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 length/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_recoverd *rec,
1232 struct ctdb_node_map *nodemap,
1238 if (ctdb->num_nodes != nodemap->num) {
1239 DEBUG(DEBUG_ERR, (__location__ " ctdb->num_nodes (%d) != nodemap->num (%d) invalid param\n",
1240 ctdb->num_nodes, nodemap->num));
1242 *culprit = ctdb->pnn;
1247 for (j=0; j<nodemap->num; j++) {
1248 /* release any existing data */
1249 if (ctdb->nodes[j]->known_public_ips) {
1250 talloc_free(ctdb->nodes[j]->known_public_ips);
1251 ctdb->nodes[j]->known_public_ips = NULL;
1253 if (ctdb->nodes[j]->available_public_ips) {
1254 talloc_free(ctdb->nodes[j]->available_public_ips);
1255 ctdb->nodes[j]->available_public_ips = NULL;
1258 if (nodemap->nodes[j].flags & NODE_FLAGS_INACTIVE) {
1262 /* grab a new shiny list of public ips from the node */
1263 ret = ctdb_ctrl_get_public_ips_flags(ctdb,
1265 ctdb->nodes[j]->pnn,
1268 &ctdb->nodes[j]->known_public_ips);
1270 DEBUG(DEBUG_ERR,("Failed to read known public ips from node : %u\n",
1271 ctdb->nodes[j]->pnn));
1273 *culprit = ctdb->nodes[j]->pnn;
1278 if (verify_remote_ip_allocation(ctdb, ctdb->nodes[j]->known_public_ips)) {
1279 DEBUG(DEBUG_ERR,("Node %d has inconsistent public ip allocation and needs update.\n", ctdb->nodes[j]->pnn));
1280 rec->need_takeover_run = true;
1283 /* grab a new shiny list of public ips from the node */
1284 ret = ctdb_ctrl_get_public_ips_flags(ctdb,
1286 ctdb->nodes[j]->pnn,
1288 CTDB_PUBLIC_IP_FLAGS_ONLY_AVAILABLE,
1289 &ctdb->nodes[j]->available_public_ips);
1291 DEBUG(DEBUG_ERR,("Failed to read available public ips from node : %u\n",
1292 ctdb->nodes[j]->pnn));
1294 *culprit = ctdb->nodes[j]->pnn;
1304 we are the recmaster, and recovery is needed - start a recovery run
1306 static int do_recovery(struct ctdb_recoverd *rec,
1307 TALLOC_CTX *mem_ctx, uint32_t pnn,
1308 struct ctdb_node_map *nodemap, struct ctdb_vnn_map *vnnmap)
1310 struct ctdb_context *ctdb = rec->ctdb;
1312 uint32_t generation;
1313 struct ctdb_dbid_map *dbmap;
1316 struct timeval start_time;
1317 uint32_t culprit = (uint32_t)-1;
1319 DEBUG(DEBUG_NOTICE, (__location__ " Starting do_recovery\n"));
1321 /* if recovery fails, force it again */
1322 rec->need_recovery = true;
1324 for (i=0; i<ctdb->num_nodes; i++) {
1325 struct ctdb_banning_state *ban_state;
1327 if (ctdb->nodes[i]->ban_state == NULL) {
1330 ban_state = (struct ctdb_banning_state *)ctdb->nodes[i]->ban_state;
1331 if (ban_state->count < 2*ctdb->num_nodes) {
1334 DEBUG(DEBUG_NOTICE,("Node %u has caused %u recoveries recently - banning it for %u seconds\n",
1335 ctdb->nodes[i]->pnn, ban_state->count,
1336 ctdb->tunable.recovery_ban_period));
1337 ctdb_ban_node(rec, ctdb->nodes[i]->pnn, ctdb->tunable.recovery_ban_period);
1338 ban_state->count = 0;
1342 if (ctdb->tunable.verify_recovery_lock != 0) {
1343 DEBUG(DEBUG_ERR,("Taking out recovery lock from recovery daemon\n"));
1344 start_time = timeval_current();
1345 if (!ctdb_recovery_lock(ctdb, true)) {
1346 ctdb_set_culprit(rec, pnn);
1347 DEBUG(DEBUG_ERR,("Unable to get recovery lock - aborting recovery\n"));
1350 ctdb_ctrl_report_recd_lock_latency(ctdb, CONTROL_TIMEOUT(), timeval_elapsed(&start_time));
1351 DEBUG(DEBUG_NOTICE,("Recovery lock taken successfully by recovery daemon\n"));
1354 DEBUG(DEBUG_NOTICE, (__location__ " Recovery initiated due to problem with node %u\n", rec->last_culprit_node));
1356 /* get a list of all databases */
1357 ret = ctdb_ctrl_getdbmap(ctdb, CONTROL_TIMEOUT(), pnn, mem_ctx, &dbmap);
1359 DEBUG(DEBUG_ERR, (__location__ " Unable to get dbids from node :%u\n", pnn));
1363 /* we do the db creation before we set the recovery mode, so the freeze happens
1364 on all databases we will be dealing with. */
1366 /* verify that we have all the databases any other node has */
1367 ret = create_missing_local_databases(ctdb, nodemap, pnn, &dbmap, mem_ctx);
1369 DEBUG(DEBUG_ERR, (__location__ " Unable to create missing local databases\n"));
1373 /* verify that all other nodes have all our databases */
1374 ret = create_missing_remote_databases(ctdb, nodemap, pnn, dbmap, mem_ctx);
1376 DEBUG(DEBUG_ERR, (__location__ " Unable to create missing remote databases\n"));
1379 DEBUG(DEBUG_NOTICE, (__location__ " Recovery - created remote databases\n"));
1381 /* update the database priority for all remote databases */
1382 ret = update_db_priority_on_remote_nodes(ctdb, nodemap, pnn, dbmap, mem_ctx);
1384 DEBUG(DEBUG_ERR, (__location__ " Unable to set db priority on remote nodes\n"));
1386 DEBUG(DEBUG_NOTICE, (__location__ " Recovery - updated db priority for all databases\n"));
1389 /* set recovery mode to active on all nodes */
1390 ret = set_recovery_mode(ctdb, rec, nodemap, CTDB_RECOVERY_ACTIVE);
1392 DEBUG(DEBUG_ERR, (__location__ " Unable to set recovery mode to active on cluster\n"));
1396 /* execute the "startrecovery" event script on all nodes */
1397 ret = run_startrecovery_eventscript(rec, nodemap);
1399 DEBUG(DEBUG_ERR, (__location__ " Unable to run the 'startrecovery' event on cluster\n"));
1404 update all nodes to have the same flags that we have
1406 for (i=0;i<nodemap->num;i++) {
1407 if (nodemap->nodes[i].flags & NODE_FLAGS_DISCONNECTED) {
1411 ret = update_flags_on_all_nodes(ctdb, nodemap, i, nodemap->nodes[i].flags);
1413 DEBUG(DEBUG_ERR, (__location__ " Unable to update flags on all nodes for node %d\n", i));
1418 DEBUG(DEBUG_NOTICE, (__location__ " Recovery - updated flags\n"));
1420 /* pick a new generation number */
1421 generation = new_generation();
1423 /* change the vnnmap on this node to use the new generation
1424 number but not on any other nodes.
1425 this guarantees that if we abort the recovery prematurely
1426 for some reason (a node stops responding?)
1427 that we can just return immediately and we will reenter
1428 recovery shortly again.
1429 I.e. we deliberately leave the cluster with an inconsistent
1430 generation id to allow us to abort recovery at any stage and
1431 just restart it from scratch.
