X-Git-Url: http://info.iut-bm.univ-fcomte.fr/pub/gitweb/simgrid.git/blobdiff_plain/eb5b6b0c3c4d9a38bb205b2c8bc9aeeba8674a25..4aa1cd22325b98b31547d515f870d99ba2057d5f:/src/msg/msg_vm.c diff --git a/src/msg/msg_vm.c b/src/msg/msg_vm.c index 7972de267b..64c3a49680 100644 --- a/src/msg/msg_vm.c +++ b/src/msg/msg_vm.c @@ -4,19 +4,13 @@ /* This program is free software; you can redistribute it and/or modify it * under the terms of the license (GNU LGPL) which comes with this package. */ -// QUESTIONS: -// 1./ check how and where a new VM is added to the list of the hosts -// 2./ Diff between SIMIX_Actions and SURF_Actions -// => SIMIX_actions : point synchro entre processus de niveau (theoretically speaking I do not have to create such SIMIX_ACTION -// => Surf_Actions - -// TODO -// MSG_TRACE can be revisited in order to use the host -// To implement a mixed model between workstation and vm_workstation, -// please give a look at surf_model_private_t model_private at SURF Level and to the share resource functions -// double (*share_resources) (double now); -// For the action into the vm workstation model, we should be able to leverage the usual one (and if needed, look at -// the workstation model. +/* TODO: + * 1. add the support of trace + * 2. use parallel tasks to simulate CPU overhead and remove the very + * experimental code generating micro computation tasks + */ + + #include "msg_private.h" #include "xbt/sysdep.h" @@ -103,7 +97,7 @@ static inline int __MSG_vm_is_state(msg_vm_t vm, e_surf_vm_state_t state) return simcall_vm_get_state(vm) == state; } -/** @brief Returns whether the given VM has just reated, not running. +/** @brief Returns whether the given VM has just created, not running. * @ingroup msg_VMs */ int MSG_vm_is_created(msg_vm_t vm) @@ -124,7 +118,8 @@ int MSG_vm_is_running(msg_vm_t vm) */ int MSG_vm_is_migrating(msg_vm_t vm) { - return __MSG_vm_is_state(vm, SURF_VM_STATE_MIGRATING); + msg_host_priv_t priv = msg_host_resource_priv(vm); + return priv->is_migrating; } /** @brief Returns whether the given VM is currently suspended, not running. @@ -180,7 +175,7 @@ msg_vm_t MSG_vm_create(msg_host_t ind_pm, const char *name, * bandwidth */ double host_speed = MSG_get_host_speed(ind_pm); double update_speed = ((double)dp_intensity/100) * mig_netspeed; - + msg_vm_t vm = MSG_vm_create_core(ind_pm, name); s_ws_params_t params; memset(¶ms, 0, sizeof(params)); @@ -238,6 +233,9 @@ 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_name(vm)); + /* First, terminate all processes on the VM if necessary */ if (MSG_vm_is_running(vm)) simcall_vm_shutdown(vm); @@ -294,67 +292,101 @@ void MSG_vm_shutdown(msg_vm_t vm) /* We have two mailboxes. mbox is used to transfer migration data between - * source and destiantion PMs. mbox_ctl is used to detect the completion of a + * 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(const char *vm_name, const char *src_pm_name, const char *dst_pm_name) +static inline char *get_mig_mbox_src_dst(msg_vm_t vm, msg_host_t src_pm, msg_host_t dst_pm) { + char *vm_name = sg_host_name(vm); + char *src_pm_name = sg_host_name(src_pm); + char *dst_pm_name = sg_host_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(const char *vm_name, const char *src_pm_name, const char *dst_pm_name) +static inline char *get_mig_mbox_ctl(msg_vm_t vm, msg_host_t src_pm, msg_host_t dst_pm) { + char *vm_name = sg_host_name(vm); + char *src_pm_name = sg_host_name(src_pm); + char *dst_pm_name = sg_host_name(dst_pm); + return