server/ctdb_monitor.c

   1 /*
   2    monitoring links to all other nodes to detect dead nodes
   3
   4
   5    Copyright (C) Ronnie Sahlberg 2007
   6
   7    This program is free software; you can redistribute it and/or modify
   8    it under the terms of the GNU General Public License as published by
   9    the Free Software Foundation; either version 3 of the License, or
  10    (at your option) any later version.
  11
  12    This program is distributed in the hope that it will be useful,
  13    but WITHOUT ANY WARRANTY; without even the implied warranty of
  14    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  15    GNU General Public License for more details.
  16
  17    You should have received a copy of the GNU General Public License
  18    along with this program; if not, see <http://www.gnu.org/licenses/>.
  19 */
  20
  21 #include "includes.h"
  22 #include "lib/events/events.h"
  23 #include "system/filesys.h"
  24 #include "system/wait.h"
  25 #include "../include/ctdb_private.h"
  26
  27 struct ctdb_monitor_state {
  28         uint32_t monitoring_mode;
  29         TALLOC_CTX *monitor_context;
  30         uint32_t next_interval;
  31 };
  32
  33 static void ctdb_check_health(struct event_context *ev, struct timed_event *te,
  34                               struct timeval t, void *private_data);
  35
  36 /*
  37   setup the notification script
  38 */
  39 int ctdb_set_notification_script(struct ctdb_context *ctdb, const char *script)
  40 {
  41         ctdb->notification_script = talloc_strdup(ctdb, script);
  42         CTDB_NO_MEMORY(ctdb, ctdb->notification_script);
  43         return 0;
  44 }
  45
  46 static int ctdb_run_notification_script_child(struct ctdb_context *ctdb, const char *event)
  47 {
  48         struct stat st;
  49         int ret;
  50         char *cmd;
  51
  52         if (stat(ctdb->notification_script, &st) != 0) {
  53                 DEBUG(DEBUG_ERR,("Could not stat notification script %s. Can not send notifications.\n", ctdb->notification_script));
  54                 return -1;
  55         }
  56         if (!(st.st_mode & S_IXUSR)) {
  57                 DEBUG(DEBUG_ERR,("Notification script %s is not executable.\n", ctdb->notification_script));
  58                 return -1;
  59         }
  60
  61         cmd = talloc_asprintf(ctdb, "%s %s\n", ctdb->notification_script, event);
  62         CTDB_NO_MEMORY(ctdb, cmd);
  63
  64         ret = system(cmd);
  65         /* if the system() call was successful, translate ret into the
  66            return code from the command
  67         */
  68         if (ret != -1) {
  69                 ret = WEXITSTATUS(ret);
  70         }
  71         if (ret != 0) {
  72                 DEBUG(DEBUG_ERR,("Notification script \"%s\" failed with error %d\n", cmd, ret));
  73         }
  74
  75         return ret;
  76 }
  77
  78 static void ctdb_run_notification_script(struct ctdb_context *ctdb, const char *event)
  79 {
  80         pid_t child;
  81
  82         if (ctdb->notification_script == NULL) {
  83                 return;
  84         }
  85
  86         child = fork();
  87         if (child == (pid_t)-1) {
  88                 DEBUG(DEBUG_ERR,("Failed to fork() a notification child process\n"));
  89                 return;
  90         }
  91         if (child == 0) {
  92                 int ret;
  93
  94                 ret = ctdb_run_notification_script_child(ctdb, event);
  95                 if (ret != 0) {
  96                         DEBUG(DEBUG_ERR,(__location__ " Notification script failed\n"));
  97                 }
  98                 _exit(0);
  99         }
 100
 101         return;
 102 }
 103
 104 /*
 105   called when a health monitoring event script finishes
 106  */
