]>
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 | ||
14 | #include <Python.h> | |
20effc67 | 15 | #include <boost/algorithm/string/replace.hpp> |
7c673cae FG |
16 | |
17 | #include "common/errno.h" | |
31f18b77 FG |
18 | #include "common/signal.h" |
19 | #include "include/compat.h" | |
7c673cae FG |
20 | |
21 | #include "include/stringify.h" | |
22 | #include "global/global_context.h" | |
23 | #include "global/signal_handler.h" | |
24 | ||
25 | #include "mgr/MgrContext.h" | |
11fdf7f2 | 26 | #include "mgr/mgr_commands.h" |
20effc67 | 27 | #include "mgr/mgr_perf_counters.h" |
7c673cae FG |
28 | |
29 | #include "messages/MMgrBeacon.h" | |
30 | #include "messages/MMgrMap.h" | |
31 | #include "Mgr.h" | |
32 | ||
33 | #include "MgrStandby.h" | |
34 | ||
35 | #define dout_context g_ceph_context | |
36 | #define dout_subsys ceph_subsys_mgr | |
37 | #undef dout_prefix | |
38 | #define dout_prefix *_dout << "mgr " << __func__ << " " | |
39 | ||
20effc67 TL |
40 | using std::map; |
41 | using std::string; | |
42 | using std::vector; | |
7c673cae | 43 | |
31f18b77 | 44 | MgrStandby::MgrStandby(int argc, const char **argv) : |
7c673cae | 45 | Dispatcher(g_ceph_context), |
f67539c2 | 46 | monc{g_ceph_context, poolctx}, |
11fdf7f2 TL |
47 | client_messenger(Messenger::create( |
48 | g_ceph_context, | |
20effc67 TL |
49 | cct->_conf.get_val<std::string>("ms_public_type").empty() ? |
50 | cct->_conf.get_val<std::string>("ms_type") : cct->_conf.get_val<std::string>("ms_public_type"), | |
11fdf7f2 TL |
51 | entity_name_t::MGR(), |
52 | "mgr", | |
f67539c2 TL |
53 | Messenger::get_pid_nonce())), |
54 | objecter{g_ceph_context, client_messenger.get(), &monc, poolctx}, | |
7c673cae | 55 | client{client_messenger.get(), &monc, &objecter}, |
9f95a23c | 56 | mgrc(g_ceph_context, client_messenger.get(), &monc.monmap), |
7c673cae FG |
57 | log_client(g_ceph_context, client_messenger.get(), &monc.monmap, LogClient::NO_FLAGS), |
58 | clog(log_client.create_channel(CLOG_CHANNEL_CLUSTER)), | |
59 | audit_clog(log_client.create_channel(CLOG_CHANNEL_AUDIT)), | |
11fdf7f2 | 60 | finisher(g_ceph_context, "MgrStandby", "mgrsb-fin"), |
7c673cae | 61 | timer(g_ceph_context, lock), |
3efd9988 | 62 | py_module_registry(clog), |
31f18b77 FG |
63 | active_mgr(nullptr), |
64 | orig_argc(argc), | |
c07f9fc5 FG |
65 | orig_argv(argv), |
66 | available_in_map(false) | |
7c673cae FG |
67 | { |
68 | } | |
69 | ||
70 | MgrStandby::~MgrStandby() = default; | |
71 | ||
72 | const char** MgrStandby::get_tracked_conf_keys() const | |
73 | { | |
74 | static const char* KEYS[] = { | |
75 | // clog & admin clog | |
76 | "clog_to_monitors", | |
77 | "clog_to_syslog", | |
78 | "clog_to_syslog_facility", | |
79 | "clog_to_syslog_level", | |
7c673cae FG |
80 | "clog_to_graylog", |
81 | "clog_to_graylog_host", | |
82 | "clog_to_graylog_port", | |
b3b6e05e | 83 | "mgr_standby_modules", |
7c673cae FG |
84 | "host", |
85 | "fsid", | |
86 | NULL | |
87 | }; | |
88 | return KEYS; | |
89 | } | |
90 | ||
91 | void MgrStandby::handle_conf_change( | |
11fdf7f2 | 92 | const ConfigProxy& conf, |
7c673cae FG |
93 | const std::set <std::string> &changed) |
94 | { | |
95 | if (changed.count("clog_to_monitors") || | |
