ClientEON.cpp

Go to the documentation of this file.

/*
** This file is part of eOn.
**
** SPDX-License-Identifier: BSD-3-Clause
**
** Copyright (c) 2010--present, eOn Development Team
** All rights reserved.
**
** Repo:
** https://github.com/TheochemUI/eOn
*/
#ifdef _WIN32
#define WIN32_LEAN_AND_MEAN
#include "eon/EonLogger.h"
#include <windows.h>
#endif
 
#include "eon/BaseStructures.h"
#include "eon/Bundling.h"
#include "eon/CommandLine.h"
#include "eon/EpiCenters.h"
#include "eon/HelperFunctions.h"
#include "eon/Job.h"
#include "eon/Parameters.h"
#include "eon/PotRegistry.h"
#include "eon/Potential.h"
#include "version.h"
#include <cstdlib>
#include <exception>
#include <format>
#include <fstream>
#include <iostream>
#include <string_view>
 
#include <cerrno>
#include <chrono>
#include <cstring>
#include <ctime>
#include <filesystem>
 
#ifdef EONMPI
#include <Python.h>
#include <cstdlib>
#include <fcntl.h>
#include <mpi.h>
#include <sstream>
#endif
 
#if defined WITH_ASE_ORCA || EMBED_PYTHON || WITH_ASE_NWCHEM
#include "eon/PyGuard.h"
#endif
 
#ifdef EONMPIBGP
#include <libgen.h>
#endif
 
// Includes for FPE trapping
#include "eon/fpe_handler.h"
 
#ifdef _WIN32
#include <float.h>
#endif
 
#ifndef _WIN32
#include <sys/resource.h>
#include <sys/time.h>
#include <sys/utsname.h>
#include <unistd.h>
#endif
 
#ifdef __APPLE__
#ifndef __aarch64__
#include <mach/mach.h>
#include <mach/task_info.h>
 
void print_memory_usage() {
  auto *log = eonc::log::get();
  struct task_basic_info t_info;
  mach_msg_type_number_t t_info_count = TASK_BASIC_INFO_COUNT;
 
  if (KERN_SUCCESS != task_info(mach_task_self(), TASK_BASIC_INFO,
                                (task_info_t)&t_info, &t_info_count)) {
    QUILL_LOG_ERROR(log, "Failed to get task info");
    return;
  }
 
  unsigned int rss = t_info.resident_size;
  unsigned int vs = t_info.virtual_size;
  QUILL_LOG_INFO(
      log,
      "\nmemory usage:\nresident size (MB): {:8.2f}\nvirtual size (MB):  "
      "{:8.2f}",
      static_cast<double>(rss) / 1024 / 1024,
      static_cast<double>(vs) / 1024 / 1024);
}
#endif
#endif
 
void printSystemInfo() {
  auto *log = eonc::log::get();
  QUILL_LOG_INFO(log, "eOn Client");
  QUILL_LOG_INFO(log, "{}", VERSION_STRING);
#ifndef __aarch64__
  QUILL_LOG_INFO(log, "OS: {}", OS_INFO);
  QUILL_LOG_INFO(log, "Arch: {}", ARCH);
#endif
 
#ifdef _WIN32
  TCHAR hostname[MAX_COMPUTERNAME_LENGTH + 1];
  DWORD size = sizeof(hostname) / sizeof(hostname[0]);
  if (GetComputerName(hostname, &size)) {
    QUILL_LOG_INFO(log, "Hostname: {}", hostname);
  } else {
    QUILL_LOG_ERROR(log, "Failed to get hostname");
  }
  QUILL_LOG_INFO(log, "PID: {}", GetCurrentProcessId());
#else
  struct utsname systemInfo;
  int status = uname(&systemInfo);
  if (status == 0) {
    QUILL_LOG_INFO(log, "Hostname: {}", systemInfo.nodename);
    QUILL_LOG_INFO(log, "PID: {}", getpid());
  } else {
    QUILL_LOG_ERROR(log, "Failed to get system information");
  }
#endif
 
