]> git.proxmox.com Git - ceph.git/blame - ceph/src/messages/MMDSBeacon.h
bump version to 18.2.2-pve1
[ceph.git] / ceph / src / messages / MMDSBeacon.h
CommitLineData
7c673cae
FG
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#ifndef CEPH_MMDSBEACON_H
16#define CEPH_MMDSBEACON_H
17
11fdf7f2 18#include <string_view>
94b18763 19
11fdf7f2 20#include "msg/Message.h"
7c673cae
FG
21#include "messages/PaxosServiceMessage.h"
22
23#include "include/types.h"
24
25#include "mds/MDSMap.h"
26
27
28
29/**
30 * Unique ID for each type of metric we can send to the mon, so that if the mon
31 * knows about the IDs then it can implement special behaviour for certain
32 * messages.
33 */
34enum mds_metric_t {
35 MDS_HEALTH_NULL = 0,
36 MDS_HEALTH_TRIM,
37 MDS_HEALTH_CLIENT_RECALL,
38 MDS_HEALTH_CLIENT_LATE_RELEASE,
39 MDS_HEALTH_CLIENT_RECALL_MANY,
40 MDS_HEALTH_CLIENT_LATE_RELEASE_MANY,
41 MDS_HEALTH_CLIENT_OLDEST_TID,
42 MDS_HEALTH_CLIENT_OLDEST_TID_MANY,
43 MDS_HEALTH_DAMAGE,
44 MDS_HEALTH_READ_ONLY,
45 MDS_HEALTH_SLOW_REQUEST,
91327a77
AA
46 MDS_HEALTH_CACHE_OVERSIZED,
47 MDS_HEALTH_SLOW_METADATA_IO,
2a845540 48 MDS_HEALTH_DUMMY, // not a real health warning, for testing
7c673cae
FG
49};
50
11fdf7f2 51inline const char *mds_metric_name(mds_metric_t m)
224ce89b
WB
52{
53 switch (m) {
54 case MDS_HEALTH_TRIM: return "MDS_TRIM";
55 case MDS_HEALTH_CLIENT_RECALL: return "MDS_CLIENT_RECALL";
56 case MDS_HEALTH_CLIENT_LATE_RELEASE: return "MDS_CLIENT_LATE_RELEASE";
57 case MDS_HEALTH_CLIENT_RECALL_MANY: return "MDS_CLIENT_RECALL_MANY";
58 case MDS_HEALTH_CLIENT_LATE_RELEASE_MANY: return "MDS_CLIENT_LATE_RELEASE_MANY";
59 case MDS_HEALTH_CLIENT_OLDEST_TID: return "MDS_CLIENT_OLDEST_TID";
60 case MDS_HEALTH_CLIENT_OLDEST_TID_MANY: return "MDS_CLIENT_OLDEST_TID_MANY";
61 case MDS_HEALTH_DAMAGE: return "MDS_DAMAGE";
62 case MDS_HEALTH_READ_ONLY: return "MDS_READ_ONLY";
63 case MDS_HEALTH_SLOW_REQUEST: return "MDS_SLOW_REQUEST";
64 case MDS_HEALTH_CACHE_OVERSIZED: return "MDS_CACHE_OVERSIZED";
91327a77 65 case MDS_HEALTH_SLOW_METADATA_IO: return "MDS_SLOW_METADATA_IO";
2a845540 66 case MDS_HEALTH_DUMMY: return "MDS_DUMMY";
224ce89b
WB
67 default:
68 return "???";
69 }
70}
71
11fdf7f2 72inline const char *mds_metric_summary(mds_metric_t m)
224ce89b
WB
73{
74 switch (m) {
75 case MDS_HEALTH_TRIM:
76 return "%num% MDSs behind on trimming";
77 case MDS_HEALTH_CLIENT_RECALL:
78 return "%num% clients failing to respond to cache pressure";
79 case MDS_HEALTH_CLIENT_LATE_RELEASE:
80 return "%num% clients failing to respond to capability release";
81 case MDS_HEALTH_CLIENT_RECALL_MANY:
82 return "%num% MDSs have many clients failing to respond to cache pressure";
83 case MDS_HEALTH_CLIENT_LATE_RELEASE_MANY:
84 return "%num% MDSs have many clients failing to respond to capability "
85 "release";
86 case MDS_HEALTH_CLIENT_OLDEST_TID:
87 return "%num% clients failing to advance oldest client/flush tid";
88 case MDS_HEALTH_CLIENT_OLDEST_TID_MANY:
89 return "%num% MDSs have clients failing to advance oldest client/flush tid";
90 case MDS_HEALTH_DAMAGE:
91 return "%num% MDSs report damaged metadata";
92 case MDS_HEALTH_READ_ONLY:
93 return "%num% MDSs are read only";
94 case MDS_HEALTH_SLOW_REQUEST:
95 return "%num% MDSs report slow requests";
96 case MDS_HEALTH_CACHE_OVERSIZED:
97 return "%num% MDSs report oversized cache";
91327a77
AA
98 case MDS_HEALTH_SLOW_METADATA_IO:
99 return "%num% MDSs report slow metadata IOs";
224ce89b
WB
100 default:
101 return "???";
102 }
103}
104
7c673cae
FG
105/**
106 * This structure is designed to allow some flexibility in how we emit health
107 * complaints, such that:
108 * - The mon doesn't have to have foreknowledge of all possible metrics: we can
109 * implement new messages in the MDS and have the mon pass them through to the user
110 * (enables us to do complex checks inside the MDS, and allows mon to be older version
111 * than MDS)
112 * - The mon has enough information to perform reductions on some types of metric, for
113 * example complaints about the same client from multiple MDSs where we might want
114 * to reduce three "client X is stale on MDS y" metrics into one "client X is stale
115 * on 3 MDSs" message.