96 | changed.count("clog_to_syslog") || | |
97 | changed.count("clog_to_syslog_level") || | |
98 | changed.count("clog_to_syslog_facility") || | |
99 | changed.count("clog_to_graylog") || | |
100 | changed.count("clog_to_graylog_host") || | |
101 | changed.count("clog_to_graylog_port") || | |
102 | changed.count("host") || | |
103 | changed.count("fsid")) { | |
104 | _update_log_config(); | |
105 | } | |
b3b6e05e TL |
106 | if (changed.count("mgr_standby_modules") && !active_mgr) { |
107 | if (g_conf().get_val<bool>("mgr_standby_modules") != py_module_registry.have_standby_modules()) { | |
108 | dout(1) << "mgr_standby_modules now " | |
109 | << (int)g_conf().get_val<bool>("mgr_standby_modules") | |
110 | << ", standby modules are " | |
111 | << (py_module_registry.have_standby_modules() ? "":"not ") | |
112 | << "active, respawning" | |
113 | << dendl; | |
114 | respawn(); | |
115 | } | |
116 | } | |
7c673cae FG |
117 | } |
118 | ||
119 | int MgrStandby::init() | |
120 | { | |
11fdf7f2 TL |
121 | init_async_signal_handler(); |
122 | register_async_signal_handler(SIGHUP, sighup_handler); | |
123 | ||
b3b6e05e TL |
124 | cct->_conf.add_observer(this); |
125 | ||
11fdf7f2 TL |
126 | std::lock_guard l(lock); |
127 | ||
128 | // Start finisher | |
129 | finisher.start(); | |
7c673cae FG |
130 | |
131 | // Initialize Messenger | |
132 | client_messenger->add_dispatcher_tail(this); | |
133 | client_messenger->add_dispatcher_head(&objecter); | |
134 | client_messenger->add_dispatcher_tail(&client); | |
135 | client_messenger->start(); | |
136 | ||
f67539c2 TL |
137 | poolctx.start(2); |
138 | ||
7c673cae FG |
139 | // Initialize MonClient |
140 | if (monc.build_initial_monmap() < 0) { | |
141 | client_messenger->shutdown(); | |
142 | client_messenger->wait(); | |
143 | return -1; | |
144 | } | |
145 | ||
146 | monc.sub_want("mgrmap", 0, 0); | |
147 | ||
148 | monc.set_want_keys(CEPH_ENTITY_TYPE_MON|CEPH_ENTITY_TYPE_OSD | |
149 | |CEPH_ENTITY_TYPE_MDS|CEPH_ENTITY_TYPE_MGR); | |
150 | monc.set_messenger(client_messenger.get()); | |
11fdf7f2 TL |
151 | |
152 | // We must register our config callback before calling init(), so | |
153 | // that we see the initial configuration message | |
154 | monc.register_config_callback([this](const std::string &k, const std::string &v){ | |
adb31ebb TL |
155 | // removing value to hide sensitive data going into mgr logs |
156 | // leaving this for debugging purposes | |
157 | // dout(10) << "config_callback: " << k << " : " << v << dendl; | |
158 | dout(10) << "config_callback: " << k << " : " << dendl; | |
11fdf7f2 | 159 | if (k.substr(0, 4) == "mgr/") { |
f67539c2 | 160 | py_module_registry.handle_config(k, v); |
11fdf7f2 TL |
161 | return true; |
162 | } | |
163 | return false; | |
164 | }); | |
165 | monc.register_config_notify_callback([this]() { | |
166 | py_module_registry.handle_config_notify(); | |
167 | }); | |
168 | dout(4) << "Registered monc callback" << dendl; | |
169 | ||
7c673cae FG |
170 | int r = monc.init(); |
171 | if (r < 0) { | |
172 | monc.shutdown(); | |
173 | client_messenger->shutdown(); | |
174 | client_messenger->wait(); | |
175 | return r; | |
176 | } | |
11fdf7f2 TL |
177 | mgrc.init(); |
