1 /* Copyright (c) 2010-2019. 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 static const char* instr_node_name(xbt_node_t node)
29 return static_cast<char*>(xbt_graph_node_get_data(node));
32 static container_t lowestCommonAncestor(container_t a1, container_t a2)
34 // this is only an optimization (since most of a1 and a2 share the same parent)
35 if (a1->father_ == a2->father_)
38 // create an array with all ancestors of a1
39 std::vector<container_t> ancestors_a1;
40 container_t p = a1->father_;
42 ancestors_a1.push_back(p);
46 // create an array with all ancestors of a2
47 std::vector<container_t> ancestors_a2;
50 ancestors_a2.push_back(p);
54 // find the lowest ancestor
56 int i = ancestors_a1.size() - 1;
57 int j = ancestors_a2.size() - 1;
58 while (i >= 0 && j >= 0) {
59 container_t a1p = ancestors_a1.at(i);
60 container_t a2p = ancestors_a2.at(j);
72 static void linkContainers(container_t src, container_t dst, std::set<std::string>* filter)
75 if (src->get_name() == "__loopback__" || dst->get_name() == "__loopback__") {
76 XBT_DEBUG(" linkContainers: ignoring loopback link");
81 container_t father = lowestCommonAncestor(src, dst);
83 xbt_die("common father unknown, this is a tracing problem");
86 // check if we already register this pair (we only need one direction)
87 std::string aux1 = src->get_name() + dst->get_name();
88 std::string aux2 = dst->get_name() + src->get_name();
89 if (filter->find(aux1) != filter->end()) {
90 XBT_DEBUG(" linkContainers: already registered %s <-> %s (1)", src->get_cname(), dst->get_cname());
93 if (filter->find(aux2) != filter->end()) {
94 XBT_DEBUG(" linkContainers: already registered %s <-> %s (2)", dst->get_cname(), src->get_cname());
98 // ok, not found, register it
100 filter->insert(aux2);
103 std::string link_typename = father->type_->get_name() + "-" + src->type_->get_name() +
104 std::to_string(src->type_->get_id()) + "-" + dst->type_->get_name() +
105 std::to_string(dst->type_->get_id());
106 simgrid::instr::LinkType* link = father->type_->by_name_or_create(link_typename, src->type_, dst->type_);
107 link->set_calling_container(father);
109 // register EDGE types for triva configuration
110 trivaEdgeTypes.insert(link->get_name());
113 static long long counter = 0;
115 std::string key = std::to_string(counter);
118 link->start_event(src, "topology", key);
119 link->end_event(dst, "topology", key);
121 XBT_DEBUG(" linkContainers %s <-> %s", src->get_cname(), dst->get_cname());
124 static void recursiveGraphExtraction(simgrid::s4u::NetZone* netzone, container_t container,
125 std::set<std::string>* filter)
127 if (not TRACE_platform_topology()) {
128 XBT_DEBUG("Graph extraction disabled by user.");
131 XBT_DEBUG("Graph extraction for NetZone = %s", netzone->get_cname());
132 if (not netzone->get_children().empty()) {
133 // bottom-up recursion
134 for (auto const& nz_son : netzone->get_children()) {
135 container_t child_container = container->children_.at(nz_son->get_name());
136 recursiveGraphExtraction(nz_son, child_container, filter);
140 xbt_graph_t graph = xbt_graph_new_graph(0, nullptr);
141 std::map<std::string, xbt_node_t>* nodes = new std::map<std::string, xbt_node_t>;
142 std::map<std::string, xbt_edge_t>* edges = new std::map<std::string, xbt_edge_t>;
144 netzone->get_impl()->get_graph(graph, nodes, edges);
145 for (auto elm : *edges) {
146 xbt_edge_t edge = elm.second;
