1 /* Copyright (c) 2010-2020. 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 "src/instr/instr_private.hpp"
8 #include "simgrid/kernel/routing/NetPoint.hpp"
9 #include "simgrid/kernel/routing/NetZoneImpl.hpp"
10 #include "simgrid/s4u/Actor.hpp"
11 #include "simgrid/s4u/Comm.hpp"
12 #include "simgrid/s4u/Engine.hpp"
13 #include "simgrid/s4u/Exec.hpp"
14 #include "simgrid/s4u/Host.hpp"
15 #include "simgrid/s4u/VirtualMachine.hpp"
16 #include "src/surf/cpu_interface.hpp"
17 #include "src/surf/network_interface.hpp"
18 #include "src/surf/surf_interface.hpp"
19 #include "src/surf/xml/platf_private.hpp"
20 #include "surf/surf.hpp"
21 #include "xbt/graph.h"
23 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(instr_routing, instr, "Tracing platform hierarchy");
25 static std::vector<simgrid::instr::NetZoneContainer*> currentContainer; /* push and pop, used only in creation */
27 std::string instr_pid(simgrid::s4u::Actor const& proc)
29 return std::string(proc.get_name()) + "-" + std::to_string(proc.get_pid());
32 static const char* instr_node_name(const s_xbt_node_t* node)
34 return static_cast<char*>(xbt_graph_node_get_data(node));
37 static container_t lowestCommonAncestor(const simgrid::instr::Container* a1, const simgrid::instr::Container* a2)
39 // this is only an optimization (since most of a1 and a2 share the same parent)
40 if (a1->father_ == a2->father_)
43 // create an array with all ancestors of a1
44 std::vector<container_t> ancestors_a1;
45 container_t p = a1->father_;
47 ancestors_a1.push_back(p);
51 // create an array with all ancestors of a2
52 std::vector<container_t> ancestors_a2;
55 ancestors_a2.push_back(p);
59 // find the lowest ancestor
61 int i = ancestors_a1.size() - 1;
62 int j = ancestors_a2.size() - 1;
63 while (i >= 0 && j >= 0) {
64 container_t a1p = ancestors_a1.at(i);
65 const simgrid::instr::Container* a2p = ancestors_a2.at(j);
77 static void linkContainers(container_t src, container_t dst, std::set<std::string>* filter)
80 if (src->get_name() == "__loopback__" || dst->get_name() == "__loopback__") {
81 XBT_DEBUG(" linkContainers: ignoring loopback link");
86 container_t father = lowestCommonAncestor(src, dst);
88 xbt_die("common father unknown, this is a tracing problem");
91 // check if we already register this pair (we only need one direction)
92 std::string aux1 = src->get_name() + dst->get_name();
93 std::string aux2 = dst->get_name() + src->get_name();
94 if (filter->find(aux1) != filter->end()) {
95 XBT_DEBUG(" linkContainers: already registered %s <-> %s (1)", src->get_cname(), dst->get_cname());
98 if (filter->find(aux2) != filter->end()) {
99 XBT_DEBUG(" linkContainers: already registered %s <-> %s (2)", dst->get_cname(), src->get_cname());
103 // ok, not found, register it
104 filter->insert(aux1);
105 filter->insert(aux2);
108 std::string link_typename = father->type_->get_name() + "-" + src->type_->get_name() +
109 std::to_string(src->type_->get_id()) + "-" + dst->type_->get_name() +
110 std::to_string(dst->type_->get_id());
111 simgrid::instr::LinkType* link = father->type_->by_name_or_create(link_typename, src->type_, dst->type_);
112 link->set_calling_container(father);
115 static long long counter = 0;
117 std::string key = std::to_string(counter);
120 link->start_event(src, "topology", key);
121 link->end_event(dst, "topology", key);
123 XBT_DEBUG(" linkContainers %s <-> %s", src->get_cname(), dst->get_cname());
126 static void recursiveGraphExtraction(const simgrid::s4u::NetZone* netzone, container_t container,
127 std::set<std::string>* filter)
129 if (not TRACE_platform_topology()) {
130 XBT_DEBUG("Graph extraction disabled by user.");
