4 Copyright (C) Ronnie Sahlberg 2007
6 This program is free software; you can redistribute it and/or modify
7 it under the terms of the GNU General Public License as published by
8 the Free Software Foundation; either version 3 of the License, or
9 (at your option) any later version.
11 This program is distributed in the hope that it will be useful,
12 but WITHOUT ANY WARRANTY; without even the implied warranty of
13 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 GNU General Public License for more details.
16 You should have received a copy of the GNU General Public License
17 along with this program; if not, see <http://www.gnu.org/licenses/>.
21 #include "lib/events/events.h"
22 #include "system/filesys.h"
23 #include "system/time.h"
24 #include "system/network.h"
25 #include "system/wait.h"
28 #include "../include/ctdb.h"
29 #include "../include/ctdb_private.h"
31 #include "dlinklist.h"
34 /* list of "ctdb ipreallocate" processes to call back when we have
35 finished the takeover run.
37 struct ip_reallocate_list {
38 struct ip_reallocate_list *next;
39 struct rd_memdump_reply *rd;
42 struct ctdb_banning_state {
44 struct timeval last_reported_time;
48 private state of recovery daemon
50 struct ctdb_recoverd {
51 struct ctdb_context *ctdb;
54 uint32_t num_connected;
55 uint32_t last_culprit_node;
56 struct ctdb_node_map *nodemap;
57 struct timeval priority_time;
58 bool need_takeover_run;
61 struct timed_event *send_election_te;
62 struct timed_event *election_timeout;
63 struct vacuum_info *vacuum_info;
64 TALLOC_CTX *ip_reallocate_ctx;
65 struct ip_reallocate_list *reallocate_callers;
66 TALLOC_CTX *ip_check_disable_ctx;
67 struct ctdb_control_get_ifaces *ifaces;
70 #define CONTROL_TIMEOUT() timeval_current_ofs(ctdb->tunable.recover_timeout, 0)
71 #define MONITOR_TIMEOUT() timeval_current_ofs(ctdb->tunable.recover_interval, 0)
75 ban a node for a period of time
77 static void ctdb_ban_node(struct ctdb_recoverd *rec, uint32_t pnn, uint32_t ban_time)
80 struct ctdb_context *ctdb = rec->ctdb;
81 struct ctdb_ban_time bantime;
83 DEBUG(DEBUG_NOTICE,("Banning node %u for %u seconds\n", pnn, ban_time));
85 if (!ctdb_validate_pnn(ctdb, pnn)) {
86 DEBUG(DEBUG_ERR,("Bad pnn %u in ctdb_ban_node\n", pnn));
91 bantime.time = ban_time;
93 ret = ctdb_ctrl_set_ban(ctdb, CONTROL_TIMEOUT(), pnn, &bantime);
95 DEBUG(DEBUG_ERR,(__location__ " Failed to ban node %d\n", pnn));
101 enum monitor_result { MONITOR_OK, MONITOR_RECOVERY_NEEDED, MONITOR_ELECTION_NEEDED, MONITOR_FAILED};
105 run the "recovered" eventscript on all nodes
107 static int run_recovered_eventscript(struct ctdb_context *ctdb, struct ctdb_node_map *nodemap, const char *caller)
112 tmp_ctx = talloc_new(ctdb);
113 CTDB_NO_MEMORY(ctdb, tmp_ctx);
115 nodes = list_of_active_nodes(ctdb, nodemap, tmp_ctx, true);
116 if (ctdb_client_async_control(ctdb, CTDB_CONTROL_END_RECOVERY,
118 CONTROL_TIMEOUT(), false, tdb_null,
121 DEBUG(DEBUG_ERR, (__location__ " Unable to run the 'recovered' event when called from %s\n", caller));
123 talloc_free(tmp_ctx);
127 talloc_free(tmp_ctx);
132 remember the trouble maker
134 static void ctdb_set_culprit_count(struct ctdb_recoverd *rec, uint32_t culprit, uint32_t count)
136 struct ctdb_context *ctdb = talloc_get_type(rec->ctdb, struct ctdb_context);
137 struct ctdb_banning_state *ban_state;
139 if (culprit > ctdb->num_nodes) {
140 DEBUG(DEBUG_ERR,("Trying to set culprit %d but num_nodes is %d\n", culprit, ctdb->num_nodes));
144 if (ctdb->nodes[culprit]->ban_state == NULL) {
145 ctdb->nodes[culprit]->ban_state = talloc_zero(ctdb->nodes[culprit], struct ctdb_banning_state);
146 CTDB_NO_MEMORY_VOID(ctdb, ctdb->nodes[culprit]->ban_state);
150 ban_state = ctdb->nodes[culprit]->ban_state;
151 if (timeval_elapsed(&ban_state->last_reported_time) > ctdb->tunable.recovery_grace_period) {
152 /* this was the first time in a long while this node
153 misbehaved so we will forgive any old transgressions.
155 ban_state->count = 0;
158 ban_state->count += count;
159 ban_state->last_reported_time = timeval_current();
160 rec->last_culprit_node = culprit;
164 remember the trouble maker
166 static void ctdb_set_culprit(struct ctdb_recoverd *rec, uint32_t culprit)
168 ctdb_set_culprit_count(rec, culprit, 1);
172 /* this callback is called for every node that failed to execute the
175 static void startrecovery_fail_callback(struct ctdb_context *ctdb, uint32_t node_pnn, int32_t res, TDB_DATA outdata, void *callback_data)
177 struct ctdb_recoverd *rec = talloc_get_type(callback_data, struct ctdb_recoverd);
179 DEBUG(DEBUG_ERR, (__location__ " Node %u failed the startrecovery event. Setting it as recovery fail culprit\n", node_pnn));
181 ctdb_set_culprit(rec, node_pnn);
185 run the "startrecovery" eventscript on all nodes
187 static int run_startrecovery_eventscript(struct ctdb_recoverd *rec, struct ctdb_node_map *nodemap)
191 struct ctdb_context *ctdb = rec->ctdb;
193 tmp_ctx = talloc_new(ctdb);
194 CTDB_NO_MEMORY(ctdb, tmp_ctx);
196 nodes = list_of_active_nodes(ctdb, nodemap, tmp_ctx, true);
197 if (ctdb_client_async_control(ctdb, CTDB_CONTROL_START_RECOVERY,
199 CONTROL_TIMEOUT(), false, tdb_null,
201 startrecovery_fail_callback,
203 DEBUG(DEBUG_ERR, (__location__ " Unable to run the 'startrecovery' event. Recovery failed.\n"));
204 talloc_free(tmp_ctx);
208 talloc_free(tmp_ctx);
212 static void async_getcap_callback(struct ctdb_context *ctdb, uint32_t node_pnn, int32_t res, TDB_DATA outdata, void *callback_data)
214 if ( (outdata.dsize != sizeof(uint32_t)) || (outdata.dptr == NULL) ) {
215 DEBUG(DEBUG_ERR, (__location__ " Invalid lenght/pointer for getcap callback : %u %p\n", (unsigned)outdata.dsize, outdata.dptr));
218 if (node_pnn < ctdb->num_nodes) {
219 ctdb->nodes[node_pnn]->capabilities = *((uint32_t *)outdata.dptr);
224 update the node capabilities for all connected nodes
226 static int update_capabilities(struct ctdb_context *ctdb, struct ctdb_node_map *nodemap)
231 tmp_ctx = talloc_new(ctdb);
232 CTDB_NO_MEMORY(ctdb, tmp_ctx);
234 nodes = list_of_active_nodes(ctdb, nodemap, tmp_ctx, true);
235 if (ctdb_client_async_control(ctdb, CTDB_CONTROL_GET_CAPABILITIES,
239 async_getcap_callback, NULL,
241 DEBUG(DEBUG_ERR, (__location__ " Failed to read node capabilities.\n"));
242 talloc_free(tmp_ctx);
246 talloc_free(tmp_ctx);
250 static void set_recmode_fail_callback(struct ctdb_context *ctdb, uint32_t node_pnn, int32_t res, TDB_DATA outdata, void *callback_data)
252 struct ctdb_recoverd *rec = talloc_get_type(callback_data, struct ctdb_recoverd);
254 DEBUG(DEBUG_ERR,("Failed to freeze node %u during recovery. Set it as ban culprit for %d credits\n", node_pnn, rec->nodemap->num));
255 ctdb_set_culprit_count(rec, node_pnn, rec->nodemap->num);
258 static void transaction_start_fail_callback(struct ctdb_context *ctdb, uint32_t node_pnn, int32_t res, TDB_DATA outdata, void *callback_data)
260 struct ctdb_recoverd *rec = talloc_get_type(callback_data, struct ctdb_recoverd);
262 DEBUG(DEBUG_ERR,("Failed to start recovery transaction on node %u. Set it as ban culprit for %d credits\n", node_pnn, rec->nodemap->num));
263 ctdb_set_culprit_count(rec, node_pnn, rec->nodemap->num);
267 change recovery mode on all nodes
269 static int set_recovery_mode(struct ctdb_context *ctdb, struct ctdb_recoverd *rec, struct ctdb_node_map *nodemap, uint32_t rec_mode)
275 tmp_ctx = talloc_new(ctdb);
276 CTDB_NO_MEMORY(ctdb, tmp_ctx);
278 /* freeze all nodes */
279 nodes = list_of_active_nodes(ctdb, nodemap, tmp_ctx, true);
280 if (rec_mode == CTDB_RECOVERY_ACTIVE) {
283 for (i=1; i<=NUM_DB_PRIORITIES; i++) {
284 if (ctdb_client_async_control(ctdb, CTDB_CONTROL_FREEZE,
289 set_recmode_fail_callback,
291 DEBUG(DEBUG_ERR, (__location__ " Unable to freeze nodes. Recovery failed.\n"));
292 talloc_free(tmp_ctx);
299 data.dsize = sizeof(uint32_t);
300 data.dptr = (unsigned char *)&rec_mode;
302 if (ctdb_client_async_control(ctdb, CTDB_CONTROL_SET_RECMODE,
308 DEBUG(DEBUG_ERR, (__location__ " Unable to set recovery mode. Recovery failed.\n"));
309 talloc_free(tmp_ctx);
313 talloc_free(tmp_ctx);
318 change recovery master on all node
320 static int set_recovery_master(struct ctdb_context *ctdb, struct ctdb_node_map *nodemap, uint32_t pnn)
326 tmp_ctx = talloc_new(ctdb);
327 CTDB_NO_MEMORY(ctdb, tmp_ctx);
329 data.dsize = sizeof(uint32_t);
330 data.dptr = (unsigned char *)&pnn;
332 nodes = list_of_active_nodes(ctdb, nodemap, tmp_ctx, true);
333 if (ctdb_client_async_control(ctdb, CTDB_CONTROL_SET_RECMASTER,
335 CONTROL_TIMEOUT(), false, data,
338 DEBUG(DEBUG_ERR, (__location__ " Unable to set recmaster. Recovery failed.\n"));
339 talloc_free(tmp_ctx);
343 talloc_free(tmp_ctx);
347 /* update all remote nodes to use the same db priority that we have
348 this can fail if the remove node has not yet been upgraded to
349 support this function, so we always return success and never fail
350 a recovery if this call fails.
