ceph/src/mgr/ClusterState.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) 2014 John Spray <john.spray@inktank.com>
   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 "messages/MMgrDigest.h"
  15 #include "messages/MMonMgrReport.h"
  16 #include "messages/MPGStats.h"
  17
  18 #include "mgr/ClusterState.h"
  19 #include <time.h>
  20 #include <boost/range/adaptor/reversed.hpp>
  21
  22 #define dout_context g_ceph_context
  23 #define dout_subsys ceph_subsys_mgr
  24 #undef dout_prefix
  25 #define dout_prefix *_dout << "mgr " << __func__ << " "
  26
  27 ClusterState::ClusterState(
  28   MonClient *monc_,
  29   Objecter *objecter_,
  30   const MgrMap& mgrmap)
  31   : monc(monc_),
  32     objecter(objecter_),
  33     mgr_map(mgrmap),
  34     asok_hook(NULL)
  35 {}
  36
  37 void ClusterState::set_objecter(Objecter *objecter_)
  38 {
  39   std::lock_guard l(lock);
  40
  41   objecter = objecter_;
  42 }
  43
  44 void ClusterState::set_fsmap(FSMap const &new_fsmap)
  45 {
  46   std::lock_guard l(lock);
  47
  48   fsmap = new_fsmap;
  49 }
  50
  51 void ClusterState::set_mgr_map(MgrMap const &new_mgrmap)
  52 {
  53   std::lock_guard l(lock);
  54   mgr_map = new_mgrmap;
  55 }
  56
  57 void ClusterState::set_service_map(ServiceMap const &new_service_map)
  58 {
  59   std::lock_guard l(lock);
  60   servicemap = new_service_map;
  61 }
  62
  63 void ClusterState::load_digest(MMgrDigest *m)
  64 {
  65   health_json = std::move(m->health_json);
  66   mon_status_json = std::move(m->mon_status_json);
  67 }
  68
  69 void ClusterState::ingest_pgstats(ref_t<MPGStats> stats)
  70 {
  71   std::lock_guard l(lock);
  72
  73   const int from = stats->get_orig_source().num();
  74   pending_inc.update_stat(from, std::move(stats->osd_stat));
  75
  76   for (auto p : stats->pg_stat) {
  77     pg_t pgid = p.first;
  78     const auto &pg_stats = p.second;
  79
  80     // In case we're hearing about a PG that according to last
  81     // OSDMap update should not exist
  82     auto r = existing_pools.find(pgid.pool());
  83     if (r == existing_pools.end()) {
  84       dout(15) << " got " << pgid
  85                << " reported at " << pg_stats.reported_epoch << ":"
  86                << pg_stats.reported_seq
  87                << " state " << pg_state_string(pg_stats.state)
  88                << " but pool not in " << existing_pools
  89                << dendl;
  90       continue;
  91     }
  92     if (pgid.ps() >= r->second) {
  93       dout(15) << " got " << pgid
  94                << " reported at " << pg_stats.reported_epoch << ":"
  95                << pg_stats.reported_seq
  96                << " state " << pg_state_string(pg_stats.state)
  97                << " but > pg_num " << r->second
  98                << dendl;
  99       continue;
 100     }
 101     // In case we already heard about more recent stats from this PG
 102     // from another OSD
 103     const auto q = pg_map.pg_stat.find(pgid);
 104     if (q != pg_map.pg_stat.end() &&
 105         q->second.get_version_pair() > pg_stats.get_version_pair()) {
 106       dout(15) << " had " << pgid << " from "
 107                << q->second.reported_epoch << ":"
 108                << q->second.reported_seq << dendl;
 109       continue;
 110     }
 111
 112     pending_inc.pg_stat_updates[pgid] = pg_stats;
 113   }
 114   for (auto p : stats->pool_stat) {
 115     pending_inc.pool_statfs_updates[std::make_pair(p.first, from)] = p.second;
 116   }
 117 }
 118
 119 void ClusterState::update_delta_stats()
 120 {
 121   pending_inc.stamp = ceph_clock_now();
 122   pending_inc.version = pg_map.version + 1; // to make apply_incremental happy
 123   dout(10) << " v" << pending_inc.version << dendl;
 124
 125   dout(30) << " pg_map before:\n";
 126   JSONFormatter jf(true);
 127   jf.dump_object("pg_map", pg_map);
 128   jf.flush(*_dout);
 129   *_dout << dendl;
 130   dout(30) << " incremental:\n";
 131   JSONFormatter jf(true);
 132   jf.dump_object("pending_inc", pending_inc);
 133   jf.flush(*_dout);
 134   *_dout << dendl;
 135   pg_map.apply_incremental(g_ceph_context, pending_inc);
 136   pending_inc = PGMap::Incremental();
 137 }
 138
 139 void ClusterState::notify_osdmap(const OSDMap &osd_map)
 140 {
 141   assert(ceph_mutex_is_locked(lock));
 142
 143   pending_inc.stamp = ceph_clock_now();
 144   pending_inc.version = pg_map.version + 1; // to make apply_incremental happy
 145   dout(10) << " v" << pending_inc.version << dendl;
 146
 147   PGMapUpdater::check_osd_map(g_ceph_context, osd_map, pg_map, &pending_inc);
 148
 149   // update our list of pools that exist, so that we can filter pg_map updates
 150   // in synchrony with this OSDMap.
