]>
Commit | Line | Data |
---|---|---|
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) 2016 John Spray <john.spray@redhat.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 | ||
224ce89b WB |
14 | #include <boost/tokenizer.hpp> |
15 | ||
7c673cae FG |
16 | #include "messages/MMgrBeacon.h" |
17 | #include "messages/MMgrMap.h" | |
18 | #include "messages/MMgrDigest.h" | |
19 | ||
7c673cae FG |
20 | #include "include/stringify.h" |
21 | #include "mgr/MgrContext.h" | |
c07f9fc5 | 22 | #include "mgr/mgr_commands.h" |
7c673cae | 23 | #include "OSDMonitor.h" |
11fdf7f2 | 24 | #include "ConfigMonitor.h" |
7c673cae FG |
25 | |
26 | #include "MgrMonitor.h" | |
27 | ||
c07f9fc5 FG |
28 | #define MGR_METADATA_PREFIX "mgr_metadata" |
29 | ||
7c673cae FG |
30 | #define dout_subsys ceph_subsys_mon |
31 | #undef dout_prefix | |
32 | #define dout_prefix _prefix(_dout, mon, map) | |
33 | static ostream& _prefix(std::ostream *_dout, Monitor *mon, | |
34 | const MgrMap& mgrmap) { | |
35 | return *_dout << "mon." << mon->name << "@" << mon->rank | |
36 | << "(" << mon->get_state_name() | |
37 | << ").mgr e" << mgrmap.get_epoch() << " "; | |
38 | } | |
39 | ||
11fdf7f2 TL |
40 | // the system treats always_on_modules as if they provide built-in functionality |
41 | // by ensuring that they are always enabled. | |
42 | const static std::map<uint32_t, std::set<std::string>> always_on_modules = { | |
43 | { | |
44 | CEPH_RELEASE_NAUTILUS, { | |
45 | "crash", | |
46 | "status", | |
47 | "progress", | |
48 | "balancer", | |
49 | "devicehealth", | |
50 | "orchestrator_cli", | |
51 | "volumes", | |
52 | } | |
53 | } | |
54 | }; | |
55 | ||
c07f9fc5 FG |
56 | // Prefix for mon store of active mgr's command descriptions |
57 | const static std::string command_descs_prefix = "mgr_command_descs"; | |
58 | ||
11fdf7f2 TL |
59 | const Option *MgrMonitor::find_module_option(const string& name) |
60 | { | |
61 | // we have two forms of names: "mgr/$module/$option" and | |
62 | // localized "mgr/$module/$instance/$option". normalize to the | |
63 | // former by stripping out $instance. | |
64 | string real_name; | |
65 | if (name.substr(0, 4) != "mgr/") { | |
66 | return nullptr; | |
67 | } | |
68 | auto second_slash = name.find('/', 5); | |
69 | if (second_slash == std::string::npos) { | |
70 | return nullptr; | |
71 | } | |
72 | auto third_slash = name.find('/', second_slash + 1); | |
73 | if (third_slash != std::string::npos) { | |
74 | // drop the $instance part between the second and third slash | |
75 | real_name = name.substr(0, second_slash) + name.substr(third_slash); | |
76 | } else { | |
77 | real_name = name; | |
78 | } | |
79 | auto p = mgr_module_options.find(real_name); | |
80 | if (p != mgr_module_options.end()) { | |
81 | return &p->second; | |
82 | } | |
83 | return nullptr; | |
84 | } | |
31f18b77 | 85 | |
11fdf7f2 | 86 | version_t MgrMonitor::get_trim_to() const |
b32b8144 | 87 | { |
11fdf7f2 | 88 | int64_t max = g_conf().get_val<int64_t>("mon_max_mgrmap_epochs"); |
b32b8144 FG |
89 | if (map.epoch > max) { |
90 | return map.epoch - max; | |
91 | } | |
92 | return 0; | |
93 | } | |
94 | ||
7c673cae FG |
95 | void MgrMonitor::create_initial() |
96 | { | |
3efd9988 | 97 | // Take a local copy of initial_modules for tokenizer to iterate over. |
11fdf7f2 | 98 | auto initial_modules = g_conf().get_val<std::string>("mgr_initial_modules"); |
3efd9988 | 99 | boost::tokenizer<> tok(initial_modules); |
224ce89b WB |
100 | for (auto& m : tok) { |
101 | pending_map.modules.insert(m); | |
102 | } | |
11fdf7f2 | 103 | pending_map.always_on_modules = always_on_modules; |
c07f9fc5 FG |
104 | pending_command_descs = mgr_commands; |
105 | dout(10) << __func__ << " initial modules " << pending_map.modules | |
11fdf7f2 TL |
106 | << ", always on modules " << pending_map.get_always_on_modules() |
107 | << ", " << pending_command_descs.size() << " commands" | |
c07f9fc5 | 108 | << dendl; |
7c673cae FG |
109 | } |
110 | ||
11fdf7f2 | 111 | void MgrMonitor::get_store_prefixes(std::set<string>& s) const |
3efd9988 FG |
112 | { |
113 | s.insert(service_name); | |
114 | s.insert(command_descs_prefix); | |
115 | s.insert(MGR_METADATA_PREFIX); | |
116 | } | |
117 | ||
7c673cae FG |
118 | void MgrMonitor::update_from_paxos(bool *need_bootstrap) |
119 | { | |
120 | version_t version = get_last_committed(); | |
121 | if (version != map.epoch) { | |
122 | dout(4) << "loading version " << version << dendl; | |
123 | ||
124 | bufferlist bl; | |
125 | int err = get_version(version, bl); | |
11fdf7f2 | 126 | ceph_assert(err == 0); |
7c673cae | 127 | |
c07f9fc5 FG |
128 | bool old_available = map.get_available(); |
129 | uint64_t old_gid = map.get_active_gid(); | |
130 | ||
11fdf7f2 | 131 | auto p = bl.cbegin(); |
7c673cae FG |
132 | map.decode(p); |
133 | ||
11fdf7f2 | 134 | dout(4) << "active server: " << map.active_addrs |
7c673cae FG |
135 | << "(" << map.active_gid << ")" << dendl; |
136 | ||
224ce89b WB |
137 | ever_had_active_mgr = get_value("ever_had_active_mgr"); |
138 | ||
139 | load_health(); | |
140 | ||
7c673cae FG |
141 | if (map.available) { |
142 | first_seen_inactive = utime_t(); | |
143 | } else { | |
144 | first_seen_inactive = ceph_clock_now(); | |
145 | } | |
146 | ||
147 | check_subs(); | |
c07f9fc5 FG |
148 | |
149 | if (version == 1 | |
3efd9988 FG |
150 | || command_descs.empty() |
151 | || (map.get_available() | |
152 | && (!old_available || old_gid != map.get_active_gid()))) { | |
c07f9fc5 FG |
153 | dout(4) << "mkfs or daemon transitioned to available, loading commands" |
154 | << dendl; | |
155 | bufferlist loaded_commands; | |
156 | int r = mon->store->get(command_descs_prefix, "", loaded_commands); | |
157 | if (r < 0) { | |
158 | derr << "Failed to load mgr commands: " << cpp_strerror(r) << dendl; | |
