]> git.proxmox.com Git - ceph.git/blob - ceph/src/mon/DataHealthService.cc
update sources to v12.1.1
[ceph.git] / 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62 // outdated stats.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 list<pair<health_status_t,string> >& summary,
69 list<pair<health_status_t,string> > *detail)
70 {
71 dout(10) << __func__ << dendl;
72 for (map<entity_inst_t,DataStats>::iterator it = stats.begin();
73 it != stats.end(); ++it) {
74 string mon_name = mon->monmap->get_name(it->first.addr);
75 DataStats& stats = it->second;
76
77 health_status_t health_status = HEALTH_OK;
78 string health_detail;
79 if (stats.fs_stats.avail_percent <= g_conf->mon_data_avail_crit) {
80 health_status = HEALTH_ERR;
81 health_detail = "low disk space, shutdown imminent";
82 } else if (stats.fs_stats.avail_percent <= g_conf->mon_data_avail_warn) {
83 health_status = HEALTH_WARN;
84 health_detail = "low disk space";
85 }
86
87 if (stats.store_stats.bytes_total >= g_conf->mon_data_size_warn) {
88 if (health_status > HEALTH_WARN)
89 health_status = HEALTH_WARN;
90 if (!health_detail.empty())
91 health_detail.append("; ");
92 stringstream ss;
93 ss << "store is getting too big! "
94 << prettybyte_t(stats.store_stats.bytes_total)
95 << " >= " << prettybyte_t(g_conf->mon_data_size_warn);
96 health_detail.append(ss.str());
97 }
98
99 if (health_status != HEALTH_OK) {
100 stringstream ss;
101 ss << "mon." << mon_name << " " << health_detail;
102 summary.push_back(make_pair(health_status, ss.str()));
103 ss << " -- " << stats.fs_stats.avail_percent << "% avail";
104 if (detail)
105 detail->push_back(make_pair(health_status, ss.str()));
106 }
107 }
108 }
109
110 int DataHealthService::update_store_stats(DataStats &ours)
111 {
112 map<string,uint64_t> extra;
113 uint64_t store_size = mon->store->get_estimated_size(extra);
114 assert(store_size > 0);
115
116 ours.store_stats.bytes_total = store_size;
117 ours.store_stats.bytes_sst = extra["sst"];
118 ours.store_stats.bytes_log = extra["log"];
119 ours.store_stats.bytes_misc = extra["misc"];
120 ours.last_update = ceph_clock_now();
121
122 return 0;
123 }
124
125
126 int DataHealthService::update_stats()
127 {
128 entity_inst_t our_inst = mon->messenger->get_myinst();
129 DataStats& ours = stats[our_inst];
130
131 int err = get_fs_stats(ours.fs_stats, g_conf->mon_data.c_str());
132 if (err < 0) {
133 derr << __func__ << " get_fs_stats error: " << cpp_strerror(err) << dendl;
134 return err;
135 }
136 dout(0) << __func__ << " avail " << ours.fs_stats.avail_percent << "%"
137 << " total " << prettybyte_t(ours.fs_stats.byte_total)
138 << ", used " << prettybyte_t(ours.fs_stats.byte_used)
139 << ", avail " << prettybyte_t(ours.fs_stats.byte_avail) << dendl;
140 ours.last_update = ceph_clock_now();
141
142 return update_store_stats(ours);
143 }
144
145 void DataHealthService::share_stats()
146 {
147 dout(10) << __func__ << dendl;
148 if (!in_quorum())
149 return;
150
151 assert(!stats.empty());
152 entity_inst_t our_inst = mon->messenger->get_myinst();
153 assert(stats.count(our_inst) > 0);
154 DataStats &ours = stats[our_inst];
155 const set<int>& quorum = mon->get_quorum();
156 for (set<int>::const_iterator it = quorum.begin();
157 it != quorum.end(); ++it) {
158 if (mon->monmap->get_name(*it) == mon->name)
159 continue;
160 entity_inst_t inst = mon->monmap->get_inst(*it);
161 MMonHealth *m = new MMonHealth(HealthService::SERVICE_HEALTH_DATA,
162 MMonHealth::OP_TELL);
163 m->data_stats = ours;
164 dout(20) << __func__ << " send " << *m << " to " << inst << dendl;
165 mon->messenger->send_message(m, inst);
166 }
167 }
168
169 void DataHealthService::service_tick()
170 {
171 dout(10) << __func__ << dendl;
172
173 int err = update_stats();
174 if (err < 0) {
175 derr << "something went wrong obtaining our disk stats: "
176 << cpp_strerror(err) << dendl;
177 force_shutdown();
178 return;
179 }
180 if (in_quorum())
181 share_stats();
182
183 DataStats &ours = stats[mon->messenger->get_myinst()];
184
185 if (ours.fs_stats.avail_percent <= g_conf->mon_data_avail_crit) {
186 derr << "reached critical levels of available space on local monitor storage"
187 << " -- shutdown!" << dendl;
188 force_shutdown();
189 return;
190 }
191
192 // we must backoff these warnings, and track how much data is being
193 // consumed in-between reports to assess if it's worth to log this info,
194 // otherwise we may very well contribute to the consumption of the
195 // already low available disk space.
196 if (ours.fs_stats.avail_percent <= g_conf->mon_data_avail_warn) {
197 if (ours.fs_stats.avail_percent != last_warned_percent)
198 mon->clog->warn()
199 << "reached concerning levels of available space on local monitor storage"
200 << " (" << ours.fs_stats.avail_percent << "% free)";
201 last_warned_percent = ours.fs_stats.avail_percent;
202 } else {
203 last_warned_percent = 0;
204 }
205 }
206
207 void DataHealthService::handle_tell(MonOpRequestRef op)
208 {
209 op->mark_event("datahealth:handle_tell");
210 MMonHealth *m = static_cast<MMonHealth*>(op->get_req());
211 dout(10) << __func__ << " " << *m << dendl;
212 assert(m->get_service_op() == MMonHealth::OP_TELL);
213
214 stats[m->get_source_inst()] = m->data_stats;
215 }
216
217 bool DataHealthService::service_dispatch_op(MonOpRequestRef op)
218 {
219 op->mark_event("datahealth:service_dispatch_op");
220 MMonHealth *m = static_cast<MMonHealth*>(op->get_req());
221 dout(10) << __func__ << " " << *m << dendl;
222 assert(m->get_service_type() == get_type());
223 if (!in_quorum()) {
224 dout(1) << __func__ << " not in quorum -- drop message" << dendl;
225 return false;
226 }
227
228 switch (m->service_op) {
229 case MMonHealth::OP_TELL:
230 // someone is telling us their stats
231 handle_tell(op);
232 break;
233 default:
234 dout(0) << __func__ << " unknown op " << m->service_op << dendl;
235 assert(0 == "Unknown service op");
236 break;
237 }
238 return true;
239 }