178 | client_messenger->add_dispatcher_tail(&mgrc); | |
179 | ||
7c673cae FG |
180 | r = monc.authenticate(); |
181 | if (r < 0) { | |
182 | derr << "Authentication failed, did you specify a mgr ID with a valid keyring?" << dendl; | |
183 | monc.shutdown(); | |
184 | client_messenger->shutdown(); | |
185 | client_messenger->wait(); | |
186 | return r; | |
187 | } | |
11fdf7f2 TL |
188 | // only forward monmap updates after authentication finishes, otherwise |
189 | // monc.authenticate() will be waiting for MgrStandy::ms_dispatch() | |
190 | // to acquire the lock forever, as it is already locked in the beginning of | |
191 | // this method. | |
192 | monc.set_passthrough_monmap(); | |
7c673cae FG |
193 | |
194 | client_t whoami = monc.get_global_id(); | |
11fdf7f2 | 195 | client_messenger->set_myname(entity_name_t::MGR(whoami.v)); |
7c673cae FG |
196 | monc.set_log_client(&log_client); |
197 | _update_log_config(); | |
198 | objecter.set_client_incarnation(0); | |
199 | objecter.init(); | |
200 | objecter.start(); | |
201 | client.init(); | |
202 | timer.init(); | |
203 | ||
11fdf7f2 | 204 | py_module_registry.init(); |
20effc67 TL |
205 | mgr_perf_start(g_ceph_context); |
206 | ||
11fdf7f2 | 207 | |
31f18b77 | 208 | tick(); |
7c673cae FG |
209 | |
210 | dout(4) << "Complete." << dendl; | |
211 | return 0; | |
212 | } | |
213 | ||
214 | void MgrStandby::send_beacon() | |
215 | { | |
9f95a23c TL |
216 | ceph_assert(ceph_mutex_is_locked_by_me(lock)); |
217 | dout(20) << state_str() << dendl; | |
11fdf7f2 | 218 | |
9f95a23c | 219 | auto modules = py_module_registry.get_modules(); |
11fdf7f2 TL |
220 | |
221 | // Construct a list of the info about each loaded module | |
222 | // which we will transmit to the monitor. | |
223 | std::vector<MgrMap::ModuleInfo> module_info; | |
224 | for (const auto &module : modules) { | |
225 | MgrMap::ModuleInfo info; | |
226 | info.name = module->get_name(); | |
227 | info.error_string = module->get_error_string(); | |
228 | info.can_run = module->get_can_run(); | |
229 | info.module_options = module->get_options(); | |
230 | module_info.push_back(std::move(info)); | |
231 | } | |
c07f9fc5 | 232 | |
9f95a23c TL |
233 | auto clients = py_module_registry.get_clients(); |
234 | for (const auto& client : clients) { | |
f67539c2 | 235 | dout(15) << "noting RADOS client for blocklist: " << client << dendl; |
9f95a23c TL |
236 | } |
237 | ||
c07f9fc5 FG |
238 | // Whether I think I am available (request MgrMonitor to set me |
239 | // as available in the map) | |
7c673cae | 240 | bool available = active_mgr != nullptr && active_mgr->is_initialized(); |
c07f9fc5 | 241 | |
11fdf7f2 | 242 | auto addrs = available ? active_mgr->get_server_addrs() : entity_addrvec_t(); |
3efd9988 | 243 | dout(10) << "sending beacon as gid " << monc.get_global_id() << dendl; |
224ce89b | 244 | |
c07f9fc5 | 245 | map<string,string> metadata; |
11fdf7f2 TL |
246 | metadata["addr"] = client_messenger->get_myaddr_legacy().ip_only_to_str(); |
247 | metadata["addrs"] = stringify(client_messenger->get_myaddrs()); | |
c07f9fc5 FG |
248 | collect_sys_info(&metadata, g_ceph_context); |
249 | ||
9f95a23c | 250 | auto m = ceph::make_message<MMgrBeacon>(monc.get_fsid(), |
