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_client.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, double secs)
892 uint32_t timed_out = 0;
893 time_t usecs = (secs - (time_t)secs) * 1000000;
894 event_add_timed(ctdb->ev, ctdb, timeval_current_ofs(secs, usecs), ctdb_wait_handler, &timed_out);
896 event_loop_once(ctdb->ev);
901 called when an election times out (ends)
903 static void ctdb_election_timeout(struct event_context *ev, struct timed_event *te,
904 struct timeval t, void *p)
906 struct ctdb_recoverd *rec = talloc_get_type(p, struct ctdb_recoverd);
907 rec->election_timeout = NULL;
910 DEBUG(DEBUG_WARNING,(__location__ " Election timed out\n"));
915 wait for an election to finish. It finished election_timeout seconds after
916 the last election packet is received
918 static void ctdb_wait_election(struct ctdb_recoverd *rec)
920 struct ctdb_context *ctdb = rec->ctdb;
921 while (rec->election_timeout) {
922 event_loop_once(ctdb->ev);
927 Update our local flags from all remote connected nodes.
928 This is only run when we are or we belive we are the recovery master
930 static int update_local_flags(struct ctdb_recoverd *rec, struct ctdb_node_map *nodemap)
933 struct ctdb_context *ctdb = rec->ctdb;
934 TALLOC_CTX *mem_ctx = talloc_new(ctdb);
936 /* get the nodemap for all active remote nodes and verify
937 they are the same as for this node
939 for (j=0; j<nodemap->num; j++) {
940 struct ctdb_node_map *remote_nodemap=NULL;
943 if (nodemap->nodes[j].flags & NODE_FLAGS_DISCONNECTED) {
946 if (nodemap->nodes[j].pnn == ctdb->pnn) {
950 ret = ctdb_ctrl_getnodemap(ctdb, CONTROL_TIMEOUT(), nodemap->nodes[j].pnn,
951 mem_ctx, &remote_nodemap);
953 DEBUG(DEBUG_ERR, (__location__ " Unable to get nodemap from remote node %u\n",
954 nodemap->nodes[j].pnn));
955 ctdb_set_culprit(rec, nodemap->nodes[j].pnn);
956 talloc_free(mem_ctx);
957 return MONITOR_FAILED;
959 if (nodemap->nodes[j].flags != remote_nodemap->nodes[j].flags) {
960 /* We should tell our daemon about this so it
961 updates its flags or else we will log the same
962 message again in the next iteration of recovery.
963 Since we are the recovery master we can just as
964 well update the flags on all nodes.
966 ret = ctdb_ctrl_modflags(ctdb, CONTROL_TIMEOUT(), nodemap->nodes[j].pnn, nodemap->nodes[j].flags, ~nodemap->nodes[j].flags);
968 DEBUG(DEBUG_ERR, (__location__ " Unable to update nodeflags on remote nodes\n"));
972 /* Update our local copy of the flags in the recovery
975 DEBUG(DEBUG_NOTICE,("Remote node %u had flags 0x%x, local had 0x%x - updating local\n",
976 nodemap->nodes[j].pnn, remote_nodemap->nodes[j].flags,
977 nodemap->nodes[j].flags));
978 nodemap->nodes[j].flags = remote_nodemap->nodes[j].flags;
980 talloc_free(remote_nodemap);
982 talloc_free(mem_ctx);
987 /* Create a new random generation ip.
988 The generation id can not be the INVALID_GENERATION id
990 static uint32_t new_generation(void)
995 generation = random();
997 if (generation != INVALID_GENERATION) {
1007 create a temporary working database
1009 static struct tdb_wrap *create_recdb(struct ctdb_context *ctdb, TALLOC_CTX *mem_ctx)
1012 struct tdb_wrap *recdb;
1015 /* open up the temporary recovery database */
1016 name = talloc_asprintf(mem_ctx, "%s/recdb.tdb.%u",
1017 ctdb->db_directory_state,
1024 tdb_flags = TDB_NOLOCK;
1025 if (ctdb->valgrinding) {
1026 tdb_flags |= TDB_NOMMAP;
1028 tdb_flags |= TDB_DISALLOW_NESTING;
1030 recdb = tdb_wrap_open(mem_ctx, name, ctdb->tunable.database_hash_size,
1031 tdb_flags, O_RDWR|O_CREAT|O_EXCL, 0600);
1032 if (recdb == NULL) {
1033 DEBUG(DEBUG_CRIT,(__location__ " Failed to create temp recovery database '%s'\n", name));
1043 a traverse function for pulling all relevent records from recdb
1046 struct ctdb_context *ctdb;
1047 struct ctdb_marshall_buffer *recdata;
1053 static int traverse_recdb(struct tdb_context *tdb, TDB_DATA key, TDB_DATA data, void *p)
1055 struct recdb_data *params = (struct recdb_data *)p;
1056 struct ctdb_rec_data *rec;
1057 struct ctdb_ltdb_header *hdr;
1059 /* skip empty records */
1060 if (data.dsize <= sizeof(struct ctdb_ltdb_header)) {
1064 /* update the dmaster field to point to us */
1065 hdr = (struct ctdb_ltdb_header *)data.dptr;
1066 if (!params->persistent) {
1067 hdr->dmaster = params->ctdb->pnn;
1070 /* add the record to the blob ready to send to the nodes */
1071 rec = ctdb_marshall_record(params->recdata, 0, key, NULL, data);
1073 params->failed = true;
1076 params->recdata = talloc_realloc_size(NULL, params->recdata, rec->length + params->len);
1077 if (params->recdata == NULL) {
1078 DEBUG(DEBUG_CRIT,(__location__ " Failed to expand recdata to %u (%u records)\n",
1079 rec->length + params->len, params->recdata->count));
1080 params->failed = true;
1083 params->recdata->count++;
1084 memcpy(params->len+(uint8_t *)params->recdata, rec, rec->length);
1085 params->len += rec->length;
1092 push the recdb database out to all nodes
1094 static int push_recdb_database(struct ctdb_context *ctdb, uint32_t dbid,
1096 struct tdb_wrap *recdb, struct ctdb_node_map *nodemap)
1098 struct recdb_data params;
1099 struct ctdb_marshall_buffer *recdata;
1101 TALLOC_CTX *tmp_ctx;
1104 tmp_ctx = talloc_new(ctdb);
1105 CTDB_NO_MEMORY(ctdb, tmp_ctx);
1107 recdata = talloc_zero(recdb, struct ctdb_marshall_buffer);
1108 CTDB_NO_MEMORY(ctdb, recdata);
1110 recdata->db_id = dbid;
1113 params.recdata = recdata;
1114 params.len = offsetof(struct ctdb_marshall_buffer, data);
1115 params.failed = false;
1116 params.persistent = persistent;
1118 if (tdb_traverse_read(recdb->tdb, traverse_recdb, ¶ms) == -1) {
1119 DEBUG(DEBUG_ERR,(__location__ " Failed to traverse recdb database\n"));
1120 talloc_free(params.recdata);
1121 talloc_free(tmp_ctx);
1125 if (params.failed) {
1126 DEBUG(DEBUG_ERR,(__location__ " Failed to traverse recdb database\n"));
1127 talloc_free(params.recdata);
1128 talloc_free(tmp_ctx);
1132 recdata = params.recdata;
1134 outdata.dptr = (void *)recdata;
1135 outdata.dsize = params.len;
1137 nodes = list_of_active_nodes(ctdb, nodemap, tmp_ctx, true);
1138 if (ctdb_client_async_control(ctdb, CTDB_CONTROL_PUSH_DB,
1140 CONTROL_TIMEOUT(), false, outdata,
1143 DEBUG(DEBUG_ERR,(__location__ " Failed to push recdb records to nodes for db 0x%x\n", dbid));
1144 talloc_free(recdata);
1145 talloc_free(tmp_ctx);
1149 DEBUG(DEBUG_NOTICE, (__location__ " Recovery - pushed remote database 0x%x of size %u\n",
1150 dbid, recdata->count));
1152 talloc_free(recdata);
1153 talloc_free(tmp_ctx);
1160 go through a full recovery on one database
1162 static int recover_database(struct ctdb_recoverd *rec,
1163 TALLOC_CTX *mem_ctx,
1167 struct ctdb_node_map *nodemap,
1168 uint32_t transaction_id)
1170 struct tdb_wrap *recdb;
1172 struct ctdb_context *ctdb = rec->ctdb;
1174 struct ctdb_control_wipe_database w;
1177 recdb = create_recdb(ctdb, mem_ctx);
1178 if (recdb == NULL) {
1182 /* pull all remote databases onto the recdb */
1183 ret = pull_remote_database(ctdb, rec, nodemap, recdb, dbid, persistent);
1185 DEBUG(DEBUG_ERR, (__location__ " Unable to pull remote database 0x%x\n", dbid));
1189 DEBUG(DEBUG_NOTICE, (__location__ " Recovery - pulled remote database 0x%x\n", dbid));
1191 /* wipe all the remote databases. This is safe as we are in a transaction */
1193 w.transaction_id = transaction_id;
1195 data.dptr = (void *)&w;
1196 data.dsize = sizeof(w);
1198 nodes = list_of_active_nodes(ctdb, nodemap, recdb, true);
1199 if (ctdb_client_async_control(ctdb, CTDB_CONTROL_WIPE_DATABASE,
1201 CONTROL_TIMEOUT(), false, data,
1204 DEBUG(DEBUG_ERR, (__location__ " Unable to wipe database. Recovery failed.\n"));
1209 /* push out the correct database. This sets the dmaster and skips
1210 the empty records */
1211 ret = push_recdb_database(ctdb, dbid, persistent, recdb, nodemap);
1217 /* all done with this database */
1224 reload the nodes file
1226 static void reload_nodes_file(struct ctdb_context *ctdb)
1229 ctdb_load_nodes_file(ctdb);
1232 static int ctdb_reload_remote_public_ips(struct ctdb_context *ctdb,
1233 struct ctdb_recoverd *rec,
1234 struct ctdb_node_map *nodemap,
1240 if (ctdb->num_nodes != nodemap->num) {
1241 DEBUG(DEBUG_ERR, (__location__ " ctdb->num_nodes (%d) != nodemap->num (%d) invalid param\n",
1242 ctdb->num_nodes, nodemap->num));
1244 *culprit = ctdb->pnn;
1249 for (j=0; j<nodemap->num; j++) {
1250 /* release any existing data */
1251 if (ctdb->nodes[j]->known_public_ips) {
1252 talloc_free(ctdb->nodes[j]->known_public_ips);
1253 ctdb->nodes[j]->known_public_ips = NULL;
1255 if (ctdb->nodes[j]->available_public_ips) {
1256 talloc_free(ctdb->nodes[j]->available_public_ips);
1257 ctdb->nodes[j]->available_public_ips = NULL;
1260 if (nodemap->nodes[j].flags & NODE_FLAGS_INACTIVE) {
1264 /* grab a new shiny list of public ips from the node */
1265 ret = ctdb_ctrl_get_public_ips_flags(ctdb,
1267 ctdb->nodes[j]->pnn,
1270 &ctdb->nodes[j]->known_public_ips);
1272 DEBUG(DEBUG_ERR,("Failed to read known public ips from node : %u\n",
1273 ctdb->nodes[j]->pnn));
1275 *culprit = ctdb->nodes[j]->pnn;
1280 if (rec->ip_check_disable_ctx == NULL) {
1281 if (verify_remote_ip_allocation(ctdb, ctdb->nodes[j]->known_public_ips)) {
1282 DEBUG(DEBUG_ERR,("Node %d has inconsistent public ip allocation and needs update.\n", ctdb->nodes[j]->pnn));
1283 rec->need_takeover_run = true;
1287 /* grab a new shiny list of public ips from the node */
1288 ret = ctdb_ctrl_get_public_ips_flags(ctdb,
1290 ctdb->nodes[j]->pnn,
1292 CTDB_PUBLIC_IP_FLAGS_ONLY_AVAILABLE,
1293 &ctdb->nodes[j]->available_public_ips);
1295 DEBUG(DEBUG_ERR,("Failed to read available public ips from node : %u\n",
1296 ctdb->nodes[j]->pnn));
1298 *culprit = ctdb->nodes[j]->pnn;
1307 /* when we start a recovery, make sure all nodes use the same reclock file
1310 static int sync_recovery_lock_file_across_cluster(struct ctdb_recoverd *rec)
1312 struct ctdb_context *ctdb = rec->ctdb;
1313 TALLOC_CTX *tmp_ctx = talloc_new(NULL);
1317 if (ctdb->recovery_lock_file == NULL) {
1321 data.dsize = strlen(ctdb->recovery_lock_file) + 1;
1322 data.dptr = (uint8_t *)ctdb->recovery_lock_file;
1325 nodes = list_of_active_nodes(ctdb, rec->nodemap, tmp_ctx, true);