133 XBT_DEBUG("Graph extraction for NetZone = %s", netzone->get_cname());
134 if (not netzone->get_children().empty()) {
135 // bottom-up recursion
136 for (auto const& nz_son : netzone->get_children()) {
137 container_t child_container = container->children_.at(nz_son->get_name());
138 recursiveGraphExtraction(nz_son, child_container, filter);
142 xbt_graph_t graph = xbt_graph_new_graph(0, nullptr);
143 std::map<std::string, xbt_node_t>* nodes = new std::map<std::string, xbt_node_t>();
144 std::map<std::string, xbt_edge_t>* edges = new std::map<std::string, xbt_edge_t>();
146 netzone->get_impl()->get_graph(graph, nodes, edges);
147 for (auto elm : *edges) {
148 const xbt_edge* edge = elm.second;
149 linkContainers(simgrid::instr::Container::by_name(static_cast<const char*>(edge->src->data)),
150 simgrid::instr::Container::by_name(static_cast<const char*>(edge->dst->data)), filter);
154 xbt_graph_free_graph(graph, xbt_free_f, xbt_free_f, nullptr);
160 static void instr_netzone_on_creation(simgrid::s4u::NetZone const& netzone)
162 std::string id = netzone.get_name();
163 if (simgrid::instr::Container::get_root() == nullptr) {
164 simgrid::instr::NetZoneContainer* root = new simgrid::instr::NetZoneContainer(id, 0, nullptr);
165 xbt_assert(simgrid::instr::Container::get_root() == root);
167 if (TRACE_smpi_is_enabled()) {
168 simgrid::instr::ContainerType* mpi = root->type_->by_name_or_create<simgrid::instr::ContainerType>("MPI");
169 if (not TRACE_smpi_is_grouped())
170 mpi->by_name_or_create<simgrid::instr::StateType>("MPI_STATE");
171 root->type_->by_name_or_create("MPI_LINK", mpi, mpi);
172 // TODO See if we can move this to the LoadBalancer plugin
173 root->type_->by_name_or_create("MIGRATE_LINK", mpi, mpi);
174 mpi->by_name_or_create<simgrid::instr::StateType>("MIGRATE_STATE");
177 if (TRACE_needs_platform()) {
178 currentContainer.push_back(root);
183 if (TRACE_needs_platform()) {
184 simgrid::instr::NetZoneContainer* container =
185 new simgrid::instr::NetZoneContainer(id, currentContainer.size(), currentContainer.back());
186 currentContainer.push_back(container);
190 static void instr_link_on_creation(simgrid::s4u::Link const& link)
192 if (currentContainer.empty()) // No ongoing parsing. Are you creating the loopback?
195 container_t container = new simgrid::instr::Container(link.get_name(), "LINK", currentContainer.back());
197 if ((TRACE_categorized() || TRACE_uncategorized() || TRACE_platform()) && (not TRACE_disable_link())) {
198 simgrid::instr::VariableType* bandwidth = container->type_->by_name_or_create("bandwidth", "");
199 bandwidth->set_calling_container(container);
200 bandwidth->set_event(0, link.get_bandwidth());
201 simgrid::instr::VariableType* latency = container->type_->by_name_or_create("latency", "");
202 latency->set_calling_container(container);
203 latency->set_event(0, link.get_latency());
205 if (TRACE_uncategorized()) {
206 container->type_->by_name_or_create("bandwidth_used", "0.5 0.5 0.5");
210 static void instr_host_on_creation(simgrid::s4u::Host const& host)
212 simgrid::instr::Container* container = new simgrid::instr::HostContainer(host, currentContainer.back());
213 const simgrid::instr::Container* root = simgrid::instr::Container::get_root();
215 if ((TRACE_categorized() || TRACE_uncategorized() || TRACE_platform()) && (not TRACE_disable_speed())) {
216 simgrid::instr::VariableType* speed = container->type_->by_name_or_create("speed", "");
217 speed->set_calling_container(container);
218 speed->set_event(0, host.get_speed());
220 simgrid::instr::VariableType* cores = container->type_->by_name_or_create("core_count", "");
