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/tevent/tevent.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 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 = ctdb_fork(ctdb);
  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                 debug_extra = talloc_asprintf(NULL, "notification-%s:", event);
  95                 ret = ctdb_run_notification_script_child(ctdb, event);
  96                 if (ret != 0) {
  97                         DEBUG(DEBUG_ERR,(__location__ " Notification script failed\n"));
  98                 }
  99                 _exit(0);
 100         }
 101
 102         return;
 103 }
 104
 105 /*
 106   called when a health monitoring event script finishes
 107  */
 108 static void ctdb_health_callback(struct ctdb_context *ctdb, int status, void *p)
 109 {
 110         struct ctdb_node *node = ctdb->nodes[ctdb->pnn];
 111         TDB_DATA data;
 112         struct ctdb_node_flag_change c;
 113         uint32_t next_interval;
 114         int ret;
 115         TDB_DATA rddata;
 116         struct takeover_run_reply rd;
 117
 118         c.pnn = ctdb->pnn;
 119         c.old_flags = node->flags;
 120
 121         rd.pnn   = ctdb->pnn;
 122         rd.srvid = CTDB_SRVID_TAKEOVER_RUN_RESPONSE;
 123
 124         rddata.dptr = (uint8_t *)&rd;
 125         rddata.dsize = sizeof(rd);
 126
 127         if (status == -ETIME) {
 128                 ctdb->event_script_timeouts++;
 129
 130                 if (ctdb->event_script_timeouts >= ctdb->tunable.script_timeout_count) {
 131                         DEBUG(DEBUG_ERR, ("Maximum timeout count %u reached for eventscript. Making node unhealthy\n", ctdb->tunable.script_timeout_count));
 132                 } else {
 133                         /* We pretend this is OK. */
 134                         goto after_change_status;
 135                 }
 136         }
 137
 138         if (status != 0 && !(node->flags & NODE_FLAGS_UNHEALTHY)) {
 139                 DEBUG(DEBUG_NOTICE,("monitor event failed - disabling node\n"));
 140                 node->flags |= NODE_FLAGS_UNHEALTHY;
 141                 ctdb->monitor->next_interval = 5;
 142
 143                 ctdb_run_notification_script(ctdb, "unhealthy");
 144
 145                 /* ask the recmaster to reallocate all addresses */
 146                 DEBUG(DEBUG_ERR,("Node became UNHEALTHY. Ask recovery master %u to perform ip reallocation\n", ctdb->recovery_master));
 147                 ret = ctdb_daemon_send_message(ctdb, ctdb->recovery_master, CTDB_SRVID_TAKEOVER_RUN, rddata);
 148                 if (ret != 0) {
 149                         DEBUG(DEBUG_ERR,(__location__ " Failed to send ip takeover run request message to %u\n", ctdb->recovery_master));
 150                 }
 151
 152         } else if (status == 0 && (node->flags & NODE_FLAGS_UNHEALTHY)) {
 153                 DEBUG(DEBUG_NOTICE,("monitor event OK - node re-enabled\n"));
 154                 node->flags &= ~NODE_FLAGS_UNHEALTHY;
 155                 ctdb->monitor->next_interval = 5;
 156
 157                 ctdb_run_notification_script(ctdb, "healthy");
 158
 159                 /* ask the recmaster to reallocate all addresses */
 160                 DEBUG(DEBUG_ERR,("Node became HEALTHY. Ask recovery master %u to perform ip reallocation\n", ctdb->recovery_master));
 161                 ret = ctdb_daemon_send_message(ctdb, ctdb->recovery_master, CTDB_SRVID_TAKEOVER_RUN, rddata);
 162                 if (ret != 0) {
 163                         DEBUG(DEBUG_ERR,(__location__ " Failed to send ip takeover run request message to %u\n", ctdb->recovery_master));
 164                 }
 165
 166         }
 167
 168 after_change_status:
 169         next_interval = ctdb->monitor->next_interval;
 170
 171         ctdb->monitor->next_interval *= 2;
 172         if (ctdb->monitor->next_interval > ctdb->tunable.monitor_interval) {
