4 Copyright (C) Ronnie Sahlberg 2009
5 Copyright (C) Michael Adam 2010-2013
6 Copyright (C) Stefan Metzmacher 2010-2011
8 This program is free software; you can redistribute it and/or modify
9 it under the terms of the GNU General Public License as published by
10 the Free Software Foundation; either version 3 of the License, or
11 (at your option) any later version.
13 This program is distributed in the hope that it will be useful,
14 but WITHOUT ANY WARRANTY; without even the implied warranty of
15 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 GNU General Public License for more details.
18 You should have received a copy of the GNU General Public License
19 along with this program; if not, see <http://www.gnu.org/licenses/>.
23 #include "system/network.h"
24 #include "system/filesys.h"
25 #include "system/time.h"
30 #include "lib/tdb_wrap/tdb_wrap.h"
31 #include "lib/util/dlinklist.h"
32 #include "lib/util/debug.h"
33 #include "lib/util/samba_util.h"
34 #include "lib/util/sys_rw.h"
35 #include "lib/util/util_process.h"
37 #include "ctdb_private.h"
38 #include "ctdb_client.h"
40 #include "common/rb_tree.h"
41 #include "common/common.h"
42 #include "common/logging.h"
44 #define TIMELIMIT() timeval_current_ofs(10, 0)
46 enum vacuum_child_status
{ VACUUM_RUNNING
, VACUUM_OK
, VACUUM_ERROR
, VACUUM_TIMEOUT
};
48 struct ctdb_vacuum_child_context
{
49 struct ctdb_vacuum_child_context
*next
, *prev
;
50 struct ctdb_vacuum_handle
*vacuum_handle
;
51 /* fd child writes status to */
54 enum vacuum_child_status status
;
55 struct timeval start_time
;
58 struct ctdb_vacuum_handle
{
59 struct ctdb_db_context
*ctdb_db
;
60 struct ctdb_vacuum_child_context
*child_ctx
;
61 uint32_t fast_path_count
;
65 /* a list of records to possibly delete */
67 struct ctdb_context
*ctdb
;
68 struct ctdb_db_context
*ctdb_db
;
69 struct tdb_context
*dest_db
;
70 trbt_tree_t
*delete_list
;
71 struct ctdb_marshall_buffer
**vacuum_fetch_list
;
77 uint32_t added_to_vacuum_fetch_list
;
78 uint32_t added_to_delete_list
;
92 uint32_t remote_error
;
105 /* this structure contains the information for one record to be deleted */
106 struct delete_record_data
{
107 struct ctdb_context
*ctdb
;
108 struct ctdb_db_context
*ctdb_db
;
109 struct ctdb_ltdb_header hdr
;
110 uint32_t remote_fail_count
;
115 struct delete_records_list
{
116 struct ctdb_marshall_buffer
*records
;
117 struct vacuum_data
*vdata
;
120 static int insert_record_into_delete_queue(struct ctdb_db_context
*ctdb_db
,
121 const struct ctdb_ltdb_header
*hdr
,
125 * Store key and header in a tree, indexed by the key hash.
127 static int insert_delete_record_data_into_tree(struct ctdb_context
*ctdb
,
128 struct ctdb_db_context
*ctdb_db
,
130 const struct ctdb_ltdb_header
*hdr
,
133 struct delete_record_data
*dd
;
137 len
= offsetof(struct delete_record_data
, keydata
) + key
.dsize
;
139 dd
= (struct delete_record_data
*)talloc_size(tree
, len
);
141 DEBUG(DEBUG_ERR
,(__location__
" Out of memory\n"));
144 talloc_set_name_const(dd
, "struct delete_record_data");
147 dd
->ctdb_db
= ctdb_db
;
148 dd
->key
.dsize
= key
.dsize
;
149 dd
->key
.dptr
= dd
->keydata
;
150 memcpy(dd
->keydata
, key
.dptr
, key
.dsize
);
153 dd
->remote_fail_count
= 0;
155 hash
= ctdb_hash(&key
);
157 trbt_insert32(tree
, hash
, dd
);
162 static int add_record_to_delete_list(struct vacuum_data
*vdata
, TDB_DATA key
,
163 struct ctdb_ltdb_header
*hdr
)
165 struct ctdb_context
*ctdb
= vdata
->ctdb
;
166 struct ctdb_db_context
*ctdb_db
= vdata
->ctdb_db
;
170 hash
= ctdb_hash(&key
);
172 if (trbt_lookup32(vdata
->delete_list
, hash
)) {
173 DEBUG(DEBUG_INFO
, (__location__
" Hash collision when vacuuming, skipping this record.\n"));
177 ret
= insert_delete_record_data_into_tree(ctdb
, ctdb_db
,
184 vdata
->count
.delete_list
.total
++;
190 * Add a record to the list of records to be sent
191 * to their lmaster with VACUUM_FETCH.
193 static int add_record_to_vacuum_fetch_list(struct vacuum_data
*vdata
,
196 struct ctdb_context
*ctdb
= vdata
->ctdb
;
198 struct ctdb_marshall_buffer
*vfl
;
200 lmaster
= ctdb_lmaster(ctdb
, &key
);
202 vfl
= vdata
->vacuum_fetch_list
[lmaster
];
204 vfl
= ctdb_marshall_add(ctdb
, vfl
, vfl
->db_id
, ctdb
->pnn
,
205 key
, NULL
, tdb_null
);
207 DEBUG(DEBUG_ERR
,(__location__
" Out of memory\n"));
208 vdata
->traverse_error
= true;
212 vdata
->vacuum_fetch_list
[lmaster
] = vfl
;
218 static void ctdb_vacuum_event(struct tevent_context
*ev
,
219 struct tevent_timer
*te
,
220 struct timeval t
, void *private_data
);
222 static int vacuum_record_parser(TDB_DATA key
, TDB_DATA data
, void *private_data
)
224 struct ctdb_ltdb_header
*header
=
225 (struct ctdb_ltdb_header
*)private_data
;
227 if (data
.dsize
!= sizeof(struct ctdb_ltdb_header
)) {
231 *header
= *(struct ctdb_ltdb_header
*)data
.dptr
;
237 * traverse function for gathering the records that can be deleted
239 static int vacuum_traverse(struct tdb_context
*tdb
, TDB_DATA key
, TDB_DATA data
,
242 struct vacuum_data
*vdata
= talloc_get_type(private_data
,
244 struct ctdb_context
*ctdb
= vdata
->ctdb
;
245 struct ctdb_db_context
*ctdb_db
= vdata
->ctdb_db
;
247 struct ctdb_ltdb_header
*hdr
;
250 vdata
->count
.db_traverse
.total
++;
252 lmaster
= ctdb_lmaster(ctdb
, &key
);
253 if (lmaster
>= ctdb
->num_nodes
) {
254 vdata
->count
.db_traverse
.error
++;
255 DEBUG(DEBUG_CRIT
, (__location__
256 " lmaster[%u] >= ctdb->num_nodes[%u] for key"
259 (unsigned)ctdb
->num_nodes
,
260 (unsigned)ctdb_hash(&key
)));
264 if (data
.dsize
!= sizeof(struct ctdb_ltdb_header
)) {
265 /* it is not a deleted record */
266 vdata
->count
.db_traverse
.skipped
++;
270 hdr
= (struct ctdb_ltdb_header
*)data
.dptr
;
272 if (hdr
->dmaster
!= ctdb
->pnn
) {
273 vdata
->count
.db_traverse
.skipped
++;
278 * Add the record to this process's delete_queue for processing
279 * in the subsequent traverse in the fast vacuum run.