352 static int update_db_priority_on_remote_nodes(struct ctdb_context *ctdb,
353 struct ctdb_node_map *nodemap,
354 uint32_t pnn, struct ctdb_dbid_map *dbmap, TALLOC_CTX *mem_ctx)
359 nodes = list_of_active_nodes(ctdb, nodemap, mem_ctx, true);
361 /* step through all local databases */
362 for (db=0; db<dbmap->num;db++) {
364 struct ctdb_db_priority db_prio;
367 db_prio.db_id = dbmap->dbs[db].dbid;
368 ret = ctdb_ctrl_get_db_priority(ctdb, CONTROL_TIMEOUT(), CTDB_CURRENT_NODE, dbmap->dbs[db].dbid, &db_prio.priority);
370 DEBUG(DEBUG_ERR,(__location__ " Failed to read database priority from local node for db 0x%08x\n", dbmap->dbs[db].dbid));
374 DEBUG(DEBUG_INFO,("Update DB priority for db 0x%08x to %u\n", dbmap->dbs[db].dbid, db_prio.priority));
376 data.dptr = (uint8_t *)&db_prio;
377 data.dsize = sizeof(db_prio);
379 if (ctdb_client_async_control(ctdb,
380 CTDB_CONTROL_SET_DB_PRIORITY,
382 CONTROL_TIMEOUT(), false, data,
385 DEBUG(DEBUG_ERR,(__location__ " Failed to set DB priority for 0x%08x\n", db_prio.db_id));
393 ensure all other nodes have attached to any databases that we have
395 static int create_missing_remote_databases(struct ctdb_context *ctdb, struct ctdb_node_map *nodemap,
396 uint32_t pnn, struct ctdb_dbid_map *dbmap, TALLOC_CTX *mem_ctx)
399 struct ctdb_dbid_map *remote_dbmap;
401 /* verify that all other nodes have all our databases */
402 for (j=0; j<nodemap->num; j++) {
403 /* we dont need to ourself ourselves */
404 if (nodemap->nodes[j].pnn == pnn) {
407 /* dont check nodes that are unavailable */
408 if (nodemap->nodes[j].flags & NODE_FLAGS_INACTIVE) {
412 ret = ctdb_ctrl_getdbmap(ctdb, CONTROL_TIMEOUT(), nodemap->nodes[j].pnn,
413 mem_ctx, &remote_dbmap);
415 DEBUG(DEBUG_ERR, (__location__ " Unable to get dbids from node %u\n", pnn));
419 /* step through all local databases */
420 for (db=0; db<dbmap->num;db++) {
424 for (i=0;i<remote_dbmap->num;i++) {
425 if (dbmap->dbs[db].dbid == remote_dbmap->dbs[i].dbid) {
429 /* the remote node already have this database */
430 if (i!=remote_dbmap->num) {
433 /* ok so we need to create this database */
434 ctdb_ctrl_getdbname(ctdb, CONTROL_TIMEOUT(), pnn, dbmap->dbs[db].dbid,
437 DEBUG(DEBUG_ERR, (__location__ " Unable to get dbname from node %u\n", pnn));
440 ctdb_ctrl_createdb(ctdb, CONTROL_TIMEOUT(), nodemap->nodes[j].pnn,
441 mem_ctx, name, dbmap->dbs[db].persistent);
443 DEBUG(DEBUG_ERR, (__location__ " Unable to create remote db:%s\n", name));
454 ensure we are attached to any databases that anyone else is attached to
456 static int create_missing_local_databases(struct ctdb_context *ctdb, struct ctdb_node_map *nodemap,
457 uint32_t pnn, struct ctdb_dbid_map **dbmap, TALLOC_CTX *mem_ctx)
460 struct ctdb_dbid_map *remote_dbmap;
462 /* verify that we have all database any other node has */
463 for (j=0; j<nodemap->num; j++) {
464 /* we dont need to ourself ourselves */
465 if (nodemap->nodes[j].pnn == pnn) {
468 /* dont check nodes that are unavailable */
469 if (nodemap->nodes[j].flags & NODE_FLAGS_INACTIVE) {
473 ret = ctdb_ctrl_getdbmap(ctdb, CONTROL_TIMEOUT(), nodemap->nodes[j].pnn,
474 mem_ctx, &remote_dbmap);
476 DEBUG(DEBUG_ERR, (__location__ " Unable to get dbids from node %u\n", pnn));
480 /* step through all databases on the remote node */
481 for (db=0; db<remote_dbmap->num;db++) {
484 for (i=0;i<(*dbmap)->num;i++) {
485 if (remote_dbmap->dbs[db].dbid == (*dbmap)->dbs[i].dbid) {
489 /* we already have this db locally */
490 if (i!=(*dbmap)->num) {
493 /* ok so we need to create this database and
496 ctdb_ctrl_getdbname(ctdb, CONTROL_TIMEOUT(), nodemap->nodes[j].pnn,
497 remote_dbmap->dbs[db].dbid, mem_ctx, &name);
499 DEBUG(DEBUG_ERR, (__location__ " Unable to get dbname from node %u\n",
500 nodemap->nodes[j].pnn));
503 ctdb_ctrl_createdb(ctdb, CONTROL_TIMEOUT(), pnn, mem_ctx, name,
504 remote_dbmap->dbs[db].persistent);
506 DEBUG(DEBUG_ERR, (__location__ " Unable to create local db:%s\n", name));
509 ret = ctdb_ctrl_getdbmap(ctdb, CONTROL_TIMEOUT(), pnn, mem_ctx, dbmap);
511 DEBUG(DEBUG_ERR, (__location__ " Unable to reread dbmap on node %u\n", pnn));
522 pull the remote database contents from one node into the recdb
524 static int pull_one_remote_database(struct ctdb_context *ctdb, uint32_t srcnode,
525 struct tdb_wrap *recdb, uint32_t dbid,
530 struct ctdb_marshall_buffer *reply;
531 struct ctdb_rec_data *rec;
533 TALLOC_CTX *tmp_ctx = talloc_new(recdb);
535 ret = ctdb_ctrl_pulldb(ctdb, srcnode, dbid, CTDB_LMASTER_ANY, tmp_ctx,
536 CONTROL_TIMEOUT(), &outdata);
538 DEBUG(DEBUG_ERR,(__location__ " Unable to copy db from node %u\n", srcnode));
539 talloc_free(tmp_ctx);
543 reply = (struct ctdb_marshall_buffer *)outdata.dptr;
545 if (outdata.dsize < offsetof(struct ctdb_marshall_buffer, data)) {
546 DEBUG(DEBUG_ERR,(__location__ " invalid data in pulldb reply\n"));
547 talloc_free(tmp_ctx);
551 rec = (struct ctdb_rec_data *)&reply->data[0];
555 rec = (struct ctdb_rec_data *)(rec->length + (uint8_t *)rec), i++) {
557 struct ctdb_ltdb_header *hdr;
560 key.dptr = &rec->data[0];
561 key.dsize = rec->keylen;
562 data.dptr = &rec->data[key.dsize];
563 data.dsize = rec->datalen;
565 hdr = (struct ctdb_ltdb_header *)data.dptr;
567 if (data.dsize < sizeof(struct ctdb_ltdb_header)) {
568 DEBUG(DEBUG_CRIT,(__location__ " bad ltdb record\n"));
569 talloc_free(tmp_ctx);
573 /* fetch the existing record, if any */
574 existing = tdb_fetch(recdb->tdb, key);
576 if (existing.dptr != NULL) {
577 struct ctdb_ltdb_header header;
578 if (existing.dsize < sizeof(struct ctdb_ltdb_header)) {
579 DEBUG(DEBUG_CRIT,(__location__ " Bad record size %u from node %u\n",
580 (unsigned)existing.dsize, srcnode));
582 talloc_free(tmp_ctx);
585 header = *(struct ctdb_ltdb_header *)existing.dptr;
587 if (!(header.rsn < hdr->rsn ||
588 (header.dmaster != ctdb->recovery_master && header.rsn == hdr->rsn))) {
593 if (tdb_store(recdb->tdb, key, data, TDB_REPLACE) != 0) {
594 DEBUG(DEBUG_CRIT,(__location__ " Failed to store record\n"));
595 talloc_free(tmp_ctx);
600 talloc_free(tmp_ctx);
606 pull all the remote database contents into the recdb
608 static int pull_remote_database(struct ctdb_context *ctdb,
609 struct ctdb_recoverd *rec,
610 struct ctdb_node_map *nodemap,
611 struct tdb_wrap *recdb, uint32_t dbid,
616 /* pull all records from all other nodes across onto this node
617 (this merges based on rsn)
619 for (j=0; j<nodemap->num; j++) {
620 /* dont merge from nodes that are unavailable */
621 if (nodemap->nodes[j].flags & NODE_FLAGS_INACTIVE) {
624 if (pull_one_remote_database(ctdb, nodemap->nodes[j].pnn, recdb, dbid, persistent) != 0) {
625 DEBUG(DEBUG_ERR,(__location__ " Failed to pull remote database from node %u\n",
626 nodemap->nodes[j].pnn));
627 ctdb_set_culprit_count(rec, nodemap->nodes[j].pnn, nodemap->num);
637 update flags on all active nodes
639 static int update_flags_on_all_nodes(struct ctdb_context *ctdb, struct ctdb_node_map *nodemap, uint32_t pnn, uint32_t flags)
643 ret = ctdb_ctrl_modflags(ctdb, CONTROL_TIMEOUT(), pnn, flags, ~flags);
645 DEBUG(DEBUG_ERR, (__location__ " Unable to update nodeflags on remote nodes\n"));
653 ensure all nodes have the same vnnmap we do
655 static int update_vnnmap_on_all_nodes(struct ctdb_context *ctdb, struct ctdb_node_map *nodemap,
656 uint32_t pnn, struct ctdb_vnn_map *vnnmap, TALLOC_CTX *mem_ctx)
660 /* push the new vnn map out to all the nodes */
661 for (j=0; j<nodemap->num; j++) {
662 /* dont push to nodes that are unavailable */
663 if (nodemap->nodes[j].flags & NODE_FLAGS_INACTIVE) {
667 ret = ctdb_ctrl_setvnnmap(ctdb, CONTROL_TIMEOUT(), nodemap->nodes[j].pnn, mem_ctx, vnnmap);
669 DEBUG(DEBUG_ERR, (__location__ " Unable to set vnnmap for node %u\n", pnn));
679 struct vacuum_info *next, *prev;
680 struct ctdb_recoverd *rec;
682 struct ctdb_db_context *ctdb_db;
683 struct ctdb_marshall_buffer *recs;
684 struct ctdb_rec_data *r;
687 static void vacuum_fetch_next(struct vacuum_info *v);
690 called when a vacuum fetch has completed - just free it and do the next one
692 static void vacuum_fetch_callback(struct ctdb_client_call_state *state)
694 struct vacuum_info *v = talloc_get_type(state->async.private_data, struct vacuum_info);
696 vacuum_fetch_next(v);
701 process the next element from the vacuum list
703 static void vacuum_fetch_next(struct vacuum_info *v)
705 struct ctdb_call call;
706 struct ctdb_rec_data *r;
708 while (v->recs->count) {
709 struct ctdb_client_call_state *state;
711 struct ctdb_ltdb_header *hdr;
714 call.call_id = CTDB_NULL_FUNC;
715 call.flags = CTDB_IMMEDIATE_MIGRATION;
718 v->r = (struct ctdb_rec_data *)(r->length + (uint8_t *)r);
721 call.key.dptr = &r->data[0];
722 call.key.dsize = r->keylen;
724 /* ensure we don't block this daemon - just skip a record if we can't get
726 if (tdb_chainlock_nonblock(v->ctdb_db->ltdb->tdb, call.key) != 0) {
730 data = tdb_fetch(v->ctdb_db->ltdb->tdb, call.key);
731 if (data.dptr == NULL) {
732 tdb_chainunlock(v->ctdb_db->ltdb->tdb, call.key);
736 if (data.dsize < sizeof(struct ctdb_ltdb_header)) {
738 tdb_chainunlock(v->ctdb_db->ltdb->tdb, call.key);
742 hdr = (struct ctdb_ltdb_header *)data.dptr;
743 if (hdr->dmaster == v->rec->ctdb->pnn) {
744 /* its already local */
746 tdb_chainunlock(v->ctdb_db->ltdb->tdb, call.key);
752 state = ctdb_call_send(v->ctdb_db, &call);
753 tdb_chainunlock(v->ctdb_db->ltdb->tdb, call.key);
755 DEBUG(DEBUG_ERR,(__location__ " Failed to setup vacuum fetch call\n"));
759 state->async.fn = vacuum_fetch_callback;
760 state->async.private_data = v;
769 destroy a vacuum info structure
771 static int vacuum_info_destructor(struct vacuum_info *v)
773 DLIST_REMOVE(v->rec->vacuum_info, v);
779 handler for vacuum fetch
781 static void vacuum_fetch_handler(struct ctdb_context *ctdb, uint64_t srvid,
782 TDB_DATA data, void *private_data)
784 struct ctdb_recoverd *rec = talloc_get_type(private_data, struct ctdb_recoverd);
785 struct ctdb_marshall_buffer *recs;
787 TALLOC_CTX *tmp_ctx = talloc_new(ctdb);
789 struct ctdb_dbid_map *dbmap=NULL;
790 bool persistent = false;
791 struct ctdb_db_context *ctdb_db;
792 struct ctdb_rec_data *r;
794 struct vacuum_info *v;
796 recs = (struct ctdb_marshall_buffer *)data.dptr;
797 r = (struct ctdb_rec_data *)&recs->data[0];
799 if (recs->count == 0) {
800 talloc_free(tmp_ctx);
806 for (v=rec->vacuum_info;v;v=v->next) {
807 if (srcnode == v->srcnode && recs->db_id == v->ctdb_db->db_id) {
808 /* we're already working on records from this node */
809 talloc_free(tmp_ctx);
814 /* work out if the database is persistent */
815 ret = ctdb_ctrl_getdbmap(ctdb, CONTROL_TIMEOUT(), CTDB_CURRENT_NODE, tmp_ctx, &dbmap);
817 DEBUG(DEBUG_ERR, (__location__ " Unable to get dbids from local node\n"));
818 talloc_free(tmp_ctx);
822 for (i=0;i<dbmap->num;i++) {
823 if (dbmap->dbs[i].dbid == recs->db_id) {
824 persistent = dbmap->dbs[i].persistent;
828 if (i == dbmap->num) {
829 DEBUG(DEBUG_ERR, (__location__ " Unable to find db_id 0x%x on local node\n", recs->db_id));
830 talloc_free(tmp_ctx);
834 /* find the name of this database */
835 if (ctdb_ctrl_getdbname(ctdb, CONTROL_TIMEOUT(), CTDB_CURRENT_NODE, recs->db_id, tmp_ctx, &name) != 0) {
836 DEBUG(DEBUG_ERR,(__location__ " Failed to get name of db 0x%x\n", recs->db_id));
837 talloc_free(tmp_ctx);
842 ctdb_db = ctdb_attach(ctdb, name, persistent, 0);
843 if (ctdb_db == NULL) {
844 DEBUG(DEBUG_ERR,(__location__ " Failed to attach to database '%s'\n", name));
845 talloc_free(tmp_ctx);
849 v = talloc_zero(rec, struct vacuum_info);
851 DEBUG(DEBUG_CRIT,(__location__ " Out of memory\n"));
852 talloc_free(tmp_ctx);
857 v->srcnode = srcnode;
858 v->ctdb_db = ctdb_db;
859 v->recs = talloc_memdup(v, recs, data.dsize);
860 if (v->recs == NULL) {
861 DEBUG(DEBUG_CRIT,(__location__ " Out of memory\n"));
863 talloc_free(tmp_ctx);
866 v->r = (struct ctdb_rec_data *)&v->recs->data[0];
868 DLIST_ADD(rec->vacuum_info, v);
870 talloc_set_destructor(v, vacuum_info_destructor);
872 vacuum_fetch_next(v);
873 talloc_free(tmp_ctx);
878 called when ctdb_wait_timeout should finish
880 static void ctdb_wait_handler(struct event_context *ev, struct timed_event *te,
881 struct timeval yt, void *p)
883 uint32_t *timed_out = (uint32_t *)p;
888 wait for a given number of seconds
890 static void ctdb_wait_timeout(struct ctdb_context *ctdb, uint32_t secs)
892 uint32_t timed_out = 0;
893 event_add_timed(ctdb->ev, ctdb, timeval_current_ofs(secs, 0), ctdb_wait_handler, &timed_out);
895 event_loop_once(ctdb->ev);
900 called when an election times out (ends)
902 static void ctdb_election_timeout(struct event_context *ev, struct timed_event *te,
903 struct timeval t, void *p)
905 struct ctdb_recoverd *rec = talloc_get_type(p, struct ctdb_recoverd);
906 rec->election_timeout = NULL;
908 DEBUG(DEBUG_WARNING,(__location__ " Election timed out\n"));
913 wait for an election to finish. It finished election_timeout seconds after
914 the last election packet is received
916 static void ctdb_wait_election(struct ctdb_recoverd *rec)
918 struct ctdb_context *ctdb = rec->ctdb;
919 while (rec->election_timeout) {
920 event_loop_once(ctdb->ev);
925 Update our local flags from all remote connected nodes.
926 This is only run when we are or we belive we are the recovery master
928 static int update_local_flags(struct ctdb_recoverd *rec, struct ctdb_node_map *nodemap)
931 struct ctdb_context *ctdb = rec->ctdb;
932 TALLOC_CTX *mem_ctx = talloc_new(ctdb);
934 /* get the nodemap for all active remote nodes and verify
935 they are the same as for this node
937 for (j=0; j<nodemap->num; j++) {
938 struct ctdb_node_map *remote_nodemap=NULL;
941 if (nodemap->nodes[j].flags & NODE_FLAGS_DISCONNECTED) {
944 if (nodemap->nodes[j].pnn == ctdb->pnn) {
948 ret = ctdb_ctrl_getnodemap(ctdb, CONTROL_TIMEOUT(), nodemap->nodes[j].pnn,
949 mem_ctx, &remote_nodemap);
951 DEBUG(DEBUG_ERR, (__location__ " Unable to get nodemap from remote node %u\n",
952 nodemap->nodes[j].pnn));
953 ctdb_set_culprit(rec, nodemap->nodes[j].pnn);
954 talloc_free(mem_ctx);
955 return MONITOR_FAILED;
957 if (nodemap->nodes[j].flags != remote_nodemap->nodes[j].flags) {
958 /* We should tell our daemon about this so it
959 updates its flags or else we will log the same
960 message again in the next iteration of recovery.