  std::filesystem::path cwd = std::filesystem::current_path();
  QUILL_LOG_INFO(log, "DIR: {}", cwd.string());
}
 
static int eonClientMain(int argc, char **argv) {
  // --- Start Logging setup
  // Configure backend for optimal performance (see BackendOptions.h)
  quill::BackendOptions backend_options;
  // Use 10us sleep for balanced performance (10x faster than 100us default)
  backend_options.sleep_duration = std::chrono::microseconds{10};
  // Larger initial buffer avoids reallocs in NEB (43 LOG calls/iter)
  backend_options.transit_event_buffer_initial_capacity = 2048;
  // eOn is single-threaded; SPSC queue guarantees ordering, no grace needed
  backend_options.log_timestamp_ordering_grace_period =
      std::chrono::microseconds{0};
  // Flush more frequently for better responsiveness
  backend_options.sink_min_flush_interval = std::chrono::milliseconds{100};
  // Disable per-string printable char scan (eOn logs numeric data only)
  backend_options.check_printable_char = {};
  quill::Backend::start(backend_options);
  auto console_sink =
      quill::Frontend::create_or_get_sink<quill::ConsoleSink>("console");
  auto file_sink = quill::Frontend::create_or_get_sink<quill::FileSink>(
      "client_quill.log",
      []() {
        quill::FileSinkConfig cfg;
        cfg.set_open_mode('w');
        return cfg;
      }(),
      quill::FileEventNotifier{});
  auto *logger = quill::Frontend::create_or_get_logger(
      "combi", {std::move(console_sink), std::move(file_sink)},
      quill::PatternFormatterOptions{"%(message)"},
      quill::ClockSourceType::System);
  logger->set_log_level(quill::LogLevel::TraceL3);
  // Traceback logger
  auto trace_csink =
      quill::Frontend::create_or_get_sink<quill::ConsoleSink>("trace_console");
  auto trace_fsink = quill::Frontend::create_or_get_sink<quill::FileSink>(
      "client_traceback.log",
      []() {
        quill::FileSinkConfig cfg;
        cfg.set_open_mode('w');
        return cfg;
      }(),
      quill::FileEventNotifier{});
  quill::Frontend::create_or_get_logger(
      "_traceback", {std::move(trace_csink), std::move(trace_fsink)},
      quill::PatternFormatterOptions{
          " [%(log_level)] [%(source_location)] [%(caller_function)] \n "
          "%(message)\n[end %(log_level)]"},
      quill::ClockSourceType::System);
  //--- End logging setup
  Parameters parameters;
 
#if defined WITH_ASE_ORCA || EMBED_PYTHON || WITH_ASE_NWCHEM
  eonc::ensure_interpreter();
#endif
 
#ifdef EONMPI
  bool client_standalone = false;
  if (getenv("EON_CLIENT_STANDALONE") != nullptr) {
    client_standalone = true;
  }
  int number_of_clients;
  if (!client_standalone) {
    if (getenv("EON_SERVER_PATH") == nullptr) {
      QUILL_LOG_ERROR(logger, "error: must set the env var EON_SERVER_PATH");
      logger->flush_log();
      return 1;
    }
    if (getenv("EON_NUMBER_OF_CLIENTS") == nullptr) {
      QUILL_LOG_ERROR(logger,
                      "error: must set the env var EON_NUMBER_OF_CLIENTS");
      logger->flush_log();
      return 1;
    }
    number_of_clients = atoi(getenv("EON_NUMBER_OF_CLIENTS"));
  } else {
    number_of_clients = 1;
  }
 
  int eon_mpi_inited = 0;
  MPI_Initialized(&eon_mpi_inited);
  if (!eon_mpi_inited) {
    int eon_mpi_provided = 0;
    MPI_Init_thread(&argc, &argv, MPI_THREAD_MULTIPLE, &eon_mpi_provided);
  }
 
  int error;
  std::string config_file = "config.ini";
  if (client_standalone) {
    if (eonc::helpers::existsFile("config_0.ini")) {
      config_file = "config_0.ini";
    }
    QUILL_LOG_INFO(logger, "Loading parameter file {}", config_file);
    error = parameters.load(config_file);
  } else {
    QUILL_LOG_INFO(logger, "Loading parameter file {}",
                   parameters.main_options.iniFilename);
    error = parameters.load(parameters.main_options.iniFilename);
  }
  if (error) {
    QUILL_LOG_ERROR(logger, "problem loading parameter file");
    logger->flush_log();
    MPI_Abort(MPI_COMM_WORLD, 1);
  }
 
