Logo AND Algorithmique Numérique Distribuée

Public GIT Repository
Merge branch 'dev/s4u_tuto_fixes' into 'master'
[simgrid.git] / src / instr / instr_platform.cpp
1 /* Copyright (c) 2010-2022. 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 #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>
15
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"
20
21 #include <fstream>
22
23 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(instr_routing, instr, "Tracing platform hierarchy");
24
25 std::string instr_pid(simgrid::s4u::Actor const& proc)
26 {
27   return std::string(proc.get_name()) + "-" + std::to_string(proc.get_pid());
28 }
29
30 static simgrid::instr::Container* lowestCommonAncestor(const simgrid::instr::Container* a1,
31                                                        const simgrid::instr::Container* a2)
32 {
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();
36
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);
41
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);
46
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     const simgrid::instr::Container* a2p = ancestors_a2.at(j);
54     if (a1p == a2p) {
55       p = a1p;
56     } else {
57       break;
58     }
59     i--;
60     j--;
61   }
62   return p;
63 }
64
65 static void linkContainers(simgrid::instr::Container* src, simgrid::instr::Container* dst,
66                            std::set<std::string, std::less<>>* filter)
67 {
68   // ignore loopback
69   if (src->get_name() == "__loopback__" || dst->get_name() == "__loopback__") {
70     XBT_DEBUG("  linkContainers: ignoring loopback link");
71     return;
72   }
73
74   // find common parent
75   simgrid::instr::Container* parent = lowestCommonAncestor(src, dst);
76   xbt_assert(parent, "common parent unknown, this is a tracing problem");
77
78   // check if we already register this pair (we only need one direction)
79   std::string aux1 = src->get_name() + dst->get_name();
80   std::string aux2 = dst->get_name() + src->get_name();
81   if (filter->find(aux1) != filter->end()) {
82     XBT_DEBUG("  linkContainers: already registered %s <-> %s (1)", src->get_cname(), dst->get_cname());
83     return;
84   }
85   if (filter->find(aux2) != filter->end()) {
86     XBT_DEBUG("  linkContainers: already registered %s <-> %s (2)", dst->get_cname(), src->get_cname());
87     return;
88   }
89
90   // ok, not found, register it
91   filter->insert(aux1);
92   filter->insert(aux2);
93
94   // declare type
95   std::string link_typename = parent->get_type()->get_name() + "-" + src->get_type()->get_name() +
96                               std::to_string(src->get_type()->get_id()) + "-" + dst->get_type()->get_name() +
97                               std::to_string(dst->get_type()->get_id());
98   simgrid::instr::LinkType* link =
99       parent->get_type()->by_name_or_create(link_typename, src->get_type(), dst->get_type());
100   link->set_calling_container(parent);
101
102   // create the link
103   static long long counter = 0;
104
105   std::string key = std::to_string(counter);
106   counter++;
107
108   link->start_event(src, "topology", key);
109   link->end_event(dst, "topology", key);
110
111   XBT_DEBUG("  linkContainers %s <-> %s", src->get_cname(), dst->get_cname());
112 }
113
114 static void recursiveGraphExtraction(const simgrid::s4u::NetZone* netzone, const simgrid::instr::Container* container,
115                                      std::set<std::string, std::less<>>* filter)
116 {
117   if (not TRACE_platform_topology()) {
118     XBT_DEBUG("Graph extraction disabled by user.");
119     return;
120   }
121   XBT_DEBUG("Graph extraction for NetZone = %s", netzone->get_cname());
122
123   // bottom-up recursion
124   for (auto const& nz_son : netzone->get_children()) {
125     const simgrid::instr::Container* child_container = container->get_child_by_name(nz_son->get_name());
126     recursiveGraphExtraction(nz_son, child_container, filter);
127   }
128
129   auto* graph = xbt_graph_new_graph(0, nullptr);
130   std::map<std::string, xbt_node_t, std::less<>> nodes;
131   std::map<std::string, xbt_edge_t, std::less<>> edges;
132
133   netzone->get_impl()->get_graph(graph, &nodes, &edges);
134   for (auto const& elm : edges) {
135     const xbt_edge* edge = elm.second;
136     linkContainers(simgrid::instr::Container::by_name(static_cast<const char*>(edge->src->data)),
