#include "src/plugins/vm/VirtualMachineImpl.hpp"
#include "src/plugins/vm/VmHostExt.hpp"
+#include "src/simix/ActorImpl.hpp"
#include <simgrid/s4u/VirtualMachine.hpp>
#include <simgrid/s4u/host.hpp>
/** @brief Start a vm (i.e., boot the guest operating system)
* @ingroup msg_VMs
*
- * If the VM cannot be started, an exception is generated.
+ * If the VM cannot be started (because of memory overprovisionning), an exception is generated.
*/
void MSG_vm_start(msg_vm_t vm)
{
- simcall_vm_start(vm);
+ simgrid::simix::kernelImmediate([vm]() {
+ simgrid::vm::VmHostExt::ensureVmExtInstalled();
+
+ simgrid::s4u::VirtualMachine* typedVM = static_cast<simgrid::s4u::VirtualMachine*>(vm);
+ simgrid::s4u::Host* pm = typedVM->pimpl_vm_->getPm();
+ if (pm->extension<simgrid::vm::VmHostExt>() == nullptr)
+ pm->extension_set(new simgrid::vm::VmHostExt());
+
+ long pm_ramsize = pm->extension<simgrid::vm::VmHostExt>()->ramsize;
+ int pm_overcommit = pm->extension<simgrid::vm::VmHostExt>()->overcommit;
+ long vm_ramsize = typedVM->getRamsize();
+
+ if (pm_ramsize && !pm_overcommit) { /* Only verify that we don't overcommit on need */
+ /* Retrieve the memory occupied by the VMs on that host. Yep, we have to traverse all VMs of all hosts for that */
+ long total_ramsize_of_vms = 0;
+ for (simgrid::s4u::VirtualMachine* ws_vm : simgrid::surf::VirtualMachineImpl::allVms_)
+ if (pm == ws_vm->pimpl_vm_->getPm())
+ total_ramsize_of_vms += ws_vm->pimpl_vm_->getRamsize();
+
+ if (vm_ramsize > pm_ramsize - total_ramsize_of_vms) {
+ XBT_WARN("cannnot start %s@%s due to memory shortage: vm_ramsize %ld, free %ld, pm_ramsize %ld (bytes).",
+ sg_host_get_name(vm), sg_host_get_name(pm), vm_ramsize, pm_ramsize - total_ramsize_of_vms, pm_ramsize);
+ THROWF(vm_error, 0, "Memory shortage on host '%s', VM '%s' cannot be started", pm->cname(), vm->cname());
+ }
+ }
+
+ typedVM->pimpl_vm_->setState(SURF_VM_STATE_RUNNING);
+ });
if (TRACE_msg_vm_is_enabled()) {
container_t vm_container = PJ_container_get(vm->name().c_str());
return (sg_size_t) updated_size;
}
-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;
- const sg_size_t chunksize = (sg_size_t)1024 * 1024 * 100000;
- sg_size_t remaining = ramsize;
- double computed_total = 0;
-
- while (remaining > 0) {
- sg_size_t datasize = chunksize;
- if (remaining < chunksize)
- datasize = remaining;
-
- remaining -= datasize;
- send_migration_data(ms->vm, ms->src_pm, ms->dst_pm, datasize, ms->mbox, 1, 0, mig_speed, -1);
- double computed = lookup_computed_flop_counts(ms->vm, 1, 0);
- computed_total += computed;
- }
-
- return computed_total;
-}
-
-static double get_threshold_value(double bandwidth, double max_downtime)
-{
- return max_downtime * bandwidth;
-}
-
static int migration_tx_fun(int argc, char *argv[])
{
XBT_DEBUG("mig: tx_start");
const double mig_speed = params.mig_speed;
double max_downtime = params.max_downtime;
- /* hard code it temporally. Fix Me */
-#define MIGRATION_TIMEOUT_DO_NOT_HARDCODE_ME 10000000.0
- double mig_timeout = MIGRATION_TIMEOUT_DO_NOT_HARDCODE_ME;
+ double mig_timeout = 10000000.0;
double remaining_size = (double) (ramsize + devsize);
double threshold = 0.0;
double clock_post_send = MSG_get_clock();
mig_timeout -= (clock_post_send - clock_prev_send);
if (mig_timeout < 0) {
- XBT_VERB("The duration of stage 1 exceeds the timeout value (%lf > %lf), skip stage 2",
- (clock_post_send - clock_prev_send), MIGRATION_TIMEOUT_DO_NOT_HARDCODE_ME);
+ XBT_VERB("The duration of stage 1 exceeds the timeout value, skip stage 2");
skip_stage2 = 1;
}
/* estimate bandwidth */
double bandwidth = ramsize / (clock_post_send - clock_prev_send);
- threshold = get_threshold_value(bandwidth, max_downtime);
+ threshold = bandwidth * max_downtime;
XBT_DEBUG("actual bandwidth %f (MB/s), threshold %f", bandwidth / 1024 / 1024, threshold);
}
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);
+ threshold = bandwidth * max_downtime;
XBT_DEBUG("actual bandwidth %f, threshold %f", bandwidth / 1024 / 1024, threshold);
remaining_size -= sent;
stage2_round += 1;
return 0;
}
-static int do_migration(msg_vm_t vm, msg_host_t src_pm, msg_host_t dst_pm)
+/** @brief Migrate the VM to the given host.
+ * @ingroup msg_VMs
+ */
+void MSG_vm_migrate(msg_vm_t vm, msg_host_t dst_pm)
{
+ /* some thoughts:
+ * - One approach is ...
