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
 228         if (ctdb->vnn_map->generation == INVALID_GENERATION) {
 229                 ctdb->db_persistent_startup_generation = INVALID_GENERATION;
 230
 231                 DEBUG(DEBUG_NOTICE,(__location__ " generation is INVALID. Wait one more second\n"));
 232                 event_add_timed(ctdb->ev, ctdb->monitor->monitor_context,
 233                                      timeval_current_ofs(1, 0),
 234                                      ctdb_wait_until_recovered, ctdb);
 235                 return;
 236         }
 237
 238         if (ctdb->recovery_mode != CTDB_RECOVERY_NORMAL) {
 239                 ctdb->db_persistent_startup_generation = INVALID_GENERATION;
 240
 241                 DEBUG(DEBUG_NOTICE,(__location__ " in recovery. Wait one more second\n"));
 242                 event_add_timed(ctdb->ev, ctdb->monitor->monitor_context,
 243                                      timeval_current_ofs(1, 0),
 244                                      ctdb_wait_until_recovered, ctdb);
 245                 return;
 246         }
 247
 248
 249         if (!fast_start && timeval_elapsed(&ctdb->last_recovery_finished) < (ctdb->tunable.rerecovery_timeout + 3)) {
 250                 ctdb->db_persistent_startup_generation = INVALID_GENERATION;
 251
 252                 DEBUG(DEBUG_NOTICE,(__location__ " wait for pending recoveries to end. Wait one more second.\n"));
 253
 254                 event_add_timed(ctdb->ev, ctdb->monitor->monitor_context,
 255                                      timeval_current_ofs(1, 0),
 256                                      ctdb_wait_until_recovered, ctdb);
 257                 return;
 258         }
 259
 260         if (ctdb->vnn_map->generation == ctdb->db_persistent_startup_generation) {
 261                 DEBUG(DEBUG_INFO,(__location__ " skip ctdb_recheck_persistent_health() "
 262                                   "until the next recovery\n"));
 263                 event_add_timed(ctdb->ev, ctdb->monitor->monitor_context,
 264                                      timeval_current_ofs(1, 0),
 265                                      ctdb_wait_until_recovered, ctdb);
 266                 return;
 267         }
 268
 269         ctdb->db_persistent_startup_generation = ctdb->vnn_map->generation;
 270         ret = ctdb_recheck_persistent_health(ctdb);
 271         if (ret != 0) {
 272                 ctdb->db_persistent_check_errors++;
 273                 if (ctdb->db_persistent_check_errors < ctdb->max_persistent_check_errors) {
 274                         DEBUG(ctdb->db_persistent_check_errors==1?DEBUG_ERR:DEBUG_WARNING,
 275                               (__location__ "ctdb_recheck_persistent_health() "
 276                               "failed (%llu of %llu times) - retry later\n",
 277                               (unsigned long long)ctdb->db_persistent_check_errors,
 278                               (unsigned long long)ctdb->max_persistent_check_errors));
 279                         event_add_timed(ctdb->ev,
 280                                         ctdb->monitor->monitor_context,
 281                                         timeval_current_ofs(1, 0),
 282                                         ctdb_wait_until_recovered, ctdb);
 283                         return;
 284                 }
 285                 DEBUG(DEBUG_ALERT,(__location__
 286                                   "ctdb_recheck_persistent_health() failed (%llu times) - prepare shutdown\n",
 287                                   (unsigned long long)ctdb->db_persistent_check_errors));
 288                 ctdb_stop_recoverd(ctdb);
 289                 ctdb_stop_keepalive(ctdb);
 290                 ctdb_stop_monitoring(ctdb);
 291                 ctdb_release_all_ips(ctdb);
 292                 if (ctdb->methods != NULL) {
 293                         ctdb->methods->shutdown(ctdb);
 294                 }
 295                 ctdb_event_script(ctdb, CTDB_EVENT_SHUTDOWN);
 296                 DEBUG(DEBUG_ALERT,("ctdb_recheck_persistent_health() failed - Stopping CTDB daemon\n"));
 297                 exit(11);
 298         }
 299         ctdb->db_persistent_check_errors = 0;
 300         DEBUG(DEBUG_NOTICE,(__location__
 301                            "ctdb_start_monitoring: ctdb_recheck_persistent_health() OK\n"));
 302
 303         DEBUG(DEBUG_NOTICE,(__location__ " Recoveries finished. Running the \"startup\" event.\n"));
 304         event_add_timed(ctdb->ev, ctdb->monitor->monitor_context,
