-/* Copyright (c) 2013-2019. The SimGrid Team. All rights reserved. */
+/* Copyright (c) 2013-2020. The SimGrid Team. All rights reserved. */
/* 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. */
#include "src/plugins/vm/VirtualMachineImpl.hpp"
#include "src/plugins/vm/VmHostExt.hpp"
-XBT_LOG_NEW_DEFAULT_CATEGORY(vm_live_migration, "S4U virtual machines live migration");
+XBT_LOG_NEW_DEFAULT_SUBCATEGORY(vm_live_migration, s4u, "S4U virtual machines live migration");
namespace simgrid {
namespace vm {
-simgrid::xbt::Extension<s4u::Host, VmMigrationExt> VmMigrationExt::EXTENSION_ID;
+xbt::Extension<s4u::Host, VmMigrationExt> VmMigrationExt::EXTENSION_ID;
void VmMigrationExt::ensureVmMigrationExtInstalled()
{
if (not EXTENSION_ID.valid())
- EXTENSION_ID = simgrid::s4u::Host::extension_create<VmMigrationExt>();
+ EXTENSION_ID = s4u::Host::extension_create<VmMigrationExt>();
}
void MigrationRx::operator()()
std::string("__mig_stage3:") + vm_->get_cname() + "(" + src_pm_->get_cname() + "-" + dst_pm_->get_cname() + ")";
while (not received_finalize) {
- std::string* payload = static_cast<std::string*>(mbox->get());
+ const std::string* payload = static_cast<std::string*>(mbox->get());
if (finalize_task_name == *payload)
received_finalize = true;
vm_->resume();
// Now the VM is running on the new host (the migration is completed) (even if the SRC crash)
- vm_->get_impl()->is_migrating_ = false;
+ vm_->get_impl()->end_migration();
XBT_DEBUG("VM(%s) moved from PM(%s) to PM(%s)", vm_->get_cname(), src_pm_->get_cname(), dst_pm_->get_cname());
if (TRACE_vm_is_enabled()) {
counter++;
// start link
- container_t msg = simgrid::instr::Container::by_name(vm_->get_name());
- simgrid::instr::Container::get_root()->get_link("VM_LINK")->start_event(msg, "M", key);
+ container_t msg = instr::Container::by_name(vm_->get_name());
+ instr::Container::get_root()->get_link("VM_LINK")->start_event(msg, "M", key);
// destroy existing container of this vm
- simgrid::instr::Container::by_name(vm_->get_name())->remove_from_parent();
+ instr::Container::by_name(vm_->get_name())->remove_from_parent();
// create new container on the new_host location
- new simgrid::instr::Container(vm_->get_name(), "VM", simgrid::instr::Container::by_name(dst_pm_->get_name()));
+ new instr::Container(vm_->get_name(), "VM", instr::Container::by_name(dst_pm_->get_name()));
// end link
- msg = simgrid::instr::Container::by_name(vm_->get_name());
- simgrid::instr::Container::get_root()->get_link("VM_LINK")->end_event(msg, "M", key);
+ msg = instr::Container::by_name(vm_->get_name());
+ instr::Container::get_root()->get_link("VM_LINK")->end_event(msg, "M", key);
}
// Inform the SRC that the migration has been correctly performed
std::string* payload = new std::string("__mig_stage4:");
double clock_sta = s4u::Engine::get_clock();
- s4u::Activity* comm = nullptr;
+ s4u::CommPtr comm = mbox->put_init(msg, size);
+ if (mig_speed > 0)
+ comm->set_rate(mig_speed);
try {
- if (mig_speed > 0)
- comm = mbox->put_init(msg, size)->set_rate(mig_speed)->wait_for(timeout);
- else
- comm = mbox->put_async(msg, size)->wait_for(timeout);
- } catch (xbt_ex& e) {
- if (comm) {
- sg_size_t remaining = static_cast<sg_size_t>(comm->get_remaining());
- XBT_VERB("timeout (%lf s) in sending_migration_data, remaining %llu bytes of %llu", timeout, remaining, size);
- sent -= remaining;
- }
+ comm->wait_for(timeout);
+ } catch (const Exception&) {
+ sg_size_t remaining = static_cast<sg_size_t>(comm->get_remaining());
+ XBT_VERB("timeout (%lf s) in sending_migration_data, remaining %llu bytes of %llu", timeout, remaining, size);
+ sent -= remaining;
delete msg;
}
skip_stage2 = true;
} else if (sent > ramsize)
XBT_CRITICAL("bug");
-
- } catch (xbt_ex& e) {
+ } catch (const Exception&) {
// 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)
sg_vm_stop_dirty_page_tracking(vm_);
/* Stage2: send update pages iteratively until the size of remaining states becomes smaller than threshold value. */
if (not 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 = sg_vm_lookup_computed_flops(vm_);
- updated_size = get_updated_size(computed, dp_rate, dp_cap);
- }
-
+ /* just after stage1, nothing has been updated. But, we have to send the data updated during stage1 */
+ sg_size_t updated_size = get_updated_size(computed_during_stage1, dp_rate, dp_cap);
+ remaining_size += updated_size;
+ XBT_DEBUG("mig-stage2.%d: remaining_size %zu (%s threshold %zu)", stage2_round, remaining_size,
+ (remaining_size < threshold) ? "<" : ">", threshold);
+
+ /* When the remaining size is below the threshold value, move to stage 3. */
+ while (threshold < remaining_size) {
