X-Git-Url: http://info.iut-bm.univ-fcomte.fr/pub/gitweb/simgrid.git/blobdiff_plain/0586543235cb7550a5952d01af0d4f656670c622..1915318b8c2435908c6471242a7de6e034710355:/src/msg/msg_vm.cpp diff --git a/src/msg/msg_vm.cpp b/src/msg/msg_vm.cpp index c70ff31bd4..5c268ac091 100644 --- a/src/msg/msg_vm.cpp +++ b/src/msg/msg_vm.cpp @@ -125,7 +125,7 @@ int MSG_vm_is_restoring(msg_vm_t vm) * @param mig_netspeed Amount of Mbyte/s allocated to the migration (cannot be larger than net_cap). Use 0 if unsure. * @param dp_intensity Dirty page percentage according to migNetSpeed, [0-100]. Use 0 if unsure. */ -msg_vm_t MSG_vm_create(msg_host_t pm, const char *name, int ncpus, int ramsize, int net_cap, char *disk_path, +msg_vm_t MSG_vm_create(msg_host_t pm, const char* name, int ncpus, int ramsize, int net_cap, char* disk_path, int disksize, int mig_netspeed, int dp_intensity) { simgrid::vm::VmHostExt::ensureVmExtInstalled(); @@ -158,28 +158,12 @@ msg_vm_t MSG_vm_create(msg_host_t pm, const char *name, int ncpus, int ramsize, * * A VM is treated as a host. The name of the VM must be unique among all hosts. */ -msg_vm_t MSG_vm_create_core(msg_host_t ind_pm, const char *name) +msg_vm_t MSG_vm_create_core(msg_host_t pm, const char* name) { - /* make sure the VM of the same name does not exit */ - { - simgrid::s4u::Host* ind_host_tmp = sg_host_by_name(name); - if (ind_host_tmp != nullptr && sg_host_simix(ind_host_tmp) != nullptr) { - XBT_ERROR("host %s already exits", name); - return nullptr; - } - } - - /* Note: ind_vm and vm_workstation point to the same elm object. */ - /* Ask the SIMIX layer to create the surf vm resource */ - sg_host_t ind_vm_workstation = simcall_vm_create(name, ind_pm); - - msg_vm_t ind_vm = (msg_vm_t) __MSG_host_create(ind_vm_workstation); - - XBT_DEBUG("A new VM (%s) has been created", name); + xbt_assert(sg_host_by_name(name) == nullptr, + "Cannot create a VM named %s: this name is already used by an host or a VM", name); - TRACE_msg_vm_create(name, ind_pm); - - return ind_vm; + return new simgrid::s4u::VirtualMachine(name, pm); } /** @brief Destroy a VM. Destroy the VM object from the simulation. @@ -188,21 +172,24 @@ msg_vm_t MSG_vm_create_core(msg_host_t ind_pm, const char *name) void MSG_vm_destroy(msg_vm_t vm) { if (MSG_vm_is_migrating(vm)) - THROWF(vm_error, 0, "VM(%s) is migrating", sg_host_get_name(vm)); + THROWF(vm_error, 0, "Cannot destroy VM '%s', which is migrating.", vm->cname()); /* First, terminate all processes on the VM if necessary */ if (MSG_vm_is_running(vm)) - simcall_vm_shutdown(vm); + MSG_vm_shutdown(vm); - if (!MSG_vm_is_created(vm)) { - XBT_CRITICAL("shutdown the given VM before destroying it"); - DIE_IMPOSSIBLE; - } + xbt_assert(MSG_vm_is_created(vm), "shutdown the given VM before destroying it"); /* Then, destroy the VM object */ - simcall_vm_destroy(vm); + simgrid::simix::kernelImmediate([vm]() { + vm->destroy(); + }); - TRACE_msg_vm_end(vm); + if (TRACE_msg_vm_is_enabled()) { + container_t container = PJ_container_get(vm->name().c_str()); + PJ_container_remove_from_parent(container); + PJ_container_free(container); + } } /** @brief Start a vm (i.e., boot the guest operating system) @@ -214,7 +201,12 @@ void MSG_vm_start(msg_vm_t vm) { simcall_vm_start(vm); - TRACE_msg_vm_start(vm); + if (TRACE_msg_vm_is_enabled()) { + container_t vm_container = PJ_container_get(vm->name().c_str()); + type_t type = PJ_type_get("MSG_VM_STATE", vm_container->type); + val_t