]>
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) 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 <sys/types.h> | |
16 | #include <sys/stat.h> | |
17 | #include <fcntl.h> | |
18 | #include <pthread.h> | |
19 | ||
20 | #include <iostream> | |
21 | #include <string> | |
7c673cae | 22 | |
f67539c2 | 23 | #include "common/async/context_pool.h" |
7c673cae FG |
24 | #include "include/ceph_features.h" |
25 | #include "include/compat.h" | |
11fdf7f2 | 26 | #include "include/random.h" |
7c673cae FG |
27 | |
28 | #include "common/config.h" | |
29 | #include "common/strtol.h" | |
9f95a23c | 30 | #include "common/numa.h" |
7c673cae FG |
31 | |
32 | #include "mon/MonMap.h" | |
33 | #include "mds/MDSDaemon.h" | |
34 | ||
35 | #include "msg/Messenger.h" | |
36 | ||
37 | #include "common/Timer.h" | |
38 | #include "common/ceph_argparse.h" | |
39 | #include "common/pick_address.h" | |
11fdf7f2 | 40 | #include "common/Preforker.h" |
7c673cae FG |
41 | |
42 | #include "global/global_init.h" | |
43 | #include "global/signal_handler.h" | |
44 | #include "global/pidfile.h" | |
45 | ||
46 | #include "mon/MonClient.h" | |
47 | ||
48 | #include "auth/KeyRing.h" | |
49 | ||
50 | #include "perfglue/heap_profiler.h" | |
51 | ||
11fdf7f2 | 52 | #include "include/ceph_assert.h" |
7c673cae FG |
53 | |
54 | #define dout_context g_ceph_context | |
55 | #define dout_subsys ceph_subsys_mds | |
56 | ||
20effc67 TL |
57 | using std::cerr; |
58 | using std::cout; | |
59 | using std::vector; | |
60 | ||
7c673cae FG |
61 | static void usage() |
62 | { | |
11fdf7f2 | 63 | cout << "usage: ceph-mds -i <ID> [flags]\n" |
7c673cae FG |
64 | << " -m monitorip:port\n" |
65 | << " connect to monitor at given address\n" | |
66 | << " --debug_mds n\n" | |
67 | << " debug MDS level (e.g. 10)\n" | |
7c673cae FG |
68 | << std::endl; |
69 | generic_server_usage(); | |
70 | } | |
71 | ||
72 | ||
7c673cae FG |
73 | MDSDaemon *mds = NULL; |
74 | ||
75 | ||
76 | static void handle_mds_signal(int signum) | |
77 | { | |
78 | if (mds) | |
79 | mds->handle_signal(signum); | |
80 | } | |
81 | ||
7c673cae | 82 | int main(int argc, const char **argv) |
7c673cae FG |
83 | { |
84 | ceph_pthread_setname(pthread_self(), "ceph-mds"); | |
85 | ||
20effc67 | 86 | auto args = argv_to_vec(argc, argv); |
11fdf7f2 TL |
87 | if (args.empty()) { |
88 | cerr << argv[0] << ": -h or --help for usage" << std::endl; | |
89 | exit(1); | |
90 | } | |
91 | if (ceph_argparse_need_usage(args)) { | |
92 | usage(); | |
93 | exit(0); | |
94 | } | |
7c673cae FG |
95 | |
96 | auto cct = global_init(NULL, args, | |
f67539c2 | 97 | CEPH_ENTITY_TYPE_MDS, CODE_ENVIRONMENT_DAEMON, 0); |
7c673cae FG |
98 | ceph_heap_profiler_init(); |
99 | ||
9f95a23c TL |
100 | int numa_node = g_conf().get_val<int64_t>("mds_numa_node"); |
101 | size_t numa_cpu_set_size = 0; | |
102 | cpu_set_t numa_cpu_set; | |
103 | if (numa_node >= 0) { | |
104 | int r = get_numa_node_cpu_set(numa_node, &numa_cpu_set_size, &numa_cpu_set); | |
105 | if (r < 0) { | |
106 | dout(1) << __func__ << " unable to determine mds numa node " << numa_node | |