 107 static void ctdb_health_callback(struct ctdb_context *ctdb, int status, void *p)
 108 {
 109         struct ctdb_node *node = ctdb->nodes[ctdb->pnn];
 110         TDB_DATA data;
 111         struct ctdb_node_flag_change c;
 112         uint32_t next_interval;
 113         int ret;
 114         TDB_DATA rddata;
 115         struct takeover_run_reply rd;
 116
 117         c.pnn = ctdb->pnn;
 118         c.old_flags = node->flags;
 119
 120         rd.pnn   = ctdb->pnn;
 121         rd.srvid = CTDB_SRVID_TAKEOVER_RUN_RESPONSE;
 122
 123         rddata.dptr = (uint8_t *)&rd;
 124         rddata.dsize = sizeof(rd);
 125
 126         if (status != 0 && !(node->flags & NODE_FLAGS_UNHEALTHY)) {
 127                 DEBUG(DEBUG_NOTICE,("monitor event failed - disabling node\n"));
 128                 node->flags |= NODE_FLAGS_UNHEALTHY;
 129                 ctdb->monitor->next_interval = 5;
 130                 if (ctdb->tunable.disable_when_unhealthy != 0) {
 131                         DEBUG(DEBUG_INFO, ("DISABLING node since it became unhealthy\n"));
 132                         node->flags |= NODE_FLAGS_DISABLED;
 133                 }
 134
 135                 ctdb_run_notification_script(ctdb, "unhealthy");
 136
 137                 /* ask the recmaster to reallocate all addresses */
 138                 DEBUG(DEBUG_ERR,("Node became UNHEALTHY. Ask recovery master %u to perform ip reallocation\n", ctdb->recovery_master));
 139                 ret = ctdb_daemon_send_message(ctdb, ctdb->recovery_master, CTDB_SRVID_TAKEOVER_RUN, rddata);
 140                 if (ret != 0) {
 141                         DEBUG(DEBUG_ERR,(__location__ " Failed to send ip takeover run request message to %u\n", ctdb->recovery_master));
 142                 }
 143
 144         } else if (status == 0 && (node->flags & NODE_FLAGS_UNHEALTHY)) {
 145                 DEBUG(DEBUG_NOTICE,("monitor event OK - node re-enabled\n"));
 146                 node->flags &= ~NODE_FLAGS_UNHEALTHY;
 147                 ctdb->monitor->next_interval = 5;
 148
 149                 ctdb_run_notification_script(ctdb, "healthy");
 150
 151                 /* ask the recmaster to reallocate all addresses */
 152                 DEBUG(DEBUG_ERR,("Node became HEALTHY. Ask recovery master %u to perform ip reallocation\n", ctdb->recovery_master));
 153                 ret = ctdb_daemon_send_message(ctdb, ctdb->recovery_master, CTDB_SRVID_TAKEOVER_RUN, rddata);
 154                 if (ret != 0) {
 155                         DEBUG(DEBUG_ERR,(__location__ " Failed to send ip takeover run request message to %u\n", ctdb->recovery_master));
 156                 }
 157
 158         }
 159
 160         next_interval = ctdb->monitor->next_interval;
 161
 162         ctdb->monitor->next_interval *= 2;
 163         if (ctdb->monitor->next_interval > ctdb->tunable.monitor_interval) {
 164                 ctdb->monitor->next_interval = ctdb->tunable.monitor_interval;
 165         }
 166
 167         event_add_timed(ctdb->ev, ctdb->monitor->monitor_context,
 168                                 timeval_current_ofs(next_interval, 0),
 169                                 ctdb_check_health, ctdb);
 170
 171         if (c.old_flags == node->flags) {
 172                 return;
 173         }
 174
 175         c.new_flags = node->flags;
 176
 177         data.dptr = (uint8_t *)&c;
 178         data.dsize = sizeof(c);
 179
 180         /* ask the recovery daemon to push these changes out to all nodes */
 181         ctdb_daemon_send_message(ctdb, ctdb->pnn,
 182                                  CTDB_SRVID_PUSH_NODE_FLAGS, data);
 183
 184 }
 185
 186
 187 /*
 188   called when the startup event script finishes
 189  */
 190 static void ctdb_startup_callback(struct ctdb_context *ctdb, int status, void *p)
 191 {
 192         if (status != 0) {