961 Since we are the recovery master we can just as
962 well update the flags on all nodes.
964 ret = ctdb_ctrl_modflags(ctdb, CONTROL_TIMEOUT(), nodemap->nodes[j].pnn, nodemap->nodes[j].flags, ~nodemap->nodes[j].flags);
966 DEBUG(DEBUG_ERR, (__location__ " Unable to update nodeflags on remote nodes\n"));
970 /* Update our local copy of the flags in the recovery
973 DEBUG(DEBUG_NOTICE,("Remote node %u had flags 0x%x, local had 0x%x - updating local\n",
974 nodemap->nodes[j].pnn, remote_nodemap->nodes[j].flags,
975 nodemap->nodes[j].flags));
976 nodemap->nodes[j].flags = remote_nodemap->nodes[j].flags;
978 talloc_free(remote_nodemap);
980 talloc_free(mem_ctx);
985 /* Create a new random generation ip.
986 The generation id can not be the INVALID_GENERATION id
988 static uint32_t new_generation(void)
993 generation = random();
995 if (generation != INVALID_GENERATION) {
1005 create a temporary working database
1007 static struct tdb_wrap *create_recdb(struct ctdb_context *ctdb, TALLOC_CTX *mem_ctx)
1010 struct tdb_wrap *recdb;
1013 /* open up the temporary recovery database */
1014 name = talloc_asprintf(mem_ctx, "%s/recdb.tdb.%u",
1015 ctdb->db_directory_state,
1022 tdb_flags = TDB_NOLOCK;
1023 if (ctdb->valgrinding) {
1024 tdb_flags |= TDB_NOMMAP;
1026 tdb_flags |= TDB_DISALLOW_NESTING;
1028 recdb = tdb_wrap_open(mem_ctx, name, ctdb->tunable.database_hash_size,
1029 tdb_flags, O_RDWR|O_CREAT|O_EXCL, 0600);
1030 if (recdb == NULL) {
1031 DEBUG(DEBUG_CRIT,(__location__ " Failed to create temp recovery database '%s'\n", name));
1041 a traverse function for pulling all relevent records from recdb
1044 struct ctdb_context *ctdb;
1045 struct ctdb_marshall_buffer *recdata;
1051 static int traverse_recdb(struct tdb_context *tdb, TDB_DATA key, TDB_DATA data, void *p)
1053 struct recdb_data *params = (struct recdb_data *)p;
1054 struct ctdb_rec_data *rec;
1055 struct ctdb_ltdb_header *hdr;
1057 /* skip empty records */
1058 if (data.dsize <= sizeof(struct ctdb_ltdb_header)) {
1062 /* update the dmaster field to point to us */
1063 hdr = (struct ctdb_ltdb_header *)data.dptr;
1064 if (!params->persistent) {
1065 hdr->dmaster = params->ctdb->pnn;
1068 /* add the record to the blob ready to send to the nodes */
1069 rec = ctdb_marshall_record(params->recdata, 0, key, NULL, data);
1071 params->failed = true;
1074 params->recdata = talloc_realloc_size(NULL, params->recdata, rec->length + params->len);
1075 if (params->recdata == NULL) {
1076 DEBUG(DEBUG_CRIT,(__location__ " Failed to expand recdata to %u (%u records)\n",
1077 rec->length + params->len, params->recdata->count));
1078 params->failed = true;
1081 params->recdata->count++;
1082 memcpy(params->len+(uint8_t *)params->recdata, rec, rec->length);
1083 params->len += rec->length;
1090 push the recdb database out to all nodes
1092 static int push_recdb_database(struct ctdb_context *ctdb, uint32_t dbid,
1094 struct tdb_wrap *recdb, struct ctdb_node_map *nodemap)
1096 struct recdb_data params;
1097 struct ctdb_marshall_buffer *recdata;
1099 TALLOC_CTX *tmp_ctx;
1102 tmp_ctx = talloc_new(ctdb);
1103 CTDB_NO_MEMORY(ctdb, tmp_ctx);
1105 recdata = talloc_zero(recdb, struct ctdb_marshall_buffer);
1106 CTDB_NO_MEMORY(ctdb, recdata);
1108 recdata->db_id = dbid;
1111 params.recdata = recdata;
1112 params.len = offsetof(struct ctdb_marshall_buffer, data);
1113 params.failed = false;
1114 params.persistent = persistent;
1116 if (tdb_traverse_read(recdb->tdb, traverse_recdb, ¶ms) == -1) {
1117 DEBUG(DEBUG_ERR,(__location__ " Failed to traverse recdb database\n"));
1118 talloc_free(params.recdata);
1119 talloc_free(tmp_ctx);
1123 if (params.failed) {
1124 DEBUG(DEBUG_ERR,(__location__ " Failed to traverse recdb database\n"));
1125 talloc_free(params.recdata);
1126 talloc_free(tmp_ctx);
1130 recdata = params.recdata;
1132 outdata.dptr = (void *)recdata;
1133 outdata.dsize = params.len;
1135 nodes = list_of_active_nodes(ctdb, nodemap, tmp_ctx, true);
1136 if (ctdb_client_async_control(ctdb, CTDB_CONTROL_PUSH_DB,
1138 CONTROL_TIMEOUT(), false, outdata,
1141 DEBUG(DEBUG_ERR,(__location__ " Failed to push recdb records to nodes for db 0x%x\n", dbid));
1142 talloc_free(recdata);
1143 talloc_free(tmp_ctx);
1147 DEBUG(DEBUG_NOTICE, (__location__ " Recovery - pushed remote database 0x%x of size %u\n",
1148 dbid, recdata->count));
1150 talloc_free(recdata);
1151 talloc_free(tmp_ctx);
1158 go through a full recovery on one database
1160 static int recover_database(struct ctdb_recoverd *rec,
1161 TALLOC_CTX *mem_ctx,
1165 struct ctdb_node_map *nodemap,
1166 uint32_t transaction_id)
1168 struct tdb_wrap *recdb;
1170 struct ctdb_context *ctdb = rec->ctdb;
1172 struct ctdb_control_wipe_database w;
1175 recdb = create_recdb(ctdb, mem_ctx);
1176 if (recdb == NULL) {
1180 /* pull all remote databases onto the recdb */
1181 ret = pull_remote_database(ctdb, rec, nodemap, recdb, dbid, persistent);
1183 DEBUG(DEBUG_ERR, (__location__ " Unable to pull remote database 0x%x\n", dbid));
1187 DEBUG(DEBUG_NOTICE, (__location__ " Recovery - pulled remote database 0x%x\n", dbid));
1189 /* wipe all the remote databases. This is safe as we are in a transaction */
1191 w.transaction_id = transaction_id;
1193 data.dptr = (void *)&w;
1194 data.dsize = sizeof(w);
1196 nodes = list_of_active_nodes(ctdb, nodemap, recdb, true);
1197 if (ctdb_client_async_control(ctdb, CTDB_CONTROL_WIPE_DATABASE,
1199 CONTROL_TIMEOUT(), false, data,
1202 DEBUG(DEBUG_ERR, (__location__ " Unable to wipe database. Recovery failed.\n"));
1207 /* push out the correct database. This sets the dmaster and skips
1208 the empty records */
1209 ret = push_recdb_database(ctdb, dbid, persistent, recdb, nodemap);
1215 /* all done with this database */
1222 reload the nodes file
1224 static void reload_nodes_file(struct ctdb_context *ctdb)
1227 ctdb_load_nodes_file(ctdb);
1230 static int ctdb_reload_remote_public_ips(struct ctdb_context *ctdb,
1231 struct ctdb_node_map *nodemap,
1237 if (ctdb->num_nodes != nodemap->num) {
1238 DEBUG(DEBUG_ERR, (__location__ " ctdb->num_nodes (%d) != nodemap->num (%d) invalid param\n",
1239 ctdb->num_nodes, nodemap->num));
1241 *culprit = ctdb->pnn;
1246 for (j=0; j<nodemap->num; j++) {
1247 /* release any existing data */
1248 if (ctdb->nodes[j]->known_public_ips) {
1249 talloc_free(ctdb->nodes[j]->known_public_ips);
1250 ctdb->nodes[j]->known_public_ips = NULL;
1252 if (ctdb->nodes[j]->available_public_ips) {
1253 talloc_free(ctdb->nodes[j]->available_public_ips);
1254 ctdb->nodes[j]->available_public_ips = NULL;
1257 if (nodemap->nodes[j].flags & NODE_FLAGS_INACTIVE) {
1261 /* grab a new shiny list of public ips from the node */
1262 ret = ctdb_ctrl_get_public_ips_flags(ctdb,
1264 ctdb->nodes[j]->pnn,
1267 &ctdb->nodes[j]->known_public_ips);
1269 DEBUG(DEBUG_ERR,("Failed to read known public ips from node : %u\n",
1270 ctdb->nodes[j]->pnn));
1272 *culprit = ctdb->nodes[j]->pnn;
1277 /* grab a new shiny list of public ips from the node */
1278 ret = ctdb_ctrl_get_public_ips_flags(ctdb,
1280 ctdb->nodes[j]->pnn,
1282 CTDB_PUBLIC_IP_FLAGS_ONLY_AVAILABLE,
1283 &ctdb->nodes[j]->available_public_ips);
1285 DEBUG(DEBUG_ERR,("Failed to read available public ips from node : %u\n",
1286 ctdb->nodes[j]->pnn));
1288 *culprit = ctdb->nodes[j]->pnn;
1298 we are the recmaster, and recovery is needed - start a recovery run
1300 static int do_recovery(struct ctdb_recoverd *rec,
1301 TALLOC_CTX *mem_ctx, uint32_t pnn,
1302 struct ctdb_node_map *nodemap, struct ctdb_vnn_map *vnnmap)
1304 struct ctdb_context *ctdb = rec->ctdb;
1306 uint32_t generation;
1307 struct ctdb_dbid_map *dbmap;
1310 struct timeval start_time;
1311 uint32_t culprit = (uint32_t)-1;
1313 DEBUG(DEBUG_NOTICE, (__location__ " Starting do_recovery\n"));
1315 /* if recovery fails, force it again */
1316 rec->need_recovery = true;
1318 for (i=0; i<ctdb->num_nodes; i++) {
1319 struct ctdb_banning_state *ban_state;
1321 if (ctdb->nodes[i]->ban_state == NULL) {
1324 ban_state = (struct ctdb_banning_state *)ctdb->nodes[i]->ban_state;
1325 if (ban_state->count < 2*ctdb->num_nodes) {
1328 DEBUG(DEBUG_NOTICE,("Node %u has caused %u recoveries recently - banning it for %u seconds\n",
1329 ctdb->nodes[i]->pnn, ban_state->count,
1330 ctdb->tunable.recovery_ban_period));
1331 ctdb_ban_node(rec, ctdb->nodes[i]->pnn, ctdb->tunable.recovery_ban_period);
1332 ban_state->count = 0;
1336 if (ctdb->tunable.verify_recovery_lock != 0) {
1337 DEBUG(DEBUG_ERR,("Taking out recovery lock from recovery daemon\n"));
1338 start_time = timeval_current();
1339 if (!ctdb_recovery_lock(ctdb, true)) {
1340 ctdb_set_culprit(rec, pnn);
1341 DEBUG(DEBUG_ERR,("Unable to get recovery lock - aborting recovery\n"));
1344 ctdb_ctrl_report_recd_lock_latency(ctdb, CONTROL_TIMEOUT(), timeval_elapsed(&start_time));
1345 DEBUG(DEBUG_ERR,("Recovery lock taken successfully by recovery daemon\n"));
1348 DEBUG(DEBUG_NOTICE, (__location__ " Recovery initiated due to problem with node %u\n", rec->last_culprit_node));
1350 /* get a list of all databases */
1351 ret = ctdb_ctrl_getdbmap(ctdb, CONTROL_TIMEOUT(), pnn, mem_ctx, &dbmap);
1353 DEBUG(DEBUG_ERR, (__location__ " Unable to get dbids from node :%u\n", pnn));
1357 /* we do the db creation before we set the recovery mode, so the freeze happens
1358 on all databases we will be dealing with. */
1360 /* verify that we have all the databases any other node has */
1361 ret = create_missing_local_databases(ctdb, nodemap, pnn, &dbmap, mem_ctx);
1363 DEBUG(DEBUG_ERR, (__location__ " Unable to create missing local databases\n"));
1367 /* verify that all other nodes have all our databases */
1368 ret = create_missing_remote_databases(ctdb, nodemap, pnn, dbmap, mem_ctx);
1370 DEBUG(DEBUG_ERR, (__location__ " Unable to create missing remote databases\n"));
1373 DEBUG(DEBUG_NOTICE, (__location__ " Recovery - created remote databases\n"));
1375 /* update the database priority for all remote databases */
1376 ret = update_db_priority_on_remote_nodes(ctdb, nodemap, pnn, dbmap, mem_ctx);
1378 DEBUG(DEBUG_ERR, (__location__ " Unable to set db priority on remote nodes\n"));
1380 DEBUG(DEBUG_NOTICE, (__location__ " Recovery - updated db priority for all databases\n"));
1383 /* set recovery mode to active on all nodes */
1384 ret = set_recovery_mode(ctdb, rec, nodemap, CTDB_RECOVERY_ACTIVE);
1386 DEBUG(DEBUG_ERR, (__location__ " Unable to set recovery mode to active on cluster\n"));
1390 /* execute the "startrecovery" event script on all nodes */
1391 ret = run_startrecovery_eventscript(rec, nodemap);
1393 DEBUG(DEBUG_ERR, (__location__ " Unable to run the 'startrecovery' event on cluster\n"));
1398 update all nodes to have the same flags that we have
1400 for (i=0;i<nodemap->num;i++) {
1401 if (nodemap->nodes[i].flags & NODE_FLAGS_DISCONNECTED) {
1405 ret = update_flags_on_all_nodes(ctdb, nodemap, i, nodemap->nodes[i].flags);
1407 DEBUG(DEBUG_ERR, (__location__ " Unable to update flags on all nodes for node %d\n", i));
1412 DEBUG(DEBUG_NOTICE, (__location__ " Recovery - updated flags\n"));
1414 /* pick a new generation number */
1415 generation = new_generation();
1417 /* change the vnnmap on this node to use the new generation
1418 number but not on any other nodes.