  // XXX: Barrier for gpaw-python
  MPI_Barrier(MPI_COMM_WORLD);
 
  int irank;
  MPI_Comm_rank(MPI_COMM_WORLD, &irank);
  int isize;
  MPI_Comm_size(MPI_COMM_WORLD, &isize);
 
  std::vector<int> process_types(isize);
  int process_type;
 
  process_type = 1;
 
  MPI_Allgather(&process_type, 1, MPI_INT, &process_types[0], 1, MPI_INT,
                MPI_COMM_WORLD);
 
  int i, servers = 0, clients = 0, potentials = 0;
  int server_rank = -1;
  int my_client_number = -1;
  std::vector<int> client_ranks;
  for (i = 0; i < isize; i++) {
    switch (process_types[i]) {
    case 0:
      servers++;
      break;
    case 1:
      if (i == irank) {
        my_client_number = clients;
      }
      clients++;
      client_ranks.push_back(i);
      break;
    case 2:
      potentials++;
      break;
    }
  }
 
  if (clients < number_of_clients) {
    QUILL_LOG_ERROR(logger,
                    "didn't launch as many mpi client ranks as specified in "
                    "EON_NUMBER_OF_CLIENTS");
    logger->flush_log();
    MPI_Abort(MPI_COMM_WORLD, 1);
  }
  clients = number_of_clients;
 
  if (parameters.potential_options.potential == PotType::MPI) {
    std::vector<int> potential_ranks(potentials);
    int j;
    for (i = 0, j = 0; i < isize; i++) {
      if (process_types[i] == 2) {
        potential_ranks[j] = i;
        j++;
      }
    }
    int potential_group_size = potentials / clients;
 
    for (i = 0; i < clients; i++) {
      MPI_Group orig_group, new_group;
      MPI_Comm_group(MPI_COMM_WORLD, &orig_group);
      int offset = i * potential_group_size;
      MPI_Group_incl(orig_group, potential_group_size, &potential_ranks[offset],
                     &new_group);
      MPI_Comm pot_comm;
      MPI_Comm_create(MPI_COMM_WORLD, new_group, &pot_comm);
    }
 
    if (my_client_number < number_of_clients) {
      parameters.potential_options.MPIPotentialRank =
          potential_ranks[my_client_number * potential_group_size];
    }
  }
 
  // LAMMPS MPI communicator setup (runtime check, not compile-time)
  if (parameters.potential_options.potential == PotType::LAMMPS) {
    for (i = 0; i < static_cast<int>(client_ranks.size()); i++) {
      MPI_Group world_group, new_group;
      MPI_Comm_group(MPI_COMM_WORLD, &world_group);
      int r = client_ranks[i];
      MPI_Group_incl(world_group, 1, &r, &new_group);
      MPI_Comm new_comm;
      MPI_Comm_create(MPI_COMM_WORLD, new_group, &new_comm);
      if (new_comm != MPI_COMM_NULL) {
        parameters.potential_options.MPIClientComm = new_comm;
      }
      QUILL_LOG_INFO(logger, "creating group with ranks: {}", r);
    }
  }
 
  if (!client_standalone) {
    server_rank = client_ranks[number_of_clients];
    if (my_client_number == number_of_clients) {
      std::ostringstream oss;
      oss << client_ranks.at(0);
      for (i = 1; i < number_of_clients; i++) {
        oss << ":" << client_ranks.at(i);
      }
      setenv("EON_CLIENT_RANKS", oss.str().c_str(), 1);
 
      wchar_t **py_argv =
          static_cast<wchar_t **>(malloc(sizeof(wchar_t *) * 2));
      py_argv[0] = Py_DecodeLocale(argv[0], nullptr);
      char *program = getenv("EON_SERVER_PATH");
      py_argv[1] = Py_DecodeLocale(program, nullptr);
      QUILL_LOG_INFO(logger, "rank: {} becoming {}", irank, program);
      Py_Initialize();
      Py_Main(2, py_argv);
      Py_FinalizeEx();
      // GH
      MPI_Finalize();
      return 0;
    } else if (my_client_number > number_of_clients) {
      MPI_Finalize();
      return 0;
    }
  }
#endif
 