 193                 DEBUG(DEBUG_ERR,("startup event failed\n"));
 194         } else if (status == 0) {
 195                 DEBUG(DEBUG_NOTICE,("startup event OK - enabling monitoring\n"));
 196                 ctdb->done_startup = true;
 197                 ctdb->monitor->next_interval = 5;
 198                 ctdb_run_notification_script(ctdb, "startup");
 199         }
 200
 201         event_add_timed(ctdb->ev, ctdb->monitor->monitor_context,
 202                         timeval_current_ofs(ctdb->monitor->next_interval, 0),
 203                         ctdb_check_health, ctdb);
 204 }
 205
 206
 207 /*
 208   wait until we have finished initial recoveries before we start the
 209   monitoring events
 210  */
 211 static void ctdb_wait_until_recovered(struct event_context *ev, struct timed_event *te,
 212                               struct timeval t, void *private_data)
 213 {
 214         struct ctdb_context *ctdb = talloc_get_type(private_data, struct ctdb_context);
 215
 216         DEBUG(DEBUG_NOTICE,("CTDB_WAIT_UNTIL_RECOVERED\n"));
 217
 218         if (ctdb->vnn_map->generation == INVALID_GENERATION) {
 219                 DEBUG(DEBUG_NOTICE,(__location__ " generation is INVALID. Wait one more second\n"));
 220                 event_add_timed(ctdb->ev, ctdb->monitor->monitor_context,
 221                                      timeval_current_ofs(1, 0),
 222                                      ctdb_wait_until_recovered, ctdb);
 223                 return;
 224         }
 225
 226         if (ctdb->recovery_mode != CTDB_RECOVERY_NORMAL) {
 227                 DEBUG(DEBUG_NOTICE,(__location__ " in recovery. Wait one more second\n"));
 228                 event_add_timed(ctdb->ev, ctdb->monitor->monitor_context,
 229                                      timeval_current_ofs(1, 0),
 230                                      ctdb_wait_until_recovered, ctdb);
 231                 return;
 232         }
 233
 234
 235         if (timeval_elapsed(&ctdb->last_recovery_finished) < (ctdb->tunable.rerecovery_timeout + 3)) {
 236                 DEBUG(DEBUG_NOTICE,(__location__ " wait for pending recoveries to end. Wait one more second.\n"));
 237
 238                 event_add_timed(ctdb->ev, ctdb->monitor->monitor_context,
 239                                      timeval_current_ofs(1, 0),
 240                                      ctdb_wait_until_recovered, ctdb);
 241                 return;
 242         }
 243
 244
 245         DEBUG(DEBUG_NOTICE,(__location__ " Recoveries finished. Running the \"startup\" event.\n"));
 246         event_add_timed(ctdb->ev, ctdb->monitor->monitor_context,
 247                              timeval_current_ofs(1, 0),
 248                              ctdb_check_health, ctdb);
 249 }
 250
 251
 252 /*
 253   see if the event scripts think we are healthy
 254  */
 255 static void ctdb_check_health(struct event_context *ev, struct timed_event *te,
 256                               struct timeval t, void *private_data)
 257 {
 258         struct ctdb_context *ctdb = talloc_get_type(private_data, struct ctdb_context);
 259         int ret = 0;
 260
 261         if (ctdb->recovery_mode != CTDB_RECOVERY_NORMAL ||
 262             (ctdb->monitor->monitoring_mode == CTDB_MONITORING_DISABLED && ctdb->done_startup)) {
 263                 event_add_timed(ctdb->ev, ctdb->monitor->monitor_context,
 264                                 timeval_current_ofs(ctdb->monitor->next_interval, 0),
 265                                 ctdb_check_health, ctdb);
 266                 return;
 267         }
 268
 269         if (!ctdb->done_startup) {
 270                 ret = ctdb_event_script_callback(ctdb,
 271                                                  ctdb->monitor->monitor_context, ctdb_startup_callback,
