Logo AND Algorithmique Numérique Distribuée

Public GIT Repository
7bbf6bf3c4cbfefaf0efb2a3870aff1778876ee7
[simgrid.git] / src / instr / instr_platform.cpp
1 /* Copyright (c) 2010-2018. 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 "src/instr/instr_private.hpp"
7
8 #include "simgrid/kernel/routing/NetPoint.hpp"
9 #include "simgrid/kernel/routing/NetZoneImpl.hpp"
10 #include "simgrid/s4u/Actor.hpp"
11 #include "simgrid/s4u/Engine.hpp"
12 #include "simgrid/s4u/Host.hpp"
13 #include "simgrid/s4u/VirtualMachine.hpp"
14 #include "src/surf/cpu_interface.hpp"
15 #include "src/surf/network_interface.hpp"
16 #include "src/surf/surf_interface.hpp"
17 #include "src/surf/xml/platf_private.hpp"
18 #include "surf/surf.hpp"
19 #include "xbt/graph.h"
20
21 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(instr_routing, instr, "Tracing platform hierarchy");
22
23 static std::vector<simgrid::instr::NetZoneContainer*> currentContainer; /* push and pop, used only in creation */
24
25 static const char* instr_node_name(xbt_node_t node)
26 {
27   return static_cast<char*>(xbt_graph_node_get_data(node));
28 }
29
30 static container_t lowestCommonAncestor(container_t a1, container_t a2)
31 {
32   // this is only an optimization (since most of a1 and a2 share the same parent)
33   if (a1->father_ == a2->father_)
34     return a1->father_;
35
36   // create an array with all ancestors of a1
37   std::vector<container_t> ancestors_a1;
38   container_t p = a1->father_;
39   while (p) {
40     ancestors_a1.push_back(p);
41     p = p->father_;
42   }
43
44   // create an array with all ancestors of a2
45   std::vector<container_t> ancestors_a2;
46   p = a2->father_;
47   while (p) {
48     ancestors_a2.push_back(p);
49     p = p->father_;
50   }
51
52   // find the lowest ancestor
53   p     = nullptr;
54   int i = ancestors_a1.size() - 1;
55   int j = ancestors_a2.size() - 1;
56   while (i >= 0 && j >= 0) {
57     container_t a1p = ancestors_a1.at(i);
58     container_t a2p = ancestors_a2.at(j);
59     if (a1p == a2p) {
60       p = a1p;
61     } else {
62       break;
63     }
64     i--;
65     j--;
66   }
67   return p;
68 }
69
70 static void linkContainers(container_t src, container_t dst, std::set<std::string>* filter)
71 {
72   // ignore loopback
73   if (src->get_name() == "__loopback__" || dst->get_name() == "__loopback__") {
74     XBT_DEBUG("  linkContainers: ignoring loopback link");
75     return;
76   }
77
78   // find common father
79   container_t father = lowestCommonAncestor(src, dst);
80   if (not father) {
81     xbt_die("common father unknown, this is a tracing problem");
82   }
83
84   // check if we already register this pair (we only need one direction)
85   std::string aux1 = src->get_name() + dst->get_name();
86   std::string aux2 = dst->get_name() + src->get_name();
87   if (filter->find(aux1) != filter->end()) {
88     XBT_DEBUG("  linkContainers: already registered %s <-> %s (1)", src->get_cname(), dst->get_cname());
89     return;
90   }
91   if (filter->find(aux2) != filter->end()) {
92     XBT_DEBUG("  linkContainers: already registered %s <-> %s (2)", dst->get_cname(), src->get_cname());
93     return;
94   }
95
96   // ok, not found, register it
97   filter->insert(aux1);
98   filter->insert(aux2);
99
100   // declare type
101   std::string link_typename = father->type_->get_name() + "-" + src->type_->get_name() +
102                               std::to_string(src->type_->get_id()) + "-" + dst->type_->get_name() +
