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 #ifndef NDEBUG
  17 /* Many tests are failing after pybind11 commit ad6bf5cd39ca64b4a9bf846b84b11c4c8df1c8e1 "Adding PyGILState_Check() in
  18  *  object_api<>::operator(). (#2919)".
  19  * See https://github.com/pybind/pybind11/commit/ad6bf5cd39ca64b4a9bf846b84b11c4c8df1c8e1
  20  *
  21  * The failing tests are mostly those with boost/raw/sysv contexts. As a workaround, define NDEBUG before pybind11
  22  * includes.
  23  */
  24 #define NDEBUG
  25 #define NDEBUG_LOCALLY_DEFINED
  26 #endif
  27
  28 #include <pybind11/pybind11.h> // Must come before our own stuff
  29
  30 #include <pybind11/functional.h>
  31 #include <pybind11/stl.h>
  32
  33 #ifdef NDEBUG_LOCALLY_DEFINED
  34 #undef NDEBUG_LOCALLY_DEFINED
  35 #undef NDEBUG
  36 #endif
  37
  38 #if defined(__GNUG__)
  39 #pragma GCC diagnostic pop
  40 #endif
  41
  42 #include "simgrid/kernel/routing/NetPoint.hpp"
  43 #include "src/kernel/context/Context.hpp"
  44 #include <simgrid/Exception.hpp>
  45 #include <simgrid/s4u/Actor.hpp>
  46 #include <simgrid/s4u/Comm.hpp>
  47 #include <simgrid/s4u/Engine.hpp>
  48 #include <simgrid/s4u/Exec.hpp>
  49 #include <simgrid/s4u/Host.hpp>
  50 #include <simgrid/s4u/Link.hpp>
  51 #include <simgrid/s4u/Mailbox.hpp>
  52 #include <simgrid/s4u/NetZone.hpp>
  53 #include <simgrid/version.h>
  54
  55 #include <algorithm>
  56 #include <memory>
  57 #include <string>
  58 #include <vector>
  59
  60 namespace py = pybind11;
  61 using simgrid::s4u::Actor;
  62 using simgrid::s4u::ActorPtr;
  63 using simgrid::s4u::Engine;
  64 using simgrid::s4u::Host;
  65 using simgrid::s4u::Mailbox;
  66
  67 XBT_LOG_NEW_DEFAULT_CATEGORY(python, "python");
  68
  69 namespace {
  70
  71 std::string get_simgrid_version()
  72 {
  73   int major;
  74   int minor;
  75   int patch;
  76   sg_version_get(&major, &minor, &patch);
  77   return simgrid::xbt::string_printf("%i.%i.%i", major, minor, patch);
  78 }
  79
  80 /** @brief Wrap for mailbox::get_async */
  81 class PyGetAsync {
  82   std::unique_ptr<PyObject*> data = std::make_unique<PyObject*>();
  83
  84 public:
  85   PyObject** get() const { return data.get(); }
  86 };
  87
  88 /* Classes GilScopedAcquire and GilScopedRelease have the same purpose as pybind11::gil_scoped_acquire and
  89  * pybind11::gil_scoped_release.  Refer to the manual of pybind11 for details:
  90  * https://pybind11.readthedocs.io/en/stable/advanced/misc.html#global-interpreter-lock-gil
  91  *
  92  * The pybind11 versions are however too sophisticated (using TLS for example) and don't work well with all kinds of
  93  * contexts.
  94  * See also https://github.com/pybind/pybind11/issues/1276, which may be related.
  95  *
  96  * Briefly, GilScopedAcquire can be used on actor creation to acquire a new PyThreadState.  The PyThreadState has to be
  97  * released for context switches (i.e. before simcalls). That's the purpose of GilScopedRelease.
  98  *
  99  * Like their pybind11 counterparts, both classes use a RAII pattern.