 173                 ctdb->monitor->next_interval = ctdb->tunable.monitor_interval;
 174         }
 175
 176         event_add_timed(ctdb->ev, ctdb->monitor->monitor_context,
 177                                 timeval_current_ofs(next_interval, 0),
 178                                 ctdb_check_health, ctdb);
 179
 180         if (c.old_flags == node->flags) {
 181                 return;
 182         }
 183
 184         c.new_flags = node->flags;
 185
 186         data.dptr = (uint8_t *)&c;
 187         data.dsize = sizeof(c);
 188
 189         /* ask the recovery daemon to push these changes out to all nodes */
 190         ctdb_daemon_send_message(ctdb, ctdb->pnn,
 191                                  CTDB_SRVID_PUSH_NODE_FLAGS, data);
 192
 193 }
 194
 195
 196 /*
 197   called when the startup event script finishes
 198  */
 199 static void ctdb_startup_callback(struct ctdb_context *ctdb, int status, void *p)
 200 {
 201         if (status != 0) {
 202                 DEBUG(DEBUG_ERR,("startup event failed\n"));
 203         } else if (status == 0) {
 204                 DEBUG(DEBUG_NOTICE,("startup event OK - enabling monitoring\n"));
 205                 ctdb->done_startup = true;
 206                 ctdb->monitor->next_interval = 2;
 207                 ctdb_run_notification_script(ctdb, "startup");
 208         }
 209
 210         event_add_timed(ctdb->ev, ctdb->monitor->monitor_context,
 211                         timeval_current_ofs(ctdb->monitor->next_interval, 0),
 212                         ctdb_check_health, ctdb);
 213 }
 214
 215
 216 /*
 217   wait until we have finished initial recoveries before we start the
 218   monitoring events
 219  */
 220 static void ctdb_wait_until_recovered(struct event_context *ev, struct timed_event *te,
 221                               struct timeval t, void *private_data)
 222 {
 223         struct ctdb_context *ctdb = talloc_get_type(private_data, struct ctdb_context);
 224         int ret;
 225
 226         DEBUG(DEBUG_NOTICE,("CTDB_WAIT_UNTIL_RECOVERED\n"));
 227         if (ctdb->nodes[ctdb->pnn]->flags & NODE_FLAGS_STOPPED) {
 228                 DEBUG(DEBUG_NOTICE,("Node is STOPPED. Node will NOT recover.\n"));
 229         }
 230
 231         if (ctdb->vnn_map->generation == INVALID_GENERATION) {
 232                 ctdb->db_persistent_startup_generation = INVALID_GENERATION;
 233
 234                 DEBUG(DEBUG_NOTICE,(__location__ " generation is INVALID. Wait one more second\n"));
 235                 event_add_timed(ctdb->ev, ctdb->monitor->monitor_context,
 236                                      timeval_current_ofs(1, 0),
 237                                      ctdb_wait_until_recovered, ctdb);
 238                 return;
 239         }
 240
 241         if (ctdb->recovery_mode != CTDB_RECOVERY_NORMAL) {
 242                 ctdb->db_persistent_startup_generation = INVALID_GENERATION;
 243
 244                 DEBUG(DEBUG_NOTICE,(__location__ " in recovery. Wait one more second\n"));
 245                 event_add_timed(ctdb->ev, ctdb->monitor->monitor_context,
 246                                      timeval_current_ofs(1, 0),
 247                                      ctdb_wait_until_recovered, ctdb);
 248                 return;
 249         }
 250
 251
 252         if (!fast_start && timeval_elapsed(&ctdb->last_recovery_finished) < (ctdb->tunable.rerecovery_timeout + 3)) {
 253                 ctdb->db_persistent_startup_generation = INVALID_GENERATION;
 254
 255                 DEBUG(DEBUG_NOTICE,(__location__ " wait for pending recoveries to end. Wait one more second.\n"));
 256
 257                 event_add_timed(ctdb->ev, ctdb->monitor->monitor_context,
 258                                      timeval_current_ofs(1, 0),
 259                                      ctdb_wait_until_recovered, ctdb);
 260                 return;
 261         }
 262
 263         if (ctdb->vnn_map->generation == ctdb->db_persistent_startup_generation) {