116 */
117struct MDSHealthMetric
118{
119 mds_metric_t type;
120 health_status_t sev;
121 std::string message;
122 std::map<std::string, std::string> metadata;
123
f67539c2 124 void encode(ceph::buffer::list& bl) const {
7c673cae 125 ENCODE_START(1, 1, bl);
11fdf7f2
TL
126 ceph_assert(type != MDS_HEALTH_NULL);
127 encode((uint16_t)type, bl);
128 encode((uint8_t)sev, bl);
129 encode(message, bl);
130 encode(metadata, bl);
7c673cae
FG
131 ENCODE_FINISH(bl);
132 }
133
f67539c2 134 void decode(ceph::buffer::list::const_iterator& bl) {
7c673cae 135 DECODE_START(1, bl);
f91f0fd5
TL
136 uint16_t raw_type;
137 decode(raw_type, bl);
138 type = (mds_metric_t)raw_type;
11fdf7f2 139 ceph_assert(type != MDS_HEALTH_NULL);
f91f0fd5
TL
140 uint8_t raw_sev;
141 decode(raw_sev, bl);
142 sev = (health_status_t)raw_sev;
11fdf7f2
TL
143 decode(message, bl);
144 decode(metadata, bl);
7c673cae
FG
145 DECODE_FINISH(bl);
146 }
147
148 bool operator==(MDSHealthMetric const &other) const
149 {
150 return (type == other.type && sev == other.sev && message == other.message);
151 }
152
153 MDSHealthMetric() : type(MDS_HEALTH_NULL), sev(HEALTH_OK) {}
11fdf7f2 154 MDSHealthMetric(mds_metric_t type_, health_status_t sev_, std::string_view message_)
7c673cae
FG
155 : type(type_), sev(sev_), message(message_) {}
156};
157WRITE_CLASS_ENCODER(MDSHealthMetric)
158
159
160/**
161 * Health metrics send by the MDS to the mon, so that the mon can generate
162 * user friendly warnings about undesirable states.
163 */
164struct MDSHealth
165{
9f95a23c 166 std::vector<MDSHealthMetric> metrics;
7c673cae 167
f67539c2 168 void encode(ceph::buffer::list& bl) const {
7c673cae 169 ENCODE_START(1, 1, bl);
11fdf7f2 170 encode(metrics, bl);
7c673cae
FG
171 ENCODE_FINISH(bl);
172 }
173
f67539c2 174 void decode(ceph::buffer::list::const_iterator& bl) {
7c673cae 175 DECODE_START(1, bl);
11fdf7f2 176 decode(metrics, bl);
7c673cae
FG
177 DECODE_FINISH(bl);
178 }
179
180 bool operator==(MDSHealth const &other) const
181 {
182 return metrics == other.metrics;
183 }
184};
185WRITE_CLASS_ENCODER(MDSHealth)
186
187
f67539c2 188class MMDSBeacon final : public PaxosServiceMessage {
11fdf7f2 189private:
7c673cae 190
9f95a23c 191 static constexpr int HEAD_VERSION = 8;
11fdf7f2 192 static constexpr int COMPAT_VERSION = 6;
7c673cae
FG
193
194 uuid_d fsid;
11fdf7f2 195 mds_gid_t global_id = MDS_GID_NONE;
f67539c2 196 std::string name;
7c673cae 197
11fdf7f2
TL
198 MDSMap::DaemonState state = MDSMap::STATE_NULL;
199 version_t seq = 0;
7c673cae
FG
200
201 CompatSet compat;
202
203 MDSHealth health;
204
f67539c2 205 std::map<std::string, std::string> sys_info;
7c673cae 206
11fdf7f2 207 uint64_t mds_features = 0;
7c673cae 208
f67539c2 209 std::string fs;
9f95a23c 210
11fdf7f2 211protected:
9f95a23c 212 MMDSBeacon() : PaxosServiceMessage(MSG_MDS_BEACON, 0, HEAD_VERSION, COMPAT_VERSION)
11fdf7f2 213 {
91327a77
AA
214 set_priority(CEPH_MSG_PRIO_HIGH);
215 }
f67539c2
TL
216 MMDSBeacon(const uuid_d &f, mds_gid_t g, const std::string& n, epoch_t les,
217 MDSMap::DaemonState st, version_t se, uint64_t feat) :
9f95a23c 218 PaxosServiceMessage(MSG_MDS_BEACON, les, HEAD_VERSION, COMPAT_VERSION),
7c673cae 219 fsid(f), global_id(g), name(n), state(st), seq(se),
11fdf7f2 220 mds_features(feat) {
91327a77 221 set_priority(CEPH_MSG_PRIO_HIGH);
7c673cae 222 }
f67539c2 223 ~MMDSBeacon() final {}
7c673cae
FG
224
225public:
11fdf7f2
TL
226 const uuid_d& get_fsid() const { return fsid; }