103                               std::to_string(dst->type_->get_id());
104   simgrid::instr::LinkType* link = father->type_->by_name_or_create(link_typename, src->type_, dst->type_);
105   link->set_calling_container(father);
106
107   // register EDGE types for triva configuration
108   trivaEdgeTypes.insert(link->get_name());
109
110   // create the link
111   static long long counter = 0;
112
113   std::string key = std::to_string(counter);
114   counter++;
115
116   link->start_event(src, "topology", key);
117   link->end_event(dst, "topology", key);
118
119   XBT_DEBUG("  linkContainers %s <-> %s", src->get_cname(), dst->get_cname());
120 }
121
122 static void recursiveGraphExtraction(simgrid::s4u::NetZone* netzone, container_t container,
123                                      std::set<std::string>* filter)
124 {
125   if (not TRACE_platform_topology()) {
126     XBT_DEBUG("Graph extraction disabled by user.");
127     return;
128   }
129   XBT_DEBUG("Graph extraction for NetZone = %s", netzone->get_cname());
130   if (not netzone->getChildren()->empty()) {
131     // bottom-up recursion
132     for (auto const& nz_son : *netzone->getChildren()) {
133       container_t child_container = container->children_.at(nz_son->get_cname());
134       recursiveGraphExtraction(nz_son, child_container, filter);
135     }
136   }
137
138   xbt_graph_t graph                        = xbt_graph_new_graph(0, nullptr);
139   std::map<std::string, xbt_node_t>* nodes = new std::map<std::string, xbt_node_t>;
140   std::map<std::string, xbt_edge_t>* edges = new std::map<std::string, xbt_edge_t>;
141
142   static_cast<simgrid::kernel::routing::NetZoneImpl*>(netzone)->get_graph(graph, nodes, edges);
143   for (auto elm : *edges) {
144     xbt_edge_t edge = elm.second;
145     linkContainers(simgrid::instr::Container::by_name(static_cast<const char*>(edge->src->data)),
146                    simgrid::instr::Container::by_name(static_cast<const char*>(edge->dst->data)), filter);
147   }
148   delete nodes;
149   delete edges;
150   xbt_graph_free_graph(graph, xbt_free_f, xbt_free_f, nullptr);
151 }
152
153 /*
154  * Callbacks
155  */
156 static void instr_netzone_on_creation(simgrid::s4u::NetZone& netzone)
157 {
158   std::string id = netzone.get_name();
159   if (simgrid::instr::Container::get_root() == nullptr) {
160     simgrid::instr::NetZoneContainer* root = new simgrid::instr::NetZoneContainer(id, 0, nullptr);
161
162     if (TRACE_smpi_is_enabled()) {
163       simgrid::instr::ContainerType* mpi = root->type_->by_name_or_create<simgrid::instr::ContainerType>("MPI");
164       if (not TRACE_smpi_is_grouped())
165         mpi->by_name_or_create<simgrid::instr::StateType>("MPI_STATE");
166       root->type_->by_name_or_create("MPI_LINK", mpi, mpi);
167       // TODO See if we can move this to the LoadBalancer plugin
168       root->type_->by_name_or_create("MIGRATE_LINK", mpi, mpi);
169       mpi->by_name_or_create<simgrid::instr::StateType>("MIGRATE_STATE");
170     }
171
172     if (TRACE_needs_platform()) {
173       currentContainer.push_back(root);
174     }
175     return;
176   }
177
178   if (TRACE_needs_platform()) {
179     simgrid::instr::NetZoneContainer* container =
180         new simgrid::instr::NetZoneContainer(id, currentContainer.size(), currentContainer.back());
181     currentContainer.push_back(container);
182   }
183 }
184
185 static void instr_link_on_creation(simgrid::s4u::Link& link)
186 {
187   if (currentContainer.empty()) // No ongoing parsing. Are you creating the loopback?