1326 if (ctdb_client_async_control(ctdb, CTDB_CONTROL_SET_RECLOCK_FILE,
1332 DEBUG(DEBUG_ERR, (__location__ " Failed to sync reclock file settings\n"));
1333 talloc_free(tmp_ctx);
1337 talloc_free(tmp_ctx);
1343 we are the recmaster, and recovery is needed - start a recovery run
1345 static int do_recovery(struct ctdb_recoverd *rec,
1346 TALLOC_CTX *mem_ctx, uint32_t pnn,
1347 struct ctdb_node_map *nodemap, struct ctdb_vnn_map *vnnmap)
1349 struct ctdb_context *ctdb = rec->ctdb;
1351 uint32_t generation;
1352 struct ctdb_dbid_map *dbmap;
1355 struct timeval start_time;
1356 uint32_t culprit = (uint32_t)-1;
1358 DEBUG(DEBUG_NOTICE, (__location__ " Starting do_recovery\n"));
1360 /* if recovery fails, force it again */
1361 rec->need_recovery = true;
1363 for (i=0; i<ctdb->num_nodes; i++) {
1364 struct ctdb_banning_state *ban_state;
1366 if (ctdb->nodes[i]->ban_state == NULL) {
1369 ban_state = (struct ctdb_banning_state *)ctdb->nodes[i]->ban_state;
1370 if (ban_state->count < 2*ctdb->num_nodes) {
1373 DEBUG(DEBUG_NOTICE,("Node %u has caused %u recoveries recently - banning it for %u seconds\n",
1374 ctdb->nodes[i]->pnn, ban_state->count,
1375 ctdb->tunable.recovery_ban_period));
1376 ctdb_ban_node(rec, ctdb->nodes[i]->pnn, ctdb->tunable.recovery_ban_period);
1377 ban_state->count = 0;
1381 if (ctdb->tunable.verify_recovery_lock != 0) {
1382 DEBUG(DEBUG_ERR,("Taking out recovery lock from recovery daemon\n"));
1383 start_time = timeval_current();
1384 if (!ctdb_recovery_lock(ctdb, true)) {
1385 ctdb_set_culprit(rec, pnn);
1386 DEBUG(DEBUG_ERR,("Unable to get recovery lock - aborting recovery\n"));
1389 ctdb_ctrl_report_recd_lock_latency(ctdb, CONTROL_TIMEOUT(), timeval_elapsed(&start_time));
1390 DEBUG(DEBUG_NOTICE,("Recovery lock taken successfully by recovery daemon\n"));
1393 DEBUG(DEBUG_NOTICE, (__location__ " Recovery initiated due to problem with node %u\n", rec->last_culprit_node));
1395 /* get a list of all databases */
1396 ret = ctdb_ctrl_getdbmap(ctdb, CONTROL_TIMEOUT(), pnn, mem_ctx, &dbmap);
1398 DEBUG(DEBUG_ERR, (__location__ " Unable to get dbids from node :%u\n", pnn));
1402 /* we do the db creation before we set the recovery mode, so the freeze happens
1403 on all databases we will be dealing with. */
1405 /* verify that we have all the databases any other node has */
1406 ret = create_missing_local_databases(ctdb, nodemap, pnn, &dbmap, mem_ctx);
1408 DEBUG(DEBUG_ERR, (__location__ " Unable to create missing local databases\n"));
1412 /* verify that all other nodes have all our databases */
1413 ret = create_missing_remote_databases(ctdb, nodemap, pnn, dbmap, mem_ctx);
1415 DEBUG(DEBUG_ERR, (__location__ " Unable to create missing remote databases\n"));
1418 DEBUG(DEBUG_NOTICE, (__location__ " Recovery - created remote databases\n"));
1420 /* update the database priority for all remote databases */
1421 ret = update_db_priority_on_remote_nodes(ctdb, nodemap, pnn, dbmap, mem_ctx);
1423 DEBUG(DEBUG_ERR, (__location__ " Unable to set db priority on remote nodes\n"));
1425 DEBUG(DEBUG_NOTICE, (__location__ " Recovery - updated db priority for all databases\n"));
1428 /* update all other nodes to use the same setting for reclock files
1429 as the local recovery master.
1431 sync_recovery_lock_file_across_cluster(rec);
1433 /* set recovery mode to active on all nodes */
1434 ret = set_recovery_mode(ctdb, rec, nodemap, CTDB_RECOVERY_ACTIVE);
1436 DEBUG(DEBUG_ERR, (__location__ " Unable to set recovery mode to active on cluster\n"));
1440 /* execute the "startrecovery" event script on all nodes */
1441 ret = run_startrecovery_eventscript(rec, nodemap);
1443 DEBUG(DEBUG_ERR, (__location__ " Unable to run the 'startrecovery' event on cluster\n"));
1448 update all nodes to have the same flags that we have
1450 for (i=0;i<nodemap->num;i++) {
1451 if (nodemap->nodes[i].flags & NODE_FLAGS_DISCONNECTED) {
1455 ret = update_flags_on_all_nodes(ctdb, nodemap, i, nodemap->nodes[i].flags);
1457 DEBUG(DEBUG_ERR, (__location__ " Unable to update flags on all nodes for node %d\n", i));
1462 DEBUG(DEBUG_NOTICE, (__location__ " Recovery - updated flags\n"));
1464 /* pick a new generation number */
1465 generation = new_generation();
1467 /* change the vnnmap on this node to use the new generation
1468 number but not on any other nodes.
1469 this guarantees that if we abort the recovery prematurely
1470 for some reason (a node stops responding?)
1471 that we can just return immediately and we will reenter
1472 recovery shortly again.
1473 I.e. we deliberately leave the cluster with an inconsistent
1474 generation id to allow us to abort recovery at any stage and
1475 just restart it from scratch.
1477 vnnmap->generation = generation;
1478 ret = ctdb_ctrl_setvnnmap(ctdb, CONTROL_TIMEOUT(), pnn, mem_ctx, vnnmap);
1480 DEBUG(DEBUG_ERR, (__location__ " Unable to set vnnmap for node %u\n", pnn));
1484 data.dptr = (void *)&generation;
1485 data.dsize = sizeof(uint32_t);
1487 nodes = list_of_active_nodes(ctdb, nodemap, mem_ctx, true);
1488 if (ctdb_client_async_control(ctdb, CTDB_CONTROL_TRANSACTION_START,
1490 CONTROL_TIMEOUT(), false, data,
1492 transaction_start_fail_callback,
1494 DEBUG(DEBUG_ERR, (__location__ " Unable to start transactions. Recovery failed.\n"));
1495 if (ctdb_client_async_control(ctdb, CTDB_CONTROL_TRANSACTION_CANCEL,
1497 CONTROL_TIMEOUT(), false, tdb_null,
1501 DEBUG(DEBUG_ERR,("Failed to cancel recovery transaction\n"));
1506 DEBUG(DEBUG_NOTICE,(__location__ " started transactions on all nodes\n"));
1508 for (i=0;i<dbmap->num;i++) {
1509 ret = recover_database(rec, mem_ctx,
1511 dbmap->dbs[i].persistent,
1512 pnn, nodemap, generation);
1514 DEBUG(DEBUG_ERR, (__location__ " Failed to recover database 0x%x\n", dbmap->dbs[i].dbid));
1519 DEBUG(DEBUG_NOTICE, (__location__ " Recovery - starting database commits\n"));
1521 /* commit all the changes */
1522 if (ctdb_client_async_control(ctdb, CTDB_CONTROL_TRANSACTION_COMMIT,
1524 CONTROL_TIMEOUT(), false, data,
1527 DEBUG(DEBUG_ERR, (__location__ " Unable to commit recovery changes. Recovery failed.\n"));
1531 DEBUG(DEBUG_NOTICE, (__location__ " Recovery - committed databases\n"));
1534 /* update the capabilities for all nodes */
1535 ret = update_capabilities(ctdb, nodemap);
1537 DEBUG(DEBUG_ERR, (__location__ " Unable to update node capabilities.\n"));
1541 /* build a new vnn map with all the currently active and
1543 generation = new_generation();
1544 vnnmap = talloc(mem_ctx, struct ctdb_vnn_map);
1545 CTDB_NO_MEMORY(ctdb, vnnmap);
1546 vnnmap->generation = generation;
1548 vnnmap->map = talloc_zero_array(vnnmap, uint32_t, vnnmap->size);
1549 CTDB_NO_MEMORY(ctdb, vnnmap->map);
1550 for (i=j=0;i<nodemap->num;i++) {
1551 if (nodemap->nodes[i].flags & NODE_FLAGS_INACTIVE) {
1554 if (!(ctdb->nodes[i]->capabilities & CTDB_CAP_LMASTER)) {
1555 /* this node can not be an lmaster */
1556 DEBUG(DEBUG_DEBUG, ("Node %d cant be a LMASTER, skipping it\n", i));
1561 vnnmap->map = talloc_realloc(vnnmap, vnnmap->map, uint32_t, vnnmap->size);
1562 CTDB_NO_MEMORY(ctdb, vnnmap->map);
1563 vnnmap->map[j++] = nodemap->nodes[i].pnn;
1566 if (vnnmap->size == 0) {
1567 DEBUG(DEBUG_NOTICE, ("No suitable lmasters found. Adding local node (recmaster) anyway.\n"));
1569 vnnmap->map = talloc_realloc(vnnmap, vnnmap->map, uint32_t, vnnmap->size);
1570 CTDB_NO_MEMORY(ctdb, vnnmap->map);
1571 vnnmap->map[0] = pnn;
1574 /* update to the new vnnmap on all nodes */
1575 ret = update_vnnmap_on_all_nodes(ctdb, nodemap, pnn, vnnmap, mem_ctx);
1577 DEBUG(DEBUG_ERR, (__location__ " Unable to update vnnmap on all nodes\n"));
1581 DEBUG(DEBUG_NOTICE, (__location__ " Recovery - updated vnnmap\n"));
1583 /* update recmaster to point to us for all nodes */
1584 ret = set_recovery_master(ctdb, nodemap, pnn);
1586 DEBUG(DEBUG_ERR, (__location__ " Unable to set recovery master\n"));
1590 DEBUG(DEBUG_NOTICE, (__location__ " Recovery - updated recmaster\n"));
1593 update all nodes to have the same flags that we have
1595 for (i=0;i<nodemap->num;i++) {
1596 if (nodemap->nodes[i].flags & NODE_FLAGS_DISCONNECTED) {
1600 ret = update_flags_on_all_nodes(ctdb, nodemap, i, nodemap->nodes[i].flags);
1602 DEBUG(DEBUG_ERR, (__location__ " Unable to update flags on all nodes for node %d\n", i));
1607 DEBUG(DEBUG_NOTICE, (__location__ " Recovery - updated flags\n"));
1609 /* disable recovery mode */
1610 ret = set_recovery_mode(ctdb, rec, nodemap, CTDB_RECOVERY_NORMAL);
1612 DEBUG(DEBUG_ERR, (__location__ " Unable to set recovery mode to normal on cluster\n"));
1616 DEBUG(DEBUG_NOTICE, (__location__ " Recovery - disabled recovery mode\n"));
1619 tell nodes to takeover their public IPs
1621 ret = ctdb_reload_remote_public_ips(ctdb, rec, nodemap, &culprit);
1623 DEBUG(DEBUG_ERR,("Failed to read public ips from remote node %d\n",
1627 rec->need_takeover_run = false;
1628 ret = ctdb_takeover_run(ctdb, nodemap);
1630 DEBUG(DEBUG_ERR, (__location__ " Unable to setup public takeover addresses\n"));
1633 DEBUG(DEBUG_NOTICE, (__location__ " Recovery - takeip finished\n"));
1635 /* execute the "recovered" event script on all nodes */
1636 ret = run_recovered_eventscript(ctdb, nodemap, "do_recovery");
1638 DEBUG(DEBUG_ERR, (__location__ " Unable to run the 'recovered' event on cluster. Recovery process failed.\n"));
1642 DEBUG(DEBUG_NOTICE, (__location__ " Recovery - finished the recovered event\n"));
1644 /* send a message to all clients telling them that the cluster
1645 has been reconfigured */
1646 ctdb_client_send_message(ctdb, CTDB_BROADCAST_CONNECTED, CTDB_SRVID_RECONFIGURE, tdb_null);
1648 DEBUG(DEBUG_NOTICE, (__location__ " Recovery complete\n"));
1650 rec->need_recovery = false;
1652 /* we managed to complete a full recovery, make sure to forgive
1653 any past sins by the nodes that could now participate in the
1656 DEBUG(DEBUG_ERR,("Resetting ban count to 0 for all nodes\n"));
1657 for (i=0;i<nodemap->num;i++) {
1658 struct ctdb_banning_state *ban_state;
1660 if (nodemap->nodes[i].flags & NODE_FLAGS_DISCONNECTED) {
1664 ban_state = (struct ctdb_banning_state *)ctdb->nodes[nodemap->nodes[i].pnn]->ban_state;
1665 if (ban_state == NULL) {
1669 ban_state->count = 0;
1673 /* We just finished a recovery successfully.