bprintf("__mbox_mig_ctl:%s(%s-%s)", vm_name, src_pm_name, dst_pm_name); } -static inline char *get_mig_process_tx_name(const char *vm_name, const char *src_pm_name, const char *dst_pm_name) +static inline char *get_mig_process_tx_name(msg_vm_t vm, msg_host_t src_pm, msg_host_t dst_pm) { + char *vm_name = sg_host_name(vm); + char *src_pm_name = sg_host_name(src_pm); + char *dst_pm_name = sg_host_name(dst_pm); + return bprintf("__pr_mig_tx:%s(%s-%s)", vm_name, src_pm_name, dst_pm_name); } -static inline char *get_mig_process_rx_name(const char *vm_name, const char *src_pm_name, const char *dst_pm_name) +static inline char *get_mig_process_rx_name(msg_vm_t vm, msg_host_t src_pm, msg_host_t dst_pm) { + char *vm_name = sg_host_name(vm); + char *src_pm_name = sg_host_name(src_pm); + char *dst_pm_name = sg_host_name(dst_pm); + return bprintf("__pr_mig_rx:%s(%s-%s)", vm_name, src_pm_name, dst_pm_name); } -static inline char *get_mig_task_name(const char *vm_name, const char *src_pm_name, const char *dst_pm_name, int stage) +static inline char *get_mig_task_name(msg_vm_t vm, msg_host_t src_pm, msg_host_t dst_pm, int stage) { + char *vm_name = sg_host_name(vm); + char *src_pm_name = sg_host_name(src_pm); + char *dst_pm_name = sg_host_name(dst_pm); + return bprintf("__task_mig_stage%d:%s(%s-%s)", stage, vm_name, src_pm_name, dst_pm_name); } static void launch_deferred_exec_process(msg_host_t host, double computation, double prio); + +struct migration_session { + msg_vm_t vm; + msg_host_t src_pm; + msg_host_t dst_pm; + + /* The miration_rx process uses mbox_ctl to let the caller of do_migration() + * know the completion of the migration. */ + char *mbox_ctl; + /* The migration_rx and migration_tx processes use mbox to transfer migration + * data. */ + char *mbox; +}; + + static int migration_rx_fun(int argc, char *argv[]) { XBT_DEBUG("mig: rx_start"); - xbt_assert(argc == 4); - const char *vm_name = argv[1]; - const char *src_pm_name = argv[2]; - const char *dst_pm_name = argv[3]; - msg_vm_t vm = MSG_get_host_by_name(vm_name); - msg_host_t src_pm = MSG_get_host_by_name(src_pm_name); - msg_host_t dst_pm = MSG_get_host_by_name(dst_pm_name); - + // The structure has been created in the do_migration function and should only be freed in the same place ;) + struct migration_session *ms = MSG_process_get_data(MSG_process_self()); s_ws_params_t params; - simcall_host_get_params(vm, ¶ms); + simcall_host_get_params(ms->vm, ¶ms); const double xfer_cpu_overhead = params.xfer_cpu_overhead; - int need_exit = 0; - char *mbox = get_mig_mbox_src_dst(vm_name, src_pm_name, dst_pm_name); - char *mbox_ctl = get_mig_mbox_ctl(vm_name, src_pm_name, dst_pm_name); - char *finalize_task_name = get_mig_task_name(vm_name, src_pm_name, dst_pm_name, 3); + char *finalize_task_name = get_mig_task_name(ms->vm, ms->src_pm, ms->dst_pm, 3); for (;;) { msg_task_t task = NULL; - MSG_task_recv(&task, mbox); + MSG_task_recv(&task, ms->mbox); { - double received = MSG_task_get_data_size(task); + double received ; + if (task) + received = MSG_task_get_data_size(task); + else{ + // An error occured, 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; + } /* TODO: clean up */ // const double alpha = 0.22L * 1.0E8 / (80L * 1024 * 1024); - launch_deferred_exec_process(vm, received * xfer_cpu_overhead, 1); + launch_deferred_exec_process(ms->vm, received * xfer_cpu_overhead, 1); } if (strcmp(task->name, finalize_task_name) == 0) @@ -366,35 +398,43 @@ static int migration_rx_fun(int argc, char *argv[]) break; } + // Here Stage 1, 