value = PJ_value_get_or_new("start", "0 0 1", type); // start is blue + new_pajePushState(MSG_get_clock(), vm_container, type, value); + } } /** @brief Immediately kills all processes within the given VM. Any memory that they allocated will be leaked. @@ -225,58 +217,26 @@ void MSG_vm_start(msg_vm_t vm) */ void MSG_vm_shutdown(msg_vm_t vm) { - /* msg_vm_t equals to msg_host_t */ simcall_vm_shutdown(vm); - MSG_process_sleep(0.); // Make sure that the processes in the VM are killed in this scheduling round before processing - // (eg with the VM destroy) - // TRACE_msg_vm_(vm); -} -/* We have two mailboxes. mbox is used to transfer migration data between source and destination PMs. mbox_ctl is used - * to detect the completion of a migration. The names of these mailboxes must not conflict with others. */ -static inline char *get_mig_mbox_src_dst(msg_vm_t vm, msg_host_t src_pm, msg_host_t dst_pm) -{ - const char *vm_name = sg_host_get_name(vm); - const char *src_pm_name = sg_host_get_name(src_pm); - const char *dst_pm_name = sg_host_get_name(dst_pm); - - return bprintf("__mbox_mig_src_dst:%s(%s-%s)", vm_name, src_pm_name, dst_pm_name); -} - -static inline char *get_mig_mbox_ctl(msg_vm_t vm, msg_host_t src_pm, msg_host_t dst_pm) -{ - const char *vm_name = sg_host_get_name(vm); - const char *src_pm_name = sg_host_get_name(src_pm); - const char *dst_pm_name = sg_host_get_name(dst_pm); - - return bprintf("__mbox_mig_ctl:%s(%s-%s)", vm_name, src_pm_name, dst_pm_name); + // Make sure that the processes in the VM are killed in this scheduling round before processing + // (eg with the VM destroy) + MSG_process_sleep(0.); } static inline char *get_mig_process_tx_name(msg_vm_t vm, msg_host_t src_pm, msg_host_t dst_pm) { - const char *vm_name = sg_host_get_name(vm); - const char *src_pm_name = sg_host_get_name(src_pm); - const char *dst_pm_name = sg_host_get_name(dst_pm); - - return bprintf("__pr_mig_tx:%s(%s-%s)", vm_name, src_pm_name, dst_pm_name); + return bprintf("__pr_mig_tx:%s(%s-%s)", vm->cname(), src_pm->cname(), dst_pm->cname()); } static inline char *get_mig_process_rx_name(msg_vm_t vm, msg_host_t src_pm, msg_host_t dst_pm) { - const char *vm_name = sg_host_get_name(vm); - const char *src_pm_name = sg_host_get_name(src_pm); - const char *dst_pm_name = sg_host_get_name(dst_pm); - - return bprintf("__pr_mig_rx:%s(%s-%s)", vm_name, src_pm_name, dst_pm_name); + return bprintf("__pr_mig_rx:%s(%s-%s)", vm->cname(), src_pm->cname(), dst_pm->cname()); } static inline char *get_mig_task_name(msg_vm_t vm, msg_host_t src_pm, msg_host_t dst_pm, int stage) { - const char *vm_name = sg_host_get_name(vm); - const char *src_pm_name = sg_host_get_name(src_pm); - const char *dst_pm_name = sg_host_get_name(dst_pm); - - return bprintf("__task_mig_stage%d:%s(%s-%s)", stage, vm_name, src_pm_name, dst_pm_name); + return bprintf("__task_mig_stage%d:%s(%s-%s)", stage, vm->cname(), src_pm->cname(), dst_pm->cname()); } struct migration_session { @@ -298,34 +258,26 @@ static int migration_rx_fun(int argc, char *argv[]) // The structure has been created in the do_migration function and should only be freed in the same place ;) struct migration_session *ms = (migration_session *) MSG_process_get_data(MSG_process_self()); - s_vm_params_t params; - static_cast(ms->vm)->parameters(¶ms); - - int need_exit = 0; + bool received_finalize = false; char *finalize_task_name = get_mig_task_name(ms->vm, ms->src_pm, ms->dst_pm, 3); - - int ret = 0; - for (;;) { + while (!received_finalize) { msg_task_t task = nullptr; - ret = MSG_task_recv(&task, ms->mbox); - { - if (ret != MSG_OK) { - // An error occurred, clean the code and return - // The owner did not change, hence the task should be only destroyed on the other side - xbt_free(finalize_task_name); - return 0; - } + int ret = MSG_task_recv(&task, ms->mbox); + + if (ret != MSG_OK) { + // An error occurred, clean the code and return + // The owner did not change, hence the task should be only destroyed on the other side + xbt_free(finalize_task_name); + return 0; } if (strcmp(task->name, finalize_task_name) == 0) - need_exit = 1; + received_finalize = 1; MSG_task_destroy(task); - - if (need_exit) - break; } + xbt_free(finalize_task_name); // Here Stage 1, 2 and 3 have been performed. // Hence complete the migration @@ -351,11 +303,33 @@ static int migration_rx_fun(int argc, char *argv[]) }); { - // Now the VM is running on the new host (the migration is completed) (even if the SRC crash) - vm->pimpl_vm_->isMigrating = false; - XBT_DEBUG("VM(%s) moved from PM(%s) to PM(%s)", sg_host_get_name(ms->vm), sg_host_get_name(ms->src_pm), - sg_host_get_name(ms->dst_pm)); - TRACE_msg_vm_change_host(ms->vm, ms->src_pm, ms->dst_pm); + // Now the VM is running on the new host (the migration is completed) (even if the SRC crash) + vm->pimpl_vm_->isMigrating = false; + XBT_DEBUG("VM(%s) moved from PM(%s) to PM(%s)", ms->vm->cname(), ms->src_pm->cname(), ms->dst_pm->cname()); + + if (TRACE_msg_vm_is_enabled()) { + static long long int counter = 0; + char key[INSTR_DEFAULT_STR_SIZE]; + snprintf(key, INSTR_DEFAULT_STR_SIZE, "%lld", counter++); + + // start link + container_t msg = PJ_container_get(vm->name().c_str()); + type_t type = PJ_type_get("MSG_VM_LINK", PJ_type_get_root()); + new_pajeStartLink(MSG_get_clock(), PJ_container_get_root(), type, msg, "M", key); + + // destroy existing container of this vm + container_t existing_container = PJ_container_get(vm->name().c_str()); + PJ_container_remove_from_parent(existing_container); + PJ_container_free(existing_container); + + // create new container on the new_host location + PJ_container_new(vm->cname(), INSTR_MSG_VM, PJ_container_get(sg_host_get_name(ms->dst_pm))); + + // end link + msg = PJ_container_get(vm->name().c_str()); + type = PJ_type_get("MSG_VM_LINK", PJ_type_get_root()); + new_pajeEndLink(MSG_get_clock(), PJ_container_get_root(), type, msg, "M", key); + } } // Inform the SRC that the migration has been correctly performed { @@ -375,21 +349,21 @@ static int migration_rx_fun(int argc, char *argv[]) xbt_free(task_name); } - xbt_free(finalize_task_name); - XBT_DEBUG("mig: rx_done"); return 0; } -static void reset_dirty_pages(msg_vm_t vm) +static void start_dirty_page_tracking(msg_vm_t vm) { - simgrid::surf::VirtualMachineImpl* pimpl = static_cast(vm->pimpl_); + simgrid::surf::VirtualMachineImpl* pimpl = static_cast(vm)->pimpl_vm_; + + pimpl->dp_enabled = 1; + if (!pimpl->dp_objs) + return; char *key = nullptr; xbt_dict_cursor_t cursor = nullptr; dirty_page_t dp = nullptr; - if (!pimpl->dp_objs) - return; xbt_dict_foreach (pimpl->dp_objs, cursor, key, dp) { double remaining = MSG_task_get_flops_amount(dp->task); dp->prev_clock = MSG_get_clock(); @@ -399,13 +373,6 @@ static void reset_dirty_pages(msg_vm_t vm) } } -static void start_dirty_page_tracking(msg_vm_t vm) -{ - static_cast(vm)->pimpl_vm_->dp_enabled = 1; - - reset_dirty_pages(vm); -} - static void stop_dirty_page_tracking(msg_vm_t vm) { static_cast(vm)->pimpl_vm_->dp_enabled = 0; @@ -460,7 +427,7 @@ void MSG_host_add_task(msg_host_t host, msg_task_t task) simgrid::s4u::VirtualMachine* vm = dynamic_cast(host); if (vm == nullptr) return; - simgrid::surf::VirtualMachineImpl* pimpl = static_cast(vm->pimpl_); + simgrid::surf::VirtualMachineImpl* pimpl = static_cast(vm)->pimpl_vm_; double remaining = MSG_task_get_flops_amount(task); char *key = bprintf("%s-%p", task->name, task); @@ -485,7 +452,7 @@ void MSG_host_del_task(msg_host_t host, msg_task_t task) simgrid::s4u::VirtualMachine* vm = dynamic_cast(host); if (vm == nullptr) return; - simgrid::surf::VirtualMachineImpl* pimpl = static_cast(vm->pimpl_); + simgrid::surf::VirtualMachineImpl* pimpl = static_cast(vm)->pimpl_vm_; char *key = bprintf("%s-%p", task->name, task); dirty_page_t dp = (dirty_page_t)(pimpl->dp_objs ? xbt_dict_get_or_null(pimpl->dp_objs, key) : NULL); @@ -509,7 +476,7 @@ void MSG_host_del_task(msg_host_t host, msg_task_t task) xbt_free(key); } -static sg_size_t send_migration_data(msg_vm_t vm, msg_host_t src_pm, msg_host_t dst_pm, sg_size_t size, char *mbox, +static sg_size_t send_migration_data(msg_vm_t vm, msg_host_t src_pm, msg_host_t dst_pm, sg_size_t size, char* mbox, int stage, int stage2_round, double mig_speed, double timeout) { sg_size_t sent = 0; @@ -552,11 +519,11 @@ static sg_size_t send_migration_data(msg_vm_t vm, msg_host_t src_pm, msg_host_t double actual_speed = size / duration; if (stage == 2) - XBT_DEBUG("mig-stage%d.%d: sent %llu duration %f actual_speed %f (target %f)", - stage, stage2_round, size, duration, actual_speed, mig_speed); + XBT_DEBUG("mig-stage%d.%d: sent %llu duration %f actual_speed %f (target %f)", stage, stage2_round, size, duration, + actual_speed, mig_speed); else - XBT_DEBUG("mig-stage%d: sent %llu duration %f actual_speed %f (target %f)", - stage, size, duration, actual_speed, mig_speed); + XBT_DEBUG("mig-stage%d: sent %llu duration %f actual_speed %f (target %f)", stage, size, duration, actual_speed, + mig_speed); return sent; } @@ -574,7 +541,7 @@ static sg_size_t get_updated_size(double computed, double dp_rate, double dp_cap return (sg_size_t) updated_size; } -static double send_stage1(struct migration_session *ms, sg_size_t ramsize, double mig_speed, double dp_rate, +static double send_stage1(struct migration_session* ms, sg_size_t ramsize, double mig_speed, double dp_rate, double dp_cap) { // const long chunksize = (sg_size_t)1024 * 1024 * 100; @@ -681,70 +648,70 @@ static int migration_tx_fun(int argc, char *argv[]) /* Stage2: send update pages iteratively until the size of remaining states becomes smaller than threshold value. */ - if (! skip_stage2) { - - int stage2_round = 0; - for (;;) { - - sg_size_t updated_size = 0; - if (stage2_round == 0) { - /* just after stage1, nothing has been updated. But, we have to send the data updated during stage1 */ - updated_size = get_updated_size(computed_during_stage1, dp_rate, dp_cap); - } else { - double computed = lookup_computed_flop_counts(ms->vm, 2, stage2_round); - updated_size = get_updated_size(computed, dp_rate, dp_cap); - } - - XBT_DEBUG("mig-stage 2:%d updated_size %llu computed_during_stage1 %f dp_rate %f dp_cap %f", - stage2_round, updated_size, computed_during_stage1, dp_rate, dp_cap); - - /* Check whether the remaining size is below the threshold value. If so, move to stage 3. */ - remaining_size += updated_size; - XBT_DEBUG("mig-stage2.