1419 this guarantees that if we abort the recovery prematurely
1420 for some reason (a node stops responding?)
1421 that we can just return immediately and we will reenter
1422 recovery shortly again.
1423 I.e. we deliberately leave the cluster with an inconsistent
1424 generation id to allow us to abort recovery at any stage and
1425 just restart it from scratch.
1427 vnnmap->generation = generation;
1428 ret = ctdb_ctrl_setvnnmap(ctdb, CONTROL_TIMEOUT(), pnn, mem_ctx, vnnmap);
1430 DEBUG(DEBUG_ERR, (__location__ " Unable to set vnnmap for node %u\n", pnn));
1434 data.dptr = (void *)&generation;
1435 data.dsize = sizeof(uint32_t);
1437 nodes = list_of_active_nodes(ctdb, nodemap, mem_ctx, true);
1438 if (ctdb_client_async_control(ctdb, CTDB_CONTROL_TRANSACTION_START,
1440 CONTROL_TIMEOUT(), false, data,
1442 transaction_start_fail_callback,
1444 DEBUG(DEBUG_ERR, (__location__ " Unable to start transactions. Recovery failed.\n"));
1445 if (ctdb_client_async_control(ctdb, CTDB_CONTROL_TRANSACTION_CANCEL,
1447 CONTROL_TIMEOUT(), false, tdb_null,
1451 DEBUG(DEBUG_ERR,("Failed to cancel recovery transaction\n"));
1456 DEBUG(DEBUG_NOTICE,(__location__ " started transactions on all nodes\n"));
1458 for (i=0;i<dbmap->num;i++) {
1459 ret = recover_database(rec, mem_ctx,
1461 dbmap->dbs[i].persistent,
1462 pnn, nodemap, generation);
1464 DEBUG(DEBUG_ERR, (__location__ " Failed to recover database 0x%x\n", dbmap->dbs[i].dbid));
1469 DEBUG(DEBUG_NOTICE, (__location__ " Recovery - starting database commits\n"));
1471 /* commit all the changes */
1472 if (ctdb_client_async_control(ctdb, CTDB_CONTROL_TRANSACTION_COMMIT,
1474 CONTROL_TIMEOUT(), false, data,
1477 DEBUG(DEBUG_ERR, (__location__ " Unable to commit recovery changes. Recovery failed.\n"));
1481 DEBUG(DEBUG_NOTICE, (__location__ " Recovery - committed databases\n"));
1484 /* update the capabilities for all nodes */
1485 ret = update_capabilities(ctdb, nodemap);
1487 DEBUG(DEBUG_ERR, (__location__ " Unable to update node capabilities.\n"));
1491 /* build a new vnn map with all the currently active and
1493 generation = new_generation();
1494 vnnmap = talloc(mem_ctx, struct ctdb_vnn_map);
1495 CTDB_NO_MEMORY(ctdb, vnnmap);
1496 vnnmap->generation = generation;
1498 vnnmap->map = talloc_zero_array(vnnmap, uint32_t, vnnmap->size);
1499 CTDB_NO_MEMORY(ctdb, vnnmap->map);
1500 for (i=j=0;i<nodemap->num;i++) {
1501 if (nodemap->nodes[i].flags & NODE_FLAGS_INACTIVE) {
1504 if (!(ctdb->nodes[i]->capabilities & CTDB_CAP_LMASTER)) {
1505 /* this node can not be an lmaster */
1506 DEBUG(DEBUG_DEBUG, ("Node %d cant be a LMASTER, skipping it\n", i));
1511 vnnmap->map = talloc_realloc(vnnmap, vnnmap->map, uint32_t, vnnmap->size);
1512 CTDB_NO_MEMORY(ctdb, vnnmap->map);
1513 vnnmap->map[j++] = nodemap->nodes[i].pnn;
1516 if (vnnmap->size == 0) {
1517 DEBUG(DEBUG_NOTICE, ("No suitable lmasters found. Adding local node (recmaster) anyway.\n"));
1519 vnnmap->map = talloc_realloc(vnnmap, vnnmap->map, uint32_t, vnnmap->size);
1520 CTDB_NO_MEMORY(ctdb, vnnmap->map);
1521 vnnmap->map[0] = pnn;
1524 /* update to the new vnnmap on all nodes */
1525 ret = update_vnnmap_on_all_nodes(ctdb, nodemap, pnn, vnnmap, mem_ctx);
1527 DEBUG(DEBUG_ERR, (__location__ " Unable to update vnnmap on all nodes\n"));
1531 DEBUG(DEBUG_NOTICE, (__location__ " Recovery - updated vnnmap\n"));
1533 /* update recmaster to point to us for all nodes */
1534 ret = set_recovery_master(ctdb, nodemap, pnn);
1536 DEBUG(DEBUG_ERR, (__location__ " Unable to set recovery master\n"));
1540 DEBUG(DEBUG_NOTICE, (__location__ " Recovery - updated recmaster\n"));
1543 update all nodes to have the same flags that we have
1545 for (i=0;i<nodemap->num;i++) {
1546 if (nodemap->nodes[i].flags & NODE_FLAGS_DISCONNECTED) {
1550 ret = update_flags_on_all_nodes(ctdb, nodemap, i, nodemap->nodes[i].flags);
1552 DEBUG(DEBUG_ERR, (__location__ " Unable to update flags on all nodes for node %d\n", i));
1557 DEBUG(DEBUG_NOTICE, (__location__ " Recovery - updated flags\n"));
1559 /* disable recovery mode */
1560 ret = set_recovery_mode(ctdb, rec, nodemap, CTDB_RECOVERY_NORMAL);
1562 DEBUG(DEBUG_ERR, (__location__ " Unable to set recovery mode to normal on cluster\n"));
1566 DEBUG(DEBUG_NOTICE, (__location__ " Recovery - disabled recovery mode\n"));
1569 tell nodes to takeover their public IPs
1571 ret = ctdb_reload_remote_public_ips(ctdb, nodemap, &culprit);
1573 DEBUG(DEBUG_ERR,("Failed to read public ips from remote node %d\n",
1577 rec->need_takeover_run = false;
1578 ret = ctdb_takeover_run(ctdb, nodemap);
1580 DEBUG(DEBUG_ERR, (__location__ " Unable to setup public takeover addresses\n"));
1583 DEBUG(DEBUG_NOTICE, (__location__ " Recovery - takeip finished\n"));
1585 /* execute the "recovered" event script on all nodes */
1586 ret = run_recovered_eventscript(ctdb, nodemap, "do_recovery");
1588 DEBUG(DEBUG_ERR, (__location__ " Unable to run the 'recovered' event on cluster. Recovery process failed.\n"));
1592 DEBUG(DEBUG_NOTICE, (__location__ " Recovery - finished the recovered event\n"));
1594 /* send a message to all clients telling them that the cluster
1595 has been reconfigured */
1596 ctdb_send_message(ctdb, CTDB_BROADCAST_CONNECTED, CTDB_SRVID_RECONFIGURE, tdb_null);
1598 DEBUG(DEBUG_NOTICE, (__location__ " Recovery complete\n"));
1600 rec->need_recovery = false;
1602 /* we managed to complete a full recovery, make sure to forgive
1603 any past sins by the nodes that could now participate in the
1606 DEBUG(DEBUG_ERR,("Resetting ban count to 0 for all nodes\n"));
1607 for (i=0;i<nodemap->num;i++) {
1608 struct ctdb_banning_state *ban_state;
1610 if (nodemap->nodes[i].flags & NODE_FLAGS_DISCONNECTED) {
1614 ban_state = (struct ctdb_banning_state *)ctdb->nodes[nodemap->nodes[i].pnn]->ban_state;
1615 if (ban_state == NULL) {
1619 ban_state->count = 0;
1623 /* We just finished a recovery successfully.
1624 We now wait for rerecovery_timeout before we allow
1625 another recovery to take place.
1627 DEBUG(DEBUG_NOTICE, (__location__ " New recoveries supressed for the rerecovery timeout\n"));
1628 ctdb_wait_timeout(ctdb, ctdb->tunable.rerecovery_timeout);
1629 DEBUG(DEBUG_NOTICE, (__location__ " Rerecovery timeout elapsed. Recovery reactivated.\n"));
1636 elections are won by first checking the number of connected nodes, then
1637 the priority time, then the pnn
1639 struct election_message {
1640 uint32_t num_connected;
1641 struct timeval priority_time;
1643 uint32_t node_flags;
1647 form this nodes election data
1649 static void ctdb_election_data(struct ctdb_recoverd *rec, struct election_message *em)
1652 struct ctdb_node_map *nodemap;
1653 struct ctdb_context *ctdb = rec->ctdb;
1657 em->pnn = rec->ctdb->pnn;
1658 em->priority_time = rec->priority_time;
1660 ret = ctdb_ctrl_getnodemap(ctdb, CONTROL_TIMEOUT(), CTDB_CURRENT_NODE, rec, &nodemap);
1662 DEBUG(DEBUG_ERR,(__location__ " unable to get election data\n"));
1666 rec->node_flags = nodemap->nodes[ctdb->pnn].flags;
1667 em->node_flags = rec->node_flags;
1669 for (i=0;i<nodemap->num;i++) {
1670 if (!(nodemap->nodes[i].flags & NODE_FLAGS_DISCONNECTED)) {
1671 em->num_connected++;
1675 /* we shouldnt try to win this election if we cant be a recmaster */
1676 if ((ctdb->capabilities & CTDB_CAP_RECMASTER) == 0) {
1677 em->num_connected = 0;
1678 em->priority_time = timeval_current();
1681 talloc_free(nodemap);
1685 see if the given election data wins
1687 static bool ctdb_election_win(struct ctdb_recoverd *rec, struct election_message *em)
1689 struct election_message myem;
1692 ctdb_election_data(rec, &myem);
1694 /* we cant win if we dont have the recmaster capability */
1695 if ((rec->ctdb->capabilities & CTDB_CAP_RECMASTER) == 0) {
1699 /* we cant win if we are banned */
1700 if (rec->node_flags & NODE_FLAGS_BANNED) {
1704 /* we cant win if we are stopped */
1705 if (rec->node_flags & NODE_FLAGS_STOPPED) {
1709 /* we will automatically win if the other node is banned */
1710 if (em->node_flags & NODE_FLAGS_BANNED) {
1714 /* we will automatically win if the other node is banned */
1715 if (em->node_flags & NODE_FLAGS_STOPPED) {
1719 /* try to use the most connected node */
1721 cmp = (int)myem.num_connected - (int)em->num_connected;
1724 /* then the longest running node */
1726 cmp = timeval_compare(&em->priority_time, &myem.priority_time);
1730 cmp = (int)myem.pnn - (int)em->pnn;
1737 send out an election request
1739 static int send_election_request(struct ctdb_recoverd *rec, uint32_t pnn, bool update_recmaster)
1742 TDB_DATA election_data;
1743 struct election_message emsg;
1745 struct ctdb_context *ctdb = rec->ctdb;
1747 srvid = CTDB_SRVID_RECOVERY;
1749 ctdb_election_data(rec, &emsg);
1751 election_data.dsize = sizeof(struct election_message);
1752 election_data.dptr = (unsigned char *)&emsg;
1755 /* send an election message to all active nodes */
1756 DEBUG(DEBUG_INFO,(__location__ " Send election request to all active nodes\n"));
1757 ctdb_send_message(ctdb, CTDB_BROADCAST_ALL, srvid, election_data);
1760 /* A new node that is already frozen has entered the cluster.