159 | } else { | |
11fdf7f2 TL |
160 | auto p = loaded_commands.cbegin(); |
161 | decode(command_descs, p); | |
162 | } | |
163 | } | |
164 | } | |
165 | ||
166 | // populate module options | |
167 | mgr_module_options.clear(); | |
168 | misc_option_strings.clear(); | |
169 | for (auto& i : map.available_modules) { | |
170 | for (auto& j : i.module_options) { | |
171 | string name = string("mgr/") + i.name + "/" + j.second.name; | |
172 | auto p = mgr_module_options.emplace( | |
173 | name, | |
174 | Option(name, static_cast<Option::type_t>(j.second.type), | |
175 | static_cast<Option::level_t>(j.second.level))); | |
176 | Option& opt = p.first->second; | |
177 | opt.set_flags(static_cast<Option::flag_t>(j.second.flags)); | |
178 | opt.set_flag(Option::FLAG_MGR); | |
179 | opt.set_description(j.second.desc.c_str()); | |
180 | opt.set_long_description(j.second.long_desc.c_str()); | |
181 | for (auto& k : j.second.tags) { | |
182 | opt.add_tag(k.c_str()); | |
183 | } | |
184 | for (auto& k : j.second.see_also) { | |
185 | if (i.module_options.count(k)) { | |
186 | // it's another module option | |
187 | misc_option_strings.push_back(string("mgr/") + i.name + "/" + k); | |
188 | opt.add_see_also(misc_option_strings.back().c_str()); | |
189 | } else { | |
190 | // it's a native option | |
191 | opt.add_see_also(k.c_str()); | |
192 | } | |
193 | } | |
194 | Option::value_t v, v2; | |
195 | std::string err; | |
196 | if (j.second.default_value.size() && | |
197 | !opt.parse_value(j.second.default_value, &v, &err)) { | |
198 | opt.set_default(v); | |
199 | } | |
200 | if (j.second.min.size() && | |
201 | j.second.max.size() && | |
202 | !opt.parse_value(j.second.min, &v, &err) && | |
203 | !opt.parse_value(j.second.max, &v2, &err)) { | |
204 | opt.set_min_max(v, v2); | |
c07f9fc5 | 205 | } |
11fdf7f2 TL |
206 | std::vector<const char *> enum_allowed; |
207 | for (auto& k : j.second.enum_allowed) { | |
208 | enum_allowed.push_back(k.c_str()); | |
209 | } | |
210 | opt.set_enum_allowed(enum_allowed); | |
c07f9fc5 | 211 | } |
7c673cae | 212 | } |
11fdf7f2 TL |
213 | // force ConfigMonitor to refresh, since it uses const Option * |
214 | // pointers into our mgr_module_options (which we just rebuilt). | |
215 | mon->configmon()->load_config(); | |
7c673cae | 216 | |
11fdf7f2 TL |
217 | if (!mon->is_init()) { |
218 | // feed our pet MgrClient, unless we are in Monitor::[pre]init() | |
219 | prime_mgr_client(); | |
220 | } | |
221 | } | |
222 | ||
223 | void MgrMonitor::prime_mgr_client() | |
224 | { | |
225 | dout(10) << __func__ << dendl; | |
7c673cae FG |
226 | mon->mgr_client.ms_dispatch(new MMgrMap(map)); |
227 | } | |
228 | ||
229 | void MgrMonitor::create_pending() | |
230 | { | |
231 | pending_map = map; | |
232 | pending_map.epoch++; | |
233 | } | |
234 | ||
224ce89b WB |
235 | health_status_t MgrMonitor::should_warn_about_mgr_down() |
236 | { | |
237 | utime_t now = ceph_clock_now(); | |
238 | // we warn if | |
239 | // - we've ever had an active mgr, or | |
240 | // - we have osds AND we've exceeded the grace period | |
241 | // which means a new mon cluster and be HEALTH_OK indefinitely as long as | |
242 | // no OSDs are ever created. | |
243 | if (ever_had_active_mgr || | |
244 | (mon->osdmon()->osdmap.get_num_osds() > 0 && | |
11fdf7f2 | 245 | now > mon->monmap->created + g_conf().get_val<int64_t>("mon_mgr_mkfs_grace"))) { |
224ce89b WB |
246 | health_status_t level = HEALTH_WARN; |
247 | if (first_seen_inactive != utime_t() && | |
11fdf7f2 | 248 | now - first_seen_inactive > g_conf().get_val<int64_t>("mon_mgr_inactive_grace")) { |
224ce89b WB |
249 | level = HEALTH_ERR; |
250 | } | |
251 | return level; | |
252 | } | |
253 | return HEALTH_OK; | |
254 | } | |
255 | ||
11fdf7f2 TL |
256 | void MgrMonitor::post_paxos_update() |
257 | { | |
258 | // are we handling digest subscribers? | |
259 | if (digest_event) { | |
260 | bool send = false; | |
261 | if (prev_health_checks.empty()) { | |
262 | prev_health_checks.resize(mon->paxos_service.size()); | |
263 | send = true; | |
264 | } | |
265 | ceph_assert(prev_health_checks.size() == mon->paxos_service.size()); | |
266 | for (auto i = 0u; i < prev_health_checks.size(); i++) { | |
267 | const auto& curr = mon->paxos_service[i]->get_health_checks(); | |
268 | if (!send && curr != prev_health_checks[i]) { | |
269 | send = true; | |
270 | } | |
271 | prev_health_checks[i] = curr; | |
272 | } | |
273 | if (send) { | |
274 | if (is_active()) { | |
275 | send_digests(); | |
276 | } else { | |
277 | cancel_timer(); | |
278 | wait_for_active_ctx(new C_MonContext(mon, [this](int) { | |
279 | send_digests(); | |
280 | })); | |
281 | } | |
282 | } | |
283 | } | |
284 | } | |
285 | ||
7c673cae FG |
286 | void MgrMonitor::encode_pending(MonitorDBStore::TransactionRef t) |
287 | { | |
288 | dout(10) << __func__ << " " << pending_map << dendl; | |
289 | bufferlist bl; | |
290 | pending_map.encode(bl, mon->get_quorum_con_features()); | |
291 | put_version(t, pending_map.epoch, bl); | |
292 | put_last_committed(t, pending_map.epoch); | |
224ce89b | 293 | |
c07f9fc5 FG |
294 | for (auto& p : pending_metadata) { |
295 | dout(10) << __func__ << " set metadata for " << p.first << dendl; | |
296 | t->put(MGR_METADATA_PREFIX, p.first, p.second); | |
297 | } | |
298 | for (auto& name : pending_metadata_rm) { | |
299 | dout(10) << __func__ << " rm metadata for " << name << dendl; | |
300 | t->erase(MGR_METADATA_PREFIX, name); | |
301 | } | |
302 | pending_metadata.clear(); | |
303 | pending_metadata_rm.clear(); | |
304 | ||
224ce89b WB |
305 | health_check_map_t next; |
306 | if (pending_map.active_gid == 0) { | |
307 | auto level = should_warn_about_mgr_down(); | |
308 | if (level != HEALTH_OK) { | |