 151   existing_pools.clear();
 152   for (auto& p : osd_map.get_pools()) {
 153     existing_pools[p.first] = p.second.get_pg_num();
 154   }
 155
 156   // brute force this for now (don't bother being clever by only
 157   // checking osds that went up/down)
 158   set<int> need_check_down_pg_osds;
 159   PGMapUpdater::check_down_pgs(osd_map, pg_map, true,
 160                                need_check_down_pg_osds, &pending_inc);
 161
 162   dout(30) << " pg_map before:\n";
 163   JSONFormatter jf(true);
 164   jf.dump_object("pg_map", pg_map);
 165   jf.flush(*_dout);
 166   *_dout << dendl;
 167   dout(30) << " incremental:\n";
 168   JSONFormatter jf(true);
 169   jf.dump_object("pending_inc", pending_inc);
 170   jf.flush(*_dout);
 171   *_dout << dendl;
 172
 173   pg_map.apply_incremental(g_ceph_context, pending_inc);
 174   pending_inc = PGMap::Incremental();
 175   // TODO: Complete the separation of PG state handling so
 176   // that a cut-down set of functionality remains in PGMonitor
 177   // while the full-blown PGMap lives only here.
 178 }
 179
 180 class ClusterSocketHook : public AdminSocketHook {
 181   ClusterState *cluster_state;
 182 public:
 183   explicit ClusterSocketHook(ClusterState *o) : cluster_state(o) {}
 184   int call(std::string_view admin_command, const cmdmap_t& cmdmap,
 185            Formatter *f,
 186            std::ostream& errss,
 187            bufferlist& out) override {
 188     stringstream outss;
 189     int r = 0;
 190     try {
 191       r = cluster_state->asok_command(admin_command, cmdmap, f, outss);
 192       out.append(outss);
 193     } catch (const TOPNSPC::common::bad_cmd_get& e) {
 194       errss << e.what();
 195       r = -EINVAL;
 196     }
 197     return r;
 198   }
 199 };
 200
 201 void ClusterState::final_init()
 202 {
 203   AdminSocket *admin_socket = g_ceph_context->get_admin_socket();
 204   asok_hook = new ClusterSocketHook(this);
 205   int r = admin_socket->register_command(
 206     "dump_osd_network name=value,type=CephInt,req=false", asok_hook,
 207     "Dump osd heartbeat network ping times");
 208   ceph_assert(r == 0);
 209 }
 210
 211 void ClusterState::shutdown()
 212 {
 213   // unregister commands
 214   g_ceph_context->get_admin_socket()->unregister_commands(asok_hook);
 215   delete asok_hook;
 216   asok_hook = NULL;
 217 }
 218
 219 bool ClusterState::asok_command(
 220   std::string_view admin_command,
 221   const cmdmap_t& cmdmap,
 222   Formatter *f,
 223   ostream& ss)
 224 {
 225   std::lock_guard l(lock);
 226   if (admin_command == "dump_osd_network") {
 227     int64_t value = 0;
 228     // Default to health warning level if nothing specified
 229     if (!(TOPNSPC::common::cmd_getval(cmdmap, "value", value))) {
 230       // Convert milliseconds to microseconds
 231       value = static_cast<int64_t>(g_ceph_context->_conf.get_val<double>("mon_warn_on_slow_ping_time")) * 1000;
 232       if (value == 0) {
 233         double ratio = g_conf().get_val<double>("mon_warn_on_slow_ping_ratio");
 234         value = g_conf().get_val<int64_t>("osd_heartbeat_grace");
 235         value *= 1000000 * ratio; // Seconds of grace to microseconds at ratio
 236       }
 237     } else {
 238       // Convert user input to microseconds
 239       value *= 1000;
 240     }
 241     if (value < 0)
 242       value = 0;
 243
 244     struct mgr_ping_time_t {
 245       uint32_t pingtime;
 246       int from;
 247       int to;
 248       bool back;
 249       std::array<uint32_t,3> times;
 250       std::array<uint32_t,3> min;
 251       std::array<uint32_t,3> max;
 252       uint32_t last;
 253       uint32_t last_update;
 254
 255       bool operator<(const mgr_ping_time_t& rhs) const {
 256         if (pingtime < rhs.pingtime)
 257           return true;
 258         if (pingtime > rhs.pingtime)
 259           return false;
 260         if (from < rhs.from)
 261           return true;
 262         if (from > rhs.from)
 263           return false;
 264         if (to < rhs.to)
 265           return true;
 266         if (to > rhs.to)
 267           return false;
 268         return back;
 269       }
 270     };
 271
 272     set<mgr_ping_time_t> sorted;
 273     utime_t now = ceph_clock_now();
 274     for (auto i : pg_map.osd_stat) {
 275       for (auto j : i.second.hb_pingtime) {
 276
 277         if (j.second.last_update == 0)
 278           continue;
 279         auto stale_time = g_ceph_context->_conf.get_val<int64_t>("osd_mon_heartbeat_stat_stale");
 280         if (now.sec() - j.second.last_update > stale_time) {
