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