1433 vnnmap->generation = generation;
1434 ret = ctdb_ctrl_setvnnmap(ctdb, CONTROL_TIMEOUT(), pnn, mem_ctx, vnnmap);
1436 DEBUG(DEBUG_ERR, (__location__ " Unable to set vnnmap for node %u\n", pnn));
1440 data.dptr = (void *)&generation;
1441 data.dsize = sizeof(uint32_t);
1443 nodes = list_of_active_nodes(ctdb, nodemap, mem_ctx, true);
1444 if (ctdb_client_async_control(ctdb, CTDB_CONTROL_TRANSACTION_START,
1446 CONTROL_TIMEOUT(), false, data,
1448 transaction_start_fail_callback,
1450 DEBUG(DEBUG_ERR, (__location__ " Unable to start transactions. Recovery failed.\n"));
1451 if (ctdb_client_async_control(ctdb, CTDB_CONTROL_TRANSACTION_CANCEL,
1453 CONTROL_TIMEOUT(), false, tdb_null,
1457 DEBUG(DEBUG_ERR,("Failed to cancel recovery transaction\n"));
1462 DEBUG(DEBUG_NOTICE,(__location__ " started transactions on all nodes\n"));
1464 for (i=0;i<dbmap->num;i++) {
1465 ret = recover_database(rec, mem_ctx,
1467 dbmap->dbs[i].persistent,
1468 pnn, nodemap, generation);
1470 DEBUG(DEBUG_ERR, (__location__ " Failed to recover database 0x%x\n", dbmap->dbs[i].dbid));
1475 DEBUG(DEBUG_NOTICE, (__location__ " Recovery - starting database commits\n"));
1477 /* commit all the changes */
1478 if (ctdb_client_async_control(ctdb, CTDB_CONTROL_TRANSACTION_COMMIT,
1480 CONTROL_TIMEOUT(), false, data,
1483 DEBUG(DEBUG_ERR, (__location__ " Unable to commit recovery changes. Recovery failed.\n"));
1487 DEBUG(DEBUG_NOTICE, (__location__ " Recovery - committed databases\n"));
1490 /* update the capabilities for all nodes */
1491 ret = update_capabilities(ctdb, nodemap);
1493 DEBUG(DEBUG_ERR, (__location__ " Unable to update node capabilities.\n"));
1497 /* build a new vnn map with all the currently active and
1499 generation = new_generation();
1500 vnnmap = talloc(mem_ctx, struct ctdb_vnn_map);
1501 CTDB_NO_MEMORY(ctdb, vnnmap);
1502 vnnmap->generation = generation;
1504 vnnmap->map = talloc_zero_array(vnnmap, uint32_t, vnnmap->size);
1505 CTDB_NO_MEMORY(ctdb, vnnmap->map);
1506 for (i=j=0;i<nodemap->num;i++) {
1507 if (nodemap->nodes[i].flags & NODE_FLAGS_INACTIVE) {
1510 if (!(ctdb->nodes[i]->capabilities & CTDB_CAP_LMASTER)) {
1511 /* this node can not be an lmaster */
1512 DEBUG(DEBUG_DEBUG, ("Node %d cant be a LMASTER, skipping it\n", i));
1517 vnnmap->map = talloc_realloc(vnnmap, vnnmap->map, uint32_t, vnnmap->size);
1518 CTDB_NO_MEMORY(ctdb, vnnmap->map);
1519 vnnmap->map[j++] = nodemap->nodes[i].pnn;
1522 if (vnnmap->size == 0) {
1523 DEBUG(DEBUG_NOTICE, ("No suitable lmasters found. Adding local node (recmaster) anyway.\n"));
1525 vnnmap->map = talloc_realloc(vnnmap, vnnmap->map, uint32_t, vnnmap->size);
1526 CTDB_NO_MEMORY(ctdb, vnnmap->map);
1527 vnnmap->map[0] = pnn;
1530 /* update to the new vnnmap on all nodes */
1531 ret = update_vnnmap_on_all_nodes(ctdb, nodemap, pnn, vnnmap, mem_ctx);
1533 DEBUG(DEBUG_ERR, (__location__ " Unable to update vnnmap on all nodes\n"));
1537 DEBUG(DEBUG_NOTICE, (__location__ " Recovery - updated vnnmap\n"));
1539 /* update recmaster to point to us for all nodes */
1540 ret = set_recovery_master(ctdb, nodemap, pnn);
1542 DEBUG(DEBUG_ERR, (__location__ " Unable to set recovery master\n"));
1546 DEBUG(DEBUG_NOTICE, (__location__ " Recovery - updated recmaster\n"));
1549 update all nodes to have the same flags that we have
1551 for (i=0;i<nodemap->num;i++) {
1552 if (nodemap->nodes[i].flags & NODE_FLAGS_DISCONNECTED) {
1556 ret = update_flags_on_all_nodes(ctdb, nodemap, i, nodemap->nodes[i].flags);
1558 DEBUG(DEBUG_ERR, (__location__ " Unable to update flags on all nodes for node %d\n", i));
1563 DEBUG(DEBUG_NOTICE, (__location__ " Recovery - updated flags\n"));
1565 /* disable recovery mode */
1566 ret = set_recovery_mode(ctdb, rec, nodemap, CTDB_RECOVERY_NORMAL);
1568 DEBUG(DEBUG_ERR, (__location__ " Unable to set recovery mode to normal on cluster\n"));
1572 DEBUG(DEBUG_NOTICE, (__location__ " Recovery - disabled recovery mode\n"));
1575 tell nodes to takeover their public IPs
1577 ret = ctdb_reload_remote_public_ips(ctdb, rec, nodemap, &culprit);
1579 DEBUG(DEBUG_ERR,("Failed to read public ips from remote node %d\n",
1583 rec->need_takeover_run = false;
1584 ret = ctdb_takeover_run(ctdb, nodemap);
1586 DEBUG(DEBUG_ERR, (__location__ " Unable to setup public takeover addresses\n"));
1589 DEBUG(DEBUG_NOTICE, (__location__ " Recovery - takeip finished\n"));
1591 /* execute the "recovered" event script on all nodes */
1592 ret = run_recovered_eventscript(ctdb, nodemap, "do_recovery");
1594 DEBUG(DEBUG_ERR, (__location__ " Unable to run the 'recovered' event on cluster. Recovery process failed.\n"));
1598 DEBUG(DEBUG_NOTICE, (__location__ " Recovery - finished the recovered event\n"));
1600 /* send a message to all clients telling them that the cluster
1601 has been reconfigured */
1602 ctdb_send_message(ctdb, CTDB_BROADCAST_CONNECTED, CTDB_SRVID_RECONFIGURE, tdb_null);
1604 DEBUG(DEBUG_NOTICE, (__location__ " Recovery complete\n"));
1606 rec->need_recovery = false;
1608 /* we managed to complete a full recovery, make sure to forgive
1609 any past sins by the nodes that could now participate in the
1612 DEBUG(DEBUG_ERR,("Resetting ban count to 0 for all nodes\n"));
1613 for (i=0;i<nodemap->num;i++) {
1614 struct ctdb_banning_state *ban_state;
1616 if (nodemap->nodes[i].flags & NODE_FLAGS_DISCONNECTED) {
1620 ban_state = (struct ctdb_banning_state *)ctdb->nodes[nodemap->nodes[i].pnn]->ban_state;
1621 if (ban_state == NULL) {
1625 ban_state->count = 0;
1629 /* We just finished a recovery successfully.
1630 We now wait for rerecovery_timeout before we allow
1631 another recovery to take place.
1633 DEBUG(DEBUG_NOTICE, (__location__ " New recoveries supressed for the rerecovery timeout\n"));
1634 ctdb_wait_timeout(ctdb, ctdb->tunable.rerecovery_timeout);
1635 DEBUG(DEBUG_NOTICE, (__location__ " Rerecovery timeout elapsed. Recovery reactivated.\n"));
1642 elections are won by first checking the number of connected nodes, then
1643 the priority time, then the pnn
1645 struct election_message {
1646 uint32_t num_connected;
1647 struct timeval priority_time;
1649 uint32_t node_flags;
1653 form this nodes election data
1655 static void ctdb_election_data(struct ctdb_recoverd *rec, struct election_message *em)
1658 struct ctdb_node_map *nodemap;
1659 struct ctdb_context *ctdb = rec->ctdb;
1663 em->pnn = rec->ctdb->pnn;
1664 em->priority_time = rec->priority_time;
1666 ret = ctdb_ctrl_getnodemap(ctdb, CONTROL_TIMEOUT(), CTDB_CURRENT_NODE, rec, &nodemap);
1668 DEBUG(DEBUG_ERR,(__location__ " unable to get election data\n"));
1672 rec->node_flags = nodemap->nodes[ctdb->pnn].flags;
1673 em->node_flags = rec->node_flags;
1675 for (i=0;i<nodemap->num;i++) {
1676 if (!(nodemap->nodes[i].flags & NODE_FLAGS_DISCONNECTED)) {
1677 em->num_connected++;
1681 /* we shouldnt try to win this election if we cant be a recmaster */
1682 if ((ctdb->capabilities & CTDB_CAP_RECMASTER) == 0) {
1683 em->num_connected = 0;
1684 em->priority_time = timeval_current();
1687 talloc_free(nodemap);
1691 see if the given election data wins
1693 static bool ctdb_election_win(struct ctdb_recoverd *rec, struct election_message *em)
1695 struct election_message myem;
1698 ctdb_election_data(rec, &myem);
1700 /* we cant win if we dont have the recmaster capability */
1701 if ((rec->ctdb->capabilities & CTDB_CAP_RECMASTER) == 0) {
1705 /* we cant win if we are banned */
1706 if (rec->node_flags & NODE_FLAGS_BANNED) {
1710 /* we cant win if we are stopped */
1711 if (rec->node_flags & NODE_FLAGS_STOPPED) {
1715 /* we will automatically win if the other node is banned */
1716 if (em->node_flags & NODE_FLAGS_BANNED) {
1720 /* we will automatically win if the other node is banned */
1721 if (em->node_flags & NODE_FLAGS_STOPPED) {
1725 /* try to use the most connected node */
1727 cmp = (int)myem.num_connected - (int)em->num_connected;
1730 /* then the longest running node */
1732 cmp = timeval_compare(&em->priority_time, &myem.priority_time);
1736 cmp = (int)myem.pnn - (int)em->pnn;
1743 send out an election request
1745 static int send_election_request(struct ctdb_recoverd *rec, uint32_t pnn, bool update_recmaster)
1748 TDB_DATA election_data;
1749 struct election_message emsg;
1751 struct ctdb_context *ctdb = rec->ctdb;
1753 srvid = CTDB_SRVID_RECOVERY;
1755 ctdb_election_data(rec, &emsg);
1757 election_data.dsize = sizeof(struct election_message);
1758 election_data.dptr = (unsigned char *)&emsg;
1761 /* send an election message to all active nodes */
1762 DEBUG(DEBUG_INFO,(__location__ " Send election request to all active nodes\n"));
1763 ctdb_send_message(ctdb, CTDB_BROADCAST_ALL, srvid, election_data);
1766 /* A new node that is already frozen has entered the cluster.