137                    simgrid::instr::Container::by_name(static_cast<const char*>(edge->dst->data)), filter);
138   }
139   xbt_graph_free_graph(graph, xbt_free_f, xbt_free_f, nullptr);
140 }
141
142 /*
143  * user categories support
144  */
145 static void recursiveNewVariableType(const std::string& new_typename, const std::string& color,
146                                      simgrid::instr::Type* root)
147 {
148   if (root->get_name() == "HOST" || root->get_name() == "VM")
149     root->by_name_or_create(std::string("p") + new_typename, color);
150
151   if (root->get_name() == "LINK")
152     root->by_name_or_create(std::string("b") + new_typename, color);
153
154   for (auto const& elm : root->get_children()) {
155     recursiveNewVariableType(new_typename, color, elm.second.get());
156   }
157 }
158
159 void instr_new_variable_type(const std::string& new_typename, const std::string& color)
160 {
161   recursiveNewVariableType(new_typename, color, simgrid::instr::Container::get_root()->get_type());
162 }
163
164 static void recursiveNewUserVariableType(const std::string& parent_type, const std::string& new_typename,
165                                          const std::string& color, simgrid::instr::Type* root)
166 {
167   if (root->get_name() == parent_type) {
168     root->by_name_or_create(new_typename, color);
169   }
170   for (auto const& elm : root->get_children())
171     recursiveNewUserVariableType(parent_type, new_typename, color, elm.second.get());
172 }
173
174 void instr_new_user_variable_type(const std::string& parent_type, const std::string& new_typename,
175                                   const std::string& color)
176 {
177   recursiveNewUserVariableType(parent_type, new_typename, color, simgrid::instr::Container::get_root()->get_type());
178 }
179
180 static void recursiveNewUserStateType(const std::string& parent_type, const std::string& new_typename,
181                                       simgrid::instr::Type* root)
182 {
183   if (root->get_name() == parent_type)
184     root->by_name_or_create<simgrid::instr::StateType>(new_typename);
185
186   for (auto const& elm : root->get_children())
187     recursiveNewUserStateType(parent_type, new_typename, elm.second.get());
188 }
189
190 void instr_new_user_state_type(const std::string& parent_type, const std::string& new_typename)
191 {
192   recursiveNewUserStateType(parent_type, new_typename, simgrid::instr::Container::get_root()->get_type());
193 }
194
195 static void recursiveNewValueForUserStateType(const std::string& type_name, const char* val, const std::string& color,
196                                               simgrid::instr::Type* root)
197 {
198   if (root->get_name() == type_name)
199     static_cast<simgrid::instr::StateType*>(root)->add_entity_value(val, color);
200
201   for (auto const& elm : root->get_children())
202     recursiveNewValueForUserStateType(type_name, val, color, elm.second.get());
203 }
204
205 void instr_new_value_for_user_state_type(const std::string& type_name, const char* value, const std::string& color)
206 {
207   recursiveNewValueForUserStateType(type_name, value, color, simgrid::instr::Container::get_root()->get_type());
208 }
209
210 namespace simgrid {
211 namespace instr {
212
213 /** @brief Creates a file with the topology of the platform file used for the simulator.
214  *
215  *  The graph topology will have the following properties: all hosts, links and routers of the platform file are mapped
216  *  to graph nodes; routes are mapped to edges. The platform's zones are not represented in the output.
217  */
218 void platform_graph_export_graphviz(const std::string& output_filename)
219 {
220   auto* g     = xbt_graph_new_graph(0, nullptr);
221   std::map<std::string, xbt_node_t, std::less<>> nodes;
222   std::map<std::string, xbt_edge_t, std::less<>> edges;
223   s4u::Engine::get_instance()->get_netzone_root()->extract_xbt_graph(g, &nodes, &edges);
224
225   std::ofstream fs;
226   fs.open(output_filename, std::ofstream::out);
227   xbt_assert(not fs.fail(), "Failed to open %s", output_filename.c_str());
228
229   if (g->directed)
230     fs << "digraph test {" << std::endl;
231   else
232     fs << "graph test {" << std::endl;
233
234   fs << "  graph [overlap=scale]" << std::endl;
235
236   fs << "  node [shape=box, style=filled]" << std::endl;
237   fs << "  node [width=.3, height=.3, style=filled, color=skyblue]" << std::endl << std::endl;