188     return;
189
190   container_t container = new simgrid::instr::Container(link.get_name(), "LINK", currentContainer.back());
191
192   if ((TRACE_categorized() || TRACE_uncategorized() || TRACE_platform()) && (not TRACE_disable_link())) {
193     simgrid::instr::VariableType* bandwidth = container->type_->by_name_or_create("bandwidth", "");
194     bandwidth->set_calling_container(container);
195     bandwidth->set_event(0, link.get_bandwidth());
196     simgrid::instr::VariableType* latency = container->type_->by_name_or_create("latency", "");
197     latency->set_calling_container(container);
198     latency->set_event(0, link.get_latency());
199   }
200   if (TRACE_uncategorized()) {
201     container->type_->by_name_or_create("bandwidth_used", "0.5 0.5 0.5");
202   }
203 }
204
205 static void instr_host_on_creation(simgrid::s4u::Host& host)
206 {
207   container_t container = new simgrid::instr::HostContainer(host, currentContainer.back());
208   container_t root      = simgrid::instr::Container::get_root();
209
210   if ((TRACE_categorized() || TRACE_uncategorized() || TRACE_platform()) && (not TRACE_disable_speed())) {
211     simgrid::instr::VariableType* power = container->type_->by_name_or_create("power", "");
212     power->set_calling_container(container);
213     power->set_event(0, host.getSpeed());
214   }
215
216   if (TRACE_uncategorized())
217     container->type_->by_name_or_create("power_used", "0.5 0.5 0.5");
218
219   if (TRACE_smpi_is_enabled() && TRACE_smpi_is_grouped()) {
220     simgrid::instr::ContainerType* mpi = container->type_->by_name_or_create<simgrid::instr::ContainerType>("MPI");
221     mpi->by_name_or_create<simgrid::instr::StateType>("MPI_STATE");
222     // TODO See if we can move this to the LoadBalancer plugin
223     root->type_->by_name_or_create("MIGRATE_LINK", mpi, mpi);
224     mpi->by_name_or_create<simgrid::instr::StateType>("MIGRATE_STATE");
225   }
226 }
227
228 static void instr_host_on_speed_change(simgrid::s4u::Host& host)
229 {
230   simgrid::instr::Container::by_name(host.get_cname())
231       ->get_variable("power")
232       ->set_event(surf_get_clock(), host.getCoreCount() * host.get_available_speed());
233 }
234
235 static void instr_cpu_action_on_state_change(simgrid::surf::CpuAction* action,
236                                              simgrid::kernel::resource::Action::State /* previous */)
237 {
238   simgrid::surf::Cpu* cpu = static_cast<simgrid::surf::Cpu*>(action->get_variable()->get_constraint(0)->get_id());
239   TRACE_surf_resource_set_utilization("HOST", "power_used", cpu->get_cname(), action->get_category(),
240                                       action->get_variable()->get_value(), action->get_last_update(),
241                                       SIMIX_get_clock() - action->get_last_update());
242 }
243
244 static void instr_link_on_communication_state_change(simgrid::kernel::resource::NetworkAction* action)
245 {
246   int n = action->get_variable()->get_number_of_constraint();
247
248   for (int i = 0; i < n; i++) {
249     simgrid::kernel::lmm::Constraint* constraint = action->get_variable()->get_constraint(i);
250     simgrid::kernel::resource::LinkImpl* link = static_cast<simgrid::kernel::resource::LinkImpl*>(constraint->get_id());
251     double value = action->get_variable()->get_value() * action->get_variable()->get_constraint_weight(i);
252     TRACE_surf_resource_set_utilization("LINK", "bandwidth_used", link->get_cname(), action->get_category(), value,
253                                         action->get_last_update(), SIMIX_get_clock() - action->get_last_update());
254   }
255 }
256 static void instr_link_on_bandwidth_change(simgrid::s4u::Link& link)
257 {
258   simgrid::instr::Container::by_name(link.get_cname())
259       ->get_variable("bandwidth")
260       ->set_event(surf_get_clock(), sg_bandwidth_factor * link.get_bandwidth());
261 }
262
263 static void instr_netpoint_on_creation(simgrid::kernel::routing::NetPoint* netpoint)
264 {
265   if (netpoint->is_router())