1767 The existing nodes are not frozen and dont need to be frozen
1768 until the election has ended and we start the actual recovery
1770 if (update_recmaster == true) {
1771 /* first we assume we will win the election and set
1772 recoverymaster to be ourself on the current node
1774 ret = ctdb_ctrl_setrecmaster(ctdb, CONTROL_TIMEOUT(), pnn, pnn);
1776 DEBUG(DEBUG_ERR, (__location__ " failed to send recmaster election request\n"));
1786 this function will unban all nodes in the cluster
1788 static void unban_all_nodes(struct ctdb_context *ctdb)
1791 struct ctdb_node_map *nodemap;
1792 TALLOC_CTX *tmp_ctx = talloc_new(ctdb);
1794 ret = ctdb_ctrl_getnodemap(ctdb, CONTROL_TIMEOUT(), CTDB_CURRENT_NODE, tmp_ctx, &nodemap);
1796 DEBUG(DEBUG_ERR,(__location__ " failed to get nodemap to unban all nodes\n"));
1800 for (i=0;i<nodemap->num;i++) {
1801 if ( (!(nodemap->nodes[i].flags & NODE_FLAGS_DISCONNECTED))
1802 && (nodemap->nodes[i].flags & NODE_FLAGS_BANNED) ) {
1803 ctdb_ctrl_modflags(ctdb, CONTROL_TIMEOUT(), nodemap->nodes[i].pnn, 0, NODE_FLAGS_BANNED);
1807 talloc_free(tmp_ctx);
1812 we think we are winning the election - send a broadcast election request
1814 static void election_send_request(struct event_context *ev, struct timed_event *te, struct timeval t, void *p)
1816 struct ctdb_recoverd *rec = talloc_get_type(p, struct ctdb_recoverd);
1819 ret = send_election_request(rec, ctdb_get_pnn(rec->ctdb), false);
1821 DEBUG(DEBUG_ERR,("Failed to send election request!\n"));
1824 talloc_free(rec->send_election_te);
1825 rec->send_election_te = NULL;
1829 handler for memory dumps
1831 static void mem_dump_handler(struct ctdb_context *ctdb, uint64_t srvid,
1832 TDB_DATA data, void *private_data)
1834 TALLOC_CTX *tmp_ctx = talloc_new(ctdb);
1837 struct rd_memdump_reply *rd;
1839 if (data.dsize != sizeof(struct rd_memdump_reply)) {
1840 DEBUG(DEBUG_ERR, (__location__ " Wrong size of return address.\n"));
1841 talloc_free(tmp_ctx);
1844 rd = (struct rd_memdump_reply *)data.dptr;
1846 dump = talloc_zero(tmp_ctx, TDB_DATA);
1848 DEBUG(DEBUG_ERR, (__location__ " Failed to allocate memory for memdump\n"));
1849 talloc_free(tmp_ctx);
1852 ret = ctdb_dump_memory(ctdb, dump);
1854 DEBUG(DEBUG_ERR, (__location__ " ctdb_dump_memory() failed\n"));
1855 talloc_free(tmp_ctx);
1859 DEBUG(DEBUG_ERR, ("recovery master memory dump\n"));
1861 ret = ctdb_send_message(ctdb, rd->pnn, rd->srvid, *dump);
1863 DEBUG(DEBUG_ERR,("Failed to send rd memdump reply message\n"));
1864 talloc_free(tmp_ctx);
1868 talloc_free(tmp_ctx);
1872 handler for reload_nodes
1874 static void reload_nodes_handler(struct ctdb_context *ctdb, uint64_t srvid,
1875 TDB_DATA data, void *private_data)
1877 struct ctdb_recoverd *rec = talloc_get_type(private_data, struct ctdb_recoverd);
1879 DEBUG(DEBUG_ERR, (__location__ " Reload nodes file from recovery daemon\n"));
1881 reload_nodes_file(rec->ctdb);
1885 static void reenable_ip_check(struct event_context *ev, struct timed_event *te,
1886 struct timeval yt, void *p)
1888 struct ctdb_recoverd *rec = talloc_get_type(p, struct ctdb_recoverd);
1890 talloc_free(rec->ip_check_disable_ctx);
1891 rec->ip_check_disable_ctx = NULL;
1895 static void recd_update_ip_handler(struct ctdb_context *ctdb, uint64_t srvid,
1896 TDB_DATA data, void *private_data)
1898 struct ctdb_recoverd *rec = talloc_get_type(private_data, struct ctdb_recoverd);
1899 struct ctdb_public_ip *ip;
1901 if (rec->recmaster != rec->ctdb->pnn) {
1902 DEBUG(DEBUG_INFO,("Not recmaster, ignore update ip message\n"));
1906 if (data.dsize != sizeof(struct ctdb_public_ip)) {
1907 DEBUG(DEBUG_ERR,(__location__ " Incorrect size of recd update ip message. Was %zd but expected %zd bytes\n", data.dsize, sizeof(struct ctdb_public_ip)));
1911 ip = (struct ctdb_public_ip *)data.dptr;
1913 update_ip_assignment_tree(rec->ctdb, ip);
1917 static void disable_ip_check_handler(struct ctdb_context *ctdb, uint64_t srvid,
1918 TDB_DATA data, void *private_data)
1920 struct ctdb_recoverd *rec = talloc_get_type(private_data, struct ctdb_recoverd);
1923 if (rec->ip_check_disable_ctx != NULL) {
1924 talloc_free(rec->ip_check_disable_ctx);
1925 rec->ip_check_disable_ctx = NULL;
1928 if (data.dsize != sizeof(uint32_t)) {
1929 DEBUG(DEBUG_ERR,(__location__ " Wrong size for data :%lu "
1930 "expexting %lu\n", (long unsigned)data.dsize,
1931 (long unsigned)sizeof(uint32_t)));
1934 if (data.dptr == NULL) {
1935 DEBUG(DEBUG_ERR,(__location__ " No data recaived\n"));
1939 timeout = *((uint32_t *)data.dptr);
1940 DEBUG(DEBUG_NOTICE,("Disabling ip check for %u seconds\n", timeout));
1942 rec->ip_check_disable_ctx = talloc_new(rec);
1943 CTDB_NO_MEMORY_VOID(ctdb, rec->ip_check_disable_ctx);
1945 event_add_timed(ctdb->ev, rec->ip_check_disable_ctx, timeval_current_ofs(timeout, 0), reenable_ip_check, rec);
1950 handler for ip reallocate, just add it to the list of callers and
1951 handle this later in the monitor_cluster loop so we do not recurse
1952 with other callers to takeover_run()
1954 static void ip_reallocate_handler(struct ctdb_context *ctdb, uint64_t srvid,
1955 TDB_DATA data, void *private_data)
1957 struct ctdb_recoverd *rec = talloc_get_type(private_data, struct ctdb_recoverd);
1958 struct ip_reallocate_list *caller;
1960 if (data.dsize != sizeof(struct rd_memdump_reply)) {
1961 DEBUG(DEBUG_ERR, (__location__ " Wrong size of return address.\n"));
1965 if (rec->ip_reallocate_ctx == NULL) {
1966 rec->ip_reallocate_ctx = talloc_new(rec);
1967 CTDB_NO_MEMORY_FATAL(ctdb, rec->ip_reallocate_ctx);
1970 caller = talloc(rec->ip_reallocate_ctx, struct ip_reallocate_list);
1971 CTDB_NO_MEMORY_FATAL(ctdb, caller);
1973 caller->rd = (struct rd_memdump_reply *)talloc_steal(caller, data.dptr);
1974 caller->next = rec->reallocate_callers;
1975 rec->reallocate_callers = caller;
1980 static void process_ipreallocate_requests(struct ctdb_context *ctdb, struct ctdb_recoverd *rec)
1982 TALLOC_CTX *tmp_ctx = talloc_new(ctdb);
1985 struct ip_reallocate_list *callers;
1988 DEBUG(DEBUG_INFO, ("recovery master forced ip reallocation\n"));
1990 /* update the list of public ips that a node can handle for
1993 ret = ctdb_reload_remote_public_ips(ctdb, rec, rec->nodemap, &culprit);
1995 DEBUG(DEBUG_ERR,("Failed to read public ips from remote node %d\n",
1997 rec->need_takeover_run = true;
2000 ret = ctdb_takeover_run(ctdb, rec->nodemap);
2002 DEBUG(DEBUG_ERR,("Failed to read public ips from remote node %d\n",
2004 rec->need_takeover_run = true;
2008 result.dsize = sizeof(int32_t);
2009 result.dptr = (uint8_t *)&ret;
2011 for (callers=rec->reallocate_callers; callers; callers=callers->next) {
2013 /* Someone that sent srvid==0 does not want a reply */
2014 if (callers->rd->srvid == 0) {
2017 DEBUG(DEBUG_INFO,("Sending ip reallocate reply message to "
2018 "%u:%llu\n", (unsigned)callers->rd->pnn,
2019 (unsigned long long)callers->rd->srvid));
2020 ret = ctdb_send_message(ctdb, callers->rd->pnn, callers->rd->srvid, result);
2022 DEBUG(DEBUG_ERR,("Failed to send ip reallocate reply "
2023 "message to %u:%llu\n",
2024 (unsigned)callers->rd->pnn,
2025 (unsigned long long)callers->rd->srvid));
2029 talloc_free(tmp_ctx);
2030 talloc_free(rec->ip_reallocate_ctx);
2031 rec->ip_reallocate_ctx = NULL;
2032 rec->reallocate_callers = NULL;
2038 handler for recovery master elections
2040 static void election_handler(struct ctdb_context *ctdb, uint64_t srvid,
2041 TDB_DATA data, void *private_data)
2043 struct ctdb_recoverd *rec = talloc_get_type(private_data, struct ctdb_recoverd);
2045 struct election_message *em = (struct election_message *)data.dptr;
2046 TALLOC_CTX *mem_ctx;
2048 /* we got an election packet - update the timeout for the election */
2049 talloc_free(rec->election_timeout);
2050 rec->election_timeout = event_add_timed(ctdb->ev, ctdb,
2051 timeval_current_ofs(ctdb->tunable.election_timeout, 0),
2052 ctdb_election_timeout, rec);
2054 mem_ctx = talloc_new(ctdb);
2056 /* someone called an election. check their election data
2057 and if we disagree and we would rather be the elected node,
2058 send a new election message to all other nodes
2060 if (ctdb_election_win(rec, em)) {
2061 if (!rec->send_election_te) {
2062 rec->send_election_te = event_add_timed(ctdb->ev, rec,
2063 timeval_current_ofs(0, 500000),
2064 election_send_request, rec);
2066 talloc_free(mem_ctx);
2067 /*unban_all_nodes(ctdb);*/
2072 talloc_free(rec->send_election_te);
2073 rec->send_election_te = NULL;
2075 if (ctdb->tunable.verify_recovery_lock != 0) {
2076 /* release the recmaster lock */
2077 if (em->pnn != ctdb->pnn &&
2078 ctdb->recovery_lock_fd != -1) {
2079 close(ctdb->recovery_lock_fd);
2080 ctdb->recovery_lock_fd = -1;
2081 unban_all_nodes(ctdb);
2085 /* ok, let that guy become recmaster then */
2086 ret = ctdb_ctrl_setrecmaster(ctdb, CONTROL_TIMEOUT(), ctdb_get_pnn(ctdb), em->pnn);
2088 DEBUG(DEBUG_ERR, (__location__ " failed to send recmaster election request"));
2089 talloc_free(mem_ctx);
2093 talloc_free(mem_ctx);
2099 force the start of the election process
2101 static void force_election(struct ctdb_recoverd *rec, uint32_t pnn,