 264                 DEBUG(DEBUG_INFO,(__location__ " skip ctdb_recheck_persistent_health() "
 265                                   "until the next recovery\n"));
 266                 event_add_timed(ctdb->ev, ctdb->monitor->monitor_context,
 267                                      timeval_current_ofs(1, 0),
 268                                      ctdb_wait_until_recovered, ctdb);
 269                 return;
 270         }
 271
 272         ctdb->db_persistent_startup_generation = ctdb->vnn_map->generation;
 273         ret = ctdb_recheck_persistent_health(ctdb);
 274         if (ret != 0) {
 275                 ctdb->db_persistent_check_errors++;
 276                 if (ctdb->db_persistent_check_errors < ctdb->max_persistent_check_errors) {
 277                         DEBUG(ctdb->db_persistent_check_errors==1?DEBUG_ERR:DEBUG_WARNING,
 278                               (__location__ "ctdb_recheck_persistent_health() "
 279                               "failed (%llu of %llu times) - retry later\n",
 280                               (unsigned long long)ctdb->db_persistent_check_errors,
 281                               (unsigned long long)ctdb->max_persistent_check_errors));
 282                         event_add_timed(ctdb->ev,
 283                                         ctdb->monitor->monitor_context,
 284                                         timeval_current_ofs(1, 0),
 285                                         ctdb_wait_until_recovered, ctdb);
 286                         return;
 287                 }
 288                 DEBUG(DEBUG_ALERT,(__location__
 289                                   "ctdb_recheck_persistent_health() failed (%llu times) - prepare shutdown\n",
 290                                   (unsigned long long)ctdb->db_persistent_check_errors));
 291                 ctdb_stop_recoverd(ctdb);
 292                 ctdb_stop_keepalive(ctdb);
 293                 ctdb_stop_monitoring(ctdb);
 294                 ctdb_release_all_ips(ctdb);
 295                 if (ctdb->methods != NULL) {
 296                         ctdb->methods->shutdown(ctdb);
 297                 }
 298                 ctdb_event_script(ctdb, CTDB_EVENT_SHUTDOWN);
 299                 DEBUG(DEBUG_ALERT,("ctdb_recheck_persistent_health() failed - Stopping CTDB daemon\n"));
 300                 exit(11);
 301         }
 302         ctdb->db_persistent_check_errors = 0;
 303         DEBUG(DEBUG_NOTICE,(__location__
 304                            "ctdb_start_monitoring: ctdb_recheck_persistent_health() OK\n"));
 305
 306         DEBUG(DEBUG_NOTICE,(__location__ " Recoveries finished. Running the \"startup\" event.\n"));
 307         event_add_timed(ctdb->ev, ctdb->monitor->monitor_context,
 308                              timeval_current(),
 309                              ctdb_check_health, ctdb);
 310 }
 311
 312
 313 /*
 314   see if the event scripts think we are healthy
 315  */
 316 static void ctdb_check_health(struct event_context *ev, struct timed_event *te,
 317                               struct timeval t, void *private_data)
 318 {
 319         struct ctdb_context *ctdb = talloc_get_type(private_data, struct ctdb_context);
 320         int ret = 0;
 321
 322         if (ctdb->recovery_mode != CTDB_RECOVERY_NORMAL ||
 323             (ctdb->monitor->monitoring_mode == CTDB_MONITORING_DISABLED && ctdb->done_startup)) {
 324                 event_add_timed(ctdb->ev, ctdb->monitor->monitor_context,
 325                                 timeval_current_ofs(ctdb->monitor->next_interval, 0),
 326                                 ctdb_check_health, ctdb);
 327                 return;
 328         }
 329
 330         if (!ctdb->done_startup) {
 331                 ret = ctdb_event_script_callback(ctdb,
 332                                                  ctdb->monitor->monitor_context, ctdb_startup_callback,
 333                                                  ctdb, false,
