4 Copyright (C) Ronnie Sahlberg 2009
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 "lib/tdb/include/tdb.h"
23 #include "system/network.h"
24 #include "system/filesys.h"
25 #include "system/dir.h"
26 #include "../include/ctdb_private.h"
28 #include "lib/util/dlinklist.h"
29 #include "lib/events/events.h"
30 #include "../include/ctdb_private.h"
31 #include "../common/rb_tree.h"
33 #define TIMELIMIT() timeval_current_ofs(10, 0)
34 #define TUNINGDBNAME "vactune.tdb"
36 enum vacuum_child_status { VACUUM_RUNNING, VACUUM_OK, VACUUM_ERROR, VACUUM_TIMEOUT};
38 struct ctdb_vacuum_child_context {
39 struct ctdb_vacuum_handle *vacuum_handle;
42 enum vacuum_child_status status;
43 struct timeval start_time;
46 struct ctdb_vacuum_handle {
47 struct ctdb_db_context *ctdb_db;
48 struct ctdb_vacuum_child_context *child_ctx;
52 /* a list of records to possibly delete */
54 uint32_t vacuum_limit;
55 uint32_t repack_limit;
56 struct ctdb_context *ctdb;
57 struct ctdb_db_context *ctdb_db;
58 struct tdb_context *dest_db;
59 trbt_tree_t *delete_tree;
60 uint32_t delete_count;
61 struct ctdb_marshall_buffer **list;
70 /* tuning information stored for every db */
71 struct vacuum_tuning_data {
72 uint32_t last_num_repack;
73 uint32_t last_num_empty;
74 uint32_t last_interval;
75 uint32_t new_interval;
76 struct timeval last_start;
80 /* this structure contains the information for one record to be deleted */
81 struct delete_record_data {
82 struct ctdb_context *ctdb;
83 struct ctdb_db_context *ctdb_db;
84 struct ctdb_ltdb_header hdr;
88 struct delete_records_list {
89 struct ctdb_marshall_buffer *records;
92 static void ctdb_vacuum_event(struct event_context *ev, struct timed_event *te,
93 struct timeval t, void *private_data);
97 * traverse function for gathering the records that can be deleted
99 static int vacuum_traverse(struct tdb_context *tdb, TDB_DATA key, TDB_DATA data, void *private)
101 struct vacuum_data *vdata = talloc_get_type(private, struct vacuum_data);
102 struct ctdb_context *ctdb = vdata->ctdb;
103 struct ctdb_db_context *ctdb_db = vdata->ctdb_db;
105 struct ctdb_ltdb_header *hdr;
106 struct ctdb_rec_data *rec;
109 lmaster = ctdb_lmaster(ctdb, &key);
110 if (lmaster >= ctdb->vnn_map->size) {
114 if (data.dsize != sizeof(struct ctdb_ltdb_header)) {
115 /* its not a deleted record */
119 hdr = (struct ctdb_ltdb_header *)data.dptr;
121 if (hdr->dmaster != ctdb->pnn) {
125 /* is this a records we could possibly delete? I.e.