2102 struct ctdb_node_map *nodemap)
2105 struct ctdb_context *ctdb = rec->ctdb;
2107 DEBUG(DEBUG_INFO,(__location__ " Force an election\n"));
2109 /* set all nodes to recovery mode to stop all internode traffic */
2110 ret = set_recovery_mode(ctdb, rec, nodemap, CTDB_RECOVERY_ACTIVE);
2112 DEBUG(DEBUG_ERR, (__location__ " Unable to set recovery mode to active on cluster\n"));
2116 talloc_free(rec->election_timeout);
2117 rec->election_timeout = event_add_timed(ctdb->ev, ctdb,
2118 timeval_current_ofs(ctdb->tunable.election_timeout, 0),
2119 ctdb_election_timeout, rec);
2121 ret = send_election_request(rec, pnn, true);
2123 DEBUG(DEBUG_ERR, (__location__ " failed to initiate recmaster election"));
2127 /* wait for a few seconds to collect all responses */
2128 ctdb_wait_election(rec);
2134 handler for when a node changes its flags
2136 static void monitor_handler(struct ctdb_context *ctdb, uint64_t srvid,
2137 TDB_DATA data, void *private_data)
2140 struct ctdb_node_flag_change *c = (struct ctdb_node_flag_change *)data.dptr;
2141 struct ctdb_node_map *nodemap=NULL;
2142 TALLOC_CTX *tmp_ctx;
2143 uint32_t changed_flags;
2145 struct ctdb_recoverd *rec = talloc_get_type(private_data, struct ctdb_recoverd);
2146 int disabled_flag_changed;
2148 if (data.dsize != sizeof(*c)) {
2149 DEBUG(DEBUG_ERR,(__location__ "Invalid data in ctdb_node_flag_change\n"));
2153 tmp_ctx = talloc_new(ctdb);
2154 CTDB_NO_MEMORY_VOID(ctdb, tmp_ctx);
2156 ret = ctdb_ctrl_getnodemap(ctdb, CONTROL_TIMEOUT(), CTDB_CURRENT_NODE, tmp_ctx, &nodemap);
2158 DEBUG(DEBUG_ERR,(__location__ "ctdb_ctrl_getnodemap failed in monitor_handler\n"));
2159 talloc_free(tmp_ctx);
2164 for (i=0;i<nodemap->num;i++) {
2165 if (nodemap->nodes[i].pnn == c->pnn) break;
2168 if (i == nodemap->num) {
2169 DEBUG(DEBUG_CRIT,(__location__ "Flag change for non-existant node %u\n", c->pnn));
2170 talloc_free(tmp_ctx);
2174 changed_flags = c->old_flags ^ c->new_flags;
2176 if (nodemap->nodes[i].flags != c->new_flags) {
2177 DEBUG(DEBUG_NOTICE,("Node %u has changed flags - now 0x%x was 0x%x\n", c->pnn, c->new_flags, c->old_flags));
2180 disabled_flag_changed = (nodemap->nodes[i].flags ^ c->new_flags) & NODE_FLAGS_DISABLED;
2182 nodemap->nodes[i].flags = c->new_flags;
2184 ret = ctdb_ctrl_getrecmaster(ctdb, tmp_ctx, CONTROL_TIMEOUT(),
2185 CTDB_CURRENT_NODE, &ctdb->recovery_master);
2188 ret = ctdb_ctrl_getrecmode(ctdb, tmp_ctx, CONTROL_TIMEOUT(),
2189 CTDB_CURRENT_NODE, &ctdb->recovery_mode);
2193 ctdb->recovery_master == ctdb->pnn &&
2194 ctdb->recovery_mode == CTDB_RECOVERY_NORMAL) {
2195 /* Only do the takeover run if the perm disabled or unhealthy
2196 flags changed since these will cause an ip failover but not
2198 If the node became disconnected or banned this will also
2199 lead to an ip address failover but that is handled
2202 if (disabled_flag_changed) {
2203 rec->need_takeover_run = true;
2207 talloc_free(tmp_ctx);
2211 handler for when we need to push out flag changes ot all other nodes
2213 static void push_flags_handler(struct ctdb_context *ctdb, uint64_t srvid,
2214 TDB_DATA data, void *private_data)
2217 struct ctdb_node_flag_change *c = (struct ctdb_node_flag_change *)data.dptr;
2218 struct ctdb_node_map *nodemap=NULL;
2219 TALLOC_CTX *tmp_ctx = talloc_new(ctdb);
2223 /* find the recovery master */
2224 ret = ctdb_ctrl_getrecmaster(ctdb, tmp_ctx, CONTROL_TIMEOUT(), CTDB_CURRENT_NODE, &recmaster);
2226 DEBUG(DEBUG_ERR, (__location__ " Unable to get recmaster from local node\n"));
2227 talloc_free(tmp_ctx);
2231 /* read the node flags from the recmaster */
2232 ret = ctdb_ctrl_getnodemap(ctdb, CONTROL_TIMEOUT(), recmaster, tmp_ctx, &nodemap);
2234 DEBUG(DEBUG_ERR, (__location__ " Unable to get nodemap from node %u\n", c->pnn));
2235 talloc_free(tmp_ctx);
2238 if (c->pnn >= nodemap->num) {
2239 DEBUG(DEBUG_ERR,(__location__ " Nodemap from recmaster does not contain node %d\n", c->pnn));
2240 talloc_free(tmp_ctx);
2244 /* send the flags update to all connected nodes */
2245 nodes = list_of_connected_nodes(ctdb, nodemap, tmp_ctx, true);
2247 if (ctdb_client_async_control(ctdb, CTDB_CONTROL_MODIFY_FLAGS,
2248 nodes, 0, CONTROL_TIMEOUT(),
2252 DEBUG(DEBUG_ERR, (__location__ " ctdb_control to modify node flags failed\n"));
2254 talloc_free(tmp_ctx);
2258 talloc_free(tmp_ctx);
2262 struct verify_recmode_normal_data {
2264 enum monitor_result status;
2267 static void verify_recmode_normal_callback(struct ctdb_client_control_state *state)
2269 struct verify_recmode_normal_data *rmdata = talloc_get_type(state->async.private_data, struct verify_recmode_normal_data);
2272 /* one more node has responded with recmode data*/
2275 /* if we failed to get the recmode, then return an error and let
2276 the main loop try again.
2278 if (state->state != CTDB_CONTROL_DONE) {
2279 if (rmdata->status == MONITOR_OK) {
2280 rmdata->status = MONITOR_FAILED;
2285 /* if we got a response, then the recmode will be stored in the
2288 if (state->status != CTDB_RECOVERY_NORMAL) {
2289 DEBUG(DEBUG_NOTICE, (__location__ " Node:%u was in recovery mode. Restart recovery process\n", state->c->hdr.destnode));
2290 rmdata->status = MONITOR_RECOVERY_NEEDED;
2297 /* verify that all nodes are in normal recovery mode */
2298 static enum monitor_result verify_recmode(struct ctdb_context *ctdb, struct ctdb_node_map *nodemap)
2300 struct verify_recmode_normal_data *rmdata;
2301 TALLOC_CTX *mem_ctx = talloc_new(ctdb);
2302 struct ctdb_client_control_state *state;
2303 enum monitor_result status;
2306 rmdata = talloc(mem_ctx, struct verify_recmode_normal_data);
2307 CTDB_NO_MEMORY_FATAL(ctdb, rmdata);
2309 rmdata->status = MONITOR_OK;
2311 /* loop over all active nodes and send an async getrecmode call to
2313 for (j=0; j<nodemap->num; j++) {
2314 if (nodemap->nodes[j].flags & NODE_FLAGS_INACTIVE) {
2317 state = ctdb_ctrl_getrecmode_send(ctdb, mem_ctx,
2319 nodemap->nodes[j].pnn);
2320 if (state == NULL) {
2321 /* we failed to send the control, treat this as
2322 an error and try again next iteration
2324 DEBUG(DEBUG_ERR,("Failed to call ctdb_ctrl_getrecmode_send during monitoring\n"));
2325 talloc_free(mem_ctx);
2326 return MONITOR_FAILED;
2329 /* set up the callback functions */
2330 state->async.fn = verify_recmode_normal_callback;
2331 state->async.private_data = rmdata;
2333 /* one more control to wait for to complete */
2338 /* now wait for up to the maximum number of seconds allowed
2339 or until all nodes we expect a response from has replied
2341 while (rmdata->count > 0) {
2342 event_loop_once(ctdb->ev);
2345 status = rmdata->status;
2346 talloc_free(mem_ctx);
2351 struct verify_recmaster_data {
2352 struct ctdb_recoverd *rec;
2355 enum monitor_result status;
2358 static void verify_recmaster_callback(struct ctdb_client_control_state *state)
2360 struct verify_recmaster_data *rmdata = talloc_get_type(state->async.private_data, struct verify_recmaster_data);
2363 /* one more node has responded with recmaster data*/
2366 /* if we failed to get the recmaster, then return an error and let
2367 the main loop try again.
2369 if (state->state != CTDB_CONTROL_DONE) {
2370 if (rmdata->status == MONITOR_OK) {
2371 rmdata->status = MONITOR_FAILED;
2376 /* if we got a response, then the recmaster will be stored in the
2379 if (state->status != rmdata->pnn) {
2380 DEBUG(DEBUG_ERR,("Node %d does not agree we are the recmaster. Need a new recmaster election\n", state->c->hdr.destnode));
2381 ctdb_set_culprit(rmdata->rec, state->c->hdr.destnode);
2382 rmdata->status = MONITOR_ELECTION_NEEDED;
2389 /* verify that all nodes agree that we are the recmaster */
2390 static enum monitor_result verify_recmaster(struct ctdb_recoverd *rec, struct ctdb_node_map *nodemap, uint32_t pnn)
2392 struct ctdb_context *ctdb = rec->ctdb;
2393 struct verify_recmaster_data *rmdata;
2394 TALLOC_CTX *mem_ctx = talloc_new(ctdb);
2395 struct ctdb_client_control_state *state;
2396 enum monitor_result status;
2399 rmdata = talloc(mem_ctx, struct verify_recmaster_data);
2400 CTDB_NO_MEMORY_FATAL(ctdb, rmdata);
2404 rmdata->status = MONITOR_OK;
2406 /* loop over all active nodes and send an async getrecmaster call to
2408 for (j=0; j<nodemap->num; j++) {
2409 if (nodemap->nodes[j].flags & NODE_FLAGS_INACTIVE) {
2412 state = ctdb_ctrl_getrecmaster_send(ctdb, mem_ctx,
2414 nodemap->nodes[j].pnn);
2415 if (state == NULL) {
2416 /* we failed to send the control, treat this as
2417 an error and try again next iteration
2419 DEBUG(DEBUG_ERR,("Failed to call ctdb_ctrl_getrecmaster_send during monitoring\n"));
2420 talloc_free(mem_ctx);
2421 return MONITOR_FAILED;
2424 /* set up the callback functions */
2425 state->async.fn = verify_recmaster_callback;
2426 state->async.private_data = rmdata;
2428 /* one more control to wait for to complete */
2433 /* now wait for up to the maximum number of seconds allowed
2434 or until all nodes we expect a response from has replied
2436 while (rmdata->count > 0) {
2437 event_loop_once(ctdb->ev);
2440 status = rmdata->status;
2441 talloc_free(mem_ctx);
2446 /* called to check that the local allocation of public ip addresses is ok.