 100  */
 101 class XBT_PRIVATE GilScopedAcquire {
 102   static PyThreadState* acquire()
 103   {
 104     PyThreadState* state = PyThreadState_New(PyInterpreterState_Head());
 105     PyEval_AcquireThread(state);
 106     return state;
 107   }
 108   static void release(PyThreadState* state)
 109   {
 110     PyEval_ReleaseThread(state);
 111     PyThreadState_Clear(state);
 112     PyThreadState_Delete(state);
 113   }
 114
 115   std::unique_ptr<PyThreadState, decltype(&release)> thread_state{acquire(), &release};
 116
 117 public:
 118   void reset() { thread_state.reset(); }
 119 };
 120
 121 class XBT_PRIVATE GilScopedRelease {
 122   std::unique_ptr<PyThreadState, decltype(&PyEval_RestoreThread)> thread_state{PyEval_SaveThread(),
 123                                                                                &PyEval_RestoreThread};
 124 };
 125
 126 } // namespace
 127
 128 PYBIND11_DECLARE_HOLDER_TYPE(T, boost::intrusive_ptr<T>)
 129
 130 PYBIND11_MODULE(simgrid, m)
 131 {
 132   m.doc() = "SimGrid userspace API";
 133
 134   m.attr("simgrid_version") = get_simgrid_version();
 135
 136   // Internal exception used to kill actors and sweep the RAII chimney (free objects living on the stack)
 137   static py::object pyForcefulKillEx(py::register_exception<simgrid::ForcefulKillException>(m, "ActorKilled"));
 138
 139   /* this_actor namespace */
 140   m.def_submodule("this_actor", "Bindings of the s4u::this_actor namespace.")
 141       .def(
 142           "info", [](const char* s) { XBT_INFO("%s", s); }, "Display a logging message of 'info' priority.")
 143       .def(
 144           "error", [](const char* s) { XBT_ERROR("%s", s); }, "Display a logging message of 'error' priority.")
 145       .def("execute", py::overload_cast<double, double>(&simgrid::s4u::this_actor::execute),
 146            py::call_guard<GilScopedRelease>(),
 147            "Block the current actor, computing the given amount of flops at the given priority.", py::arg("flops"),
 148            py::arg("priority") = 1)
 149       .def("exec_init", py::overload_cast<double>(&simgrid::s4u::this_actor::exec_init),
 150            py::call_guard<GilScopedRelease>())
 151       .def("get_host", &simgrid::s4u::this_actor::get_host, "Retrieves host on which the current actor is located")
 152       .def("set_host", &simgrid::s4u::this_actor::set_host, py::call_guard<GilScopedRelease>(),
 153            "Moves the current actor to another host.", py::arg("dest"))
 154       .def("sleep_for", static_cast<void (*)(double)>(&simgrid::s4u::this_actor::sleep_for),
 155            py::call_guard<GilScopedRelease>(), "Block the actor sleeping for that amount of seconds.",
 156            py::arg("duration"))
 157       .def("sleep_until", static_cast<void (*)(double)>(&simgrid::s4u::this_actor::sleep_until),
 158            py::call_guard<GilScopedRelease>(), "Block the actor sleeping until the specified timestamp.",
 159            py::arg("duration"))
 160       .def("suspend", &simgrid::s4u::this_actor::suspend, py::call_guard<GilScopedRelease>(),
 161            "Suspend the current actor, that is blocked until resume()ed by another actor.")
 162       .def("yield_", &simgrid::s4u::this_actor::yield, py::call_guard<GilScopedRelease>(), "Yield the actor")
 163       .def("exit", &simgrid::s4u::this_actor::exit, py::call_guard<GilScopedRelease>(), "kill the current actor")
 164       .def(
 165           "on_exit",
 166           [](py::object fun) {
 167             fun.inc_ref(); // FIXME: why is this needed for tests like actor-kill and actor-lifetime?
 168             simgrid::s4u::this_actor::on_exit([fun](bool /*failed*/) {
 169               GilScopedAcquire py_context; // need a new context for callback
 170               try {
 171                 fun();
 172               } catch (const py::error_already_set& e) {
 173                 std::string what = e.what();
 174                 py_context.reset();
 175                 xbt_die("Error while executing the on_exit lambda: %s", what.c_str());
 176               }
 177             });
 178           },
 179           py::call_guard<GilScopedRelease>(), "");
 180
 181   /* Class Engine */
 182   py::class_<Engine>(m, "Engine", "Simulation Engine")
 183       .def(py::init([](std::vector<std::string> args) {
 184         auto argc           = static_cast<int>(args.size());
 185         std::vector<char*> argv(args.size() + 1); // argv[argc] is nullptr
 186         std::transform(begin(args), end(args), begin(argv), [](std::string& s) { return &s.front(); });
 187         // Currently this can be dangling, we should wrap this somehow.