 334                                                  CTDB_EVENT_STARTUP, "%s", "");
 335         } else {
 336                 int i;
 337                 int skip_monitoring = 0;
 338
 339                 if (ctdb->recovery_mode != CTDB_RECOVERY_NORMAL) {
 340                         skip_monitoring = 1;
 341                         DEBUG(DEBUG_ERR,("Skip monitoring during recovery\n"));
 342                 }
 343                 for (i=1; i<=NUM_DB_PRIORITIES; i++) {
 344                         if (ctdb->freeze_handles[i] != NULL) {
 345                                 DEBUG(DEBUG_ERR,("Skip monitoring since databases are frozen\n"));
 346                                 skip_monitoring = 1;
 347                                 break;
 348                         }
 349                 }
 350                 if (skip_monitoring != 0) {
 351                         event_add_timed(ctdb->ev, ctdb->monitor->monitor_context,
 352                                         timeval_current_ofs(ctdb->monitor->next_interval, 0),
 353                                         ctdb_check_health, ctdb);
 354                         return;
 355                 } else {
 356                         ret = ctdb_event_script_callback(ctdb,
 357                                         ctdb->monitor->monitor_context, ctdb_health_callback,
 358                                         ctdb, false,
 359                                         CTDB_EVENT_MONITOR, "%s", "");
 360                 }
 361         }
 362
 363         if (ret != 0) {
 364                 DEBUG(DEBUG_ERR,("Unable to launch monitor event script\n"));
 365                 ctdb->monitor->next_interval = 5;
 366                 event_add_timed(ctdb->ev, ctdb->monitor->monitor_context,
 367                         timeval_current_ofs(5, 0),
 368                         ctdb_check_health, ctdb);
 369         }
 370 }
 371
 372 /*
 373   (Temporaily) Disabling monitoring will stop the monitor event scripts
 374   from running   but node health checks will still occur
 375 */
 376 void ctdb_disable_monitoring(struct ctdb_context *ctdb)
 377 {
 378         ctdb->monitor->monitoring_mode = CTDB_MONITORING_DISABLED;
 379         DEBUG(DEBUG_INFO,("Monitoring has been disabled\n"));
 380 }
 381
 382 /*
 383    Re-enable running monitor events after they have been disabled
 384  */
 385 void ctdb_enable_monitoring(struct ctdb_context *ctdb)
 386 {
 387         ctdb->monitor->monitoring_mode  = CTDB_MONITORING_ACTIVE;
 388         ctdb->monitor->next_interval = 5;
 389         DEBUG(DEBUG_INFO,("Monitoring has been enabled\n"));
 390 }
 391
 392 /* stop any monitoring
 393    this should only be done when shutting down the daemon
 394 */
 395 void ctdb_stop_monitoring(struct ctdb_context *ctdb)
 396 {
 397         talloc_free(ctdb->monitor->monitor_context);
 398         ctdb->monitor->monitor_context = NULL;
 399
 400         ctdb->monitor->monitoring_mode  = CTDB_MONITORING_DISABLED;
 401         ctdb->monitor->next_interval = 5;
 402         DEBUG(DEBUG_NOTICE,("Monitoring has been stopped\n"));
 403 }
 404
 405 /*
 406   start watching for nodes that might be dead
 407  */
 408 void ctdb_start_monitoring(struct ctdb_context *ctdb)
 409 {
 410         if (ctdb->monitor != NULL) {
 411                 return;
 412         }
 413
 414         ctdb->monitor = talloc(ctdb, struct ctdb_monitor_state);
 415         CTDB_NO_MEMORY_FATAL(ctdb, ctdb->monitor);
 416
 417         ctdb->monitor->next_interval = 5;
 418
 419         ctdb->monitor->monitor_context = talloc_new(ctdb->monitor);
 420         CTDB_NO_MEMORY_FATAL(ctdb, ctdb->monitor->monitor_context);
 421
 422         event_add_timed(ctdb->ev, ctdb->monitor->monitor_context,
 423                              timeval_current_ofs(1, 0),
 424                              ctdb_wait_until_recovered, ctdb);
 425
 426         ctdb->monitor->monitoring_mode  = CTDB_MONITORING_ACTIVE;