309 | next.add("MGR_DOWN", level, "no active mgr"); | |
310 | } else { | |
311 | dout(10) << __func__ << " no health warning (never active and new cluster)" | |
312 | << dendl; | |
313 | } | |
314 | } else { | |
315 | put_value(t, "ever_had_active_mgr", 1); | |
316 | } | |
317 | encode_health(next, t); | |
c07f9fc5 FG |
318 | |
319 | if (pending_command_descs.size()) { | |
320 | dout(4) << __func__ << " encoding " << pending_command_descs.size() | |
321 | << " command_descs" << dendl; | |
322 | for (auto& p : pending_command_descs) { | |
323 | p.set_flag(MonCommand::FLAG_MGR); | |
324 | } | |
325 | bufferlist bl; | |
11fdf7f2 | 326 | encode(pending_command_descs, bl); |
c07f9fc5 FG |
327 | t->put(command_descs_prefix, "", bl); |
328 | pending_command_descs.clear(); | |
329 | } | |
7c673cae FG |
330 | } |
331 | ||
332 | bool MgrMonitor::check_caps(MonOpRequestRef op, const uuid_d& fsid) | |
333 | { | |
334 | // check permissions | |
335 | MonSession *session = op->get_session(); | |
336 | if (!session) | |
337 | return false; | |
338 | if (!session->is_capable("mgr", MON_CAP_X)) { | |
339 | dout(1) << __func__ << " insufficient caps " << session->caps << dendl; | |
340 | return false; | |
341 | } | |
342 | if (fsid != mon->monmap->fsid) { | |
343 | dout(1) << __func__ << " op fsid " << fsid | |
344 | << " != " << mon->monmap->fsid << dendl; | |
345 | return false; | |
346 | } | |
347 | return true; | |
348 | } | |
349 | ||
350 | bool MgrMonitor::preprocess_query(MonOpRequestRef op) | |
351 | { | |
352 | PaxosServiceMessage *m = static_cast<PaxosServiceMessage*>(op->get_req()); | |
353 | switch (m->get_type()) { | |
354 | case MSG_MGR_BEACON: | |
355 | return preprocess_beacon(op); | |
356 | case MSG_MON_COMMAND: | |
f64942e4 AA |
357 | try { |
358 | return preprocess_command(op); | |
11fdf7f2 | 359 | } catch (const bad_cmd_get& e) { |
f64942e4 AA |
360 | bufferlist bl; |
361 | mon->reply_command(op, -EINVAL, e.what(), bl, get_last_committed()); | |
362 | return true; | |
363 | } | |
364 | ||
7c673cae FG |
365 | default: |
366 | mon->no_reply(op); | |
367 | derr << "Unhandled message type " << m->get_type() << dendl; | |
368 | return true; | |
369 | } | |
370 | } | |
371 | ||
372 | bool MgrMonitor::prepare_update(MonOpRequestRef op) | |
373 | { | |
374 | PaxosServiceMessage *m = static_cast<PaxosServiceMessage*>(op->get_req()); | |
375 | switch (m->get_type()) { | |
376 | case MSG_MGR_BEACON: | |
377 | return prepare_beacon(op); | |
378 | ||
379 | case MSG_MON_COMMAND: | |
f64942e4 AA |
380 | try { |
381 | return prepare_command(op); | |
11fdf7f2 | 382 | } catch (const bad_cmd_get& e) { |
f64942e4 AA |
383 | bufferlist bl; |
384 | mon->reply_command(op, -EINVAL, e.what(), bl, get_last_committed()); | |
385 | return true; | |
386 | } | |
7c673cae FG |
387 | |
388 | default: | |
389 | mon->no_reply(op); | |
390 | derr << "Unhandled message type " << m->get_type() << dendl; | |
391 | return true; | |
392 | } | |
393 | } | |
394 | ||
395 | ||
396 | ||
397 | class C_Updated : public Context { | |
398 | MgrMonitor *mm; | |
399 | MonOpRequestRef op; | |
400 | public: | |
401 | C_Updated(MgrMonitor *a, MonOpRequestRef c) : | |
402 | mm(a), op(c) {} | |
403 | void finish(int r) override { | |
404 | if (r >= 0) { | |
405 | // Success | |
406 | } else if (r == -ECANCELED) { | |
407 | mm->mon->no_reply(op); | |
408 | } else { | |
409 | mm->dispatch(op); // try again | |
410 | } | |
411 | } | |
412 | }; | |
413 | ||
414 | bool MgrMonitor::preprocess_beacon(MonOpRequestRef op) | |
415 | { | |
416 | MMgrBeacon *m = static_cast<MMgrBeacon*>(op->get_req()); | |
94b18763 | 417 | mon->no_reply(op); // we never reply to beacons |
7c673cae FG |
418 | dout(4) << "beacon from " << m->get_gid() << dendl; |
419 | ||
420 | if (!check_caps(op, m->get_fsid())) { | |
421 | // drop it on the floor | |
422 | return true; | |
423 | } | |
424 | ||
425 | // always send this to the leader's prepare_beacon() | |
426 | return false; | |
427 | } | |
428 | ||
429 | bool MgrMonitor::prepare_beacon(MonOpRequestRef op) | |
430 | { | |
431 | MMgrBeacon *m = static_cast<MMgrBeacon*>(op->get_req()); | |
432 | dout(4) << "beacon from " << m->get_gid() << dendl; | |
433 | ||
434 | // See if we are seeing same name, new GID for the active daemon | |
435 | if (m->get_name() == pending_map.active_name | |
436 | && m->get_gid() != pending_map.active_gid) | |
437 | { | |
438 | dout(4) << "Active daemon restart (mgr." << m->get_name() << ")" << dendl; | |
224ce89b WB |
439 | mon->clog->info() << "Active manager daemon " << m->get_name() |
440 | << " restarted"; | |
7c673cae FG |
441 | drop_active(); |
442 | } | |
443 | ||
444 | // See if we are seeing same name, new GID for any standbys | |
445 | for (const auto &i : pending_map.standbys) { | |
11fdf7f2 | 446 | const MgrMap::StandbyInfo &s = i.second; |
7c673cae FG |
447 | if (s.name == m->get_name() && s.gid != m->get_gid()) { |
448 | dout(4) << "Standby daemon restart (mgr." << m->get_name() << ")" << dendl; | |
224ce89b WB |
449 | mon->clog->debug() << "Standby manager daemon " << m->get_name() |
450 | << " restarted"; | |
7c673cae FG |
451 | drop_standby(i.first); |
452 | break; | |
453 | } | |
454 | } | |
455 | ||
31f18b77 | 456 | last_beacon[m->get_gid()] = ceph::coarse_mono_clock::now(); |
7c673cae FG |
457 | |
458 | // Track whether we modified pending_map | |
459 | bool updated = false; | |
460 | ||
461 | if (pending_map.active_gid == m->get_gid()) { | |
3efd9988 FG |
462 | if (pending_map.services != m->get_services()) { |
463 | dout(4) << "updated services from mgr." << m->get_name() | |
464 | << ": " << m->get_services() << dendl; | |
465 | pending_map.services = m->get_services(); | |
466 | updated = true; | |
467 | } | |
468 | ||