2 and 3 have been performed. + // Hence complete the migration - /* deinstall the current affinity setting */ - simcall_vm_set_affinity(vm, src_pm, 0); +// TODO: we have an issue, if the DST node is turning off during the three next calls, then the VM is in an inconsistent state +// I should check with Takahiro in order to make this portion of code atomic + /* deinstall the current affinity setting for the CPU */ + simcall_vm_set_affinity(ms->vm, ms->src_pm, 0); - simcall_vm_migrate(vm, dst_pm); - simcall_vm_resume(vm); + /* Update the vm location */ + simcall_vm_migrate(ms->vm, ms->dst_pm); + + /* Resume the VM */ + simcall_vm_resume(ms->vm); /* install the affinity setting of the VM on the destination pm */ { - msg_host_priv_t priv = msg_host_resource_priv(vm); + msg_host_priv_t priv = msg_host_resource_priv(ms->vm); - unsigned long affinity_mask = (unsigned long) xbt_dict_get_or_null_ext(priv->affinity_mask_db, (char *) dst_pm, sizeof(msg_host_t)); - simcall_vm_set_affinity(vm, dst_pm, affinity_mask); - XBT_INFO("set affinity(0x%04lx@%s) for %s", affinity_mask, MSG_host_get_name(dst_pm), MSG_host_get_name(vm)); + unsigned long affinity_mask = (unsigned long) xbt_dict_get_or_null_ext(priv->affinity_mask_db, (char *) ms->dst_pm, sizeof(msg_host_t)); + simcall_vm_set_affinity(ms->vm, ms->dst_pm, affinity_mask); + XBT_INFO("set affinity(0x%04lx@%s) for %s", affinity_mask, MSG_host_get_name(ms->dst_pm), MSG_host_get_name(ms->vm)); } { - char *task_name = get_mig_task_name(vm_name, src_pm_name, dst_pm_name, 4); + char *task_name = get_mig_task_name(ms->vm, ms->src_pm, ms->dst_pm, 4); msg_task_t task = MSG_task_create(task_name, 0, 0, NULL); - msg_error_t ret = MSG_task_send(task, mbox_ctl); - xbt_assert(ret == MSG_OK); - + msg_error_t ret = MSG_task_send(task, ms->mbox_ctl); + // xbt_assert(ret == MSG_OK); + if(ret == MSG_HOST_FAILURE){ + // The SRC has crashed, this is not a problem has the VM has been correctly migrated on the DST node + MSG_task_destroy(task); + } xbt_free(task_name); } - xbt_free(mbox); - xbt_free(mbox_ctl); xbt_free(finalize_task_name); XBT_DEBUG("mig: rx_done"); @@ -456,7 +496,7 @@ static double get_computed(char *key, msg_vm_t vm, dirty_page_t dp, double remai double computed = dp->prev_remaining - remaining; double duration = clock - dp->prev_clock; - XBT_DEBUG("%s@%s: computated %f ops (remaining %f -> %f) in %f secs (%f -> %f)", + XBT_DEBUG("%s@%s: computed %f ops (remaining %f -> %f) in %f secs (%f -> %f)", key, sg_host_name(vm), computed, dp->prev_remaining, remaining, duration, dp->prev_clock, clock); return computed; @@ -472,7 +512,7 @@ static double lookup_computed_flop_counts(msg_vm_t vm, int stage_for_fancy_debug dirty_page_t dp = NULL; xbt_dict_foreach(priv->dp_objs, cursor, key, dp) { double remaining = MSG_task_get_remaining_computation(dp->task); - + double clock = MSG_get_clock(); // total += calc_updated_pages(key, vm, dp, remaining, clock); @@ -531,7 +571,7 @@ void MSG_host_del_task(msg_host_t host, msg_task_t task) xbt_assert(dp->task == task); /* If we are in the middle of dirty page tracking, we record how much - * computaion has been done until now, and keep the information for the + * computation has been done until now, and keep the information for the * lookup_() function that will called soon. */ if (priv->dp_enabled) { double remaining = MSG_task_get_remaining_computation(task); @@ -560,7 +600,7 @@ static int deferred_exec_fun(int argc, char *argv[]) double prio = atof(prio_str); msg_task_t task = MSG_task_create("__task_deferred", computaion, 0, NULL); - // XBT_INFO("exec deferred %f", computaion); + // XBT_INFO("exec deferred %f", computation); /* dpt is the results of the VM activity */ MSG_task_set_priority(task, prio); @@ -643,7 +683,8 @@ static void shutdown_overhead_process(msg_task_t comm_task) // XBT_INFO("micro shutdown: mbox %s", mbox); msg_error_t ret = MSG_task_send(task, mbox); - xbt_assert(ret == MSG_OK); + if(ret != MSG_OK) + xbt_die("shutdown error - task not sent"); xbt_free(mbox); // XBT_INFO("shutdown done"); @@ -657,7 +698,8 @@ static void request_overhead(msg_task_t comm_task, double computation) // XBT_INFO("req overhead"); msg_error_t ret = MSG_task_send(task, mbox); - xbt_assert(ret == MSG_OK); + if(ret != MSG_OK) + xbt_die("req overhead error - task not sent"); xbt_free(mbox); } @@ -698,7 +740,8 @@ static void task_send_bounded_with_cpu_overhead(msg_task_t comm_task, char *mbox request_overhead(comm_task, data_size * alpha); msg_error_t ret = MSG_task_send(mtask, mbox); - xbt_assert(ret == MSG_OK); + if(ret != MSG_OK) + xbt_die("migration error - task not sent"); xbt_free(mtask_name); } @@ -795,10 +838,10 @@ const double alpha = 0.22L * 1.0E8 / (80L * 1024 * 1024); #endif -static void send_migration_data(const char *vm_name, const char *src_pm_name, const char *dst_pm_name, +static void 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 xfer_cpu_overhead) { - char *task_name = get_mig_task_name(vm_name, src_pm_name, dst_pm_name, stage); + char *task_name = get_mig_task_name(vm, src_pm, dst_pm, stage); msg_task_t task = MSG_task_create(task_name, 0, size, NULL); /* TODO: clean up */ @@ -815,7 +858,15 @@ static void send_migration_data(const char *vm_name, const char *src_pm_name, co ret = MSG_task_send_bounded(task, mbox, mig_speed); else ret = MSG_task_send(task, mbox); - xbt_assert(ret == MSG_OK); +// xbt_assert(ret == MSG_OK); + xbt_free(task_name); + if(ret == MSG_HOST_FAILURE){ + THROWF(host_error, 0, "host failed during migration of %s (stage %d)", sg_host_name(vm), stage); + //XBT_INFO("host failed during migration of %s (stage %d)", sg_host_name(vm), stage); + MSG_task_destroy(task); + return; + + } #endif double clock_end = MSG_get_clock(); @@ -827,18 +878,12 @@ static void send_migration_data(const char *vm_name, const char *src_pm_name, co double cpu_utilization = 0; #endif - - - if (stage == 2){ XBT_DEBUG("mig-stage%d.%d: sent %llu duration %f actual_speed %f (target %f) cpu %f", stage, stage2_round, size, duration, actual_speed, mig_speed, cpu_utilization);} else{ XBT_DEBUG("mig-stage%d: sent %llu duration %f actual_speed %f (target %f) cpu %f", stage, size, duration, actual_speed, mig_speed, cpu_utilization); } - xbt_free(task_name); - - #ifdef USE_MICRO_TASK /* The name of a micro task starts with __micro, which does not match the @@ -849,11 +894,19 @@ static void send_migration_data(const char *vm_name, const char *src_pm_name, co char *task_name = get_mig_task_name(vm_name, src_pm_name, dst_pm_name, stage); msg_task_t task = MSG_task_create(task_name, 0, 0, NULL); msg_error_t ret = MSG_task_send(task, mbox); - xbt_assert(ret == MSG_OK); +// xbt_assert(ret == MSG_OK); xbt_free(task_name); + if(ret == MSG_HOST_FAILURE){ + THROWF(host_error, 0, "host failed during migration of VM %s (stage 3)", sg_host_name(vm)); + //XBT_INFO("host failed during migration of %s (stage 3)", sg_host_name(vm)); + // The owner of the task did not change so destroy the task + MSG_task_destroy(task); + return; + } } } #endif + } static