 305                              timeval_current(),
 306                              ctdb_check_health, ctdb);
 307 }
 308
 309
 310 /*
 311   see if the event scripts think we are healthy
 312  */
 313 static void ctdb_check_health(struct event_context *ev, struct timed_event *te,
 314                               struct timeval t, void *private_data)
 315 {
 316         struct ctdb_context *ctdb = talloc_get_type(private_data, struct ctdb_context);
 317         int ret = 0;
 318
 319         if (ctdb->recovery_mode != CTDB_RECOVERY_NORMAL ||
 320             (ctdb->monitor->monitoring_mode == CTDB_MONITORING_DISABLED && ctdb->done_startup)) {
 321                 event_add_timed(ctdb->ev, ctdb->monitor->monitor_context,
 322                                 timeval_current_ofs(ctdb->monitor->next_interval, 0),
 323                                 ctdb_check_health, ctdb);
 324                 return;
 325         }
 326
 327         if (!ctdb->done_startup) {
 328                 ret = ctdb_event_script_callback(ctdb,
 329                                                  ctdb->monitor->monitor_context, ctdb_startup_callback,
 330                                                  ctdb, false,
 331                                                  CTDB_EVENT_STARTUP, "%s", "");
 332         } else {
 333                 int i;
 334                 int skip_monitoring = 0;
 335
 336                 if (ctdb->recovery_mode != CTDB_RECOVERY_NORMAL) {
 337                         skip_monitoring = 1;
 338                         DEBUG(DEBUG_ERR,("Skip monitoring during recovery\n"));
 339                 }
 340                 for (i=1; i<=NUM_DB_PRIORITIES; i++) {
 341                         if (ctdb->freeze_handles[i] != NULL) {
 342                                 DEBUG(DEBUG_ERR,("Skip monitoring since databases are frozen\n"));
 343                                 skip_monitoring = 1;
 344                                 break;
 345                         }
 346                 }
 347                 if (skip_monitoring != 0) {
 348                         event_add_timed(ctdb->ev, ctdb->monitor->monitor_context,
 349                                         timeval_current_ofs(ctdb->monitor->next_interval, 0),
 350                                         ctdb_check_health, ctdb);
 351                         return;
 352                 } else {
 353                         ret = ctdb_event_script_callback(ctdb,
 354                                         ctdb->monitor->monitor_context, ctdb_health_callback,
 355                                         ctdb, false,
 356                                         CTDB_EVENT_MONITOR, "%s", "");
 357                 }
 358         }
 359
 360         if (ret != 0) {
 361                 DEBUG(DEBUG_ERR,("Unable to launch monitor event script\n"));
 362                 ctdb->monitor->next_interval = 5;
 363                 event_add_timed(ctdb->ev, ctdb->monitor->monitor_context,
 364                         timeval_current_ofs(5, 0),
 365                         ctdb_check_health, ctdb);
 366         }
 367 }
 368
 369 /*
 370   (Temporaily) Disabling monitoring will stop the monitor event scripts
 371   from running   but node health checks will still occur
 372 */
 373 void ctdb_disable_monitoring(struct ctdb_context *ctdb)
 374 {
 375         ctdb->monitor->monitoring_mode = CTDB_MONITORING_DISABLED;
 376         DEBUG(DEBUG_INFO,("Monitoring has been disabled\n"));
 377 }
 378
 379 /*
 380    Re-enable running monitor events after they have been disabled
 381  */
 382 void ctdb_enable_monitoring(struct ctdb_context *ctdb)
 383 {
 384         ctdb->monitor->monitoring_mode  = CTDB_MONITORING_ACTIVE;
 385         ctdb->monitor->next_interval = 5;
 386         DEBUG(DEBUG_INFO,("Monitoring has been enabled\n"));
 387 }
 388
 389 /* stop any monitoring
 390    this should only be done when shutting down the daemon
 391 */
 392 void ctdb_stop_monitoring(struct ctdb_context *ctdb)
 393 {
 394         talloc_free(ctdb->monitor->monitor_context);
 395         ctdb->monitor->monitor_context = NULL;
 396
 397         ctdb->monitor->monitoring_mode  = CTDB_MONITORING_DISABLED;
 398         ctdb->monitor->next_interval = 5;
 399         DEBUG(DEBUG_NOTICE,("Monitoring has been stopped\n"));
 400 }
 401
 402 /*
 403   start watching for nodes that might be dead
 404  */
 405 void ctdb_start_monitoring(struct ctdb_context *ctdb)
 406 {
 407         if (ctdb->monitor != NULL) {
 408                 return;