1761 The existing nodes are not frozen and dont need to be frozen
1762 until the election has ended and we start the actual recovery
1764 if (update_recmaster == true) {
1765 /* first we assume we will win the election and set
1766 recoverymaster to be ourself on the current node
1768 ret = ctdb_ctrl_setrecmaster(ctdb, CONTROL_TIMEOUT(), pnn, pnn);
1770 DEBUG(DEBUG_ERR, (__location__ " failed to send recmaster election request\n"));
1780 this function will unban all nodes in the cluster
1782 static void unban_all_nodes(struct ctdb_context *ctdb)
1785 struct ctdb_node_map *nodemap;
1786 TALLOC_CTX *tmp_ctx = talloc_new(ctdb);
1788 ret = ctdb_ctrl_getnodemap(ctdb, CONTROL_TIMEOUT(), CTDB_CURRENT_NODE, tmp_ctx, &nodemap);
1790 DEBUG(DEBUG_ERR,(__location__ " failed to get nodemap to unban all nodes\n"));
1794 for (i=0;i<nodemap->num;i++) {
1795 if ( (!(nodemap->nodes[i].flags & NODE_FLAGS_DISCONNECTED))
1796 && (nodemap->nodes[i].flags & NODE_FLAGS_BANNED) ) {
1797 ctdb_ctrl_modflags(ctdb, CONTROL_TIMEOUT(), nodemap->nodes[i].pnn, 0, NODE_FLAGS_BANNED);
1801 talloc_free(tmp_ctx);
1806 we think we are winning the election - send a broadcast election request
1808 static void election_send_request(struct event_context *ev, struct timed_event *te, struct timeval t, void *p)
1810 struct ctdb_recoverd *rec = talloc_get_type(p, struct ctdb_recoverd);
1813 ret = send_election_request(rec, ctdb_get_pnn(rec->ctdb), false);
1815 DEBUG(DEBUG_ERR,("Failed to send election request!\n"));
1818 talloc_free(rec->send_election_te);
1819 rec->send_election_te = NULL;
1823 handler for memory dumps
1825 static void mem_dump_handler(struct ctdb_context *ctdb, uint64_t srvid,
1826 TDB_DATA data, void *private_data)
1828 TALLOC_CTX *tmp_ctx = talloc_new(ctdb);
1831 struct rd_memdump_reply *rd;
1833 if (data.dsize != sizeof(struct rd_memdump_reply)) {
1834 DEBUG(DEBUG_ERR, (__location__ " Wrong size of return address.\n"));
1835 talloc_free(tmp_ctx);
1838 rd = (struct rd_memdump_reply *)data.dptr;
1840 dump = talloc_zero(tmp_ctx, TDB_DATA);
1842 DEBUG(DEBUG_ERR, (__location__ " Failed to allocate memory for memdump\n"));
1843 talloc_free(tmp_ctx);
1846 ret = ctdb_dump_memory(ctdb, dump);
1848 DEBUG(DEBUG_ERR, (__location__ " ctdb_dump_memory() failed\n"));
1849 talloc_free(tmp_ctx);
1853 DEBUG(DEBUG_ERR, ("recovery master memory dump\n"));
1855 ret = ctdb_send_message(ctdb, rd->pnn, rd->srvid, *dump);
1857 DEBUG(DEBUG_ERR,("Failed to send rd memdump reply message\n"));
1858 talloc_free(tmp_ctx);
1862 talloc_free(tmp_ctx);
1866 handler for reload_nodes
1868 static void reload_nodes_handler(struct ctdb_context *ctdb, uint64_t srvid,
1869 TDB_DATA data, void *private_data)
1871 struct ctdb_recoverd *rec = talloc_get_type(private_data, struct ctdb_recoverd);
1873 DEBUG(DEBUG_ERR, (__location__ " Reload nodes file from recovery daemon\n"));
1875 reload_nodes_file(rec->ctdb);
1879 static void reenable_ip_check(struct event_context *ev, struct timed_event *te,
1880 struct timeval yt, void *p)
1882 struct ctdb_recoverd *rec = talloc_get_type(p, struct ctdb_recoverd);
1884 talloc_free(rec->ip_check_disable_ctx);
1885 rec->ip_check_disable_ctx = NULL;
1888 static void disable_ip_check_handler(struct ctdb_context *ctdb, uint64_t srvid,
1889 TDB_DATA data, void *private_data)
1891 struct ctdb_recoverd *rec = talloc_get_type(private_data, struct ctdb_recoverd);
1894 if (rec->ip_check_disable_ctx != NULL) {
1895 talloc_free(rec->ip_check_disable_ctx);
1896 rec->ip_check_disable_ctx = NULL;
1899 if (data.dsize != sizeof(uint32_t)) {
1900 DEBUG(DEBUG_ERR,(__location__ " Wrong size for data :%lu "
1901 "expexting %lu\n", (long unsigned)data.dsize,
1902 (long unsigned)sizeof(uint32_t)));
1905 if (data.dptr == NULL) {
1906 DEBUG(DEBUG_ERR,(__location__ " No data recaived\n"));
1910 timeout = *((uint32_t *)data.dptr);
1911 DEBUG(DEBUG_NOTICE,("Disabling ip check for %u seconds\n", timeout));
1913 rec->ip_check_disable_ctx = talloc_new(rec);
1914 CTDB_NO_MEMORY_VOID(ctdb, rec->ip_check_disable_ctx);
1916 event_add_timed(ctdb->ev, rec->ip_check_disable_ctx, timeval_current_ofs(timeout, 0), reenable_ip_check, rec);
1921 handler for ip reallocate, just add it to the list of callers and
1922 handle this later in the monitor_cluster loop so we do not recurse
1923 with other callers to takeover_run()
1925 static void ip_reallocate_handler(struct ctdb_context *ctdb, uint64_t srvid,
1926 TDB_DATA data, void *private_data)
1928 struct ctdb_recoverd *rec = talloc_get_type(private_data, struct ctdb_recoverd);
1929 struct ip_reallocate_list *caller;
1931 if (data.dsize != sizeof(struct rd_memdump_reply)) {
1932 DEBUG(DEBUG_ERR, (__location__ " Wrong size of return address.\n"));
1936 if (rec->ip_reallocate_ctx == NULL) {
1937 rec->ip_reallocate_ctx = talloc_new(rec);
1938 CTDB_NO_MEMORY_FATAL(ctdb, rec->ip_reallocate_ctx);
1941 caller = talloc(rec->ip_reallocate_ctx, struct ip_reallocate_list);
1942 CTDB_NO_MEMORY_FATAL(ctdb, caller);
1944 caller->rd = (struct rd_memdump_reply *)talloc_steal(caller, data.dptr);
1945 caller->next = rec->reallocate_callers;
1946 rec->reallocate_callers = caller;
1951 static void process_ipreallocate_requests(struct ctdb_context *ctdb, struct ctdb_recoverd *rec)
1953 TALLOC_CTX *tmp_ctx = talloc_new(ctdb);
1956 struct ip_reallocate_list *callers;
1959 DEBUG(DEBUG_INFO, ("recovery master forced ip reallocation\n"));
1961 /* update the list of public ips that a node can handle for
1964 ret = ctdb_reload_remote_public_ips(ctdb, rec->nodemap, &culprit);
1966 DEBUG(DEBUG_ERR,("Failed to read public ips from remote node %d\n",
1968 rec->need_takeover_run = true;
1971 ret = ctdb_takeover_run(ctdb, rec->nodemap);
1973 DEBUG(DEBUG_ERR,("Failed to read public ips from remote node %d\n",
1975 rec->need_takeover_run = true;
1979 result.dsize = sizeof(int32_t);
1980 result.dptr = (uint8_t *)&ret;
1982 for (callers=rec->reallocate_callers; callers; callers=callers->next) {
1984 /* Someone that sent srvid==0 does not want a reply */
1985 if (callers->rd->srvid == 0) {
1988 DEBUG(DEBUG_INFO,("Sending ip reallocate reply message to "
1989 "%u:%llu\n", (unsigned)callers->rd->pnn,
1990 (unsigned long long)callers->rd->srvid));
1991 ret = ctdb_send_message(ctdb, callers->rd->pnn, callers->rd->srvid, result);
1993 DEBUG(DEBUG_ERR,("Failed to send ip reallocate reply "
1994 "message to %u:%llu\n",
1995 (unsigned)callers->rd->pnn,
1996 (unsigned long long)callers->rd->srvid));
2000 talloc_free(tmp_ctx);
2001 talloc_free(rec->ip_reallocate_ctx);
2002 rec->ip_reallocate_ctx = NULL;
2003 rec->reallocate_callers = NULL;
2009 handler for recovery master elections
2011 static void election_handler(struct ctdb_context *ctdb, uint64_t srvid,
2012 TDB_DATA data, void *private_data)
2014 struct ctdb_recoverd *rec = talloc_get_type(private_data, struct ctdb_recoverd);
2016 struct election_message *em = (struct election_message *)data.dptr;
2017 TALLOC_CTX *mem_ctx;
2019 /* we got an election packet - update the timeout for the election */
2020 talloc_free(rec->election_timeout);
2021 rec->election_timeout = event_add_timed(ctdb->ev, ctdb,
2022 timeval_current_ofs(ctdb->tunable.election_timeout, 0),
2023 ctdb_election_timeout, rec);
2025 mem_ctx = talloc_new(ctdb);
2027 /* someone called an election. check their election data
2028 and if we disagree and we would rather be the elected node,
2029 send a new election message to all other nodes
2031 if (ctdb_election_win(rec, em)) {
2032 if (!rec->send_election_te) {
2033 rec->send_election_te = event_add_timed(ctdb->ev, rec,
2034 timeval_current_ofs(0, 500000),
2035 election_send_request, rec);
2037 talloc_free(mem_ctx);
2038 /*unban_all_nodes(ctdb);*/
2043 talloc_free(rec->send_election_te);
2044 rec->send_election_te = NULL;
2046 if (ctdb->tunable.verify_recovery_lock != 0) {
2047 /* release the recmaster lock */
2048 if (em->pnn != ctdb->pnn &&
2049 ctdb->recovery_lock_fd != -1) {
2050 close(ctdb->recovery_lock_fd);
2051 ctdb->recovery_lock_fd = -1;
2052 unban_all_nodes(ctdb);
2056 /* ok, let that guy become recmaster then */
2057 ret = ctdb_ctrl_setrecmaster(ctdb, CONTROL_TIMEOUT(), ctdb_get_pnn(ctdb), em->pnn);
2059 DEBUG(DEBUG_ERR, (__location__ " failed to send recmaster election request"));
2060 talloc_free(mem_ctx);
2064 talloc_free(mem_ctx);
2070 force the start of the election process
2072 static void force_election(struct ctdb_recoverd *rec, uint32_t pnn,
2073 struct ctdb_node_map *nodemap)
2076 struct ctdb_context *ctdb = rec->ctdb;
2078 DEBUG(DEBUG_INFO,(__location__ " Force an election\n"));
2080 /* set all nodes to recovery mode to stop all internode traffic */
2081 ret = set_recovery_mode(ctdb, rec, nodemap, CTDB_RECOVERY_ACTIVE);
2083 DEBUG(DEBUG_ERR, (__location__ " Unable to set recovery mode to active on cluster\n"));
2087 talloc_free(rec->election_timeout);
2088 rec->election_timeout = event_add_timed(ctdb->ev, ctdb,
2089 timeval_current_ofs(ctdb->tunable.election_timeout, 0),
2090 ctdb_election_timeout, rec);
2092 ret = send_election_request(rec, pnn, true);
2094 DEBUG(DEBUG_ERR, (__location__ " failed to initiate recmaster election"));
2098 /* wait for a few seconds to collect all responses */
2099 ctdb_wait_election(rec);
2105 handler for when a node changes its flags
2107 static void monitor_handler(struct ctdb_context *ctdb, uint64_t srvid,
2108 TDB_DATA data, void *private_data)
2111 struct ctdb_node_flag_change *c = (struct ctdb_node_flag_change *)data.dptr;
2112 struct ctdb_node_map *nodemap=NULL;
2113 TALLOC_CTX *tmp_ctx;
2114 uint32_t changed_flags;
2116 struct ctdb_recoverd *rec = talloc_get_type(private_data, struct ctdb_recoverd);
2117 int disabled_flag_changed;
2119 if (data.dsize != sizeof(*c)) {
2120 DEBUG(DEBUG_ERR,(__location__ "Invalid data in ctdb_node_flag_change\n"));
2124 tmp_ctx = talloc_new(ctdb);
2125 CTDB_NO_MEMORY_VOID(ctdb, tmp_ctx);
2127 ret = ctdb_ctrl_getnodemap(ctdb, CONTROL_TIMEOUT(), CTDB_CURRENT_NODE, tmp_ctx, &nodemap);
2129 DEBUG(DEBUG_ERR,(__location__ "ctdb_ctrl_getnodemap failed in monitor_handler\n"));
2130 talloc_free(tmp_ctx);
2135 for (i=0;i<nodemap->num;i++) {
2136 if (nodemap->nodes[i].pnn == c->pnn) break;
2139 if (i == nodemap->num) {
2140 DEBUG(DEBUG_CRIT,(__location__ "Flag change for non-existant node %u\n", c->pnn));
2141 talloc_free(tmp_ctx);
2145 changed_flags = c->old_flags ^ c->new_flags;
2147 if (nodemap->nodes[i].flags != c->new_flags) {
2148 DEBUG(DEBUG_NOTICE,("Node %u has changed flags - now 0x%x was 0x%x\n", c->pnn, c->new_flags, c->old_flags));
2151 disabled_flag_changed = (nodemap->nodes[i].flags ^ c->new_flags) & NODE_FLAGS_DISABLED;
2153 nodemap->nodes[i].flags = c->new_flags;
2155 ret = ctdb_ctrl_getrecmaster(ctdb, tmp_ctx, CONTROL_TIMEOUT(),
2156 CTDB_CURRENT_NODE, &ctdb->recovery_master);
2159 ret = ctdb_ctrl_getrecmode(ctdb, tmp_ctx, CONTROL_TIMEOUT(),
2160 CTDB_CURRENT_NODE, &ctdb->recovery_mode);
2164 ctdb->recovery_master == ctdb->pnn &&
2165 ctdb->recovery_mode == CTDB_RECOVERY_NORMAL) {
2166 /* Only do the takeover run if the perm disabled or unhealthy
2167 flags changed since these will cause an ip failover but not
2169 If the node became disconnected or banned this will also
2170 lead to an ip address failover but that is handled
2173 if (disabled_flag_changed) {
2174 rec->need_takeover_run = true;
2178 talloc_free(tmp_ctx);
2182 handler for when we need to push out flag changes ot all other nodes
2184 static void push_flags_handler(struct ctdb_context *ctdb, uint64_t srvid,
2185 TDB_DATA data, void *private_data)
2188 struct ctdb_node_flag_change *c = (struct ctdb_node_flag_change *)data.dptr;
2189 struct ctdb_node_map *nodemap=NULL;
2190 TALLOC_CTX *tmp_ctx = talloc_new(ctdb);
2194 /* find the recovery master */
2195 ret = ctdb_ctrl_getrecmaster(ctdb, tmp_ctx, CONTROL_TIMEOUT(), CTDB_CURRENT_NODE, &recmaster);
2197 DEBUG(DEBUG_ERR, (__location__ " Unable to get recmaster from local node\n"));
2198 talloc_free(tmp_ctx);
2202 /* read the node flags from the recmaster */
2203 ret = ctdb_ctrl_getnodemap(ctdb, CONTROL_TIMEOUT(), recmaster, tmp_ctx, &nodemap);
2205 DEBUG(DEBUG_ERR, (__location__ " Unable to get nodemap from node %u\n", c->pnn));
2206 talloc_free(tmp_ctx);
2209 if (c->pnn >= nodemap->num) {
2210 DEBUG(DEBUG_ERR,(__location__ " Nodemap from recmaster does not contain node %d\n", c->pnn));
2211 talloc_free(tmp_ctx);
2215 /* send the flags update to all connected nodes */
2216 nodes = list_of_connected_nodes(ctdb, nodemap, tmp_ctx, true);
2218 if (ctdb_client_async_control(ctdb, CTDB_CONTROL_MODIFY_FLAGS,
2219 nodes, 0, CONTROL_TIMEOUT(),
2223 DEBUG(DEBUG_ERR, (__location__ " ctdb_control to modify node flags failed\n"));
2225 talloc_free(tmp_ctx);
2229 talloc_free(tmp_ctx);
2233 struct verify_recmode_normal_data {
2235 enum monitor_result status;
2238 static void verify_recmode_normal_callback(struct ctdb_client_control_state *state)
2240 struct verify_recmode_normal_data *rmdata = talloc_get_type(state->async.private_data, struct verify_recmode_normal_data);
2243 /* one more node has responded with recmode data*/
2246 /* if we failed to get the recmode, then return an error and let
2247 the main loop try again.
