src/bindings/python/simgrid_python.cpp

   1 /* Copyright (c) 2018-2021. The SimGrid Team. All rights reserved.          */
   2
   3 /* This program is free software; you can redistribute it and/or modify it
   4  * under the terms of the license (GNU LGPL) which comes with this package. */
   5
   6 #ifdef _WIN32
   7 #warning Try to work around https://bugs.python.org/issue11566
   8 #define _hypot hypot
   9 #endif
  10
  11 #if defined(__GNUG__)
  12 #pragma GCC diagnostic push
  13 #pragma GCC diagnostic ignored "-Wunused-value"
  14 #endif
  15
  16 #include <pybind11/functional.h>
  17 #include <pybind11/pybind11.h> // Must come before our own stuff
  18 #include <pybind11/stl.h>
  19
  20 #if defined(__GNUG__)
  21 #pragma GCC diagnostic pop
  22 #endif
  23
  24 #include "src/kernel/context/Context.hpp"
  25 #include <simgrid/Exception.hpp>
  26 #include <simgrid/s4u/Actor.hpp>
  27 #include <simgrid/s4u/Comm.hpp>
  28 #include <simgrid/s4u/Engine.hpp>
  29 #include <simgrid/s4u/Exec.hpp>
  30 #include <simgrid/s4u/Host.hpp>
  31 #include <simgrid/s4u/Mailbox.hpp>
  32 #include <simgrid/version.h>
  33
  34 #include <algorithm>
  35 #include <memory>
  36 #include <string>
  37 #include <vector>
  38
  39 namespace py = pybind11;
  40 using simgrid::s4u::Actor;
  41 using simgrid::s4u::ActorPtr;
  42 using simgrid::s4u::Engine;
  43 using simgrid::s4u::Host;
  44 using simgrid::s4u::Mailbox;
  45
  46 XBT_LOG_NEW_DEFAULT_CATEGORY(python, "python");
  47
  48 namespace {
  49
  50 std::string get_simgrid_version()
  51 {
  52   int major;
  53   int minor;
  54   int patch;
  55   sg_version_get(&major, &minor, &patch);
  56   return simgrid::xbt::string_printf("%i.%i.%i", major, minor, patch);
  57 }
  58
  59 /* Classes GilScopedAcquire and GilScopedRelease have the same purpose as pybind11::gil_scoped_acquire and
  60  * pybind11::gil_scoped_release.  Refer to the manual of pybind11 for details:
  61  * https://pybind11.readthedocs.io/en/stable/advanced/misc.html#global-interpreter-lock-gil
  62  *
  63  * The pybind11 versions are however too sophisticated (using TLS for example) and don't work well with all kinds of
  64  * contexts.
  65  * See also https://github.com/pybind/pybind11/issues/1276, which may be related.
  66  *
  67  * Briefly, GilScopedAcquire can be used on actor creation to acquire a new PyThreadState.  The PyThreadState has to be
  68  * released for context switches (i.e. before simcalls). That's the purpose of GilScopedRelease.
  69  *
  70  * Like their pybind11 counterparts, both classes use a RAII pattern.
  71  */
  72 class XBT_PRIVATE GilScopedAcquire {
  73   static PyThreadState* acquire()
  74   {
  75     PyThreadState* state = PyThreadState_New(PyInterpreterState_Head());
  76     PyEval_AcquireThread(state);
  77     return state;
  78   }
  79   static void release(PyThreadState* state)
  80   {
  81     PyEval_ReleaseThread(state);
  82     PyThreadState_Clear(state);
  83     PyThreadState_Delete(state);
  84   }
  85
  86   std::unique_ptr<PyThreadState, decltype(&release)> thread_state{acquire(), &release};
  87
  88 public:
  89   void reset() { thread_state.reset(); }
  90 };
  91
  92 class XBT_PRIVATE GilScopedRelease {
  93   std::unique_ptr<PyThreadState, decltype(&PyEval_RestoreThread)> thread_state{PyEval_SaveThread(),
  94                                                                                &PyEval_RestoreThread};
  95 };
  96
  97 } // namespace
  98
  99 PYBIND11_DECLARE_HOLDER_TYPE(T, boost::intrusive_ptr<T>)
 100
 101 PYBIND11_MODULE(simgrid, m)
 102 {
 103   m.doc() = "SimGrid userspace API";
 104
 105   m.attr("simgrid_version") = get_simgrid_version();
 106
 107   // Internal exception used to kill actors and sweep the RAII chimney (free objects living on the stack)
 108   static py::object pyForcefulKillEx(py::register_exception<simgrid::ForcefulKillException>(m, "ActorKilled"));
 109
 110   /* this_actor namespace */
 111   m.def_submodule("this_actor", "Bindings of the s4u::this_actor namespace.")