%d: remaining_size %f (%s threshold %f)", stage2_round, - remaining_size, (remaining_size < threshold) ? "<" : ">", threshold); - if (remaining_size < threshold) - break; - - sg_size_t sent = 0; - double clock_prev_send = MSG_get_clock(); - try { - XBT_DEBUG("Stage 2, gonna send %llu", updated_size); - sent = send_migration_data(ms->vm, ms->src_pm, ms->dst_pm, updated_size, ms->mbox, 2, stage2_round, mig_speed, - mig_timeout); - } - catch (xbt_ex& e) { - //hostfailure (if you want to know whether this is the SRC or the DST check directly in send_migration_data code) - // Stop the dirty page tracking an return (there is no memory space to release) - stop_dirty_page_tracking(ms->vm); - return 0; - } - double clock_post_send = MSG_get_clock(); + if (!skip_stage2) { + + int stage2_round = 0; + for (;;) { + + sg_size_t updated_size = 0; + if (stage2_round == 0) { + /* just after stage1, nothing has been updated. But, we have to send the data updated during stage1 */ + updated_size = get_updated_size(computed_during_stage1, dp_rate, dp_cap); + } else { + double computed = lookup_computed_flop_counts(ms->vm, 2, stage2_round); + updated_size = get_updated_size(computed, dp_rate, dp_cap); + } - if (sent == updated_size) { - /* timeout did not happen */ - double bandwidth = updated_size / (clock_post_send - clock_prev_send); - threshold = get_threshold_value(bandwidth, max_downtime); - XBT_DEBUG("actual bandwidth %f, threshold %f", bandwidth / 1024 / 1024, threshold); - remaining_size -= sent; - stage2_round += 1; - mig_timeout -= (clock_post_send - clock_prev_send); - xbt_assert(mig_timeout > 0); - - } else if (sent < updated_size) { - /* When timeout happens, we move to stage 3. The size of memory pages - * updated before timeout must be added to the remaining size. */ - XBT_VERB("mig-stage2.%d: timeout, force moving to stage 3. sent %llu / %llu, eta %lf", - stage2_round, sent, updated_size, (clock_post_send - clock_prev_send)); - remaining_size -= sent; + XBT_DEBUG("mig-stage 2:%d updated_size %llu computed_during_stage1 %f dp_rate %f dp_cap %f", stage2_round, + updated_size, computed_during_stage1, dp_rate, dp_cap); - double computed = lookup_computed_flop_counts(ms->vm, 2, stage2_round); - updated_size = get_updated_size(computed, dp_rate, dp_cap); + /* Check whether the remaining size is below the threshold value. If so, move to stage 3. */ remaining_size += updated_size; - break; - } else - XBT_CRITICAL("bug"); + XBT_DEBUG("mig-stage2.%d: remaining_size %f (%s threshold %f)", stage2_round, remaining_size, + (remaining_size < threshold) ? "<" : ">", threshold); + if (remaining_size < threshold) + break; + + sg_size_t sent = 0; + double clock_prev_send = MSG_get_clock(); + try { + XBT_DEBUG("Stage 2, gonna send %llu", updated_size); + sent = send_migration_data(ms->vm, ms->src_pm, ms->dst_pm, updated_size, ms->mbox, 2, stage2_round, mig_speed, + mig_timeout); + } catch (xbt_ex& e) { + // hostfailure (if you want to know whether this is the SRC or the DST check directly in send_migration_data + // code) + // Stop the dirty page tracking an return (there is no memory space to release) + stop_dirty_page_tracking(ms->vm); + return 0; + } + double clock_post_send = MSG_get_clock(); + + if (sent == updated_size) { + /* timeout did not happen */ + double bandwidth = updated_size / (clock_post_send - clock_prev_send); + threshold = get_threshold_value(bandwidth, max_downtime); + XBT_DEBUG("actual bandwidth %f, threshold %f", bandwidth / 1024 / 1024, threshold); + remaining_size -= sent; + stage2_round += 1; + mig_timeout -= (clock_post_send - clock_prev_send); + xbt_assert(mig_timeout > 0); + + } else if (sent < updated_size) { + /* When timeout happens, we move to stage 3. The size of memory pages + * updated before timeout must be added to the remaining size. */ + XBT_VERB("mig-stage2.%d: timeout, force moving to stage 3. sent %llu / %llu, eta %lf", stage2_round, sent, + updated_size, (clock_post_send - clock_prev_send)); + remaining_size -= sent; + + double computed = lookup_computed_flop_counts(ms->vm, 2, stage2_round); + updated_size = get_updated_size(computed, dp_rate, dp_cap); + remaining_size += updated_size; + break; + } else + XBT_CRITICAL("bug"); + } } - } /* Stage3: stop the VM and copy the rest of states. */ XBT_DEBUG("mig-stage3: remaining_size %f", remaining_size); @@ -775,63 +742,54 @@ static int do_migration(msg_vm_t vm, msg_host_t src_pm, msg_host_t dst_pm) ms->vm = vm; ms->src_pm = src_pm; ms->dst_pm = dst_pm; - ms->mbox_ctl = get_mig_mbox_ctl(vm, src_pm, dst_pm); - ms->mbox = get_mig_mbox_src_dst(vm, src_pm, dst_pm); + + /* We have two mailboxes. mbox is used to transfer migration data between source and destination PMs. mbox_ctl is used + * to detect the completion of a migration. The names of these mailboxes must not conflict with others. */ + ms->mbox_ctl = bprintf("__mbox_mig_ctl:%s(%s-%s)", vm->cname(), src_pm->cname(), dst_pm->cname()); + ms->mbox = bprintf("__mbox_mig_src_dst:%s(%s-%s)", vm->cname(), src_pm->cname(), dst_pm->cname()); char *pr_rx_name = get_mig_process_rx_name(vm, src_pm, dst_pm); char *pr_tx_name = get_mig_process_tx_name(vm, src_pm, dst_pm); -// msg_process_t tx_process, rx_process; -// MSG_process_create(pr_rx_name, migration_rx_fun, ms, dst_pm); -// MSG_process_create(pr_tx_name, migration_tx_fun, ms, src_pm); -#if 1 - { - char **argv = xbt_new(char *, 2); - argv[0] = pr_rx_name; - argv[1] = nullptr; -/*rx_process = */ MSG_process_create_with_arguments(pr_rx_name, migration_rx_fun, ms, dst_pm, 1, argv); - } - { - char **argv = xbt_new(char *, 2); - argv[0] = pr_tx_name; - argv[1] = nullptr; -/* tx_process = */MSG_process_create_with_arguments(pr_tx_name, migration_tx_fun, ms, src_pm, 1, argv); - } -#endif - - /* wait until the migration have finished or on error has occurred */ - { - XBT_DEBUG("wait for reception of the final ACK (i.e. migration has been correctly performed"); - msg_task_t task = nullptr; - msg_error_t ret = MSG_TIMEOUT; - while (ret == MSG_TIMEOUT && MSG_host_is_on(dst_pm)) //Wait while you receive the message o - ret = MSG_task_receive_with_timeout(&task, ms->mbox_ctl, 4); - - xbt_free(ms->mbox_ctl); - xbt_free(ms->mbox); - xbt_free(ms); + char** argv = xbt_new(char*, 2); + argv[0] = pr_rx_name; + argv[1] = nullptr; + MSG_process_create_with_arguments(pr_rx_name, migration_rx_fun, ms, dst_pm, 1, argv); - //xbt_assert(ret == MSG_OK); - if(ret == MSG_HOST_FAILURE){ - // Note that since the communication failed, the owner did not change and the task should be destroyed on the - // other side. Hence, just throw the execption - XBT_ERROR("SRC crashes, throw an exception (m-control)"); - //MSG_process_kill(tx_process); // Adrien, I made a merge on