 188         return new simgrid::s4u::Engine(&argc, argv.data());
 189       }))
 190       .def_static("get_clock", &Engine::get_clock,
 191                   "The simulation time, ie the amount of simulated seconds since the simulation start.")
 192       .def("get_all_hosts", &Engine::get_all_hosts, "Returns the list of all hosts found in the platform")
 193       .def("load_platform", &Engine::load_platform, "Load a platform file describing the environment")
 194       .def("load_deployment", &Engine::load_deployment, "Load a deployment file and launch the actors that it contains")
 195       .def("run", &Engine::run, py::call_guard<GilScopedRelease>(), "Run the simulation")
 196       .def(
 197           "register_actor",
 198           [](Engine* e, const std::string& name, py::object fun_or_class) {
 199             e->register_actor(name, [fun_or_class](std::vector<std::string> args) {
 200               GilScopedAcquire py_context;
 201               try {
 202                 /* Convert the std::vector into a py::tuple */
 203                 py::tuple params(args.size() - 1);
 204                 for (size_t i = 1; i < args.size(); i++)
 205                   params[i - 1] = py::cast(args[i]);
 206
 207                 py::object res = fun_or_class(*params);
 208                 /* If I was passed a class, I just built an instance, so I need to call it now */
 209                 if (py::isinstance<py::function>(res))
 210                   res();
 211               } catch (const py::error_already_set& ex) {
 212                 bool ffk = ex.matches(pyForcefulKillEx);
 213                 py_context.reset();
 214                 if (ffk) {
 215                   XBT_VERB("Actor killed");
 216                   simgrid::ForcefulKillException::do_throw(); // Forward that ForcefulKill exception
 217                 }
 218                 throw;
 219               }
 220             });
 221           },
 222           "Registers the main function of an actor");
 223
 224   /* Class Netzone */
 225   py::class_<simgrid::s4u::NetZone, std::unique_ptr<simgrid::s4u::NetZone, py::nodelete>>(m, "NetZone",
 226                                                                                           "Networking Zones")
 227       .def_static("create_full_zone", &simgrid::s4u::create_full_zone, "Creates a netzone of type FullZone")
 228       .def("add_route",
 229            py::overload_cast<simgrid::kernel::routing::NetPoint*, simgrid::kernel::routing::NetPoint*,
 230                              simgrid::kernel::routing::NetPoint*, simgrid::kernel::routing::NetPoint*,
 231                              const std::vector<simgrid::s4u::LinkInRoute>&, bool>(&simgrid::s4u::NetZone::add_route),
 232            "Add a route between 2 netpoints")
 233       .def("create_host", py::overload_cast<const std::string&, double>(&simgrid::s4u::NetZone::create_host),
 234            "Creates a host")
 235       .def("create_split_duplex_link",
 236            py::overload_cast<const std::string&, double>(&simgrid::s4u::NetZone::create_split_duplex_link),
 237            "Creates a split-duplex link")
 238       .def("seal", &simgrid::s4u::NetZone::seal, "Seal this NetZone");
 239
 240   /* Class Host */
 241   py::class_<simgrid::s4u::Host, std::unique_ptr<Host, py::nodelete>>(m, "Host", "Simulated host")
 242       .def("by_name", &Host::by_name, "Retrieves a host from its name, or die")
 243       .def("get_pstate_count", &Host::get_pstate_count, "Retrieve the count of defined pstate levels")
 244       .def("get_pstate_speed", &Host::get_pstate_speed, "Retrieve the maximal speed at the given pstate")
 245       .def("get_netpoint", &Host::get_netpoint, "Retrieve the netpoint associated to this host")
 246       .def("seal", &Host::seal, "Seal this host")
 247       .def_property(
 248           "pstate", &Host::get_pstate,
 249           [](Host* h, int i) {
 250             GilScopedRelease gil_guard;
 251             h->set_pstate(i);
 252           },
 253           "The current pstate")
 254       .def("current", &Host::current, py::call_guard<GilScopedRelease>(),
 255            "Retrieves the host on which the running actor is located.")