 272                                                  ctdb, false,
 273                                                  CTDB_EVENT_STARTUP, "%s", "");
 274         } else {
 275                 int i;
 276                 int skip_monitoring = 0;
 277
 278                 if (ctdb->recovery_mode != CTDB_RECOVERY_NORMAL) {
 279                         skip_monitoring = 1;
 280                         DEBUG(DEBUG_ERR,("Skip monitoring during recovery\n"));
 281                 }
 282                 for (i=1; i<=NUM_DB_PRIORITIES; i++) {
 283                         if (ctdb->freeze_handles[i] != NULL) {
 284                                 DEBUG(DEBUG_ERR,("Skip monitoring since databases are frozen\n"));
 285                                 skip_monitoring = 1;
 286                                 break;
 287                         }
 288                 }
 289                 if (skip_monitoring != 0) {
 290                         event_add_timed(ctdb->ev, ctdb->monitor->monitor_context,
 291                                         timeval_current_ofs(ctdb->monitor->next_interval, 0),
 292                                         ctdb_check_health, ctdb);
 293                         return;
 294                 } else {
 295                         ret = ctdb_event_script_callback(ctdb,
 296                                         ctdb->monitor->monitor_context, ctdb_health_callback,
 297                                         ctdb, false,
 298                                         CTDB_EVENT_MONITOR, "%s", "");
 299                 }
 300         }
 301
 302         if (ret != 0) {
 303                 DEBUG(DEBUG_ERR,("Unable to launch monitor event script\n"));
 304                 ctdb->monitor->next_interval = 5;
 305                 event_add_timed(ctdb->ev, ctdb->monitor->monitor_context,
 306                         timeval_current_ofs(5, 0),
 307                         ctdb_check_health, ctdb);
 308         }
 309 }
 310
 311 /*
 312   (Temporaily) Disabling monitoring will stop the monitor event scripts
 313   from running   but node health checks will still occur
 314 */
 315 void ctdb_disable_monitoring(struct ctdb_context *ctdb)
 316 {
 317         ctdb->monitor->monitoring_mode = CTDB_MONITORING_DISABLED;
 318         DEBUG(DEBUG_INFO,("Monitoring has been disabled\n"));
 319 }
 320
 321 /*
 322    Re-enable running monitor events after they have been disabled
 323  */
 324 void ctdb_enable_monitoring(struct ctdb_context *ctdb)
 325 {
 326         ctdb->monitor->monitoring_mode  = CTDB_MONITORING_ACTIVE;
 327         ctdb->monitor->next_interval = 5;
 328         DEBUG(DEBUG_INFO,("Monitoring has been enabled\n"));
 329 }
 330
 331 /* stop any monitoring
 332    this should only be done when shutting down the daemon
 333 */
 334 void ctdb_stop_monitoring(struct ctdb_context *ctdb)
 335 {
 336         talloc_free(ctdb->monitor->monitor_context);
 337         ctdb->monitor->monitor_context = NULL;
 338
 339         ctdb->monitor->monitoring_mode  = CTDB_MONITORING_DISABLED;
 340         ctdb->monitor->next_interval = 5;
 341         DEBUG(DEBUG_NOTICE,("Monitoring has been stopped\n"));
 342 }
 343
 344 /*
 345   start watching for nodes that might be dead
 346  */
 347 void ctdb_start_monitoring(struct ctdb_context *ctdb)
 348 {
 349         if (ctdb->monitor != NULL) {
 350                 return;
 351         }
 352
 353         ctdb->monitor = talloc(ctdb, struct ctdb_monitor_state);
 354         CTDB_NO_MEMORY_FATAL(ctdb, ctdb->monitor);
 355
 356         ctdb->monitor->next_interval = 5;
 357
 358         ctdb->monitor->monitor_context = talloc_new(ctdb->monitor);
 359         CTDB_NO_MEMORY_FATAL(ctdb, ctdb->monitor->monitor_context);
 360
 361         event_add_timed(ctdb->ev, ctdb->monitor->monitor_context,