281 res
= insert_record_into_delete_queue(ctdb_db
, hdr
, key
);
283 vdata
->count
.db_traverse
.error
++;
285 vdata
->count
.db_traverse
.scheduled
++;
292 * traverse the tree of records to delete and marshall them into
295 static int delete_marshall_traverse(void *param
, void *data
)
297 struct delete_record_data
*dd
= talloc_get_type(data
, struct delete_record_data
);
298 struct delete_records_list
*recs
= talloc_get_type(param
, struct delete_records_list
);
299 struct ctdb_marshall_buffer
*m
;
301 m
= ctdb_marshall_add(recs
, recs
->records
, recs
->records
->db_id
,
302 recs
->records
->db_id
,
303 dd
->key
, &dd
->hdr
, tdb_null
);
305 DEBUG(DEBUG_ERR
, (__location__
" failed to marshall record\n"));
314 * traverse function for the traversal of the delete_queue,
315 * the fast-path vacuuming list.
317 * - If the record has been migrated off the node
318 * or has been revived (filled with data) on the node,
319 * then skip the record.
321 * - If the current node is the record's lmaster and it is
322 * a record that has never been migrated with data, then
323 * delete the record from the local tdb.
325 * - If the current node is the record's lmaster and it has
326 * been migrated with data, then schedule it for the normal
327 * vacuuming procedure (i.e. add it to the delete_list).
329 * - If the current node is NOT the record's lmaster then
330 * add it to the list of records that are to be sent to
331 * the lmaster with the VACUUM_FETCH message.
333 static int delete_queue_traverse(void *param
, void *data
)
335 struct delete_record_data
*dd
=
336 talloc_get_type(data
, struct delete_record_data
);
337 struct vacuum_data
*vdata
= talloc_get_type(param
, struct vacuum_data
);
338 struct ctdb_db_context
*ctdb_db
= dd
->ctdb_db
;
339 struct ctdb_context
*ctdb
= ctdb_db
->ctdb
; /* or dd->ctdb ??? */
341 struct ctdb_ltdb_header header
;
343 uint32_t hash
= ctdb_hash(&(dd
->key
));
345 vdata
->count
.delete_queue
.total
++;
347 res
= tdb_chainlock_nonblock(ctdb_db
->ltdb
->tdb
, dd
->key
);
349 vdata
->count
.delete_queue
.error
++;
353 res
= tdb_parse_record(ctdb_db
->ltdb
->tdb
, dd
->key
,
354 vacuum_record_parser
, &header
);
359 if (header
.dmaster
!= ctdb
->pnn
) {
360 /* The record has been migrated off the node. Skip. */
364 if (header
.rsn
!= dd
->hdr
.rsn
) {
366 * The record has been migrated off the node and back again.
367 * But not requeued for deletion. Skip it.
373 * We are dmaster, and the record has no data, and it has
374 * not been migrated after it has been queued for deletion.
376 * At this stage, the record could still have been revived locally
377 * and last been written with empty data. This can only be
378 * fixed with the addition of an active or delete flag. (TODO)
381 lmaster
= ctdb_lmaster(ctdb_db
->ctdb
, &dd
->key
);
383 if (lmaster
!= ctdb
->pnn
) {
384 res
= add_record_to_vacuum_fetch_list(vdata
, dd
->key
);
388 (__location__
" Error adding record to list "
389 "of records to send to lmaster.\n"));
390 vdata
->count
.delete_queue
.error
++;
392 vdata
->count
.delete_queue
.added_to_vacuum_fetch_list
++;
397 /* use header->flags or dd->hdr.flags ?? */
398 if (dd
->hdr
.flags
& CTDB_REC_FLAG_MIGRATED_WITH_DATA
) {
399 res
= add_record_to_delete_list(vdata
, dd
->key
, &dd
->hdr
);
403 (__location__
" Error adding record to list "
404 "of records for deletion on lmaster.\n"));
405 vdata
->count
.delete_queue
.error
++;
407 vdata
->count
.delete_queue
.added_to_delete_list
++;
410 res
= tdb_delete(ctdb_db
->ltdb
->tdb
, dd
->key
);
414 (__location__
" Error deleting record with key "
415 "hash [0x%08x] from local data base db[%s].\n",
416 hash
, ctdb_db
->db_name
));
417 vdata
->count
.delete_queue
.error
++;
422 (__location__
" Deleted record with key hash "
423 "[0x%08x] from local data base db[%s].\n",
424 hash
, ctdb_db
->db_name
));
425 vdata
->count
.delete_queue
.deleted
++;
431 vdata
->count
.delete_queue
.skipped
++;
434 tdb_chainunlock(ctdb_db
->ltdb
->tdb
, dd
->key
);
440 * Delete the records that we are lmaster and dmaster for and
441 * that could be deleted on all other nodes via the TRY_DELETE_RECORDS
444 static int delete_record_traverse(void *param
, void *data
)
446 struct delete_record_data
*dd
=
447 talloc_get_type(data
, struct delete_record_data
);
448 struct vacuum_data
*vdata
= talloc_get_type(param
, struct vacuum_data
);
449 struct ctdb_db_context
*ctdb_db
= dd
->ctdb_db
;
450 struct ctdb_context
*ctdb
= ctdb_db
->ctdb
;
452 struct ctdb_ltdb_header header
;
454 uint32_t hash
= ctdb_hash(&(dd
->key
));
456 if (dd
->remote_fail_count
> 0) {
457 vdata
->count
.delete_list
.remote_error
++;
458 vdata
->count
.delete_list
.left
--;
463 res
= tdb_chainlock(ctdb_db
->ltdb
->tdb
, dd
->key
);
466 (__location__
" Error getting chainlock on record with "
467 "key hash [0x%08x] on database db[%s].\n",
468 hash
, ctdb_db
->db_name
));
469 vdata
->count
.delete_list
.local_error
++;
470 vdata
->count
.delete_list
.left
--;
476 * Verify that the record is still empty, its RSN has not
477 * changed and that we are still its lmaster and dmaster.
