ceph/src/mds/RecoveryQueue.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) 2004-2006 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 #include "CInode.h"
  16 #include "MDCache.h"
  17 #include "MDSRank.h"
  18 #include "Locker.h"
  19 #include "osdc/Filer.h"
  20
  21 #include "RecoveryQueue.h"
  22
  23 #define dout_context g_ceph_context
  24 #define dout_subsys ceph_subsys_mds
  25 #undef dout_prefix
  26 #define dout_prefix *_dout << "mds." << mds->get_nodeid() << " RecoveryQueue::" << __func__ << " "
  27
  28 class C_MDC_Recover : public MDSIOContextBase {
  29 protected:
  30   RecoveryQueue *rq;
  31   CInode *in;
  32   void finish(int r) override {
  33     rq->_recovered(in, r, size, mtime);
  34   }
  35
  36   MDSRank *get_mds() override {
  37     return rq->mds;
  38   }
  39
  40 public:
  41   uint64_t size;
  42   utime_t mtime;
  43
  44   C_MDC_Recover(RecoveryQueue *rq_, CInode *i) :
  45     MDSIOContextBase(false), rq(rq_), in(i), size(0) {
  46     assert(rq != NULL);
  47   }
  48   void print(ostream& out) const override {
  49     out << "file_recover(" << in->ino() << ")";
  50   }
  51 };
  52
  53
  54 RecoveryQueue::RecoveryQueue(MDSRank *mds_) :
  55   file_recover_queue(member_offset(CInode, item_dirty_dirfrag_dir)),
  56   file_recover_queue_front(member_offset(CInode, item_dirty_dirfrag_nest)),
  57   mds(mds_), logger(NULL), filer(mds_->objecter, mds_->finisher)
  58 { }
  59
  60
  61 /**
  62  * Progress the queue.  Call this after enqueuing something or on
  63  * completion of something.
  64  */
  65 void RecoveryQueue::advance()
  66 {
  67   dout(10) << file_recover_queue_size << " queued, "
  68            << file_recover_queue_front_size << " prioritized, "
  69            << file_recovering.size() << " recovering" << dendl;
  70
  71   while (file_recovering.size() < g_conf->mds_max_file_recover) {
  72     if (!file_recover_queue_front.empty()) {
  73       CInode *in = file_recover_queue_front.front();
  74       in->item_recover_queue_front.remove_myself();
  75       file_recover_queue_front_size--;
  76       _start(in);
  77     } else if (!file_recover_queue.empty()) {
  78       CInode *in = file_recover_queue.front();
  79       in->item_recover_queue.remove_myself();
  80       file_recover_queue_size--;
  81       _start(in);
  82     } else {
  83       break;
  84     }
  85   }
  86
  87   logger->set(l_mdc_num_recovering_processing, file_recovering.size());
  88   logger->set(l_mdc_num_recovering_enqueued, file_recover_queue_size + file_recover_queue_front_size);
  89   logger->set(l_mdc_num_recovering_prioritized, file_recover_queue_front_size);
  90 }
  91
  92 void RecoveryQueue::_start(CInode *in)
  93 {
  94   auto pi = in->get_projected_inode();
  95
  96   // blech
  97   if (pi->client_ranges.size() && !pi->get_max_size()) {
  98     mds->clog->warn() << "bad client_range " << pi->client_ranges
  99                       << " on ino " << pi->ino;
 100   }
 101
 102   auto p = file_recovering.find(in);
 103   if (pi->client_ranges.size() && pi->get_max_size()) {
 104     dout(10) << "starting " << in->inode.size << " " << pi->client_ranges
 105              << " " << *in << dendl;
 106     if (p == file_recovering.end()) {
 107       file_recovering.insert(make_pair(in, false));
 108
 109       C_MDC_Recover *fin = new C_MDC_Recover(this, in);
 110       filer.probe(in->inode.ino, &in->inode.layout, in->last,
 111                   pi->get_max_size(), &fin->size, &fin->mtime, false,
 112                   0, fin);
 113     } else {
 114       p->second = true;
 115       dout(10) << "already working on " << *in << ", set need_restart flag" << dendl;
 116     }
 117   } else {
 118     dout(10) << "skipping " << in->inode.size << " " << *in << dendl;
 119     if (p == file_recovering.end()) {
 120       in->state_clear(CInode::STATE_RECOVERING);
 121       mds->locker->eval(in, CEPH_LOCK_IFILE);
 122       in->auth_unpin(this);
 123     }
 124   }
 125 }
 126
 127 void RecoveryQueue::prioritize(CInode *in)
 128 {
 129   if (file_recovering.count(in)) {
 130     dout(10) << "already working on " << *in << dendl;
 131     return;
 132   }
 133
 134   if (!in->item_recover_queue_front.is_on_list()) {
 135     dout(20) << *in << dendl;
 136
 137     assert(in->item_recover_queue.is_on_list());
 138     in->item_recover_queue.remove_myself();
 139     file_recover_queue_size--;
 140
 141     file_recover_queue_front.push_back(&in->item_recover_queue_front);
 142
 143     file_recover_queue_front_size++;
 144     logger->set(l_mdc_num_recovering_prioritized, file_recover_queue_front_size);
 145     return;
 146   }
 147
 148   dout(10) << "not queued " << *in << dendl;
 149 }
 150
 151 static bool _is_in_any_recover_queue(CInode *in)
 152 {
 153   return in->item_recover_queue.is_on_list() ||
 154          in->item_recover_queue_front.is_on_list();
 155 }
 156
 157 /**
 158  * Given an authoritative inode which is in the cache,
 159  * enqueue it for recovery.
