bmc/remote_state_ping.cc

#include "bmc/remote_state_ping.h"

#include <iterator>
#include <memory>
#include <string>
#include <utility>
#include <vector>

#include "bmc/daemon_context_bmc.h"
#include "daemon_context.h"
#include "safepower_agent.pb.h"
#include "safepower_agent_config.pb.h"
#include "state_updater.h"
#include "one/network_interfaces.pb.h"
#include "one/offline_node_entities.pb.h"
#include "one/resolved_entities.pb.h"
#include "absl/functional/bind_front.h"
#include "absl/log/check.h"
#include "absl/log/log.h"
#include "absl/status/status.h"
#include "absl/strings/str_cat.h"
#include "absl/strings/str_join.h"
#include "absl/synchronization/mutex.h"
#include "absl/time/time.h"
// NOLINTBEGIN(readability/boost) // Runs on BMC.
#include "boost/process.hpp"
#include "boost/process/args.hpp"
#include "boost/system.hpp"
// NOLINTEND(readability/boost)

namespace safepower_agent {

PingStateMonitor::PingStateMonitor(
    std::string entity_tag, std::string hostname,
    safepower_agent_config::PingConfig ping_config,
    std::shared_ptr<StateUpdater<safepower_agent_proto::SystemState>>
        system_state_updater)
    : entity_tag_(std::move(entity_tag)),
      hostname_(std::move(hostname)),
      command_({std::move(*ping_config.mutable_command())}),
      interval_(absl::Milliseconds(ping_config.interval_ms())),
      system_state_updater_(std::move(system_state_updater)),
      args_(std::make_move_iterator(ping_config.mutable_extra_args()->begin()),
            std::make_move_iterator(ping_config.mutable_extra_args()->end())) {
  args_.push_back("-c");
  args_.push_back(absl::StrCat(ping_config.count_per_check()));
  args_.push_back(hostname_);
  LOG(INFO) << "Ping command: " << command_ << " " << absl::StrJoin(args_, " ");
}

PingStateMonitor::PingStateMonitor(PingStateMonitor&& other)
    : entity_tag_(std::move(other.entity_tag_)),
      hostname_(std::move(other.hostname_)),
      command_(std::move(other.command_)),
      interval_(std::move(other.interval_)),
      system_state_updater_(std::move(other.system_state_updater_)),
      args_(std::move(other.args_)),
      reachable_state_(other.reachable_state_) {
  CHECK(!other.Started()) << "PingStateMonitor moved while started";
}

PingStateMonitor& PingStateMonitor::operator=(PingStateMonitor&& other) {
  if (Started()) {
    LOG(DFATAL) << "PingStateMonitor replaced while started";
    return *this;
  }

  if (other.Started()) {
    LOG(DFATAL) << "PingStateMonitor moved while started";
    return *this;
  }

  entity_tag_ = std::move(other.entity_tag_);
  hostname_ = std::move(other.hostname_);
  command_ = std::move(other.command_);
  interval_ = std::move(other.interval_);
  system_state_updater_ = std::move(other.system_state_updater_);
  args_ = std::move(other.args_);
  reachable_state_ = other.reachable_state_;
  return *this;
}

bool PingStateMonitor::Started() const {
  absl::MutexLock lock(task_mutex_);
  return !task_name_.empty() || activate_callback_handle_.pending() ||
         idle_callback_handle_.pending();
}

void PingStateMonitor::StopTimer() {
  if (task_name_.empty()) {
    return;
  }
  LOG(INFO) << "Stopping ping state monitor for " << entity_tag_;
  absl::Status status = DaemonContext::Get().scheduler().CancelCall(task_name_);
  if (!status.ok()) {
    LOG(WARNING) << "Failed to cancel ping state monitor: " << status;
  }
  task_name_.clear();
}

void PingStateMonitor::StartTimer(absl::Duration interval) {
  LOG(INFO) << "Starting ping state monitor for " << entity_tag_
            << " with interval " << interval;
  std::string task_name = absl::StrCat("ping_state_monitor_", entity_tag_);
  absl::Status status = DaemonContext::Get().scheduler().PeriodicCall(
      absl::bind_front(&PingStateMonitor::Ping, this), interval, task_name);
  if (!status.ok()) {
    LOG(DFATAL) << "Failed to start ping state monitor: " << status;
    return;
  }
  task_name_ = std::move(task_name);
}