147 linkContainers(simgrid::instr::Container::by_name(static_cast<const char*>(edge->src->data)),
148 simgrid::instr::Container::by_name(static_cast<const char*>(edge->dst->data)), filter);
152 xbt_graph_free_graph(graph, xbt_free_f, xbt_free_f, nullptr);
158 static void instr_netzone_on_creation(simgrid::s4u::NetZone const& netzone)
160 std::string id = netzone.get_name();
161 if (simgrid::instr::Container::get_root() == nullptr) {
162 simgrid::instr::NetZoneContainer* root = new simgrid::instr::NetZoneContainer(id, 0, nullptr);
163 xbt_assert(simgrid::instr::Container::get_root() == root);
165 if (TRACE_smpi_is_enabled()) {
166 simgrid::instr::ContainerType* mpi = root->type_->by_name_or_create<simgrid::instr::ContainerType>("MPI");
167 if (not TRACE_smpi_is_grouped())
168 mpi->by_name_or_create<simgrid::instr::StateType>("MPI_STATE");
169 root->type_->by_name_or_create("MPI_LINK", mpi, mpi);
170 // TODO See if we can move this to the LoadBalancer plugin
171 root->type_->by_name_or_create("MIGRATE_LINK", mpi, mpi);
172 mpi->by_name_or_create<simgrid::instr::StateType>("MIGRATE_STATE");
175 if (TRACE_needs_platform()) {
176 currentContainer.push_back(root);
181 if (TRACE_needs_platform()) {
182 simgrid::instr::NetZoneContainer* container =
183 new simgrid::instr::NetZoneContainer(id, currentContainer.size(), currentContainer.back());
184 currentContainer.push_back(container);
188 static void instr_link_on_creation(simgrid::s4u::Link const& link)
190 if (currentContainer.empty()) // No ongoing parsing. Are you creating the loopback?
193 container_t container = new simgrid::instr::Container(link.get_name(), "LINK", currentContainer.back());
195 if ((TRACE_categorized() || TRACE_uncategorized() || TRACE_platform()) && (not TRACE_disable_link())) {
196 simgrid::instr::VariableType* bandwidth = container->type_->by_name_or_create("bandwidth", "");
197 bandwidth->set_calling_container(container);
198 bandwidth->set_event(0, link.get_bandwidth());
199 simgrid::instr::VariableType* latency = container->type_->by_name_or_create("latency", "");
200 latency->set_calling_container(container);
201 latency->set_event(0, link.get_latency());
203 if (TRACE_uncategorized()) {
204 container->type_->by_name_or_create("bandwidth_used", "0.5 0.5 0.5");
208 static void instr_host_on_creation(simgrid::s4u::Host const& host)
210 container_t container = new simgrid::instr::HostContainer(host, currentContainer.back());
211 container_t root = simgrid::instr::Container::get_root();
213 if ((TRACE_categorized() || TRACE_uncategorized() || TRACE_platform()) && (not TRACE_disable_speed())) {
214 simgrid::instr::VariableType* speed = container->type_->by_name_or_create("speed", "");
215 speed->set_calling_container(container);
216 speed->set_event(0, host.get_speed());
218 simgrid::instr::VariableType* cores = container->type_->by_name_or_create("core_count", "");
219 cores->set_calling_container(container);
220 cores->set_event(0, host.get_core_count());
223 if (TRACE_uncategorized())
224 container->type_->by_name_or_create("speed_used", "0.5 0.5 0.5");
226 if (TRACE_smpi_is_enabled() && TRACE_smpi_is_grouped()) {
227 simgrid::instr::ContainerType* mpi = container->type_->by_name_or_create<simgrid::instr::ContainerType>("MPI");
228 mpi->by_name_or_create<simgrid::instr::StateType>("MPI_STATE");
229 // TODO See if we can move this to the LoadBalancer plugin
230 root->type_->by_name_or_create("MIGRATE_LINK", mpi, mpi);
231 mpi->by_name_or_create<simgrid::instr::StateType>("MIGRATE_STATE");