107 | << " CPUs" << dendl; | |
108 | numa_node = -1; | |
109 | } else { | |
110 | r = set_cpu_affinity_all_threads(numa_cpu_set_size, &numa_cpu_set); | |
111 | if (r < 0) { | |
112 | derr << __func__ << " failed to set numa affinity: " << cpp_strerror(r) | |
113 | << dendl; | |
114 | } | |
115 | } | |
116 | } else { | |
117 | dout(1) << __func__ << " not setting numa affinity" << dendl; | |
118 | } | |
7c673cae FG |
119 | std::string val, action; |
120 | for (std::vector<const char*>::iterator i = args.begin(); i != args.end(); ) { | |
121 | if (ceph_argparse_double_dash(args, i)) { | |
122 | break; | |
123 | } | |
7c673cae | 124 | else if (ceph_argparse_witharg(args, i, &val, "--hot-standby", (char*)NULL)) { |
11fdf7f2 | 125 | dout(0) << "--hot-standby is obsolete and has no effect" << dendl; |
7c673cae FG |
126 | } |
127 | else { | |
128 | derr << "Error: can't understand argument: " << *i << "\n" << dendl; | |
11fdf7f2 | 129 | exit(1); |
7c673cae FG |
130 | } |
131 | } | |
132 | ||
11fdf7f2 TL |
133 | Preforker forker; |
134 | ||
135 | entity_addrvec_t addrs; | |
136 | pick_addresses(g_ceph_context, CEPH_PICK_ADDRESS_PUBLIC, &addrs); | |
7c673cae FG |
137 | |
138 | // Normal startup | |
11fdf7f2 | 139 | if (g_conf()->name.has_default_id()) { |
7c673cae | 140 | derr << "must specify '-i name' with the ceph-mds instance name" << dendl; |
11fdf7f2 | 141 | exit(1); |
7c673cae FG |
142 | } |
143 | ||
11fdf7f2 TL |
144 | if (g_conf()->name.get_id().empty() || |
145 | (g_conf()->name.get_id()[0] >= '0' && g_conf()->name.get_id()[0] <= '9')) { | |
146 | derr << "MDS id '" << g_conf()->name << "' is invalid. " | |
7c673cae | 147 | "MDS names may not start with a numeric digit." << dendl; |
11fdf7f2 TL |
148 | exit(1); |
149 | } | |
150 | ||
151 | if (global_init_prefork(g_ceph_context) >= 0) { | |
152 | std::string err; | |
153 | int r = forker.prefork(err); | |
154 | if (r < 0) { | |
155 | cerr << err << std::endl; | |
156 | return r; | |
157 | } | |
158 | if (forker.is_parent()) { | |
159 | if (forker.parent_wait(err) != 0) { | |
160 | return -ENXIO; | |
161 | } | |
162 | return 0; | |
163 | } | |
164 | global_init_postfork_start(g_ceph_context); | |
7c673cae | 165 | } |
11fdf7f2 TL |
166 | common_init_finish(g_ceph_context); |
167 | global_init_chdir(g_ceph_context); | |
7c673cae | 168 | |
11fdf7f2 | 169 | std::string public_msgr_type = g_conf()->ms_public_type.empty() ? g_conf().get_val<std::string>("ms_type") : g_conf()->ms_public_type; |
7c673cae FG |
170 | Messenger *msgr = Messenger::create(g_ceph_context, public_msgr_type, |
171 | entity_name_t::MDS(-1), "mds", | |
f67539c2 | 172 | Messenger::get_random_nonce()); |
7c673cae | 173 | if (!msgr) |
11fdf7f2 | 174 | forker.exit(1); |
7c673cae FG |
175 | msgr->set_cluster_protocol(CEPH_MDS_PROTOCOL); |
176 | ||
11fdf7f2 | 177 | cout << "starting " << g_conf()->name << " at " << msgr->get_myaddrs() |
7c673cae FG |
178 | << std::endl; |
179 | uint64_t required = | |
180 | CEPH_FEATURE_OSDREPLYMUX; | |
181 | ||
182 | msgr->set_default_policy(Messenger::Policy::lossy_client(required)); | |