238
239   for (auto const& elm : nodes)
240     fs << "  \"" << elm.first << "\";" << std::endl;
241
242   for (auto const& elm : edges) {
243     const char* src_s = static_cast<char*>(elm.second->src->data);
244     const char* dst_s = static_cast<char*>(elm.second->dst->data);
245     if (g->directed)
246       fs << "  \"" << src_s << "\" -> \"" << dst_s << "\";" << std::endl;
247     else
248       fs << "  \"" << src_s << "\" -- \"" << dst_s << "\";" << std::endl;
249   }
250   fs << "}" << std::endl;
251   fs.close();
252
253   xbt_graph_free_graph(g, xbt_free_f, xbt_free_f, nullptr);
254 }
255
256 /* Callbacks */
257 static std::vector<NetZoneContainer*> currentContainer; /* push and pop, used only in creation */
258 static void on_netzone_creation(s4u::NetZone const& netzone)
259 {
260   std::string id = netzone.get_name();
261   if (Container::get_root() == nullptr) {
262     auto* root = new NetZoneContainer(id, 0, nullptr);
263     xbt_assert(Container::get_root() == root);
264
265     if (TRACE_smpi_is_enabled()) {
266       auto* mpi = root->get_type()->by_name_or_create<ContainerType>("MPI");
267       if (not TRACE_smpi_is_grouped())
268         mpi->by_name_or_create<StateType>("MPI_STATE");
269       root->get_type()->by_name_or_create("MPI_LINK", mpi, mpi);
270       root->get_type()->by_name_or_create("MIGRATE_LINK", mpi, mpi);
271       mpi->by_name_or_create<StateType>("MIGRATE_STATE");
272     }
273
274     if (TRACE_needs_platform()) {
275       currentContainer.push_back(root);
276     }
277     return;
278   }
279
280   if (TRACE_needs_platform()) {
281     auto level      = static_cast<unsigned>(currentContainer.size());
282     auto* container = new NetZoneContainer(id, level, currentContainer.back());
283     currentContainer.push_back(container);
284   }
285 }
286
287 static void on_link_creation(s4u::Link const& link)
288 {
289   if (currentContainer.empty()) // No ongoing parsing. Are you creating the loopback?
290     return;
291
292   auto* container = new Container(link.get_name(), "LINK", currentContainer.back());
293
294   if ((TRACE_categorized() || TRACE_uncategorized() || TRACE_platform()) && (not TRACE_disable_link())) {
295     VariableType* bandwidth = container->get_type()->by_name_or_create("bandwidth", "");
296     bandwidth->set_calling_container(container);
297     bandwidth->set_event(0, link.get_bandwidth());
298     VariableType* latency = container->get_type()->by_name_or_create("latency", "");
299     latency->set_calling_container(container);
300     latency->set_event(0, link.get_latency());
301   }
302
303   if (TRACE_uncategorized()) {
304     container->get_type()->by_name_or_create("bandwidth_used", "0.5 0.5 0.5");
305   }
306 }
307
308 static void on_host_creation(s4u::Host const& host)
309 {
310   if (Container::by_name_or_null(host.get_name())) // This host already exists, do nothing
311     return;
312
313   Container* container  = new HostContainer(host, currentContainer.back());
314   const Container* root = Container::get_root();
315
316   if ((TRACE_categorized() || TRACE_uncategorized() || TRACE_platform()) && (not TRACE_disable_speed())) {
317     VariableType* speed = container->get_type()->by_name_or_create("speed", "");
318     speed->set_calling_container(container);
319     speed->set_event(0, host.get_speed());
320
321     VariableType* cores = container->get_type()->by_name_or_create("core_count", "");
322     cores->set_calling_container(container);
323     cores->set_event(0, host.get_core_count());
324   }
325
326   if (TRACE_uncategorized())
327     container->get_type()->by_name_or_create("speed_used", "0.5 0.5 0.5");
328
329   if (TRACE_smpi_is_enabled() && TRACE_smpi_is_grouped()) {
330     auto* mpi = container->get_type()->by_name_or_create<ContainerType>("MPI");
331     mpi->by_name_or_create<StateType>("MPI_STATE");
332     root->get_type()->by_name_or_create("MIGRATE_LINK", mpi, mpi);
333     mpi->by_name_or_create<StateType>("MIGRATE_STATE");
334   }
335 }
336
337 static void on_action_state_change(kernel::resource::Action const& action,
338                                    kernel::resource::Action::State /* previous */)
339 {
340   auto n = static_cast<unsigned>(action.get_variable()->get_number_of_constraint());
341
342   for (unsigned i = 0; i < n; i++) {