 256       .def_property_readonly(
 257           "name",
 258           [](const Host* self) {
 259             return std::string(self->get_name().c_str()); // Convert from xbt::string because of MC
 260           },
 261           "The name of this host")
 262       .def_property_readonly(
 263           "load", &Host::get_load,
 264           "Returns the current computation load (in flops per second). This is the currently achieved speed.")
 265       .def_property_readonly(
 266           "speed", &Host::get_speed,
 267           "The peak computing speed in flops/s at the current pstate, taking the external load into account. "
 268           "This is the max potential speed.");
 269
 270   /* Class NetPoint */
 271   py::class_<simgrid::kernel::routing::NetPoint, std::unique_ptr<simgrid::kernel::routing::NetPoint, py::nodelete>>(
 272       m, "NetPoint", "NetPoint object");
 273
 274   /* Class Link */
 275   py::class_<simgrid::s4u::Link, std::unique_ptr<simgrid::s4u::Link, py::nodelete>> link(m, "Link", "Network link");
 276   link.def("set_latency", py::overload_cast<const std::string&>(&simgrid::s4u::Link::set_latency), "Set the latency");
 277   link.def("set_latency", py::overload_cast<double>(&simgrid::s4u::Link::set_latency), "Set the latency");
 278   link.def("set_sharing_policy", &simgrid::s4u::Link::set_sharing_policy, "Set sharing policy for this link");
 279   link.def("seal", &simgrid::s4u::Link::seal, "Seal this link");
 280   link.def_property_readonly(
 281       "name",
 282       [](const simgrid::s4u::Link* self) {
 283         return std::string(self->get_name().c_str()); // Convert from xbt::string because of MC
 284       },
 285       "The name of this link");
 286   py::enum_<simgrid::s4u::Link::SharingPolicy>(link, "SharingPolicy")
 287       .value("NONLINEAR", simgrid::s4u::Link::SharingPolicy::NONLINEAR)
 288       .value("WIFI", simgrid::s4u::Link::SharingPolicy::WIFI)
 289       .value("SPLITDUPLEX", simgrid::s4u::Link::SharingPolicy::SPLITDUPLEX)
 290       .value("SHARED", simgrid::s4u::Link::SharingPolicy::SHARED)
 291       .value("FATPIPE", simgrid::s4u::Link::SharingPolicy::FATPIPE)
 292       .export_values();
 293
 294   /* Class LinkInRoute */
 295   py::class_<simgrid::s4u::LinkInRoute> linkinroute(m, "LinkInRoute", "Abstraction to add link in routes");
 296   linkinroute.def(py::init<const simgrid::s4u::Link*>());
 297   linkinroute.def(py::init<const simgrid::s4u::Link*, simgrid::s4u::LinkInRoute::Direction>());
 298   py::enum_<simgrid::s4u::LinkInRoute::Direction>(linkinroute, "Direction")
 299       .value("UP", simgrid::s4u::LinkInRoute::Direction::UP)
 300       .value("DOWN", simgrid::s4u::LinkInRoute::Direction::DOWN)
 301       .value("NONE", simgrid::s4u::LinkInRoute::Direction::NONE)
 302       .export_values();
 303
 304   /* Class Split-Duplex Link */
 305   py::class_<simgrid::s4u::SplitDuplexLink, simgrid::s4u::Link,
 306              std::unique_ptr<simgrid::s4u::SplitDuplexLink, py::nodelete>>(m, "SplitDuplexLink",
 307                                                                            "Network split-duplex link")
 308       .def("get_link_up", &simgrid::s4u::SplitDuplexLink::get_link_up, "Get link direction up")
 309       .def("get_link_down", &simgrid::s4u::SplitDuplexLink::get_link_down, "Get link direction down");
 310
 311   /* Class Mailbox */
 312   py::class_<simgrid::s4u::Mailbox, std::unique_ptr<Mailbox, py::nodelete>>(m, "Mailbox", "Mailbox")
 313       .def(
 314           "__str__", [](const Mailbox* self) { return std::string("Mailbox(") + self->get_cname() + ")"; },
 315           "Textual representation of the Mailbox`")
 316       .def("by_name", &Mailbox::by_name, py::call_guard<GilScopedRelease>(), "Retrieve a Mailbox from its name")
 317       .def_property_readonly(
 318           "name",
 319           [](const Mailbox* self) {
 320             return std::string(self->get_name().c_str()); // Convert from xbt::string because of MC
 321           },
 322           "The name of that mailbox")
 323       .def(
 324           "put",
 325           [](Mailbox* self, py::object data, int size) {
 326             data.inc_ref();
 327             self->put(data.ptr(), size);
 328           },
 329           py::call_guard<GilScopedRelease>(), "Blocking data transmission")
 330       .def(
 331           "put_async",
 332           [](Mailbox* self, py::object data, int size) {
 333             data.inc_ref();
 334             return self->put_async(data.ptr(), size);
 335           },
 336           py::call_guard<GilScopedRelease>(), "Non-blocking data transmission")
 337       .def(
 338           "get",
 339           [](Mailbox* self) {
 340             py::object data = pybind11::reinterpret_steal<py::object>(self->get<PyObject>());
 341             // data.dec_ref(); // FIXME: why does it break python-actor-create?