221 cores->set_calling_container(container);
222 cores->set_event(0, host.get_core_count());
225 if (TRACE_uncategorized())
226 container->type_->by_name_or_create("speed_used", "0.5 0.5 0.5");
228 if (TRACE_smpi_is_enabled() && TRACE_smpi_is_grouped()) {
229 simgrid::instr::ContainerType* mpi = container->type_->by_name_or_create<simgrid::instr::ContainerType>("MPI");
230 mpi->by_name_or_create<simgrid::instr::StateType>("MPI_STATE");
231 // TODO See if we can move this to the LoadBalancer plugin
232 root->type_->by_name_or_create("MIGRATE_LINK", mpi, mpi);
233 mpi->by_name_or_create<simgrid::instr::StateType>("MIGRATE_STATE");
237 static void instr_host_on_speed_change(simgrid::s4u::Host const& host)
239 simgrid::instr::Container::by_name(host.get_name())
240 ->get_variable("speed")
241 ->set_event(surf_get_clock(), host.get_core_count() * host.get_available_speed());
244 static void instr_action_on_state_change(simgrid::kernel::resource::Action const& action,
245 simgrid::kernel::resource::Action::State /* previous */)
247 int n = action.get_variable()->get_number_of_constraint();
249 for (int i = 0; i < n; i++) {
250 double value = action.get_variable()->get_value() * action.get_variable()->get_constraint_weight(i);
251 /* Beware of composite actions: ptasks put links and cpus together. Extra pb: we cannot dynamic_cast from void* */
252 simgrid::kernel::resource::Resource* resource = action.get_variable()->get_constraint(i)->get_id();
253 const simgrid::kernel::resource::Cpu* cpu = dynamic_cast<simgrid::kernel::resource::Cpu*>(resource);
256 TRACE_surf_resource_set_utilization("HOST", "speed_used", cpu->get_cname(), action.get_category(), value,
257 action.get_last_update(), SIMIX_get_clock() - action.get_last_update());
259 const simgrid::kernel::resource::LinkImpl* link = dynamic_cast<simgrid::kernel::resource::LinkImpl*>(resource);
262 TRACE_surf_resource_set_utilization("LINK", "bandwidth_used", link->get_cname(), action.get_category(), value,
263 action.get_last_update(), SIMIX_get_clock() - action.get_last_update());
267 static void instr_link_on_bandwidth_change(simgrid::s4u::Link const& link)
269 simgrid::instr::Container::by_name(link.get_name())
270 ->get_variable("bandwidth")
271 ->set_event(surf_get_clock(), sg_bandwidth_factor * link.get_bandwidth());
274 static void instr_netpoint_on_creation(simgrid::kernel::routing::NetPoint const& netpoint)
276 if (netpoint.is_router())
277 new simgrid::instr::RouterContainer(netpoint.get_name(), currentContainer.back());
280 static void instr_on_platform_created()
282 currentContainer.clear();
283 std::set<std::string>* filter = new std::set<std::string>();
284 XBT_DEBUG("Starting graph extraction.");
285 recursiveGraphExtraction(simgrid::s4u::Engine::get_instance()->get_netzone_root(),
286 simgrid::instr::Container::get_root(), filter);
287 XBT_DEBUG("Graph extraction finished.");
289 TRACE_paje_dump_buffer(true);
292 static void instr_actor_on_creation(simgrid::s4u::Actor const& actor)
294 const simgrid::instr::Container* root = simgrid::instr::Container::get_root();
295 simgrid::instr::Container* container = simgrid::instr::Container::by_name(actor.get_host()->get_name());
297 container->create_child(instr_pid(actor), "ACTOR");
298 simgrid::instr::ContainerType* actor_type =
299 container->type_->by_name_or_create<simgrid::instr::ContainerType>("ACTOR");
300 simgrid::instr::StateType* state = actor_type->by_name_or_create<simgrid::instr::StateType>("ACTOR_STATE");
301 state->add_entity_value("suspend", "1 0 1");
302 state->add_entity_value("sleep", "1 1 0");
303 state->add_entity_value("receive", "1 0 0");