7c673cae | 469 | // A beacon from the currently active daemon |
11fdf7f2 TL |
470 | if (pending_map.active_addrs != m->get_server_addrs()) { |
471 | dout(4) << "learned address " << m->get_server_addrs() | |
472 | << " (was " << pending_map.active_addrs << ")" << dendl; | |
473 | pending_map.active_addrs = m->get_server_addrs(); | |
7c673cae FG |
474 | updated = true; |
475 | } | |
476 | ||
477 | if (pending_map.get_available() != m->get_available()) { | |
478 | dout(4) << "available " << m->get_gid() << dendl; | |
224ce89b WB |
479 | mon->clog->info() << "Manager daemon " << pending_map.active_name |
480 | << " is now available"; | |
c07f9fc5 FG |
481 | |
482 | // This beacon should include command descriptions | |
483 | pending_command_descs = m->get_command_descs(); | |
484 | if (pending_command_descs.empty()) { | |
485 | // This should not happen, but it also isn't fatal: we just | |
486 | // won't successfully update our list of commands. | |
487 | dout(4) << "First available beacon from " << pending_map.active_name | |
488 | << "(" << m->get_gid() << ") does not include command descs" | |
489 | << dendl; | |
490 | } else { | |
491 | dout(4) << "First available beacon from " << pending_map.active_name | |
492 | << "(" << m->get_gid() << ") includes " | |
493 | << pending_command_descs.size() << " command descs" << dendl; | |
494 | } | |
495 | ||
7c673cae FG |
496 | pending_map.available = m->get_available(); |
497 | updated = true; | |
498 | } | |
224ce89b WB |
499 | if (pending_map.available_modules != m->get_available_modules()) { |
500 | dout(4) << "available_modules " << m->get_available_modules() | |
501 | << " (was " << pending_map.available_modules << ")" << dendl; | |
502 | pending_map.available_modules = m->get_available_modules(); | |
503 | updated = true; | |
504 | } | |
7c673cae FG |
505 | } else if (pending_map.active_gid == 0) { |
506 | // There is no currently active daemon, select this one. | |
507 | if (pending_map.standbys.count(m->get_gid())) { | |
181888fb | 508 | drop_standby(m->get_gid(), false); |
7c673cae FG |
509 | } |
510 | dout(4) << "selecting new active " << m->get_gid() | |
511 | << " " << m->get_name() | |
512 | << " (was " << pending_map.active_gid << " " | |
513 | << pending_map.active_name << ")" << dendl; | |
514 | pending_map.active_gid = m->get_gid(); | |
515 | pending_map.active_name = m->get_name(); | |
11fdf7f2 | 516 | pending_map.active_change = ceph_clock_now(); |
224ce89b | 517 | pending_map.available_modules = m->get_available_modules(); |
11fdf7f2 | 518 | encode(m->get_metadata(), pending_metadata[m->get_name()]); |
c07f9fc5 | 519 | pending_metadata_rm.erase(m->get_name()); |
224ce89b WB |
520 | |
521 | mon->clog->info() << "Activating manager daemon " | |
522 | << pending_map.active_name; | |
7c673cae FG |
523 | |
524 | updated = true; | |
525 | } else { | |
526 | if (pending_map.standbys.count(m->get_gid()) > 0) { | |
527 | dout(10) << "from existing standby " << m->get_gid() << dendl; | |
224ce89b WB |
528 | if (pending_map.standbys[m->get_gid()].available_modules != |
529 | m->get_available_modules()) { | |
530 | dout(10) << "existing standby " << m->get_gid() << " available_modules " | |
531 | << m->get_available_modules() << " (was " | |
532 | << pending_map.standbys[m->get_gid()].available_modules << ")" | |
533 | << dendl; | |
534 | pending_map.standbys[m->get_gid()].available_modules = | |
535 | m->get_available_modules(); | |
536 | updated = true; | |
537 | } | |
7c673cae FG |
538 | } else { |
539 | dout(10) << "new standby " << m->get_gid() << dendl; | |
224ce89b WB |
540 | mon->clog->debug() << "Standby manager daemon " << m->get_name() |
541 | << " started"; | |
c07f9fc5 FG |
542 | pending_map.standbys[m->get_gid()] = {m->get_gid(), m->get_name(), |
543 | m->get_available_modules()}; | |
11fdf7f2 | 544 | encode(m->get_metadata(), pending_metadata[m->get_name()]); |
c07f9fc5 | 545 | pending_metadata_rm.erase(m->get_name()); |
7c673cae FG |
546 | updated = true; |
547 | } | |
548 | } | |
549 | ||
550 | if (updated) { | |
551 | dout(4) << "updating map" << dendl; | |
552 | wait_for_finished_proposal(op, new C_Updated(this, op)); | |
553 | } else { | |
554 | dout(10) << "no change" << dendl; | |
555 | } | |
556 | ||
557 | return updated; | |
558 | } | |
559 | ||
560 | void MgrMonitor::check_subs() | |
561 | { | |
562 | const std::string type = "mgrmap"; | |
563 | if (mon->session_map.subs.count(type) == 0) | |
564 | return; | |
565 | for (auto sub : *(mon->session_map.subs[type])) { | |
566 | check_sub(sub); | |
567 | } | |
568 | } | |
569 | ||
570 | void MgrMonitor::check_sub(Subscription *sub) | |
571 | { | |
572 | if (sub->type == "mgrmap") { | |
573 | if (sub->next <= map.get_epoch()) { | |
224ce89b WB |
574 | dout(20) << "Sending map to subscriber " << sub->session->con |
575 | << " " << sub->session->con->get_peer_addr() << dendl; | |
7c673cae FG |
576 | sub->session->con->send_message(new MMgrMap(map)); |
577 | if (sub->onetime) { | |
578 | mon->session_map.remove_sub(sub); | |
579 | } else { | |
580 | sub->next = map.get_epoch() + 1; | |
581 | } | |
582 | } | |
583 | } else { | |
11fdf7f2 | 584 | ceph_assert(sub->type == "mgrdigest"); |
c07f9fc5 FG |
585 | if (sub->next == 0) { |
586 | // new registration; cancel previous timer | |
587 | cancel_timer(); | |
588 | } | |
31f18b77 | 589 | if (digest_event == nullptr) { |
7c673cae FG |
590 | send_digests(); |
591 | } | |
592 | } | |
593 | } | |
594 | ||
595 | /** | |
596 | * Handle digest subscriptions separately (outside of check_sub) because | |
597 | * they are going to be periodic rather than version-driven. | |
598 | */ | |
599 | void MgrMonitor::send_digests() | |
600 | { | |
31f18b77 FG |
601 | cancel_timer(); |
602 | ||