 112       .def(
 113           "info", [](const char* s) { XBT_INFO("%s", s); }, "Display a logging message of 'info' priority.")
 114       .def(
 115           "error", [](const char* s) { XBT_ERROR("%s", s); }, "Display a logging message of 'error' priority.")
 116       .def("execute", py::overload_cast<double, double>(&simgrid::s4u::this_actor::execute),
 117            py::call_guard<GilScopedRelease>(),
 118            "Block the current actor, computing the given amount of flops at the given priority.", py::arg("flops"),
 119            py::arg("priority") = 1)
 120       .def("exec_init", py::overload_cast<double>(&simgrid::s4u::this_actor::exec_init),
 121            py::call_guard<GilScopedRelease>())
 122       .def("get_host", &simgrid::s4u::this_actor::get_host, "Retrieves host on which the current actor is located")
 123       .def("set_host", &simgrid::s4u::this_actor::set_host, py::call_guard<GilScopedRelease>(),
 124            "Moves the current actor to another host.", py::arg("dest"))
 125       .def("sleep_for", static_cast<void (*)(double)>(&simgrid::s4u::this_actor::sleep_for),
 126            py::call_guard<GilScopedRelease>(), "Block the actor sleeping for that amount of seconds.",
 127            py::arg("duration"))
 128       .def("sleep_until", static_cast<void (*)(double)>(&simgrid::s4u::this_actor::sleep_until),
 129            py::call_guard<GilScopedRelease>(), "Block the actor sleeping until the specified timestamp.",
 130            py::arg("duration"))
 131       .def("suspend", &simgrid::s4u::this_actor::suspend, py::call_guard<GilScopedRelease>(),
 132            "Suspend the current actor, that is blocked until resume()ed by another actor.")
 133       .def("yield_", &simgrid::s4u::this_actor::yield, py::call_guard<GilScopedRelease>(), "Yield the actor")
 134       .def("exit", &simgrid::s4u::this_actor::exit, py::call_guard<GilScopedRelease>(), "kill the current actor")
 135       .def(
 136           "on_exit",
 137           [](py::object fun) {
 138             fun.inc_ref(); // FIXME: why is this needed for tests like actor-kill and actor-lifetime?
 139             simgrid::s4u::this_actor::on_exit([fun](bool /*failed*/) {
 140               GilScopedAcquire py_context; // need a new context for callback
 141               try {
 142                 fun();
 143               } catch (const py::error_already_set& e) {
 144                 std::string what = e.what();
 145                 py_context.reset();
 146                 xbt_die("Error while executing the on_exit lambda: %s", what.c_str());
 147               }
 148             });
 149           },
 150           py::call_guard<GilScopedRelease>(), "");
 151
 152   /* Class Engine */
 153   py::class_<Engine>(m, "Engine", "Simulation Engine")
 154       .def(py::init([](std::vector<std::string> args) {
 155         auto argc           = static_cast<int>(args.size());
 156         std::vector<char*> argv(args.size() + 1); // argv[argc] is nullptr
 157         std::transform(begin(args), end(args), begin(argv), [](std::string& s) { return &s.front(); });
 158         // Currently this can be dangling, we should wrap this somehow.
 159         return new simgrid::s4u::Engine(&argc, argv.data());
 160       }))
 161       .def_static("get_clock", &Engine::get_clock,
 162                   "The simulation time, ie the amount of simulated seconds since the simulation start.")