304 state->add_entity_value("send", "0 0 1");
305 state->add_entity_value("execute", "0 1 1");
306 root->type_->by_name_or_create("ACTOR_LINK", actor_type, actor_type);
307 root->type_->by_name_or_create("ACTOR_TASK_LINK", actor_type, actor_type);
309 std::string container_name = instr_pid(actor);
310 actor.on_exit([container_name](bool failed) {
312 // kill means that this actor no longer exists, let's destroy it
313 simgrid::instr::Container::by_name(container_name)->remove_from_parent();
317 static void instr_actor_on_host_change(simgrid::s4u::Actor const& actor,
318 simgrid::s4u::Host const& /*previous_location*/)
320 static long long int counter = 0;
321 container_t container = simgrid::instr::Container::by_name(instr_pid(actor));
322 simgrid::instr::LinkType* link = simgrid::instr::Container::get_root()->get_link("ACTOR_LINK");
325 link->start_event(container, "M", std::to_string(counter));
326 // destroy existing container of this process
327 container->remove_from_parent();
328 // create new container on the new_host location
329 simgrid::instr::Container::by_name(actor.get_host()->get_name())->create_child(instr_pid(actor), "ACTOR");
331 link->end_event(simgrid::instr::Container::by_name(instr_pid(actor)), "M", std::to_string(counter));
335 static void instr_vm_on_creation(simgrid::s4u::Host const& host)
337 const simgrid::instr::Container* container = new simgrid::instr::HostContainer(host, currentContainer.back());
338 const simgrid::instr::Container* root = simgrid::instr::Container::get_root();
339 simgrid::instr::ContainerType* vm = container->type_->by_name_or_create<simgrid::instr::ContainerType>("VM");
340 simgrid::instr::StateType* state = vm->by_name_or_create<simgrid::instr::StateType>("VM_STATE");
341 state->add_entity_value("suspend", "1 0 1");
342 state->add_entity_value("sleep", "1 1 0");
343 state->add_entity_value("receive", "1 0 0");
344 state->add_entity_value("send", "0 0 1");
345 state->add_entity_value("execute", "0 1 1");
346 root->type_->by_name_or_create("VM_LINK", vm, vm);
347 root->type_->by_name_or_create("VM_ACTOR_LINK", vm, vm);
350 void instr_define_callbacks()
352 // always need the callbacks to zones (we need only the root zone), to create the rootContainer and the rootType
354 if (TRACE_needs_platform()) {
355 simgrid::s4u::Engine::on_platform_created.connect(instr_on_platform_created);
356 simgrid::s4u::Host::on_creation.connect(instr_host_on_creation);
357 simgrid::s4u::Host::on_speed_change.connect(instr_host_on_speed_change);
358 simgrid::s4u::Link::on_creation.connect(instr_link_on_creation);
359 simgrid::s4u::Link::on_bandwidth_change.connect(instr_link_on_bandwidth_change);
360 simgrid::s4u::NetZone::on_seal.connect(
361 [](simgrid::s4u::NetZone const& /*netzone*/) { currentContainer.pop_back(); });
362 simgrid::kernel::routing::NetPoint::on_creation.connect(instr_netpoint_on_creation);
364 simgrid::s4u::NetZone::on_creation.connect(instr_netzone_on_creation);
366 simgrid::kernel::resource::CpuAction::on_state_change.connect(instr_action_on_state_change);
367 simgrid::s4u::Link::on_communication_state_change.connect(instr_action_on_state_change);
369 if (TRACE_actor_is_enabled()) {
370 simgrid::s4u::Actor::on_creation.connect(instr_actor_on_creation);
371 simgrid::s4u::Actor::on_destruction.connect([](simgrid::s4u::Actor const& actor) {
372 auto container = simgrid::instr::Container::by_name_or_null(instr_pid(actor));
373 if (container != nullptr)
374 container->remove_from_parent();
376 simgrid::s4u::Actor::on_suspend.connect([](simgrid::s4u::Actor const& actor) {
