081 public:
00096 RoutingMasterTest(int argc, char* argv[]);
00097
00101 void go();
00102
00106 void generate_tokens();
00107
00111 void routing_master();
00112
00116 void table_receiver();
00117
00124 fhicl::ParameterSet getPset(int argc, char* argv[]) const;
00125
00126 private:
00127 enum class TestRole_t : int
00128 {
00129 TOKEN_GEN = 0,
00130 ROUTING_MASTER = 1,
00131 TABLE_RECEIVER = 2
00132 };
00133
00134 void printHost(const std::string& functionName) const;
00135
00136 fhicl::ParameterSet const pset_;
00137 fhicl::ParameterSet const daq_pset_;
00138 TestRole_t role_;
00139
00140 std::string routing_master_address_;
00141 std::string multicast_address_;
00142 int token_port_;
00143 int table_port_;
00144 int ack_port_;
00145 std::vector<int> eb_ranks_;
00146 int token_count_;
00147 size_t token_interval_us_;
00148 };
00149
00150 RoutingMasterTest::RoutingMasterTest(int argc, char* argv[]) :
00151 MPIProg(argc, argv)
00152 , pset_(getPset(argc, argv))
00153 , daq_pset_(pset_.get<fhicl::ParameterSet>("daq"))
00154 , routing_master_address_(daq_pset_.get<std::string>("routing_master_hostname", "localhost"))
00155 , multicast_address_(daq_pset_.get<std::string>("table_update_address", "227.128.12.28"))
00156 , token_port_(daq_pset_.get<int>("routing_token_port", 35555))
00157 , table_port_(daq_pset_.get<int>("table_update_port", 35556))
00158 , ack_port_(daq_pset_.get<int>("table_acknowledge_port", 35557))
00159 , token_count_(pset_.get<int>("token_count", 1000))
00160 , token_interval_us_(pset_.get<size_t>("token_interval_us", 5000))
00161 {
00162 assert(!(my_rank < 0));
00163 switch (my_rank)
00164 {
00165 case 0:
00166 role_ = TestRole_t::TOKEN_GEN;
00167 break;
00168 case 1:
00169 role_ = TestRole_t::ROUTING_MASTER;
00170 break;
00171 default:
00172 role_ = TestRole_t::TABLE_RECEIVER;
00173 break;
00174 }
00175 auto policy_pset = daq_pset_.get<fhicl::ParameterSet>("policy");
00176 eb_ranks_ = policy_pset.get<std::vector<int>>("receiver_ranks");
00177
00178 }
00179
00180 fhicl::ParameterSet RoutingMasterTest::getPset(int argc, char* argv[]) const
00181 {
00182 std::ostringstream descstr;
00183 descstr << "-- <-c <config-file>>";
00184 bpo::options_description desc(descstr.str());
00185 desc.add_options()
00186 ("config,c", bpo::value<std::string>(), "Configuration file.");
00187 bpo::variables_map vm;
00188 try
00189 {
00190 bpo::store(bpo::command_line_parser(argc, argv).
00191 options(desc).allow_unregistered().run(), vm);
00192 bpo::notify(vm);
00193 }
00194 catch (bpo::error const& e)
00195 {
00196 std::cerr << "Exception from command line processing in Config::getArtPset: " << e.what() << "\n";
00197 throw "cmdline parsing error.";
00198 }
00199 if (!vm.count("config"))
00200 {
00201 std::cerr << "Expected \"-- -c <config-file>\" fhicl file specification.\n";
00202 throw "cmdline parsing error.";
00203 }
00204 fhicl::ParameterSet pset;
00205 cet::filepath_lookup lookup_policy("FHICL_FILE_PATH");
00206 fhicl::make_ParameterSet(vm["config"].as<std::string>(), lookup_policy, pset);
00207
00208 return pset;
00209 }
00210
00211 void RoutingMasterTest::go()
00212 {
00213 if (LockFile::IsLocked("/tmp/routing_master_t.lock")) return;
00214 MPI_Barrier(MPI_COMM_WORLD);
00215 std::unique_ptr<LockFile> lock;
00216 if (my_rank == 0) {
00217 lock = std::make_unique<LockFile>("/tmp/routing_master_t.lock");
00218 }
00219
00220
00221 switch (role_)
00222 {
00223 case TestRole_t::TABLE_RECEIVER:
00224 table_receiver();
00225 break;
00226 case TestRole_t::ROUTING_MASTER:
00227 routing_master();
00228 break;
00229 case TestRole_t::TOKEN_GEN:
00230 generate_tokens();
00231 break;
00232 default:
00233 throw "No such node type";
00234 }
00235 TLOG_DEBUG("routing_master") << "Rank " << my_rank << " complete." << TLOG_ENDL;
00236 }
00237
00238 void RoutingMasterTest::generate_tokens()
00239 {
00240 TLOG_DEBUG("generate_tokens") << "Init" << TLOG_ENDL;
00241 printHost("generate_tokens");
00242 sleep(1);
00243
00244 int token_socket = socket(AF_INET, SOCK_STREAM, IPPROTO_TCP);
00245 if (token_socket < 0)
00246 {
00247 TLOG_ERROR("generate_tokens") << "I failed to create the socket for sending Routing Tokens!" << TLOG_ENDL;
00248 exit(1);
00249 }
00250 struct sockaddr_in token_addr;
00251 auto sts = ResolveHost(routing_master_address_.c_str(), token_port_, token_addr);
00252 if(sts == -1)
00253 {
00254 TLOG_ERROR("generate_tokens") << "Could not resolve host name" << TLOG_ENDL;
00255 }
00256
00257 connect(token_socket, (struct sockaddr*)&token_addr, sizeof(token_addr));
00258
00259 int sent_tokens = 0;
00260 std::map<int, int> token_counter;
00261 for(auto rank : eb_ranks_)
00262 {
00263 token_counter[rank] = 0;
00264 }
00265 while (sent_tokens < token_count_) {
00266 int this_rank = eb_ranks_[seedAndRandom() % eb_ranks_.size()];
00267 token_counter[this_rank]++;
00268 artdaq::detail::RoutingToken token;
00269 token.header = TOKEN_MAGIC;
00270 token.rank = this_rank;
00271 token.new_slots_free = 1;
00272
00273 TLOG_DEBUG("generate_tokens") << "Sending RoutingToken " << std::to_string(++sent_tokens) << " for rank " << this_rank << " to " << routing_master_address_ << TLOG_ENDL;
00274 send(token_socket, &token, sizeof(artdaq::detail::RoutingToken), 0);
00275 usleep(token_interval_us_);
00276 }
00277 auto max_rank = 0;
00278 for(auto rank : token_counter)
00279 {
00280 if (rank.second > max_rank) max_rank = rank.second;
00281 }
00282 for(auto rank : token_counter)
00283 {
00284 artdaq::detail::RoutingToken token;
00285 token.header = TOKEN_MAGIC;
00286 token.rank = rank.first;
00287 token.new_slots_free = max_rank - rank.second;
00288
00289 TLOG_DEBUG("generate_tokens") << "Sending RoutingToken " << std::to_string(++sent_tokens) << " for rank " << rank.first << " to " << routing_master_address_ << TLOG_ENDL;
00290 send(token_socket, &token, sizeof(artdaq::detail::RoutingToken), 0);
00291 usleep(token_interval_us_);
00292
00293 }
00294
00295 TLOG_INFO("generate_tokens") << "Waiting at MPI_Barrier" << TLOG_ENDL;
00296 MPI_Barrier(MPI_COMM_WORLD);
00297 TLOG_INFO("generate_tokens") << "Done with MPI_Barrier" << TLOG_ENDL;
00298 }
00299
00300 void RoutingMasterTest::table_receiver()
00301 {
00302 TLOG_DEBUG("table_receiver") << "Init" << TLOG_ENDL;
00303 printHost("table_receiver");
00304
00305
00306 auto table_socket = socket(AF_INET, SOCK_DGRAM, IPPROTO_UDP);
00307 if (table_socket < 0)
00308 {
00309 TLOG_ERROR("table_receiver") << "Error creating socket for receiving data requests!" << TLOG_ENDL;
00310 exit(1);
00311 }
00312
00313 struct sockaddr_in si_me_request;
00314
00315 int yes = 1;
00316 if (setsockopt(table_socket, SOL_SOCKET, SO_REUSEADDR, &yes, sizeof(yes)) < 0)
00317 {
00318 TLOG_ERROR("table_receiver") << " Unable to enable port reuse on request socket" << TLOG_ENDL;
00319 exit(1);
00320 }
00321 memset(&si_me_request, 0, sizeof(si_me_request));
00322 si_me_request.sin_family = AF_INET;
00323 si_me_request.sin_port = htons(table_port_);
00324 si_me_request.sin_addr.s_addr = htonl(INADDR_ANY);
00325 if (bind(table_socket, (struct sockaddr *)&si_me_request, sizeof(si_me_request)) == -1)
