1 /* Copyright (c) 2010-2022. The SimGrid Team. All rights reserved. */
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. */
6 #include <simgrid/kernel/routing/NetPoint.hpp>
7 #include <simgrid/kernel/routing/NetZoneImpl.hpp>
8 #include <simgrid/s4u/Actor.hpp>
9 #include <simgrid/s4u/Comm.hpp>
10 #include <simgrid/s4u/Engine.hpp>
11 #include <simgrid/s4u/Exec.hpp>
12 #include <simgrid/s4u/Host.hpp>
13 #include <simgrid/s4u/VirtualMachine.hpp>
14 #include <xbt/graph.h>
16 #include "src/instr/instr_private.hpp"
17 #include "src/kernel/resource/CpuImpl.hpp"
18 #include "src/kernel/resource/NetworkModel.hpp"
19 #include "src/surf/surf_interface.hpp"
23 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(instr_routing, instr, "Tracing platform hierarchy");
25 std::string instr_pid(simgrid::s4u::Actor const& proc)
27 return std::string(proc.get_name()) + "-" + std::to_string(proc.get_pid());
30 static simgrid::instr::Container* lowestCommonAncestor(const simgrid::instr::Container* a1,
31 const simgrid::instr::Container* a2)
33 // this is only an optimization (since most of a1 and a2 share the same parent)
34 if (a1->get_parent() == a2->get_parent())
35 return a1->get_parent();
37 // create an array with all ancestors of a1
38 std::vector<simgrid::instr::Container*> ancestors_a1;
39 for (auto* p = a1->get_parent(); p != nullptr; p = p->get_parent())
40 ancestors_a1.push_back(p);
42 // create an array with all ancestors of a2
43 std::vector<simgrid::instr::Container*> ancestors_a2;
44 for (auto* p = a2->get_parent(); p != nullptr; p = p->get_parent())
45 ancestors_a2.push_back(p);
47 // find the lowest ancestor
48 simgrid::instr::Container* p = nullptr;
49 int i = static_cast<int>(ancestors_a1.size()) - 1;
50 int j = static_cast<int>(ancestors_a2.size()) - 1;
51 while (i >= 0 && j >= 0) {
52 simgrid::instr::Container* a1p = ancestors_a1.at(i);
53 if (a1p != ancestors_a2.at(j))
62 static void linkContainers(simgrid::instr::Container* src, simgrid::instr::Container* dst,
63 std::set<std::string, std::less<>>* filter)
66 if (src->get_name() == "__loopback__" || dst->get_name() == "__loopback__") {
67 XBT_DEBUG(" linkContainers: ignoring loopback link");
72 simgrid::instr::Container* parent = lowestCommonAncestor(src, dst);
73 xbt_assert(parent, "common parent unknown, this is a tracing problem");
75 // check if we already register this pair (we only need one direction)
76 std::string aux1 = src->get_name() + dst->get_name();
77 std::string aux2 = dst->get_name() + src->get_name();
78 if (filter->find(aux1) != filter->end()) {
79 XBT_DEBUG(" linkContainers: already registered %s <-> %s (1)", src->get_cname(), dst->get_cname());
82 if (filter->find(aux2) != filter->end()) {
83 XBT_DEBUG(" linkContainers: already registered %s <-> %s (2)", dst->get_cname(), src->get_cname());
87 // ok, not found, register it
92 std::string link_typename = parent->get_type()->get_name() + "-" + src->get_type()->get_name() +
93 std::to_string(src->get_type()->get_id()) + "-" + dst->get_type()->get_name() +
94 std::to_string(dst->get_type()->get_id());
95 simgrid::instr::LinkType* link =
96 parent->get_type()->by_name_or_create(link_typename, src->get_type(), dst->get_type());
97 link->set_calling_container(parent);
100 static long long counter = 0;
102 std::string key = std::to_string(counter);
105 link->start_event(src, "topology", key);
106 link->end_event(dst, "topology", key);
108 XBT_DEBUG(" linkContainers %s <-> %s", src->get_cname(), dst->get_cname());
111 static void recursiveGraphExtraction(const simgrid::s4u::NetZone* netzone, const simgrid::instr::Container* container,
112 std::set<std::string, std::less<>>* filter)
114 if (not TRACE_platform_topology()) {
115 XBT_DEBUG("Graph extraction disabled by user.");