 362                              timeval_current_ofs(1, 0),
 363                              ctdb_wait_until_recovered, ctdb);
 364
 365         ctdb->monitor->monitoring_mode  = CTDB_MONITORING_ACTIVE;
 366         DEBUG(DEBUG_NOTICE,("Monitoring has been started\n"));
 367 }
 368
 369
 370 /*
 371   modify flags on a node
 372  */
 373 int32_t ctdb_control_modflags(struct ctdb_context *ctdb, TDB_DATA indata)
 374 {
 375         struct ctdb_node_flag_change *c = (struct ctdb_node_flag_change *)indata.dptr;
 376         struct ctdb_node *node;
 377         uint32_t old_flags;
 378         int i;
 379
 380         if (c->pnn >= ctdb->num_nodes) {
 381                 DEBUG(DEBUG_ERR,(__location__ " Node %d is invalid, num_nodes :%d\n", c->pnn, ctdb->num_nodes));
 382                 return -1;
 383         }
 384
 385         node         = ctdb->nodes[c->pnn];
 386         old_flags    = node->flags;
 387         if (c->pnn != ctdb->pnn) {
 388                 c->old_flags  = node->flags;
 389         }
 390         node->flags   = c->new_flags & ~NODE_FLAGS_DISCONNECTED;
 391         node->flags  |= (c->old_flags & NODE_FLAGS_DISCONNECTED);
 392
 393         /* we dont let other nodes modify our STOPPED status */
 394         if (c->pnn == ctdb->pnn) {
 395                 node->flags &= ~NODE_FLAGS_STOPPED;
 396                 if (old_flags & NODE_FLAGS_STOPPED) {
 397                         node->flags |= NODE_FLAGS_STOPPED;
 398                 }
 399         }
 400
 401         /* we dont let other nodes modify our BANNED status */
 402         if (c->pnn == ctdb->pnn) {
 403                 node->flags &= ~NODE_FLAGS_BANNED;
 404                 if (old_flags & NODE_FLAGS_BANNED) {
 405                         node->flags |= NODE_FLAGS_BANNED;
 406                 }
 407         }
 408
 409         if (node->flags == c->old_flags) {
 410                 DEBUG(DEBUG_INFO, ("Control modflags on node %u - Unchanged - flags 0x%x\n", c->pnn, node->flags));
 411                 return 0;
 412         }
 413
 414         DEBUG(DEBUG_INFO, ("Control modflags on node %u - flags now 0x%x\n", c->pnn, node->flags));
 415
 416
 417         /* tell the recovery daemon something has changed */
 418         ctdb_daemon_send_message(ctdb, ctdb->pnn,
 419                                  CTDB_SRVID_SET_NODE_FLAGS, indata);
 420
 421         /* if we have become banned, we should go into recovery mode */
 422         if ((node->flags & NODE_FLAGS_BANNED) && !(c->old_flags & NODE_FLAGS_BANNED) && (node->pnn == ctdb->pnn)) {
 423                 /* make sure we are frozen */
 424                 DEBUG(DEBUG_NOTICE,("This node has been banned - forcing freeze and recovery\n"));
 425                 /* Reset the generation id to 1 to make us ignore any
 426                    REQ/REPLY CALL/DMASTER someone sends to us.
 427                    We are now banned so we shouldnt service database calls
 428                    anymore.
 429                 */
 430                 ctdb->vnn_map->generation = INVALID_GENERATION;
 431
 432                 for (i=1; i<=NUM_DB_PRIORITIES; i++) {
 433                         if (ctdb_start_freeze(ctdb, i) != 0) {
 434                                 DEBUG(DEBUG_ERR,(__location__ " Failed to freeze db priority %u\n", i));
 435                         }
 436                 }
 437                 ctdb_release_all_ips(ctdb);
 438                 ctdb->recovery_mode = CTDB_RECOVERY_ACTIVE;
 439         }
 440
 441         return 0;
 442 }
 443
 444 /*
 445   return the monitoring mode
 446  */
 447 int32_t ctdb_monitoring_mode(struct ctdb_context *ctdb)
 448 {
 449         if (ctdb->monitor == NULL) {
 450                 return CTDB_MONITORING_DISABLED;
 451         }
 452         return ctdb->monitor->monitoring_mode;
 453 }
 454