1 /* Copyright (c) 2007-2016. 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. */
7 #include "smx_private.h"
8 #include "src/kernel/activity/CommImpl.hpp"
9 #include "src/mc/mc_replay.h"
10 #include "src/plugins/vm/VirtualMachineImpl.hpp"
11 #include "src/surf/surf_interface.hpp"
14 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(simix_host, simix, "SIMIX hosts");
18 simgrid::xbt::Extension<simgrid::s4u::Host, Host> Host::EXTENSION_ID;
22 if (not Host::EXTENSION_ID.valid())
23 Host::EXTENSION_ID = s4u::Host::extension_create<simix::Host>();
25 simgrid::simix::ActorImpl act;
26 process_list = xbt_swag_new(xbt_swag_offset(act, host_proc_hookup));
31 /* All processes should be gone when the host is turned off (by the end of the simulation). */
32 if (xbt_swag_size(process_list) != 0) {
33 std::string msg = std::string("Shutting down host, but it's not empty:");
34 smx_actor_t process = nullptr;
36 xbt_swag_foreach(process, process_list) msg = msg + "\n\t" + process->name.c_str();
38 SIMIX_display_process_status();
39 THROWF(arg_error, 0, "%s", msg.c_str());
41 for (auto const& arg : auto_restart_processes)
43 auto_restart_processes.clear();
44 for (auto const& arg : boot_processes)
46 boot_processes.clear();
47 xbt_swag_free(process_list);
50 /** Re-starts all the actors that are marked as restartable.
52 * Weird things will happen if you turn on an host that is already on. S4U is fool-proof, not this.
56 for (auto const& arg : boot_processes) {
57 XBT_DEBUG("Booting Process %s(%s) right now", arg->name.c_str(), arg->host->getCname());
58 smx_actor_t actor = simix_global->create_process_function(arg->name.c_str(), arg->code, nullptr, arg->host,
59 arg->properties.get(), nullptr);
60 if (arg->kill_time >= 0)
61 simcall_process_set_kill_time(actor, arg->kill_time);
62 if (arg->auto_restart)
63 actor->auto_restart = arg->auto_restart;
69 /** @brief Stop the host if it is on */
70 void SIMIX_host_off(sg_host_t h, smx_actor_t issuer)
72 simgrid::simix::Host* host = h->extension<simgrid::simix::Host>();
74 xbt_assert((host != nullptr), "Invalid parameters");
77 h->pimpl_cpu->turnOff();
79 /* Clean Simulator data */
80 if (xbt_swag_size(host->process_list) != 0) {
81 smx_actor_t process = nullptr;
82 xbt_swag_foreach(process, host->process_list) {
83 SIMIX_process_kill(process, issuer);
84 XBT_DEBUG("Killing %s@%s on behalf of %s which turned off that host.", process->cname(),
85 process->host->getCname(), issuer->cname());
89 XBT_INFO("Host %s is already off", h->getCname());
93 sg_host_t sg_host_self()
95 smx_actor_t process = SIMIX_process_self();
96 return (process == nullptr) ? nullptr : process->host;
99 /* needs to be public and without simcall for exceptions and logging events */
100 const char* sg_host_self_get_name()
102 sg_host_t host = sg_host_self();
103 if (host == nullptr || SIMIX_process_self() == simix_global->maestro_process)
106 return host->getCname();
110 * \brief Add a process to the list of the processes that the host will restart when it comes back
111 * This function add a process to the list of the processes that will be restarted when the host comes
112 * back. It is expected that this function is called when the host is down.
113 * The processes will only be restarted once, meaning that you will have to register the process
114 * again to restart the process again.
116 void SIMIX_host_add_auto_restart_process(sg_host_t host, const char* name, std::function<void()> code, void* data,
117 double kill_time, std::map<std::string, std::string>* properties,
120 smx_process_arg_t arg = new simgrid::simix::ProcessArg();
122 arg->code = std::move(code);
125 arg->kill_time = kill_time;
126 arg->properties.reset(properties, [](decltype(properties)) {});
127 arg->auto_restart = auto_restart;
129 if (host->isOff() && watched_hosts.find(host->getCname()) == watched_hosts.end()) {
130 watched_hosts.insert(host->getCname());
131 XBT_DEBUG("Push host %s to watched_hosts because state == SURF_RESOURCE_OFF", host->getCname());
133 host->extension<simgrid::simix::Host>()->auto_restart_processes.push_back(arg);
135 /** @brief Restart the list of processes that have been registered to the host */
136 void SIMIX_host_autorestart(sg_host_t host)
138 std::vector<simgrid::simix::ProcessArg*> process_list =
139 host->extension<simgrid::simix::Host>()->auto_restart_processes;
141 for (auto const& arg : process_list) {
142 XBT_DEBUG("Restarting Process %s@%s right now", arg->name.c_str(), arg->host->getCname());
143 smx_actor_t actor = simix_global->create_process_function(arg->name.c_str(), arg->code, nullptr, arg->host,
144 arg->properties.get(), nullptr);
145 if (arg->kill_time >= 0)
146 simcall_process_set_kill_time(actor, arg->kill_time);
147 if (arg->auto_restart)
148 actor->auto_restart = arg->auto_restart;
150 process_list.clear();