2448 static int verify_local_ip_allocation(struct ctdb_context *ctdb, struct ctdb_recoverd *rec, uint32_t pnn)
2450 TALLOC_CTX *mem_ctx = talloc_new(NULL);
2451 struct ctdb_control_get_ifaces *ifaces = NULL;
2452 struct ctdb_all_public_ips *ips = NULL;
2453 struct ctdb_uptime *uptime1 = NULL;
2454 struct ctdb_uptime *uptime2 = NULL;
2456 bool need_iface_check = false;
2457 bool need_takeover_run = false;
2459 ret = ctdb_ctrl_uptime(ctdb, mem_ctx, CONTROL_TIMEOUT(),
2460 CTDB_CURRENT_NODE, &uptime1);
2462 DEBUG(DEBUG_ERR, ("Unable to get uptime from local node %u\n", pnn));
2463 talloc_free(mem_ctx);
2468 /* read the interfaces from the local node */
2469 ret = ctdb_ctrl_get_ifaces(ctdb, CONTROL_TIMEOUT(), CTDB_CURRENT_NODE, mem_ctx, &ifaces);
2471 DEBUG(DEBUG_ERR, ("Unable to get interfaces from local node %u\n", pnn));
2472 talloc_free(mem_ctx);
2477 need_iface_check = true;
2478 } else if (rec->ifaces->num != ifaces->num) {
2479 need_iface_check = true;
2480 } else if (memcmp(rec->ifaces, ifaces, talloc_get_size(ifaces)) != 0) {
2481 need_iface_check = true;
2484 if (need_iface_check) {
2485 DEBUG(DEBUG_NOTICE, ("The interfaces status has changed on "
2486 "local node %u - force takeover run\n",
2488 need_takeover_run = true;
2491 /* read the ip allocation from the local node */
2492 ret = ctdb_ctrl_get_public_ips(ctdb, CONTROL_TIMEOUT(), CTDB_CURRENT_NODE, mem_ctx, &ips);
2494 DEBUG(DEBUG_ERR, ("Unable to get public ips from local node %u\n", pnn));
2495 talloc_free(mem_ctx);
2499 ret = ctdb_ctrl_uptime(ctdb, mem_ctx, CONTROL_TIMEOUT(),
2500 CTDB_CURRENT_NODE, &uptime2);
2502 DEBUG(DEBUG_ERR, ("Unable to get uptime from local node %u\n", pnn));
2503 talloc_free(mem_ctx);
2507 /* skip the check if the startrecovery time has changed */
2508 if (timeval_compare(&uptime1->last_recovery_started,
2509 &uptime2->last_recovery_started) != 0) {
2510 DEBUG(DEBUG_NOTICE, (__location__ " last recovery time changed while we read the public ip list. skipping public ip address check\n"));
2511 talloc_free(mem_ctx);
2515 /* skip the check if the endrecovery time has changed */
2516 if (timeval_compare(&uptime1->last_recovery_finished,
2517 &uptime2->last_recovery_finished) != 0) {
2518 DEBUG(DEBUG_NOTICE, (__location__ " last recovery time changed while we read the public ip list. skipping public ip address check\n"));
2519 talloc_free(mem_ctx);
2523 /* skip the check if we have started but not finished recovery */
2524 if (timeval_compare(&uptime1->last_recovery_finished,
2525 &uptime1->last_recovery_started) != 1) {
2526 DEBUG(DEBUG_NOTICE, (__location__ " in the middle of recovery or ip reallocation. skipping public ip address check\n"));
2527 talloc_free(mem_ctx);
2532 talloc_free(rec->ifaces);
2533 rec->ifaces = talloc_steal(rec, ifaces);
2535 /* verify that we have the ip addresses we should have
2536 and we dont have ones we shouldnt have.
2537 if we find an inconsistency we set recmode to
2538 active on the local node and wait for the recmaster
2539 to do a full blown recovery
2541 for (j=0; j<ips->num; j++) {
2542 if (ips->ips[j].pnn == pnn) {
2543 if (!ctdb_sys_have_ip(&ips->ips[j].addr)) {
2544 DEBUG(DEBUG_CRIT,("Public address '%s' is missing and we should serve this ip\n",
2545 ctdb_addr_to_str(&ips->ips[j].addr)));
2546 need_takeover_run = true;
2549 if (ctdb_sys_have_ip(&ips->ips[j].addr)) {
2550 DEBUG(DEBUG_CRIT,("We are still serving a public address '%s' that we should not be serving.\n",
2551 ctdb_addr_to_str(&ips->ips[j].addr)));
2552 need_takeover_run = true;
2557 if (need_takeover_run) {
2558 struct takeover_run_reply rd;
2561 DEBUG(DEBUG_CRIT,("Trigger takeoverrun\n"));
2565 data.dptr = (uint8_t *)&rd;
2566 data.dsize = sizeof(rd);
2568 ret = ctdb_send_message(ctdb, rec->recmaster, CTDB_SRVID_TAKEOVER_RUN, data);
2570 DEBUG(DEBUG_ERR,(__location__ " Failed to send ipreallocate to recmaster :%d\n", (int)rec->recmaster));
2573 talloc_free(mem_ctx);
2578 static void async_getnodemap_callback(struct ctdb_context *ctdb, uint32_t node_pnn, int32_t res, TDB_DATA outdata, void *callback_data)
2580 struct ctdb_node_map **remote_nodemaps = callback_data;
2582 if (node_pnn >= ctdb->num_nodes) {
2583 DEBUG(DEBUG_ERR,(__location__ " pnn from invalid node\n"));
2587 remote_nodemaps[node_pnn] = (struct ctdb_node_map *)talloc_steal(remote_nodemaps, outdata.dptr);
2591 static int get_remote_nodemaps(struct ctdb_context *ctdb, TALLOC_CTX *mem_ctx,
2592 struct ctdb_node_map *nodemap,
2593 struct ctdb_node_map **remote_nodemaps)
2597 nodes = list_of_active_nodes(ctdb, nodemap, mem_ctx, true);
2598 if (ctdb_client_async_control(ctdb, CTDB_CONTROL_GET_NODEMAP,
2600 CONTROL_TIMEOUT(), false, tdb_null,
2601 async_getnodemap_callback,
2603 remote_nodemaps) != 0) {
2604 DEBUG(DEBUG_ERR, (__location__ " Unable to pull all remote nodemaps\n"));
2612 enum reclock_child_status { RECLOCK_CHECKING, RECLOCK_OK, RECLOCK_FAILED, RECLOCK_TIMEOUT};
2613 struct ctdb_check_reclock_state {
2614 struct ctdb_context *ctdb;
2615 struct timeval start_time;
2618 struct timed_event *te;
2619 struct fd_event *fde;
2620 enum reclock_child_status status;
2623 /* when we free the reclock state we must kill any child process.
2625 static int check_reclock_destructor(struct ctdb_check_reclock_state *state)
2627 struct ctdb_context *ctdb = state->ctdb;
2629 ctdb_ctrl_report_recd_lock_latency(ctdb, CONTROL_TIMEOUT(), timeval_elapsed(&state->start_time));
2631 if (state->fd[0] != -1) {
2632 close(state->fd[0]);
2635 if (state->fd[1] != -1) {
2636 close(state->fd[1]);
2639 kill(state->child, SIGKILL);
2644 called if our check_reclock child times out. this would happen if
2645 i/o to the reclock file blocks.
2647 static void ctdb_check_reclock_timeout(struct event_context *ev, struct timed_event *te,
2648 struct timeval t, void *private_data)
2650 struct ctdb_check_reclock_state *state = talloc_get_type(private_data,
2651 struct ctdb_check_reclock_state);
2653 DEBUG(DEBUG_ERR,(__location__ " check_reclock child process hung/timedout CFS slow to grant locks?\n"));
2654 state->status = RECLOCK_TIMEOUT;
2657 /* this is called when the child process has completed checking the reclock
2658 file and has written data back to us through the pipe.