1674 We now wait for rerecovery_timeout before we allow
1675 another recovery to take place.
1677 DEBUG(DEBUG_NOTICE, (__location__ " New recoveries supressed for the rerecovery timeout\n"));
1678 ctdb_wait_timeout(ctdb, ctdb->tunable.rerecovery_timeout);
1679 DEBUG(DEBUG_NOTICE, (__location__ " Rerecovery timeout elapsed. Recovery reactivated.\n"));
1686 elections are won by first checking the number of connected nodes, then
1687 the priority time, then the pnn
1689 struct election_message {
1690 uint32_t num_connected;
1691 struct timeval priority_time;
1693 uint32_t node_flags;
1697 form this nodes election data
1699 static void ctdb_election_data(struct ctdb_recoverd *rec, struct election_message *em)
1702 struct ctdb_node_map *nodemap;
1703 struct ctdb_context *ctdb = rec->ctdb;
1707 em->pnn = rec->ctdb->pnn;
1708 em->priority_time = rec->priority_time;
1710 ret = ctdb_ctrl_getnodemap(ctdb, CONTROL_TIMEOUT(), CTDB_CURRENT_NODE, rec, &nodemap);
1712 DEBUG(DEBUG_ERR,(__location__ " unable to get election data\n"));
1716 rec->node_flags = nodemap->nodes[ctdb->pnn].flags;
1717 em->node_flags = rec->node_flags;
1719 for (i=0;i<nodemap->num;i++) {
1720 if (!(nodemap->nodes[i].flags & NODE_FLAGS_DISCONNECTED)) {
1721 em->num_connected++;
1725 /* we shouldnt try to win this election if we cant be a recmaster */
1726 if ((ctdb->capabilities & CTDB_CAP_RECMASTER) == 0) {
1727 em->num_connected = 0;
1728 em->priority_time = timeval_current();
1731 talloc_free(nodemap);
1735 see if the given election data wins
1737 static bool ctdb_election_win(struct ctdb_recoverd *rec, struct election_message *em)
1739 struct election_message myem;
1742 ctdb_election_data(rec, &myem);
1744 /* we cant win if we dont have the recmaster capability */
1745 if ((rec->ctdb->capabilities & CTDB_CAP_RECMASTER) == 0) {
1749 /* we cant win if we are banned */
1750 if (rec->node_flags & NODE_FLAGS_BANNED) {
1754 /* we cant win if we are stopped */
1755 if (rec->node_flags & NODE_FLAGS_STOPPED) {
1759 /* we will automatically win if the other node is banned */
1760 if (em->node_flags & NODE_FLAGS_BANNED) {
1764 /* we will automatically win if the other node is banned */
1765 if (em->node_flags & NODE_FLAGS_STOPPED) {
1769 /* try to use the most connected node */
1771 cmp = (int)myem.num_connected - (int)em->num_connected;
1774 /* then the longest running node */
1776 cmp = timeval_compare(&em->priority_time, &myem.priority_time);
1780 cmp = (int)myem.pnn - (int)em->pnn;
1787 send out an election request
1789 static int send_election_request(struct ctdb_recoverd *rec, uint32_t pnn, bool update_recmaster)
1792 TDB_DATA election_data;
1793 struct election_message emsg;
1795 struct ctdb_context *ctdb = rec->ctdb;
1797 srvid = CTDB_SRVID_RECOVERY;
1799 ctdb_election_data(rec, &emsg);
1801 election_data.dsize = sizeof(struct election_message);
1802 election_data.dptr = (unsigned char *)&emsg;
1805 /* send an election message to all active nodes */
1806 DEBUG(DEBUG_INFO,(__location__ " Send election request to all active nodes\n"));
1807 ctdb_client_send_message(ctdb, CTDB_BROADCAST_ALL, srvid, election_data);
1810 /* A new node that is already frozen has entered the cluster.
1811 The existing nodes are not frozen and dont need to be frozen
1812 until the election has ended and we start the actual recovery
1814 if (update_recmaster == true) {
1815 /* first we assume we will win the election and set
1816 recoverymaster to be ourself on the current node
1818 ret = ctdb_ctrl_setrecmaster(ctdb, CONTROL_TIMEOUT(), pnn, pnn);
1820 DEBUG(DEBUG_ERR, (__location__ " failed to send recmaster election request\n"));
1830 this function will unban all nodes in the cluster
1832 static void unban_all_nodes(struct ctdb_context *ctdb)
1835 struct ctdb_node_map *nodemap;
1836 TALLOC_CTX *tmp_ctx = talloc_new(ctdb);
1838 ret = ctdb_ctrl_getnodemap(ctdb, CONTROL_TIMEOUT(), CTDB_CURRENT_NODE, tmp_ctx, &nodemap);
1840 DEBUG(DEBUG_ERR,(__location__ " failed to get nodemap to unban all nodes\n"));
1844 for (i=0;i<nodemap->num;i++) {
1845 if ( (!(nodemap->nodes[i].flags & NODE_FLAGS_DISCONNECTED))
1846 && (nodemap->nodes[i].flags & NODE_FLAGS_BANNED) ) {
1847 ctdb_ctrl_modflags(ctdb, CONTROL_TIMEOUT(), nodemap->nodes[i].pnn, 0, NODE_FLAGS_BANNED);
1851 talloc_free(tmp_ctx);
1856 we think we are winning the election - send a broadcast election request
1858 static void election_send_request(struct event_context *ev, struct timed_event *te, struct timeval t, void *p)
1860 struct ctdb_recoverd *rec = talloc_get_type(p, struct ctdb_recoverd);
1863 ret = send_election_request(rec, ctdb_get_pnn(rec->ctdb), false);
1865 DEBUG(DEBUG_ERR,("Failed to send election request!\n"));
1868 talloc_free(rec->send_election_te);
1869 rec->send_election_te = NULL;
1873 handler for memory dumps
1875 static void mem_dump_handler(struct ctdb_context *ctdb, uint64_t srvid,
1876 TDB_DATA data, void *private_data)
1878 TALLOC_CTX *tmp_ctx = talloc_new(ctdb);
1881 struct rd_memdump_reply *rd;
1883 if (data.dsize != sizeof(struct rd_memdump_reply)) {
1884 DEBUG(DEBUG_ERR, (__location__ " Wrong size of return address.\n"));
1885 talloc_free(tmp_ctx);
1888 rd = (struct rd_memdump_reply *)data.dptr;
1890 dump = talloc_zero(tmp_ctx, TDB_DATA);
1892 DEBUG(DEBUG_ERR, (__location__ " Failed to allocate memory for memdump\n"));
1893 talloc_free(tmp_ctx);
1896 ret = ctdb_dump_memory(ctdb, dump);
1898 DEBUG(DEBUG_ERR, (__location__ " ctdb_dump_memory() failed\n"));
1899 talloc_free(tmp_ctx);
1903 DEBUG(DEBUG_ERR, ("recovery master memory dump\n"));
1905 ret = ctdb_client_send_message(ctdb, rd->pnn, rd->srvid, *dump);
1907 DEBUG(DEBUG_ERR,("Failed to send rd memdump reply message\n"));
1908 talloc_free(tmp_ctx);
1912 talloc_free(tmp_ctx);
1916 handler for reload_nodes
1918 static void reload_nodes_handler(struct ctdb_context *ctdb, uint64_t srvid,
1919 TDB_DATA data, void *private_data)
1921 struct ctdb_recoverd *rec = talloc_get_type(private_data, struct ctdb_recoverd);
1923 DEBUG(DEBUG_ERR, (__location__ " Reload nodes file from recovery daemon\n"));
1925 reload_nodes_file(rec->ctdb);
1929 static void reenable_ip_check(struct event_context *ev, struct timed_event *te,
1930 struct timeval yt, void *p)
1932 struct ctdb_recoverd *rec = talloc_get_type(p, struct ctdb_recoverd);
1934 talloc_free(rec->ip_check_disable_ctx);
1935 rec->ip_check_disable_ctx = NULL;
1939 static void recd_update_ip_handler(struct ctdb_context *ctdb, uint64_t srvid,
1940 TDB_DATA data, void *private_data)
1942 struct ctdb_recoverd *rec = talloc_get_type(private_data, struct ctdb_recoverd);
1943 struct ctdb_public_ip *ip;
1945 if (rec->recmaster != rec->ctdb->pnn) {
1946 DEBUG(DEBUG_INFO,("Not recmaster, ignore update ip message\n"));
1950 if (data.dsize != sizeof(struct ctdb_public_ip)) {
1951 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)));
1955 ip = (struct ctdb_public_ip *)data.dptr;
1957 update_ip_assignment_tree(rec->ctdb, ip);
1961 static void disable_ip_check_handler(struct ctdb_context *ctdb, uint64_t srvid,
1962 TDB_DATA data, void *private_data)
1964 struct ctdb_recoverd *rec = talloc_get_type(private_data, struct ctdb_recoverd);
1967 if (rec->ip_check_disable_ctx != NULL) {
1968 talloc_free(rec->ip_check_disable_ctx);
1969 rec->ip_check_disable_ctx = NULL;
1972 if (data.dsize != sizeof(uint32_t)) {
1973 DEBUG(DEBUG_ERR,(__location__ " Wrong size for data :%lu "
1974 "expexting %lu\n", (long unsigned)data.dsize,
1975 (long unsigned)sizeof(uint32_t)));
1978 if (data.dptr == NULL) {
1979 DEBUG(DEBUG_ERR,(__location__ " No data recaived\n"));
1983 timeout = *((uint32_t *)data.dptr);
1984 DEBUG(DEBUG_NOTICE,("Disabling ip check for %u seconds\n", timeout));
1986 rec->ip_check_disable_ctx = talloc_new(rec);
1987 CTDB_NO_MEMORY_VOID(ctdb, rec->ip_check_disable_ctx);
1989 event_add_timed(ctdb->ev, rec->ip_check_disable_ctx, timeval_current_ofs(timeout, 0), reenable_ip_check, rec);
1994 handler for ip reallocate, just add it to the list of callers and
1995 handle this later in the monitor_cluster loop so we do not recurse
1996 with other callers to takeover_run()
1998 static void ip_reallocate_handler(struct ctdb_context *ctdb, uint64_t srvid,
1999 TDB_DATA data, void *private_data)
2001 struct ctdb_recoverd *rec = talloc_get_type(private_data, struct ctdb_recoverd);
2002 struct ip_reallocate_list *caller;
2004 if (data.dsize != sizeof(struct rd_memdump_reply)) {
2005 DEBUG(DEBUG_ERR, (__location__ " Wrong size of return address.\n"));
2009 if (rec->ip_reallocate_ctx == NULL) {