153 boost::intrusive_ptr<simgrid::kernel::activity::ExecImpl> simcall_HANDLER_execution_start(smx_simcall_t simcall,
156 double priority, double bound)
158 return SIMIX_execution_start(simcall->issuer, name,flops_amount,priority,bound);
161 boost::intrusive_ptr<simgrid::kernel::activity::ExecImpl>
162 SIMIX_execution_start(smx_actor_t issuer, const char* name, double flops_amount, double priority, double bound)
165 /* alloc structures and initialize */
166 simgrid::kernel::activity::ExecImplPtr exec =
167 simgrid::kernel::activity::ExecImplPtr(new simgrid::kernel::activity::ExecImpl(name, issuer->host));
169 /* set surf's action */
170 if (not MC_is_active() && not MC_record_replay_is_active()) {
172 exec->surf_exec = issuer->host->pimpl_cpu->execution_start(flops_amount);
173 exec->surf_exec->setData(exec.get());
174 exec->surf_exec->setSharingWeight(priority);
177 static_cast<simgrid::surf::CpuAction*>(exec->surf_exec)->setBound(bound);
180 XBT_DEBUG("Create execute synchro %p: %s", exec.get(), exec->name.c_str());
185 boost::intrusive_ptr<simgrid::kernel::activity::ExecImpl>
186 SIMIX_execution_parallel_start(const char* name, int host_nb, sg_host_t* host_list, double* flops_amount,
187 double* bytes_amount, double rate, double timeout)
190 /* alloc structures and initialize */
191 simgrid::kernel::activity::ExecImplPtr exec =
192 simgrid::kernel::activity::ExecImplPtr(new simgrid::kernel::activity::ExecImpl(name, nullptr));
194 /* set surf's synchro */
195 sg_host_t *host_list_cpy = xbt_new0(sg_host_t, host_nb);
196 for (int i = 0; i < host_nb; i++)
197 host_list_cpy[i] = host_list[i];
199 /* Check that we are not mixing VMs and PMs in the parallel task */
200 bool is_a_vm = (nullptr != dynamic_cast<simgrid::s4u::VirtualMachine*>(host_list[0]));
201 for (int i = 1; i < host_nb; i++) {
202 bool tmp_is_a_vm = (nullptr != dynamic_cast<simgrid::s4u::VirtualMachine*>(host_list[i]));
203 xbt_assert(is_a_vm == tmp_is_a_vm, "parallel_execute: mixing VMs and PMs is not supported (yet).");
206 /* set surf's synchro */
207 if (not MC_is_active() && not MC_record_replay_is_active()) {
208 exec->surf_exec = surf_host_model->executeParallelTask(host_nb, host_list_cpy, flops_amount, bytes_amount, rate);
209 exec->surf_exec->setData(exec.get());
211 exec->timeoutDetector = host_list[0]->pimpl_cpu->sleep(timeout);
212 exec->timeoutDetector->setData(exec.get());
215 XBT_DEBUG("Create parallel execute synchro %p", exec.get());
220 void simcall_HANDLER_execution_wait(smx_simcall_t simcall, smx_activity_t synchro)
222 simgrid::kernel::activity::ExecImplPtr exec =
223 boost::static_pointer_cast<simgrid::kernel::activity::ExecImpl>(synchro);
224 XBT_DEBUG("Wait for execution of synchro %p, state %d", synchro.get(), (int)synchro->state);
226 /* Associate this simcall to the synchro */
227 synchro->simcalls.push_back(simcall);
228 simcall->issuer->waiting_synchro = synchro;
230 /* set surf's synchro */
231 if (MC_is_active() || MC_record_replay_is_active()) {
232 synchro->state = SIMIX_DONE;
233 SIMIX_execution_finish(exec);
237 /* If the synchro is already finished then perform the error handling */
238 if (synchro->state != SIMIX_RUNNING)
239 SIMIX_execution_finish(exec);
242 void SIMIX_execution_finish(simgrid::kernel::activity::ExecImplPtr exec)
244 for (smx_simcall_t const& simcall : exec->simcalls) {
245 switch (exec->state) {
248 /* do nothing, synchro done */
249 XBT_DEBUG("SIMIX_execution_finished: execution successful");
253 XBT_DEBUG("SIMIX_execution_finished: host '%s' failed", simcall->issuer->host->getCname());
254 simcall->issuer->context->iwannadie = 1;
255 SMX_EXCEPTION(simcall->issuer, host_error, 0, "Host failed");
259 XBT_DEBUG("SIMIX_execution_finished: execution canceled");
260 SMX_EXCEPTION(simcall->issuer, cancel_error, 0, "Canceled");
264 XBT_DEBUG("SIMIX_execution_finished: execution timeouted");
265 SMX_EXCEPTION(simcall->issuer, timeout_error, 0, "Timeouted");
269 xbt_die("Internal error in SIMIX_execution_finish: unexpected synchro state %d",
272 /* Fail the process if the host is down */
273 if (simcall->issuer->host->isOff())
274 simcall->issuer->context->iwannadie = 1;
276 simcall->issuer->waiting_synchro = nullptr;
277 simcall_execution_wait__set__result(simcall, exec->state);
278 SIMIX_simcall_answer(simcall);
282 void SIMIX_set_category(smx_activity_t synchro, const char *category)
284 if (synchro->state != SIMIX_RUNNING)
287 simgrid::kernel::activity::ExecImplPtr exec =
288 boost::dynamic_pointer_cast<simgrid::kernel::activity::ExecImpl>(synchro);
289 if (exec != nullptr) {
290 exec->surf_exec->setCategory(category);
294 simgrid::kernel::activity::CommImplPtr comm =
295 boost::dynamic_pointer_cast<simgrid::kernel::activity::CommImpl>(synchro);
296 if (comm != nullptr) {
297 comm->surf_comm->setCategory(category);