double get_updated_size(double computed, double dp_rate, double dp_cap) @@ -868,11 +921,9 @@ static double get_updated_size(double computed, double dp_rate, double dp_cap) return updated_size; } -static double send_stage1(msg_host_t vm, const char *src_pm_name, const char *dst_pm_name, +static double send_stage1(struct migration_session *ms, sg_size_t ramsize, double mig_speed, double xfer_cpu_overhead, double dp_rate, double dp_cap, double dpt_cpu_overhead) { - const char *vm_name = MSG_host_get_name(vm); - char *mbox = get_mig_mbox_src_dst(vm_name, src_pm_name, dst_pm_name); // const long chunksize = (sg_size_t)1024 * 1024 * 100; const sg_size_t chunksize = (sg_size_t)1024 * 1024 * 100000; @@ -886,8 +937,8 @@ static double send_stage1(msg_host_t vm, const char *src_pm_name, const char *ds remaining -= datasize; - send_migration_data(vm_name, src_pm_name, dst_pm_name, datasize, mbox, 1, 0, mig_speed, xfer_cpu_overhead); - double computed = lookup_computed_flop_counts(vm, 1, 0); + send_migration_data(ms->vm, ms->src_pm, ms->dst_pm, datasize, ms->mbox, 1, 0, mig_speed, xfer_cpu_overhead); + double computed = lookup_computed_flop_counts(ms->vm, 1, 0); computed_total += computed; // { @@ -897,7 +948,7 @@ static double send_stage1(msg_host_t vm, const char *src_pm_name, const char *ds // launch_deferred_exec_process(vm, overhead, 10000); // } } - xbt_free(mbox); + return computed_total; } @@ -916,15 +967,11 @@ static int migration_tx_fun(int argc, char *argv[]) { XBT_DEBUG("mig: tx_start"); - xbt_assert(argc == 4); - const char *vm_name = argv[1]; - const char *src_pm_name = argv[2]; - const char *dst_pm_name = argv[3]; - msg_vm_t vm = MSG_get_host_by_name(vm_name); - + // Note that the ms structure has been allocated in do_migration and hence should be freed in the same function ;) + struct migration_session *ms = MSG_process_get_data(MSG_process_self()); s_ws_params_t params; - simcall_host_get_params(vm, ¶ms); + simcall_host_get_params(ms->vm, ¶ms); const sg_size_t ramsize = params.ramsize; const sg_size_t devsize = params.devsize; const int skip_stage1 = params.skip_stage1; @@ -951,12 +998,11 @@ static int migration_tx_fun(int argc, char *argv[]) if (ramsize == 0) XBT_WARN("migrate a VM, but ramsize is zero"); - char *mbox = get_mig_mbox_src_dst(vm_name, src_pm_name, dst_pm_name); XBT_INFO("mig-stage1: remaining_size %f", remaining_size); /* Stage1: send all memory pages to the destination. */ - start_dirty_page_tracking(vm); + start_dirty_page_tracking(ms->vm); double computed_during_stage1 = 0; if (!skip_stage1) { @@ -965,13 +1011,20 @@ static int migration_tx_fun(int argc, char *argv[]) /* send ramsize, but split it */ double clock_prev_send = MSG_get_clock(); - computed_during_stage1 = send_stage1(vm, src_pm_name, dst_pm_name, ramsize, mig_speed, xfer_cpu_overhead, dp_rate, dp_cap, dpt_cpu_overhead); + TRY{ + computed_during_stage1 = send_stage1(ms, ramsize, mig_speed, xfer_cpu_overhead, dp_rate, dp_cap, dpt_cpu_overhead); + } CATCH_ANONYMOUS{ + //hostfailure + // Stop the dirty page tracking an return (there is no memory space to release) + stop_dirty_page_tracking(ms->vm); + return 0; + } remaining_size -= ramsize; double clock_post_send = MSG_get_clock(); double bandwidth = ramsize / (clock_post_send - clock_prev_send); threshold = get_threshold_value(bandwidth, max_downtime); - XBT_INFO("actual banwdidth %f (MB/s), threshold %f", bandwidth / 1024 / 1024, threshold); + XBT_INFO("actual bandwidth %f (MB/s), threshold %f", bandwidth / 1024 / 