118 XBT_DEBUG("Graph extraction for NetZone = %s", netzone->get_cname());
120 // bottom-up recursion
121 for (auto const& nz_son : netzone->get_children()) {
122 const simgrid::instr::Container* child_container = container->get_child_by_name(nz_son->get_name());
123 recursiveGraphExtraction(nz_son, child_container, filter);
126 auto* graph = xbt_graph_new_graph(0, nullptr);
127 std::map<std::string, xbt_node_t, std::less<>> nodes;
128 std::map<std::string, xbt_edge_t, std::less<>> edges;
130 netzone->get_impl()->get_graph(graph, &nodes, &edges);
131 for (auto const& [_, edge] : edges) {
132 linkContainers(simgrid::instr::Container::by_name(static_cast<const char*>(edge->src->data)),
133 simgrid::instr::Container::by_name(static_cast<const char*>(edge->dst->data)), filter);
135 xbt_graph_free_graph(graph, xbt_free_f, xbt_free_f, nullptr);
139 * user categories support
141 static void recursiveNewVariableType(const std::string& new_typename, const std::string& color,
142 simgrid::instr::Type* root)
144 if (root->get_name() == "HOST" || root->get_name() == "VM")
145 root->by_name_or_create(std::string("p") + new_typename, color);
147 if (root->get_name() == "LINK")
148 root->by_name_or_create(std::string("b") + new_typename, color);
150 for (auto const& [_, child] : root->get_children()) {
151 recursiveNewVariableType(new_typename, color, child.get());
155 void instr_new_variable_type(const std::string& new_typename, const std::string& color)
157 recursiveNewVariableType(new_typename, color, simgrid::instr::Container::get_root()->get_type());
160 static void recursiveNewUserVariableType(const std::string& parent_type, const std::string& new_typename,
161 const std::string& color, simgrid::instr::Type* root)
163 if (root->get_name() == parent_type) {
164 root->by_name_or_create(new_typename, color);
166 for (auto const& [_, child] : root->get_children())
167 recursiveNewUserVariableType(parent_type, new_typename, color, child.get());
170 void instr_new_user_variable_type(const std::string& parent_type, const std::string& new_typename,
171 const std::string& color)
173 recursiveNewUserVariableType(parent_type, new_typename, color, simgrid::instr::Container::get_root()->get_type());
176 static void recursiveNewUserStateType(const std::string& parent_type, const std::string& new_typename,
177 simgrid::instr::Type* root)
179 if (root->get_name() == parent_type)
180 root->by_name_or_create<simgrid::instr::StateType>(new_typename);
182 for (auto const& [_, child] : root->get_children())
183 recursiveNewUserStateType(parent_type, new_typename, child.get());
186 void instr_new_user_state_type(const std::string& parent_type, const std::string& new_typename)
188 recursiveNewUserStateType(parent_type, new_typename, simgrid::instr::Container::get_root()->get_type());
191 static void recursiveNewValueForUserStateType(const std::string& type_name, const char* val, const std::string& color,
192 simgrid::instr::Type* root)
194 if (root->get_name() == type_name)
195 static_cast<simgrid::instr::StateType*>(root)->add_entity_value(val, color);
197 for (auto const& [_, child] : root->get_children())
198 recursiveNewValueForUserStateType(type_name, val, color, child.get());
201 void instr_new_value_for_user_state_type(const std::string& type_name, const char* value, const std::string& color)
203 recursiveNewValueForUserStateType(type_name, value, color, simgrid::instr::Container::get_root()->get_type());
209 /** @brief Creates a file with the topology of the platform file used for the simulator.
211 * The graph topology will have the following properties: all hosts, links and routers of the platform file are mapped
212 * to graph nodes; routes are mapped to edges. The platform's zones are not represented in the output.
