ceph/src/mon/DataHealthService.cc

   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) 2013 Inktank, Inc
   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 #include "include/memory.h"
  15 #include <errno.h>
  16 #include <map>
  17 #include <list>
  18 #include <string>
  19 #include <sstream>
  20
  21 #include "acconfig.h"
  22
  23 #ifdef HAVE_SYS_VFS_H
  24 #include <sys/vfs.h>
  25 #endif
  26
  27 #ifdef HAVE_SYS_MOUNT_H
  28 #include <sys/mount.h>
  29 #endif
  30
  31 #ifdef HAVE_SYS_PARAM_H
  32 #include <sys/param.h>
  33 #endif
  34
  35 #include "messages/MMonHealth.h"
  36 #include "include/assert.h"
  37 #include "common/Formatter.h"
  38 #include "common/errno.h"
  39
  40 #include "mon/Monitor.h"
  41 #include "mon/DataHealthService.h"
  42
  43 #define dout_subsys ceph_subsys_mon
  44 #undef dout_prefix
  45 #define dout_prefix _prefix(_dout, mon, this)
  46 static ostream& _prefix(std::ostream *_dout, const Monitor *mon,
  47                         const DataHealthService *svc) {
  48   assert(mon != NULL);
  49   assert(svc != NULL);
  50   return *_dout << "mon." << mon->name << "@" << mon->rank
  51                 << "(" << mon->get_state_name() << ")." << svc->get_name()
  52                 << "(" << svc->get_epoch() << ") ";
  53 }
  54
  55 void DataHealthService::start_epoch()
  56 {
  57   dout(10) << __func__ << " epoch " << get_epoch() << dendl;
  58   // we are not bound by election epochs, but we should clear the stats
  59   // everytime an election is triggerd.  As far as we know, a monitor might
  60   // have been running out of disk space and someone fixed it.  We don't want
  61   // to hold the cluster back, even confusing the user, due to some possibly
  62   // outdated stats.
  63   stats.clear();
  64   last_warned_percent = 0;
  65 }
  66
  67 void DataHealthService::get_health(
  68     Formatter *f,
  69     list<pair<health_status_t,string> >& summary,
  70     list<pair<health_status_t,string> > *detail)
  71 {
  72   dout(10) << __func__ << dendl;
  73   if (f) {
  74     f->open_object_section("data_health");
  75     f->open_array_section("mons");
  76   }
  77
  78   for (map<entity_inst_t,DataStats>::iterator it = stats.begin();
  79        it != stats.end(); ++it) {
  80     string mon_name = mon->monmap->get_name(it->first.addr);
  81     DataStats& stats = it->second;
  82
  83     health_status_t health_status = HEALTH_OK;
  84     string health_detail;
  85     if (stats.fs_stats.avail_percent <= g_conf->mon_data_avail_crit) {
  86       health_status = HEALTH_ERR;
  87       health_detail = "low disk space, shutdown imminent";
  88     } else if (stats.fs_stats.avail_percent <= g_conf->mon_data_avail_warn) {
  89       health_status = HEALTH_WARN;
  90       health_detail = "low disk space";
  91     }
  92
  93     if (stats.store_stats.bytes_total >= g_conf->mon_data_size_warn) {
  94       if (health_status > HEALTH_WARN)
  95         health_status = HEALTH_WARN;
  96       if (!health_detail.empty())
  97         health_detail.append("; ");
  98       stringstream ss;
  99       ss << "store is getting too big! "
 100          << prettybyte_t(stats.store_stats.bytes_total)
 101          << " >= " << prettybyte_t(g_conf->mon_data_size_warn);
 102       health_detail.append(ss.str());
 103     }
 104
 105     if (health_status != HEALTH_OK) {
 106       stringstream ss;
 107       ss << "mon." << mon_name << " " << health_detail;
 108       summary.push_back(make_pair(health_status, ss.str()));
 109       ss << " -- " <<  stats.fs_stats.avail_percent << "% avail";
 110       if (detail)
 111         detail->push_back(make_pair(health_status, ss.str()));
 112     }
 113
 114     if (f) {
 115       f->open_object_section("mon");
 116       f->dump_string("name", mon_name.c_str());
 117       // leave this unenclosed by an object section to avoid breaking backward-compatibility
 118       stats.dump(f);
 119       f->dump_stream("health") << health_status;
 120       if (health_status != HEALTH_OK)
 121         f->dump_string("health_detail", health_detail);
 122       f->close_section();
 123     }
 124   }
 125
 126   if (f) {
 127     f->close_section(); // mons
 128     f->close_section(); // data_health
 129   }
 130 }
 131
 132 int DataHealthService::update_store_stats(DataStats &ours)
 133 {
 134   map<string,uint64_t> extra;
 135   uint64_t store_size = mon->store->get_estimated_size(extra);
 136   assert(store_size > 0);
 137