480 res
= tdb_parse_record(ctdb_db
->ltdb
->tdb
, dd
->key
,
481 vacuum_record_parser
, &header
);
486 if (header
.flags
& CTDB_REC_RO_FLAGS
) {
487 DEBUG(DEBUG_INFO
, (__location__
": record with hash [0x%08x] "
488 "on database db[%s] has read-only flags. "
490 hash
, ctdb_db
->db_name
));
494 if (header
.dmaster
!= ctdb
->pnn
) {
495 DEBUG(DEBUG_INFO
, (__location__
": record with hash [0x%08x] "
496 "on database db[%s] has been migrated away. "
498 hash
, ctdb_db
->db_name
));
502 if (header
.rsn
!= dd
->hdr
.rsn
) {
504 * The record has been migrated off the node and back again.
505 * But not requeued for deletion. Skip it.
507 DEBUG(DEBUG_INFO
, (__location__
": record with hash [0x%08x] "
508 "on database db[%s] seems to have been "
509 "migrated away and back again (with empty "
510 "data). skipping.\n",
511 hash
, ctdb_db
->db_name
));
515 lmaster
= ctdb_lmaster(ctdb_db
->ctdb
, &dd
->key
);
517 if (lmaster
!= ctdb
->pnn
) {
518 DEBUG(DEBUG_INFO
, (__location__
": not lmaster for record in "
519 "delete list (key hash [0x%08x], db[%s]). "
520 "Strange! skipping.\n",
521 hash
, ctdb_db
->db_name
));
525 res
= tdb_delete(ctdb_db
->ltdb
->tdb
, dd
->key
);
529 (__location__
" Error deleting record with key hash "
530 "[0x%08x] from local data base db[%s].\n",
531 hash
, ctdb_db
->db_name
));
532 vdata
->count
.delete_list
.local_error
++;
537 (__location__
" Deleted record with key hash [0x%08x] from "
538 "local data base db[%s].\n", hash
, ctdb_db
->db_name
));
540 vdata
->count
.delete_list
.deleted
++;
544 vdata
->count
.delete_list
.skipped
++;
547 tdb_chainunlock(ctdb_db
->ltdb
->tdb
, dd
->key
);
550 vdata
->count
.delete_list
.left
--;
556 * Traverse the delete_queue.
557 * Records are either deleted directly or filled
558 * into the delete list or the vacuum fetch lists
559 * for further processing.
561 static void ctdb_process_delete_queue(struct ctdb_db_context
*ctdb_db
,
562 struct vacuum_data
*vdata
)
567 ret
= trbt_traversearray32(ctdb_db
->delete_queue
, 1,
568 delete_queue_traverse
, vdata
);
571 DEBUG(DEBUG_ERR
, (__location__
" Error traversing "
572 "the delete queue.\n"));
575 sum
= vdata
->count
.delete_queue
.deleted
576 + vdata
->count
.delete_queue
.skipped
577 + vdata
->count
.delete_queue
.error
578 + vdata
->count
.delete_queue
.added_to_delete_list
579 + vdata
->count
.delete_queue
.added_to_vacuum_fetch_list
;
581 if (vdata
->count
.delete_queue
.total
!= sum
) {
582 DEBUG(DEBUG_ERR
, (__location__
" Inconsistency in fast vacuum "
583 "counts for db[%s]: total[%u] != sum[%u]\n",
585 (unsigned)vdata
->count
.delete_queue
.total
,
589 if (vdata
->count
.delete_queue
.total
> 0) {
592 " fast vacuuming delete_queue traverse statistics: "
601 (unsigned)vdata
->count
.delete_queue
.total
,
602 (unsigned)vdata
->count
.delete_queue
.deleted
,
603 (unsigned)vdata
->count
.delete_queue
.skipped
,
604 (unsigned)vdata
->count
.delete_queue
.error
,
605 (unsigned)vdata
->count
.delete_queue
.added_to_delete_list
,
606 (unsigned)vdata
->count
.delete_queue
.added_to_vacuum_fetch_list
));
613 * read-only traverse of the database, looking for records that
614 * might be able to be vacuumed.
616 * This is not done each time but only every tunable
617 * VacuumFastPathCount times.
619 static void ctdb_vacuum_traverse_db(struct ctdb_db_context
*ctdb_db
,
620 struct vacuum_data
*vdata
)
624 ret
= tdb_traverse_read(ctdb_db
->ltdb
->tdb
, vacuum_traverse
, vdata
);
625 if (ret
== -1 || vdata
->traverse_error
) {
626 DEBUG(DEBUG_ERR
, (__location__
" Traverse error in vacuuming "
627 "'%s'\n", ctdb_db
->db_name
));
631 if (vdata
->count
.db_traverse
.total
> 0) {
634 " full vacuuming db traverse statistics: "
641 (unsigned)vdata
->count
.db_traverse
.total
,
642 (unsigned)vdata
->count
.db_traverse
.skipped
,
643 (unsigned)vdata
->count
.db_traverse
.error
,
644 (unsigned)vdata
->count
.db_traverse
.scheduled
));
651 * Process the vacuum fetch lists:
652 * For records for which we are not the lmaster, tell the lmaster to
655 static void ctdb_process_vacuum_fetch_lists(struct ctdb_db_context
*ctdb_db
,
656 struct vacuum_data
*vdata
)
659 struct ctdb_context
*ctdb
= ctdb_db
->ctdb
;
661 for (i
= 0; i
< ctdb
->num_nodes
; i
++) {
663 struct ctdb_marshall_buffer
*vfl
= vdata
->vacuum_fetch_list
[i
];
665 if (ctdb
->nodes
[i
]->pnn
== ctdb
->pnn
) {
669 if (vfl
->count
== 0) {
673 DEBUG(DEBUG_INFO
, ("Found %u records for lmaster %u in '%s'\n",
674 vfl
->count
, ctdb
->nodes
[i
]->pnn
,
677 data
= ctdb_marshall_finish(vfl
);
678 if (ctdb_client_send_message(ctdb
, ctdb
->nodes
[i
]->pnn
,
679 CTDB_SRVID_VACUUM_FETCH
,
682 DEBUG(DEBUG_ERR
, (__location__
" Failed to send vacuum "
683 "fetch message to %u\n",
684 ctdb
->nodes
[i
]->pnn
));
692 * Process the delete list:
694 * This is the last step of vacuuming that consistently deletes
695 * those records that have been migrated with data and can hence
696 * not be deleted when leaving a node.