7c673cae | 603 | const std::string type = "mgrdigest"; |
11fdf7f2 TL |
604 | if (mon->session_map.subs.count(type) == 0) { |
605 | prev_health_checks.clear(); | |
7c673cae | 606 | return; |
11fdf7f2 | 607 | } |
7c673cae | 608 | |
b32b8144 FG |
609 | if (!is_active()) { |
610 | // if paxos is currently not active, don't send a digest but reenable timer | |
611 | goto timer; | |
612 | } | |
613 | dout(10) << __func__ << dendl; | |
614 | ||
7c673cae | 615 | for (auto sub : *(mon->session_map.subs[type])) { |
224ce89b WB |
616 | dout(10) << __func__ << " sending digest to subscriber " << sub->session->con |
617 | << " " << sub->session->con->get_peer_addr() << dendl; | |
7c673cae FG |
618 | MMgrDigest *mdigest = new MMgrDigest; |
619 | ||
620 | JSONFormatter f; | |
224ce89b | 621 | mon->get_health_status(true, &f, nullptr, nullptr, nullptr); |
7c673cae FG |
622 | f.flush(mdigest->health_json); |
623 | f.reset(); | |
624 | ||
625 | std::ostringstream ss; | |
626 | mon->get_mon_status(&f, ss); | |
627 | f.flush(mdigest->mon_status_json); | |
628 | f.reset(); | |
629 | ||
630 | sub->session->con->send_message(mdigest); | |
631 | } | |
632 | ||
b32b8144 | 633 | timer: |
3efd9988 | 634 | digest_event = mon->timer.add_event_after( |
11fdf7f2 | 635 | g_conf().get_val<int64_t>("mon_mgr_digest_period"), |
3efd9988 | 636 | new C_MonContext(mon, [this](int) { |
7c673cae | 637 | send_digests(); |
3efd9988 | 638 | })); |
31f18b77 FG |
639 | } |
640 | ||
641 | void MgrMonitor::cancel_timer() | |
642 | { | |
643 | if (digest_event) { | |
644 | mon->timer.cancel_event(digest_event); | |
645 | digest_event = nullptr; | |
646 | } | |
7c673cae FG |
647 | } |
648 | ||
649 | void MgrMonitor::on_active() | |
650 | { | |
224ce89b WB |
651 | if (mon->is_leader()) { |
652 | mon->clog->debug() << "mgrmap e" << map.epoch << ": " << map; | |
7c673cae | 653 | |
11fdf7f2 TL |
654 | if (HAVE_FEATURE(mon->get_quorum_con_features(), SERVER_NAUTILUS) && |
655 | pending_map.always_on_modules != always_on_modules) { | |
656 | pending_map.always_on_modules = always_on_modules; | |
657 | dout(4) << "always on modules changed, pending " | |
658 | << pending_map.get_always_on_modules() | |
659 | << " != wanted " << always_on_modules << dendl; | |
660 | propose_pending(); | |
7c673cae | 661 | } |
7c673cae FG |
662 | } |
663 | } | |
664 | ||
665 | void MgrMonitor::tick() | |
666 | { | |
667 | if (!is_active() || !mon->is_leader()) | |
668 | return; | |
669 | ||
31f18b77 | 670 | const auto now = ceph::coarse_mono_clock::now(); |
3efd9988 | 671 | |
11fdf7f2 TL |
672 | const auto mgr_beacon_grace = |
673 | g_conf().get_val<std::chrono::seconds>("mon_mgr_beacon_grace"); | |
3efd9988 FG |
674 | |
675 | // Note that this is the mgr daemon's tick period, not ours (the | |
676 | // beacon is sent with this period). | |
11fdf7f2 TL |
677 | const auto mgr_tick_period = |
678 | g_conf().get_val<std::chrono::seconds>("mgr_tick_period"); | |
3efd9988 FG |
679 | |
680 | if (last_tick != ceph::coarse_mono_clock::time_point::min() | |
681 | && (now - last_tick > (mgr_beacon_grace - mgr_tick_period))) { | |
682 | // This case handles either local slowness (calls being delayed | |
683 | // for whatever reason) or cluster election slowness (a long gap | |
684 | // between calls while an election happened) | |
685 | dout(4) << __func__ << ": resetting beacon timeouts due to mon delay " | |
686 | "(slow election?) of " << now - last_tick << " seconds" << dendl; | |
687 | for (auto &i : last_beacon) { | |
688 | i.second = now; | |
689 | } | |
690 | } | |
691 | ||
692 | last_tick = now; | |
7c673cae FG |
693 | |
694 | // Populate any missing beacons (i.e. no beacon since MgrMonitor | |
695 | // instantiation) with the current time, so that they will | |
696 | // eventually look laggy if they fail to give us a beacon. | |
697 | if (pending_map.active_gid != 0 | |
698 | && last_beacon.count(pending_map.active_gid) == 0) { | |
699 | last_beacon[pending_map.active_gid] = now; | |
700 | } | |
701 | for (auto s : pending_map.standbys) { | |
702 | if (last_beacon.count(s.first) == 0) { | |
703 | last_beacon[s.first] = now; | |
704 | } | |
705 | } | |
706 | ||
707 | // Cull standbys first so that any remaining standbys | |
708 | // will be eligible to take over from the active if we cull him. | |
709 | std::list<uint64_t> dead_standbys; | |
3efd9988 | 710 | const auto cutoff = now - mgr_beacon_grace; |
7c673cae FG |
711 | for (const auto &i : pending_map.standbys) { |
712 | auto last_beacon_time = last_beacon.at(i.first); | |
713 | if (last_beacon_time < cutoff) { | |
714 | dead_standbys.push_back(i.first); | |
715 | } | |
716 | } | |
717 | ||
718 | bool propose = false; | |
719 | ||
720 | for (auto i : dead_standbys) { | |
721 | dout(4) << "Dropping laggy standby " << i << dendl; | |
722 | drop_standby(i); | |
723 | propose = true; | |
724 | } | |
725 | ||
726 | if (pending_map.active_gid != 0 | |
727 | && last_beacon.at(pending_map.active_gid) < cutoff) { | |
224ce89b | 728 | const std::string old_active_name = pending_map.active_name; |
7c673cae FG |
729 | drop_active(); |
730 | propose = true; | |
731 | dout(4) << "Dropping active" << pending_map.active_gid << dendl; | |
732 | if (promote_standby()) { | |
733 | dout(4) << "Promoted standby " << pending_map.active_gid << dendl; | |
224ce89b WB |
734 | mon->clog->info() << "Manager daemon " << old_active_name |
735 | << " is unresponsive, replacing it with standby" | |
736 | << " daemon " << pending_map.active_name; | |
7c673cae FG |
737 | } else { |
738 | dout(4) << "Active is laggy but have no standbys to replace it" << dendl; | |
28e407b8 | 739 | mon->clog->info() << "Manager daemon " << old_active_name |
224ce89b | 740 | << " is unresponsive. No standby daemons available."; |
7c673cae FG |