2010 rec->ip_reallocate_ctx = talloc_new(rec);
2011 CTDB_NO_MEMORY_FATAL(ctdb, rec->ip_reallocate_ctx);
2014 caller = talloc(rec->ip_reallocate_ctx, struct ip_reallocate_list);
2015 CTDB_NO_MEMORY_FATAL(ctdb, caller);
2017 caller->rd = (struct rd_memdump_reply *)talloc_steal(caller, data.dptr);
2018 caller->next = rec->reallocate_callers;
2019 rec->reallocate_callers = caller;
2024 static void process_ipreallocate_requests(struct ctdb_context *ctdb, struct ctdb_recoverd *rec)
2026 TALLOC_CTX *tmp_ctx = talloc_new(ctdb);
2029 struct ip_reallocate_list *callers;
2032 DEBUG(DEBUG_INFO, ("recovery master forced ip reallocation\n"));
2034 /* update the list of public ips that a node can handle for
2037 ret = ctdb_reload_remote_public_ips(ctdb, rec, rec->nodemap, &culprit);
2039 DEBUG(DEBUG_ERR,("Failed to read public ips from remote node %d\n",
2041 rec->need_takeover_run = true;
2044 ret = ctdb_takeover_run(ctdb, rec->nodemap);
2046 DEBUG(DEBUG_ERR,("Failed to read public ips from remote node %d\n",
2048 rec->need_takeover_run = true;
2052 result.dsize = sizeof(int32_t);
2053 result.dptr = (uint8_t *)&ret;
2055 for (callers=rec->reallocate_callers; callers; callers=callers->next) {
2057 /* Someone that sent srvid==0 does not want a reply */
2058 if (callers->rd->srvid == 0) {
2061 DEBUG(DEBUG_INFO,("Sending ip reallocate reply message to "
2062 "%u:%llu\n", (unsigned)callers->rd->pnn,
2063 (unsigned long long)callers->rd->srvid));
2064 ret = ctdb_client_send_message(ctdb, callers->rd->pnn, callers->rd->srvid, result);
2066 DEBUG(DEBUG_ERR,("Failed to send ip reallocate reply "
2067 "message to %u:%llu\n",
2068 (unsigned)callers->rd->pnn,
2069 (unsigned long long)callers->rd->srvid));
2073 talloc_free(tmp_ctx);
2074 talloc_free(rec->ip_reallocate_ctx);
2075 rec->ip_reallocate_ctx = NULL;
2076 rec->reallocate_callers = NULL;
2082 handler for recovery master elections
2084 static void election_handler(struct ctdb_context *ctdb, uint64_t srvid,
2085 TDB_DATA data, void *private_data)
2087 struct ctdb_recoverd *rec = talloc_get_type(private_data, struct ctdb_recoverd);
2089 struct election_message *em = (struct election_message *)data.dptr;
2090 TALLOC_CTX *mem_ctx;
2092 /* we got an election packet - update the timeout for the election */
2093 talloc_free(rec->election_timeout);
2094 rec->election_timeout = event_add_timed(ctdb->ev, ctdb,
2096 timeval_current_ofs(0, 500000) :
2097 timeval_current_ofs(ctdb->tunable.election_timeout, 0),
2098 ctdb_election_timeout, rec);
2100 mem_ctx = talloc_new(ctdb);
2102 /* someone called an election. check their election data
2103 and if we disagree and we would rather be the elected node,
2104 send a new election message to all other nodes
2106 if (ctdb_election_win(rec, em)) {
2107 if (!rec->send_election_te) {
2108 rec->send_election_te = event_add_timed(ctdb->ev, rec,
2109 timeval_current_ofs(0, 500000),
2110 election_send_request, rec);
2112 talloc_free(mem_ctx);
2113 /*unban_all_nodes(ctdb);*/
2118 talloc_free(rec->send_election_te);
2119 rec->send_election_te = NULL;
2121 if (ctdb->tunable.verify_recovery_lock != 0) {
2122 /* release the recmaster lock */
2123 if (em->pnn != ctdb->pnn &&
2124 ctdb->recovery_lock_fd != -1) {
2125 close(ctdb->recovery_lock_fd);
2126 ctdb->recovery_lock_fd = -1;
2127 unban_all_nodes(ctdb);
2131 /* ok, let that guy become recmaster then */
2132 ret = ctdb_ctrl_setrecmaster(ctdb, CONTROL_TIMEOUT(), ctdb_get_pnn(ctdb), em->pnn);
2134 DEBUG(DEBUG_ERR, (__location__ " failed to send recmaster election request"));
2135 talloc_free(mem_ctx);
2139 talloc_free(mem_ctx);
2145 force the start of the election process
2147 static void force_election(struct ctdb_recoverd *rec, uint32_t pnn,
2148 struct ctdb_node_map *nodemap)
2151 struct ctdb_context *ctdb = rec->ctdb;
2153 DEBUG(DEBUG_INFO,(__location__ " Force an election\n"));
2155 /* set all nodes to recovery mode to stop all internode traffic */
2156 ret = set_recovery_mode(ctdb, rec, nodemap, CTDB_RECOVERY_ACTIVE);
2158 DEBUG(DEBUG_ERR, (__location__ " Unable to set recovery mode to active on cluster\n"));
2162 talloc_free(rec->election_timeout);
2163 rec->election_timeout = event_add_timed(ctdb->ev, ctdb,
2165 timeval_current_ofs(0, 500000) :
2166 timeval_current_ofs(ctdb->tunable.election_timeout, 0),
2167 ctdb_election_timeout, rec);
2169 ret = send_election_request(rec, pnn, true);
2171 DEBUG(DEBUG_ERR, (__location__ " failed to initiate recmaster election"));
2175 /* wait for a few seconds to collect all responses */
2176 ctdb_wait_election(rec);
2182 handler for when a node changes its flags
2184 static void monitor_handler(struct ctdb_context *ctdb, uint64_t srvid,
2185 TDB_DATA data, void *private_data)
2188 struct ctdb_node_flag_change *c = (struct ctdb_node_flag_change *)data.dptr;
2189 struct ctdb_node_map *nodemap=NULL;
2190 TALLOC_CTX *tmp_ctx;
2191 uint32_t changed_flags;
2193 struct ctdb_recoverd *rec = talloc_get_type(private_data, struct ctdb_recoverd);
2194 int disabled_flag_changed;
2196 if (data.dsize != sizeof(*c)) {
2197 DEBUG(DEBUG_ERR,(__location__ "Invalid data in ctdb_node_flag_change\n"));
2201 tmp_ctx = talloc_new(ctdb);
2202 CTDB_NO_MEMORY_VOID(ctdb, tmp_ctx);
2204 ret = ctdb_ctrl_getnodemap(ctdb, CONTROL_TIMEOUT(), CTDB_CURRENT_NODE, tmp_ctx, &nodemap);
2206 DEBUG(DEBUG_ERR,(__location__ "ctdb_ctrl_getnodemap failed in monitor_handler\n"));
2207 talloc_free(tmp_ctx);
2212 for (i=0;i<nodemap->num;i++) {
2213 if (nodemap->nodes[i].pnn == c->pnn) break;
2216 if (i == nodemap->num) {
2217 DEBUG(DEBUG_CRIT,(__location__ "Flag change for non-existant node %u\n", c->pnn));
2218 talloc_free(tmp_ctx);
2222 changed_flags = c->old_flags ^ c->new_flags;
2224 if (nodemap->nodes[i].flags != c->new_flags) {
2225 DEBUG(DEBUG_NOTICE,("Node %u has changed flags - now 0x%x was 0x%x\n", c->pnn, c->new_flags, c->old_flags));
2228 disabled_flag_changed = (nodemap->nodes[i].flags ^ c->new_flags) & NODE_FLAGS_DISABLED;
2230 nodemap->nodes[i].flags = c->new_flags;
2232 ret = ctdb_ctrl_getrecmaster(ctdb, tmp_ctx, CONTROL_TIMEOUT(),
2233 CTDB_CURRENT_NODE, &ctdb->recovery_master);
2236 ret = ctdb_ctrl_getrecmode(ctdb, tmp_ctx, CONTROL_TIMEOUT(),
2237 CTDB_CURRENT_NODE, &ctdb->recovery_mode);
2241 ctdb->recovery_master == ctdb->pnn &&
2242 ctdb->recovery_mode == CTDB_RECOVERY_NORMAL) {
2243 /* Only do the takeover run if the perm disabled or unhealthy
2244 flags changed since these will cause an ip failover but not
2246 If the node became disconnected or banned this will also
2247 lead to an ip address failover but that is handled
2250 if (disabled_flag_changed) {
2251 rec->need_takeover_run = true;
2255 talloc_free(tmp_ctx);
2259 handler for when we need to push out flag changes ot all other nodes
2261 static void push_flags_handler(struct ctdb_context *ctdb, uint64_t srvid,
2262 TDB_DATA data, void *private_data)
2265 struct ctdb_node_flag_change *c = (struct ctdb_node_flag_change *)data.dptr;
2266 struct ctdb_node_map *nodemap=NULL;
2267 TALLOC_CTX *tmp_ctx = talloc_new(ctdb);
2271 /* find the recovery master */
2272 ret = ctdb_ctrl_getrecmaster(ctdb, tmp_ctx, CONTROL_TIMEOUT(), CTDB_CURRENT_NODE, &recmaster);
2274 DEBUG(DEBUG_ERR, (__location__ " Unable to get recmaster from local node\n"));
2275 talloc_free(tmp_ctx);
2279 /* read the node flags from the recmaster */
2280 ret = ctdb_ctrl_getnodemap(ctdb, CONTROL_TIMEOUT(), recmaster, tmp_ctx, &nodemap);
2282 DEBUG(DEBUG_ERR, (__location__ " Unable to get nodemap from node %u\n", c->pnn));
2283 talloc_free(tmp_ctx);
2286 if (c->pnn >= nodemap->num) {
2287 DEBUG(DEBUG_ERR,(__location__ " Nodemap from recmaster does not contain node %d\n", c->pnn));
2288 talloc_free(tmp_ctx);
2292 /* send the flags update to all connected nodes */
2293 nodes = list_of_connected_nodes(ctdb, nodemap, tmp_ctx, true);
2295 if (ctdb_client_async_control(ctdb, CTDB_CONTROL_MODIFY_FLAGS,
2296 nodes, 0, CONTROL_TIMEOUT(),
2300 DEBUG(DEBUG_ERR, (__location__ " ctdb_control to modify node flags failed\n"));
2302 talloc_free(tmp_ctx);
2306 talloc_free(tmp_ctx);
2310 struct verify_recmode_normal_data {
2312 enum monitor_result status;
2315 static void verify_recmode_normal_callback(struct ctdb_client_control_state *state)
2317 struct verify_recmode_normal_data *rmdata = talloc_get_type(state->async.private_data, struct verify_recmode_normal_data);
2320 /* one more node has responded with recmode data*/
2323 /* if we failed to get the recmode, then return an error and let
2324 the main loop try again.