698 * In this step, the lmaster does the final deletion of those empty
699 * records that it is also dmaster for. It has ususally received
700 * at least some of these records previously from the former dmasters
701 * with the vacuum fetch message.
703 * 1) Send the records to all active nodes with the TRY_DELETE_RECORDS
704 * control. The remote notes delete their local copy.
705 * 2) The lmaster locally deletes its copies of all records that
706 * could successfully be deleted remotely in step #2.
708 static void ctdb_process_delete_list(struct ctdb_db_context
*ctdb_db
,
709 struct vacuum_data
*vdata
)
712 struct ctdb_context
*ctdb
= ctdb_db
->ctdb
;
713 struct delete_records_list
*recs
;
715 struct ctdb_node_map_old
*nodemap
;
716 uint32_t *active_nodes
;
717 int num_active_nodes
;
721 if (vdata
->count
.delete_list
.total
== 0) {
725 tmp_ctx
= talloc_new(vdata
);
726 if (tmp_ctx
== NULL
) {
727 DEBUG(DEBUG_ERR
,(__location__
" Out of memory\n"));
731 vdata
->count
.delete_list
.left
= vdata
->count
.delete_list
.total
;
734 * get the list of currently active nodes
737 ret
= ctdb_ctrl_getnodemap(ctdb
, TIMELIMIT(),
742 DEBUG(DEBUG_ERR
,(__location__
" unable to get node map\n"));
746 active_nodes
= list_of_active_nodes(ctdb
, nodemap
,
747 nodemap
, /* talloc context */
748 false /* include self */);
750 num_active_nodes
= talloc_get_size(active_nodes
)/sizeof(*active_nodes
);
753 * Now delete the records all active nodes in a two-phase process:
754 * 1) tell all active remote nodes to delete all their copy
755 * 2) if all remote nodes deleted their record copy, delete it locally
758 recs
= talloc_zero(tmp_ctx
, struct delete_records_list
);
760 DEBUG(DEBUG_ERR
,(__location__
" Out of memory\n"));
766 * Send all records to all active nodes for deletion.
770 * Create a marshall blob from the remaining list of records to delete.
773 recs
->records
= (struct ctdb_marshall_buffer
*)
774 talloc_zero_size(recs
,
775 offsetof(struct ctdb_marshall_buffer
, data
));
776 if (recs
->records
== NULL
) {
777 DEBUG(DEBUG_ERR
,(__location__
" Out of memory\n"));
780 recs
->records
->db_id
= ctdb_db
->db_id
;
782 ret
= trbt_traversearray32(vdata
->delete_list
, 1,
783 delete_marshall_traverse
, recs
);
785 DEBUG(DEBUG_ERR
, (__location__
" Error traversing the "
786 "delete list for second marshalling.\n"));
790 indata
= ctdb_marshall_finish(recs
->records
);
792 for (i
= 0; i
< num_active_nodes
; i
++) {
793 struct ctdb_marshall_buffer
*records
;
794 struct ctdb_rec_data_old
*rec
;
798 ret
= ctdb_control(ctdb
, active_nodes
[i
], 0,
799 CTDB_CONTROL_TRY_DELETE_RECORDS
, 0,
800 indata
, recs
, &outdata
, &res
,
802 if (ret
!= 0 || res
!= 0) {
803 DEBUG(DEBUG_ERR
, ("Failed to delete records on "
804 "node %u: ret[%d] res[%d]\n",
805 active_nodes
[i
], ret
, res
));
810 * outdata contains the list of records coming back
811 * from the node: These are the records that the
812 * remote node could not delete. We remove these from
813 * the list to delete locally.
815 records
= (struct ctdb_marshall_buffer
*)outdata
.dptr
;
816 rec
= (struct ctdb_rec_data_old
*)&records
->data
[0];
817 while (records
->count
-- > 1) {
818 TDB_DATA reckey
, recdata
;
819 struct ctdb_ltdb_header
*rechdr
;
820 struct delete_record_data
*dd
;
822 reckey
.dptr
= &rec
->data
[0];
823 reckey
.dsize
= rec
->keylen
;
824 recdata
.dptr
= &rec
->data
[reckey
.dsize
];
825 recdata
.dsize
= rec
->datalen
;
827 if (recdata
.dsize
< sizeof(struct ctdb_ltdb_header
)) {
828 DEBUG(DEBUG_CRIT
,(__location__
" bad ltdb record\n"));
831 rechdr
= (struct ctdb_ltdb_header
*)recdata
.dptr
;
832 recdata
.dptr
+= sizeof(*rechdr
);
833 recdata
.dsize
-= sizeof(*rechdr
);
835 dd
= (struct delete_record_data
*)trbt_lookup32(
840 * The remote node could not delete the
841 * record. Since other remote nodes can
842 * also fail, we just mark the record.
844 dd
->remote_fail_count
++;
846 DEBUG(DEBUG_ERR
, (__location__
" Failed to "
847 "find record with hash 0x%08x coming "
848 "back from TRY_DELETE_RECORDS "
849 "control in delete list.\n",
850 ctdb_hash(&reckey
)));
853 rec
= (struct ctdb_rec_data_old
*)(rec
->length
+ (uint8_t *)rec
);
857 if (vdata
->count
.delete_list
.left
== 0) {
863 * Delete the remaining records locally.
865 * These records have successfully been deleted on all
866 * active remote nodes.
