Logo AND Algorithmique Numérique Distribuée

Public GIT Repository
jedule: obey our coding standards
[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->get_children()->empty()) {
131     // bottom-up recursion
132     for (auto const& nz_son : *netzone->get_children()) {
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.get_core_count() * host.get_available_speed());
233 }
234
235 static void instr_cpu_action_on_state_change(simgrid::surf::CpuAction* action)
236 {
237   simgrid::surf::Cpu* cpu = static_cast<simgrid::surf::Cpu*>(action->get_variable()->get_constraint(0)->get_id());
238   TRACE_surf_resource_set_utilization("HOST", "power_used", cpu->get_cname(), action->get_category(),
239                                       action->get_variable()->get_value(), action->get_last_update(),
240                                       SIMIX_get_clock() - action->get_last_update());
241 }
242
243 static void instr_link_on_communication_state_change(simgrid::kernel::resource::NetworkAction* action)
244 {
245   int n = action->get_variable()->get_number_of_constraint();
246
247   for (int i = 0; i < n; i++) {
248     simgrid::kernel::lmm::Constraint* constraint = action->get_variable()->get_constraint(i);
249     simgrid::kernel::resource::LinkImpl* link = static_cast<simgrid::kernel::resource::LinkImpl*>(constraint->get_id());
250     double value = action->get_variable()->get_value() * action->get_variable()->get_constraint_weight(i);
251     TRACE_surf_resource_set_utilization("LINK", "bandwidth_used", link->get_cname(), action->get_category(), value,
252                                         action->get_last_update(), SIMIX_get_clock() - action->get_last_update());
253   }
254 }
255 static void instr_link_on_bandwidth_change(simgrid::s4u::Link& link)
256 {
257   simgrid::instr::Container::by_name(link.get_cname())
258       ->get_variable("bandwidth")
259       ->set_event(surf_get_clock(), sg_bandwidth_factor * link.get_bandwidth());
260 }
261
262 static void instr_netpoint_on_creation(simgrid::kernel::routing::NetPoint* netpoint)
263 {
264   if (netpoint->is_router())
265     new simgrid::instr::RouterContainer(netpoint->get_cname(), currentContainer.back());
266 }
267
268 static void instr_on_platform_created()
269 {
270   currentContainer.clear();
271   std::set<std::string>* filter = new std::set<std::string>;
272   XBT_DEBUG("Starting graph extraction.");
273   recursiveGraphExtraction(simgrid::s4u::Engine::get_instance()->get_netzone_root(),
274                            simgrid::instr::Container::get_root(), filter);
275   XBT_DEBUG("Graph extraction finished.");
276   delete filter;
277   TRACE_paje_dump_buffer(true);
278 }
279
280 static void instr_actor_on_creation(simgrid::s4u::ActorPtr actor)
281 {
282   container_t root      = simgrid::instr::Container::get_root();
283   container_t container = simgrid::instr::Container::by_name(actor->get_host()->get_name());
284
285   container->create_child(instr_pid(actor.get()), "ACTOR");
286   simgrid::instr::ContainerType* actor_type =
287       container->type_->by_name_or_create<simgrid::instr::ContainerType>("ACTOR");
288   simgrid::instr::StateType* state = actor_type->by_name_or_create<simgrid::instr::StateType>("ACTOR_STATE");
289   state->add_entity_value("suspend", "1 0 1");
290   state->add_entity_value("sleep", "1 1 0");
291   state->add_entity_value("receive", "1 0 0");
292   state->add_entity_value("send", "0 0 1");
293   state->add_entity_value("task_execute", "0 1 1");
294   root->type_->by_name_or_create("ACTOR_LINK", actor_type, actor_type);
295   root->type_->by_name_or_create("ACTOR_TASK_LINK", actor_type, actor_type);
296
297   actor->on_exit(
298       [](int status, void* actor) {
299         if (status == SMX_EXIT_FAILURE)
300           // kill means that this actor no longer exists, let's destroy it
301           simgrid::instr::Container::by_name(instr_pid(static_cast<simgrid::s4u::Actor*>(actor)))->remove_from_parent();
302       },
303       actor->get_impl());
304 }
305
306 static long long int counter = 0;
307
308 static void instr_actor_on_migration_start(simgrid::s4u::ActorPtr actor)
309 {
310   // start link
311   container_t container = simgrid::instr::Container::by_name(instr_pid(actor.get()));
312   simgrid::instr::Container::get_root()->get_link("ACTOR_LINK")->start_event(container, "M", std::to_string(counter));
313
314   // destroy existing container of this process
315   container->remove_from_parent();
316 }
317
318 static void instr_actor_on_migration_end(simgrid::s4u::ActorPtr actor)
319 {
320   // create new container on the new_host location
321   simgrid::instr::Container::by_name(actor->get_host()->get_name())->create_child(instr_pid(actor.get()), "ACTOR");
322   // end link
323   simgrid::instr::Container::get_root()