00326 {
00327 TLOG_ERROR("table_receiver") << "Cannot bind request socket to port " << table_port_ << TLOG_ENDL;
00328 exit(1);
00329 }
00330
00331 struct ip_mreq mreq;
00332 long int sts = ResolveHost(multicast_address_.c_str(), mreq.imr_multiaddr);
00333 if(sts == -1)
00334 {
00335 TLOG_ERROR("table_Receiver") << "Unable to resolve multicast hostname" << TLOG_ENDL;
00336 exit(1);
00337 }
00338 mreq.imr_interface.s_addr = htonl(INADDR_ANY);
00339 if (setsockopt(table_socket, IPPROTO_IP, IP_ADD_MEMBERSHIP, &mreq, sizeof(mreq)) < 0)
00340 {
00341 TLOG_ERROR("table_receiver") << "Unable to join multicast group" << TLOG_ENDL;
00342 exit(1);
00343 }
00344
00345 struct epoll_event ev;
00346 int table_epoll_fd = epoll_create1(0);
00347 ev.events = EPOLLIN | EPOLLPRI;
00348 ev.data.fd = table_socket;
00349 if (epoll_ctl(table_epoll_fd, EPOLL_CTL_ADD, table_socket, &ev) == -1)
00350 {
00351 TLOG_ERROR("table_receiver") << "Could not register listen socket to epoll fd" << TLOG_ENDL;
00352 exit(3);
00353 }
00354
00355 auto ack_socket = socket(AF_INET, SOCK_DGRAM, IPPROTO_UDP);
00356 struct sockaddr_in ack_addr;
00357 sts = ResolveHost(routing_master_address_.c_str(), ack_port_, ack_addr);
00358 if(sts == -1)
00359 {
00360 TLOG_ERROR("table_Receiver") << "Unable to resolve routing master hostname" << TLOG_ENDL;
00361 exit(1);
00362 }
00363
00364 if (table_socket == -1 || table_epoll_fd == -1 || ack_socket == -1)
00365 {
00366 TLOG_DEBUG("table_receiver") << "One of the listen sockets was not opened successfully." << TLOG_ENDL;
00367 exit(4);
00368 }
00369 artdaq::Fragment::sequence_id_t max_sequence_id = token_count_;
00370 artdaq::Fragment::sequence_id_t current_sequence_id = 0;
00371 std::map<artdaq::Fragment::sequence_id_t, int> routing_table;
00372 TLOG_INFO("table_receiver") << "Expecting " << std::to_string(max_sequence_id) << " as the last Sequence ID in this run" << TLOG_ENDL;
00373 while (current_sequence_id < max_sequence_id)
00374 {
00375 std::vector<epoll_event> table_events_(4);
00376 TLOG_DEBUG("table_receiver") << "Waiting for event on table socket" << TLOG_ENDL;
00377 auto nfds = epoll_wait(table_epoll_fd, &table_events_[0], table_events_.size(), -1);
00378 if (nfds == -1) {
00379 perror("epoll_wait");
00380 exit(EXIT_FAILURE);
00381 }
00382
00383 TLOG_DEBUG("table_receiver") << "Received " << nfds << " table update(s)" << TLOG_ENDL;
00384 for (auto n = 0; n < nfds; ++n) {
00385 auto first = artdaq::Fragment::InvalidSequenceID;
00386 auto last = artdaq::Fragment::InvalidSequenceID;
00387 artdaq::detail::RoutingPacketHeader hdr;
00388 recv(table_events_[n].data.fd, &hdr, sizeof(artdaq::detail::RoutingPacketHeader), 0);
00389
00390 TLOG_DEBUG("table_receiver") << "Checking for valid header" << TLOG_ENDL;
00391 if (hdr.header == ROUTING_MAGIC) {
00392 artdaq::detail::RoutingPacket buffer(hdr.nEntries);
00393 TLOG_DEBUG("table_receiver") << "Receiving data buffer" << TLOG_ENDL;
00394 sts = recv(table_events_[n].data.fd, &buffer[0], sizeof(artdaq::detail::RoutingPacketEntry) * hdr.nEntries, 0);
00395 assert(static_cast<size_t>(sts) == sizeof(artdaq::detail::RoutingPacketEntry) * hdr.nEntries);
00396
00397 first = buffer[0].sequence_id;
00398 last = buffer[buffer.size() - 1].sequence_id;
00399
00400 for (auto entry : buffer)
00401 {
00402 if (routing_table.count(entry.sequence_id))
00403 {
00404 assert(routing_table[entry.sequence_id] == entry.destination_rank);
00405 continue;
00406 }
00407 routing_table[entry.sequence_id] = entry.destination_rank;