869 ret
= trbt_traversearray32(vdata
->delete_list
, 1,
870 delete_record_traverse
, vdata
);
872 DEBUG(DEBUG_ERR
, (__location__
" Error traversing the "
873 "delete list for deletion.\n"));
878 if (vdata
->count
.delete_list
.left
!= 0) {
879 DEBUG(DEBUG_ERR
, (__location__
" Vaccum db[%s] error: "
880 "there are %u records left for deletion after "
881 "processing delete list\n",
883 (unsigned)vdata
->count
.delete_list
.left
));
886 sum
= vdata
->count
.delete_list
.deleted
887 + vdata
->count
.delete_list
.skipped
888 + vdata
->count
.delete_list
.remote_error
889 + vdata
->count
.delete_list
.local_error
890 + vdata
->count
.delete_list
.left
;
892 if (vdata
->count
.delete_list
.total
!= sum
) {
893 DEBUG(DEBUG_ERR
, (__location__
" Inconsistency in vacuum "
894 "delete list counts for db[%s]: total[%u] != sum[%u]\n",
896 (unsigned)vdata
->count
.delete_list
.total
,
900 if (vdata
->count
.delete_list
.total
> 0) {
903 " vacuum delete list statistics: "
912 (unsigned)vdata
->count
.delete_list
.total
,
913 (unsigned)vdata
->count
.delete_list
.deleted
,
914 (unsigned)vdata
->count
.delete_list
.skipped
,
915 (unsigned)vdata
->count
.delete_list
.remote_error
,
916 (unsigned)vdata
->count
.delete_list
.local_error
,
917 (unsigned)vdata
->count
.delete_list
.left
));
921 talloc_free(tmp_ctx
);
927 * initialize the vacuum_data
929 static struct vacuum_data
*ctdb_vacuum_init_vacuum_data(
930 struct ctdb_db_context
*ctdb_db
,
934 struct ctdb_context
*ctdb
= ctdb_db
->ctdb
;
935 struct vacuum_data
*vdata
;
937 vdata
= talloc_zero(mem_ctx
, struct vacuum_data
);
939 DEBUG(DEBUG_ERR
,(__location__
" Out of memory\n"));
943 vdata
->ctdb
= ctdb_db
->ctdb
;
944 vdata
->ctdb_db
= ctdb_db
;
945 vdata
->delete_list
= trbt_create(vdata
, 0);
946 if (vdata
->delete_list
== NULL
) {
947 DEBUG(DEBUG_ERR
,(__location__
" Out of memory\n"));
951 vdata
->start
= timeval_current();
953 vdata
->count
.delete_queue
.added_to_delete_list
= 0;
954 vdata
->count
.delete_queue
.added_to_vacuum_fetch_list
= 0;
955 vdata
->count
.delete_queue
.deleted
= 0;
956 vdata
->count
.delete_queue
.skipped
= 0;
957 vdata
->count
.delete_queue
.error
= 0;
958 vdata
->count
.delete_queue
.total
= 0;
959 vdata
->count
.db_traverse
.scheduled
= 0;
960 vdata
->count
.db_traverse
.skipped
= 0;
961 vdata
->count
.db_traverse
.error
= 0;
962 vdata
->count
.db_traverse
.total
= 0;
963 vdata
->count
.delete_list
.total
= 0;
964 vdata
->count
.delete_list
.left
= 0;
965 vdata
->count
.delete_list
.remote_error
= 0;
966 vdata
->count
.delete_list
.local_error
= 0;
967 vdata
->count
.delete_list
.skipped
= 0;
968 vdata
->count
.delete_list
.deleted
= 0;
970 /* the list needs to be of length num_nodes */
971 vdata
->vacuum_fetch_list
= talloc_zero_array(vdata
,
972 struct ctdb_marshall_buffer
*,
974 if (vdata
->vacuum_fetch_list
== NULL
) {
975 DEBUG(DEBUG_ERR
,(__location__
" Out of memory\n"));
978 for (i
= 0; i
< ctdb
->num_nodes
; i
++) {
979 vdata
->vacuum_fetch_list
[i
] = (struct ctdb_marshall_buffer
*)
980 talloc_zero_size(vdata
->vacuum_fetch_list
,
981 offsetof(struct ctdb_marshall_buffer
, data
));
982 if (vdata
->vacuum_fetch_list
[i
] == NULL
) {
983 DEBUG(DEBUG_ERR
,(__location__
" Out of memory\n"));
987 vdata
->vacuum_fetch_list
[i
]->db_id
= ctdb_db
->db_id
;
999 * - Always do the fast vacuuming run, which traverses
1000 * the in-memory delete queue: these records have been
1001 * scheduled for deletion.
1002 * - Only if explicitly requested, the database is traversed
1003 * in order to use the traditional heuristics on empty records
1004 * to trigger deletion.
1005 * This is done only every VacuumFastPathCount'th vacuuming run.
1007 * The traverse runs fill two lists:
1009 * - The delete_list:
1010 * This is the list of empty records the current
1011 * node is lmaster and dmaster for. These records are later
1012 * deleted first on other nodes and then locally.
1014 * The fast vacuuming run has a short cut for those records
1015 * that have never been migrated with data: these records
1016 * are immediately deleted locally, since they have left
1017 * no trace on other nodes.
1019 * - The vacuum_fetch lists
1020 * (one for each other lmaster node):
1021 * The records in this list are sent for deletion to
1022 * their lmaster in a bulk VACUUM_FETCH message.
1024 * The lmaster then migrates all these records to itelf
1025 * so that they can be vacuumed there.
1027 * This executes in the child context.