 427         DEBUG(DEBUG_NOTICE,("Monitoring has been started\n"));
 428 }
 429
 430
 431 /*
 432   modify flags on a node
 433  */
 434 int32_t ctdb_control_modflags(struct ctdb_context *ctdb, TDB_DATA indata)
 435 {
 436         struct ctdb_node_flag_change *c = (struct ctdb_node_flag_change *)indata.dptr;
 437         struct ctdb_node *node;
 438         uint32_t old_flags;
 439         int i;
 440
 441         if (c->pnn >= ctdb->num_nodes) {
 442                 DEBUG(DEBUG_ERR,(__location__ " Node %d is invalid, num_nodes :%d\n", c->pnn, ctdb->num_nodes));
 443                 return -1;
 444         }
 445
 446         node         = ctdb->nodes[c->pnn];
 447         old_flags    = node->flags;
 448         if (c->pnn != ctdb->pnn) {
 449                 c->old_flags  = node->flags;
 450         }
 451         node->flags   = c->new_flags & ~NODE_FLAGS_DISCONNECTED;
 452         node->flags  |= (c->old_flags & NODE_FLAGS_DISCONNECTED);
 453
 454         /* we dont let other nodes modify our STOPPED status */
 455         if (c->pnn == ctdb->pnn) {
 456                 node->flags &= ~NODE_FLAGS_STOPPED;
 457                 if (old_flags & NODE_FLAGS_STOPPED) {
 458                         node->flags |= NODE_FLAGS_STOPPED;
 459                 }
 460         }
 461
 462         /* we dont let other nodes modify our BANNED status */
 463         if (c->pnn == ctdb->pnn) {
 464                 node->flags &= ~NODE_FLAGS_BANNED;
 465                 if (old_flags & NODE_FLAGS_BANNED) {
 466                         node->flags |= NODE_FLAGS_BANNED;
 467                 }
 468         }
 469
 470         if (node->flags == c->old_flags) {
 471                 DEBUG(DEBUG_INFO, ("Control modflags on node %u - Unchanged - flags 0x%x\n", c->pnn, node->flags));
 472                 return 0;
 473         }
 474
 475         DEBUG(DEBUG_INFO, ("Control modflags on node %u - flags now 0x%x\n", c->pnn, node->flags));
 476
 477         if (node->flags == 0 && !ctdb->done_startup) {
 478                 DEBUG(DEBUG_ERR, (__location__ " Node %u became healthy - force recovery for startup\n",
 479                                   c->pnn));
 480                 ctdb->recovery_mode = CTDB_RECOVERY_ACTIVE;
 481         }
 482
 483         /* tell the recovery daemon something has changed */
 484         ctdb_daemon_send_message(ctdb, ctdb->pnn,
 485                                  CTDB_SRVID_SET_NODE_FLAGS, indata);
 486
 487         /* if we have become banned, we should go into recovery mode */
 488         if ((node->flags & NODE_FLAGS_BANNED) && !(c->old_flags & NODE_FLAGS_BANNED) && (node->pnn == ctdb->pnn)) {
 489                 /* make sure we are frozen */
 490                 DEBUG(DEBUG_NOTICE,("This node has been banned - forcing freeze and recovery\n"));
 491                 /* Reset the generation id to 1 to make us ignore any
 492                    REQ/REPLY CALL/DMASTER someone sends to us.
 493                    We are now banned so we shouldnt service database calls
 494                    anymore.
 495                 */
 496                 ctdb->vnn_map->generation = INVALID_GENERATION;
 497
 498                 for (i=1; i<=NUM_DB_PRIORITIES; i++) {
 499                         if (ctdb_start_freeze(ctdb, i) != 0) {
 500                                 DEBUG(DEBUG_ERR,(__location__ " Failed to freeze db priority %u\n", i));
 501                         }
 502                 }
 503                 ctdb_release_all_ips(ctdb);
 504                 ctdb->recovery_mode = CTDB_RECOVERY_ACTIVE;
 505         }
 506
 507         return 0;
 508 }
 509
 510 /*
 511   return the monitoring mode
 512  */
 513 int32_t ctdb_monitoring_mode(struct ctdb_context *ctdb)
 514 {
 515         if (ctdb->monitor == NULL) {
 516                 return CTDB_MONITORING_DISABLED;
 517         }
 518         return ctdb->monitor->monitoring_mode;
 519 }
 520