741 | } |
742 | } else if (pending_map.active_gid == 0) { | |
743 | if (promote_standby()) { | |
744 | dout(4) << "Promoted standby " << pending_map.active_gid << dendl; | |
224ce89b | 745 | mon->clog->info() << "Activating manager daemon " |
3efd9988 | 746 | << pending_map.active_name; |
7c673cae FG |
747 | propose = true; |
748 | } | |
749 | } | |
750 | ||
224ce89b | 751 | if (!pending_map.available && |
c07f9fc5 | 752 | !ever_had_active_mgr && |
224ce89b | 753 | should_warn_about_mgr_down() != HEALTH_OK) { |
3efd9988 | 754 | dout(10) << " exceeded mon_mgr_mkfs_grace " |
11fdf7f2 | 755 | << g_conf().get_val<int64_t>("mon_mgr_mkfs_grace") |
3efd9988 | 756 | << " seconds" << dendl; |
224ce89b WB |
757 | propose = true; |
758 | } | |
759 | ||
7c673cae FG |
760 | if (propose) { |
761 | propose_pending(); | |
762 | } | |
763 | } | |
764 | ||
224ce89b WB |
765 | void MgrMonitor::on_restart() |
766 | { | |
767 | // Clear out the leader-specific state. | |
768 | last_beacon.clear(); | |
3efd9988 | 769 | last_tick = ceph::coarse_mono_clock::now(); |
224ce89b WB |
770 | } |
771 | ||
772 | ||
7c673cae FG |
773 | bool MgrMonitor::promote_standby() |
774 | { | |
11fdf7f2 | 775 | ceph_assert(pending_map.active_gid == 0); |
7c673cae FG |
776 | if (pending_map.standbys.size()) { |
777 | // Promote a replacement (arbitrary choice of standby) | |
778 | auto replacement_gid = pending_map.standbys.begin()->first; | |
779 | pending_map.active_gid = replacement_gid; | |
780 | pending_map.active_name = pending_map.standbys.at(replacement_gid).name; | |
781 | pending_map.available = false; | |
11fdf7f2 TL |
782 | pending_map.active_addrs = entity_addrvec_t(); |
783 | pending_map.active_change = ceph_clock_now(); | |
7c673cae | 784 | |
181888fb FG |
785 | drop_standby(replacement_gid, false); |
786 | ||
7c673cae FG |
787 | return true; |
788 | } else { | |
789 | return false; | |
790 | } | |
791 | } | |
792 | ||
793 | void MgrMonitor::drop_active() | |
794 | { | |
795 | if (last_beacon.count(pending_map.active_gid) > 0) { | |
796 | last_beacon.erase(pending_map.active_gid); | |
797 | } | |
798 | ||
c07f9fc5 FG |
799 | pending_metadata_rm.insert(pending_map.active_name); |
800 | pending_metadata.erase(pending_map.active_name); | |
7c673cae FG |
801 | pending_map.active_name = ""; |
802 | pending_map.active_gid = 0; | |
11fdf7f2 | 803 | pending_map.active_change = ceph_clock_now(); |
7c673cae | 804 | pending_map.available = false; |
11fdf7f2 | 805 | pending_map.active_addrs = entity_addrvec_t(); |
3efd9988 | 806 | pending_map.services.clear(); |
224ce89b WB |
807 | |
808 | // So that when new active mgr subscribes to mgrdigest, it will | |
809 | // get an immediate response instead of waiting for next timer | |
810 | cancel_timer(); | |
7c673cae FG |
811 | } |
812 | ||
181888fb | 813 | void MgrMonitor::drop_standby(uint64_t gid, bool drop_meta) |
7c673cae | 814 | { |
181888fb FG |
815 | if (drop_meta) { |
816 | pending_metadata_rm.insert(pending_map.standbys[gid].name); | |
817 | pending_metadata.erase(pending_map.standbys[gid].name); | |
818 | } | |
7c673cae FG |
819 | pending_map.standbys.erase(gid); |
820 | if (last_beacon.count(gid) > 0) { | |
821 | last_beacon.erase(gid); | |
822 | } | |
7c673cae FG |
823 | } |
824 | ||
825 | bool MgrMonitor::preprocess_command(MonOpRequestRef op) | |
826 | { | |
31f18b77 FG |
827 | MMonCommand *m = static_cast<MMonCommand*>(op->get_req()); |
828 | std::stringstream ss; | |
829 | bufferlist rdata; | |
830 | ||
11fdf7f2 | 831 | cmdmap_t cmdmap; |
31f18b77 FG |
832 | if (!cmdmap_from_json(m->cmd, &cmdmap, ss)) { |
833 | string rs = ss.str(); | |
834 | mon->reply_command(op, -EINVAL, rs, rdata, get_last_committed()); | |
835 | return true; | |
836 | } | |
7c673cae | 837 | |
11fdf7f2 | 838 | MonSession *session = op->get_session(); |
31f18b77 FG |
839 | if (!session) { |
840 | mon->reply_command(op, -EACCES, "access denied", rdata, | |
841 | get_last_committed()); | |
842 | return true; | |
843 | } | |
844 | ||
845 | string format; | |
11fdf7f2 | 846 | cmd_getval(g_ceph_context, cmdmap, "format", format, string("json-pretty")); |
31f18b77 FG |
847 | boost::scoped_ptr<Formatter> f(Formatter::create(format)); |
848 | ||
849 | string prefix; | |
11fdf7f2 | 850 | cmd_getval(g_ceph_context, cmdmap, "prefix", prefix); |
31f18b77 FG |
851 | int r = 0; |
852 | ||
853 | if (prefix == "mgr dump") { | |
854 | int64_t epoch = 0; | |
11fdf7f2 | 855 | cmd_getval(g_ceph_context, cmdmap, "epoch", epoch, (int64_t)map.get_epoch()); |
31f18b77 FG |
856 | if (epoch == (int64_t)map.get_epoch()) { |
857 | f->dump_object("mgrmap", map); | |
858 | } else { | |
859 | bufferlist bl; | |
860 | int err = get_version(epoch, bl); | |
861 | if (err == -ENOENT) { | |
862 | r = -ENOENT; | |
863 | ss << "there is no map for epoch " << epoch; | |
864 | goto reply; | |
865 | } | |
866 | MgrMap m; | |
11fdf7f2 | 867 | auto p = bl.cbegin(); |
31f18b77 FG |
868 | m.decode(p); |
869 | f->dump_object("mgrmap", m); | |
870 | } | |
871 | f->flush(rdata); | |
224ce89b | 872 | } else if (prefix == "mgr module ls") { |
3efd9988 FG |
873 | f->open_object_section("modules"); |
874 | { | |
875 | f->open_array_section("enabled_modules"); | |
876 | for (auto& p : map.modules) { | |
11fdf7f2 TL |
877 | if (map.get_always_on_modules().count(p) > 0) |
878 | continue; | |
879 | // We only show the name for enabled modules. The any errors | |
880 | // etc will show up as a health checks. | |
3efd9988 FG |
881 | f->dump_string("module", p); |
882 | } | |
883 | f->close_section(); | |
884 | f->open_array_section("disabled_modules"); | |
885 | for (auto& p : map.available_modules) { | |
11fdf7f2 TL |
886 | if (map.modules.count(p.name) == 0 && |
887 | map.get_always_on_modules().count(p.name) == 0) { | |