2326 if (state->state != CTDB_CONTROL_DONE) {
2327 if (rmdata->status == MONITOR_OK) {
2328 rmdata->status = MONITOR_FAILED;
2333 /* if we got a response, then the recmode will be stored in the
2336 if (state->status != CTDB_RECOVERY_NORMAL) {
2337 DEBUG(DEBUG_NOTICE, (__location__ " Node:%u was in recovery mode. Restart recovery process\n", state->c->hdr.destnode));
2338 rmdata->status = MONITOR_RECOVERY_NEEDED;
2345 /* verify that all nodes are in normal recovery mode */
2346 static enum monitor_result verify_recmode(struct ctdb_context *ctdb, struct ctdb_node_map *nodemap)
2348 struct verify_recmode_normal_data *rmdata;
2349 TALLOC_CTX *mem_ctx = talloc_new(ctdb);
2350 struct ctdb_client_control_state *state;
2351 enum monitor_result status;
2354 rmdata = talloc(mem_ctx, struct verify_recmode_normal_data);
2355 CTDB_NO_MEMORY_FATAL(ctdb, rmdata);
2357 rmdata->status = MONITOR_OK;
2359 /* loop over all active nodes and send an async getrecmode call to
2361 for (j=0; j<nodemap->num; j++) {
2362 if (nodemap->nodes[j].flags & NODE_FLAGS_INACTIVE) {
2365 state = ctdb_ctrl_getrecmode_send(ctdb, mem_ctx,
2367 nodemap->nodes[j].pnn);
2368 if (state == NULL) {
2369 /* we failed to send the control, treat this as
2370 an error and try again next iteration
2372 DEBUG(DEBUG_ERR,("Failed to call ctdb_ctrl_getrecmode_send during monitoring\n"));
2373 talloc_free(mem_ctx);
2374 return MONITOR_FAILED;
2377 /* set up the callback functions */
2378 state->async.fn = verify_recmode_normal_callback;
2379 state->async.private_data = rmdata;
2381 /* one more control to wait for to complete */
2386 /* now wait for up to the maximum number of seconds allowed
2387 or until all nodes we expect a response from has replied
2389 while (rmdata->count > 0) {
2390 event_loop_once(ctdb->ev);
2393 status = rmdata->status;
2394 talloc_free(mem_ctx);
2399 struct verify_recmaster_data {
2400 struct ctdb_recoverd *rec;
2403 enum monitor_result status;
2406 static void verify_recmaster_callback(struct ctdb_client_control_state *state)
2408 struct verify_recmaster_data *rmdata = talloc_get_type(state->async.private_data, struct verify_recmaster_data);
2411 /* one more node has responded with recmaster data*/
2414 /* if we failed to get the recmaster, then return an error and let
2415 the main loop try again.
2417 if (state->state != CTDB_CONTROL_DONE) {
2418 if (rmdata->status == MONITOR_OK) {
2419 rmdata->status = MONITOR_FAILED;
2424 /* if we got a response, then the recmaster will be stored in the
2427 if (state->status != rmdata->pnn) {
2428 DEBUG(DEBUG_ERR,("Node %d does not agree we are the recmaster. Need a new recmaster election\n", state->c->hdr.destnode));
2429 ctdb_set_culprit(rmdata->rec, state->c->hdr.destnode);
2430 rmdata->status = MONITOR_ELECTION_NEEDED;
2437 /* verify that all nodes agree that we are the recmaster */
2438 static enum monitor_result verify_recmaster(struct ctdb_recoverd *rec, struct ctdb_node_map *nodemap, uint32_t pnn)
2440 struct ctdb_context *ctdb = rec->ctdb;
2441 struct verify_recmaster_data *rmdata;
2442 TALLOC_CTX *mem_ctx = talloc_new(ctdb);
2443 struct ctdb_client_control_state *state;
2444 enum monitor_result status;
2447 rmdata = talloc(mem_ctx, struct verify_recmaster_data);
2448 CTDB_NO_MEMORY_FATAL(ctdb, rmdata);
2452 rmdata->status = MONITOR_OK;
2454 /* loop over all active nodes and send an async getrecmaster call to
2456 for (j=0; j<nodemap->num; j++) {
2457 if (nodemap->nodes[j].flags & NODE_FLAGS_INACTIVE) {
2460 state = ctdb_ctrl_getrecmaster_send(ctdb, mem_ctx,
2462 nodemap->nodes[j].pnn);
2463 if (state == NULL) {
2464 /* we failed to send the control, treat this as
2465 an error and try again next iteration
2467 DEBUG(DEBUG_ERR,("Failed to call ctdb_ctrl_getrecmaster_send during monitoring\n"));
2468 talloc_free(mem_ctx);
2469 return MONITOR_FAILED;
2472 /* set up the callback functions */
2473 state->async.fn = verify_recmaster_callback;
2474 state->async.private_data = rmdata;
2476 /* one more control to wait for to complete */
2481 /* now wait for up to the maximum number of seconds allowed
2482 or until all nodes we expect a response from has replied
2484 while (rmdata->count > 0) {
2485 event_loop_once(ctdb->ev);
2488 status = rmdata->status;
2489 talloc_free(mem_ctx);
2494 /* called to check that the local allocation of public ip addresses is ok.
2496 static int verify_local_ip_allocation(struct ctdb_context *ctdb, struct ctdb_recoverd *rec, uint32_t pnn)
2498 TALLOC_CTX *mem_ctx = talloc_new(NULL);
2499 struct ctdb_control_get_ifaces *ifaces = NULL;
2500 struct ctdb_all_public_ips *ips = NULL;
2501 struct ctdb_uptime *uptime1 = NULL;
2502 struct ctdb_uptime *uptime2 = NULL;
2504 bool need_iface_check = false;
2505 bool need_takeover_run = false;
2507 ret = ctdb_ctrl_uptime(ctdb, mem_ctx, CONTROL_TIMEOUT(),
2508 CTDB_CURRENT_NODE, &uptime1);
2510 DEBUG(DEBUG_ERR, ("Unable to get uptime from local node %u\n", pnn));
2511 talloc_free(mem_ctx);
2516 /* read the interfaces from the local node */
2517 ret = ctdb_ctrl_get_ifaces(ctdb, CONTROL_TIMEOUT(), CTDB_CURRENT_NODE, mem_ctx, &ifaces);
2519 DEBUG(DEBUG_ERR, ("Unable to get interfaces from local node %u\n", pnn));
2520 talloc_free(mem_ctx);
2525 need_iface_check = true;
2526 } else if (rec->ifaces->num != ifaces->num) {
2527 need_iface_check = true;
2528 } else if (memcmp(rec->ifaces, ifaces, talloc_get_size(ifaces)) != 0) {
2529 need_iface_check = true;
2532 if (need_iface_check) {
2533 DEBUG(DEBUG_NOTICE, ("The interfaces status has changed on "
2534 "local node %u - force takeover run\n",
2536 need_takeover_run = true;
2539 /* read the ip allocation from the local node */
2540 ret = ctdb_ctrl_get_public_ips(ctdb, CONTROL_TIMEOUT(), CTDB_CURRENT_NODE, mem_ctx, &ips);
2542 DEBUG(DEBUG_ERR, ("Unable to get public ips from local node %u\n", pnn));
2543 talloc_free(mem_ctx);
2547 ret = ctdb_ctrl_uptime(ctdb, mem_ctx, CONTROL_TIMEOUT(),
2548 CTDB_CURRENT_NODE, &uptime2);
2550 DEBUG(DEBUG_ERR, ("Unable to get uptime from local node %u\n", pnn));
2551 talloc_free(mem_ctx);
2555 /* skip the check if the startrecovery time has changed */
2556 if (timeval_compare(&uptime1->last_recovery_started,
2557 &uptime2->last_recovery_started) != 0) {
2558 DEBUG(DEBUG_NOTICE, (__location__ " last recovery time changed while we read the public ip list. skipping public ip address check\n"));
2559 talloc_free(mem_ctx);
2563 /* skip the check if the endrecovery time has changed */
2564 if (timeval_compare(&uptime1->last_recovery_finished,
2565 &uptime2->last_recovery_finished) != 0) {
2566 DEBUG(DEBUG_NOTICE, (__location__ " last recovery time changed while we read the public ip list. skipping public ip address check\n"));
2567 talloc_free(mem_ctx);
2571 /* skip the check if we have started but not finished recovery */
2572 if (timeval_compare(&uptime1->last_recovery_finished,
2573 &uptime1->last_recovery_started) != 1) {
2574 DEBUG(DEBUG_NOTICE, (__location__ " in the middle of recovery or ip reallocation. skipping public ip address check\n"));
2575 talloc_free(mem_ctx);
2580 talloc_free(rec->ifaces);
2581 rec->ifaces = talloc_steal(rec, ifaces);
2583 /* verify that we have the ip addresses we should have
2584 and we dont have ones we shouldnt have.
2585 if we find an inconsistency we set recmode to
2586 active on the local node and wait for the recmaster
2587 to do a full blown recovery
2589 for (j=0; j<ips->num; j++) {
2590 if (ips->ips[j].pnn == pnn) {
2591 if (!ctdb_sys_have_ip(&ips->ips[j].addr)) {
2592 DEBUG(DEBUG_CRIT,("Public address '%s' is missing and we should serve this ip\n",
2593 ctdb_addr_to_str(&ips->ips[j].addr)));
2594 need_takeover_run = true;
2597 if (ctdb_sys_have_ip(&ips->ips[j].addr)) {
2598 DEBUG(DEBUG_CRIT,("We are still serving a public address '%s' that we should not be serving.\n",
2599 ctdb_addr_to_str(&ips->ips[j].addr)));
2600 need_takeover_run = true;
2605 if (need_takeover_run) {
2606 struct takeover_run_reply rd;
2609 DEBUG(DEBUG_CRIT,("Trigger takeoverrun\n"));
2613 data.dptr = (uint8_t *)&rd;
2614 data.dsize = sizeof(rd);
2616 ret = ctdb_client_send_message(ctdb, rec->recmaster, CTDB_SRVID_TAKEOVER_RUN, data);
2618 DEBUG(DEBUG_ERR,(__location__ " Failed to send ipreallocate to recmaster :%d\n", (int)rec->recmaster));
2621 talloc_free(mem_ctx);
2626 static void async_getnodemap_callback(struct ctdb_context *ctdb, uint32_t node_pnn, int32_t res, TDB_DATA outdata, void *callback_data)
2628 struct ctdb_node_map **remote_nodemaps = callback_data;
2630 if (node_pnn >= ctdb->num_nodes) {
2631 DEBUG(DEBUG_ERR,(__location__ " pnn from invalid node\n"));
2635 remote_nodemaps[node_pnn] = (struct ctdb_node_map *)talloc_steal(remote_nodemaps, outdata.dptr);
2639 static int get_remote_nodemaps(struct ctdb_context *ctdb, TALLOC_CTX *mem_ctx,
2640 struct ctdb_node_map *nodemap,
2641 struct ctdb_node_map **remote_nodemaps)
2645 nodes = list_of_active_nodes(ctdb, nodemap, mem_ctx, true);
2646 if (ctdb_client_async_control(ctdb, CTDB_CONTROL_GET_NODEMAP,
2648 CONTROL_TIMEOUT(), false, tdb_null,
2649 async_getnodemap_callback,
2651 remote_nodemaps) != 0) {
2652 DEBUG(DEBUG_ERR, (__location__ " Unable to pull all remote nodemaps\n"));
2660 enum reclock_child_status { RECLOCK_CHECKING, RECLOCK_OK, RECLOCK_FAILED, RECLOCK_TIMEOUT};
2661 struct ctdb_check_reclock_state {
2662 struct ctdb_context *ctdb;
2663 struct timeval start_time;
2666 struct timed_event *te;
2667 struct fd_event *fde;
2668 enum reclock_child_status status;
2671 /* when we free the reclock state we must kill any child process.