266     new simgrid::instr::RouterContainer(netpoint->get_cname(), currentContainer.back());
267 }
268
269 static void instr_on_platform_created()
270 {
271   currentContainer.clear();
272   std::set<std::string>* filter = new std::set<std::string>;
273   XBT_DEBUG("Starting graph extraction.");
274   recursiveGraphExtraction(simgrid::s4u::Engine::get_instance()->get_netzone_root(),
275                            simgrid::instr::Container::get_root(), filter);
276   XBT_DEBUG("Graph extraction finished.");
277   delete filter;
278   TRACE_paje_dump_buffer(true);
279 }
280
281 static void TRACE_actor_kill(smx_process_exit_status_t status, simgrid::s4u::Actor* actor)
282 {
283   if (status == SMX_EXIT_FAILURE)
284     // kill means that this actor no longer exists, let's destroy it
285     simgrid::instr::Container::by_name(instr_pid(actor))->remove_from_parent();
286 }
287
288 static void instr_actor_on_creation(simgrid::s4u::ActorPtr actor)
289 {
290   container_t root      = simgrid::instr::Container::get_root();
291   container_t container = simgrid::instr::Container::by_name(actor->get_host()->get_name());
292
293   container->create_child(instr_pid(actor.get()), "ACTOR");
294   simgrid::instr::ContainerType* actor_type =
295       container->type_->by_name_or_create<simgrid::instr::ContainerType>("ACTOR");
296   simgrid::instr::StateType* state = actor_type->by_name_or_create<simgrid::instr::StateType>("ACTOR_STATE");
297   state->add_entity_value("suspend", "1 0 1");
298   state->add_entity_value("sleep", "1 1 0");
299   state->add_entity_value("receive", "1 0 0");
300   state->add_entity_value("send", "0 0 1");
301   state->add_entity_value("task_execute", "0 1 1");
302   root->type_->by_name_or_create("ACTOR_LINK", actor_type, actor_type);
303   root->type_->by_name_or_create("ACTOR_TASK_LINK", actor_type, actor_type);
304
305   actor->on_exit((int_f_pvoid_pvoid_t)TRACE_actor_kill, actor->get_impl());
306 }
307
308 static long long int counter = 0;
309
310 static void instr_actor_on_migration_start(simgrid::s4u::ActorPtr actor)
311 {
312   // start link
313   container_t container = simgrid::instr::Container::by_name(instr_pid(actor.get()));
314   simgrid::instr::Container::get_root()->get_link("ACTOR_LINK")->start_event(container, "M", std::to_string(counter));
315
316   // destroy existing container of this process
317   container->remove_from_parent();
318 }
319
320 static void instr_actor_on_migration_end(simgrid::s4u::ActorPtr actor)
321 {
322   // create new container on the new_host location
323   simgrid::instr::Container::by_name(actor->get_host()->get_name())->create_child(instr_pid(actor.get()), "ACTOR");
324   // end link
325   simgrid::instr::Container::get_root()
326       ->get_link("ACTOR_LINK")
327       ->end_event(simgrid::instr::Container::by_name(instr_pid(actor.get())), "M", std::to_string(counter));
328   counter++;
329 }
330
331 static void instr_vm_on_creation(simgrid::s4u::Host& host)
332 {
333   container_t container             = new simgrid::instr::HostContainer(host, currentContainer.back());
334   container_t root                  = simgrid::instr::Container::get_root();
335   simgrid::instr::ContainerType* vm = container->type_->by_name_or_create<simgrid::instr::ContainerType>("VM");
336   simgrid::instr::StateType* state  = vm->by_name_or_create<simgrid::instr::StateType>("VM_STATE");
337   state->add_entity_value("suspend", "1 0 1");
338   state->add_entity_value("sleep", "1 1 0");
339   state->add_entity_value("receive", "1 0 0");
340   state->add_entity_value("send", "0 0 1");
341   state->add_entity_value("task_execute", "0 1 1");
342   root->type_->by_name_or_create("VM_LINK", vm, vm);
343   root->type_->by_name_or_create("VM_ACTOR_LINK", vm, vm);
344 }
345
346 void instr_define_callbacks()
347 {
348   // always need the callbacks to zones (we need only the root zone), to create the rootContainer and the rootType
349   // properly
350   if (TRACE_needs_platform()) {
351     simgrid::s4u::on_platform_created.connect(instr_on_platform_created);
352     simgrid::s4u::Host::on_creation.connect(instr_host_on_creation);