235 static void instr_host_on_speed_change(simgrid::s4u::Host const& host)
237 simgrid::instr::Container::by_name(host.get_name())
238 ->get_variable("speed")
239 ->set_event(surf_get_clock(), host.get_core_count() * host.get_available_speed());
242 static void instr_action_on_state_change(simgrid::kernel::resource::Action const& action,
243 simgrid::kernel::resource::Action::State /* previous */)
245 int n = action.get_variable()->get_number_of_constraint();
247 for (int i = 0; i < n; i++) {
248 double value = action.get_variable()->get_value() * action.get_variable()->get_constraint_weight(i);
249 /* Beware of composite actions: ptasks put links and cpus together. Extra pb: we cannot dynamic_cast from void* */
250 simgrid::kernel::resource::Resource* resource = action.get_variable()->get_constraint(i)->get_id();
251 simgrid::kernel::resource::Cpu* cpu = dynamic_cast<simgrid::kernel::resource::Cpu*>(resource);
254 TRACE_surf_resource_set_utilization("HOST", "speed_used", cpu->get_cname(), action.get_category(), value,
255 action.get_last_update(), SIMIX_get_clock() - action.get_last_update());
257 simgrid::kernel::resource::LinkImpl* link = dynamic_cast<simgrid::kernel::resource::LinkImpl*>(resource);
260 TRACE_surf_resource_set_utilization("LINK", "bandwidth_used", link->get_cname(), action.get_category(), value,
261 action.get_last_update(), SIMIX_get_clock() - action.get_last_update());
265 static void instr_link_on_bandwidth_change(simgrid::s4u::Link const& link)
267 simgrid::instr::Container::by_name(link.get_name())
268 ->get_variable("bandwidth")
269 ->set_event(surf_get_clock(), sg_bandwidth_factor * link.get_bandwidth());
272 static void instr_netpoint_on_creation(simgrid::kernel::routing::NetPoint const& netpoint)
274 if (netpoint.is_router())
275 new simgrid::instr::RouterContainer(netpoint.get_name(), currentContainer.back());
278 static void instr_on_platform_created()
280 currentContainer.clear();
281 std::set<std::string>* filter = new std::set<std::string>;
282 XBT_DEBUG("Starting graph extraction.");
283 recursiveGraphExtraction(simgrid::s4u::Engine::get_instance()->get_netzone_root(),
284 simgrid::instr::Container::get_root(), filter);
285 XBT_DEBUG("Graph extraction finished.");
287 TRACE_paje_dump_buffer(true);
290 static void instr_actor_on_creation(simgrid::s4u::Actor const& actor)
292 container_t root = simgrid::instr::Container::get_root();
293 container_t container = simgrid::instr::Container::by_name(actor.get_host()->get_name());
295 container->create_child(instr_pid(actor), "ACTOR");
296 simgrid::instr::ContainerType* actor_type =
297 container->type_->by_name_or_create<simgrid::instr::ContainerType>("ACTOR");
298 simgrid::instr::StateType* state = actor_type->by_name_or_create<simgrid::instr::StateType>("ACTOR_STATE");
299 state->add_entity_value("suspend", "1 0 1");
300 state->add_entity_value("sleep", "1 1 0");
301 state->add_entity_value("receive", "1 0 0");
302 state->add_entity_value("send", "0 0 1");
303 state->add_entity_value("execute", "0 1 1");
304 root->type_->by_name_or_create("ACTOR_LINK", actor_type, actor_type);
305 root->type_->by_name_or_create("ACTOR_TASK_LINK", actor_type, actor_type);
307 std::string container_name = instr_pid(actor);
308 actor.on_exit([container_name](bool failed) {
310 // kill means that this actor no longer exists, let's destroy it
311 simgrid::instr::Container::by_name(container_name)->remove_from_parent();