2673 static int check_reclock_destructor(struct ctdb_check_reclock_state *state)
2675 struct ctdb_context *ctdb = state->ctdb;
2677 ctdb_ctrl_report_recd_lock_latency(ctdb, CONTROL_TIMEOUT(), timeval_elapsed(&state->start_time));
2679 if (state->fd[0] != -1) {
2680 close(state->fd[0]);
2683 if (state->fd[1] != -1) {
2684 close(state->fd[1]);
2687 kill(state->child, SIGKILL);
2692 called if our check_reclock child times out. this would happen if
2693 i/o to the reclock file blocks.
2695 static void ctdb_check_reclock_timeout(struct event_context *ev, struct timed_event *te,
2696 struct timeval t, void *private_data)
2698 struct ctdb_check_reclock_state *state = talloc_get_type(private_data,
2699 struct ctdb_check_reclock_state);
2701 DEBUG(DEBUG_ERR,(__location__ " check_reclock child process hung/timedout CFS slow to grant locks?\n"));
2702 state->status = RECLOCK_TIMEOUT;
2705 /* this is called when the child process has completed checking the reclock
2706 file and has written data back to us through the pipe.
2708 static void reclock_child_handler(struct event_context *ev, struct fd_event *fde,
2709 uint16_t flags, void *private_data)
2711 struct ctdb_check_reclock_state *state= talloc_get_type(private_data,
2712 struct ctdb_check_reclock_state);
2716 /* we got a response from our child process so we can abort the
2719 talloc_free(state->te);
2722 ret = read(state->fd[0], &c, 1);
2723 if (ret != 1 || c != RECLOCK_OK) {
2724 DEBUG(DEBUG_ERR,(__location__ " reclock child process returned error %d\n", c));
2725 state->status = RECLOCK_FAILED;
2730 state->status = RECLOCK_OK;
2734 static int check_recovery_lock(struct ctdb_context *ctdb)
2737 struct ctdb_check_reclock_state *state;
2738 pid_t parent = getpid();
2740 if (ctdb->recovery_lock_fd == -1) {
2741 DEBUG(DEBUG_CRIT,("recovery master doesn't have the recovery lock\n"));
2745 state = talloc(ctdb, struct ctdb_check_reclock_state);
2746 CTDB_NO_MEMORY(ctdb, state);
2749 state->start_time = timeval_current();
2750 state->status = RECLOCK_CHECKING;
2754 ret = pipe(state->fd);
2757 DEBUG(DEBUG_CRIT,(__location__ " Failed to open pipe for check_reclock child\n"));
2761 state->child = fork();
2762 if (state->child == (pid_t)-1) {
2763 DEBUG(DEBUG_CRIT,(__location__ " fork() failed in check_reclock child\n"));
2764 close(state->fd[0]);
2766 close(state->fd[1]);
2772 if (state->child == 0) {
2773 char cc = RECLOCK_OK;
2774 close(state->fd[0]);
2777 if (pread(ctdb->recovery_lock_fd, &cc, 1, 0) == -1) {
2778 DEBUG(DEBUG_CRIT,("failed read from recovery_lock_fd - %s\n", strerror(errno)));
2779 cc = RECLOCK_FAILED;
2782 write(state->fd[1], &cc, 1);
2783 /* make sure we die when our parent dies */
2784 while (kill(parent, 0) == 0 || errno != ESRCH) {
2786 write(state->fd[1], &cc, 1);
2790 close(state->fd[1]);
2792 set_close_on_exec(state->fd[0]);
2794 DEBUG(DEBUG_DEBUG, (__location__ " Created PIPE FD:%d for check_recovery_lock\n", state->fd[0]));
2796 talloc_set_destructor(state, check_reclock_destructor);
2798 state->te = event_add_timed(ctdb->ev, state, timeval_current_ofs(15, 0),
2799 ctdb_check_reclock_timeout, state);
2800 if (state->te == NULL) {
2801 DEBUG(DEBUG_CRIT,(__location__ " Failed to create a timed event for reclock child\n"));
2806 state->fde = event_add_fd(ctdb->ev, state, state->fd[0],
2807 EVENT_FD_READ|EVENT_FD_AUTOCLOSE,
2808 reclock_child_handler,
2811 if (state->fde == NULL) {
2812 DEBUG(DEBUG_CRIT,(__location__ " Failed to create an fd event for reclock child\n"));
2817 while (state->status == RECLOCK_CHECKING) {
2818 event_loop_once(ctdb->ev);
2821 if (state->status == RECLOCK_FAILED) {
2822 DEBUG(DEBUG_ERR,(__location__ " reclock child failed when checking file\n"));
2823 close(ctdb->recovery_lock_fd);
2824 ctdb->recovery_lock_fd = -1;
2833 static int update_recovery_lock_file(struct ctdb_context *ctdb)
2835 TALLOC_CTX *tmp_ctx = talloc_new(NULL);
2836 const char *reclockfile;
2838 if (ctdb_ctrl_getreclock(ctdb, CONTROL_TIMEOUT(), CTDB_CURRENT_NODE, tmp_ctx, &reclockfile) != 0) {
2839 DEBUG(DEBUG_ERR,("Failed to read reclock file from daemon\n"));
2840 talloc_free(tmp_ctx);
2844 if (reclockfile == NULL) {
2845 if (ctdb->recovery_lock_file != NULL) {
2846 DEBUG(DEBUG_ERR,("Reclock file disabled\n"));
2847 talloc_free(ctdb->recovery_lock_file);
2848 ctdb->recovery_lock_file = NULL;
2849 if (ctdb->recovery_lock_fd != -1) {
2850 close(ctdb->recovery_lock_fd);
2851 ctdb->recovery_lock_fd = -1;
2854 ctdb->tunable.verify_recovery_lock = 0;
2855 talloc_free(tmp_ctx);
2859 if (ctdb->recovery_lock_file == NULL) {
2860 ctdb->recovery_lock_file = talloc_strdup(ctdb, reclockfile);
2861 if (ctdb->recovery_lock_fd != -1) {
2862 close(ctdb->recovery_lock_fd);
2863 ctdb->recovery_lock_fd = -1;
2865 talloc_free(tmp_ctx);
2870 if (!strcmp(reclockfile, ctdb->recovery_lock_file)) {
2871 talloc_free(tmp_ctx);
2875 talloc_free(ctdb->recovery_lock_file);
2876 ctdb->recovery_lock_file = talloc_strdup(ctdb, reclockfile);
2877 ctdb->tunable.verify_recovery_lock = 0;
2878 if (ctdb->recovery_lock_fd != -1) {
2879 close(ctdb->recovery_lock_fd);
2880 ctdb->recovery_lock_fd = -1;
2883 talloc_free(tmp_ctx);
2887 static void main_loop(struct ctdb_context *ctdb, struct ctdb_recoverd *rec,
2888 TALLOC_CTX *mem_ctx)
2891 struct ctdb_node_map *nodemap=NULL;
2892 struct ctdb_node_map *recmaster_nodemap=NULL;
2893 struct ctdb_node_map **remote_nodemaps=NULL;
2894 struct ctdb_vnn_map *vnnmap=NULL;
2895 struct ctdb_vnn_map *remote_vnnmap=NULL;
2896 int32_t debug_level;
2901 /* verify that the main daemon is still running */
2902 if (kill(ctdb->ctdbd_pid, 0) != 0) {
2903 DEBUG(DEBUG_CRIT,("CTDB daemon is no longer available. Shutting down recovery daemon\n"));
2907 /* ping the local daemon to tell it we are alive */
2908 ctdb_ctrl_recd_ping(ctdb);
2910 if (rec->election_timeout) {
2911 /* an election is in progress */
2915 /* read the debug level from the parent and update locally */
2916 ret = ctdb_ctrl_get_debuglevel(ctdb, CTDB_CURRENT_NODE, &debug_level);
2918 DEBUG(DEBUG_ERR, (__location__ " Failed to read debuglevel from parent\n"));
2921 LogLevel = debug_level;
2924 /* We must check if we need to ban a node here but we want to do this
2925 as early as possible so we dont wait until we have pulled the node
2926 map from the local node. thats why we have the hardcoded value 20
2928 for (i=0; i<ctdb->num_nodes; i++) {
2929 struct ctdb_banning_state *ban_state;
2931 if (ctdb->nodes[i]->ban_state == NULL) {
2934 ban_state = (struct ctdb_banning_state *)ctdb->nodes[i]->ban_state;
2935 if (ban_state->count < 20) {
2938 DEBUG(DEBUG_NOTICE,("Node %u has caused %u recoveries recently - banning it for %u seconds\n",
2939 ctdb->nodes[i]->pnn, ban_state->count,
2940 ctdb->tunable.recovery_ban_period));
2941 ctdb_ban_node(rec, ctdb->nodes[i]->pnn, ctdb->tunable.recovery_ban_period);
2942 ban_state->count = 0;
2945 /* get relevant tunables */
2946 ret = ctdb_ctrl_get_all_tunables(ctdb, CONTROL_TIMEOUT(), CTDB_CURRENT_NODE, &ctdb->tunable);
2948 DEBUG(DEBUG_ERR,("Failed to get tunables - retrying\n"));
2952 /* get the current recovery lock file from the server */
2953 if (update_recovery_lock_file(ctdb) != 0) {
2954 DEBUG(DEBUG_ERR,("Failed to update the recovery lock file\n"));
2958 /* Make sure that if recovery lock verification becomes disabled when
2961 if (ctdb->tunable.verify_recovery_lock == 0) {
2962 if (ctdb->recovery_lock_fd != -1) {
2963 close(ctdb->recovery_lock_fd);
2964 ctdb->recovery_lock_fd = -1;
2968 pnn = ctdb_ctrl_getpnn(ctdb, CONTROL_TIMEOUT(), CTDB_CURRENT_NODE);
2969 if (pnn == (uint32_t)-1) {
2970 DEBUG(DEBUG_ERR,("Failed to get local pnn - retrying\n"));
2974 /* get the vnnmap */
2975 ret = ctdb_ctrl_getvnnmap(ctdb, CONTROL_TIMEOUT(), pnn, mem_ctx, &vnnmap);
2977 DEBUG(DEBUG_ERR, (__location__ " Unable to get vnnmap from node %u\n", pnn));
2982 /* get number of nodes */
2984 talloc_free(rec->nodemap);
2985 rec->nodemap = NULL;
2988 ret = ctdb_ctrl_getnodemap(ctdb, CONTROL_TIMEOUT(), pnn, rec, &rec->nodemap);
2990 DEBUG(DEBUG_ERR, (__location__ " Unable to get nodemap from node %u\n", pnn));
2993 nodemap = rec->nodemap;
2995 /* check which node is the recovery master */
2996 ret = ctdb_ctrl_getrecmaster(ctdb, mem_ctx, CONTROL_TIMEOUT(), pnn, &rec->recmaster);
2998 DEBUG(DEBUG_ERR, (__location__ " Unable to get recmaster from node %u\n", pnn));
3002 /* if we are not the recmaster we can safely ignore any ip reallocate requests */
3003 if (rec->recmaster != pnn) {
3004 if (rec->ip_reallocate_ctx != NULL) {
3005 talloc_free(rec->ip_reallocate_ctx);
3006 rec->ip_reallocate_ctx = NULL;
3007 rec->reallocate_callers = NULL;