214 void platform_graph_export_graphviz(const std::string& output_filename)
216 auto* g = xbt_graph_new_graph(0, nullptr);
217 std::map<std::string, xbt_node_t, std::less<>> nodes;
218 std::map<std::string, xbt_edge_t, std::less<>> edges;
219 s4u::Engine::get_instance()->get_netzone_root()->extract_xbt_graph(g, &nodes, &edges);
222 fs.open(output_filename, std::ofstream::out);
223 xbt_assert(not fs.fail(), "Failed to open %s", output_filename.c_str());
226 fs << "digraph test {" << std::endl;
228 fs << "graph test {" << std::endl;
230 fs << " graph [overlap=scale]" << std::endl;
232 fs << " node [shape=box, style=filled]" << std::endl;
233 fs << " node [width=.3, height=.3, style=filled, color=skyblue]" << std::endl << std::endl;
235 for (auto const& [node, _] : nodes)
236 fs << " \"" << node << "\";" << std::endl;
238 for (auto const& [_, edge] : edges) {
239 const char* src_s = static_cast<char*>(edge->src->data);
240 const char* dst_s = static_cast<char*>(edge->dst->data);
242 fs << " \"" << src_s << "\" -> \"" << dst_s << "\";" << std::endl;
244 fs << " \"" << src_s << "\" -- \"" << dst_s << "\";" << std::endl;
246 fs << "}" << std::endl;
249 xbt_graph_free_graph(g, xbt_free_f, xbt_free_f, nullptr);
253 static std::vector<NetZoneContainer*> currentContainer; /* push and pop, used only in creation */
254 static void on_netzone_creation(s4u::NetZone const& netzone)
256 std::string id = netzone.get_name();
257 if (Container::get_root() == nullptr) {
258 auto* root = new NetZoneContainer(id, 0, nullptr);
259 xbt_assert(Container::get_root() == root);
261 if (TRACE_smpi_is_enabled()) {
262 auto* mpi = root->get_type()->by_name_or_create<ContainerType>("MPI");
263 if (not TRACE_smpi_is_grouped())
264 mpi->by_name_or_create<StateType>("MPI_STATE");
265 root->get_type()->by_name_or_create("MPI_LINK", mpi, mpi);
266 root->get_type()->by_name_or_create("MIGRATE_LINK", mpi, mpi);
267 mpi->by_name_or_create<StateType>("MIGRATE_STATE");
270 if (TRACE_needs_platform()) {
271 currentContainer.push_back(root);
276 if (TRACE_needs_platform()) {
277 auto level = static_cast<unsigned>(currentContainer.size());
278 auto* container = new NetZoneContainer(id, level, currentContainer.back());
279 currentContainer.push_back(container);
283 static void on_link_creation(s4u::Link const& link)
285 if (currentContainer.empty()) // No ongoing parsing. Are you creating the loopback?
288 auto* container = new Container(link.get_name(), "LINK", currentContainer.back());
290 if ((TRACE_categorized() || TRACE_uncategorized() || TRACE_platform()) && (not TRACE_disable_link())) {
291 VariableType* bandwidth = container->get_type()->by_name_or_create("bandwidth", "");
292 bandwidth->set_calling_container(container);
293 bandwidth->set_event(0, link.get_bandwidth());
294 VariableType* latency = container->get_type()->by_name_or_create("latency", "");
295 latency->set_calling_container(container);
296 latency->set_event(0, link.get_latency());
299 if (TRACE_uncategorized()) {
300 container->get_type()->by_name_or_create("bandwidth_used", "0.5 0.5 0.5");
304 static void on_host_creation(s4u::Host const& host)
306 if (Container::by_name_or_null(host.get_name())) // This host already exists, do nothing
309 Container* container = new HostContainer(host, currentContainer.back());
310 const Container* root = Container::get_root();
312 if ((TRACE_categorized() || TRACE_uncategorized() || TRACE_platform()) && (not TRACE_disable_speed())) {
313 VariableType* speed = container->get_type()->by_name_or_create("speed", "");
314 speed->set_calling_container(container);
315 speed->set_event(0, host.get_speed());
317 VariableType* cores = container->get_type()->by_name_or_create("core_count", "");
318 cores->set_calling_container(container);
319 cores->set_event(0, host.get_core_count());
322 if (TRACE_uncategorized())
323 container->get_type()->by_name_or_create("speed_used", "0.5 0.5 0.5");
325 if (TRACE_smpi_is_enabled() && TRACE_smpi_is_grouped()) {
326 auto* mpi = container->get_type()->by_name_or_create<ContainerType>("MPI");