315 static void instr_actor_on_host_change(simgrid::s4u::Actor const& actor,
316 simgrid::s4u::Host const& /*previous_location*/)
318 static long long int counter = 0;
319 container_t container = simgrid::instr::Container::by_name(instr_pid(actor));
320 simgrid::instr::LinkType* link = simgrid::instr::Container::get_root()->get_link("ACTOR_LINK");
323 link->start_event(container, "M", std::to_string(counter));
324 // destroy existing container of this process
325 container->remove_from_parent();
326 // create new container on the new_host location
327 simgrid::instr::Container::by_name(actor.get_host()->get_name())->create_child(instr_pid(actor), "ACTOR");
329 link->end_event(simgrid::instr::Container::by_name(instr_pid(actor)), "M", std::to_string(counter));
333 static void instr_vm_on_creation(simgrid::s4u::Host const& host)
335 container_t container = new simgrid::instr::HostContainer(host, currentContainer.back());
336 container_t root = simgrid::instr::Container::get_root();
337 simgrid::instr::ContainerType* vm = container->type_->by_name_or_create<simgrid::instr::ContainerType>("VM");
338 simgrid::instr::StateType* state = vm->by_name_or_create<simgrid::instr::StateType>("VM_STATE");
339 state->add_entity_value("suspend", "1 0 1");
340 state->add_entity_value("sleep", "1 1 0");
341 state->add_entity_value("receive", "1 0 0");
342 state->add_entity_value("send", "0 0 1");
343 state->add_entity_value("execute", "0 1 1");
344 root->type_->by_name_or_create("VM_LINK", vm, vm);
345 root->type_->by_name_or_create("VM_ACTOR_LINK", vm, vm);
348 void instr_define_callbacks()
350 // always need the callbacks to zones (we need only the root zone), to create the rootContainer and the rootType
352 if (TRACE_needs_platform()) {
353 simgrid::s4u::Engine::on_platform_created.connect(instr_on_platform_created);
354 simgrid::s4u::Host::on_creation.connect(instr_host_on_creation);
355 simgrid::s4u::Host::on_speed_change.connect(instr_host_on_speed_change);
356 simgrid::s4u::Link::on_creation.connect(instr_link_on_creation);
357 simgrid::s4u::Link::on_bandwidth_change.connect(instr_link_on_bandwidth_change);
358 simgrid::s4u::NetZone::on_seal.connect(
359 [](simgrid::s4u::NetZone const& /*netzone*/) { currentContainer.pop_back(); });
360 simgrid::kernel::routing::NetPoint::on_creation.connect(instr_netpoint_on_creation);
362 simgrid::s4u::NetZone::on_creation.connect(instr_netzone_on_creation);
364 simgrid::kernel::resource::CpuAction::on_state_change.connect(instr_action_on_state_change);
365 simgrid::s4u::Link::on_communication_state_change.connect(instr_action_on_state_change);
367 if (TRACE_actor_is_enabled()) {
368 simgrid::s4u::Actor::on_creation.connect(instr_actor_on_creation);
369 simgrid::s4u::Actor::on_destruction.connect([](simgrid::s4u::Actor const& actor) {
370 auto container = simgrid::instr::Container::by_name_or_null(instr_pid(actor));
371 if (container != nullptr)
372 container->remove_from_parent();
374 simgrid::s4u::Actor::on_suspend.connect([](simgrid::s4u::Actor const& actor) {
375 simgrid::instr::Container::by_name(instr_pid(actor))->get_state("ACTOR_STATE")->push_event("suspend");
377 simgrid::s4u::Actor::on_resume.connect([](simgrid::s4u::Actor const& actor) {
378 simgrid::instr::Container::by_name(instr_pid(actor))->get_state("ACTOR_STATE")->pop_event();
380 simgrid::s4u::Actor::on_sleep.connect([](simgrid::s4u::Actor const& actor) {