void PingStateMonitor::Pause() {
  {
    absl::MutexLock lock(task_mutex_);
    if (task_name_.empty()) {
      LOG(DFATAL) << "PingStateMonitor paused while task not started";
    }
    StopTimer();
    // When the state updater is paused, we don't know the current state of the
    // node. e.g. it is possible that the node goes down or comes back up while
    // the updater is paused. When a new listener is added, setting an unknown
    // state ensures that we never report a stale state. Since state comparisons
    // are usually equality checks, this should not break any assumptions.
    UpdateReachableState(ReachableState::kUnknown);
  }
  activate_callback_handle_ = system_state_updater_->OnActive(
      absl::bind_front(&PingStateMonitor::Resume, this));
}

void PingStateMonitor::Resume() {
  {
    absl::MutexLock lock(task_mutex_);
    if (!task_name_.empty()) {
      LOG(DFATAL) << "PingStateMonitor resumed while task started";
      return;
    }
    StartTimer(interval_);
  }
  idle_callback_handle_ = system_state_updater_->OnIdle(
      absl::bind_front(&PingStateMonitor::Pause, this));
}

absl::Status PingStateMonitor::Start() {
  // Will start immediately if the state updater is idle.
  activate_callback_handle_ = system_state_updater_->OnActive(
      absl::bind_front(&PingStateMonitor::Resume, this));
  return absl::OkStatus();
}

PingStateMonitor::~PingStateMonitor() {
  idle_callback_handle_.TryCancel();
  activate_callback_handle_.TryCancel();
  if (!task_mutex_.try_lock()) {
    LOG(DFATAL) << "PingStateMonitor destroyed while mutex was held";
    task_mutex_.lock();
  }
  StopTimer();
  task_mutex_.unlock();
}

// TODO(jebr): b/450140775 - occasionally log the ping command output
void PingStateMonitor::Ping() {
  boost::process::async_system(
      DaemonContextBMC::Get().get_io_context(),
      absl::bind_front(&PingStateMonitor::ProcessPingResult, this),
      boost::process::exe = command_, boost::process::args = args_,
      (boost::process::std_out & boost::process::std_err) >
          boost::process::null);
}

void PingStateMonitor::ProcessPingResult(boost::system::error_code ec,
                                         int exit_code) {
  if (ec) {
    LOG(ERROR) << "Failed to ping " << hostname_ << ": " << ec.message();
    return;
  }
  ReachableState new_reachable_state = exit_code == 0
                                           ? ReachableState::kReachable
                                           : ReachableState::kUnreachable;
  UpdateReachableState(new_reachable_state);
}

void PingStateMonitor::UpdateReachableState(
    ReachableState new_reachable_state) {
  if (new_reachable_state == reachable_state_) {
    return;
  }
  safepower_agent_proto::BootStateSpecifier state;
  switch (new_reachable_state) {
    case ReachableState::kUnknown:
      LOG(INFO) << entity_tag_ << " is reachability state is unknown";
      state = safepower_agent_proto::BOOT_STATE_UNSPECIFIED;
      break;
    case ReachableState::kReachable:
      LOG(INFO) << entity_tag_ << " is reachable";
      state = safepower_agent_proto::BOOT_STATE_BOOTING;
      break;
    case ReachableState::kUnreachable:
      LOG(INFO) << entity_tag_ << " is unreachable";
      state = safepower_agent_proto::BOOT_STATE_NOT_BOOTED;
      break;
    default:
      LOG(DFATAL) << "Unknown reachable state: "
                  << static_cast<int>(new_reachable_state);
      return;
  }
  safepower_agent_proto::SystemState new_state;
  (*new_state.mutable_node_state())[entity_tag_]
      .mutable_boot_state()
      ->set_state(state);
  system_state_updater_->UpdateState(new_state);
  reachable_state_ = new_reachable_state;
}

}  // namespace safepower_agent