 409         }
 410
 411         ctdb->monitor = talloc(ctdb, struct ctdb_monitor_state);
 412         CTDB_NO_MEMORY_FATAL(ctdb, ctdb->monitor);
 413
 414         ctdb->monitor->next_interval = 5;
 415
 416         ctdb->monitor->monitor_context = talloc_new(ctdb->monitor);
 417         CTDB_NO_MEMORY_FATAL(ctdb, ctdb->monitor->monitor_context);
 418
 419         event_add_timed(ctdb->ev, ctdb->monitor->monitor_context,
 420                              timeval_current_ofs(1, 0),
 421                              ctdb_wait_until_recovered, ctdb);
 422
 423         ctdb->monitor->monitoring_mode  = CTDB_MONITORING_ACTIVE;
 424         DEBUG(DEBUG_NOTICE,("Monitoring has been started\n"));
 425 }
 426
 427
 428 /*
 429   modify flags on a node
 430  */
 431 int32_t ctdb_control_modflags(struct ctdb_context *ctdb, TDB_DATA indata)
 432 {
 433         struct ctdb_node_flag_change *c = (struct ctdb_node_flag_change *)indata.dptr;
 434         struct ctdb_node *node;
 435         uint32_t old_flags;
 436         int i;
 437
 438         if (c->pnn >= ctdb->num_nodes) {
 439                 DEBUG(DEBUG_ERR,(__location__ " Node %d is invalid, num_nodes :%d\n", c->pnn, ctdb->num_nodes));
 440                 return -1;
 441         }
 442
 443         node         = ctdb->nodes[c->pnn];
 444         old_flags    = node->flags;
 445         if (c->pnn != ctdb->pnn) {
 446                 c->old_flags  = node->flags;
 447         }
 448         node->flags   = c->new_flags & ~NODE_FLAGS_DISCONNECTED;
 449         node->flags  |= (c->old_flags & NODE_FLAGS_DISCONNECTED);
 450
 451         /* we dont let other nodes modify our STOPPED status */
 452         if (c->pnn == ctdb->pnn) {
 453                 node->flags &= ~NODE_FLAGS_STOPPED;
 454                 if (old_flags & NODE_FLAGS_STOPPED) {
 455                         node->flags |= NODE_FLAGS_STOPPED;
 456                 }
 457         }
 458
 459         /* we dont let other nodes modify our BANNED status */
 460         if (c->pnn == ctdb->pnn) {
 461                 node->flags &= ~NODE_FLAGS_BANNED;
 462                 if (old_flags & NODE_FLAGS_BANNED) {
 463                         node->flags |= NODE_FLAGS_BANNED;
 464                 }
 465         }
 466
 467         if (node->flags == c->old_flags) {
 468                 DEBUG(DEBUG_INFO, ("Control modflags on node %u - Unchanged - flags 0x%x\n", c->pnn, node->flags));
 469                 return 0;
 470         }
 471
 472         DEBUG(DEBUG_INFO, ("Control modflags on node %u - flags now 0x%x\n", c->pnn, node->flags));
 473
 474         if (node->flags == 0 && !ctdb->done_startup) {
 475                 DEBUG(DEBUG_ERR, (__location__ " Node %u became healthy - force recovery for startup\n",
 476                                   c->pnn));
 477                 ctdb->recovery_mode = CTDB_RECOVERY_ACTIVE;
 478         }
 479
 480         /* tell the recovery daemon something has changed */
 481         ctdb_daemon_send_message(ctdb, ctdb->pnn,
 482                                  CTDB_SRVID_SET_NODE_FLAGS, indata);
 483
 484         /* if we have become banned, we should go into recovery mode */
 485         if ((node->flags & NODE_FLAGS_BANNED) && !(c->old_flags & NODE_FLAGS_BANNED) && (node->pnn == ctdb->pnn)) {
 486                 /* make sure we are frozen */
 487                 DEBUG(DEBUG_NOTICE,("This node has been banned - forcing freeze and recovery\n"));
 488                 /* Reset the generation id to 1 to make us ignore any
 489                    REQ/REPLY CALL/DMASTER someone sends to us.
 490                    We are now banned so we shouldnt service database calls
 491                    anymore.
 492                 */
 493                 ctdb->vnn_map->generation = INVALID_GENERATION;
 494
 495                 for (i=1; i<=NUM_DB_PRIORITIES; i++) {
 496                         if (ctdb_start_freeze(ctdb, i) != 0) {
 497                                 DEBUG(DEBUG_ERR,(__location__ " Failed to freeze db priority %u\n", i));
 498                         }
 499                 }
 500                 ctdb_release_all_ips(ctdb);
 501                 ctdb->recovery_mode = CTDB_RECOVERY_ACTIVE;
 502         }
 503
 504         return 0;
 505 }
 506
 507 /*
 508   return the monitoring mode
 509  */
 510 int32_t ctdb_monitoring_mode(struct ctdb_context *ctdb)
 511 {
 512         if (ctdb->monitor == NULL) {
 513                 return CTDB_MONITORING_DISABLED;
 514         }
 515         return ctdb->monitor->monitoring_mode;
 516 }
 517