2660 static void reclock_child_handler(struct event_context *ev, struct fd_event *fde,
2661 uint16_t flags, void *private_data)
2663 struct ctdb_check_reclock_state *state= talloc_get_type(private_data,
2664 struct ctdb_check_reclock_state);
2668 /* we got a response from our child process so we can abort the
2671 talloc_free(state->te);
2674 ret = read(state->fd[0], &c, 1);
2675 if (ret != 1 || c != RECLOCK_OK) {
2676 DEBUG(DEBUG_ERR,(__location__ " reclock child process returned error %d\n", c));
2677 state->status = RECLOCK_FAILED;
2682 state->status = RECLOCK_OK;
2686 static int check_recovery_lock(struct ctdb_context *ctdb)
2689 struct ctdb_check_reclock_state *state;
2690 pid_t parent = getpid();
2692 if (ctdb->recovery_lock_fd == -1) {
2693 DEBUG(DEBUG_CRIT,("recovery master doesn't have the recovery lock\n"));
2697 state = talloc(ctdb, struct ctdb_check_reclock_state);
2698 CTDB_NO_MEMORY(ctdb, state);
2701 state->start_time = timeval_current();
2702 state->status = RECLOCK_CHECKING;
2706 ret = pipe(state->fd);
2709 DEBUG(DEBUG_CRIT,(__location__ " Failed to open pipe for check_reclock child\n"));
2713 state->child = fork();
2714 if (state->child == (pid_t)-1) {
2715 DEBUG(DEBUG_CRIT,(__location__ " fork() failed in check_reclock child\n"));
2716 close(state->fd[0]);
2718 close(state->fd[1]);
2724 if (state->child == 0) {
2725 char cc = RECLOCK_OK;
2726 close(state->fd[0]);
2729 if (pread(ctdb->recovery_lock_fd, &cc, 1, 0) == -1) {
2730 DEBUG(DEBUG_CRIT,("failed read from recovery_lock_fd - %s\n", strerror(errno)));
2731 cc = RECLOCK_FAILED;
2734 write(state->fd[1], &cc, 1);
2735 /* make sure we die when our parent dies */
2736 while (kill(parent, 0) == 0 || errno != ESRCH) {
2738 write(state->fd[1], &cc, 1);
2742 close(state->fd[1]);
2744 set_close_on_exec(state->fd[0]);
2746 DEBUG(DEBUG_DEBUG, (__location__ " Created PIPE FD:%d for check_recovery_lock\n", state->fd[0]));
2748 talloc_set_destructor(state, check_reclock_destructor);
2750 state->te = event_add_timed(ctdb->ev, state, timeval_current_ofs(15, 0),
2751 ctdb_check_reclock_timeout, state);
2752 if (state->te == NULL) {
2753 DEBUG(DEBUG_CRIT,(__location__ " Failed to create a timed event for reclock child\n"));
2758 state->fde = event_add_fd(ctdb->ev, state, state->fd[0],
2759 EVENT_FD_READ|EVENT_FD_AUTOCLOSE,
2760 reclock_child_handler,
2763 if (state->fde == NULL) {
2764 DEBUG(DEBUG_CRIT,(__location__ " Failed to create an fd event for reclock child\n"));
2769 while (state->status == RECLOCK_CHECKING) {
2770 event_loop_once(ctdb->ev);
2773 if (state->status == RECLOCK_FAILED) {
2774 DEBUG(DEBUG_ERR,(__location__ " reclock child failed when checking file\n"));
2775 close(ctdb->recovery_lock_fd);
2776 ctdb->recovery_lock_fd = -1;
2785 static int update_recovery_lock_file(struct ctdb_context *ctdb)
2787 TALLOC_CTX *tmp_ctx = talloc_new(NULL);
2788 const char *reclockfile;
2790 if (ctdb_ctrl_getreclock(ctdb, CONTROL_TIMEOUT(), CTDB_CURRENT_NODE, tmp_ctx, &reclockfile) != 0) {
2791 DEBUG(DEBUG_ERR,("Failed to read reclock file from daemon\n"));
2792 talloc_free(tmp_ctx);
2796 if (reclockfile == NULL) {
2797 if (ctdb->recovery_lock_file != NULL) {
2798 DEBUG(DEBUG_ERR,("Reclock file disabled\n"));
2799 talloc_free(ctdb->recovery_lock_file);
2800 ctdb->recovery_lock_file = NULL;
2801 if (ctdb->recovery_lock_fd != -1) {
2802 close(ctdb->recovery_lock_fd);
2803 ctdb->recovery_lock_fd = -1;
2806 ctdb->tunable.verify_recovery_lock = 0;
2807 talloc_free(tmp_ctx);
2811 if (ctdb->recovery_lock_file == NULL) {
2812 ctdb->recovery_lock_file = talloc_strdup(ctdb, reclockfile);
2813 if (ctdb->recovery_lock_fd != -1) {
2814 close(ctdb->recovery_lock_fd);
2815 ctdb->recovery_lock_fd = -1;
2817 talloc_free(tmp_ctx);
2822 if (!strcmp(reclockfile, ctdb->recovery_lock_file)) {
2823 talloc_free(tmp_ctx);
2827 talloc_free(ctdb->recovery_lock_file);
2828 ctdb->recovery_lock_file = talloc_strdup(ctdb, reclockfile);
2829 ctdb->tunable.verify_recovery_lock = 0;
2830 if (ctdb->recovery_lock_fd != -1) {
2831 close(ctdb->recovery_lock_fd);
2832 ctdb->recovery_lock_fd = -1;
2835 talloc_free(tmp_ctx);
2840 the main monitoring loop
2842 static void monitor_cluster(struct ctdb_context *ctdb)
2845 TALLOC_CTX *mem_ctx=NULL;
2846 struct ctdb_node_map *nodemap=NULL;
2847 struct ctdb_node_map *recmaster_nodemap=NULL;
2848 struct ctdb_node_map **remote_nodemaps=NULL;
2849 struct ctdb_vnn_map *vnnmap=NULL;
2850 struct ctdb_vnn_map *remote_vnnmap=NULL;
2851 int32_t debug_level;
2853 struct ctdb_recoverd *rec;
2855 DEBUG(DEBUG_NOTICE,("monitor_cluster starting\n"));
2857 rec = talloc_zero(ctdb, struct ctdb_recoverd);
2858 CTDB_NO_MEMORY_FATAL(ctdb, rec);
2862 rec->priority_time = timeval_current();
2864 /* register a message port for sending memory dumps */
2865 ctdb_set_message_handler(ctdb, CTDB_SRVID_MEM_DUMP, mem_dump_handler, rec);
2867 /* register a message port for recovery elections */
2868 ctdb_set_message_handler(ctdb, CTDB_SRVID_RECOVERY, election_handler, rec);
2870 /* when nodes are disabled/enabled */
2871 ctdb_set_message_handler(ctdb, CTDB_SRVID_SET_NODE_FLAGS, monitor_handler, rec);
2873 /* when we are asked to puch out a flag change */
2874 ctdb_set_message_handler(ctdb, CTDB_SRVID_PUSH_NODE_FLAGS, push_flags_handler, rec);
2876 /* register a message port for vacuum fetch */
2877 ctdb_set_message_handler(ctdb, CTDB_SRVID_VACUUM_FETCH, vacuum_fetch_handler, rec);
2879 /* register a message port for reloadnodes */
2880 ctdb_set_message_handler(ctdb, CTDB_SRVID_RELOAD_NODES, reload_nodes_handler, rec);
2882 /* register a message port for performing a takeover run */
2883 ctdb_set_message_handler(ctdb, CTDB_SRVID_TAKEOVER_RUN, ip_reallocate_handler, rec);
2885 /* register a message port for disabling the ip check for a short while */
2886 ctdb_set_message_handler(ctdb, CTDB_SRVID_DISABLE_IP_CHECK, disable_ip_check_handler, rec);
2888 /* register a message port for updating the recovery daemons node assignment for an ip */
2889 ctdb_set_message_handler(ctdb, CTDB_SRVID_RECD_UPDATE_IP, recd_update_ip_handler, rec);
2893 talloc_free(mem_ctx);
2896 mem_ctx = talloc_new(ctdb);
2898 DEBUG(DEBUG_CRIT,(__location__ " Failed to create temporary context\n"));
2902 /* we only check for recovery once every second */
2903 ctdb_wait_timeout(ctdb, ctdb->tunable.recover_interval);
2905 /* verify that the main daemon is still running */
2906 if (kill(ctdb->ctdbd_pid, 0) != 0) {
2907 DEBUG(DEBUG_CRIT,("CTDB daemon is no longer available. Shutting down recovery daemon\n"));
2911 /* ping the local daemon to tell it we are alive */
2912 ctdb_ctrl_recd_ping(ctdb);
2914 if (rec->election_timeout) {
2915 /* an election is in progress */
2919 /* read the debug level from the parent and update locally */
2920 ret = ctdb_ctrl_get_debuglevel(ctdb, CTDB_CURRENT_NODE, &debug_level);
2922 DEBUG(DEBUG_ERR, (__location__ " Failed to read debuglevel from parent\n"));
2925 LogLevel = debug_level;
2928 /* We must check if we need to ban a node here but we want to do this
2929 as early as possible so we dont wait until we have pulled the node
2930 map from the local node. thats why we have the hardcoded value 20
2932 for (i=0; i<ctdb->num_nodes; i++) {
2933 struct ctdb_banning_state *ban_state;
2935 if (ctdb->nodes[i]->ban_state == NULL) {
2938 ban_state = (struct ctdb_banning_state *)ctdb->nodes[i]->ban_state;
2939 if (ban_state->count < 20) {
2942 DEBUG(DEBUG_NOTICE,("Node %u has caused %u recoveries recently - banning it for %u seconds\n",
2943 ctdb->nodes[i]->pnn, ban_state->count,
2944 ctdb->tunable.recovery_ban_period));
2945 ctdb_ban_node(rec, ctdb->nodes[i]->pnn, ctdb->tunable.recovery_ban_period);
2946 ban_state->count = 0;
2949 /* get relevant tunables */
2950 ret = ctdb_ctrl_get_all_tunables(ctdb, CONTROL_TIMEOUT(), CTDB_CURRENT_NODE, &ctdb->tunable);
2952 DEBUG(DEBUG_ERR,("Failed to get tunables - retrying\n"));
2956 /* get the current recovery lock file from the server */
2957 if (update_recovery_lock_file(ctdb) != 0) {
2958 DEBUG(DEBUG_ERR,("Failed to update the recovery lock file\n"));
2962 /* Make sure that if recovery lock verification becomes disabled when
2965 if (ctdb->tunable.verify_recovery_lock == 0) {