3010 /* if there are takeovers requested, perform it and notify the waiters */
3011 if (rec->reallocate_callers) {
3012 process_ipreallocate_requests(ctdb, rec);
3015 if (rec->recmaster == (uint32_t)-1) {
3016 DEBUG(DEBUG_NOTICE,(__location__ " Initial recovery master set - forcing election\n"));
3017 force_election(rec, pnn, nodemap);
3022 /* if the local daemon is STOPPED, we verify that the databases are
3023 also frozen and thet the recmode is set to active
3025 if (nodemap->nodes[pnn].flags & NODE_FLAGS_STOPPED) {
3026 ret = ctdb_ctrl_getrecmode(ctdb, mem_ctx, CONTROL_TIMEOUT(), CTDB_CURRENT_NODE, &ctdb->recovery_mode);
3028 DEBUG(DEBUG_ERR,(__location__ " Failed to read recmode from local node\n"));
3030 if (ctdb->recovery_mode == CTDB_RECOVERY_NORMAL) {
3031 DEBUG(DEBUG_ERR,("Node is stopped but recovery mode is not active. Activate recovery mode and lock databases\n"));
3033 ret = ctdb_ctrl_freeze_priority(ctdb, CONTROL_TIMEOUT(), CTDB_CURRENT_NODE, 1);
3035 DEBUG(DEBUG_ERR,(__location__ " Failed to freeze node due to node being STOPPED\n"));
3038 ret = ctdb_ctrl_setrecmode(ctdb, CONTROL_TIMEOUT(), CTDB_CURRENT_NODE, CTDB_RECOVERY_ACTIVE);
3040 DEBUG(DEBUG_ERR,(__location__ " Failed to activate recovery mode due to node being stopped\n"));
3047 /* If the local node is stopped, verify we are not the recmaster
3048 and yield this role if so
3050 if ((nodemap->nodes[pnn].flags & NODE_FLAGS_STOPPED) && (rec->recmaster == pnn)) {
3051 DEBUG(DEBUG_ERR,("Local node is STOPPED. Yielding recmaster role\n"));
3052 force_election(rec, pnn, nodemap);
3056 /* check that we (recovery daemon) and the local ctdb daemon
3057 agrees on whether we are banned or not
3061 /* remember our own node flags */
3062 rec->node_flags = nodemap->nodes[pnn].flags;
3064 /* count how many active nodes there are */
3065 rec->num_active = 0;
3066 rec->num_connected = 0;
3067 for (i=0; i<nodemap->num; i++) {
3068 if (!(nodemap->nodes[i].flags & NODE_FLAGS_INACTIVE)) {
3071 if (!(nodemap->nodes[i].flags & NODE_FLAGS_DISCONNECTED)) {
3072 rec->num_connected++;
3077 /* verify that the recmaster node is still active */
3078 for (j=0; j<nodemap->num; j++) {
3079 if (nodemap->nodes[j].pnn==rec->recmaster) {
3084 if (j == nodemap->num) {
3085 DEBUG(DEBUG_ERR, ("Recmaster node %u not in list. Force reelection\n", rec->recmaster));
3086 force_election(rec, pnn, nodemap);
3090 /* if recovery master is disconnected we must elect a new recmaster */
3091 if (nodemap->nodes[j].flags & NODE_FLAGS_DISCONNECTED) {
3092 DEBUG(DEBUG_NOTICE, ("Recmaster node %u is disconnected. Force reelection\n", nodemap->nodes[j].pnn));
3093 force_election(rec, pnn, nodemap);
3097 /* grap the nodemap from the recovery master to check if it is banned */
3098 ret = ctdb_ctrl_getnodemap(ctdb, CONTROL_TIMEOUT(), nodemap->nodes[j].pnn,
3099 mem_ctx, &recmaster_nodemap);
3101 DEBUG(DEBUG_ERR, (__location__ " Unable to get nodemap from recovery master %u\n",
3102 nodemap->nodes[j].pnn));
3107 if (recmaster_nodemap->nodes[j].flags & NODE_FLAGS_INACTIVE) {
3108 DEBUG(DEBUG_NOTICE, ("Recmaster node %u no longer available. Force reelection\n", nodemap->nodes[j].pnn));
3109 force_election(rec, pnn, nodemap);
3114 /* verify that we have all ip addresses we should have and we dont
3115 * have addresses we shouldnt have.
3117 if (ctdb->do_checkpublicip) {
3118 if (rec->ip_check_disable_ctx == NULL) {
3119 if (verify_local_ip_allocation(ctdb, rec, pnn) != 0) {
3120 DEBUG(DEBUG_ERR, (__location__ " Public IPs were inconsistent.\n"));
3126 /* if we are not the recmaster then we do not need to check
3127 if recovery is needed
3129 if (pnn != rec->recmaster) {
3134 /* ensure our local copies of flags are right */
3135 ret = update_local_flags(rec, nodemap);
3136 if (ret == MONITOR_ELECTION_NEEDED) {
3137 DEBUG(DEBUG_NOTICE,("update_local_flags() called for a re-election.\n"));
3138 force_election(rec, pnn, nodemap);
3141 if (ret != MONITOR_OK) {
3142 DEBUG(DEBUG_ERR,("Unable to update local flags\n"));
3146 if (ctdb->num_nodes != nodemap->num) {
3147 DEBUG(DEBUG_ERR, (__location__ " ctdb->num_nodes (%d) != nodemap->num (%d) reloading nodes file\n", ctdb->num_nodes, nodemap->num));
3148 reload_nodes_file(ctdb);
3152 /* verify that all active nodes agree that we are the recmaster */
3153 switch (verify_recmaster(rec, nodemap, pnn)) {
3154 case MONITOR_RECOVERY_NEEDED:
3155 /* can not happen */
3157 case MONITOR_ELECTION_NEEDED:
3158 force_election(rec, pnn, nodemap);
3162 case MONITOR_FAILED:
3167 if (rec->need_recovery) {
3168 /* a previous recovery didn't finish */
3169 do_recovery(rec, mem_ctx, pnn, nodemap, vnnmap);
3173 /* verify that all active nodes are in normal mode
3174 and not in recovery mode
3176 switch (verify_recmode(ctdb, nodemap)) {
3177 case MONITOR_RECOVERY_NEEDED:
3178 do_recovery(rec, mem_ctx, pnn, nodemap, vnnmap);
3180 case MONITOR_FAILED:
3182 case MONITOR_ELECTION_NEEDED:
3183 /* can not happen */
3189 if (ctdb->tunable.verify_recovery_lock != 0) {
3190 /* we should have the reclock - check its not stale */
3191 ret = check_recovery_lock(ctdb);
3193 DEBUG(DEBUG_ERR,("Failed check_recovery_lock. Force a recovery\n"));
3194 ctdb_set_culprit(rec, ctdb->pnn);
3195 do_recovery(rec, mem_ctx, pnn, nodemap, vnnmap);
3200 /* get the nodemap for all active remote nodes
3202 remote_nodemaps = talloc_array(mem_ctx, struct ctdb_node_map *, nodemap->num);
3203 if (remote_nodemaps == NULL) {
3204 DEBUG(DEBUG_ERR, (__location__ " failed to allocate remote nodemap array\n"));
3207 for(i=0; i<nodemap->num; i++) {
3208 remote_nodemaps[i] = NULL;
3210 if (get_remote_nodemaps(ctdb, mem_ctx, nodemap, remote_nodemaps) != 0) {
3211 DEBUG(DEBUG_ERR,(__location__ " Failed to read remote nodemaps\n"));
3215 /* verify that all other nodes have the same nodemap as we have
3217 for (j=0; j<nodemap->num; j++) {
3218 if (nodemap->nodes[j].flags & NODE_FLAGS_INACTIVE) {
3222 if (remote_nodemaps[j] == NULL) {
3223 DEBUG(DEBUG_ERR,(__location__ " Did not get a remote nodemap for node %d, restarting monitoring\n", j));
3224 ctdb_set_culprit(rec, j);
3229 /* if the nodes disagree on how many nodes there are
3230 then this is a good reason to try recovery
3232 if (remote_nodemaps[j]->num != nodemap->num) {
3233 DEBUG(DEBUG_ERR, (__location__ " Remote node:%u has different node count. %u vs %u of the local node\n",
3234 nodemap->nodes[j].pnn, remote_nodemaps[j]->num, nodemap->num));
3235 ctdb_set_culprit(rec, nodemap->nodes[j].pnn);
3236 do_recovery(rec, mem_ctx, pnn, nodemap, vnnmap);
3240 /* if the nodes disagree on which nodes exist and are
3241 active, then that is also a good reason to do recovery
3243 for (i=0;i<nodemap->num;i++) {
3244 if (remote_nodemaps[j]->nodes[i].pnn != nodemap->nodes[i].pnn) {
3245 DEBUG(DEBUG_ERR, (__location__ " Remote node:%u has different nodemap pnn for %d (%u vs %u).\n",
3246 nodemap->nodes[j].pnn, i,
3247 remote_nodemaps[j]->nodes[i].pnn, nodemap->nodes[i].pnn));
3248 ctdb_set_culprit(rec, nodemap->nodes[j].pnn);
3249 do_recovery(rec, mem_ctx, pnn, nodemap,
3255 /* verify the flags are consistent
3257 for (i=0; i<nodemap->num; i++) {
3258 if (nodemap->nodes[i].flags & NODE_FLAGS_DISCONNECTED) {
3262 if (nodemap->nodes[i].flags != remote_nodemaps[j]->nodes[i].flags) {
3263 DEBUG(DEBUG_ERR, (__location__ " Remote node:%u has different flags for node %u. It has 0x%02x vs our 0x%02x\n",
3264 nodemap->nodes[j].pnn,
3265 nodemap->nodes[i].pnn,
3266 remote_nodemaps[j]->nodes[i].flags,
3267 nodemap->nodes[j].flags));
3269 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));
3270 update_flags_on_all_nodes(ctdb, nodemap, nodemap->nodes[i].pnn, remote_nodemaps[j]->nodes[i].flags);
3271 ctdb_set_culprit(rec, nodemap->nodes[j].pnn);
3272 do_recovery(rec, mem_ctx, pnn, nodemap,
3276 DEBUG(DEBUG_ERR,("Use flags 0x%02x from local recmaster node for cluster update of node %d flags\n", nodemap->nodes[i].flags, i));
3277 update_flags_on_all_nodes(ctdb, nodemap, nodemap->nodes[i].pnn, nodemap->nodes[i].flags);
3278 ctdb_set_culprit(rec, nodemap->nodes[j].pnn);
3279 do_recovery(rec, mem_ctx, pnn, nodemap,
3288 /* there better be the same number of lmasters in the vnn map
3289 as there are active nodes or we will have to do a recovery
3291 if (vnnmap->size != rec->num_active) {
3292 DEBUG(DEBUG_ERR, (__location__ " The vnnmap count is different from the number of active nodes. %u vs %u\n",
3293 vnnmap->size, rec->num_active));
3294 ctdb_set_culprit(rec, ctdb->pnn);
3295 do_recovery(rec, mem_ctx, pnn, nodemap, vnnmap);
3299 /* verify that all active nodes in the nodemap also exist in
3302 for (j=0; j<nodemap->num; j++) {