Nov 28th 2014, I'm not sure whether this line is - // required or not - return -1; - } - else if((ret == MSG_TRANSFER_FAILURE) || (ret == MSG_TIMEOUT)){ - // MSG_TIMEOUT here means that MSG_host_is_avail() returned false. - XBT_ERROR("DST crashes, throw an exception (m-control)"); - return -2; - } + argv = xbt_new(char*, 2); + argv[0] = pr_tx_name; + argv[1] = nullptr; + MSG_process_create_with_arguments(pr_tx_name, migration_tx_fun, ms, src_pm, 1, argv); - char *expected_task_name = get_mig_task_name(vm, src_pm, dst_pm, 4); - xbt_assert(strcmp(task->name, expected_task_name) == 0); - xbt_free(expected_task_name); - MSG_task_destroy(task); - return 0; + /* wait until the migration have finished or on error has occurred */ + XBT_DEBUG("wait for reception of the final ACK (i.e. migration has been correctly performed"); + msg_task_t task = nullptr; + msg_error_t ret = MSG_TIMEOUT; + while (ret == MSG_TIMEOUT && MSG_host_is_on(dst_pm)) // Wait while you receive the message o + ret = MSG_task_receive_with_timeout(&task, ms->mbox_ctl, 4); + + xbt_free(ms->mbox_ctl); + xbt_free(ms->mbox); + xbt_free(ms); + + if (ret == MSG_HOST_FAILURE) { + // Note that since the communication failed, the owner did not change and the task should be destroyed on the + // other side. Hence, just throw the execption + XBT_ERROR("SRC crashes, throw an exception (m-control)"); + // MSG_process_kill(tx_process); // Adrien, I made a merge on Nov 28th 2014, I'm not sure whether this line is + // required or not + return -1; + } else if ((ret == MSG_TRANSFER_FAILURE) || (ret == MSG_TIMEOUT)) { + // MSG_TIMEOUT here means that MSG_host_is_avail() returned false. + XBT_ERROR("DST crashes, throw an exception (m-control)"); + return -2; } + + char* expected_task_name = get_mig_task_name(vm, src_pm, dst_pm, 4); + xbt_assert(strcmp(task->name, expected_task_name) == 0); + xbt_free(expected_task_name); + MSG_task_destroy(task); + return 0; } /** @brief Migrate the VM to the given host. @@ -858,39 +816,33 @@ void MSG_vm_migrate(msg_vm_t vm, msg_host_t new_pm) * The second one would be easier. */ - msg_host_t old_pm = (msg_host_t) simcall_vm_get_pm(vm); + simgrid::s4u::VirtualMachine* typedVm = static_cast(vm); + simgrid::surf::VirtualMachineImpl* pimpl = typedVm->pimpl_vm_; + msg_host_t old_pm = pimpl->getPm(); - if(MSG_host_is_off(old_pm)) - THROWF(vm_error, 0, "SRC host(%s) seems off, cannot start a migration", sg_host_get_name(old_pm)); + if (old_pm->isOff()) + THROWF(vm_error, 0, "Cannot migrate VM '%s' from host '%s', which is offline.", vm->cname(), old_pm->cname()); - if(MSG_host_is_off(new_pm)) - THROWF(vm_error, 0, "DST host(%s) seems off, cannot start a migration", sg_host_get_name(new_pm)); + if (new_pm->isOff()) + THROWF(vm_error, 0, "Cannot migrate VM '%s' to host '%s', which is offline.", vm->cname(), new_pm->cname()); if (!MSG_vm_is_running(vm)) - THROWF(vm_error, 0, "VM(%s) is not running", sg_host_get_name(vm)); + THROWF(vm_error, 0, "Cannot migrate VM '%s' that is not running yet.", vm->cname()); - if (MSG_vm_is_migrating(vm)) - THROWF(vm_error, 0, "VM(%s) is already migrating", sg_host_get_name(vm)); + if (typedVm->isMigrating()) + THROWF(vm_error, 0, "Cannot migrate VM '%s' that is already migrating.", vm->cname()); - simgrid::surf::VirtualMachineImpl* pimpl = static_cast(vm->pimpl_); - pimpl->isMigrating = 1; + pimpl->isMigrating = true; - { - int ret = do_migration(vm, old_pm, new_pm); - if (ret == -1){ - pimpl->isMigrating = 0; - THROWF(host_error, 0, "SRC host failed during migration"); - } - else if(ret == -2){ - pimpl->isMigrating = 0; - THROWF(host_error, 0, "DST host failed during migration"); - } + int ret = do_migration(vm, old_pm, new_pm); + if (ret == -1) { + pimpl->isMigrating = false; + THROWF(host_error, 0, "Source host '%s' failed during the migration of VM '%s'.", old_pm->cname(), vm->cname()); + } else if (ret == -2) { + pimpl->isMigrating = false; + THROWF(host_error, 0, "Destination host '%s' failed during the migration of VM '%s'.", new_pm->cname(), + vm->cname()); } - - // This part is done in the RX code, to handle the corner case where SRC can crash just at the end of the migration - // process. In that case, the VM has been already assigned to the DST node. - //XBT_DEBUG("VM(%s) moved from PM(%s) to PM(%s)", vm->key, old_pm->key, new_pm->key); - //TRACE_msg_vm_change_host(vm, old_pm, new_pm); } /** @brief Immediately suspend the execution of all processes within the given VM. @@ -904,13 +856,18 @@ void MSG_vm_migrate(msg_vm_t vm, msg_host_t new_pm) void MSG_vm_suspend(msg_vm_t vm) { if (MSG_vm_is_migrating(vm)) - THROWF(vm_error, 0, "VM(%s) is migrating", sg_host_get_name(vm)); + THROWF(vm_error, 0, "Cannot suspend VM '%s', which is migrating", vm->cname()); simcall_vm_suspend(vm); XBT_DEBUG("vm_suspend done"); - TRACE_msg_vm_suspend(vm); + if (TRACE_msg_vm_is_enabled()) { + container_t vm_container = PJ_container_get(vm->cname()); + type_t type = PJ_type_get("MSG_VM_STATE", vm_container->type); + val_t value = PJ_value_get_or_new("suspend", "1 0 0", type); // suspend is red + new_pajePushState(MSG_get_clock(), vm_container, type, value); + } } /** @brief Resume the execution of the VM. All processes on the VM run again. @@ -922,7 +879,11 @@ void MSG_vm_resume(msg_vm_t vm) { simcall_vm_resume(vm); - TRACE_msg_vm_resume(vm); + if (TRACE_msg_vm_is_enabled()) { + container_t vm_container = PJ_container_get(vm->cname()); + type_t type = PJ_type_get("MSG_VM_STATE", vm_container->type); + new_pajePopState(MSG_get_clock(), vm_container, type); + } } @@ -938,10 +899,16 @@ void MSG_vm_resume(msg_vm_t vm) void MSG_vm_save(msg_vm_t vm) { if (MSG_vm_is_migrating(vm)) - THROWF(vm_error, 0, "VM(%s) is migrating", sg_host_get_name(vm)); + THROWF(vm_error, 0, "Cannot save VM '%s', which is migrating.", vm->cname()); simcall_vm_save(vm); - TRACE_msg_vm_save(vm); + + if (TRACE_msg_vm_is_enabled()) { + container_t vm_container = PJ_container_get(vm->cname()); + type_t type = PJ_type_get("MSG_VM_STATE", vm_container->type); + val_t value = PJ_value_get_or_new("save", "0 1 0", type); // save is green + new_pajePushState(MSG_get_clock(), vm_container, type, value); + } } /** @brief Restore the execution of the VM. All processes on the VM run again. @@ -954,7 +921,11 @@ void MSG_vm_restore(msg_vm_t vm) { simcall_vm_restore(vm); - TRACE_msg_vm_restore(vm); + if (TRACE_msg_vm_is_enabled()) { + container_t vm_container = PJ_container_get(vm->cname()); + type_t type = PJ_type_get("MSG_VM_STATE", vm_container->type); + new_pajePopState(MSG_get_clock(), vm_container, type); + } } /** @brief Get the physical host of a given VM. @@ -962,7 +933,7 @@ void MSG_vm_restore(msg_vm_t vm) */ msg_host_t MSG_vm_get_pm(msg_vm_t vm) { - return (msg_host_t) simcall_vm_get_pm(vm); + return static_cast(vm)->pimpl_vm_->getPm(); } /** @brief Set a CPU bound for a given VM.