 342             return data;
 343           },
 344           py::call_guard<GilScopedRelease>(), "Blocking data reception")
 345       .def(
 346           "get_async",
 347           [](Mailbox* self) -> std::tuple<simgrid::s4u::CommPtr, PyGetAsync> {
 348             PyGetAsync wrap;
 349             auto comm = self->get_async(wrap.get());
 350             return std::make_tuple(std::move(comm), std::move(wrap));
 351           },
 352           py::call_guard<GilScopedRelease>(),
 353           "Non-blocking data reception. Use data.get() to get the python object after the communication has finished")
 354       .def(
 355           "set_receiver", [](Mailbox* self, ActorPtr actor) { self->set_receiver(actor); },
 356           py::call_guard<GilScopedRelease>(), "Sets the actor as permanent receiver");
 357
 358   /* Class PyGetAsync */
 359   py::class_<PyGetAsync>(m, "PyGetAsync", "Wrapper for async get communications")
 360       .def(py::init<>())
 361       .def(
 362           "get", [](PyGetAsync* self) { return py::reinterpret_steal<py::object>(*(self->get())); },
 363           "Get python object after async communication in receiver side");
 364
 365   /* Class Comm */
 366   py::class_<simgrid::s4u::Comm, simgrid::s4u::CommPtr>(m, "Comm", "Communication")
 367       .def("test", &simgrid::s4u::Comm::test, py::call_guard<GilScopedRelease>(),
 368            "Test whether the communication is terminated.")
 369       .def("wait", &simgrid::s4u::Comm::wait, py::call_guard<GilScopedRelease>(),
 370            "Block until the completion of that communication.")
 371       // use py::overload_cast for wait_all/wait_any, until the overload marked XBT_ATTRIB_DEPRECATED_v332 is removed
 372       .def_static("wait_all",
 373                   py::overload_cast<const std::vector<simgrid::s4u::CommPtr>&>(&simgrid::s4u::Comm::wait_all),
 374                   py::call_guard<GilScopedRelease>(), "Block until the completion of all communications in the list.")
 375       .def_static(
 376           "wait_any", py::overload_cast<const std::vector<simgrid::s4u::CommPtr>&>(&simgrid::s4u::Comm::wait_any),
 377           py::call_guard<GilScopedRelease>(),
 378           "Block until the completion of any communication in the list and return the index of the terminated one.");
 379
 380   /* Class Exec */
 381   py::class_<simgrid::s4u::Exec, simgrid::s4u::ExecPtr>(m, "Exec", "Execution")
 382       .def_property_readonly(
 383           "remaining",
 384           [](simgrid::s4u::ExecPtr self) {
 385             GilScopedRelease gil_guard;
 386             return self->get_remaining();
 387           },
 388           "Amount of flops that remain to be computed until completion.")