 163       .def("get_all_hosts", &Engine::get_all_hosts, "Returns the list of all hosts found in the platform")
 164       .def("load_platform", &Engine::load_platform, "Load a platform file describing the environment")
 165       .def("load_deployment", &Engine::load_deployment, "Load a deployment file and launch the actors that it contains")
 166       .def("run", &Engine::run, py::call_guard<GilScopedRelease>(), "Run the simulation")
 167       .def(
 168           "register_actor",
 169           [](Engine* e, const std::string& name, py::object fun_or_class) {
 170             e->register_actor(name, [fun_or_class](std::vector<std::string> args) {
 171               GilScopedAcquire py_context;
 172               try {
 173                 /* Convert the std::vector into a py::tuple */
 174                 py::tuple params(args.size() - 1);
 175                 for (size_t i = 1; i < args.size(); i++)
 176                   params[i - 1] = py::cast(args[i]);
 177
 178                 py::object res = fun_or_class(*params);
 179
 180                 /* If I was passed a class, I just built an instance, so I need to call it now */
 181                 if (py::isinstance<py::function>(res))
 182                   res();
 183               } catch (const py::error_already_set& ex) {
 184                 bool ffk = ex.matches(pyForcefulKillEx);
 185                 py_context.reset();
 186                 if (ffk) {
 187                   XBT_VERB("Actor killed");
 188                   /* Forward that ForcefulKill exception */
 189                   simgrid::ForcefulKillException::do_throw();
 190                 }
 191                 throw;
 192               }
 193             });
 194           },
 195           "Registers the main function of an actor");
 196
 197   /* Class Host */
 198   py::class_<simgrid::s4u::Host, std::unique_ptr<Host, py::nodelete>>(m, "Host", "Simulated host")
 199       .def("by_name", &Host::by_name, "Retrieves a host from its name, or die")
 200       .def("get_pstate_count", &Host::get_pstate_count, "Retrieve the count of defined pstate levels")
 201       .def("get_pstate_speed", &Host::get_pstate_speed, "Retrieve the maximal speed at the given pstate")
 202       .def_property(
 203           "pstate", &Host::get_pstate,
 204           [](Host* h, int i) {
 205             GilScopedRelease gil_guard;
 206             h->set_pstate(i);
 207           },
 208           "The current pstate")
 209       .def("current", &Host::current, py::call_guard<GilScopedRelease>(),
 210            "Retrieves the host on which the running actor is located.")
 211       .def_property_readonly(
 212           "name",
 213           [](const Host* self) {
 214             return std::string(self->get_name().c_str()); // Convert from xbt::string because of MC
 215           },
 216           "The name of this host")
 217       .def_property_readonly(
 218           "load", &Host::get_load,
 219           "Returns the current computation load (in flops per second). This is the currently achieved speed.")
 220       .def_property_readonly(
 221           "speed", &Host::get_speed,
 222           "The peak computing speed in flops/s at the current pstate, taking the external load into account. "
 223           "This is the max potential speed.");
 224
 225   /* Class Mailbox */
 226   py::class_<simgrid::s4u::Mailbox, std::unique_ptr<Mailbox, py::nodelete>>(m, "Mailbox", "Mailbox")
 227       .def(
 228           "__str__", [](const Mailbox* self) { return std::string("Mailbox(") + self->get_cname() + ")"; },
 229           "Textual representation of the Mailbox`")
 230       .def("by_name", &Mailbox::by_name, py::call_guard<GilScopedRelease>(), "Retrieve a Mailbox from its name")
 231       .def_property_readonly(
 232           "name",
 233           [](const Mailbox* self) {
 234             return std::string(self->get_name().c_str()); // Convert from xbt::string because of MC
 235           },
 236           "The name of that mailbox")
 237       .def(
 238           "put",
 239           [](Mailbox* self, py::object data, int size) {
 240             data.inc_ref();
 241             self->put(data.ptr(), size);
 242           },
 243           py::call_guard<GilScopedRelease>(), "Blocking data transmission")
 244       .def(
 245           "put_async",
 246           [](Mailbox* self, py::object data, int size) {
 247             data.inc_ref();
 248             return self->put_async(data.ptr(), size);
 249           },
 250           py::call_guard<GilScopedRelease>(), "Non-blocking data transmission")
 251       .def(
 252           "get",
 253           [](Mailbox* self) {
 254             py::object data = pybind11::reinterpret_steal<py::object>(self->get<PyObject>());
 255             // data.dec_ref(); // FIXME: why does it break python-actor-create?
 256             return data;
 257           },
 258           py::call_guard<GilScopedRelease>(), "Blocking data reception")
 259       .def("set_receiver",
 260          [](Mailbox* self, ActorPtr actor) {
 261            self->set_receiver(actor);
 262          },
 263          py::call_guard<GilScopedRelease>(),
 264          "Sets the actor as permanent receiver");
 265
 266   /* Class Comm */
 267   py::class_<simgrid::s4u::Comm, simgrid::s4u::CommPtr>(m, "Comm", "Communication")
 268       .def("test", &simgrid::s4u::Comm::test, py::call_guard<GilScopedRelease>(),
 269            "Test whether the communication is terminated.")
 270       .def("wait", &simgrid::s4u::Comm::wait, py::call_guard<GilScopedRelease>(),
 271            "Block until the completion of that communication.")
 272       .def("wait_all", &simgrid::s4u::Comm::wait_all, py::call_guard<GilScopedRelease>(),
 273            "Block until the completion of all communications in the list.")