324       ->get_link("ACTOR_LINK")
325       ->end_event(simgrid::instr::Container::by_name(instr_pid(actor.get())), "M", std::to_string(counter));
326   counter++;
327 }
328
329 static void instr_vm_on_creation(simgrid::s4u::Host& host)
330 {
331   container_t container             = new simgrid::instr::HostContainer(host, currentContainer.back());
332   container_t root                  = simgrid::instr::Container::get_root();
333   simgrid::instr::ContainerType* vm = container->type_->by_name_or_create<simgrid::instr::ContainerType>("VM");
334   simgrid::instr::StateType* state  = vm->by_name_or_create<simgrid::instr::StateType>("VM_STATE");
335   state->add_entity_value("suspend", "1 0 1");
336   state->add_entity_value("sleep", "1 1 0");
337   state->add_entity_value("receive", "1 0 0");
338   state->add_entity_value("send", "0 0 1");
339   state->add_entity_value("task_execute", "0 1 1");
340   root->type_->by_name_or_create("VM_LINK", vm, vm);
341   root->type_->by_name_or_create("VM_ACTOR_LINK", vm, vm);
342 }
343
344 void instr_define_callbacks()
345 {
346   // always need the callbacks to zones (we need only the root zone), to create the rootContainer and the rootType
347   // properly
348   if (TRACE_needs_platform()) {
349     simgrid::s4u::on_platform_created.connect(instr_on_platform_created);
350     simgrid::s4u::Host::on_creation.connect(instr_host_on_creation);
351     simgrid::s4u::Host::on_speed_change.connect(instr_host_on_speed_change);
352     simgrid::s4u::Link::on_creation.connect(instr_link_on_creation);
353     simgrid::s4u::Link::on_bandwidth_change.connect(instr_link_on_bandwidth_change);
354     simgrid::s4u::NetZone::on_seal.connect([](simgrid::s4u::NetZone& /*netzone*/) { currentContainer.pop_back(); });
355     simgrid::kernel::routing::NetPoint::onCreation.connect(instr_netpoint_on_creation);
356   }
357   simgrid::s4u::NetZone::on_creation.connect(instr_netzone_on_creation);
358
359   simgrid::surf::CpuAction::on_state_change.connect(instr_cpu_action_on_state_change);
360   simgrid::s4u::Link::on_communication_state_change.connect(instr_link_on_communication_state_change);
361
362   if (TRACE_actor_is_enabled()) {
363     simgrid::s4u::Actor::on_creation.connect(instr_actor_on_creation);
364     simgrid::s4u::Actor::on_suspend.connect([](simgrid::s4u::ActorPtr actor) {
365       simgrid::instr::Container::by_name(instr_pid(actor.get()))->get_state("ACTOR_STATE")->push_event("suspend");
366     });
367     simgrid::s4u::Actor::on_resume.connect([](simgrid::s4u::ActorPtr actor) {
368       simgrid::instr::Container::by_name(instr_pid(actor.get()))->get_state("ACTOR_STATE")->pop_event();
369     });
370     simgrid::s4u::Actor::on_sleep.connect([](simgrid::s4u::ActorPtr actor) {
371       simgrid::instr::Container::by_name(instr_pid(actor.get()))->get_state("ACTOR_STATE")->push_event("sleep");
372     });
373     simgrid::s4u::Actor::on_wake_up.connect([](simgrid::s4u::ActorPtr actor) {
374       simgrid::instr::Container::by_name(instr_pid(actor.get()))->get_state("ACTOR_STATE")->pop_event();
375     });
376     simgrid::s4u::Actor::on_migration_start.connect(instr_actor_on_migration_start);
377     simgrid::s4u::Actor::on_migration_end.connect(instr_actor_on_migration_end);
378   }
379
380   if (TRACE_vm_is_enabled()) {
381     simgrid::s4u::Host::on_creation.connect(instr_vm_on_creation);
382     simgrid::s4u::VirtualMachine::on_start.connect([](simgrid::s4u::VirtualMachine& vm) {
383       simgrid::instr::Container::by_name(vm.get_name())->get_state("VM_STATE")->push_event("start");
384     });
385     simgrid::s4u::VirtualMachine::on_started.connect([](simgrid::s4u::VirtualMachine& vm) {
386       simgrid::instr::Container::by_name(vm.get_name())->get_state("VM_STATE")->pop_event();
387     });
388     simgrid::s4u::VirtualMachine::on_suspend.connect([](simgrid::s4u::VirtualMachine& vm) {
389       simgrid::instr::Container::by_name(vm.get_name())->get_state("VM_STATE")->push_event("suspend");
390     });
391     simgrid::s4u::VirtualMachine::on_resume.connect([](simgrid::s4u::VirtualMachine& vm) {
392       simgrid::instr::Container::by_name(vm.get_name())->get_state("VM_STATE")->pop_event();
393     });
394     simgrid::s4u::Host::on_destruction.connect(
395         [](simgrid::s4u::Host& host) { simgrid::instr::Container::by_name(host.get_name())->remove_from_parent(); });
396   }
397 }
398 /*
399  * user categories support
400  */
401 static void recursiveNewVariableType(std::string new_typename, std::string color, simgrid::instr::Type* root)
402 {
403   if (root->get_name() == "HOST" || root->get_name() == "VM")
404     root->by_name_or_create(std::string("p") + new_typename, color);
405
406   if (root->get_name() == "LINK")
407     root->by_name_or_create(std::string("b") + new_typename, color);