2966 if (ctdb->recovery_lock_fd != -1) {
2967 close(ctdb->recovery_lock_fd);
2968 ctdb->recovery_lock_fd = -1;
2972 pnn = ctdb_ctrl_getpnn(ctdb, CONTROL_TIMEOUT(), CTDB_CURRENT_NODE);
2973 if (pnn == (uint32_t)-1) {
2974 DEBUG(DEBUG_ERR,("Failed to get local pnn - retrying\n"));
2978 /* get the vnnmap */
2979 ret = ctdb_ctrl_getvnnmap(ctdb, CONTROL_TIMEOUT(), pnn, mem_ctx, &vnnmap);
2981 DEBUG(DEBUG_ERR, (__location__ " Unable to get vnnmap from node %u\n", pnn));
2986 /* get number of nodes */
2988 talloc_free(rec->nodemap);
2989 rec->nodemap = NULL;
2992 ret = ctdb_ctrl_getnodemap(ctdb, CONTROL_TIMEOUT(), pnn, rec, &rec->nodemap);
2994 DEBUG(DEBUG_ERR, (__location__ " Unable to get nodemap from node %u\n", pnn));
2997 nodemap = rec->nodemap;
2999 /* check which node is the recovery master */
3000 ret = ctdb_ctrl_getrecmaster(ctdb, mem_ctx, CONTROL_TIMEOUT(), pnn, &rec->recmaster);
3002 DEBUG(DEBUG_ERR, (__location__ " Unable to get recmaster from node %u\n", pnn));
3006 /* if we are not the recmaster we can safely ignore any ip reallocate requests */
3007 if (rec->recmaster != pnn) {
3008 if (rec->ip_reallocate_ctx != NULL) {
3009 talloc_free(rec->ip_reallocate_ctx);
3010 rec->ip_reallocate_ctx = NULL;
3011 rec->reallocate_callers = NULL;
3014 /* if there are takeovers requested, perform it and notify the waiters */
3015 if (rec->reallocate_callers) {
3016 process_ipreallocate_requests(ctdb, rec);
3019 if (rec->recmaster == (uint32_t)-1) {
3020 DEBUG(DEBUG_NOTICE,(__location__ " Initial recovery master set - forcing election\n"));
3021 force_election(rec, pnn, nodemap);
3026 /* if the local daemon is STOPPED, we verify that the databases are
3027 also frozen and thet the recmode is set to active
3029 if (nodemap->nodes[pnn].flags & NODE_FLAGS_STOPPED) {
3030 ret = ctdb_ctrl_getrecmode(ctdb, mem_ctx, CONTROL_TIMEOUT(), CTDB_CURRENT_NODE, &ctdb->recovery_mode);
3032 DEBUG(DEBUG_ERR,(__location__ " Failed to read recmode from local node\n"));
3034 if (ctdb->recovery_mode == CTDB_RECOVERY_NORMAL) {
3035 DEBUG(DEBUG_ERR,("Node is stopped but recovery mode is not active. Activate recovery mode and lock databases\n"));
3037 ret = ctdb_ctrl_freeze_priority(ctdb, CONTROL_TIMEOUT(), CTDB_CURRENT_NODE, 1);
3039 DEBUG(DEBUG_ERR,(__location__ " Failed to freeze node due to node being STOPPED\n"));
3042 ret = ctdb_ctrl_setrecmode(ctdb, CONTROL_TIMEOUT(), CTDB_CURRENT_NODE, CTDB_RECOVERY_ACTIVE);
3044 DEBUG(DEBUG_ERR,(__location__ " Failed to activate recovery mode due to node being stopped\n"));
3051 /* If the local node is stopped, verify we are not the recmaster
3052 and yield this role if so
3054 if ((nodemap->nodes[pnn].flags & NODE_FLAGS_STOPPED) && (rec->recmaster == pnn)) {
3055 DEBUG(DEBUG_ERR,("Local node is STOPPED. Yielding recmaster role\n"));
3056 force_election(rec, pnn, nodemap);
3060 /* check that we (recovery daemon) and the local ctdb daemon
3061 agrees on whether we are banned or not
3065 /* remember our own node flags */
3066 rec->node_flags = nodemap->nodes[pnn].flags;
3068 /* count how many active nodes there are */
3069 rec->num_active = 0;
3070 rec->num_connected = 0;
3071 for (i=0; i<nodemap->num; i++) {
3072 if (!(nodemap->nodes[i].flags & NODE_FLAGS_INACTIVE)) {
3075 if (!(nodemap->nodes[i].flags & NODE_FLAGS_DISCONNECTED)) {
3076 rec->num_connected++;
3081 /* verify that the recmaster node is still active */
3082 for (j=0; j<nodemap->num; j++) {
3083 if (nodemap->nodes[j].pnn==rec->recmaster) {
3088 if (j == nodemap->num) {
3089 DEBUG(DEBUG_ERR, ("Recmaster node %u not in list. Force reelection\n", rec->recmaster));
3090 force_election(rec, pnn, nodemap);
3094 /* if recovery master is disconnected we must elect a new recmaster */
3095 if (nodemap->nodes[j].flags & NODE_FLAGS_DISCONNECTED) {
3096 DEBUG(DEBUG_NOTICE, ("Recmaster node %u is disconnected. Force reelection\n", nodemap->nodes[j].pnn));
3097 force_election(rec, pnn, nodemap);
3101 /* grap the nodemap from the recovery master to check if it is banned */
3102 ret = ctdb_ctrl_getnodemap(ctdb, CONTROL_TIMEOUT(), nodemap->nodes[j].pnn,
3103 mem_ctx, &recmaster_nodemap);
3105 DEBUG(DEBUG_ERR, (__location__ " Unable to get nodemap from recovery master %u\n",
3106 nodemap->nodes[j].pnn));
3111 if (recmaster_nodemap->nodes[j].flags & NODE_FLAGS_INACTIVE) {
3112 DEBUG(DEBUG_NOTICE, ("Recmaster node %u no longer available. Force reelection\n", nodemap->nodes[j].pnn));
3113 force_election(rec, pnn, nodemap);
3118 /* verify that we have all ip addresses we should have and we dont
3119 * have addresses we shouldnt have.
3121 if (ctdb->do_checkpublicip) {
3122 if (rec->ip_check_disable_ctx == NULL) {
3123 if (verify_local_ip_allocation(ctdb, rec, pnn) != 0) {
3124 DEBUG(DEBUG_ERR, (__location__ " Public IPs were inconsistent.\n"));
3130 /* if we are not the recmaster then we do not need to check
3131 if recovery is needed
3133 if (pnn != rec->recmaster) {
3138 /* ensure our local copies of flags are right */
3139 ret = update_local_flags(rec, nodemap);
3140 if (ret == MONITOR_ELECTION_NEEDED) {
3141 DEBUG(DEBUG_NOTICE,("update_local_flags() called for a re-election.\n"));
3142 force_election(rec, pnn, nodemap);
3145 if (ret != MONITOR_OK) {
3146 DEBUG(DEBUG_ERR,("Unable to update local flags\n"));
3150 if (ctdb->num_nodes != nodemap->num) {
3151 DEBUG(DEBUG_ERR, (__location__ " ctdb->num_nodes (%d) != nodemap->num (%d) reloading nodes file\n", ctdb->num_nodes, nodemap->num));
3152 reload_nodes_file(ctdb);
3156 /* verify that all active nodes agree that we are the recmaster */
3157 switch (verify_recmaster(rec, nodemap, pnn)) {
3158 case MONITOR_RECOVERY_NEEDED:
3159 /* can not happen */
3161 case MONITOR_ELECTION_NEEDED:
3162 force_election(rec, pnn, nodemap);
3166 case MONITOR_FAILED:
3171 if (rec->need_recovery) {
3172 /* a previous recovery didn't finish */
3173 do_recovery(rec, mem_ctx, pnn, nodemap, vnnmap);
3177 /* verify that all active nodes are in normal mode
3178 and not in recovery mode
3180 switch (verify_recmode(ctdb, nodemap)) {
3181 case MONITOR_RECOVERY_NEEDED:
3182 do_recovery(rec, mem_ctx, pnn, nodemap, vnnmap);
3184 case MONITOR_FAILED:
3186 case MONITOR_ELECTION_NEEDED:
3187 /* can not happen */
3193 if (ctdb->tunable.verify_recovery_lock != 0) {
3194 /* we should have the reclock - check its not stale */
3195 ret = check_recovery_lock(ctdb);
3197 DEBUG(DEBUG_ERR,("Failed check_recovery_lock. Force a recovery\n"));
3198 ctdb_set_culprit(rec, ctdb->pnn);
3199 do_recovery(rec, mem_ctx, pnn, nodemap, vnnmap);
3204 /* get the nodemap for all active remote nodes
3206 remote_nodemaps = talloc_array(mem_ctx, struct ctdb_node_map *, nodemap->num);
3207 if (remote_nodemaps == NULL) {
3208 DEBUG(DEBUG_ERR, (__location__ " failed to allocate remote nodemap array\n"));
3211 for(i=0; i<nodemap->num; i++) {
3212 remote_nodemaps[i] = NULL;
3214 if (get_remote_nodemaps(ctdb, mem_ctx, nodemap, remote_nodemaps) != 0) {
3215 DEBUG(DEBUG_ERR,(__location__ " Failed to read remote nodemaps\n"));
3219 /* verify that all other nodes have the same nodemap as we have
3221 for (j=0; j<nodemap->num; j++) {
3222 if (nodemap->nodes[j].flags & NODE_FLAGS_INACTIVE) {
3226 if (remote_nodemaps[j] == NULL) {
3227 DEBUG(DEBUG_ERR,(__location__ " Did not get a remote nodemap for node %d, restarting monitoring\n", j));
3228 ctdb_set_culprit(rec, j);
3233 /* if the nodes disagree on how many nodes there are
3234 then this is a good reason to try recovery
3236 if (remote_nodemaps[j]->num != nodemap->num) {
3237 DEBUG(DEBUG_ERR, (__location__ " Remote node:%u has different node count. %u vs %u of the local node\n",
3238 nodemap->nodes[j].pnn, remote_nodemaps[j]->num, nodemap->num));
3239 ctdb_set_culprit(rec, nodemap->nodes[j].pnn);
3240 do_recovery(rec, mem_ctx, pnn, nodemap, vnnmap);
3244 /* if the nodes disagree on which nodes exist and are
3245 active, then that is also a good reason to do recovery
3247 for (i=0;i<nodemap->num;i++) {
3248 if (remote_nodemaps[j]->nodes[i].pnn != nodemap->nodes[i].pnn) {
3249 DEBUG(DEBUG_ERR, (__location__ " Remote node:%u has different nodemap pnn for %d (%u vs %u).\n",
3250 nodemap->nodes[j].pnn, i,