00408 TLOG_DEBUG("table_receiver") << "table_receiver " << std::to_string(my_rank) << ": received update: SeqID " << std::to_string(entry.sequence_id) << " -> Rank " << std::to_string(entry.destination_rank) << TLOG_ENDL;
00409 }
00410
00411 artdaq::detail::RoutingAckPacket ack;
00412 ack.rank = my_rank;
00413 ack.first_sequence_id = first;
00414 ack.last_sequence_id = last;
00415
00416 TLOG_DEBUG("table_receiver") << "Sending RoutingAckPacket with first= " << std::to_string(first) << " and last= " << std::to_string(last) << " to " << routing_master_address_ << ", port " << ack_port_ << TLOG_ENDL;
00417 sendto(ack_socket, &ack, sizeof(artdaq::detail::RoutingAckPacket), 0, (struct sockaddr *)&ack_addr, sizeof(ack_addr));
00418 current_sequence_id = last;
00419 }
00420 }
00421 }
00422
00423 TLOG_INFO("table_receiver") << "Waiting at MPI_Barrier" << TLOG_ENDL;
00424 MPI_Barrier(MPI_COMM_WORLD);
00425 TLOG_INFO("table_receiver") << "Done with MPI_Barrier" << TLOG_ENDL;
00426 }
00427
00428 void RoutingMasterTest::routing_master()
00429 {
00430 TLOG_DEBUG("routing_master") << "Init" << TLOG_ENDL;
00431 printHost("routing_master");
00432
00433 auto app = std::make_unique<artdaq::RoutingMasterApp>(MPI_COMM_WORLD, "RoutingMaster");
00434
00435 app->initialize(pset_, 0, 0);
00436 app->do_start(art::RunID(1), 0, 0);
00437 TLOG_INFO("routing_master") << "Waiting at MPI_Barrier" << TLOG_ENDL;
00438 MPI_Barrier(MPI_COMM_WORLD);
00439 TLOG_INFO("routing_master") << "Done with MPI_Barrier, calling RoutingMasterCore::stop" << TLOG_ENDL;
00440 app->do_stop(0, 0);
00441 TLOG_INFO("routing_master") << "Done with RoutingMasterCore::stop, calling shutdown" << TLOG_ENDL;
00442 app->do_shutdown(0);
00443 TLOG_INFO("routing_master") << "Done with RoutingMasterCore::shutdown" << TLOG_ENDL;
00444 }
00445
00446 void RoutingMasterTest::printHost(const std::string& functionName) const
00447 {
00448 char* doPrint = getenv("PRINT_HOST");
00449 if (doPrint == 0) { return; }
00450 const int ARRSIZE = 80;
00451 char hostname[ARRSIZE];
00452 std::string hostString;
00453 if (!gethostname(hostname, ARRSIZE))
00454 {
00455 hostString = hostname;
00456 }
00457 else
00458 {
00459 hostString = "unknown";
00460 }
00461 TLOG_DEBUG("routing_master") << "Running " << functionName
00462 << " on host " << hostString
00463 << " with rank " << my_rank << "."
00464 << TLOG_ENDL;
00465 }
00466
00467 void printUsage()
00468 {
00469 int myid = 0;
00470 struct rusage usage;
00471 getrusage(RUSAGE_SELF, &usage);
00472 std::cout << myid << ":"
00473 << " user=" << artdaq::TimeUtils::convertUnixTimeToSeconds(usage.ru_utime)
00474 << " sys=" << artdaq::TimeUtils::convertUnixTimeToSeconds(usage.ru_stime)
00475 << std::endl;
00476 }
00477
00478 int main(int argc, char* argv[])
00479 {
00480 artdaq::configureMessageFacility("routing_master", false);
00481 int rc = 1;
00482 try
00483 {
00484 RoutingMasterTest p(argc, argv);
00485 std::cerr << "Started process " << my_rank << " of " << p.procs_ << ".\n";
00486 p.go();
00487 rc = 0;
00488 }
00489 catch (std::string& x)
00490 {
00491 std::cerr << "Exception (type string) caught in routing_master: "
00492 << x
00493 << '\n';
00494 return 1;
00495 }
00496 catch (char const* m)
00497 {
00498 std::cerr << "Exception (type char const*) caught in routing_master: ";
00499 if (m)
00500 {
00501 std::cerr << m;
00502 }
00503 else
00504 {
00505 std::cerr << "[the value was a null pointer, so no message is available]";
00506 }
00507 std::cerr << '\n';
00508 }
00509 return rc;
00510 }