327 mpi->by_name_or_create<StateType>("MPI_STATE");
328 root->get_type()->by_name_or_create("MIGRATE_LINK", mpi, mpi);
329 mpi->by_name_or_create<StateType>("MIGRATE_STATE");
333 static void on_action_state_change(kernel::resource::Action const& action,
334 kernel::resource::Action::State /* previous */)
336 auto n = static_cast<unsigned>(action.get_variable()->get_number_of_constraint());
338 for (unsigned i = 0; i < n; i++) {
339 double value = action.get_rate() * action.get_variable()->get_constraint_weight(i);
340 /* Beware of composite actions: ptasks put links and cpus together. Extra pb: we cannot dynamic_cast from void* */
341 kernel::resource::Resource* resource = action.get_variable()->get_constraint(i)->get_id();
342 const kernel::resource::CpuImpl* cpu = dynamic_cast<kernel::resource::CpuImpl*>(resource);
345 resource_set_utilization("HOST", "speed_used", cpu->get_cname(), action.get_category(), value,
346 action.get_last_update(), simgrid_get_clock() - action.get_last_update());
348 const kernel::resource::StandardLinkImpl* link = dynamic_cast<kernel::resource::StandardLinkImpl*>(resource);
351 resource_set_utilization("LINK", "bandwidth_used", link->get_cname(), action.get_category(), value,
352 action.get_last_update(), simgrid_get_clock() - action.get_last_update());
356 static void on_platform_created()
358 currentContainer.clear();
359 std::set<std::string, std::less<>> filter;
360 XBT_DEBUG("Starting graph extraction.");
361 recursiveGraphExtraction(s4u::Engine::get_instance()->get_netzone_root(), Container::get_root(), &filter);
362 XBT_DEBUG("Graph extraction finished.");
366 static void on_actor_creation(s4u::Actor const& actor)
368 const Container* root = Container::get_root();
369 Container* container = Container::by_name(actor.get_host()->get_name());
370 std::string container_name = instr_pid(actor);
372 container->create_child(container_name, "ACTOR");
373 auto* actor_type = container->get_type()->by_name_or_create<ContainerType>("ACTOR");
374 auto* state = actor_type->by_name_or_create<StateType>("ACTOR_STATE");
375 state->add_entity_value("suspend", "1 0 1");
376 state->add_entity_value("sleep", "1 1 0");
377 state->add_entity_value("receive", "1 0 0");
378 state->add_entity_value("send", "0 0 1");
379 state->add_entity_value("execute", "0 1 1");
380 root->get_type()->by_name_or_create("ACTOR_LINK", actor_type, actor_type);
382 actor.on_exit([container_name](bool failed) {
384 // kill means that this actor no longer exists, let's destroy it
385 Container::by_name(container_name)->remove_from_parent();
389 static void on_actor_host_change(s4u::Actor const& actor, s4u::Host const& /*previous_location*/)
391 static long long int counter = 0;
392 Container* container = Container::by_name(instr_pid(actor));
393 LinkType* link = Container::get_root()->get_link("ACTOR_LINK");
396 link->start_event(container, "M", std::to_string(counter));
397 // destroy existing container of this process
398 container->remove_from_parent();
399 // create new container on the new_host location
400 Container::by_name(actor.get_host()->get_name())->create_child(instr_pid(actor), "ACTOR");
402 link->end_event(Container::by_name(instr_pid(actor)), "M", std::to_string(counter));
406 static void on_vm_creation(s4u::Host const& host)
408 const Container* container = new HostContainer(host, currentContainer.back());
409 const Container* root = Container::get_root();
410 auto* vm = container->get_type()->by_name_or_create<ContainerType>("VM");
411 auto* state = vm->by_name_or_create<StateType>("VM_STATE");
412 state->add_entity_value("suspend", "1 0 1");
413 state->add_entity_value("sleep", "1 1 0");
414 state->add_entity_value("receive", "1 0 0");
415 state->add_entity_value("send", "0 0 1");
416 state->add_entity_value("execute", "0 1 1");
417 root->get_type()->by_name_or_create("VM_LINK", vm, vm);
418 root->get_type()->by_name_or_create("VM_ACTOR_LINK", vm, vm);
421 void define_callbacks()
423 // always need the callbacks to zones (we need only the root zone), to create the rootContainer and the rootType