2249 if (state->state != CTDB_CONTROL_DONE) {
2250 if (rmdata->status == MONITOR_OK) {
2251 rmdata->status = MONITOR_FAILED;
2256 /* if we got a response, then the recmode will be stored in the
2259 if (state->status != CTDB_RECOVERY_NORMAL) {
2260 DEBUG(DEBUG_NOTICE, (__location__ " Node:%u was in recovery mode. Restart recovery process\n", state->c->hdr.destnode));
2261 rmdata->status = MONITOR_RECOVERY_NEEDED;
2268 /* verify that all nodes are in normal recovery mode */
2269 static enum monitor_result verify_recmode(struct ctdb_context *ctdb, struct ctdb_node_map *nodemap)
2271 struct verify_recmode_normal_data *rmdata;
2272 TALLOC_CTX *mem_ctx = talloc_new(ctdb);
2273 struct ctdb_client_control_state *state;
2274 enum monitor_result status;
2277 rmdata = talloc(mem_ctx, struct verify_recmode_normal_data);
2278 CTDB_NO_MEMORY_FATAL(ctdb, rmdata);
2280 rmdata->status = MONITOR_OK;
2282 /* loop over all active nodes and send an async getrecmode call to
2284 for (j=0; j<nodemap->num; j++) {
2285 if (nodemap->nodes[j].flags & NODE_FLAGS_INACTIVE) {
2288 state = ctdb_ctrl_getrecmode_send(ctdb, mem_ctx,
2290 nodemap->nodes[j].pnn);
2291 if (state == NULL) {
2292 /* we failed to send the control, treat this as
2293 an error and try again next iteration
2295 DEBUG(DEBUG_ERR,("Failed to call ctdb_ctrl_getrecmode_send during monitoring\n"));
2296 talloc_free(mem_ctx);
2297 return MONITOR_FAILED;
2300 /* set up the callback functions */
2301 state->async.fn = verify_recmode_normal_callback;
2302 state->async.private_data = rmdata;
2304 /* one more control to wait for to complete */
2309 /* now wait for up to the maximum number of seconds allowed
2310 or until all nodes we expect a response from has replied
2312 while (rmdata->count > 0) {
2313 event_loop_once(ctdb->ev);
2316 status = rmdata->status;
2317 talloc_free(mem_ctx);
2322 struct verify_recmaster_data {
2323 struct ctdb_recoverd *rec;
2326 enum monitor_result status;
2329 static void verify_recmaster_callback(struct ctdb_client_control_state *state)
2331 struct verify_recmaster_data *rmdata = talloc_get_type(state->async.private_data, struct verify_recmaster_data);
2334 /* one more node has responded with recmaster data*/
2337 /* if we failed to get the recmaster, then return an error and let
2338 the main loop try again.
2340 if (state->state != CTDB_CONTROL_DONE) {
2341 if (rmdata->status == MONITOR_OK) {
2342 rmdata->status = MONITOR_FAILED;
2347 /* if we got a response, then the recmaster will be stored in the
2350 if (state->status != rmdata->pnn) {
2351 DEBUG(DEBUG_ERR,("Node %d does not agree we are the recmaster. Need a new recmaster election\n", state->c->hdr.destnode));
2352 ctdb_set_culprit(rmdata->rec, state->c->hdr.destnode);
2353 rmdata->status = MONITOR_ELECTION_NEEDED;
2360 /* verify that all nodes agree that we are the recmaster */
2361 static enum monitor_result verify_recmaster(struct ctdb_recoverd *rec, struct ctdb_node_map *nodemap, uint32_t pnn)
2363 struct ctdb_context *ctdb = rec->ctdb;
2364 struct verify_recmaster_data *rmdata;
2365 TALLOC_CTX *mem_ctx = talloc_new(ctdb);
2366 struct ctdb_client_control_state *state;
2367 enum monitor_result status;
2370 rmdata = talloc(mem_ctx, struct verify_recmaster_data);
2371 CTDB_NO_MEMORY_FATAL(ctdb, rmdata);
2375 rmdata->status = MONITOR_OK;
2377 /* loop over all active nodes and send an async getrecmaster call to
2379 for (j=0; j<nodemap->num; j++) {
2380 if (nodemap->nodes[j].flags & NODE_FLAGS_INACTIVE) {
2383 state = ctdb_ctrl_getrecmaster_send(ctdb, mem_ctx,
2385 nodemap->nodes[j].pnn);
2386 if (state == NULL) {
2387 /* we failed to send the control, treat this as
2388 an error and try again next iteration
2390 DEBUG(DEBUG_ERR,("Failed to call ctdb_ctrl_getrecmaster_send during monitoring\n"));
2391 talloc_free(mem_ctx);
2392 return MONITOR_FAILED;
2395 /* set up the callback functions */
2396 state->async.fn = verify_recmaster_callback;
2397 state->async.private_data = rmdata;
2399 /* one more control to wait for to complete */
2404 /* now wait for up to the maximum number of seconds allowed
2405 or until all nodes we expect a response from has replied
2407 while (rmdata->count > 0) {
2408 event_loop_once(ctdb->ev);
2411 status = rmdata->status;
2412 talloc_free(mem_ctx);
2417 /* called to check that the allocation of public ip addresses is ok.
2419 static int verify_ip_allocation(struct ctdb_context *ctdb, struct ctdb_recoverd *rec, uint32_t pnn)
2421 TALLOC_CTX *mem_ctx = talloc_new(NULL);
2422 struct ctdb_control_get_ifaces *ifaces = NULL;
2423 struct ctdb_all_public_ips *ips = NULL;
2424 struct ctdb_uptime *uptime1 = NULL;
2425 struct ctdb_uptime *uptime2 = NULL;
2427 bool need_iface_check = false;
2428 bool need_takeover_run = false;
2430 ret = ctdb_ctrl_uptime(ctdb, mem_ctx, CONTROL_TIMEOUT(),
2431 CTDB_CURRENT_NODE, &uptime1);
2433 DEBUG(DEBUG_ERR, ("Unable to get uptime from local node %u\n", pnn));
2434 talloc_free(mem_ctx);
2439 /* read the interfaces from the local node */
2440 ret = ctdb_ctrl_get_ifaces(ctdb, CONTROL_TIMEOUT(), CTDB_CURRENT_NODE, mem_ctx, &ifaces);
2442 DEBUG(DEBUG_ERR, ("Unable to get interfaces from local node %u\n", pnn));
2443 talloc_free(mem_ctx);
2448 need_iface_check = true;
2449 } else if (rec->ifaces->num != ifaces->num) {
2450 need_iface_check = true;
2451 } else if (memcmp(rec->ifaces, ifaces, talloc_get_size(ifaces)) != 0) {
2452 need_iface_check = true;
2455 if (need_iface_check) {
2456 DEBUG(DEBUG_NOTICE, ("The interfaces status has changed on "
2457 "local node %u - force takeover run\n",
2459 need_takeover_run = true;
2462 /* read the ip allocation from the local node */
2463 ret = ctdb_ctrl_get_public_ips(ctdb, CONTROL_TIMEOUT(), CTDB_CURRENT_NODE, mem_ctx, &ips);
2465 DEBUG(DEBUG_ERR, ("Unable to get public ips from local node %u\n", pnn));
2466 talloc_free(mem_ctx);
2470 ret = ctdb_ctrl_uptime(ctdb, mem_ctx, CONTROL_TIMEOUT(),
2471 CTDB_CURRENT_NODE, &uptime2);
2473 DEBUG(DEBUG_ERR, ("Unable to get uptime from local node %u\n", pnn));
2474 talloc_free(mem_ctx);
2478 /* skip the check if the startrecovery time has changed */
2479 if (timeval_compare(&uptime1->last_recovery_started,
2480 &uptime2->last_recovery_started) != 0) {
2481 DEBUG(DEBUG_NOTICE, (__location__ " last recovery time changed while we read the public ip list. skipping public ip address check\n"));
2482 talloc_free(mem_ctx);
2486 /* skip the check if the endrecovery time has changed */
2487 if (timeval_compare(&uptime1->last_recovery_finished,
2488 &uptime2->last_recovery_finished) != 0) {
2489 DEBUG(DEBUG_NOTICE, (__location__ " last recovery time changed while we read the public ip list. skipping public ip address check\n"));
2490 talloc_free(mem_ctx);
2494 /* skip the check if we have started but not finished recovery */
2495 if (timeval_compare(&uptime1->last_recovery_finished,
2496 &uptime1->last_recovery_started) != 1) {
2497 DEBUG(DEBUG_NOTICE, (__location__ " in the middle of recovery or ip reallocation. skipping public ip address check\n"));
2498 talloc_free(mem_ctx);
2503 talloc_free(rec->ifaces);
2504 rec->ifaces = talloc_steal(rec, ifaces);
2506 /* verify that we have the ip addresses we should have
2507 and we dont have ones we shouldnt have.
2508 if we find an inconsistency we set recmode to
2509 active on the local node and wait for the recmaster
2510 to do a full blown recovery
2512 for (j=0; j<ips->num; j++) {
2513 if (ips->ips[j].pnn == pnn) {
2514 if (!ctdb_sys_have_ip(&ips->ips[j].addr)) {
2515 DEBUG(DEBUG_CRIT,("Public address '%s' is missing and we should serve this ip\n",
2516 ctdb_addr_to_str(&ips->ips[j].addr)));
2517 need_takeover_run = true;
2520 if (ctdb_sys_have_ip(&ips->ips[j].addr)) {
2521 DEBUG(DEBUG_CRIT,("We are still serving a public address '%s' that we should not be serving.\n",
2522 ctdb_addr_to_str(&ips->ips[j].addr)));
2523 need_takeover_run = true;
2528 if (need_takeover_run) {
2529 struct takeover_run_reply rd;
2532 DEBUG(DEBUG_CRIT,("Trigger takeoverrun\n"));
2536 data.dptr = (uint8_t *)&rd;
2537 data.dsize = sizeof(rd);
2539 ret = ctdb_send_message(ctdb, rec->recmaster, CTDB_SRVID_TAKEOVER_RUN, data);
2541 DEBUG(DEBUG_ERR,(__location__ " Failed to send ipreallocate to recmaster :%d\n", (int)rec->recmaster));
2544 talloc_free(mem_ctx);
2549 static void async_getnodemap_callback(struct ctdb_context *ctdb, uint32_t node_pnn, int32_t res, TDB_DATA outdata, void *callback_data)
2551 struct ctdb_node_map **remote_nodemaps = callback_data;
2553 if (node_pnn >= ctdb->num_nodes) {
2554 DEBUG(DEBUG_ERR,(__location__ " pnn from invalid node\n"));
2558 remote_nodemaps[node_pnn] = (struct ctdb_node_map *)talloc_steal(remote_nodemaps, outdata.dptr);
2562 static int get_remote_nodemaps(struct ctdb_context *ctdb, TALLOC_CTX *mem_ctx,
2563 struct ctdb_node_map *nodemap,
2564 struct ctdb_node_map **remote_nodemaps)
2568 nodes = list_of_active_nodes(ctdb, nodemap, mem_ctx, true);
2569 if (ctdb_client_async_control(ctdb, CTDB_CONTROL_GET_NODEMAP,
2571 CONTROL_TIMEOUT(), false, tdb_null,
2572 async_getnodemap_callback,
2574 remote_nodemaps) != 0) {
2575 DEBUG(DEBUG_ERR, (__location__ " Unable to pull all remote nodemaps\n"));
2583 enum reclock_child_status { RECLOCK_CHECKING, RECLOCK_OK, RECLOCK_FAILED, RECLOCK_TIMEOUT};
2584 struct ctdb_check_reclock_state {
2585 struct ctdb_context *ctdb;
2586 struct timeval start_time;
2589 struct timed_event *te;
2590 struct fd_event *fde;
2591 enum reclock_child_status status;
2594 /* when we free the reclock state we must kill any child process.
2596 static int check_reclock_destructor(struct ctdb_check_reclock_state *state)
2598 struct ctdb_context *ctdb = state->ctdb;
2600 ctdb_ctrl_report_recd_lock_latency(ctdb, CONTROL_TIMEOUT(), timeval_elapsed(&state->start_time));
2602 if (state->fd[0] != -1) {
2603 close(state->fd[0]);
2606 if (state->fd[1] != -1) {
2607 close(state->fd[1]);
2610 kill(state->child, SIGKILL);
2615 called if our check_reclock child times out. this would happen if
2616 i/o to the reclock file blocks.
2618 static void ctdb_check_reclock_timeout(struct event_context *ev, struct timed_event *te,
2619 struct timeval t, void *private_data)
2621 struct ctdb_check_reclock_state *state = talloc_get_type(private_data,
2622 struct ctdb_check_reclock_state);
2624 DEBUG(DEBUG_ERR,(__location__ " check_reclock child process hung/timedout CFS slow to grant locks?\n"));
2625 state->status = RECLOCK_TIMEOUT;
2628 /* this is called when the child process has completed checking the reclock
2629 file and has written data back to us through the pipe.