343     double value = action.get_rate() * action.get_variable()->get_constraint_weight(i);
344     /* Beware of composite actions: ptasks put links and cpus together. Extra pb: we cannot dynamic_cast from void* */
345     kernel::resource::Resource* resource = action.get_variable()->get_constraint(i)->get_id();
346     const kernel::resource::CpuImpl* cpu = dynamic_cast<kernel::resource::CpuImpl*>(resource);
347
348     if (cpu != nullptr)
349       resource_set_utilization("HOST", "speed_used", cpu->get_cname(), action.get_category(), value,
350                                action.get_last_update(), simgrid_get_clock() - action.get_last_update());
351
352     const kernel::resource::StandardLinkImpl* link = dynamic_cast<kernel::resource::StandardLinkImpl*>(resource);
353
354     if (link != nullptr)
355       resource_set_utilization("LINK", "bandwidth_used", link->get_cname(), action.get_category(), value,
356                                action.get_last_update(), simgrid_get_clock() - action.get_last_update());
357   }
358 }
359
360 static void on_platform_created()
361 {
362   currentContainer.clear();
363   std::set<std::string, std::less<>> filter;
364   XBT_DEBUG("Starting graph extraction.");
365   recursiveGraphExtraction(s4u::Engine::get_instance()->get_netzone_root(), Container::get_root(), &filter);
366   XBT_DEBUG("Graph extraction finished.");
367   dump_buffer(true);
368 }
369
370 static void on_actor_creation(s4u::Actor const& actor)
371 {
372   const Container* root      = Container::get_root();
373   Container* container       = Container::by_name(actor.get_host()->get_name());
374   std::string container_name = instr_pid(actor);
375
376   container->create_child(container_name, "ACTOR");
377   auto* actor_type = container->get_type()->by_name_or_create<ContainerType>("ACTOR");
378   auto* state      = actor_type->by_name_or_create<StateType>("ACTOR_STATE");
379   state->add_entity_value("suspend", "1 0 1");
380   state->add_entity_value("sleep", "1 1 0");
381   state->add_entity_value("receive", "1 0 0");
382   state->add_entity_value("send", "0 0 1");
383   state->add_entity_value("execute", "0 1 1");
384   root->get_type()->by_name_or_create("ACTOR_LINK", actor_type, actor_type);
385
386   actor.on_exit([container_name](bool failed) {
387     if (failed)
388       // kill means that this actor no longer exists, let's destroy it
389       Container::by_name(container_name)->remove_from_parent();
390   });
391 }
392
393 static void on_actor_host_change(s4u::Actor const& actor, s4u::Host const& /*previous_location*/)
394 {
395   static long long int counter = 0;
396   Container* container         = Container::by_name(instr_pid(actor));
397   LinkType* link               = Container::get_root()->get_link("ACTOR_LINK");
398
399   // start link
400   link->start_event(container, "M", std::to_string(counter));
401   // destroy existing container of this process
402   container->remove_from_parent();
403   // create new container on the new_host location
404   Container::by_name(actor.get_host()->get_name())->create_child(instr_pid(actor), "ACTOR");
405   // end link
406   link->end_event(Container::by_name(instr_pid(actor)), "M", std::to_string(counter));
407   counter++;
408 }
409
410 static void on_vm_creation(s4u::Host const& host)
411 {
412   const Container* container = new HostContainer(host, currentContainer.back());
413   const Container* root      = Container::get_root();
414   auto* vm                   = container->get_type()->by_name_or_create<ContainerType>("VM");
415   auto* state                = vm->by_name_or_create<StateType>("VM_STATE");
416   state->add_entity_value("suspend", "1 0 1");
417   state->add_entity_value("sleep", "1 1 0");
418   state->add_entity_value("receive", "1 0 0");
419   state->add_entity_value("send", "0 0 1");
420   state->add_entity_value("execute", "0 1 1");
421   root->get_type()->by_name_or_create("VM_LINK", vm, vm);
422   root->get_type()->by_name_or_create("VM_ACTOR_LINK", vm, vm);
423 }
424
425 void define_callbacks()
426 {
427   // always need the callbacks to zones (we need only the root zone), to create the rootContainer and the rootType
428   // properly
429   if (TRACE_needs_platform()) {
430     s4u::Engine::on_platform_created_cb(on_platform_created);
431     s4u::Host::on_creation_cb(on_host_creation);
432     s4u::Host::on_speed_change_cb([](s4u::Host const& host) {
433       Container::by_name(host.get_name())