1024, threshold); } @@ -993,7 +1046,7 @@ static int migration_tx_fun(int argc, char *argv[]) /* 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(vm, 2, stage2_round); + double computed = lookup_computed_flop_counts(ms->vm, 2, stage2_round); updated_size = get_updated_size(computed, dp_rate, dp_cap); } @@ -1020,19 +1073,19 @@ static int migration_tx_fun(int argc, char *argv[]) } double clock_prev_send = MSG_get_clock(); - - send_migration_data(vm_name, src_pm_name, dst_pm_name, updated_size, mbox, 2, stage2_round, mig_speed, xfer_cpu_overhead); - + TRY{ + send_migration_data(ms->vm, ms->src_pm, ms->dst_pm, updated_size, ms->mbox, 2, stage2_round, mig_speed, xfer_cpu_overhead); + }CATCH_ANONYMOUS{ + //hostfailure + // 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(); double bandwidth = updated_size / (clock_post_send - clock_prev_send); threshold = get_threshold_value(bandwidth, max_downtime); - XBT_INFO("actual banwdidth %f, threshold %f", bandwidth / 1024 / 1024, threshold); - - - - - + XBT_INFO("actual bandwidth %f, threshold %f", bandwidth / 1024 / 1024, threshold); remaining_size -= updated_size; @@ -1043,12 +1096,19 @@ static int migration_tx_fun(int argc, char *argv[]) stage3: /* Stage3: stop the VM and copy the rest of states. */ XBT_INFO("mig-stage3: remaining_size %f", remaining_size); - simcall_vm_suspend(vm); - stop_dirty_page_tracking(vm); - - send_migration_data(vm_name, src_pm_name, dst_pm_name, remaining_size, mbox, 3, 0, mig_speed, xfer_cpu_overhead); - - xbt_free(mbox); + simcall_vm_suspend(ms->vm); + stop_dirty_page_tracking(ms->vm); + + TRY{ + send_migration_data(ms->vm, ms->src_pm, ms->dst_pm, remaining_size, ms->mbox, 3, 0, mig_speed, xfer_cpu_overhead); + }CATCH_ANONYMOUS{ + //hostfailure + // Stop the dirty page tracking an return (there is no memory space to release) + simcall_vm_resume(ms->vm); + return 0; + } + + // At that point the Migration is considered valid for the SRC node but remind that the DST side should relocate effectively the VM on the DST node. XBT_DEBUG("mig: tx_done"); @@ -1059,50 +1119,60 @@ stage3: static void do_migration(msg_vm_t vm, msg_host_t src_pm, msg_host_t dst_pm) { - char *mbox_ctl = get_mig_mbox_ctl(sg_host_name(vm), sg_host_name(src_pm), sg_host_name(dst_pm)); - - { - char *pr_name = get_mig_process_rx_name(sg_host_name(vm), sg_host_name(src_pm), sg_host_name(dst_pm)); - int nargvs = 5; - char **argv = xbt_new(char *, nargvs); - argv[0] = pr_name; - argv[1] = xbt_strdup(sg_host_name(vm)); - argv[2] = xbt_strdup(sg_host_name(src_pm)); - argv[3] = xbt_strdup(sg_host_name(dst_pm)); - argv[4] = NULL; - - MSG_process_create_with_arguments(pr_name, migration_rx_fun, NULL, dst_pm, nargvs - 1, argv); - } - - { - char *pr_name = get_mig_process_tx_name(sg_host_name(vm), sg_host_name(src_pm), sg_host_name(dst_pm)); - int nargvs = 5; - char **argv = xbt_new(char *, nargvs); - argv[0] = pr_name; - argv[1] = xbt_strdup(sg_host_name(vm)); - argv[2] = xbt_strdup(sg_host_name(src_pm)); - argv[3] = xbt_strdup(sg_host_name(dst_pm)); - argv[4] = NULL; - MSG_process_create_with_arguments(pr_name, migration_tx_fun, NULL, src_pm, nargvs - 1, argv); - } + struct migration_session *ms = xbt_new(struct migration_session, 1); + 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); + + + 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_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] = NULL; + 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] = NULL; + MSG_process_create_with_arguments(pr_tx_name, migration_tx_fun, ms, src_pm, 