 281           dout(20) << __func__ << " time out heartbeat for osd " << i.first
 282                    << " last_update " << j.second.last_update << dendl;
 283            continue;
 284         }
 285         mgr_ping_time_t item;
 286         item.pingtime = std::max(j.second.back_pingtime[0], j.second.back_pingtime[1]);
 287         item.pingtime = std::max(item.pingtime, j.second.back_pingtime[2]);
 288         if (!value || item.pingtime >= value) {
 289           item.from = i.first;
 290           item.to = j.first;
 291           item.times[0] = j.second.back_pingtime[0];
 292           item.times[1] = j.second.back_pingtime[1];
 293           item.times[2] = j.second.back_pingtime[2];
 294           item.min[0] = j.second.back_min[0];
 295           item.min[1] = j.second.back_min[1];
 296           item.min[2] = j.second.back_min[2];
 297           item.max[0] = j.second.back_max[0];
 298           item.max[1] = j.second.back_max[1];
 299           item.max[2] = j.second.back_max[2];
 300           item.last = j.second.back_last;
 301           item.back = true;
 302           item.last_update = j.second.last_update;
 303           sorted.emplace(item);
 304         }
 305
 306         if (j.second.front_last == 0)
 307           continue;
 308         item.pingtime = std::max(j.second.front_pingtime[0], j.second.front_pingtime[1]);
 309         item.pingtime = std::max(item.pingtime, j.second.front_pingtime[2]);
 310         if (!value || item.pingtime >= value) {
 311           item.from = i.first;
 312           item.to = j.first;
 313           item.times[0] = j.second.front_pingtime[0];
 314           item.times[1] = j.second.front_pingtime[1];
 315           item.times[2] = j.second.front_pingtime[2];
 316           item.min[0] = j.second.front_min[0];
 317           item.min[1] = j.second.front_min[1];
 318           item.min[2] = j.second.front_min[2];
 319           item.max[0] = j.second.front_max[0];
 320           item.max[1] = j.second.front_max[1];
 321           item.max[2] = j.second.front_max[2];
 322           item.last = j.second.front_last;
 323           item.back = false;
 324           item.last_update = j.second.last_update;
 325           sorted.emplace(item);
 326         }
 327       }
 328     }
 329
 330     // Network ping times (1min 5min 15min)
 331     f->open_object_section("network_ping_times");
 332     f->dump_int("threshold", value / 1000);
 333     f->open_array_section("entries");
 334     for (auto &sitem : boost::adaptors::reverse(sorted)) {
 335       ceph_assert(!value || sitem.pingtime >= value);
 336
 337       f->open_object_section("entry");
 338
 339       const time_t lu(sitem.last_update);
 340       char buffer[26];
 341       string lustr(ctime_r(&lu, buffer));
 342       lustr.pop_back();   // Remove trailing \n
 343       auto stale = g_ceph_context->_conf.get_val<int64_t>("osd_heartbeat_stale");
 344       f->dump_string("last update", lustr);
 345       f->dump_bool("stale", ceph_clock_now().sec() - sitem.last_update > stale);
 346       f->dump_int("from osd", sitem.from);
 347       f->dump_int("to osd", sitem.to);
 348       f->dump_string("interface", (sitem.back ? "back" : "front"));
 349       f->open_object_section("average");
 350       f->dump_format_unquoted("1min", "%s", fixed_u_to_string(sitem.times[0],3).c_str());
 351       f->dump_format_unquoted("5min", "%s", fixed_u_to_string(sitem.times[1],3).c_str());
 352       f->dump_format_unquoted("15min", "%s", fixed_u_to_string(sitem.times[2],3).c_str());
 353       f->close_section(); // average
 354       f->open_object_section("min");
 355       f->dump_format_unquoted("1min", "%s", fixed_u_to_string(sitem.min[0],3).c_str());
 356       f->dump_format_unquoted("5min", "%s", fixed_u_to_string(sitem.min[1],3).c_str());
 357       f->dump_format_unquoted("15min", "%s", fixed_u_to_string(sitem.min[2],3).c_str());
 358       f->close_section(); // min
 359       f->open_object_section("max");
 360       f->dump_format_unquoted("1min", "%s", fixed_u_to_string(sitem.max[0],3).c_str());
 361       f->dump_format_unquoted("5min", "%s", fixed_u_to_string(sitem.max[1],3).c_str());
 362       f->dump_format_unquoted("15min", "%s", fixed_u_to_string(sitem.max[2],3).c_str());
 363       f->close_section(); // max
 364       f->dump_format_unquoted("last", "%s", fixed_u_to_string(sitem.last,3).c_str());
 365       f->close_section(); // entry
 366     }
 367     f->close_section(); // entries
 368     f->close_section(); // network_ping_times
 369   } else {
 370     ceph_abort_msg("broken asok registration");
 371   }
 372   return true;
 373 }