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
  22 #include <pthread.h>
  23
  24 #include "RWLock.h"
  25
  26 class CephContext;
  27
  28 namespace ceph {
  29
  30 /*
  31  * HeartbeatMap -
  32  *
  33  * Maintain a set of handles for internal subsystems to periodically
  34  * check in with a health check and timeout.  Each user can register
  35  * and get a handle they can use to set or reset a timeout.
  36  *
  37  * A simple is_healthy() method checks for any users who are not within
  38  * their grace period for a heartbeat.
  39  */
  40
  41 struct heartbeat_handle_d {
  42   const std::string name;
  43   pthread_t thread_id;
  44   std::atomic<unsigned> timeout = { 0 }, suicide_timeout = { 0 };
  45   time_t grace, suicide_grace;
  46   std::list<heartbeat_handle_d*>::iterator list_item;
  47
  48   explicit heartbeat_handle_d(const std::string& n)
  49     : name(n), thread_id(0), grace(0), suicide_grace(0)
  50   { }
  51 };
  52
  53 class HeartbeatMap {
  54  public:
  55   // register/unregister
  56   heartbeat_handle_d *add_worker(const std::string& name, pthread_t thread_id);
  57   void remove_worker(const heartbeat_handle_d *h);
  58
  59   // reset the timeout so that it expects another touch within grace amount of time
  60   void reset_timeout(heartbeat_handle_d *h, time_t grace, time_t suicide_grace);
  61   // clear the timeout so that it's not checked on
  62   void clear_timeout(heartbeat_handle_d *h);
  63
  64   // return false if any of the timeouts are currently expired.
  65   bool is_healthy();
  66
  67   // touch cct->_conf->heartbeat_file if is_healthy()
  68   void check_touch_file();
  69
  70   // get the number of unhealthy workers
  71   int get_unhealthy_workers() const;
  72
  73   // get the number of total workers
  74   int get_total_workers() const;
  75
  76   explicit HeartbeatMap(CephContext *cct);
  77   ~HeartbeatMap();
  78
  79  private:
  80   CephContext *m_cct;
  81   RWLock m_rwlock;
  82   time_t m_inject_unhealthy_until;
  83   std::list<heartbeat_handle_d*> m_workers;
  84   std::atomic<unsigned> m_unhealthy_workers = { 0 };
  85   std::atomic<unsigned> m_total_workers = { 0 };
  86
  87   bool _check(const heartbeat_handle_d *h, const char *who, time_t now);
  88 };
  89
  90 }
  91 #endif