434           ->get_variable("speed")
435           ->set_event(simgrid_get_clock(), host.get_core_count() * host.get_available_speed());
436     });
437     s4u::Link::on_creation_cb(on_link_creation);
438     s4u::Link::on_bandwidth_change_cb([](s4u::Link const& link) {
439       Container::by_name(link.get_name())
440           ->get_variable("bandwidth")
441           ->set_event(simgrid_get_clock(), sg_bandwidth_factor * link.get_bandwidth());
442     });
443     s4u::NetZone::on_seal_cb([](s4u::NetZone const& /*netzone*/) { currentContainer.pop_back(); });
444     kernel::routing::NetPoint::on_creation.connect([](kernel::routing::NetPoint const& netpoint) {
445       if (netpoint.is_router())
446         new RouterContainer(netpoint.get_name(), currentContainer.back());
447     });
448   }
449
450   s4u::NetZone::on_creation_cb(on_netzone_creation);
451
452   kernel::resource::CpuAction::on_state_change.connect(on_action_state_change);
453   s4u::Link::on_communication_state_change_cb(on_action_state_change);
454
455   if (TRACE_actor_is_enabled()) {
456     s4u::Actor::on_creation_cb(on_actor_creation);
457     s4u::Actor::on_destruction_cb([](s4u::Actor const& actor) {
458       auto container = Container::by_name_or_null(instr_pid(actor));
459       if (container != nullptr)
460         container->remove_from_parent();
461     });
462     s4u::Actor::on_suspend_cb([](s4u::Actor const& actor) {
463       Container::by_name(instr_pid(actor))->get_state("ACTOR_STATE")->push_event("suspend");
464     });
465     s4u::Actor::on_resume_cb(
466         [](s4u::Actor const& actor) { Container::by_name(instr_pid(actor))->get_state("ACTOR_STATE")->pop_event(); });
467     s4u::Actor::on_sleep_cb([](s4u::Actor const& actor) {
468       Container::by_name(instr_pid(actor))->get_state("ACTOR_STATE")->push_event("sleep");
469     });
470     s4u::Actor::on_wake_up_cb(
471         [](s4u::Actor const& actor) { Container::by_name(instr_pid(actor))->get_state("ACTOR_STATE")->pop_event(); });
472     s4u::Exec::on_start_cb([](s4u::Exec const&) {
473       Container::by_name(instr_pid(*s4u::Actor::self()))->get_state("ACTOR_STATE")->push_event("execute");
474     });
475     s4u::Activity::on_completion_cb([](const s4u::Activity&) {
476       Container::by_name(instr_pid(*s4u::Actor::self()))->get_state("ACTOR_STATE")->pop_event();
477     });
478     s4u::Comm::on_send_cb([](s4u::Comm const&) {
479       Container::by_name(instr_pid(*s4u::Actor::self()))->get_state("ACTOR_STATE")->push_event("send");
480     });
481     s4u::Comm::on_recv_cb([](s4u::Comm const&) {
482       Container::by_name(instr_pid(*s4u::Actor::self()))->get_state("ACTOR_STATE")->push_event("receive");
483     });
484     s4u::Actor::on_host_change_cb(on_actor_host_change);
485   }
486
487   if (TRACE_smpi_is_enabled() && TRACE_smpi_is_computing()) {
488     s4u::Exec::on_start_cb([](s4u::Exec const& exec) {
489       Container::by_name(std::string("rank-") + std::to_string(s4u::Actor::self()->get_pid()))
490           ->get_state("MPI_STATE")
491           ->push_event("computing", new CpuTIData("compute", exec.get_cost()));
492     });
493     s4u::Activity::on_completion_cb([](const s4u::Activity&) {
494       Container::by_name(std::string("rank-") + std::to_string(s4u::Actor::self()->get_pid()))
495           ->get_state("MPI_STATE")
496           ->pop_event();
497     });
498   }
499
500   if (TRACE_vm_is_enabled()) {
501     s4u::Host::on_creation_cb(on_vm_creation);
502     s4u::VirtualMachine::on_start_cb([](s4u::VirtualMachine const& vm) {
503       Container::by_name(vm.get_name())->get_state("VM_STATE")->push_event("start");
504     });
505     s4u::VirtualMachine::on_started_cb(
506         [](s4u::VirtualMachine const& vm) { Container::by_name(vm.get_name())->get_state("VM_STATE")->pop_event(); });
507     s4u::VirtualMachine::on_suspend_cb([](s4u::VirtualMachine const& vm) {
508       Container::by_name(vm.get_name())->get_state("VM_STATE")->push_event("suspend");
509     });
510     s4u::VirtualMachine::on_resume_cb(
511         [](s4u::VirtualMachine const& vm) { Container::by_name(vm.get_name())->get_state("VM_STATE")->pop_event(); });
512     s4u::Host::on_destruction_cb(
513         [](s4u::Host const& host) { Container::by_name(host.get_name())->remove_from_parent(); });
514   }
515 }
516 } // namespace instr
517 } // namespace simgrid