888 | // For disabled modules, we show the full info, to | |
889 | // give a hint about whether enabling it will work | |
890 | p.dump(f.get()); | |
3efd9988 FG |
891 | } |
892 | } | |
893 | f->close_section(); | |
894 | } | |
895 | f->close_section(); | |
896 | f->flush(rdata); | |
897 | } else if (prefix == "mgr services") { | |
898 | f->open_object_section("services"); | |
899 | for (const auto &i : map.services) { | |
900 | f->dump_string(i.first.c_str(), i.second); | |
224ce89b WB |
901 | } |
902 | f->close_section(); | |
903 | f->flush(rdata); | |
c07f9fc5 FG |
904 | } else if (prefix == "mgr metadata") { |
905 | string name; | |
11fdf7f2 | 906 | cmd_getval(g_ceph_context, cmdmap, "who", name); |
c07f9fc5 FG |
907 | if (name.size() > 0 && !map.have_name(name)) { |
908 | ss << "mgr." << name << " does not exist"; | |
909 | r = -ENOENT; | |
910 | goto reply; | |
911 | } | |
912 | string format; | |
11fdf7f2 | 913 | cmd_getval(g_ceph_context, cmdmap, "format", format); |
c07f9fc5 FG |
914 | boost::scoped_ptr<Formatter> f(Formatter::create(format, "json-pretty", "json-pretty")); |
915 | if (name.size()) { | |
916 | f->open_object_section("mgr_metadata"); | |
11fdf7f2 | 917 | f->dump_string("name", name); |
c07f9fc5 FG |
918 | r = dump_metadata(name, f.get(), &ss); |
919 | if (r < 0) | |
920 | goto reply; | |
921 | f->close_section(); | |
922 | } else { | |
923 | r = 0; | |
924 | f->open_array_section("mgr_metadata"); | |
925 | for (auto& i : map.get_all_names()) { | |
926 | f->open_object_section("mgr"); | |
11fdf7f2 | 927 | f->dump_string("name", i); |
c07f9fc5 FG |
928 | r = dump_metadata(i, f.get(), NULL); |
929 | if (r == -EINVAL || r == -ENOENT) { | |
930 | // Drop error, continue to get other daemons' metadata | |
931 | dout(4) << "No metadata for mgr." << i << dendl; | |
932 | r = 0; | |
933 | } else if (r < 0) { | |
934 | // Unexpected error | |
935 | goto reply; | |
936 | } | |
937 | f->close_section(); | |
938 | } | |
939 | f->close_section(); | |
940 | } | |
941 | f->flush(rdata); | |
942 | } else if (prefix == "mgr versions") { | |
943 | if (!f) | |
944 | f.reset(Formatter::create("json-pretty")); | |
945 | count_metadata("ceph_version", f.get()); | |
946 | f->flush(rdata); | |
947 | r = 0; | |
948 | } else if (prefix == "mgr count-metadata") { | |
949 | if (!f) | |
950 | f.reset(Formatter::create("json-pretty")); | |
951 | string field; | |
11fdf7f2 | 952 | cmd_getval(g_ceph_context, cmdmap, "property", field); |
c07f9fc5 FG |
953 | count_metadata(field, f.get()); |
954 | f->flush(rdata); | |
955 | r = 0; | |
31f18b77 FG |
956 | } else { |
957 | return false; | |
958 | } | |
959 | ||
960 | reply: | |
961 | string rs; | |
962 | getline(ss, rs); | |
963 | mon->reply_command(op, r, rs, rdata, get_last_committed()); | |
964 | return true; | |
7c673cae FG |
965 | } |
966 | ||
967 | bool MgrMonitor::prepare_command(MonOpRequestRef op) | |
968 | { | |
969 | MMonCommand *m = static_cast<MMonCommand*>(op->get_req()); | |
970 | ||
971 | std::stringstream ss; | |
972 | bufferlist rdata; | |
973 | ||
11fdf7f2 | 974 | cmdmap_t cmdmap; |
7c673cae FG |
975 | if (!cmdmap_from_json(m->cmd, &cmdmap, ss)) { |
976 | string rs = ss.str(); | |
977 | mon->reply_command(op, -EINVAL, rs, rdata, get_last_committed()); | |
978 | return true; | |
979 | } | |
980 | ||
11fdf7f2 | 981 | MonSession *session = op->get_session(); |
7c673cae FG |
982 | if (!session) { |
983 | mon->reply_command(op, -EACCES, "access denied", rdata, get_last_committed()); | |
984 | return true; | |
985 | } | |
986 | ||
224ce89b | 987 | string format; |
11fdf7f2 | 988 | cmd_getval(g_ceph_context, cmdmap, "format", format, string("plain")); |
224ce89b WB |
989 | boost::scoped_ptr<Formatter> f(Formatter::create(format)); |
990 | ||
7c673cae | 991 | string prefix; |
11fdf7f2 | 992 | cmd_getval(g_ceph_context, cmdmap, "prefix", prefix); |
7c673cae FG |
993 | |
994 | int r = 0; | |
995 | ||
996 | if (prefix == "mgr fail") { | |
997 | string who; | |
11fdf7f2 | 998 | cmd_getval(g_ceph_context, cmdmap, "who", who); |
7c673cae FG |
999 | |
1000 | std::string err; | |
1001 | uint64_t gid = strict_strtol(who.c_str(), 10, &err); | |
1002 | bool changed = false; | |
1003 | if (!err.empty()) { | |
1004 | // Does not parse as a gid, treat it as a name | |
1005 | if (pending_map.active_name == who) { | |
1006 | drop_active(); | |
1007 | changed = true; | |
1008 | } else { | |
1009 | gid = 0; | |
1010 | for (const auto &i : pending_map.standbys) { | |
1011 | if (i.second.name == who) { | |
1012 | gid = i.first; | |
1013 | break; | |
1014 | } | |
1015 | } | |
1016 | if (gid != 0) { | |
1017 | drop_standby(gid); | |
1018 | changed = true; | |
1019 | } else { | |
1020 | ss << "Daemon not found '" << who << "', already failed?"; | |
1021 | } | |
1022 | } | |
1023 | } else { | |
1024 | if (pending_map.active_gid == gid) { | |
1025 | drop_active(); | |
1026 | changed = true; | |
1027 | } else if (pending_map.standbys.count(gid) > 0) { | |
1028 | drop_standby(gid); | |
1029 | changed = true; | |
1030 | } else { | |
1031 | ss << "Daemon not found '" << gid << "', already failed?"; | |
1032 | } | |
1033 | } | |
1034 | ||
1035 | if (changed && pending_map.active_gid == 0) { | |
1036 | promote_standby(); | |
1037 | } | |
224ce89b WB |
1038 | } else if (prefix == "mgr module enable") { |
1039 | string module; | |
11fdf7f2 | 1040 | cmd_getval(g_ceph_context, cmdmap, "module", module); |
224ce89b WB |
1041 | if (module.empty()) { |
1042 | r = -EINVAL; | |
1043 | goto out; | |
1044 | } | |
11fdf7f2 TL |
1045 | if (pending_map.get_always_on_modules().count(module) > 0) { |
1046 | ss << "module '" << module << "' is already enabled (always-on)"; | |
1047 | goto out; | |
1048 | } | |
224ce89b | 1049 | string force; |