 389       .def_property_readonly(
 390           "remaining_ratio",
 391           [](simgrid::s4u::ExecPtr self) {
 392             GilScopedRelease gil_guard;
 393             return self->get_remaining_ratio();
 394           },
 395           "Amount of work remaining until completion from 0 (completely done) to 1 (nothing done "
 396           "yet).")
 397       .def_property("host", &simgrid::s4u::Exec::get_host, &simgrid::s4u::Exec::set_host,
 398                     "Host on which this execution runs. Only the first host is returned for parallel executions.")
 399       .def("test", &simgrid::s4u::Exec::test, py::call_guard<GilScopedRelease>(),
 400            "Test whether the execution is terminated.")
 401       .def("cancel", &simgrid::s4u::Exec::cancel, py::call_guard<GilScopedRelease>(), "Cancel that execution.")
 402       .def("start", &simgrid::s4u::Exec::start, py::call_guard<GilScopedRelease>(), "Start that execution.")
 403       .def("wait", &simgrid::s4u::Exec::wait, py::call_guard<GilScopedRelease>(),
 404            "Block until the completion of that execution.");
 405
 406   /* Class Actor */
 407   py::class_<simgrid::s4u::Actor, ActorPtr>(m, "Actor",
 408                                             "An actor is an independent stream of execution in your distributed "
 409                                             "application")
 410       .def(
 411           "create",
 412           [](py::str name, Host* host, py::object fun, py::args args) {
 413             fun.inc_ref();  // FIXME: why is this needed for tests like exec-async, exec-dvfs and exec-remote?
 414             args.inc_ref(); // FIXME: why is this needed for tests like actor-migrate?
 415             return simgrid::s4u::Actor::create(name, host, [fun, args]() {
 416               GilScopedAcquire py_context;
 417               try {
 418                 fun(*args);
 419               } catch (const py::error_already_set& ex) {
 420                 bool ffk = ex.matches(pyForcefulKillEx);
 421                 py_context.reset();
 422                 if (ffk) {
 423                   XBT_VERB("Actor killed");
 424                   simgrid::ForcefulKillException::do_throw(); // Forward that ForcefulKill exception
 425                 }
 426                 throw;
 427               }
 428             });
 429           },
 430           py::call_guard<GilScopedRelease>(), "Create an actor from a function or an object.")
 431       .def_property(
 432           "host", &Actor::get_host,
 433           [](Actor* a, Host* h) {
 434             GilScopedRelease gil_guard;
 435             a->set_host(h);
 436           },
 437           "The host on which this actor is located")
 438       .def_property_readonly("name", &Actor::get_cname, "The name of this actor.")
 439       .def_property_readonly("pid", &Actor::get_pid, "The PID (unique identifier) of this actor.")
 440       .def_property_readonly("ppid", &Actor::get_ppid,
 441                              "The PID (unique identifier) of the actor that created this one.")
 442       .def("by_pid", &Actor::by_pid, "Retrieve an actor by its PID")
 443       .def("daemonize", &Actor::daemonize, py::call_guard<GilScopedRelease>(),
 444            "This actor will be automatically terminated when the last non-daemon actor finishes (more info in the C++ "
 445            "documentation).")
 446       .def("is_daemon", &Actor::is_daemon,
 447            "Returns True if that actor is a daemon and will be terminated automatically when the last non-daemon actor "
 448            "terminates.")
 449       .def("join", py::overload_cast<double>(&Actor::join, py::const_), py::call_guard<GilScopedRelease>(),
 450            "Wait for the actor to finish (more info in the C++ documentation).", py::arg("timeout"))
 451       .def("kill", &Actor::kill, py::call_guard<GilScopedRelease>(), "Kill that actor")
 452       .def("kill_all", &Actor::kill_all, py::call_guard<GilScopedRelease>(), "Kill all actors but the caller.")
 453       .def("self", &Actor::self, "Retrieves the current actor.")
 454       .def("is_suspended", &Actor::is_suspended, "Returns True if that actor is currently suspended.")
 455       .def("suspend", &Actor::suspend, py::call_guard<GilScopedRelease>(),
 456            "Suspend that actor, that is blocked until resume()ed by another actor.")
 457       .def("resume", &Actor::resume, py::call_guard<GilScopedRelease>(),
 458            "Resume that actor, that was previously suspend()ed.");
 459 }