7c673cae | 251 | monc.get_global_id(), |
11fdf7f2 TL |
252 | g_conf()->name.get_id(), |
253 | addrs, | |
224ce89b | 254 | available, |
11fdf7f2 | 255 | std::move(module_info), |
9f95a23c TL |
256 | std::move(metadata), |
257 | std::move(clients), | |
258 | CEPH_FEATURES_ALL); | |
c07f9fc5 | 259 | |
3efd9988 FG |
260 | if (available) { |
261 | if (!available_in_map) { | |
262 | // We are informing the mon that we are done initializing: inform | |
263 | // it of our command set. This has to happen after init() because | |
264 | // it needs the python modules to have loaded. | |
11fdf7f2 TL |
265 | std::vector<MonCommand> commands = mgr_commands; |
266 | std::vector<MonCommand> py_commands = py_module_registry.get_commands(); | |
267 | commands.insert(commands.end(), py_commands.begin(), py_commands.end()); | |
20effc67 TL |
268 | if (monc.monmap.min_mon_release < ceph_release_t::quincy) { |
269 | dout(10) << " stripping out positional=false quincy-ism" << dendl; | |
270 | for (auto& i : commands) { | |
271 | boost::replace_all(i.cmdstring, ",positional=false", ""); | |
272 | } | |
273 | } | |
11fdf7f2 | 274 | m->set_command_descs(commands); |
3efd9988 FG |
275 | dout(4) << "going active, including " << m->get_command_descs().size() |
276 | << " commands in beacon" << dendl; | |
277 | } | |
278 | ||
279 | m->set_services(active_mgr->get_services()); | |
c07f9fc5 | 280 | } |
20effc67 | 281 | |
9f95a23c | 282 | monc.send_mon_message(std::move(m)); |
31f18b77 FG |
283 | } |
284 | ||
285 | void MgrStandby::tick() | |
286 | { | |
224ce89b | 287 | dout(10) << __func__ << dendl; |
31f18b77 FG |
288 | send_beacon(); |
289 | ||
11fdf7f2 TL |
290 | timer.add_event_after( |
291 | g_conf().get_val<std::chrono::seconds>("mgr_tick_period").count(), | |
9f95a23c | 292 | new LambdaContext([this](int r){ |
31f18b77 | 293 | tick(); |
3efd9988 | 294 | } |
20effc67 | 295 | )); |
7c673cae FG |
296 | } |
297 | ||
7c673cae FG |
298 | void MgrStandby::shutdown() |
299 | { | |
9f95a23c | 300 | finisher.queue(new LambdaContext([&](int) { |
11fdf7f2 TL |
301 | std::lock_guard l(lock); |
302 | ||
303 | dout(4) << "Shutting down" << dendl; | |
304 | ||
f67539c2 TL |
305 | py_module_registry.shutdown(); |
306 | // stop sending beacon first, I use monc to talk with monitors | |
11fdf7f2 TL |
307 | timer.shutdown(); |
308 | // client uses monc and objecter | |
309 | client.shutdown(); | |
310 | mgrc.shutdown(); | |
f67539c2 TL |
311 | // Stop asio threads, so leftover events won't call into shut down |
312 | // monclient/objecter. | |
313 | poolctx.finish(); | |
11fdf7f2 TL |
314 | // stop monc, so mon won't be able to instruct me to shutdown/activate after |
315 | // the active_mgr is stopped | |
316 | monc.shutdown(); | |
317 | if (active_mgr) { | |
318 | active_mgr->shutdown(); | |
319 | } | |
11fdf7f2 TL |
320 | // objecter is used by monc and active_mgr |
321 | objecter.shutdown(); | |
322 | // client_messenger is used by all of them, so stop it in the end | |
323 | client_messenger->shutdown(); | |
324 | })); | |
325 | ||
326 | // Then stop the finisher to ensure its enqueued contexts aren't going | |
327 | // to touch references to the things we're about to tear down | |
328 | finisher.wait_for_empty(); | |