377 simgrid::instr::Container::by_name(instr_pid(actor))->get_state("ACTOR_STATE")->push_event("suspend");
379 simgrid::s4u::Actor::on_resume.connect([](simgrid::s4u::Actor const& actor) {
380 simgrid::instr::Container::by_name(instr_pid(actor))->get_state("ACTOR_STATE")->pop_event();
382 simgrid::s4u::Actor::on_sleep.connect([](simgrid::s4u::Actor const& actor) {
383 simgrid::instr::Container::by_name(instr_pid(actor))->get_state("ACTOR_STATE")->push_event("sleep");
385 simgrid::s4u::Actor::on_wake_up.connect([](simgrid::s4u::Actor const& actor) {
386 simgrid::instr::Container::by_name(instr_pid(actor))->get_state("ACTOR_STATE")->pop_event();
388 simgrid::s4u::Exec::on_start.connect([](simgrid::s4u::Actor const& actor, simgrid::s4u::Exec const&) {
389 simgrid::instr::Container::by_name(instr_pid(actor))->get_state("ACTOR_STATE")->push_event("execute");
391 simgrid::s4u::Exec::on_completion.connect([](simgrid::s4u::Actor const& actor, simgrid::s4u::Exec const&) {
392 simgrid::instr::Container::by_name(instr_pid(actor))->get_state("ACTOR_STATE")->pop_event();
394 simgrid::s4u::Comm::on_sender_start.connect([](simgrid::s4u::Actor const& actor) {
395 simgrid::instr::Container::by_name(instr_pid(actor))->get_state("ACTOR_STATE")->push_event("send");
397 simgrid::s4u::Comm::on_receiver_start.connect([](simgrid::s4u::Actor const& actor) {
398 simgrid::instr::Container::by_name(instr_pid(actor))->get_state("ACTOR_STATE")->push_event("receive");
400 simgrid::s4u::Comm::on_completion.connect([](simgrid::s4u::Actor const& actor) {
401 simgrid::instr::Container::by_name(instr_pid(actor))->get_state("ACTOR_STATE")->pop_event();
403 simgrid::s4u::Actor::on_host_change.connect(instr_actor_on_host_change);
406 if (TRACE_vm_is_enabled()) {
407 simgrid::s4u::Host::on_creation.connect(instr_vm_on_creation);
408 simgrid::s4u::VirtualMachine::on_start.connect([](simgrid::s4u::VirtualMachine const& vm) {
409 simgrid::instr::Container::by_name(vm.get_name())->get_state("VM_STATE")->push_event("start");
411 simgrid::s4u::VirtualMachine::on_started.connect([](simgrid::s4u::VirtualMachine const& vm) {
412 simgrid::instr::Container::by_name(vm.get_name())->get_state("VM_STATE")->pop_event();
414 simgrid::s4u::VirtualMachine::on_suspend.connect([](simgrid::s4u::VirtualMachine const& vm) {
415 simgrid::instr::Container::by_name(vm.get_name())->get_state("VM_STATE")->push_event("suspend");
417 simgrid::s4u::VirtualMachine::on_resume.connect([](simgrid::s4u::VirtualMachine const& vm) {
418 simgrid::instr::Container::by_name(vm.get_name())->get_state("VM_STATE")->pop_event();
420 simgrid::s4u::Host::on_destruction.connect([](simgrid::s4u::Host const& host) {
421 simgrid::instr::Container::by_name(host.get_name())->remove_from_parent();
426 * user categories support
428 static void recursiveNewVariableType(const std::string& new_typename, const std::string& color,
429 simgrid::instr::Type* root)
431 if (root->get_name() == "HOST" || root->get_name() == "VM")
432 root->by_name_or_create(std::string("p") + new_typename, color);
434 if (root->get_name() == "LINK")
435 root->by_name_or_create(std::string("b") + new_typename, color);
437 for (auto const& elm : root->children_) {
438 recursiveNewVariableType(new_typename, color, elm.second.get());
442 void instr_new_variable_type(const std::string& new_typename, const std::string& color)
444 recursiveNewVariableType(new_typename, color, simgrid::instr::Container::get_root()->type_);
447 static void recursiveNewUserVariableType(const std::string& father_type, const std::string& new_typename,
448 const std::string& color, simgrid::instr::Type* root)
450 if (root->get_name() == father_type) {