425 if (TRACE_needs_platform()) {
426 s4u::Engine::on_platform_created_cb(on_platform_created);
427 s4u::Host::on_creation_cb(on_host_creation);
428 s4u::Host::on_speed_change_cb([](s4u::Host const& host) {
429 Container::by_name(host.get_name())
430 ->get_variable("speed")
431 ->set_event(simgrid_get_clock(), host.get_core_count() * host.get_available_speed());
433 s4u::Link::on_creation_cb(on_link_creation);
434 s4u::Link::on_bandwidth_change_cb([](s4u::Link const& link) {
435 Container::by_name(link.get_name())
436 ->get_variable("bandwidth")
437 ->set_event(simgrid_get_clock(), sg_bandwidth_factor * link.get_bandwidth());
439 s4u::NetZone::on_seal_cb([](s4u::NetZone const& /*netzone*/) { currentContainer.pop_back(); });
440 kernel::routing::NetPoint::on_creation.connect([](kernel::routing::NetPoint const& netpoint) {
441 if (netpoint.is_router())
442 new RouterContainer(netpoint.get_name(), currentContainer.back());
446 s4u::NetZone::on_creation_cb(on_netzone_creation);
448 kernel::resource::CpuAction::on_state_change.connect(on_action_state_change);
449 s4u::Link::on_communication_state_change_cb(on_action_state_change);
451 if (TRACE_actor_is_enabled()) {
452 s4u::Actor::on_creation_cb(on_actor_creation);
453 s4u::Actor::on_destruction_cb([](s4u::Actor const& actor) {
454 auto container = Container::by_name_or_null(instr_pid(actor));
455 if (container != nullptr)
456 container->remove_from_parent();
458 s4u::Actor::on_suspend_cb([](s4u::Actor const& actor) {
459 Container::by_name(instr_pid(actor))->get_state("ACTOR_STATE")->push_event("suspend");
461 s4u::Actor::on_resume_cb(
462 [](s4u::Actor const& actor) { Container::by_name(instr_pid(actor))->get_state("ACTOR_STATE")->pop_event(); });
463 s4u::Actor::on_sleep_cb([](s4u::Actor const& actor) {
464 Container::by_name(instr_pid(actor))->get_state("ACTOR_STATE")->push_event("sleep");
466 s4u::Actor::on_wake_up_cb(
467 [](s4u::Actor const& actor) { Container::by_name(instr_pid(actor))->get_state("ACTOR_STATE")->pop_event(); });
468 s4u::Exec::on_start_cb([](s4u::Exec const&) {
469 Container::by_name(instr_pid(*s4u::Actor::self()))->get_state("ACTOR_STATE")->push_event("execute");
471 s4u::Activity::on_completion_cb([](const s4u::Activity&) {
472 Container::by_name(instr_pid(*s4u::Actor::self()))->get_state("ACTOR_STATE")->pop_event();
474 s4u::Comm::on_send_cb([](s4u::Comm const&) {
475 Container::by_name(instr_pid(*s4u::Actor::self()))->get_state("ACTOR_STATE")->push_event("send");
477 s4u::Comm::on_recv_cb([](s4u::Comm const&) {
478 Container::by_name(instr_pid(*s4u::Actor::self()))->get_state("ACTOR_STATE")->push_event("receive");
480 s4u::Actor::on_host_change_cb(on_actor_host_change);
483 if (TRACE_smpi_is_enabled() && TRACE_smpi_is_computing()) {
484 s4u::Exec::on_start_cb([](s4u::Exec const& exec) {
485 Container::by_name(std::string("rank-") + std::to_string(s4u::Actor::self()->get_pid()))
486 ->get_state("MPI_STATE")
487 ->push_event("computing", new CpuTIData("compute", exec.get_cost()));
489 s4u::Activity::on_completion_cb([](const s4u::Activity&) {
490 Container::by_name(std::string("rank-") + std::to_string(s4u::Actor::self()->get_pid()))
491 ->get_state("MPI_STATE")
496 if (TRACE_vm_is_enabled()) {
497 s4u::Host::on_creation_cb(on_vm_creation);
498 s4u::VirtualMachine::on_start_cb([](s4u::VirtualMachine const& vm) {
499 Container::by_name(vm.get_name())->get_state("VM_STATE")->push_event("start");
501 s4u::VirtualMachine::on_started_cb(
502 [](s4u::VirtualMachine const& vm) { Container::by_name(vm.get_name())->get_state("VM_STATE")->pop_event(); });
503 s4u::VirtualMachine::on_suspend_cb([](s4u::VirtualMachine const& vm) {
504 Container::by_name(vm.get_name())->get_state("VM_STATE")->push_event("suspend");
506 s4u::VirtualMachine::on_resume_cb(
507 [](s4u::VirtualMachine const& vm) { Container::by_name(vm.get_name())->get_state("VM_STATE")->pop_event(); });
508 s4u::Host::on_destruction_cb(
509 [](s4u::Host const& host) { Container::by_name(host.get_name())->remove_from_parent(); });
513 } // namespace simgrid