 274       .def("wait_any", &simgrid::s4u::Comm::wait_any, py::call_guard<GilScopedRelease>(),
 275            "Block until the completion of any communication in the list and return the index of the terminated one.");
 276
 277   /* Class Exec */
 278   py::class_<simgrid::s4u::Exec, simgrid::s4u::ExecPtr>(m, "Exec", "Execution")
 279       .def_property_readonly(
 280           "remaining",
 281           [](simgrid::s4u::ExecPtr self) {
 282             GilScopedRelease gil_guard;
 283             return self->get_remaining();
 284           },
 285           "Amount of flops that remain to be computed until completion.")
 286       .def_property_readonly(
 287           "remaining_ratio",
 288           [](simgrid::s4u::ExecPtr self) {
 289             GilScopedRelease gil_guard;
 290             return self->get_remaining_ratio();
 291           },
 292           "Amount of work remaining until completion from 0 (completely done) to 1 (nothing done "
 293           "yet).")
 294       .def_property("host", &simgrid::s4u::Exec::get_host, &simgrid::s4u::Exec::set_host,
 295                     "Host on which this execution runs. Only the first host is returned for parallel executions.")
 296       .def("test", &simgrid::s4u::Exec::test, py::call_guard<GilScopedRelease>(),
 297            "Test whether the execution is terminated.")
 298       .def("cancel", &simgrid::s4u::Exec::cancel, py::call_guard<GilScopedRelease>(), "Cancel that execution.")
 299       .def("start", &simgrid::s4u::Exec::start, py::call_guard<GilScopedRelease>(), "Start that execution.")
 300       .def("wait", &simgrid::s4u::Exec::wait, py::call_guard<GilScopedRelease>(),
 301            "Block until the completion of that execution.");
 302
 303   /* Class Actor */
 304   py::class_<simgrid::s4u::Actor, ActorPtr>(m, "Actor",
 305                                             "An actor is an independent stream of execution in your distributed "
 306                                             "application")
 307       .def(
 308           "create",
 309           [](py::str name, Host* host, py::object fun, py::args args) {
 310             fun.inc_ref();  // FIXME: why is this needed for tests like exec-async, exec-dvfs and exec-remote?
 311             args.inc_ref(); // FIXME: why is this needed for tests like actor-migrate?
 312             return simgrid::s4u::Actor::create(name, host, [fun, args]() {
 313               GilScopedAcquire py_context;
 314               try {
 315                 fun(*args);
 316               } catch (const py::error_already_set& ex) {
 317                 bool ffk = ex.matches(pyForcefulKillEx);
 318                 py_context.reset();
 319                 if (ffk) {
 320                   XBT_VERB("Actor killed");
 321                   /* Forward that ForcefulKill exception */
 322                   simgrid::ForcefulKillException::do_throw();
 323                 }
 324                 throw;
 325               }
 326             });
 327           },
 328           py::call_guard<GilScopedRelease>(), "Create an actor from a function or an object.")
 329       .def_property(
 330           "host", &Actor::get_host,
 331           [](Actor* a, Host* h) {
 332             GilScopedRelease gil_guard;
 333             a->set_host(h);
 334           },
 335           "The host on which this actor is located")
 336       .def_property_readonly("name", &Actor::get_cname, "The name of this actor.")
 337       .def_property_readonly("pid", &Actor::get_pid, "The PID (unique identifier) of this actor.")
 338       .def_property_readonly("ppid", &Actor::get_ppid,
 339                              "The PID (unique identifier) of the actor that created this one.")
 340       .def("by_pid", &Actor::by_pid, "Retrieve an actor by its PID")
 341       .def("daemonize", &Actor::daemonize, py::call_guard<GilScopedRelease>(),
 342            "This actor will be automatically terminated when the last non-daemon actor finishes (more info in the C++ "
 343            "documentation).")
 344       .def("is_daemon", &Actor::is_daemon,
 345            "Returns True if that actor is a daemon and will be terminated automatically when the last non-daemon actor "
 346            "terminates.")
 347       .def("join", py::overload_cast<double>(&Actor::join, py::const_), py::call_guard<GilScopedRelease>(),
 348            "Wait for the actor to finish (more info in the C++ documentation).", py::arg("timeout"))
 349       .def("kill", &Actor::kill, py::call_guard<GilScopedRelease>(), "Kill that actor")
 350       .def("kill_all", &Actor::kill_all, py::call_guard<GilScopedRelease>(), "Kill all actors but the caller.")
 351       .def("self", &Actor::self, "Retrieves the current actor.")
 352       .def("is_suspended", &Actor::is_suspended, "Returns True if that actor is currently suspended.")
 353       .def("suspend", &Actor::suspend, py::call_guard<GilScopedRelease>(),
 354            "Suspend that actor, that is blocked until resume()ed by another actor.")
 355       .def("resume", &Actor::resume, py::call_guard<GilScopedRelease>(),
 356            "Resume that actor, that was previously suspend()ed.");
 357 }