451 root->by_name_or_create(new_typename, color);
453 for (auto const& elm : root->children_)
454 recursiveNewUserVariableType(father_type, new_typename, color, elm.second.get());
457 void instr_new_user_variable_type(const std::string& father_type, const std::string& new_typename,
458 const std::string& color)
460 recursiveNewUserVariableType(father_type, new_typename, color, simgrid::instr::Container::get_root()->type_);
463 static void recursiveNewUserStateType(const std::string& father_type, const std::string& new_typename,
464 simgrid::instr::Type* root)
466 if (root->get_name() == father_type)
467 root->by_name_or_create<simgrid::instr::StateType>(new_typename);
469 for (auto const& elm : root->children_)
470 recursiveNewUserStateType(father_type, new_typename, elm.second.get());
473 void instr_new_user_state_type(const std::string& father_type, const std::string& new_typename)
475 recursiveNewUserStateType(father_type, new_typename, simgrid::instr::Container::get_root()->type_);
478 static void recursiveNewValueForUserStateType(const std::string& type_name, const char* val, const std::string& color,
479 simgrid::instr::Type* root)
481 if (root->get_name() == type_name)
482 static_cast<simgrid::instr::StateType*>(root)->add_entity_value(val, color);
484 for (auto const& elm : root->children_)
485 recursiveNewValueForUserStateType(type_name, val, color, elm.second.get());
488 void instr_new_value_for_user_state_type(const std::string& type_name, const char* value, const std::string& color)
490 recursiveNewValueForUserStateType(type_name, value, color, simgrid::instr::Container::get_root()->type_);
493 #define GRAPHICATOR_SUPPORT_FUNCTIONS
495 static void recursiveXBTGraphExtraction(const s_xbt_graph_t* graph, std::map<std::string, xbt_node_t>* nodes,
496 std::map<std::string, xbt_edge_t>* edges, const_sg_netzone_t netzone,
497 container_t container)
499 if (not netzone->get_children().empty()) {
500 // bottom-up recursion
501 for (auto const& netzone_child : netzone->get_children()) {
502 container_t child_container = container->children_.at(netzone_child->get_name());
503 recursiveXBTGraphExtraction(graph, nodes, edges, netzone_child, child_container);
507 netzone->get_impl()->get_graph(graph, nodes, edges);
510 xbt_graph_t instr_routing_platform_graph()
512 xbt_graph_t ret = xbt_graph_new_graph(0, nullptr);
513 std::map<std::string, xbt_node_t>* nodes = new std::map<std::string, xbt_node_t>();
514 std::map<std::string, xbt_edge_t>* edges = new std::map<std::string, xbt_edge_t>();
515 recursiveXBTGraphExtraction(ret, nodes, edges, simgrid::s4u::Engine::get_instance()->get_netzone_root(),
516 simgrid::instr::Container::get_root());
522 void instr_routing_platform_graph_export_graphviz(const s_xbt_graph_t* g, const char* filename)
524 unsigned int cursor = 0;
525 xbt_node_t node = nullptr;
526 xbt_edge_t edge = nullptr;
528 FILE* file = fopen(filename, "w");
529 xbt_assert(file, "Failed to open %s \n", filename);
532 fprintf(file, "digraph test {\n");
534 fprintf(file, "graph test {\n");
536 fprintf(file, " graph [overlap=scale]\n");
538 fprintf(file, " node [shape=box, style=filled]\n");
539 fprintf(file, " node [width=.3, height=.3, style=filled, color=skyblue]\n\n");
541 xbt_dynar_foreach (g->nodes, cursor, node) {
542 fprintf(file, " \"%s\";\n", instr_node_name(node));
544 xbt_dynar_foreach (g->edges, cursor, edge) {
545 const char* src_s = instr_node_name(edge->src);
546 const char* dst_s = instr_node_name(edge->dst);
548 fprintf(file, " \"%s\" -> \"%s\";\n", src_s, dst_s);
550 fprintf(file, " \"%s\" -- \"%s\";\n", src_s, dst_s);
552 fprintf(file, "}\n");