#ifndef EONMPI
  if (argc > 1) {
    eonc::commandLine(argc, argv);
    return 0;
  }
#endif
 
  eonc::enableFPE(); // from ExceptionsEON.h
 
  auto start_time = std::chrono::steady_clock::now();
 
#ifdef EONMPI
  // XXX: When do we stop? The server should probably tell everyone when to
  // stop.
  char logfilename[1024];
  snprintf(logfilename, 1024, "eonclient_%i.log", my_client_number);
 
  auto orig_path = std::filesystem::current_path();
  while (true) {
    std::filesystem::current_path(orig_path);
    std::string path(1024, '\0');
    int ready = 1;
    if (!client_standalone) {
      QUILL_LOG_INFO(
          logger, "client: rank {} is ready, posting send to server rank: {}!",
          irank, server_rank);
      // Tag "1" is to interrupt the main loop and tell the communicator that a
      // client is ready
      {
        MPI_Request eon_rq;
        MPI_Isend(&ready, 1, MPI_INT, server_rank, 1, MPI_COMM_WORLD, &eon_rq);
        MPI_Request_free(&eon_rq);
      }
 
      // Get the path we should run in from the server
      MPI_Recv(&path[0], 1024, MPI_CHAR, server_rank, 0, MPI_COMM_WORLD,
               MPI_STATUS_IGNORE);
      if (path.starts_with("STOPCAR")) {
        QUILL_LOG_INFO(logger, "rank {} got STOPCAR", irank);
        MPI_Finalize();
        return 0;
      }
      QUILL_LOG_INFO(logger, "client: rank: {} chdir to {}", irank, path);
 
      try {
        std::filesystem::current_path(path.c_str());
      } catch (const std::filesystem::filesystem_error &e) {
        QUILL_LOG_ERROR(logger, "error: chdir: {}", e.what());
      }
    }
#endif
 
    printSystemInfo();
 
    // XXX(rg): Be more gentle here
    bool bundlingEnabled = false;
    int bundleSize = -1; // eonc::getBundleSize();
    if (bundleSize == 0) {
      bundleSize = 1;
    } else if (bundleSize == -1) {
      // Not using bundling
      bundleSize = 1;
      bundlingEnabled = false;
    }
 
    std::vector<std::string> bundledFilenames;
    for (int i = 0; i < bundleSize; i++) {
      if (bundleSize > 1)
        QUILL_LOG_INFO(logger, "Beginning Job {} of {}", i + 1, bundleSize);
      std::vector<std::string> unbundledFilenames;
      if (bundlingEnabled) {
        unbundledFilenames = eonc::unbundle(i);
      }
 
      // check to see if parameters file exists before loading
      int error = 0;
      std::string config_file =
          eonc::helpers::getRelevantFile(parameters.main_options.iniFilename);
      QUILL_LOG_INFO(logger, "Loading parameter file {}", config_file);
      error = parameters.load(config_file);
 
      if (error) {
        QUILL_LOG_ERROR(logger, "problem loading parameter file, stopping");
        logger->flush_log();
        exit(1);
        abort();
      }
 