2631 static void reclock_child_handler(struct event_context *ev, struct fd_event *fde,
2632 uint16_t flags, void *private_data)
2634 struct ctdb_check_reclock_state *state= talloc_get_type(private_data,
2635 struct ctdb_check_reclock_state);
2639 /* we got a response from our child process so we can abort the
2642 talloc_free(state->te);
2645 ret = read(state->fd[0], &c, 1);
2646 if (ret != 1 || c != RECLOCK_OK) {
2647 DEBUG(DEBUG_ERR,(__location__ " reclock child process returned error %d\n", c));
2648 state->status = RECLOCK_FAILED;
2653 state->status = RECLOCK_OK;
2657 static int check_recovery_lock(struct ctdb_context *ctdb)
2660 struct ctdb_check_reclock_state *state;
2661 pid_t parent = getpid();
2663 if (ctdb->recovery_lock_fd == -1) {
2664 DEBUG(DEBUG_CRIT,("recovery master doesn't have the recovery lock\n"));
2668 state = talloc(ctdb, struct ctdb_check_reclock_state);
2669 CTDB_NO_MEMORY(ctdb, state);
2672 state->start_time = timeval_current();
2673 state->status = RECLOCK_CHECKING;
2677 ret = pipe(state->fd);
2680 DEBUG(DEBUG_CRIT,(__location__ " Failed to open pipe for check_reclock child\n"));
2684 state->child = fork();
2685 if (state->child == (pid_t)-1) {
2686 DEBUG(DEBUG_CRIT,(__location__ " fork() failed in check_reclock child\n"));
2687 close(state->fd[0]);
2689 close(state->fd[1]);
2695 if (state->child == 0) {
2696 char cc = RECLOCK_OK;
2697 close(state->fd[0]);
2700 if (pread(ctdb->recovery_lock_fd, &cc, 1, 0) == -1) {
2701 DEBUG(DEBUG_CRIT,("failed read from recovery_lock_fd - %s\n", strerror(errno)));
2702 cc = RECLOCK_FAILED;
2705 write(state->fd[1], &cc, 1);
2706 /* make sure we die when our parent dies */
2707 while (kill(parent, 0) == 0 || errno != ESRCH) {
2709 write(state->fd[1], &cc, 1);
2713 close(state->fd[1]);
2715 set_close_on_exec(state->fd[0]);
2717 DEBUG(DEBUG_DEBUG, (__location__ " Created PIPE FD:%d for check_recovery_lock\n", state->fd[0]));
2719 talloc_set_destructor(state, check_reclock_destructor);
2721 state->te = event_add_timed(ctdb->ev, state, timeval_current_ofs(15, 0),
2722 ctdb_check_reclock_timeout, state);
2723 if (state->te == NULL) {
2724 DEBUG(DEBUG_CRIT,(__location__ " Failed to create a timed event for reclock child\n"));
2729 state->fde = event_add_fd(ctdb->ev, state, state->fd[0],
2730 EVENT_FD_READ|EVENT_FD_AUTOCLOSE,
2731 reclock_child_handler,
2734 if (state->fde == NULL) {
2735 DEBUG(DEBUG_CRIT,(__location__ " Failed to create an fd event for reclock child\n"));
2740 while (state->status == RECLOCK_CHECKING) {
2741 event_loop_once(ctdb->ev);
2744 if (state->status == RECLOCK_FAILED) {
2745 DEBUG(DEBUG_ERR,(__location__ " reclock child failed when checking file\n"));
2746 close(ctdb->recovery_lock_fd);
2747 ctdb->recovery_lock_fd = -1;
2756 static int update_recovery_lock_file(struct ctdb_context *ctdb)
2758 TALLOC_CTX *tmp_ctx = talloc_new(NULL);
2759 const char *reclockfile;
2761 if (ctdb_ctrl_getreclock(ctdb, CONTROL_TIMEOUT(), CTDB_CURRENT_NODE, tmp_ctx, &reclockfile) != 0) {
2762 DEBUG(DEBUG_ERR,("Failed to read reclock file from daemon\n"));
2763 talloc_free(tmp_ctx);
2767 if (reclockfile == NULL) {
2768 if (ctdb->recovery_lock_file != NULL) {
2769 DEBUG(DEBUG_ERR,("Reclock file disabled\n"));
2770 talloc_free(ctdb->recovery_lock_file);
2771 ctdb->recovery_lock_file = NULL;
2772 if (ctdb->recovery_lock_fd != -1) {
2773 close(ctdb->recovery_lock_fd);
2774 ctdb->recovery_lock_fd = -1;
2777 ctdb->tunable.verify_recovery_lock = 0;
2778 talloc_free(tmp_ctx);
2782 if (ctdb->recovery_lock_file == NULL) {
2783 ctdb->recovery_lock_file = talloc_strdup(ctdb, reclockfile);
2784 if (ctdb->recovery_lock_fd != -1) {
2785 close(ctdb->recovery_lock_fd);
2786 ctdb->recovery_lock_fd = -1;
2788 talloc_free(tmp_ctx);
2793 if (!strcmp(reclockfile, ctdb->recovery_lock_file)) {
2794 talloc_free(tmp_ctx);
2798 talloc_free(ctdb->recovery_lock_file);
2799 ctdb->recovery_lock_file = talloc_strdup(ctdb, reclockfile);
2800 ctdb->tunable.verify_recovery_lock = 0;
2801 if (ctdb->recovery_lock_fd != -1) {
2802 close(ctdb->recovery_lock_fd);
2803 ctdb->recovery_lock_fd = -1;
2806 talloc_free(tmp_ctx);
2811 the main monitoring loop
2813 static void monitor_cluster(struct ctdb_context *ctdb)
2816 TALLOC_CTX *mem_ctx=NULL;
2817 struct ctdb_node_map *nodemap=NULL;
2818 struct ctdb_node_map *recmaster_nodemap=NULL;
2819 struct ctdb_node_map **remote_nodemaps=NULL;
2820 struct ctdb_vnn_map *vnnmap=NULL;
2821 struct ctdb_vnn_map *remote_vnnmap=NULL;
2822 int32_t debug_level;
2824 struct ctdb_recoverd *rec;
2826 DEBUG(DEBUG_NOTICE,("monitor_cluster starting\n"));
2828 rec = talloc_zero(ctdb, struct ctdb_recoverd);
2829 CTDB_NO_MEMORY_FATAL(ctdb, rec);
2833 rec->priority_time = timeval_current();
2835 /* register a message port for sending memory dumps */
2836 ctdb_set_message_handler(ctdb, CTDB_SRVID_MEM_DUMP, mem_dump_handler, rec);
2838 /* register a message port for recovery elections */
2839 ctdb_set_message_handler(ctdb, CTDB_SRVID_RECOVERY, election_handler, rec);
2841 /* when nodes are disabled/enabled */
2842 ctdb_set_message_handler(ctdb, CTDB_SRVID_SET_NODE_FLAGS, monitor_handler, rec);
2844 /* when we are asked to puch out a flag change */
2845 ctdb_set_message_handler(ctdb, CTDB_SRVID_PUSH_NODE_FLAGS, push_flags_handler, rec);
2847 /* register a message port for vacuum fetch */
2848 ctdb_set_message_handler(ctdb, CTDB_SRVID_VACUUM_FETCH, vacuum_fetch_handler, rec);
2850 /* register a message port for reloadnodes */
2851 ctdb_set_message_handler(ctdb, CTDB_SRVID_RELOAD_NODES, reload_nodes_handler, rec);
2853 /* register a message port for performing a takeover run */
2854 ctdb_set_message_handler(ctdb, CTDB_SRVID_TAKEOVER_RUN, ip_reallocate_handler, rec);
2856 /* register a message port for disabling the ip check for a short while */
2857 ctdb_set_message_handler(ctdb, CTDB_SRVID_DISABLE_IP_CHECK, disable_ip_check_handler, rec);
2861 talloc_free(mem_ctx);
2864 mem_ctx = talloc_new(ctdb);
2866 DEBUG(DEBUG_CRIT,(__location__ " Failed to create temporary context\n"));
2870 /* we only check for recovery once every second */
2871 ctdb_wait_timeout(ctdb, ctdb->tunable.recover_interval);
2873 /* verify that the main daemon is still running */
2874 if (kill(ctdb->ctdbd_pid, 0) != 0) {
2875 DEBUG(DEBUG_CRIT,("CTDB daemon is no longer available. Shutting down recovery daemon\n"));
2879 /* ping the local daemon to tell it we are alive */
2880 ctdb_ctrl_recd_ping(ctdb);
2882 if (rec->election_timeout) {
2883 /* an election is in progress */
2887 /* read the debug level from the parent and update locally */
2888 ret = ctdb_ctrl_get_debuglevel(ctdb, CTDB_CURRENT_NODE, &debug_level);
2890 DEBUG(DEBUG_ERR, (__location__ " Failed to read debuglevel from parent\n"));
2893 LogLevel = debug_level;
2896 /* We must check if we need to ban a node here but we want to do this
2897 as early as possible so we dont wait until we have pulled the node
2898 map from the local node. thats why we have the hardcoded value 20
2900 for (i=0; i<ctdb->num_nodes; i++) {
2901 struct ctdb_banning_state *ban_state;
2903 if (ctdb->nodes[i]->ban_state == NULL) {
2906 ban_state = (struct ctdb_banning_state *)ctdb->nodes[i]->ban_state;
2907 if (ban_state->count < 20) {
2910 DEBUG(DEBUG_NOTICE,("Node %u has caused %u recoveries recently - banning it for %u seconds\n",
2911 ctdb->nodes[i]->pnn, ban_state->count,
2912 ctdb->tunable.recovery_ban_period));
2913 ctdb_ban_node(rec, ctdb->nodes[i]->pnn, ctdb->tunable.recovery_ban_period);
2914 ban_state->count = 0;
2917 /* get relevant tunables */
2918 ret = ctdb_ctrl_get_all_tunables(ctdb, CONTROL_TIMEOUT(), CTDB_CURRENT_NODE, &ctdb->tunable);
2920 DEBUG(DEBUG_ERR,("Failed to get tunables - retrying\n"));
2924 /* get the current recovery lock file from the server */
2925 if (update_recovery_lock_file(ctdb) != 0) {
2926 DEBUG(DEBUG_ERR,("Failed to update the recovery lock file\n"));
2930 /* Make sure that if recovery lock verification becomes disabled when
2933 if (ctdb->tunable.verify_recovery_lock == 0) {
2934 if (ctdb->recovery_lock_fd != -1) {
2935 close(ctdb->recovery_lock_fd);
2936 ctdb->recovery_lock_fd = -1;
2940 pnn = ctdb_ctrl_getpnn(ctdb, CONTROL_TIMEOUT(), CTDB_CURRENT_NODE);
2941 if (pnn == (uint32_t)-1) {
2942 DEBUG(DEBUG_ERR,("Failed to get local pnn - retrying\n"));
2946 /* get the vnnmap */
2947 ret = ctdb_ctrl_getvnnmap(ctdb, CONTROL_TIMEOUT(), pnn, mem_ctx, &vnnmap);
2949 DEBUG(DEBUG_ERR, (__location__ " Unable to get vnnmap from node %u\n", pnn));
2954 /* get number of nodes */
2956 talloc_free(rec->nodemap);
2957 rec->nodemap = NULL;
2960 ret = ctdb_ctrl_getnodemap(ctdb, CONTROL_TIMEOUT(), pnn, rec, &rec->nodemap);
2962 DEBUG(DEBUG_ERR, (__location__ " Unable to get nodemap from node %u\n", pnn));
2965 nodemap = rec->nodemap;
2967 /* check which node is the recovery master */
2968 ret = ctdb_ctrl_getrecmaster(ctdb, mem_ctx, CONTROL_TIMEOUT(), pnn, &rec->recmaster);
2970 DEBUG(DEBUG_ERR, (__location__ " Unable to get recmaster from node %u\n", pnn));
2974 /* if we are not the recmaster we can safely ignore any ip reallocate requests */
2975 if (rec->recmaster != pnn) {
2976 if (rec->ip_reallocate_ctx != NULL) {
2977 talloc_free(rec->ip_reallocate_ctx);
2978 rec->ip_reallocate_ctx = NULL;
2979 rec->reallocate_callers = NULL;
2982 /* if there are takeovers requested, perform it and notify the waiters */
2983 if (rec->reallocate_callers) {
2984 process_ipreallocate_requests(ctdb, rec);
2987 if (rec->recmaster == (uint32_t)-1) {
2988 DEBUG(DEBUG_NOTICE,(__location__ " Initial recovery master set - forcing election\n"));
2989 force_election(rec, pnn, nodemap);
2994 /* if the local daemon is STOPPED, we verify that the databases are
2995 also frozen and thet the recmode is set to active
2997 if (nodemap->nodes[pnn].flags & NODE_FLAGS_STOPPED) {
2998 ret = ctdb_ctrl_getrecmode(ctdb, mem_ctx, CONTROL_TIMEOUT(), CTDB_CURRENT_NODE, &ctdb->recovery_mode);
3000 DEBUG(DEBUG_ERR,(__location__ " Failed to read recmode from local node\n"));
3002 if (ctdb->recovery_mode == CTDB_RECOVERY_NORMAL) {
3003 DEBUG(DEBUG_ERR,("Node is stopped but recovery mode is not active. Activate recovery mode and lock databases\n"));
3005 ret = ctdb_ctrl_freeze_priority(ctdb, CONTROL_TIMEOUT(), CTDB_CURRENT_NODE, 1);
3007 DEBUG(DEBUG_ERR,(__location__ " Failed to freeze node due to node being STOPPED\n"));
3010 ret = ctdb_ctrl_setrecmode(ctdb, CONTROL_TIMEOUT(), CTDB_CURRENT_NODE, CTDB_RECOVERY_ACTIVE);
3012 DEBUG(DEBUG_ERR,(__location__ " Failed to activate recovery mode due to node being stopped\n"));
3019 /* If the local node is stopped, verify we are not the recmaster
3020 and yield this role if so
3022 if ((nodemap->nodes[pnn].flags & NODE_FLAGS_STOPPED) && (rec->recmaster == pnn)) {
3023 DEBUG(DEBUG_ERR,("Local node is STOPPED. Yielding recmaster role\n"));
3024 force_election(rec, pnn, nodemap);
3028 /* check that we (recovery daemon) and the local ctdb daemon
3029 agrees on whether we are banned or not
3033 /* remember our own node flags */
3034 rec->node_flags = nodemap->nodes[pnn].flags;
3036 /* count how many active nodes there are */
3037 rec->num_active = 0;
3038 rec->num_connected = 0;
3039 for (i=0; i<nodemap->num; i++) {
3040 if (!(nodemap->nodes[i].flags & NODE_FLAGS_INACTIVE)) {
3043 if (!(nodemap->nodes[i].flags & NODE_FLAGS_DISCONNECTED)) {
3044 rec->num_connected++;
3049 /* verify that the recmaster node is still active */
3050 for (j=0; j<nodemap->num; j++) {
3051 if (nodemap->nodes[j].pnn==rec->recmaster) {
3056 if (j == nodemap->num) {
3057 DEBUG(DEBUG_ERR, ("Recmaster node %u not in list. Force reelection\n", rec->recmaster));
3058 force_election(rec, pnn, nodemap);
3062 /* if recovery master is disconnected we must elect a new recmaster */
3063 if (nodemap->nodes[j].flags & NODE_FLAGS_DISCONNECTED) {
3064 DEBUG(DEBUG_NOTICE, ("Recmaster node %u is disconnected. Force reelection\n", nodemap->nodes[j].pnn));
3065 force_election(rec, pnn, nodemap);
3069 /* grap the nodemap from the recovery master to check if it is banned */
3070 ret = ctdb_ctrl_getnodemap(ctdb, CONTROL_TIMEOUT(), nodemap->nodes[j].pnn,
3071 mem_ctx, &recmaster_nodemap);
3073 DEBUG(DEBUG_ERR, (__location__ " Unable to get nodemap from recovery master %u\n",
3074 nodemap->nodes[j].pnn));
3079 if (recmaster_nodemap->nodes[j].flags & NODE_FLAGS_INACTIVE) {
3080 DEBUG(DEBUG_NOTICE, ("Recmaster node %u no longer available. Force reelection\n", nodemap->nodes[j].pnn));
3081 force_election(rec, pnn, nodemap);
3086 /* verify that we have all ip addresses we should have and we dont
3087 * have addresses we shouldnt have.