329 | finisher.stop(); | |
20effc67 | 330 | mgr_perf_stop(g_ceph_context); |
7c673cae FG |
331 | } |
332 | ||
31f18b77 FG |
333 | void MgrStandby::respawn() |
334 | { | |
11fdf7f2 TL |
335 | // --- WARNING TO FUTURE COPY/PASTERS --- |
336 | // You must also add a call like | |
337 | // | |
338 | // ceph_pthread_setname(pthread_self(), "ceph-mgr"); | |
339 | // | |
340 | // to main() so that /proc/$pid/stat field 2 contains "(ceph-mgr)" | |
341 | // instead of "(exe)", so that killall (and log rotation) will work. | |
342 | ||
31f18b77 FG |
343 | char *new_argv[orig_argc+1]; |
344 | dout(1) << " e: '" << orig_argv[0] << "'" << dendl; | |
345 | for (int i=0; i<orig_argc; i++) { | |
346 | new_argv[i] = (char *)orig_argv[i]; | |
347 | dout(1) << " " << i << ": '" << orig_argv[i] << "'" << dendl; | |
348 | } | |
349 | new_argv[orig_argc] = NULL; | |
350 | ||
351 | /* Determine the path to our executable, test if Linux /proc/self/exe exists. | |
352 | * This allows us to exec the same executable even if it has since been | |
353 | * unlinked. | |
354 | */ | |
355 | char exe_path[PATH_MAX] = ""; | |
356 | if (readlink(PROCPREFIX "/proc/self/exe", exe_path, PATH_MAX-1) == -1) { | |
357 | /* Print CWD for the user's interest */ | |
358 | char buf[PATH_MAX]; | |
359 | char *cwd = getcwd(buf, sizeof(buf)); | |
11fdf7f2 | 360 | ceph_assert(cwd); |
31f18b77 FG |
361 | dout(1) << " cwd " << cwd << dendl; |
362 | ||
363 | /* Fall back to a best-effort: just running in our CWD */ | |
364 | strncpy(exe_path, orig_argv[0], PATH_MAX-1); | |
365 | } else { | |
366 | dout(1) << "respawning with exe " << exe_path << dendl; | |
367 | strcpy(exe_path, PROCPREFIX "/proc/self/exe"); | |
368 | } | |
369 | ||
370 | dout(1) << " exe_path " << exe_path << dendl; | |
371 | ||
372 | unblock_all_signals(NULL); | |
373 | execv(exe_path, new_argv); | |
374 | ||
375 | derr << "respawn execv " << orig_argv[0] | |
376 | << " failed with " << cpp_strerror(errno) << dendl; | |
377 | ceph_abort(); | |
378 | } | |
379 | ||
7c673cae FG |
380 | void MgrStandby::_update_log_config() |
381 | { | |
20effc67 TL |
382 | clog->parse_client_options(cct); |
383 | audit_clog->parse_client_options(cct); | |
7c673cae FG |
384 | } |
385 | ||
9f95a23c | 386 | void MgrStandby::handle_mgr_map(ref_t<MMgrMap> mmap) |
7c673cae | 387 | { |
c07f9fc5 | 388 | auto &map = mmap->get_map(); |
7c673cae FG |
389 | dout(4) << "received map epoch " << map.get_epoch() << dendl; |
390 | const bool active_in_map = map.active_gid == monc.get_global_id(); | |
391 | dout(4) << "active in map: " << active_in_map | |
392 | << " active is " << map.active_gid << dendl; | |
3efd9988 | 393 | |
11fdf7f2 TL |
394 | // PyModuleRegistry may ask us to respawn if it sees that |
395 | // this MgrMap is changing its set of enabled modules | |
396 | bool need_respawn = py_module_registry.handle_mgr_map(map); | |
397 | if (need_respawn) { | |
adb31ebb | 398 | dout(1) << "respawning because set of enabled modules changed!" << dendl; |
11fdf7f2 | 399 | respawn(); |
3efd9988 FG |
400 | } |
401 | ||
7c673cae FG |
402 | if (active_in_map) { |
403 | if (!active_mgr) { | |
404 | dout(1) << "Activating!" << dendl; | |
3efd9988 FG |
