ceph/src/common/HeartbeatMap.h

   1 // -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*-
   2 // vim: ts=8 sw=2 smarttab
   3 /*
   4  * Ceph - scalable distributed file system
   5  *
   6  * Copyright (C) 2011 Sage Weil <sage@newdream.net>
   7  *
   8  * This is free software; you can redistribute it and/or
   9  * modify it under the terms of the GNU Lesser General Public
  10  * License version 2.1, as published by the Free Software
  11  * Foundation.  See file COPYING.
  12  *
  13  */
  14
  15 #ifndef CEPH_HEARTBEATMAP_H
  16 #define CEPH_HEARTBEATMAP_H
  17
  18 #include <list>
  19 #include <atomic>
  20 #include <string>
  21 #include <pthread.h>
  22
  23 #include "common/ceph_time.h"
  24 #include "common/ceph_mutex.h"
  25 #include "include/common_fwd.h"
  26
  27 namespace ceph {
  28
  29 /*
  30  * HeartbeatMap -
  31  *
  32  * Maintain a set of handles for internal subsystems to periodically
  33  * check in with a health check and timeout.  Each user can register
  34  * and get a handle they can use to set or reset a timeout.
  35  *
  36  * A simple is_healthy() method checks for any users who are not within
  37  * their grace period for a heartbeat.
  38  */
  39
  40 struct heartbeat_handle_d {
  41   const std::string name;
  42   pthread_t thread_id = 0;
  43   using clock = ceph::coarse_mono_clock;
  44   using time = ceph::coarse_mono_time;
  45   std::atomic<time> timeout = clock::zero();
  46   std::atomic<time> suicide_timeout = clock::zero();
  47   ceph::timespan grace = ceph::timespan::zero();
  48   ceph::timespan suicide_grace = ceph::timespan::zero();
  49   std::list<heartbeat_handle_d*>::iterator list_item;
  50
  51   explicit heartbeat_handle_d(const std::string& n)
  52     : name(n)
  53   { }
  54 };
  55
  56 class HeartbeatMap {
  57  public:
  58   // register/unregister
  59   heartbeat_handle_d *add_worker(const std::string& name, pthread_t thread_id);
  60   void remove_worker(const heartbeat_handle_d *h);
  61
  62   // reset the timeout so that it expects another touch within grace amount of time
  63   void reset_timeout(heartbeat_handle_d *h,
  64                      ceph::timespan grace,
  65                      ceph::timespan suicide_grace);
  66   // clear the timeout so that it's not checked on
  67   void clear_timeout(heartbeat_handle_d *h);
  68
  69   // return false if any of the timeouts are currently expired.
  70   bool is_healthy();
  71
  72   // touch cct->_conf->heartbeat_file if is_healthy()
  73   void check_touch_file();
  74
  75   // get the number of unhealthy workers
  76   int get_unhealthy_workers() const;
  77
  78   // get the number of total workers
  79   int get_total_workers() const;
  80
  81   explicit HeartbeatMap(CephContext *cct);
  82   ~HeartbeatMap();
  83
  84  private:
  85   using clock = ceph::coarse_mono_clock;
  86   CephContext *m_cct;
  87   ceph::shared_mutex m_rwlock =
  88     ceph::make_shared_mutex("HeartbeatMap::m_rwlock");
  89   clock::time_point m_inject_unhealthy_until;
  90   std::list<heartbeat_handle_d*> m_workers;
  91   std::atomic<unsigned> m_unhealthy_workers = { 0 };
  92   std::atomic<unsigned> m_total_workers = { 0 };
  93
  94   bool _check(const heartbeat_handle_d *h, const char *who,
  95               ceph::coarse_mono_time now);
  96 };
  97
  98 }
  99 #endif