3303 if (nodemap->nodes[j].flags & NODE_FLAGS_INACTIVE) {
3306 if (nodemap->nodes[j].pnn == pnn) {
3310 for (i=0; i<vnnmap->size; i++) {
3311 if (vnnmap->map[i] == nodemap->nodes[j].pnn) {
3315 if (i == vnnmap->size) {
3316 DEBUG(DEBUG_ERR, (__location__ " Node %u is active in the nodemap but did not exist in the vnnmap\n",
3317 nodemap->nodes[j].pnn));
3318 ctdb_set_culprit(rec, nodemap->nodes[j].pnn);
3319 do_recovery(rec, mem_ctx, pnn, nodemap, vnnmap);
3325 /* verify that all other nodes have the same vnnmap
3326 and are from the same generation
3328 for (j=0; j<nodemap->num; j++) {
3329 if (nodemap->nodes[j].flags & NODE_FLAGS_INACTIVE) {
3332 if (nodemap->nodes[j].pnn == pnn) {
3336 ret = ctdb_ctrl_getvnnmap(ctdb, CONTROL_TIMEOUT(), nodemap->nodes[j].pnn,
3337 mem_ctx, &remote_vnnmap);
3339 DEBUG(DEBUG_ERR, (__location__ " Unable to get vnnmap from remote node %u\n",
3340 nodemap->nodes[j].pnn));
3344 /* verify the vnnmap generation is the same */
3345 if (vnnmap->generation != remote_vnnmap->generation) {
3346 DEBUG(DEBUG_ERR, (__location__ " Remote node %u has different generation of vnnmap. %u vs %u (ours)\n",
3347 nodemap->nodes[j].pnn, remote_vnnmap->generation, vnnmap->generation));
3348 ctdb_set_culprit(rec, nodemap->nodes[j].pnn);
3349 do_recovery(rec, mem_ctx, pnn, nodemap, vnnmap);
3353 /* verify the vnnmap size is the same */
3354 if (vnnmap->size != remote_vnnmap->size) {
3355 DEBUG(DEBUG_ERR, (__location__ " Remote node %u has different size of vnnmap. %u vs %u (ours)\n",
3356 nodemap->nodes[j].pnn, remote_vnnmap->size, vnnmap->size));
3357 ctdb_set_culprit(rec, nodemap->nodes[j].pnn);
3358 do_recovery(rec, mem_ctx, pnn, nodemap, vnnmap);
3362 /* verify the vnnmap is the same */
3363 for (i=0;i<vnnmap->size;i++) {
3364 if (remote_vnnmap->map[i] != vnnmap->map[i]) {
3365 DEBUG(DEBUG_ERR, (__location__ " Remote node %u has different vnnmap.\n",
3366 nodemap->nodes[j].pnn));
3367 ctdb_set_culprit(rec, nodemap->nodes[j].pnn);
3368 do_recovery(rec, mem_ctx, pnn, nodemap,
3375 /* we might need to change who has what IP assigned */
3376 if (rec->need_takeover_run) {
3377 uint32_t culprit = (uint32_t)-1;
3379 rec->need_takeover_run = false;
3381 /* update the list of public ips that a node can handle for
3384 ret = ctdb_reload_remote_public_ips(ctdb, rec, nodemap, &culprit);
3386 DEBUG(DEBUG_ERR,("Failed to read public ips from remote node %d\n",
3388 ctdb_set_culprit(rec, culprit);
3389 do_recovery(rec, mem_ctx, pnn, nodemap, vnnmap);
3393 /* execute the "startrecovery" event script on all nodes */
3394 ret = run_startrecovery_eventscript(rec, nodemap);
3396 DEBUG(DEBUG_ERR, (__location__ " Unable to run the 'startrecovery' event on cluster\n"));
3397 ctdb_set_culprit(rec, ctdb->pnn);
3398 do_recovery(rec, mem_ctx, pnn, nodemap, vnnmap);
3402 ret = ctdb_takeover_run(ctdb, nodemap);
3404 DEBUG(DEBUG_ERR, (__location__ " Unable to setup public takeover addresses - starting recovery\n"));
3405 ctdb_set_culprit(rec, ctdb->pnn);
3406 do_recovery(rec, mem_ctx, pnn, nodemap, vnnmap);
3410 /* execute the "recovered" event script on all nodes */
3411 ret = run_recovered_eventscript(ctdb, nodemap, "monitor_cluster");
3413 // we cant check whether the event completed successfully
3414 // since this script WILL fail if the node is in recovery mode
3415 // and if that race happens, the code here would just cause a second
3416 // cascading recovery.
3418 DEBUG(DEBUG_ERR, (__location__ " Unable to run the 'recovered' event on cluster. Update of public ips failed.\n"));
3419 ctdb_set_culprit(rec, ctdb->pnn);
3420 do_recovery(rec, mem_ctx, pnn, nodemap, vnnmap);
3427 the main monitoring loop
3429 static void monitor_cluster(struct ctdb_context *ctdb)
3431 struct ctdb_recoverd *rec;
3433 DEBUG(DEBUG_NOTICE,("monitor_cluster starting\n"));
3435 rec = talloc_zero(ctdb, struct ctdb_recoverd);
3436 CTDB_NO_MEMORY_FATAL(ctdb, rec);
3440 rec->priority_time = timeval_current();
3442 /* register a message port for sending memory dumps */
3443 ctdb_client_set_message_handler(ctdb, CTDB_SRVID_MEM_DUMP, mem_dump_handler, rec);
3445 /* register a message port for recovery elections */
3446 ctdb_client_set_message_handler(ctdb, CTDB_SRVID_RECOVERY, election_handler, rec);
3448 /* when nodes are disabled/enabled */
3449 ctdb_client_set_message_handler(ctdb, CTDB_SRVID_SET_NODE_FLAGS, monitor_handler, rec);
3451 /* when we are asked to puch out a flag change */
3452 ctdb_client_set_message_handler(ctdb, CTDB_SRVID_PUSH_NODE_FLAGS, push_flags_handler, rec);
3454 /* register a message port for vacuum fetch */
3455 ctdb_client_set_message_handler(ctdb, CTDB_SRVID_VACUUM_FETCH, vacuum_fetch_handler, rec);
3457 /* register a message port for reloadnodes */
3458 ctdb_client_set_message_handler(ctdb, CTDB_SRVID_RELOAD_NODES, reload_nodes_handler, rec);
3460 /* register a message port for performing a takeover run */
3461 ctdb_client_set_message_handler(ctdb, CTDB_SRVID_TAKEOVER_RUN, ip_reallocate_handler, rec);
3463 /* register a message port for disabling the ip check for a short while */
3464 ctdb_client_set_message_handler(ctdb, CTDB_SRVID_DISABLE_IP_CHECK, disable_ip_check_handler, rec);
3466 /* register a message port for updating the recovery daemons node assignment for an ip */
3467 ctdb_client_set_message_handler(ctdb, CTDB_SRVID_RECD_UPDATE_IP, recd_update_ip_handler, rec);
3470 TALLOC_CTX *mem_ctx = talloc_new(ctdb);
3471 struct timeval start;
3475 DEBUG(DEBUG_CRIT,(__location__
3476 " Failed to create temp context\n"));
3480 start = timeval_current();
3481 main_loop(ctdb, rec, mem_ctx);
3482 talloc_free(mem_ctx);
3484 /* we only check for recovery once every second */
3485 elapsed = timeval_elapsed(&start);
3486 if (elapsed < ctdb->tunable.recover_interval) {
3487 ctdb_wait_timeout(ctdb, ctdb->tunable.recover_interval
3494 event handler for when the main ctdbd dies
3496 static void ctdb_recoverd_parent(struct event_context *ev, struct fd_event *fde,
3497 uint16_t flags, void *private_data)
3499 DEBUG(DEBUG_ALERT,("recovery daemon parent died - exiting\n"));
3504 called regularly to verify that the recovery daemon is still running
3506 static void ctdb_check_recd(struct event_context *ev, struct timed_event *te,
3507 struct timeval yt, void *p)
3509 struct ctdb_context *ctdb = talloc_get_type(p, struct ctdb_context);
3511 if (kill(ctdb->recoverd_pid, 0) != 0) {
3512 DEBUG(DEBUG_ERR,("Recovery daemon (pid:%d) is no longer running. Shutting down main daemon\n", (int)ctdb->recoverd_pid));
3514 ctdb_stop_recoverd(ctdb);
3515 ctdb_stop_keepalive(ctdb);
3516 ctdb_stop_monitoring(ctdb);
3517 ctdb_release_all_ips(ctdb);
3518 if (ctdb->methods != NULL) {
3519 ctdb->methods->shutdown(ctdb);
3521 ctdb_event_script(ctdb, CTDB_EVENT_SHUTDOWN);
3526 event_add_timed(ctdb->ev, ctdb,
3527 timeval_current_ofs(30, 0),
3528 ctdb_check_recd, ctdb);
3531 static void recd_sig_child_handler(struct event_context *ev,
3532 struct signal_event *se, int signum, int count,
3536 // struct ctdb_context *ctdb = talloc_get_type(private_data, struct ctdb_context);
3541 pid = waitpid(-1, &status, WNOHANG);
3543 if (errno != ECHILD) {
3544 DEBUG(DEBUG_ERR, (__location__ " waitpid() returned error. errno:%s(%d)\n", strerror(errno),errno));
3549 DEBUG(DEBUG_DEBUG, ("RECD SIGCHLD from %d\n", (int)pid));
3555 startup the recovery daemon as a child of the main ctdb daemon
3557 int ctdb_start_recoverd(struct ctdb_context *ctdb)
3560 struct signal_event *se;
3562 if (pipe(fd) != 0) {
3566 ctdb->ctdbd_pid = getpid();
3568 ctdb->recoverd_pid = fork();
3569 if (ctdb->recoverd_pid == -1) {
3573 if (ctdb->recoverd_pid != 0) {
3575 event_add_timed(ctdb->ev, ctdb,
3576 timeval_current_ofs(30, 0),
3577 ctdb_check_recd, ctdb);
3583 srandom(getpid() ^ time(NULL));
3585 if (switch_from_server_to_client(ctdb) != 0) {
3586 DEBUG(DEBUG_CRIT, (__location__ "ERROR: failed to switch recovery daemon into client mode. shutting down.\n"));
3590 DEBUG(DEBUG_DEBUG, (__location__ " Created PIPE FD:%d to recovery daemon\n", fd[0]));
3592 event_add_fd(ctdb->ev, ctdb, fd[0], EVENT_FD_READ|EVENT_FD_AUTOCLOSE,
3593 ctdb_recoverd_parent, &fd[0]);
3595 /* set up a handler to pick up sigchld */
3596 se = event_add_signal(ctdb->ev, ctdb,
3598 recd_sig_child_handler,
3601 DEBUG(DEBUG_CRIT,("Failed to set up signal handler for SIGCHLD in recovery daemon\n"));
3605 monitor_cluster(ctdb);
3607 DEBUG(DEBUG_ALERT,("ERROR: ctdb_recoverd finished!?\n"));
3612 shutdown the recovery daemon
3614 void ctdb_stop_recoverd(struct ctdb_context *ctdb)
3616 if (ctdb->recoverd_pid == 0) {
3620 DEBUG(DEBUG_NOTICE,("Shutting down recovery daemon\n"));
3621 kill(ctdb->recoverd_pid, SIGTERM);