353     simgrid::s4u::Host::on_speed_change.connect(instr_host_on_speed_change);
354     simgrid::s4u::Link::on_creation.connect(instr_link_on_creation);
355     simgrid::s4u::Link::on_bandwidth_change.connect(instr_link_on_bandwidth_change);
356     simgrid::s4u::NetZone::onSeal.connect([](simgrid::s4u::NetZone& /*netzone*/) { currentContainer.pop_back(); });
357     simgrid::kernel::routing::NetPoint::onCreation.connect(instr_netpoint_on_creation);
358   }
359   simgrid::s4u::NetZone::onCreation.connect(instr_netzone_on_creation);
360
361   simgrid::surf::CpuAction::onStateChange.connect(instr_cpu_action_on_state_change);
362   simgrid::s4u::Link::on_communication_state_change.connect(instr_link_on_communication_state_change);
363
364   if (TRACE_actor_is_enabled()) {
365     simgrid::s4u::Actor::on_creation.connect(instr_actor_on_creation);
366     simgrid::s4u::Actor::on_suspend.connect([](simgrid::s4u::ActorPtr actor) {
367       simgrid::instr::Container::by_name(instr_pid(actor.get()))->get_state("ACTOR_STATE")->push_event("suspend");
368     });
369     simgrid::s4u::Actor::on_resume.connect([](simgrid::s4u::ActorPtr actor) {
370       simgrid::instr::Container::by_name(instr_pid(actor.get()))->get_state("ACTOR_STATE")->pop_event();
371     });
372     simgrid::s4u::Actor::on_sleep.connect([](simgrid::s4u::ActorPtr actor) {
373       simgrid::instr::Container::by_name(instr_pid(actor.get()))->get_state("ACTOR_STATE")->push_event("sleep");
374     });
375     simgrid::s4u::Actor::on_wake_up.connect([](simgrid::s4u::ActorPtr actor) {
376       simgrid::instr::Container::by_name(instr_pid(actor.get()))->get_state("ACTOR_STATE")->pop_event();
377     });
378     simgrid::s4u::Actor::on_migration_start.connect(instr_actor_on_migration_start);
379     simgrid::s4u::Actor::on_migration_end.connect(instr_actor_on_migration_end);
380   }
381
382   if (TRACE_vm_is_enabled()) {
383     simgrid::s4u::Host::on_creation.connect(instr_vm_on_creation);
384     simgrid::s4u::VirtualMachine::on_start.connect([](simgrid::s4u::VirtualMachine& vm) {
385       simgrid::instr::Container::by_name(vm.get_name())->get_state("VM_STATE")->push_event("start");
386     });
387     simgrid::s4u::VirtualMachine::on_started.connect([](simgrid::s4u::VirtualMachine& vm) {
388       simgrid::instr::Container::by_name(vm.get_name())->get_state("VM_STATE")->pop_event();
389     });
390     simgrid::s4u::VirtualMachine::on_suspend.connect([](simgrid::s4u::VirtualMachine& vm) {
391       simgrid::instr::Container::by_name(vm.get_name())->get_state("VM_STATE")->push_event("suspend");
392     });
393     simgrid::s4u::VirtualMachine::on_resume.connect([](simgrid::s4u::VirtualMachine& vm) {
394       simgrid::instr::Container::by_name(vm.get_name())->get_state("VM_STATE")->pop_event();
395     });
396     simgrid::s4u::Host::on_destruction.connect(
397         [](simgrid::s4u::Host& host) { simgrid::instr::Container::by_name(host.get_name())->remove_from_parent(); });
398   }
399 }
400 /*
401  * user categories support
402  */
403 static void recursiveNewVariableType(std::string new_typename, std::string color, simgrid::instr::Type* root)
404 {
405   if (root->get_name() == "HOST" || root->get_name() == "VM")
406     root->by_name_or_create(std::string("p") + new_typename, color);
407
408   if (root->get_name() == "LINK")
409     root->by_name_or_create(std::string("b") + new_typename, color);
410
411   for (auto elm : root->children_) {
412     recursiveNewVariableType(new_typename, color, elm.second);
413   }
414 }
415
416 void instr_new_variable_type(std::string new_typename, std::string color)
417 {
418   recursiveNewVariableType(new_typename, color, simgrid::instr::Container::get_root()->type_);
419 }
420
421 static void recursiveNewUserVariableType(std::string father_type, std::string new_typename, std::string color,
422                                          simgrid::instr::Type* root)
423 {
424   if (root->get_name() == father_type) {
425     root->by_name_or_create(new_typename, color);
426   }
427   for (auto elm : root->children_)