+ * We first create a new VM (i.e., destination VM) on the destination physical host. The destination VM will
+ * receive the state of the source
+ * VM over network. We will finally destroy the source VM.
+ * - This behavior is similar to the way of migration in the real world.
+ * Even before a migration is completed, we will see a destination VM, consuming resources.
+ * - We have to relocate all processes. The existing process migration code will work for this?
+ * - The name of the VM is a somewhat unique ID in the code. It is tricky for the destination VM?
+ *
+ * - Another one is ...
+ * We update the information of the given VM to place it to the destination physical host.
+ *
+ * The second one would be easier.
+ */
+
+ simgrid::s4u::VirtualMachine* typedVm = static_cast<simgrid::s4u::VirtualMachine*>(vm);
+ simgrid::surf::VirtualMachineImpl* pimpl = typedVm->pimpl_vm_;
+ msg_host_t src_pm = pimpl->getPm();
+
+ if (src_pm->isOff())
+ THROWF(vm_error, 0, "Cannot migrate VM '%s' from host '%s', which is offline.", vm->cname(), src_pm->cname());
+ if (dst_pm->isOff())
+ THROWF(vm_error, 0, "Cannot migrate VM '%s' to host '%s', which is offline.", vm->cname(), dst_pm->cname());
+ if (!MSG_vm_is_running(vm))
+ THROWF(vm_error, 0, "Cannot migrate VM '%s' that is not running yet.", vm->cname());
+ if (typedVm->isMigrating())
+ THROWF(vm_error, 0, "Cannot migrate VM '%s' that is already migrating.", vm->cname());
+
+ pimpl->isMigrating = true;
+
struct migration_session *ms = xbt_new(struct migration_session, 1);
ms->vm = vm;
ms->src_pm = src_pm;
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
+ while (ret == MSG_TIMEOUT && dst_pm->isOn()) // The rx will tell me when he gots the VM
ret = MSG_task_receive_with_timeout(&task, ms->mbox_ctl, 4);
+ pimpl->isMigrating = false;
+
xbt_free(ms->mbox_ctl);
xbt_free(ms->mbox);
xbt_free(ms);
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;
+ THROWF(host_error, 0, "Source host '%s' failed during the migration of VM '%s'.", src_pm->cname(), vm->cname());
} 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;
+ THROWF(host_error, 0, "Destination host '%s' failed during the migration of VM '%s'.", dst_pm->cname(),
+ vm->cname());
}
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.
- * @ingroup msg_VMs
- *
- * FIXME: No migration cost occurs. If you want to simulate this too, you want to use a MSG_task_send() before or after,
- * depending on whether you want to do cold or hot migration.
- */
-void MSG_vm_migrate(msg_vm_t vm, msg_host_t new_pm)
-{
- /* some thoughts:
- * - One approach is ...
- * We first create a new VM (i.e., destination VM) on the destination physical host. The destination VM will
- * receive the state of the source
- * VM over network. We will finally destroy the source VM.
- * - This behavior is similar to the way of migration in the real world.
- * Even before a migration is completed, we will see a destination VM, consuming resources.
- * - We have to relocate all processes. The existing process migration code will work for this?
- * - The name of the VM is a somewhat unique ID in the code. It is tricky for the destination VM?
- *
- * - Another one is ...
- * We update the information of the given VM to place it to the destination physical host.
- *
- * The second one would be easier.
- */
-
- simgrid::s4u::VirtualMachine* typedVm = static_cast<simgrid::s4u::VirtualMachine*>(vm);
- simgrid::surf::VirtualMachineImpl* pimpl = typedVm->pimpl_vm_;
- msg_host_t old_pm = pimpl->getPm();
-
- if (old_pm->isOff())
- THROWF(vm_error, 0, "Cannot migrate VM '%s' from host '%s', which is offline.", vm->cname(), old_pm->cname());
-
- 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, "Cannot migrate VM '%s' that is not running yet.", vm->cname());
-
- if (typedVm->isMigrating())
- THROWF(vm_error, 0, "Cannot migrate VM '%s' that is already migrating.", vm->cname());
-
- pimpl->isMigrating = true;
-
- 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());
- }
}
/** @brief Immediately suspend the execution of all processes within the given VM.
*/
void MSG_vm_restore(msg_vm_t vm)
{
- simcall_vm_restore(vm);
+ simgrid::simix::kernelImmediate([vm]() {
+ if (static_cast<simgrid::s4u::VirtualMachine*>(vm)->pimpl_vm_->getState() != SURF_VM_STATE_SAVED)
+ THROWF(vm_error, 0, "VM(%s) was not saved", vm->name().c_str());
+
+ XBT_DEBUG("restore VM(%s), where %d processes exist", vm->name().c_str(),
+ xbt_swag_size(sg_host_simix(vm)->process_list));
+
+ /* jump to vm_ws_restore() */
+ static_cast<simgrid::s4u::VirtualMachine*>(vm)->pimpl_vm_->restore();
+
+ smx_actor_t smx_process, smx_process_safe;
+ xbt_swag_foreach_safe(smx_process, smx_process_safe, sg_host_simix(vm)->process_list)
+ {
+ XBT_DEBUG("resume %s", smx_process->name.c_str());
+ SIMIX_process_resume(smx_process);
+ }
+ });
if (TRACE_msg_vm_is_enabled()) {
container_t vm_container = PJ_container_get(vm->cname());
*/
void MSG_vm_set_bound(msg_vm_t vm, double bound)
{
- simcall_vm_set_bound(vm, bound);
+ simgrid::simix::kernelImmediate(
+ [vm, bound]() { static_cast<simgrid::s4u::VirtualMachine*>(vm)->pimpl_vm_->setBound(bound); });
}