1, argv); + } +#endif - /* wait until the migration have finished */ + /* wait until the migration have finished or on error has occured */ { msg_task_t task = NULL; - msg_error_t ret = MSG_task_recv(&task, mbox_ctl); - - xbt_assert(ret == MSG_OK); - - char *expected_task_name = get_mig_task_name(sg_host_name(vm), sg_host_name(src_pm), sg_host_name(dst_pm), 4); + msg_error_t ret = MSG_task_recv(&task, ms->mbox_ctl); + + xbt_free(ms->mbox_ctl); + xbt_free(ms->mbox); + xbt_free(ms); + + //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 + THROWF(host_error, 0, "DST host failed during the migration of %s", sg_host_name(vm)); + } + + 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); } - - xbt_free(mbox_ctl); } + + /** @brief Migrate the VM to the given host. * @ingroup msg_VMs * @@ -1135,12 +1205,22 @@ void MSG_vm_migrate(msg_vm_t vm, msg_host_t new_pm) msg_host_t old_pm = simcall_vm_get_pm(vm); - if (simcall_vm_get_state(vm) != SURF_VM_STATE_RUNNING) + if (!MSG_vm_is_running(vm)) THROWF(vm_error, 0, "VM(%s) is not running", sg_host_name(vm)); - do_migration(vm, old_pm, new_pm); + if (MSG_vm_is_migrating(vm)) + THROWF(vm_error, 0, "VM(%s) is already migrating", sg_host_name(vm)); + msg_host_priv_t priv = msg_host_resource_priv(vm); + priv->is_migrating = 1; + TRY{ + do_migration(vm, old_pm, new_pm); + } CATCH_ANONYMOUS{ + RETHROW; + // TODO clean the code Adrien + } + priv->is_migrating = 0; XBT_DEBUG("VM(%s) moved from PM(%s) to PM(%s)", vm->key, old_pm->key, new_pm->key); @@ -1153,13 +1233,16 @@ void MSG_vm_migrate(msg_vm_t vm, msg_host_t new_pm) /** @brief Immediately suspend the execution of all processes within the given VM. * @ingroup msg_VMs * - * This function stops the exection of the VM. All the processes on this VM - * will pause. The state of the VM is perserved. We can later resume it again. + * This function stops the execution of the VM. All the processes on this VM + * will pause. The state of the VM is preserved. We can later resume it again. * * No suspension cost occurs. */ void MSG_vm_suspend(msg_vm_t vm) { + if (MSG_vm_is_migrating(vm)) + THROWF(vm_error, 0, "VM(%s) is migrating", sg_host_name(vm)); + simcall_vm_suspend(vm); XBT_DEBUG("vm_suspend done"); @@ -1188,8 +1271,8 @@ void MSG_vm_resume(msg_vm_t vm) /** @brief Immediately save the execution of all processes within the given VM. * @ingroup msg_VMs * - * This function stops the exection of the VM. All the processes on this VM - * will pause. The state of the VM is perserved. We can later resume it again. + * This function stops the execution of the VM. All the processes on this VM + * will pause. The state of the VM is preserved. We can later resume it again. * * FIXME: No suspension cost occurs. If you want to simulate this too, you want to * use a \ref MSG_file_write() before or after, depending on the exact semantic @@ -1197,6 +1280,9 @@ 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_name(vm)); + simcall_vm_save(vm); #ifdef HAVE_TRACING TRACE_msg_vm_save(vm); @@ -1250,7 +1336,7 @@ msg_host_t MSG_vm_get_pm(msg_vm_t vm) * * There are some solutions for this problem. One option is to update the bound * of the dummy CPU action automatically. It should be the sum of all tasks on - * the VM. But, this solution might be costy, because we have to scan all tasks + * the VM. But, this solution might be costly, because we have to scan all tasks * on the VM in share_resource() or we have to trap both the start and end of * task execution. *