1029 static int ctdb_vacuum_db(struct ctdb_db_context
*ctdb_db
,
1030 bool full_vacuum_run
)
1032 struct ctdb_context
*ctdb
= ctdb_db
->ctdb
;
1034 struct vacuum_data
*vdata
;
1035 TALLOC_CTX
*tmp_ctx
;
1037 DEBUG(DEBUG_INFO
, (__location__
" Entering %s vacuum run for db "
1038 "%s db_id[0x%08x]\n",
1039 full_vacuum_run
? "full" : "fast",
1040 ctdb_db
->db_name
, ctdb_db
->db_id
));
1042 ret
= ctdb_ctrl_getvnnmap(ctdb
, TIMELIMIT(), CTDB_CURRENT_NODE
, ctdb
, &ctdb
->vnn_map
);
1044 DEBUG(DEBUG_ERR
, ("Unable to get vnnmap from local node\n"));
1048 pnn
= ctdb_ctrl_getpnn(ctdb
, TIMELIMIT(), CTDB_CURRENT_NODE
);
1050 DEBUG(DEBUG_ERR
, ("Unable to get pnn from local node\n"));
1056 tmp_ctx
= talloc_new(ctdb_db
);
1057 if (tmp_ctx
== NULL
) {
1058 DEBUG(DEBUG_ERR
, ("Out of memory!\n"));
1062 vdata
= ctdb_vacuum_init_vacuum_data(ctdb_db
, tmp_ctx
);
1063 if (vdata
== NULL
) {
1064 talloc_free(tmp_ctx
);
1068 if (full_vacuum_run
) {
1069 ctdb_vacuum_traverse_db(ctdb_db
, vdata
);
1072 ctdb_process_delete_queue(ctdb_db
, vdata
);
1074 ctdb_process_vacuum_fetch_lists(ctdb_db
, vdata
);
1076 ctdb_process_delete_list(ctdb_db
, vdata
);
1078 talloc_free(tmp_ctx
);
1080 /* this ensures we run our event queue */
1081 ctdb_ctrl_getpnn(ctdb
, TIMELIMIT(), CTDB_CURRENT_NODE
);
1087 * repack and vaccum a db
1088 * called from the child context
1090 static int ctdb_vacuum_and_repack_db(struct ctdb_db_context
*ctdb_db
,
1091 bool full_vacuum_run
)
1093 uint32_t repack_limit
= ctdb_db
->ctdb
->tunable
.repack_limit
;
1094 const char *name
= ctdb_db
->db_name
;
1095 int freelist_size
= 0;
1098 if (ctdb_vacuum_db(ctdb_db
, full_vacuum_run
) != 0) {
1099 DEBUG(DEBUG_ERR
,(__location__
" Failed to vacuum '%s'\n", name
));
1102 freelist_size
= tdb_freelist_size(ctdb_db
->ltdb
->tdb
);
1103 if (freelist_size
== -1) {
1104 DEBUG(DEBUG_ERR
,(__location__
" Failed to get freelist size for '%s'\n", name
));
1109 * decide if a repack is necessary
1111 if ((repack_limit
== 0 || (uint32_t)freelist_size
< repack_limit
))
1116 DEBUG(DEBUG_INFO
, ("Repacking %s with %u freelist entries\n",
1117 name
, freelist_size
));
1119 ret
= tdb_repack(ctdb_db
->ltdb
->tdb
);
1121 DEBUG(DEBUG_ERR
,(__location__
" Failed to repack '%s'\n", name
));
1128 static uint32_t get_vacuum_interval(struct ctdb_db_context
*ctdb_db
)
1130 uint32_t interval
= ctdb_db
->ctdb
->tunable
.vacuum_interval
;
1135 static int vacuum_child_destructor(struct ctdb_vacuum_child_context
*child_ctx
)
1137 double l
= timeval_elapsed(&child_ctx
->start_time
);
1138 struct ctdb_db_context
*ctdb_db
= child_ctx
->vacuum_handle
->ctdb_db
;
1139 struct ctdb_context
*ctdb
= ctdb_db
->ctdb
;
1141 CTDB_UPDATE_DB_LATENCY(ctdb_db
, "vacuum", vacuum
.latency
, l
);
1142 DEBUG(DEBUG_INFO
,("Vacuuming took %.3f seconds for database %s\n", l
, ctdb_db
->db_name
));
1144 if (child_ctx
->child_pid
!= -1) {
1145 ctdb_kill(ctdb
, child_ctx
->child_pid
, SIGKILL
);
1147 /* Bump the number of successful fast-path runs. */
1148 child_ctx
->vacuum_handle
->fast_path_count
++;
1151 DLIST_REMOVE(ctdb
->vacuumers
, child_ctx
);
1153 tevent_add_timer(ctdb
->ev
, child_ctx
->vacuum_handle
,
1154 timeval_current_ofs(get_vacuum_interval(ctdb_db
), 0),
1155 ctdb_vacuum_event
, child_ctx
->vacuum_handle
);
1161 * this event is generated when a vacuum child process times out
1163 static void vacuum_child_timeout(struct tevent_context
*ev
,
1164 struct tevent_timer
*te
,
1165 struct timeval t
, void *private_data
)
1167 struct ctdb_vacuum_child_context
*child_ctx
= talloc_get_type(private_data
, struct ctdb_vacuum_child_context
);
1169 DEBUG(DEBUG_ERR
,("Vacuuming child process timed out for db %s\n", child_ctx
->vacuum_handle
->ctdb_db
->db_name
));
1171 child_ctx
->status
= VACUUM_TIMEOUT
;
1173 talloc_free(child_ctx
);
1178 * this event is generated when a vacuum child process has completed
1180 static void vacuum_child_handler(struct tevent_context
*ev
,
1181 struct tevent_fd
*fde
,
1182 uint16_t flags
, void *private_data
)
1184 struct ctdb_vacuum_child_context
*child_ctx
= talloc_get_type(private_data
, struct ctdb_vacuum_child_context
);
1188 DEBUG(DEBUG_INFO
,("Vacuuming child process %d finished for db %s\n", child_ctx
->child_pid
, child_ctx
->vacuum_handle
->ctdb_db
->db_name
));
1189 child_ctx
->child_pid
= -1;
1191 ret
= sys_read(child_ctx
->fd
[0], &c
, 1);
1192 if (ret
!= 1 || c
!= 0) {
1193 child_ctx
->status
= VACUUM_ERROR
;
1194 DEBUG(DEBUG_ERR
, ("A vacuum child process failed with an error for database %s. ret=%d c=%d\n", child_ctx
->vacuum_handle
->ctdb_db
->db_name
, ret
, c
));
1196 child_ctx
->status
= VACUUM_OK
;
1199 talloc_free(child_ctx
);
1203 * this event is called every time we need to start a new vacuum process
1205 static void ctdb_vacuum_event(struct tevent_context
*ev
,
1206 struct tevent_timer
*te
,
1207 struct timeval t
, void *private_data
)
1209 struct ctdb_vacuum_handle
*vacuum_handle
= talloc_get_type(private_data
, struct ctdb_vacuum_handle
);
1210 struct ctdb_db_context
*ctdb_db
= vacuum_handle
->ctdb_db
;
1211 struct ctdb_context
*ctdb
= ctdb_db
->ctdb
;
1212 struct ctdb_vacuum_child_context
*child_ctx
;
1213 struct tevent_fd
*fde
;
1216 /* we don't vacuum if we are in recovery mode, or db frozen */
1217 if (ctdb
->recovery_mode
== CTDB_RECOVERY_ACTIVE
||
1218 ctdb_db_frozen(ctdb_db
)) {
1219 DEBUG(DEBUG_INFO
, ("Not vacuuming %s (%s)\n", ctdb_db
->db_name
,
1220 ctdb
->recovery_mode
== CTDB_RECOVERY_ACTIVE
?