 160  */
 161 void RecoveryQueue::enqueue(CInode *in)
 162 {
 163   dout(15) << "RecoveryQueue::enqueue " << *in << dendl;
 164   assert(logger);  // Caller should have done set_logger before using me
 165   assert(in->is_auth());
 166
 167   in->state_clear(CInode::STATE_NEEDSRECOVER);
 168   if (!in->state_test(CInode::STATE_RECOVERING)) {
 169     in->state_set(CInode::STATE_RECOVERING);
 170     in->auth_pin(this);
 171     logger->inc(l_mdc_recovery_started);
 172   }
 173
 174   if (!_is_in_any_recover_queue(in)) {
 175     file_recover_queue.push_back(&in->item_recover_queue);
 176     file_recover_queue_size++;
 177     logger->set(l_mdc_num_recovering_enqueued, file_recover_queue_size + file_recover_queue_front_size);
 178   }
 179 }
 180
 181
 182 /**
 183  * Call back on completion of Filer probe on an inode.
 184  */
 185 void RecoveryQueue::_recovered(CInode *in, int r, uint64_t size, utime_t mtime)
 186 {
 187   dout(10) << "_recovered r=" << r << " size=" << size << " mtime=" << mtime
 188            << " for " << *in << dendl;
 189
 190   if (r != 0) {
 191     dout(0) << "recovery error! " << r << dendl;
 192     if (r == -EBLACKLISTED) {
 193       mds->respawn();
 194       return;
 195     } else {
 196       // Something wrong on the OSD side trying to recover the size
 197       // of this inode.  In principle we could record this as a piece
 198       // of per-inode damage, but it's actually more likely that
 199       // this indicates something wrong with the MDS (like maybe
 200       // it has the wrong auth caps?)
 201       mds->clog->error() << " OSD read error while recovering size for inode 0x"
 202                          << std::hex << in->ino() << std::dec;
 203       mds->damaged();
 204     }
 205   }
 206
 207   auto p = file_recovering.find(in);
 208   assert(p != file_recovering.end());
 209   bool restart = p->second;
 210   file_recovering.erase(p);
 211
 212   logger->set(l_mdc_num_recovering_processing, file_recovering.size());
 213   logger->inc(l_mdc_recovery_completed);
 214   in->state_clear(CInode::STATE_RECOVERING);
 215
 216   if (restart) {
 217     if (in->item_recover_queue.is_on_list()) {
 218       in->item_recover_queue.remove_myself();
 219       file_recover_queue_size--;
 220     }
 221     if (in->item_recover_queue_front.is_on_list()) {
 222       in->item_recover_queue_front.remove_myself();
 223       file_recover_queue_front_size--;
 224     }
 225     logger->set(l_mdc_num_recovering_enqueued, file_recover_queue_size + file_recover_queue_front_size);
 226     logger->set(l_mdc_num_recovering_prioritized, file_recover_queue_front_size);
 227     _start(in);
 228   } else if (!_is_in_any_recover_queue(in)) {
 229     // journal
 230     mds->locker->check_inode_max_size(in, true, 0,  size, mtime);
 231     mds->locker->eval(in, CEPH_LOCK_IFILE);
 232     in->auth_unpin(this);
 233   }
 234
 235   advance();
 236 }
 237