381 simgrid::instr::Container::by_name(instr_pid(actor))->get_state("ACTOR_STATE")->push_event("sleep");
383 simgrid::s4u::Actor::on_wake_up.connect([](simgrid::s4u::Actor const& actor) {
384 simgrid::instr::Container::by_name(instr_pid(actor))->get_state("ACTOR_STATE")->pop_event();
386 simgrid::s4u::Exec::on_start.connect([](simgrid::s4u::Actor const& actor, simgrid::s4u::Exec const&) {
387 simgrid::instr::Container::by_name(instr_pid(actor))->get_state("ACTOR_STATE")->push_event("execute");
389 simgrid::s4u::Exec::on_completion.connect([](simgrid::s4u::Actor const& actor, simgrid::s4u::Exec const&) {
390 simgrid::instr::Container::by_name(instr_pid(actor))->get_state("ACTOR_STATE")->pop_event();
392 simgrid::s4u::Comm::on_sender_start.connect([](simgrid::s4u::Actor const& actor) {
393 simgrid::instr::Container::by_name(instr_pid(actor))->get_state("ACTOR_STATE")->push_event("send");
395 simgrid::s4u::Comm::on_receiver_start.connect([](simgrid::s4u::Actor const& actor) {
396 simgrid::instr::Container::by_name(instr_pid(actor))->get_state("ACTOR_STATE")->push_event("receive");
398 simgrid::s4u::Comm::on_completion.connect([](simgrid::s4u::Actor const& actor) {
399 simgrid::instr::Container::by_name(instr_pid(actor))->get_state("ACTOR_STATE")->pop_event();
401 simgrid::s4u::Actor::on_host_change.connect(instr_actor_on_host_change);
404 if (TRACE_vm_is_enabled()) {
405 simgrid::s4u::Host::on_creation.connect(instr_vm_on_creation);
406 simgrid::s4u::VirtualMachine::on_start.connect([](simgrid::s4u::VirtualMachine const& vm) {
407 simgrid::instr::Container::by_name(vm.get_name())->get_state("VM_STATE")->push_event("start");
409 simgrid::s4u::VirtualMachine::on_started.connect([](simgrid::s4u::VirtualMachine const& vm) {
410 simgrid::instr::Container::by_name(vm.get_name())->get_state("VM_STATE")->pop_event();
412 simgrid::s4u::VirtualMachine::on_suspend.connect([](simgrid::s4u::VirtualMachine const& vm) {
413 simgrid::instr::Container::by_name(vm.get_name())->get_state("VM_STATE")->push_event("suspend");
415 simgrid::s4u::VirtualMachine::on_resume.connect([](simgrid::s4u::VirtualMachine const& vm) {
416 simgrid::instr::Container::by_name(vm.get_name())->get_state("VM_STATE")->pop_event();
418 simgrid::s4u::Host::on_destruction.connect([](simgrid::s4u::Host const& host) {
419 simgrid::instr::Container::by_name(host.get_name())->remove_from_parent();
424 * user categories support
426 static void recursiveNewVariableType(const std::string& new_typename, const std::string& color,
427 simgrid::instr::Type* root)
429 if (root->get_name() == "HOST" || root->get_name() == "VM")
430 root->by_name_or_create(std::string("p") + new_typename, color);
432 if (root->get_name() == "LINK")
433 root->by_name_or_create(std::string("b") + new_typename, color);
435 for (auto const& elm : root->children_) {
436 recursiveNewVariableType(new_typename, color, elm.second.get());
440 void instr_new_variable_type(const std::string& new_typename, const std::string& color)
442 recursiveNewVariableType(new_typename, color, simgrid::instr::Container::get_root()->type_);
445 static void recursiveNewUserVariableType(const std::string& father_type, const std::string& new_typename,
446 const std::string& color, simgrid::instr::Type* root)
448 if (root->get_name() == father_type) {
449 root->by_name_or_create(new_typename, color);
451 for (auto const& elm : root->children_)
452 recursiveNewUserVariableType(father_type, new_typename, color, elm.second.get());