3251 remote_nodemaps[j]->nodes[i].pnn, nodemap->nodes[i].pnn));
3252 ctdb_set_culprit(rec, nodemap->nodes[j].pnn);
3253 do_recovery(rec, mem_ctx, pnn, nodemap,
3259 /* verify the flags are consistent
3261 for (i=0; i<nodemap->num; i++) {
3262 if (nodemap->nodes[i].flags & NODE_FLAGS_DISCONNECTED) {
3266 if (nodemap->nodes[i].flags != remote_nodemaps[j]->nodes[i].flags) {
3267 DEBUG(DEBUG_ERR, (__location__ " Remote node:%u has different flags for node %u. It has 0x%02x vs our 0x%02x\n",
3268 nodemap->nodes[j].pnn,
3269 nodemap->nodes[i].pnn,
3270 remote_nodemaps[j]->nodes[i].flags,
3271 nodemap->nodes[j].flags));
3273 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));
3274 update_flags_on_all_nodes(ctdb, nodemap, nodemap->nodes[i].pnn, remote_nodemaps[j]->nodes[i].flags);
3275 ctdb_set_culprit(rec, nodemap->nodes[j].pnn);
3276 do_recovery(rec, mem_ctx, pnn, nodemap,
3280 DEBUG(DEBUG_ERR,("Use flags 0x%02x from local recmaster node for cluster update of node %d flags\n", nodemap->nodes[i].flags, i));
3281 update_flags_on_all_nodes(ctdb, nodemap, nodemap->nodes[i].pnn, nodemap->nodes[i].flags);
3282 ctdb_set_culprit(rec, nodemap->nodes[j].pnn);
3283 do_recovery(rec, mem_ctx, pnn, nodemap,
3292 /* there better be the same number of lmasters in the vnn map
3293 as there are active nodes or we will have to do a recovery
3295 if (vnnmap->size != rec->num_active) {
3296 DEBUG(DEBUG_ERR, (__location__ " The vnnmap count is different from the number of active nodes. %u vs %u\n",
3297 vnnmap->size, rec->num_active));
3298 ctdb_set_culprit(rec, ctdb->pnn);
3299 do_recovery(rec, mem_ctx, pnn, nodemap, vnnmap);
3303 /* verify that all active nodes in the nodemap also exist in
3306 for (j=0; j<nodemap->num; j++) {
3307 if (nodemap->nodes[j].flags & NODE_FLAGS_INACTIVE) {
3310 if (nodemap->nodes[j].pnn == pnn) {
3314 for (i=0; i<vnnmap->size; i++) {
3315 if (vnnmap->map[i] == nodemap->nodes[j].pnn) {
3319 if (i == vnnmap->size) {
3320 DEBUG(DEBUG_ERR, (__location__ " Node %u is active in the nodemap but did not exist in the vnnmap\n",
3321 nodemap->nodes[j].pnn));
3322 ctdb_set_culprit(rec, nodemap->nodes[j].pnn);
3323 do_recovery(rec, mem_ctx, pnn, nodemap, vnnmap);
3329 /* verify that all other nodes have the same vnnmap
3330 and are from the same generation
3332 for (j=0; j<nodemap->num; j++) {
3333 if (nodemap->nodes[j].flags & NODE_FLAGS_INACTIVE) {
3336 if (nodemap->nodes[j].pnn == pnn) {
3340 ret = ctdb_ctrl_getvnnmap(ctdb, CONTROL_TIMEOUT(), nodemap->nodes[j].pnn,
3341 mem_ctx, &remote_vnnmap);
3343 DEBUG(DEBUG_ERR, (__location__ " Unable to get vnnmap from remote node %u\n",
3344 nodemap->nodes[j].pnn));
3348 /* verify the vnnmap generation is the same */
3349 if (vnnmap->generation != remote_vnnmap->generation) {
3350 DEBUG(DEBUG_ERR, (__location__ " Remote node %u has different generation of vnnmap. %u vs %u (ours)\n",
3351 nodemap->nodes[j].pnn, remote_vnnmap->generation, vnnmap->generation));
3352 ctdb_set_culprit(rec, nodemap->nodes[j].pnn);
3353 do_recovery(rec, mem_ctx, pnn, nodemap, vnnmap);
3357 /* verify the vnnmap size is the same */
3358 if (vnnmap->size != remote_vnnmap->size) {
3359 DEBUG(DEBUG_ERR, (__location__ " Remote node %u has different size of vnnmap. %u vs %u (ours)\n",
3360 nodemap->nodes[j].pnn, remote_vnnmap->size, vnnmap->size));
3361 ctdb_set_culprit(rec, nodemap->nodes[j].pnn);
3362 do_recovery(rec, mem_ctx, pnn, nodemap, vnnmap);
3366 /* verify the vnnmap is the same */
3367 for (i=0;i<vnnmap->size;i++) {
3368 if (remote_vnnmap->map[i] != vnnmap->map[i]) {
3369 DEBUG(DEBUG_ERR, (__location__ " Remote node %u has different vnnmap.\n",
3370 nodemap->nodes[j].pnn));
3371 ctdb_set_culprit(rec, nodemap->nodes[j].pnn);
3372 do_recovery(rec, mem_ctx, pnn, nodemap,
3379 /* we might need to change who has what IP assigned */
3380 if (rec->need_takeover_run) {
3381 uint32_t culprit = (uint32_t)-1;
3383 rec->need_takeover_run = false;
3385 /* update the list of public ips that a node can handle for
3388 ret = ctdb_reload_remote_public_ips(ctdb, rec, nodemap, &culprit);
3390 DEBUG(DEBUG_ERR,("Failed to read public ips from remote node %d\n",
3392 ctdb_set_culprit(rec, culprit);
3393 do_recovery(rec, mem_ctx, pnn, nodemap, vnnmap);
3397 /* execute the "startrecovery" event script on all nodes */
3398 ret = run_startrecovery_eventscript(rec, nodemap);
3400 DEBUG(DEBUG_ERR, (__location__ " Unable to run the 'startrecovery' event on cluster\n"));
3401 ctdb_set_culprit(rec, ctdb->pnn);
3402 do_recovery(rec, mem_ctx, pnn, nodemap, vnnmap);
3406 ret = ctdb_takeover_run(ctdb, nodemap);
3408 DEBUG(DEBUG_ERR, (__location__ " Unable to setup public takeover addresses - starting recovery\n"));
3409 ctdb_set_culprit(rec, ctdb->pnn);
3410 do_recovery(rec, mem_ctx, pnn, nodemap, vnnmap);
3414 /* execute the "recovered" event script on all nodes */
3415 ret = run_recovered_eventscript(ctdb, nodemap, "monitor_cluster");
3417 // we cant check whether the event completed successfully
3418 // since this script WILL fail if the node is in recovery mode
3419 // and if that race happens, the code here would just cause a second
3420 // cascading recovery.
3422 DEBUG(DEBUG_ERR, (__location__ " Unable to run the 'recovered' event on cluster. Update of public ips failed.\n"));
3423 ctdb_set_culprit(rec, ctdb->pnn);
3424 do_recovery(rec, mem_ctx, pnn, nodemap, vnnmap);
3435 event handler for when the main ctdbd dies
3437 static void ctdb_recoverd_parent(struct event_context *ev, struct fd_event *fde,
3438 uint16_t flags, void *private_data)
3440 DEBUG(DEBUG_ALERT,("recovery daemon parent died - exiting\n"));
3445 called regularly to verify that the recovery daemon is still running
3447 static void ctdb_check_recd(struct event_context *ev, struct timed_event *te,
3448 struct timeval yt, void *p)
3450 struct ctdb_context *ctdb = talloc_get_type(p, struct ctdb_context);
3452 if (kill(ctdb->recoverd_pid, 0) != 0) {
3453 DEBUG(DEBUG_ERR,("Recovery daemon (pid:%d) is no longer running. Shutting down main daemon\n", (int)ctdb->recoverd_pid));
3455 ctdb_stop_recoverd(ctdb);
3456 ctdb_stop_keepalive(ctdb);
3457 ctdb_stop_monitoring(ctdb);
3458 ctdb_release_all_ips(ctdb);
3459 if (ctdb->methods != NULL) {
3460 ctdb->methods->shutdown(ctdb);
3462 ctdb_event_script(ctdb, CTDB_EVENT_SHUTDOWN);
3467 event_add_timed(ctdb->ev, ctdb,
3468 timeval_current_ofs(30, 0),
3469 ctdb_check_recd, ctdb);
3472 static void recd_sig_child_handler(struct event_context *ev,
3473 struct signal_event *se, int signum, int count,
3477 // struct ctdb_context *ctdb = talloc_get_type(private_data, struct ctdb_context);
3482 pid = waitpid(-1, &status, WNOHANG);
3484 if (errno != ECHILD) {
3485 DEBUG(DEBUG_ERR, (__location__ " waitpid() returned error. errno:%s(%d)\n", strerror(errno),errno));
3490 DEBUG(DEBUG_DEBUG, ("RECD SIGCHLD from %d\n", (int)pid));
3496 startup the recovery daemon as a child of the main ctdb daemon
3498 int ctdb_start_recoverd(struct ctdb_context *ctdb)
3501 struct signal_event *se;
3503 if (pipe(fd) != 0) {
3507 ctdb->ctdbd_pid = getpid();
3509 ctdb->recoverd_pid = fork();
3510 if (ctdb->recoverd_pid == -1) {
3514 if (ctdb->recoverd_pid != 0) {
3516 event_add_timed(ctdb->ev, ctdb,
3517 timeval_current_ofs(30, 0),
3518 ctdb_check_recd, ctdb);
3524 srandom(getpid() ^ time(NULL));
3526 if (switch_from_server_to_client(ctdb) != 0) {
3527 DEBUG(DEBUG_CRIT, (__location__ "ERROR: failed to switch recovery daemon into client mode. shutting down.\n"));
3531 DEBUG(DEBUG_DEBUG, (__location__ " Created PIPE FD:%d to recovery daemon\n", fd[0]));
3533 event_add_fd(ctdb->ev, ctdb, fd[0], EVENT_FD_READ|EVENT_FD_AUTOCLOSE,
3534 ctdb_recoverd_parent, &fd[0]);
3536 /* set up a handler to pick up sigchld */
3537 se = event_add_signal(ctdb->ev, ctdb,
3539 recd_sig_child_handler,
3542 DEBUG(DEBUG_CRIT,("Failed to set up signal handler for SIGCHLD in recovery daemon\n"));
3546 monitor_cluster(ctdb);
3548 DEBUG(DEBUG_ALERT,("ERROR: ctdb_recoverd finished!?\n"));
3553 shutdown the recovery daemon
3555 void ctdb_stop_recoverd(struct ctdb_context *ctdb)
3557 if (ctdb->recoverd_pid == 0) {
3561 DEBUG(DEBUG_NOTICE,("Shutting down recovery daemon\n"));
3562 kill(ctdb->recoverd_pid, SIGTERM);