405 | active_mgr.reset(new Mgr(&monc, map, &py_module_registry, |
406 | client_messenger.get(), &objecter, | |
7c673cae | 407 | &client, clog, audit_clog)); |
9f95a23c | 408 | active_mgr->background_init(new LambdaContext( |
224ce89b WB |
409 | [this](int r){ |
410 | // Advertise our active-ness ASAP instead of waiting for | |
411 | // next tick. | |
11fdf7f2 | 412 | std::lock_guard l(lock); |
224ce89b WB |
413 | send_beacon(); |
414 | })); | |
415 | dout(1) << "I am now activating" << dendl; | |
7c673cae FG |
416 | } else { |
417 | dout(10) << "I was already active" << dendl; | |
224ce89b WB |
418 | bool need_respawn = active_mgr->got_mgr_map(map); |
419 | if (need_respawn) { | |
420 | respawn(); | |
421 | } | |
7c673cae | 422 | } |
c07f9fc5 FG |
423 | |
424 | if (!available_in_map && map.get_available()) { | |
425 | dout(4) << "Map now says I am available" << dendl; | |
426 | available_in_map = true; | |
427 | } | |
3efd9988 FG |
428 | } else if (active_mgr != nullptr) { |
429 | derr << "I was active but no longer am" << dendl; | |
430 | respawn(); | |
7c673cae | 431 | } else { |
11fdf7f2 | 432 | if (map.active_gid != 0 && map.active_name != g_conf()->name.get_id()) { |
3efd9988 FG |
433 | // I am the standby and someone else is active, start modules |
434 | // in standby mode to do redirects if needed | |
b3b6e05e TL |
435 | if (!py_module_registry.is_standby_running() && |
436 | g_conf().get_val<bool>("mgr_standby_modules")) { | |
11fdf7f2 | 437 | py_module_registry.standby_start(monc, finisher); |
3efd9988 | 438 | } |
7c673cae FG |
439 | } |
440 | } | |
7c673cae FG |
441 | } |
442 | ||
9f95a23c | 443 | bool MgrStandby::ms_dispatch2(const ref_t<Message>& m) |
7c673cae | 444 | { |
11fdf7f2 | 445 | std::lock_guard l(lock); |
9f95a23c | 446 | dout(10) << state_str() << " " << *m << dendl; |
7c673cae | 447 | |
31f18b77 | 448 | if (m->get_type() == MSG_MGR_MAP) { |
9f95a23c | 449 | handle_mgr_map(ref_cast<MMgrMap>(m)); |
11fdf7f2 TL |
450 | } |
451 | bool handled = false; | |
452 | if (active_mgr) { | |
31f18b77 | 453 | auto am = active_mgr; |
9f95a23c TL |
454 | lock.unlock(); |
455 | handled = am->ms_dispatch2(m); | |
456 | lock.lock(); | |
7c673cae | 457 | } |
94b18763 FG |
458 | if (m->get_type() == MSG_MGR_MAP) { |
459 | // let this pass through for mgrc | |
460 | handled = false; | |
461 | } | |
462 | return handled; | |
7c673cae FG |
463 | } |
464 | ||
465 | ||
7c673cae FG |
466 | bool MgrStandby::ms_handle_refused(Connection *con) |
467 | { | |
468 | // do nothing for now | |
469 | return false; | |
470 | } | |
471 | ||
7c673cae FG |
472 | int MgrStandby::main(vector<const char *> args) |
473 | { | |
7c673cae FG |
474 | client_messenger->wait(); |
475 | ||
476 | // Disable signal handlers | |
477 | unregister_async_signal_handler(SIGHUP, sighup_handler); | |
7c673cae | 478 | shutdown_async_signal_handler(); |
7c673cae FG |
479 | |
480 | return 0; | |
481 | } | |
482 | ||
483 | ||
484 | std::string MgrStandby::state_str() | |
485 | { | |
3efd9988 FG |
486 | if (active_mgr == nullptr) { |
487 | return "standby"; | |
488 | } else if (active_mgr->is_initialized()) { | |
489 | return "active"; | |
490 | } else { | |
491 | return "active (starting)"; | |
492 | } | |
7c673cae | 493 | } |