11fdf7f2 | 1050 | cmd_getval(g_ceph_context, cmdmap, "force", force); |
224ce89b WB |
1051 | if (!pending_map.all_support_module(module) && |
1052 | force != "--force") { | |
1053 | ss << "all mgr daemons do not support module '" << module << "', pass " | |
1054 | << "--force to force enablement"; | |
1055 | r = -ENOENT; | |
1056 | goto out; | |
1057 | } | |
11fdf7f2 TL |
1058 | |
1059 | std::string can_run_error; | |
1060 | if (force != "--force" && !pending_map.can_run_module(module, &can_run_error)) { | |
1061 | ss << "module '" << module << "' reports that it cannot run on the active " | |
1062 | "manager daemon: " << can_run_error << " (pass --force to force " | |
1063 | "enablement)"; | |
1064 | r = -ENOENT; | |
1065 | goto out; | |
1066 | } | |
1067 | ||
1068 | if (pending_map.module_enabled(module)) { | |
1069 | ss << "module '" << module << "' is already enabled"; | |
1070 | r = 0; | |
1071 | goto out; | |
1072 | } | |
224ce89b WB |
1073 | pending_map.modules.insert(module); |
1074 | } else if (prefix == "mgr module disable") { | |
1075 | string module; | |
11fdf7f2 | 1076 | cmd_getval(g_ceph_context, cmdmap, "module", module); |
224ce89b WB |
1077 | if (module.empty()) { |
1078 | r = -EINVAL; | |
1079 | goto out; | |
1080 | } | |
11fdf7f2 TL |
1081 | if (pending_map.get_always_on_modules().count(module) > 0) { |
1082 | ss << "module '" << module << "' cannot be disabled (always-on)"; | |
1083 | r = -EINVAL; | |
1084 | goto out; | |
1085 | } | |
1086 | if (!pending_map.module_enabled(module)) { | |
1087 | ss << "module '" << module << "' is already disabled"; | |
1088 | r = 0; | |
1089 | goto out; | |
1090 | } | |
1091 | if (!pending_map.any_supports_module(module)) { | |
1092 | ss << "module '" << module << "' does not exist"; | |
1093 | } | |
224ce89b | 1094 | pending_map.modules.erase(module); |
7c673cae | 1095 | } else { |
224ce89b | 1096 | ss << "Command '" << prefix << "' not implemented!"; |
7c673cae FG |
1097 | r = -ENOSYS; |
1098 | } | |
1099 | ||
224ce89b | 1100 | out: |
7c673cae FG |
1101 | dout(4) << __func__ << " done, r=" << r << dendl; |
1102 | /* Compose response */ | |
1103 | string rs; | |
1104 | getline(ss, rs); | |
1105 | ||
1106 | if (r >= 0) { | |
1107 | // success.. delay reply | |
1108 | wait_for_finished_proposal(op, new Monitor::C_Command(mon, op, r, rs, | |
1109 | get_last_committed() + 1)); | |
1110 | return true; | |
1111 | } else { | |
1112 | // reply immediately | |
1113 | mon->reply_command(op, r, rs, rdata, get_last_committed()); | |
1114 | return false; | |
1115 | } | |
1116 | } | |
1117 | ||
1118 | void MgrMonitor::init() | |
1119 | { | |
31f18b77 | 1120 | if (digest_event == nullptr) { |
7c673cae FG |
1121 | send_digests(); // To get it to schedule its own event |
1122 | } | |
1123 | } | |
1124 | ||
1125 | void MgrMonitor::on_shutdown() | |
1126 | { | |
31f18b77 | 1127 | cancel_timer(); |
7c673cae FG |
1128 | } |
1129 | ||
c07f9fc5 | 1130 | int MgrMonitor::load_metadata(const string& name, std::map<string, string>& m, |
11fdf7f2 | 1131 | ostream *err) const |
c07f9fc5 FG |
1132 | { |
1133 | bufferlist bl; | |
1134 | int r = mon->store->get(MGR_METADATA_PREFIX, name, bl); | |
1135 | if (r < 0) | |
1136 | return r; | |
1137 | try { | |
11fdf7f2 TL |
1138 | auto p = bl.cbegin(); |
1139 | decode(m, p); | |
c07f9fc5 FG |
1140 | } |
1141 | catch (buffer::error& e) { | |
1142 | if (err) | |
1143 | *err << "mgr." << name << " metadata is corrupt"; | |
1144 | return -EIO; | |
1145 | } | |
1146 | return 0; | |
1147 | } | |
1148 | ||
1149 | void MgrMonitor::count_metadata(const string& field, std::map<string,int> *out) | |
1150 | { | |
1151 | std::set<string> ls = map.get_all_names(); | |
1152 | for (auto& name : ls) { | |
1153 | std::map<string,string> meta; | |
1154 | load_metadata(name, meta, nullptr); | |
1155 | auto p = meta.find(field); | |
1156 | if (p == meta.end()) { | |
1157 | (*out)["unknown"]++; | |
1158 | } else { | |
1159 | (*out)[p->second]++; | |
1160 | } | |
1161 | } | |
1162 | } | |
1163 | ||
1164 | void MgrMonitor::count_metadata(const string& field, Formatter *f) | |
1165 | { | |
1166 | std::map<string,int> by_val; | |
1167 | count_metadata(field, &by_val); | |
1168 | f->open_object_section(field.c_str()); | |
1169 | for (auto& p : by_val) { | |
1170 | f->dump_int(p.first.c_str(), p.second); | |
1171 | } | |
1172 | f->close_section(); | |
1173 | } | |
1174 | ||
1175 | int MgrMonitor::dump_metadata(const string& name, Formatter *f, ostream *err) | |
1176 | { | |
1177 | std::map<string,string> m; | |
1178 | if (int r = load_metadata(name, m, err)) | |
1179 | return r; | |
1180 | for (auto& p : m) { | |
1181 | f->dump_string(p.first.c_str(), p.second); | |
1182 | } | |
1183 | return 0; | |
1184 | } | |
31f18b77 | 1185 | |
11fdf7f2 TL |
1186 | void MgrMonitor::print_nodes(Formatter *f) const |
1187 | { | |
1188 | ceph_assert(f); | |
1189 | ||
1190 | std::map<string, list<string> > mgrs; // hostname => mgr | |
1191 | auto ls = map.get_all_names(); | |
1192 | for (auto& name : ls) { | |
1193 | std::map<string,string> meta; | |
1194 | if (load_metadata(name, meta, nullptr)) { | |
1195 | continue; | |
1196 | } | |
1197 | auto hostname = meta.find("hostname"); | |
1198 | if (hostname == meta.end()) { | |
1199 | // not likely though | |
1200 | continue; | |
1201 | } | |
1202 | mgrs[hostname->second].push_back(name); | |
1203 | } | |
1204 | ||
1205 | dump_services(f, mgrs, "mgr"); | |
1206 | } | |
1207 | ||
d2e6a577 FG |
1208 | const std::vector<MonCommand> &MgrMonitor::get_command_descs() const |
1209 | { | |
1210 | if (command_descs.empty()) { | |
1211 | // must have just upgraded; fallback to static commands | |
1212 | return mgr_commands; | |
1213 | } else { | |
1214 | return command_descs; | |
1215 | } | |
1216 | } |