 138   ours.store_stats.bytes_total = store_size;
 139   ours.store_stats.bytes_sst = extra["sst"];
 140   ours.store_stats.bytes_log = extra["log"];
 141   ours.store_stats.bytes_misc = extra["misc"];
 142   ours.last_update = ceph_clock_now();
 143
 144   return 0;
 145 }
 146
 147
 148 int DataHealthService::update_stats()
 149 {
 150   entity_inst_t our_inst = mon->messenger->get_myinst();
 151   DataStats& ours = stats[our_inst];
 152
 153   int err = get_fs_stats(ours.fs_stats, g_conf->mon_data.c_str());
 154   if (err < 0) {
 155     derr << __func__ << " get_fs_stats error: " << cpp_strerror(err) << dendl;
 156     return err;
 157   }
 158   dout(0) << __func__ << " avail " << ours.fs_stats.avail_percent << "%"
 159           << " total " << prettybyte_t(ours.fs_stats.byte_total)
 160           << ", used " << prettybyte_t(ours.fs_stats.byte_used)
 161           << ", avail " << prettybyte_t(ours.fs_stats.byte_avail) << dendl;
 162   ours.last_update = ceph_clock_now();
 163
 164   return update_store_stats(ours);
 165 }
 166
 167 void DataHealthService::share_stats()
 168 {
 169   dout(10) << __func__ << dendl;
 170   if (!in_quorum())
 171     return;
 172
 173   assert(!stats.empty());
 174   entity_inst_t our_inst = mon->messenger->get_myinst();
 175   assert(stats.count(our_inst) > 0);
 176   DataStats &ours = stats[our_inst];
 177   const set<int>& quorum = mon->get_quorum();
 178   for (set<int>::const_iterator it = quorum.begin();
 179        it != quorum.end(); ++it) {
 180     if (mon->monmap->get_name(*it) == mon->name)
 181       continue;
 182     entity_inst_t inst = mon->monmap->get_inst(*it);
 183     MMonHealth *m = new MMonHealth(HealthService::SERVICE_HEALTH_DATA,
 184                                    MMonHealth::OP_TELL);
 185     m->data_stats = ours;
 186     dout(20) << __func__ << " send " << *m << " to " << inst << dendl;
 187     mon->messenger->send_message(m, inst);
 188   }
 189 }
 190
 191 void DataHealthService::service_tick()
 192 {
 193   dout(10) << __func__ << dendl;
 194
 195   int err = update_stats();
 196   if (err < 0) {
 197     derr << "something went wrong obtaining our disk stats: "
 198          << cpp_strerror(err) << dendl;
 199     force_shutdown();
 200     return;
 201   }
 202   if (in_quorum())
 203     share_stats();
 204
 205   DataStats &ours = stats[mon->messenger->get_myinst()];
 206
 207   if (ours.fs_stats.avail_percent <= g_conf->mon_data_avail_crit) {
 208     derr << "reached critical levels of available space on local monitor storage"
 209          << " -- shutdown!" << dendl;
 210     force_shutdown();
 211     return;
 212   }
 213
 214   // we must backoff these warnings, and track how much data is being
 215   // consumed in-between reports to assess if it's worth to log this info,
 216   // otherwise we may very well contribute to the consumption of the
 217   // already low available disk space.
 218   if (ours.fs_stats.avail_percent <= g_conf->mon_data_avail_warn) {
 219     if (ours.fs_stats.avail_percent != last_warned_percent)
 220       mon->clog->warn()
 221         << "reached concerning levels of available space on local monitor storage"
 222         << " (" << ours.fs_stats.avail_percent << "% free)";
 223     last_warned_percent = ours.fs_stats.avail_percent;
 224   } else {
 225     last_warned_percent = 0;
 226   }
 227 }
 228
 229 void DataHealthService::handle_tell(MonOpRequestRef op)
 230 {
 231   op->mark_event("datahealth:handle_tell");
 232   MMonHealth *m = static_cast<MMonHealth*>(op->get_req());
 233   dout(10) << __func__ << " " << *m << dendl;
 234   assert(m->get_service_op() == MMonHealth::OP_TELL);
 235
 236   stats[m->get_source_inst()] = m->data_stats;
 237 }
 238
 239 bool DataHealthService::service_dispatch_op(MonOpRequestRef op)
 240 {
 241   op->mark_event("datahealth:service_dispatch_op");
 242   MMonHealth *m = static_cast<MMonHealth*>(op->get_req());
 243   dout(10) << __func__ << " " << *m << dendl;
 244   assert(m->get_service_type() == get_type());
 245   if (!in_quorum()) {
 246     dout(1) << __func__ << " not in quorum -- drop message" << dendl;
 247     return false;
 248   }
 249
 250   switch (m->service_op) {
 251     case MMonHealth::OP_TELL:
 252       // someone is telling us their stats
 253       handle_tell(op);
 254       break;
 255     default:
 256       dout(0) << __func__ << " unknown op " << m->service_op << dendl;
 257       assert(0 == "Unknown service op");
 258       break;
 259   }
 260   return true;
 261 }