XBT_DEBUG("mig-stage 2:%d updated_size %llu computed_during_stage1 %f dp_rate %f dp_cap %llu", 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 %zu (%s threshold %zu)", stage2_round, remaining_size,
- (remaining_size < threshold) ? "<" : ">", threshold);
- if (remaining_size < threshold)
- break;
-
sg_size_t sent = 0;
clock_prev_send = s4u::Engine::get_clock();
try {
XBT_DEBUG("Stage 2, gonna send %llu", updated_size);
sent = sendMigrationData(updated_size, 2, stage2_round, mig_speed, mig_timeout);
- } catch (xbt_ex& e) {
+ } catch (const Exception&) {
// 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)
sg_vm_stop_dirty_page_tracking(vm_);
return;
}
+
+ remaining_size -= sent;
+ double computed = sg_vm_lookup_computed_flops(vm_);
+
clock_post_send = s4u::Engine::get_clock();
if (sent == updated_size) {
- /* timeout did not happen */
bandwidth = updated_size / (clock_post_send - clock_prev_send);
threshold = bandwidth * max_downtime;
XBT_DEBUG("actual bandwidth %f, threshold %zu", 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) {
+ XBT_DEBUG("mig-stage2.%d: remaining_size %zu (%s threshold %zu)", stage2_round, remaining_size,
+ (remaining_size < threshold) ? "<" : ">", threshold);
+ updated_size = get_updated_size(computed, dp_rate, dp_cap);
+ remaining_size += updated_size;
+ } else {
/* 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 = sg_vm_lookup_computed_flops(vm_);
updated_size = get_updated_size(computed, dp_rate, dp_cap);
remaining_size += updated_size;
break;
- } else
- XBT_CRITICAL("bug");
+ }
}
}
try {
XBT_DEBUG("Stage 3: Gonna send %zu bytes", remaining_size);
sendMigrationData(remaining_size, 3, 0, mig_speed, -1);
- } catch (xbt_ex& e) {
+ } catch (const Exception&) {
// 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)
vm_->resume();
// effectively the VM on the DST node.
XBT_DEBUG("mig: tx_done");
}
-}
-}
+} // namespace vm
+} // namespace simgrid
-static void onVirtualMachineShutdown(simgrid::s4u::VirtualMachine& vm)
+static void onVirtualMachineShutdown(simgrid::s4u::VirtualMachine const& vm)
{
- if (vm.get_impl()->is_migrating_) {
+ if (vm.get_impl()->is_migrating()) {
vm.extension<simgrid::vm::VmMigrationExt>()->rx_->kill();
vm.extension<simgrid::vm::VmMigrationExt>()->tx_->kill();
vm.extension<simgrid::vm::VmMigrationExt>()->issuer_->kill();
- vm.get_impl()->is_migrating_ = false;
+ vm.get_impl()->end_migration();
}
}
return vm;
}
-int sg_vm_is_migrating(simgrid::s4u::VirtualMachine* vm)
+int sg_vm_is_migrating(const simgrid::s4u::VirtualMachine* vm)
{
- return vm->get_impl()->is_migrating_;
+ return vm->get_impl()->is_migrating();
}
void sg_vm_migrate(simgrid::s4u::VirtualMachine* vm, simgrid::s4u::Host* dst_pm)
simgrid::s4u::Host* src_pm = vm->get_pm();
if (not src_pm->is_on())
- THROWF(vm_error, 0, "Cannot migrate VM '%s' from host '%s', which is offline.", vm->get_cname(),
- src_pm->get_cname());
+ throw simgrid::VmFailureException(
+ XBT_THROW_POINT, simgrid::xbt::string_printf("Cannot migrate VM '%s' from host '%s', which is offline.",
+ vm->get_cname(), src_pm->get_cname()));
if (not dst_pm->is_on())
- THROWF(vm_error, 0, "Cannot migrate VM '%s' to host '%s', which is offline.", vm->get_cname(), dst_pm->get_cname());
+ throw simgrid::VmFailureException(
+ XBT_THROW_POINT, simgrid::xbt::string_printf("Cannot migrate VM '%s' to host '%s', which is offline.",
+ vm->get_cname(), dst_pm->get_cname()));
if (vm->get_state() != simgrid::s4u::VirtualMachine::state::RUNNING)
- THROWF(vm_error, 0, "Cannot migrate VM '%s' that is not running yet.", vm->get_cname());
- if (vm->get_impl()->is_migrating_)
- THROWF(vm_error, 0, "Cannot migrate VM '%s' that is already migrating.", vm->get_cname());
-
- vm->get_impl()->is_migrating_ = true;
+ throw simgrid::VmFailureException(
+ XBT_THROW_POINT,
+ simgrid::xbt::string_printf("Cannot migrate VM '%s' that is not running yet.", vm->get_cname()));
+ if (vm->get_impl()->is_migrating())
+ throw simgrid::VmFailureException(
+ XBT_THROW_POINT,
+ simgrid::xbt::string_printf("Cannot migrate VM '%s' that is already migrating.", vm->get_cname()));
+
+ vm->get_impl()->start_migration();
simgrid::s4u::VirtualMachine::on_migration_start(*vm);
std::string rx_name =
/* 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");
- simgrid::s4u::MailboxPtr mbox_ctl = simgrid::s4u::Mailbox::by_name(
+ simgrid::s4u::Mailbox* mbox_ctl = simgrid::s4u::Mailbox::by_name(
std::string("__mbox_mig_ctl:") + vm->get_cname() + "(" + src_pm->get_cname() + "-" + dst_pm->get_cname() + ")");
delete static_cast<std::string*>(mbox_ctl->get());
tx->join();
rx->join();
- vm->get_impl()->is_migrating_ = false;
+ vm->get_impl()->end_migration();
simgrid::s4u::VirtualMachine::on_migration_end(*vm);
}