408
409   for (auto elm : root->children_) {
410     recursiveNewVariableType(new_typename, color, elm.second);
411   }
412 }
413
414 void instr_new_variable_type(std::string new_typename, std::string color)
415 {
416   recursiveNewVariableType(new_typename, color, simgrid::instr::Container::get_root()->type_);
417 }
418
419 static void recursiveNewUserVariableType(std::string father_type, std::string new_typename, std::string color,
420                                          simgrid::instr::Type* root)
421 {
422   if (root->get_name() == father_type) {
423     root->by_name_or_create(new_typename, color);
424   }
425   for (auto elm : root->children_)
426     recursiveNewUserVariableType(father_type, new_typename, color, elm.second);
427 }
428
429 void instr_new_user_variable_type(std::string father_type, std::string new_typename, std::string color)
430 {
431   recursiveNewUserVariableType(father_type, new_typename, color, simgrid::instr::Container::get_root()->type_);
432 }
433
434 static void recursiveNewUserStateType(std::string father_type, std::string new_typename, simgrid::instr::Type* root)
435 {
436   if (root->get_name() == father_type)
437     root->by_name_or_create<simgrid::instr::StateType>(new_typename);
438
439   for (auto elm : root->children_)
440     recursiveNewUserStateType(father_type, new_typename, elm.second);
441 }
442
443 void instr_new_user_state_type(std::string father_type, std::string new_typename)
444 {
445   recursiveNewUserStateType(father_type, new_typename, simgrid::instr::Container::get_root()->type_);
446 }
447
448 static void recursiveNewValueForUserStateType(std::string type_name, const char* val, std::string color,
449                                               simgrid::instr::Type* root)
450 {
451   if (root->get_name() == type_name)
452     static_cast<simgrid::instr::StateType*>(root)->add_entity_value(val, color);
453
454   for (auto elm : root->children_)
455     recursiveNewValueForUserStateType(type_name, val, color, elm.second);
456 }
457
458 void instr_new_value_for_user_state_type(std::string type_name, const char* value, std::string color)
459 {
460   recursiveNewValueForUserStateType(type_name, value, color, simgrid::instr::Container::get_root()->type_);
461 }
462
463 #define GRAPHICATOR_SUPPORT_FUNCTIONS
464
465 static void recursiveXBTGraphExtraction(xbt_graph_t graph, std::map<std::string, xbt_node_t>* nodes,
466                                         std::map<std::string, xbt_edge_t>* edges, sg_netzone_t netzone,
467                                         container_t container)
468 {
469   if (not netzone->get_children()->empty()) {
470     // bottom-up recursion
471     for (auto const& netzone_child : *netzone->get_children()) {
472       container_t child_container = container->children_.at(netzone_child->get_cname());
473       recursiveXBTGraphExtraction(graph, nodes, edges, netzone_child, child_container);
474     }
475   }
476
477   static_cast<simgrid::kernel::routing::NetZoneImpl*>(netzone)->get_graph(graph, nodes, edges);
478 }
479
480 xbt_graph_t instr_routing_platform_graph()
481 {
482   xbt_graph_t ret                          = xbt_graph_new_graph(0, nullptr);
483   std::map<std::string, xbt_node_t>* nodes = new std::map<std::string, xbt_node_t>;
484   std::map<std::string, xbt_edge_t>* edges = new std::map<std::string, xbt_edge_t>;
485   recursiveXBTGraphExtraction(ret, nodes, edges, simgrid::s4u::Engine::get_instance()->get_netzone_root(),
486                               simgrid::instr::Container::get_root());
487   delete nodes;
488   delete edges;
489   return ret;
490 }
491
492 void instr_routing_platform_graph_export_graphviz(xbt_graph_t g, const char* filename)
493 {
494   unsigned int cursor = 0;
495   xbt_node_t node     = nullptr;
496   xbt_edge_t edge     = nullptr;
497
498   FILE* file = fopen(filename, "w");
499   xbt_assert(file, "Failed to open %s \n", filename);
500
501   if (g->directed)
502     fprintf(file, "digraph test {\n");
503   else
504     fprintf(file, "graph test {\n");
505
506   fprintf(file, "  graph [overlap=scale]\n");
507
508   fprintf(file, "  node [shape=box, style=filled]\n");
509   fprintf(file, "  node [width=.3, height=.3, style=filled, color=skyblue]\n\n");
510
511   xbt_dynar_foreach (g->nodes, cursor, node) {
512     fprintf(file, "  \"%s\";\n", instr_node_name(node));
513   }
514   xbt_dynar_foreach (g->edges, cursor, edge) {
515     const char* src_s = instr_node_name(edge->src);
516     const char* dst_s = instr_node_name(edge->dst);
517     if (g->directed)
518       fprintf(file, "  \"%s\" -> \"%s\";\n", src_s, dst_s);
519     else
520       fprintf(file, "  \"%s\" -- \"%s\";\n", src_s, dst_s);
521   }
522   fprintf(file, "}\n");
523   fclose(file);
524 }