455 void instr_new_user_variable_type(const std::string& father_type, const std::string& new_typename,
456 const std::string& color)
458 recursiveNewUserVariableType(father_type, new_typename, color, simgrid::instr::Container::get_root()->type_);
461 static void recursiveNewUserStateType(const std::string& father_type, const std::string& new_typename,
462 simgrid::instr::Type* root)
464 if (root->get_name() == father_type)
465 root->by_name_or_create<simgrid::instr::StateType>(new_typename);
467 for (auto const& elm : root->children_)
468 recursiveNewUserStateType(father_type, new_typename, elm.second.get());
471 void instr_new_user_state_type(const std::string& father_type, const std::string& new_typename)
473 recursiveNewUserStateType(father_type, new_typename, simgrid::instr::Container::get_root()->type_);
476 static void recursiveNewValueForUserStateType(const std::string& type_name, const char* val, const std::string& color,
477 simgrid::instr::Type* root)
479 if (root->get_name() == type_name)
480 static_cast<simgrid::instr::StateType*>(root)->add_entity_value(val, color);
482 for (auto const& elm : root->children_)
483 recursiveNewValueForUserStateType(type_name, val, color, elm.second.get());
486 void instr_new_value_for_user_state_type(const std::string& type_name, const char* value, const std::string& color)
488 recursiveNewValueForUserStateType(type_name, value, color, simgrid::instr::Container::get_root()->type_);
491 #define GRAPHICATOR_SUPPORT_FUNCTIONS
493 static void recursiveXBTGraphExtraction(xbt_graph_t graph, std::map<std::string, xbt_node_t>* nodes,
494 std::map<std::string, xbt_edge_t>* edges, sg_netzone_t netzone,
495 container_t container)
497 if (not netzone->get_children().empty()) {
498 // bottom-up recursion
499 for (auto const& netzone_child : netzone->get_children()) {
500 container_t child_container = container->children_.at(netzone_child->get_name());
501 recursiveXBTGraphExtraction(graph, nodes, edges, netzone_child, child_container);
505 netzone->get_impl()->get_graph(graph, nodes, edges);
508 xbt_graph_t instr_routing_platform_graph()
510 xbt_graph_t ret = xbt_graph_new_graph(0, nullptr);
511 std::map<std::string, xbt_node_t>* nodes = new std::map<std::string, xbt_node_t>;
512 std::map<std::string, xbt_edge_t>* edges = new std::map<std::string, xbt_edge_t>;
513 recursiveXBTGraphExtraction(ret, nodes, edges, simgrid::s4u::Engine::get_instance()->get_netzone_root(),
514 simgrid::instr::Container::get_root());
520 void instr_routing_platform_graph_export_graphviz(xbt_graph_t g, const char* filename)
522 unsigned int cursor = 0;
523 xbt_node_t node = nullptr;
524 xbt_edge_t edge = nullptr;
526 FILE* file = fopen(filename, "w");
527 xbt_assert(file, "Failed to open %s \n", filename);
530 fprintf(file, "digraph test {\n");
532 fprintf(file, "graph test {\n");
534 fprintf(file, " graph [overlap=scale]\n");
536 fprintf(file, " node [shape=box, style=filled]\n");
537 fprintf(file, " node [width=.3, height=.3, style=filled, color=skyblue]\n\n");
539 xbt_dynar_foreach (g->nodes, cursor, node) {
540 fprintf(file, " \"%s\";\n", instr_node_name(node));
542 xbt_dynar_foreach (g->edges, cursor, edge) {
543 const char* src_s = instr_node_name(edge->src);
544 const char* dst_s = instr_node_name(edge->dst);
546 fprintf(file, " \"%s\" -> \"%s\";\n", src_s, dst_s);
548 fprintf(file, " \"%s\" -- \"%s\";\n", src_s, dst_s);
550 fprintf(file, "}\n");