227 mds_gid_t get_global_id() const { return global_id; }
f67539c2 228 const std::string& get_name() const { return name; }
11fdf7f2
TL
229 epoch_t get_last_epoch_seen() const { return version; }
230 MDSMap::DaemonState get_state() const { return state; }
231 version_t get_seq() const { return seq; }
232 std::string_view get_type_name() const override { return "mdsbeacon"; }
7c673cae
FG
233 uint64_t get_mds_features() const { return mds_features; }
234
235 CompatSet const& get_compat() const { return compat; }
236 void set_compat(const CompatSet& c) { compat = c; }
237
238 MDSHealth const& get_health() const { return health; }
239 void set_health(const MDSHealth &h) { health = h; }
240
f67539c2 241 const std::string& get_fs() const { return fs; }
9f95a23c
TL
242 void set_fs(std::string_view s) { fs = s; }
243
f67539c2
TL
244 const std::map<std::string, std::string>& get_sys_info() const { return sys_info; }
245 void set_sys_info(const std::map<std::string, std::string>& i) { sys_info = i; }
7c673cae 246
f67539c2 247 void print(std::ostream& out) const override {
9f95a23c
TL
248 out << "mdsbeacon(" << global_id << "/" << name
249 << " " << ceph_mds_state_name(state);
250 if (fs.size()) {
251 out << " fs=" << fs;
252 }
253 out << " seq=" << seq << " v" << version << ")";
7c673cae
FG
254 }
255
256 void encode_payload(uint64_t features) override {
11fdf7f2 257 using ceph::encode;
7c673cae 258 paxos_encode();
11fdf7f2
TL
259 encode(fsid, payload);
260 encode(global_id, payload);
261 encode((__u32)state, payload);
262 encode(seq, payload);
263 encode(name, payload);
264 encode(MDS_RANK_NONE, payload);
265 encode(std::string(), payload);
266 encode(compat, payload);
267 encode(health, payload);
7c673cae 268 if (state == MDSMap::STATE_BOOT) {
11fdf7f2 269 encode(sys_info, payload);
7c673cae 270 }
11fdf7f2
TL
271 encode(mds_features, payload);
272 encode(FS_CLUSTER_ID_NONE, payload);
273 encode(false, payload);
9f95a23c 274 encode(fs, payload);
7c673cae
FG
275 }
276 void decode_payload() override {
11fdf7f2
TL
277 using ceph::decode;
278 auto p = payload.cbegin();
7c673cae 279 paxos_decode(p);
11fdf7f2
TL
280 decode(fsid, p);
281 decode(global_id, p);
f91f0fd5
TL
282 __u32 raw_state;
283 decode(raw_state, p);
284 state = (MDSMap::DaemonState)raw_state;
11fdf7f2
TL
285 decode(seq, p);
286 decode(name, p);
287 {
288 mds_rank_t standby_for_rank;
289 decode(standby_for_rank, p);
290 }
291 {
292 std::string standby_for_name;
293 decode(standby_for_name, p);
294 }
295 decode(compat, p);
296 decode(health, p);
7c673cae 297 if (state == MDSMap::STATE_BOOT) {
11fdf7f2
TL
298 decode(sys_info, p);
299 }
300 decode(mds_features, p);
301 {
302 fs_cluster_id_t standby_for_fscid;
303 decode(standby_for_fscid, p);
7c673cae 304 }
7c673cae 305 if (header.version >= 7) {
11fdf7f2
TL
306 bool standby_replay;
307 decode(standby_replay, p);
7c673cae
FG
308 }
309
310 if (header.version < 7 && state == MDSMap::STATE_STANDBY_REPLAY) {
311 // Old MDS daemons request the state, instead of explicitly
312 // advertising that they are configured as a replay daemon.
7c673cae
FG
313 state = MDSMap::STATE_STANDBY;
314 }
9f95a23c
TL
315 if (header.version >= 8) {
316 decode(fs, p);
317 }
7c673cae 318 }
9f95a23c
TL
319private:
320 template<class T, typename... Args>
321 friend boost::intrusive_ptr<T> ceph::make_message(Args&&... args);
20effc67
TL
322 template<class T, typename... Args>
323 friend MURef<T> crimson::make_message(Args&&... args);
7c673cae
FG
324};
325
326#endif