428     recursiveNewUserVariableType(father_type, new_typename, color, elm.second);
429 }
430
431 void instr_new_user_variable_type(std::string father_type, std::string new_typename, std::string color)
432 {
433   recursiveNewUserVariableType(father_type, new_typename, color, simgrid::instr::Container::get_root()->type_);
434 }
435
436 static void recursiveNewUserStateType(std::string father_type, std::string new_typename, simgrid::instr::Type* root)
437 {
438   if (root->get_name() == father_type)
439     root->by_name_or_create<simgrid::instr::StateType>(new_typename);
440
441   for (auto elm : root->children_)
442     recursiveNewUserStateType(father_type, new_typename, elm.second);
443 }
444
445 void instr_new_user_state_type(std::string father_type, std::string new_typename)
446 {
447   recursiveNewUserStateType(father_type, new_typename, simgrid::instr::Container::get_root()->type_);
448 }
449
450 static void recursiveNewValueForUserStateType(std::string type_name, const char* val, std::string color,
451                                               simgrid::instr::Type* root)
452 {
453   if (root->get_name() == type_name)
454     static_cast<simgrid::instr::StateType*>(root)->add_entity_value(val, color);
455
456   for (auto elm : root->children_)
457     recursiveNewValueForUserStateType(type_name, val, color, elm.second);
458 }
459
460 void instr_new_value_for_user_state_type(std::string type_name, const char* value, std::string color)
461 {
462   recursiveNewValueForUserStateType(type_name, value, color, simgrid::instr::Container::get_root()->type_);
463 }
464
465 #define GRAPHICATOR_SUPPORT_FUNCTIONS
466
467 static void recursiveXBTGraphExtraction(xbt_graph_t graph, std::map<std::string, xbt_node_t>* nodes,
468                                         std::map<std::string, xbt_edge_t>* edges, sg_netzone_t netzone,
469                                         container_t container)
470 {
471   if (not netzone->getChildren()->empty()) {
472     // bottom-up recursion
473     for (auto const& netzone_child : *netzone->getChildren()) {
474       container_t child_container = container->children_.at(netzone_child->get_cname());
475       recursiveXBTGraphExtraction(graph, nodes, edges, netzone_child, child_container);
476     }
477   }
478
479   static_cast<simgrid::kernel::routing::NetZoneImpl*>(netzone)->get_graph(graph, nodes, edges);
480 }
481
482 xbt_graph_t instr_routing_platform_graph()
483 {
484   xbt_graph_t ret                          = xbt_graph_new_graph(0, nullptr);
485   std::map<std::string, xbt_node_t>* nodes = new std::map<std::string, xbt_node_t>;
486   std::map<std::string, xbt_edge_t>* edges = new std::map<std::string, xbt_edge_t>;
487   recursiveXBTGraphExtraction(ret, nodes, edges, simgrid::s4u::Engine::get_instance()->get_netzone_root(),
488                               simgrid::instr::Container::get_root());
489   delete nodes;
490   delete edges;
491   return ret;
492 }
493
494 void instr_routing_platform_graph_export_graphviz(xbt_graph_t g, const char* filename)
495 {
496   unsigned int cursor = 0;
497   xbt_node_t node     = nullptr;
498   xbt_edge_t edge     = nullptr;
499
500   FILE* file = fopen(filename, "w");
501   xbt_assert(file, "Failed to open %s \n", filename);
502
503   if (g->directed)
504     fprintf(file, "digraph test {\n");
505   else
506     fprintf(file, "graph test {\n");
507
508   fprintf(file, "  graph [overlap=scale]\n");
509
510   fprintf(file, "  node [shape=box, style=filled]\n");
511   fprintf(file, "  node [width=.3, height=.3, style=filled, color=skyblue]\n\n");
512
513   xbt_dynar_foreach (g->nodes, cursor, node) {
514     fprintf(file, "  \"%s\";\n", instr_node_name(node));
515   }
516   xbt_dynar_foreach (g->edges, cursor, edge) {
517     const char* src_s = instr_node_name(edge->src);
518     const char* dst_s = instr_node_name(edge->dst);
519     if (g->directed)
520       fprintf(file, "  \"%s\" -> \"%s\";\n", src_s, dst_s);
521     else
522       fprintf(file, "  \"%s\" -- \"%s\";\n", src_s, dst_s);
523   }
524   fprintf(file, "}\n");
525   fclose(file);
526 }