1221 "in recovery" : "frozen"));
1222 tevent_add_timer(ctdb
->ev
, vacuum_handle
,
1223 timeval_current_ofs(get_vacuum_interval(ctdb_db
), 0),
1224 ctdb_vacuum_event
, vacuum_handle
);
1228 /* Do not allow multiple vacuuming child processes to be active at the
1229 * same time. If there is vacuuming child process active, delay
1230 * new vacuuming event to stagger vacuuming events.
1232 if (ctdb
->vacuumers
!= NULL
) {
1233 tevent_add_timer(ctdb
->ev
, vacuum_handle
,
1234 timeval_current_ofs(0, 500*1000),
1235 ctdb_vacuum_event
, vacuum_handle
);
1239 child_ctx
= talloc(vacuum_handle
, struct ctdb_vacuum_child_context
);
1240 if (child_ctx
== NULL
) {
1241 DEBUG(DEBUG_CRIT
, (__location__
" Failed to allocate child context for vacuuming of %s\n", ctdb_db
->db_name
));
1242 ctdb_fatal(ctdb
, "Out of memory when crating vacuum child context. Shutting down\n");
1246 ret
= pipe(child_ctx
->fd
);
1248 talloc_free(child_ctx
);
1249 DEBUG(DEBUG_ERR
, ("Failed to create pipe for vacuum child process.\n"));
1250 tevent_add_timer(ctdb
->ev
, vacuum_handle
,
1251 timeval_current_ofs(get_vacuum_interval(ctdb_db
), 0),
1252 ctdb_vacuum_event
, vacuum_handle
);
1256 if (vacuum_handle
->fast_path_count
> ctdb
->tunable
.vacuum_fast_path_count
) {
1257 vacuum_handle
->fast_path_count
= 0;
1260 child_ctx
->child_pid
= ctdb_fork(ctdb
);
1261 if (child_ctx
->child_pid
== (pid_t
)-1) {
1262 close(child_ctx
->fd
[0]);
1263 close(child_ctx
->fd
[1]);
1264 talloc_free(child_ctx
);
1265 DEBUG(DEBUG_ERR
, ("Failed to fork vacuum child process.\n"));
1266 tevent_add_timer(ctdb
->ev
, vacuum_handle
,
1267 timeval_current_ofs(get_vacuum_interval(ctdb_db
), 0),
1268 ctdb_vacuum_event
, vacuum_handle
);
1273 if (child_ctx
->child_pid
== 0) {
1275 bool full_vacuum_run
= false;
1276 close(child_ctx
->fd
[0]);
1278 DEBUG(DEBUG_INFO
,("Vacuuming child process %d for db %s started\n", getpid(), ctdb_db
->db_name
));
1279 prctl_set_comment("ctdb_vacuum");
1280 if (switch_from_server_to_client(ctdb
) != 0) {
1281 DEBUG(DEBUG_CRIT
, (__location__
"ERROR: failed to switch vacuum daemon into client mode. Shutting down.\n"));
1285 if ((ctdb
->tunable
.vacuum_fast_path_count
> 0) &&
1286 (vacuum_handle
->fast_path_count
== 0))
1288 full_vacuum_run
= true;
1290 cc
= ctdb_vacuum_and_repack_db(ctdb_db
, full_vacuum_run
);
1292 sys_write(child_ctx
->fd
[1], &cc
, 1);
1296 set_close_on_exec(child_ctx
->fd
[0]);
1297 close(child_ctx
->fd
[1]);
1299 child_ctx
->status
= VACUUM_RUNNING
;
1300 child_ctx
->start_time
= timeval_current();
1302 DLIST_ADD(ctdb
->vacuumers
, child_ctx
);
1303 talloc_set_destructor(child_ctx
, vacuum_child_destructor
);
1306 * Clear the fastpath vacuuming list in the parent.
1308 talloc_free(ctdb_db
->delete_queue
);
1309 ctdb_db
->delete_queue
= trbt_create(ctdb_db
, 0);
1310 if (ctdb_db
->delete_queue
== NULL
) {
1311 /* fatal here? ... */
1312 ctdb_fatal(ctdb
, "Out of memory when re-creating vacuum tree "
1313 "in parent context. Shutting down\n");
1316 tevent_add_timer(ctdb
->ev
, child_ctx
,
1317 timeval_current_ofs(ctdb
->tunable
.vacuum_max_run_time
, 0),
1318 vacuum_child_timeout
, child_ctx
);
1320 DEBUG(DEBUG_DEBUG
, (__location__
" Created PIPE FD:%d to child vacuum process\n", child_ctx
->fd
[0]));
1322 fde
= tevent_add_fd(ctdb
->ev
, child_ctx
, child_ctx
->fd
[0],
1323 TEVENT_FD_READ
, vacuum_child_handler
, child_ctx
);
1324 tevent_fd_set_auto_close(fde
);
1326 vacuum_handle
->child_ctx
= child_ctx
;
1327 child_ctx
->vacuum_handle
= vacuum_handle
;
1330 void ctdb_stop_vacuuming(struct ctdb_context
*ctdb
)
1332 /* Simply free them all. */
1333 while (ctdb
->vacuumers
) {
1334 DEBUG(DEBUG_INFO
, ("Aborting vacuuming for %s (%i)\n",
1335 ctdb
->vacuumers
->vacuum_handle
->ctdb_db
->db_name
,
1336 (int)ctdb
->vacuumers
->child_pid
));
1337 /* vacuum_child_destructor kills it, removes from list */
1338 talloc_free(ctdb
->vacuumers
);
1342 /* this function initializes the vacuuming context for a database
1343 * starts the vacuuming events
1345 int ctdb_vacuum_init(struct ctdb_db_context
*ctdb_db
)
1347 if (! ctdb_db_volatile(ctdb_db
)) {
1349 ("Vacuuming is disabled for non-volatile database %s\n",
1354 ctdb_db
->vacuum_handle
= talloc(ctdb_db
, struct ctdb_vacuum_handle
);
1355 CTDB_NO_MEMORY(ctdb_db
->ctdb
, ctdb_db
->vacuum_handle
);
1357 ctdb_db
->vacuum_handle
->ctdb_db
= ctdb_db
;
1358 ctdb_db
->vacuum_handle
->fast_path_count
= 0;
1360 tevent_add_timer(ctdb_db
->ctdb
->ev
, ctdb_db
->vacuum_handle
,
1361 timeval_current_ofs(get_vacuum_interval(ctdb_db
), 0),
1362 ctdb_vacuum_event
, ctdb_db
->vacuum_handle
);
1367 static void remove_record_from_delete_queue(struct ctdb_db_context
*ctdb_db
,
1368 const struct ctdb_ltdb_header
*hdr
,
1371 struct delete_record_data
*kd
;
1374 hash
= (uint32_t)ctdb_hash(&key
);
1376 DEBUG(DEBUG_DEBUG
, (__location__
1377 " remove_record_from_delete_queue: "
1382 "migrated_with_data[%s]\n",
1383 ctdb_db
->db_name
, ctdb_db
->db_id
,
1385 ctdb_lmaster(ctdb_db
->ctdb
, &key
),
1386 hdr
->flags
& CTDB_REC_FLAG_MIGRATED_WITH_DATA
? "yes" : "no"));
1388 kd
= (struct delete_record_data
*)trbt_lookup32(ctdb_db
->delete_queue
, hash
);
1390 DEBUG(DEBUG_DEBUG
, (__location__
1391 " remove_record_from_delete_queue: "
1392 "record not in queue (hash[0x%08x])\n.",
1397 if ((kd
->key
.dsize
!= key
.dsize
) ||
1398 (memcmp(kd
->key
.dptr
, key
.dptr
, key
.dsize
) != 0))
1400 DEBUG(DEBUG_DEBUG
, (__location__
1401 " remove_record_from_delete_queue: "
1402 "hash collision for key with hash[0x%08x] "
1403 "in db[%s] - skipping\n",
1404 hash
, ctdb_db
->db_name
));
1408 DEBUG(DEBUG_DEBUG
, (__location__
1409 " remove_record_from_delete_queue: "
1410 "removing key with hash[0x%08x]\n",
1419 * Insert a record into the ctdb_db context's delete queue,
1420 * handling hash collisions.