3089 if (ctdb->do_checkpublicip) {
3090 if (rec->ip_check_disable_ctx == NULL) {
3091 if (verify_ip_allocation(ctdb, rec, pnn) != 0) {
3092 DEBUG(DEBUG_ERR, (__location__ " Public IPs were inconsistent.\n"));
3098 /* if we are not the recmaster then we do not need to check
3099 if recovery is needed
3101 if (pnn != rec->recmaster) {
3106 /* ensure our local copies of flags are right */
3107 ret = update_local_flags(rec, nodemap);
3108 if (ret == MONITOR_ELECTION_NEEDED) {
3109 DEBUG(DEBUG_NOTICE,("update_local_flags() called for a re-election.\n"));
3110 force_election(rec, pnn, nodemap);
3113 if (ret != MONITOR_OK) {
3114 DEBUG(DEBUG_ERR,("Unable to update local flags\n"));
3118 if (ctdb->num_nodes != nodemap->num) {
3119 DEBUG(DEBUG_ERR, (__location__ " ctdb->num_nodes (%d) != nodemap->num (%d) reloading nodes file\n", ctdb->num_nodes, nodemap->num));
3120 reload_nodes_file(ctdb);
3124 /* verify that all active nodes agree that we are the recmaster */
3125 switch (verify_recmaster(rec, nodemap, pnn)) {
3126 case MONITOR_RECOVERY_NEEDED:
3127 /* can not happen */
3129 case MONITOR_ELECTION_NEEDED:
3130 force_election(rec, pnn, nodemap);
3134 case MONITOR_FAILED:
3139 if (rec->need_recovery) {
3140 /* a previous recovery didn't finish */
3141 do_recovery(rec, mem_ctx, pnn, nodemap, vnnmap);
3145 /* verify that all active nodes are in normal mode
3146 and not in recovery mode
3148 switch (verify_recmode(ctdb, nodemap)) {
3149 case MONITOR_RECOVERY_NEEDED:
3150 do_recovery(rec, mem_ctx, pnn, nodemap, vnnmap);
3152 case MONITOR_FAILED:
3154 case MONITOR_ELECTION_NEEDED:
3155 /* can not happen */
3161 if (ctdb->tunable.verify_recovery_lock != 0) {
3162 /* we should have the reclock - check its not stale */
3163 ret = check_recovery_lock(ctdb);
3165 DEBUG(DEBUG_ERR,("Failed check_recovery_lock. Force a recovery\n"));
3166 ctdb_set_culprit(rec, ctdb->pnn);
3167 do_recovery(rec, mem_ctx, pnn, nodemap, vnnmap);
3172 /* get the nodemap for all active remote nodes
3174 remote_nodemaps = talloc_array(mem_ctx, struct ctdb_node_map *, nodemap->num);
3175 if (remote_nodemaps == NULL) {
3176 DEBUG(DEBUG_ERR, (__location__ " failed to allocate remote nodemap array\n"));
3179 for(i=0; i<nodemap->num; i++) {
3180 remote_nodemaps[i] = NULL;
3182 if (get_remote_nodemaps(ctdb, mem_ctx, nodemap, remote_nodemaps) != 0) {
3183 DEBUG(DEBUG_ERR,(__location__ " Failed to read remote nodemaps\n"));
3187 /* verify that all other nodes have the same nodemap as we have
3189 for (j=0; j<nodemap->num; j++) {
3190 if (nodemap->nodes[j].flags & NODE_FLAGS_INACTIVE) {
3194 if (remote_nodemaps[j] == NULL) {
3195 DEBUG(DEBUG_ERR,(__location__ " Did not get a remote nodemap for node %d, restarting monitoring\n", j));
3196 ctdb_set_culprit(rec, j);
3201 /* if the nodes disagree on how many nodes there are
3202 then this is a good reason to try recovery
3204 if (remote_nodemaps[j]->num != nodemap->num) {
3205 DEBUG(DEBUG_ERR, (__location__ " Remote node:%u has different node count. %u vs %u of the local node\n",
3206 nodemap->nodes[j].pnn, remote_nodemaps[j]->num, nodemap->num));
3207 ctdb_set_culprit(rec, nodemap->nodes[j].pnn);
3208 do_recovery(rec, mem_ctx, pnn, nodemap, vnnmap);
3212 /* if the nodes disagree on which nodes exist and are
3213 active, then that is also a good reason to do recovery
3215 for (i=0;i<nodemap->num;i++) {
3216 if (remote_nodemaps[j]->nodes[i].pnn != nodemap->nodes[i].pnn) {
3217 DEBUG(DEBUG_ERR, (__location__ " Remote node:%u has different nodemap pnn for %d (%u vs %u).\n",
3218 nodemap->nodes[j].pnn, i,
3219 remote_nodemaps[j]->nodes[i].pnn, nodemap->nodes[i].pnn));
3220 ctdb_set_culprit(rec, nodemap->nodes[j].pnn);
3221 do_recovery(rec, mem_ctx, pnn, nodemap,
3227 /* verify the flags are consistent
3229 for (i=0; i<nodemap->num; i++) {
3230 if (nodemap->nodes[i].flags & NODE_FLAGS_DISCONNECTED) {
3234 if (nodemap->nodes[i].flags != remote_nodemaps[j]->nodes[i].flags) {
3235 DEBUG(DEBUG_ERR, (__location__ " Remote node:%u has different flags for node %u. It has 0x%02x vs our 0x%02x\n",
3236 nodemap->nodes[j].pnn,
3237 nodemap->nodes[i].pnn,
3238 remote_nodemaps[j]->nodes[i].flags,
3239 nodemap->nodes[j].flags));
3241 DEBUG(DEBUG_ERR,("Use flags 0x%02x from remote node %d for cluster update of its own flags\n", remote_nodemaps[j]->nodes[i].flags, j));
3242 update_flags_on_all_nodes(ctdb, nodemap, nodemap->nodes[i].pnn, remote_nodemaps[j]->nodes[i].flags);
3243 ctdb_set_culprit(rec, nodemap->nodes[j].pnn);
3244 do_recovery(rec, mem_ctx, pnn, nodemap,
3248 DEBUG(DEBUG_ERR,("Use flags 0x%02x from local recmaster node for cluster update of node %d flags\n", nodemap->nodes[i].flags, i));
3249 update_flags_on_all_nodes(ctdb, nodemap, nodemap->nodes[i].pnn, nodemap->nodes[i].flags);
3250 ctdb_set_culprit(rec, nodemap->nodes[j].pnn);
3251 do_recovery(rec, mem_ctx, pnn, nodemap,
3260 /* there better be the same number of lmasters in the vnn map
3261 as there are active nodes or we will have to do a recovery
3263 if (vnnmap->size != rec->num_active) {
3264 DEBUG(DEBUG_ERR, (__location__ " The vnnmap count is different from the number of active nodes. %u vs %u\n",
3265 vnnmap->size, rec->num_active));
3266 ctdb_set_culprit(rec, ctdb->pnn);
3267 do_recovery(rec, mem_ctx, pnn, nodemap, vnnmap);
3271 /* verify that all active nodes in the nodemap also exist in
3274 for (j=0; j<nodemap->num; j++) {
3275 if (nodemap->nodes[j].flags & NODE_FLAGS_INACTIVE) {
3278 if (nodemap->nodes[j].pnn == pnn) {
3282 for (i=0; i<vnnmap->size; i++) {
3283 if (vnnmap->map[i] == nodemap->nodes[j].pnn) {
3287 if (i == vnnmap->size) {
3288 DEBUG(DEBUG_ERR, (__location__ " Node %u is active in the nodemap but did not exist in the vnnmap\n",
3289 nodemap->nodes[j].pnn));
3290 ctdb_set_culprit(rec, nodemap->nodes[j].pnn);
3291 do_recovery(rec, mem_ctx, pnn, nodemap, vnnmap);
3297 /* verify that all other nodes have the same vnnmap
3298 and are from the same generation
3300 for (j=0; j<nodemap->num; j++) {
3301 if (nodemap->nodes[j].flags & NODE_FLAGS_INACTIVE) {
3304 if (nodemap->nodes[j].pnn == pnn) {
3308 ret = ctdb_ctrl_getvnnmap(ctdb, CONTROL_TIMEOUT(), nodemap->nodes[j].pnn,
3309 mem_ctx, &remote_vnnmap);
3311 DEBUG(DEBUG_ERR, (__location__ " Unable to get vnnmap from remote node %u\n",
3312 nodemap->nodes[j].pnn));
3316 /* verify the vnnmap generation is the same */
3317 if (vnnmap->generation != remote_vnnmap->generation) {
3318 DEBUG(DEBUG_ERR, (__location__ " Remote node %u has different generation of vnnmap. %u vs %u (ours)\n",
3319 nodemap->nodes[j].pnn, remote_vnnmap->generation, vnnmap->generation));
3320 ctdb_set_culprit(rec, nodemap->nodes[j].pnn);
3321 do_recovery(rec, mem_ctx, pnn, nodemap, vnnmap);
3325 /* verify the vnnmap size is the same */
3326 if (vnnmap->size != remote_vnnmap->size) {
3327 DEBUG(DEBUG_ERR, (__location__ " Remote node %u has different size of vnnmap. %u vs %u (ours)\n",
3328 nodemap->nodes[j].pnn, remote_vnnmap->size, vnnmap->size));
3329 ctdb_set_culprit(rec, nodemap->nodes[j].pnn);
3330 do_recovery(rec, mem_ctx, pnn, nodemap, vnnmap);
3334 /* verify the vnnmap is the same */
3335 for (i=0;i<vnnmap->size;i++) {
3336 if (remote_vnnmap->map[i] != vnnmap->map[i]) {
3337 DEBUG(DEBUG_ERR, (__location__ " Remote node %u has different vnnmap.\n",
3338 nodemap->nodes[j].pnn));
3339 ctdb_set_culprit(rec, nodemap->nodes[j].pnn);
3340 do_recovery(rec, mem_ctx, pnn, nodemap,
3347 /* we might need to change who has what IP assigned */
3348 if (rec->need_takeover_run) {
3349 uint32_t culprit = (uint32_t)-1;
3351 rec->need_takeover_run = false;
3353 /* update the list of public ips that a node can handle for
3356 ret = ctdb_reload_remote_public_ips(ctdb, nodemap, &culprit);
3358 DEBUG(DEBUG_ERR,("Failed to read public ips from remote node %d\n",
3360 ctdb_set_culprit(rec, culprit);
3361 do_recovery(rec, mem_ctx, pnn, nodemap, vnnmap);
3365 /* execute the "startrecovery" event script on all nodes */
3366 ret = run_startrecovery_eventscript(rec, nodemap);
3368 DEBUG(DEBUG_ERR, (__location__ " Unable to run the 'startrecovery' event on cluster\n"));
3369 ctdb_set_culprit(rec, ctdb->pnn);
3370 do_recovery(rec, mem_ctx, pnn, nodemap, vnnmap);
3374 ret = ctdb_takeover_run(ctdb, nodemap);
3376 DEBUG(DEBUG_ERR, (__location__ " Unable to setup public takeover addresses - starting recovery\n"));
3377 ctdb_set_culprit(rec, ctdb->pnn);
3378 do_recovery(rec, mem_ctx, pnn, nodemap, vnnmap);
3382 /* execute the "recovered" event script on all nodes */
3383 ret = run_recovered_eventscript(ctdb, nodemap, "monitor_cluster");
3385 // we cant check whether the event completed successfully
3386 // since this script WILL fail if the node is in recovery mode
3387 // and if that race happens, the code here would just cause a second
3388 // cascading recovery.
3390 DEBUG(DEBUG_ERR, (__location__ " Unable to run the 'recovered' event on cluster. Update of public ips failed.\n"));
3391 ctdb_set_culprit(rec, ctdb->pnn);
3392 do_recovery(rec, mem_ctx, pnn, nodemap, vnnmap);
3403 event handler for when the main ctdbd dies
3405 static void ctdb_recoverd_parent(struct event_context *ev, struct fd_event *fde,
3406 uint16_t flags, void *private_data)
3408 DEBUG(DEBUG_ALERT,("recovery daemon parent died - exiting\n"));
3413 called regularly to verify that the recovery daemon is still running
3415 static void ctdb_check_recd(struct event_context *ev, struct timed_event *te,
3416 struct timeval yt, void *p)
3418 struct ctdb_context *ctdb = talloc_get_type(p, struct ctdb_context);
3420 if (kill(ctdb->recoverd_pid, 0) != 0) {
3421 DEBUG(DEBUG_ERR,("Recovery daemon (pid:%d) is no longer running. Shutting down main daemon\n", (int)ctdb->recoverd_pid));
3423 ctdb_stop_recoverd(ctdb);
3424 ctdb_stop_keepalive(ctdb);
3425 ctdb_stop_monitoring(ctdb);
3426 ctdb_release_all_ips(ctdb);
3427 if (ctdb->methods != NULL) {
3428 ctdb->methods->shutdown(ctdb);
3430 ctdb_event_script(ctdb, CTDB_EVENT_SHUTDOWN);
3435 event_add_timed(ctdb->ev, ctdb,
3436 timeval_current_ofs(30, 0),
3437 ctdb_check_recd, ctdb);
3440 static void recd_sig_child_handler(struct event_context *ev,
3441 struct signal_event *se, int signum, int count,
3445 // struct ctdb_context *ctdb = talloc_get_type(private_data, struct ctdb_context);
3450 pid = waitpid(-1, &status, WNOHANG);
3452 if (errno != ECHILD) {
3453 DEBUG(DEBUG_ERR, (__location__ " waitpid() returned error. errno:%s(%d)\n", strerror(errno),errno));
3458 DEBUG(DEBUG_DEBUG, ("RECD SIGCHLD from %d\n", (int)pid));
3464 startup the recovery daemon as a child of the main ctdb daemon
3466 int ctdb_start_recoverd(struct ctdb_context *ctdb)
3469 struct signal_event *se;
3471 if (pipe(fd) != 0) {
3475 ctdb->ctdbd_pid = getpid();
3477 ctdb->recoverd_pid = fork();
3478 if (ctdb->recoverd_pid == -1) {
3482 if (ctdb->recoverd_pid != 0) {
3484 event_add_timed(ctdb->ev, ctdb,
3485 timeval_current_ofs(30, 0),
3486 ctdb_check_recd, ctdb);
3492 srandom(getpid() ^ time(NULL));
3494 if (switch_from_server_to_client(ctdb) != 0) {
3495 DEBUG(DEBUG_CRIT, (__location__ "ERROR: failed to switch recovery daemon into client mode. shutting down.\n"));
3499 DEBUG(DEBUG_DEBUG, (__location__ " Created PIPE FD:%d to recovery daemon\n", fd[0]));
3501 event_add_fd(ctdb->ev, ctdb, fd[0], EVENT_FD_READ|EVENT_FD_AUTOCLOSE,
3502 ctdb_recoverd_parent, &fd[0]);
3504 /* set up a handler to pick up sigchld */
3505 se = event_add_signal(ctdb->ev, ctdb,
3507 recd_sig_child_handler,
3510 DEBUG(DEBUG_CRIT,("Failed to set up signal handler for SIGCHLD in recovery daemon\n"));
3514 monitor_cluster(ctdb);
3516 DEBUG(DEBUG_ALERT,("ERROR: ctdb_recoverd finished!?\n"));
3521 shutdown the recovery daemon
3523 void ctdb_stop_recoverd(struct ctdb_context *ctdb)
3525 if (ctdb->recoverd_pid == 0) {
3529 DEBUG(DEBUG_NOTICE,("Shutting down recovery daemon\n"));
3530 kill(ctdb->recoverd_pid, SIGTERM);