183 | msgr->set_policy(entity_name_t::TYPE_MON, | |
184 | Messenger::Policy::lossy_client(CEPH_FEATURE_UID | | |
185 | CEPH_FEATURE_PGID64)); | |
186 | msgr->set_policy(entity_name_t::TYPE_MDS, | |
187 | Messenger::Policy::lossless_peer(CEPH_FEATURE_UID)); | |
188 | msgr->set_policy(entity_name_t::TYPE_CLIENT, | |
189 | Messenger::Policy::stateful_server(0)); | |
190 | ||
11fdf7f2 | 191 | int r = msgr->bindv(addrs); |
7c673cae | 192 | if (r < 0) |
11fdf7f2 | 193 | forker.exit(1); |
7c673cae | 194 | |
11fdf7f2 TL |
195 | // set up signal handlers, now that we've daemonized/forked. |
196 | init_async_signal_handler(); | |
197 | register_async_signal_handler(SIGHUP, sighup_handler); | |
198 | ||
7c673cae | 199 | // get monmap |
f67539c2 TL |
200 | ceph::async::io_context_pool ctxpool(2); |
201 | MonClient mc(g_ceph_context, ctxpool); | |
7c673cae | 202 | if (mc.build_initial_monmap() < 0) |
11fdf7f2 | 203 | forker.exit(1); |
7c673cae FG |
204 | global_init_chdir(g_ceph_context); |
205 | ||
206 | msgr->start(); | |
207 | ||
208 | // start mds | |
f67539c2 | 209 | mds = new MDSDaemon(g_conf()->name.get_id().c_str(), msgr, &mc, ctxpool); |
7c673cae FG |
210 | |
211 | // in case we have to respawn... | |
212 | mds->orig_argc = argc; | |
213 | mds->orig_argv = argv; | |
214 | ||
11fdf7f2 TL |
215 | if (g_conf()->daemonize) { |
216 | global_init_postfork_finish(g_ceph_context); | |
217 | forker.daemonize(); | |
218 | } | |
219 | ||
7c673cae FG |
220 | r = mds->init(); |
221 | if (r < 0) { | |
222 | msgr->wait(); | |
223 | goto shutdown; | |
224 | } | |
225 | ||
7c673cae FG |
226 | register_async_signal_handler_oneshot(SIGINT, handle_mds_signal); |
227 | register_async_signal_handler_oneshot(SIGTERM, handle_mds_signal); | |
228 | ||
11fdf7f2 | 229 | if (g_conf()->inject_early_sigterm) |
7c673cae FG |
230 | kill(getpid(), SIGTERM); |
231 | ||
232 | msgr->wait(); | |
233 | ||
234 | unregister_async_signal_handler(SIGHUP, sighup_handler); | |
235 | unregister_async_signal_handler(SIGINT, handle_mds_signal); | |
236 | unregister_async_signal_handler(SIGTERM, handle_mds_signal); | |
237 | shutdown_async_signal_handler(); | |
238 | ||
239 | shutdown: | |
f67539c2 | 240 | ctxpool.stop(); |
7c673cae FG |
241 | // yuck: grab the mds lock, so we can be sure that whoever in *mds |
242 | // called shutdown finishes what they were doing. | |
9f95a23c TL |
243 | mds->mds_lock.lock(); |
244 | mds->mds_lock.unlock(); | |
7c673cae FG |
245 | |
246 | pidfile_remove(); | |
247 | ||
248 | // only delete if it was a clean shutdown (to aid memory leak | |
249 | // detection, etc.). don't bother if it was a suicide. | |
250 | if (mds->is_clean_shutdown()) { | |
251 | delete mds; | |
252 | delete msgr; | |
253 | } | |
254 | ||
255 | // cd on exit, so that gmon.out (if any) goes into a separate directory for each node. | |
256 | char s[20]; | |
257 | snprintf(s, sizeof(s), "gmon/%d", getpid()); | |
258 | if ((mkdir(s, 0755) == 0) && (chdir(s) == 0)) { | |
259 | cerr << "ceph-mds: gmon.out should be in " << s << std::endl; | |
260 | } | |
261 | ||
262 | return 0; | |
263 | } |