1422 static int insert_record_into_delete_queue(struct ctdb_db_context
*ctdb_db
,
1423 const struct ctdb_ltdb_header
*hdr
,
1426 struct delete_record_data
*kd
;
1430 hash
= (uint32_t)ctdb_hash(&key
);
1432 DEBUG(DEBUG_DEBUG
, (__location__
" schedule for deletion: db[%s] "
1436 "migrated_with_data[%s]\n",
1437 ctdb_db
->db_name
, ctdb_db
->db_id
,
1439 ctdb_lmaster(ctdb_db
->ctdb
, &key
),
1440 hdr
->flags
& CTDB_REC_FLAG_MIGRATED_WITH_DATA
? "yes" : "no"));
1442 kd
= (struct delete_record_data
*)trbt_lookup32(ctdb_db
->delete_queue
, hash
);
1444 if ((kd
->key
.dsize
!= key
.dsize
) ||
1445 (memcmp(kd
->key
.dptr
, key
.dptr
, key
.dsize
) != 0))
1448 (__location__
" schedule for deletion: "
1449 "hash collision for key hash [0x%08x]. "
1450 "Skipping the record.\n", hash
));
1454 (__location__
" schedule for deletion: "
1455 "updating entry for key with hash [0x%08x].\n",
1460 ret
= insert_delete_record_data_into_tree(ctdb_db
->ctdb
, ctdb_db
,
1461 ctdb_db
->delete_queue
,
1465 (__location__
" schedule for deletion: error "
1466 "inserting key with hash [0x%08x] into delete queue\n",
1475 * Schedule a record for deletetion.
1476 * Called from the parent context.
1478 int32_t ctdb_control_schedule_for_deletion(struct ctdb_context
*ctdb
,
1481 struct ctdb_control_schedule_for_deletion
*dd
;
1482 struct ctdb_db_context
*ctdb_db
;
1486 dd
= (struct ctdb_control_schedule_for_deletion
*)indata
.dptr
;
1488 ctdb_db
= find_ctdb_db(ctdb
, dd
->db_id
);
1489 if (ctdb_db
== NULL
) {
1490 DEBUG(DEBUG_ERR
, (__location__
" Unknown db id 0x%08x\n",
1495 key
.dsize
= dd
->keylen
;
1498 ret
= insert_record_into_delete_queue(ctdb_db
, &dd
->hdr
, key
);
1503 int32_t ctdb_local_schedule_for_deletion(struct ctdb_db_context
*ctdb_db
,
1504 const struct ctdb_ltdb_header
*hdr
,
1508 struct ctdb_control_schedule_for_deletion
*dd
;
1512 if (ctdb_db
->ctdb
->ctdbd_pid
== getpid()) {
1513 /* main daemon - directly queue */
1514 ret
= insert_record_into_delete_queue(ctdb_db
, hdr
, key
);
1519 /* if we don't have a connection to the daemon we can not send
1520 a control. For example sometimes from update_record control child
1523 if (!ctdb_db
->ctdb
->can_send_controls
) {
1528 /* child process: send the main daemon a control */
1529 indata
.dsize
= offsetof(struct ctdb_control_schedule_for_deletion
, key
) + key
.dsize
;
1530 indata
.dptr
= talloc_zero_array(ctdb_db
, uint8_t, indata
.dsize
);
1531 if (indata
.dptr
== NULL
) {
1532 DEBUG(DEBUG_ERR
, (__location__
" out of memory\n"));
1535 dd
= (struct ctdb_control_schedule_for_deletion
*)(void *)indata
.dptr
;
1536 dd
->db_id
= ctdb_db
->db_id
;
1538 dd
->keylen
= key
.dsize
;
1539 memcpy(dd
->key
, key
.dptr
, key
.dsize
);
1541 ret
= ctdb_control(ctdb_db
->ctdb
,
1544 CTDB_CONTROL_SCHEDULE_FOR_DELETION
,
1545 CTDB_CTRL_FLAG_NOREPLY
, /* flags */
1550 NULL
, /* timeout : NULL == wait forever */
1551 NULL
); /* error message */
1553 talloc_free(indata
.dptr
);
1555 if (ret
!= 0 || status
!= 0) {
1556 DEBUG(DEBUG_ERR
, (__location__
" Error sending "
1557 "SCHEDULE_FOR_DELETION "
1567 void ctdb_local_remove_from_delete_queue(struct ctdb_db_context
*ctdb_db
,
1568 const struct ctdb_ltdb_header
*hdr
,
1571 if (ctdb_db
->ctdb
->ctdbd_pid
!= getpid()) {
1573 * Only remove the record from the delete queue if called
1574 * in the main daemon.
1579 remove_record_from_delete_queue(ctdb_db
, hdr
, key
);