      // Determine what type of job we are running according to the parameters
      // file.
      auto job =
          eonc::helpers::makeJob(std::make_unique<Parameters>(parameters));
      if (job == nullptr) {
        QUILL_LOG_ERROR(logger, "error: Unknown job: {}",
                        std::string{magic_enum::enum_name<JobType>(
                            parameters.main_options.job)});
        logger->flush_log();
        return 1;
      }
 
      std::vector<std::string> filenames;
      try {
        filenames = job->run();
      } catch (int e) {
        QUILL_LOG_CRITICAL(logger, "[ERROR] job exited on error {}", e);
        logger->flush_log();
      } catch (const std::exception &e) {
        QUILL_LOG_CRITICAL(logger, "[ERROR] unhandled exception: {}", e.what());
        logger->flush_log();
        std::cerr << "[ERROR] unhandled exception: " << e.what() << "\n";
        return EXIT_FAILURE;
      }
 
      job.reset(); // Force Potential destruction so PotRegistry records entries
      PotRegistry::get().write_summary();
      filenames.push_back(std::string("_potcalls.json"));
      filenames.push_back(std::string("client_quill.log"));
      filenames.push_back(std::string("client_traceback.log"));
 
      {
        std::ofstream manifest("return_files.dat");
        if (manifest) {
          for (const auto &fn : filenames) {
            manifest << fn << "\n";
          }
        }
        filenames.push_back(std::string("return_files.dat"));
      }
 
      // Finalize Timing Information
      auto end_time = std::chrono::steady_clock::now();
      std::chrono::duration<double> elapsed = end_time - start_time;
 
      double utime = 0, stime = 0, rtime = 0;
      eonc::helpers::getTime(&rtime, &utime, &stime);
 
      QUILL_LOG_INFO(logger, "Timing Information:");
      QUILL_LOG_INFO(logger, "  Real time: {:.3f} seconds", elapsed.count());
      QUILL_LOG_INFO(logger, "  User time: {:.3f} seconds", utime);
      QUILL_LOG_INFO(logger, "  System time: {:.3f} seconds", stime);
 
      // results.dat contract is "<value> <key>" (same as all job writers and
      // eon.fileio.parse_results / eon_schema.jobs adapters).
      std::ofstream result_file("results.dat", std::ios::app);
      if (result_file.is_open()) {
        result_file << std::format("{:.12e} time_seconds\n", elapsed.count());
#ifndef _WIN32
        result_file << std::format("{:.12e} user_time\n", utime);
        result_file << std::format("{:.12e} system_time\n", stime);
#endif
      } else {
        QUILL_LOG_ERROR(logger, "Failed to write timing to results.dat");
      }
 
      if (bundlingEnabled) {
        eonc::bundle(i, filenames, &bundledFilenames);
        eonc::deleteUnbundledFiles(unbundledFilenames);
      } else {
        bundledFilenames = filenames;
      }
    }
 
#ifdef EONMPI
    if (client_standalone) {
      break;
    }
    {
      MPI_Request eon_rq;
      MPI_Isend(&path[0], 1024, MPI_CHAR, server_rank, 0, MPI_COMM_WORLD,
                &eon_rq);
      MPI_Request_free(&eon_rq);
    }
 
    // End of MPI while loop
  }
#endif
 
#ifdef OSX
#ifndef __aarch64__
  print_memory_usage();
#endif
#endif
 
#ifdef EONMPI
  // Clean shutdown for both the standalone single-rank case and the
  // client/server case; the standalone path previously aborted MPI_COMM_WORLD.
  MPI_Finalize();
#endif
 
  // Ensure all queued log messages are flushed before exiting
  quill::Backend::stop();
  return EXIT_SUCCESS;
}
 
namespace {
 
int reportFatal(std::string_view what) {
  std::cerr << "eonclient: fatal error: " << what << '\n';
  // Drain the queued CRITICAL lines that say why: an escaping exception
  // reaches std::terminate without them, and so does std::exit.
  quill::Backend::stop();
#ifdef EONMPI
  int mpiReady = 0;
  MPI_Initialized(&mpiReady);
  if (mpiReady) {
    MPI_Abort(MPI_COMM_WORLD, EXIT_FAILURE);
  }
#endif
  return EXIT_FAILURE;
}
 
} // namespace
 
int main(int argc, char **argv) {
  try {
    return eonClientMain(argc, argv);
  } catch (const std::exception &e) {
    return reportFatal(e.what());
  } catch (...) {
    return reportFatal("exception of unknown type");
  }
}