126 if the record is empty and also we are both lmaster
127 and dmaster for the record we should be able to delete it
129 if (lmaster == ctdb->pnn) {
132 hash = ctdb_hash(&key);
133 if (trbt_lookup32(vdata->delete_tree, hash)) {
134 DEBUG(DEBUG_INFO, (__location__ " Hash collission when vacuuming, skipping this record.\n"));
137 struct delete_record_data *dd;
139 /* store key and header indexed by the key hash */
140 dd = talloc_zero(vdata->delete_tree, struct delete_record_data);
142 DEBUG(DEBUG_ERR,(__location__ " Out of memory\n"));
146 dd->ctdb_db = ctdb_db;
147 dd->key.dsize = key.dsize;
148 dd->key.dptr = talloc_memdup(dd, key.dptr, key.dsize);
149 if (dd->key.dptr == NULL) {
150 DEBUG(DEBUG_ERR,(__location__ " Out of memory\n"));
156 trbt_insert32(vdata->delete_tree, hash, dd);
158 vdata->delete_count++;
162 /* add the record to the blob ready to send to the nodes */
163 rec = ctdb_marshall_record(vdata->list[lmaster], ctdb->pnn, key, NULL, tdb_null);
165 DEBUG(DEBUG_ERR,(__location__ " Out of memory\n"));
166 vdata->traverse_error = true;
169 old_size = talloc_get_size(vdata->list[lmaster]);
170 vdata->list[lmaster] = talloc_realloc_size(NULL, vdata->list[lmaster],
171 old_size + rec->length);
172 if (vdata->list[lmaster] == NULL) {
173 DEBUG(DEBUG_ERR,(__location__ " Failed to expand\n"));
174 vdata->traverse_error = true;
177 vdata->list[lmaster]->count++;
178 memcpy(old_size+(uint8_t *)vdata->list[lmaster], rec, rec->length);
187 * traverse the tree of records to delete and marshall them into
190 static void delete_traverse(void *param, void *data)
192 struct delete_record_data *dd = talloc_get_type(data, struct delete_record_data);
193 struct delete_records_list *recs = talloc_get_type(param, struct delete_records_list);
194 struct ctdb_rec_data *rec;
197 rec = ctdb_marshall_record(dd, recs->records->db_id, dd->key, &dd->hdr, tdb_null);
199 DEBUG(DEBUG_ERR, (__location__ " failed to marshall record\n"));
203 old_size = talloc_get_size(recs->records);
204 recs->records = talloc_realloc_size(NULL, recs->records, old_size + rec->length);
205 if (recs->records == NULL) {
206 DEBUG(DEBUG_ERR,(__location__ " Failed to expand\n"));
209 recs->records->count++;
210 memcpy(old_size+(uint8_t *)(recs->records), rec, rec->length);
214 * read-only traverse the database in order to find
215 * records that can be deleted and try to delete these
216 * records on the other nodes
217 * this executes in the child context
219 static int ctdb_vacuum_db(struct ctdb_db_context *ctdb_db, struct vacuum_data *vdata)
221 struct ctdb_context *ctdb = ctdb_db->ctdb;
222 const char *name = ctdb_db->db_name;
225 ret = ctdb_ctrl_getvnnmap(ctdb, TIMELIMIT(), CTDB_CURRENT_NODE, ctdb, &ctdb->vnn_map);
227 DEBUG(DEBUG_ERR, ("Unable to get vnnmap from local node\n"));
231 pnn = ctdb_ctrl_getpnn(ctdb, TIMELIMIT(), CTDB_CURRENT_NODE);
233 DEBUG(DEBUG_ERR, ("Unable to get pnn from local node\n"));
238 /* the list needs to be of length num_nodes */
239 vdata->list = talloc_array(vdata, struct ctdb_marshall_buffer *, ctdb->vnn_map->size);
240 if (vdata->list == NULL) {
241 DEBUG(DEBUG_ERR,(__location__ " Out of memory\n"));
244 for (i = 0; i < ctdb->vnn_map->size; i++) {
245 vdata->list[i] = (struct ctdb_marshall_buffer *)
246 talloc_zero_size(vdata->list,
247 offsetof(struct ctdb_marshall_buffer, data));
248 if (vdata->list[i] == NULL) {
249 DEBUG(DEBUG_ERR,(__location__ " Out of memory\n"));
252 vdata->list[i]->db_id = ctdb_db->db_id;
255 /* read-only traverse, looking for records that might be able to be vacuumed */
256 if (tdb_traverse_read(ctdb_db->ltdb->tdb, vacuum_traverse, vdata) == -1 ||
257 vdata->traverse_error) {
258 DEBUG(DEBUG_ERR,(__location__ " Traverse error in vacuuming '%s'\n", name));
262 for ( i = 0; i < ctdb->vnn_map->size; i++) {
263 if (vdata->list[i]->count == 0) {
267 /* for records where we are not the lmaster, tell the lmaster to fetch the record */
268 if (ctdb->vnn_map->map[i] != ctdb->pnn) {
270 DEBUG(DEBUG_NOTICE,("Found %u records for lmaster %u in '%s'\n",
271 vdata->list[i]->count, i, name));
273 data.dsize = talloc_get_size(vdata->list[i]);
274 data.dptr = (void *)vdata->list[i];
275 if (ctdb_send_message(ctdb, ctdb->vnn_map->map[i], CTDB_SRVID_VACUUM_FETCH, data) != 0) {
276 DEBUG(DEBUG_ERR,(__location__ " Failed to send vacuum fetch message to %u\n",
277 ctdb->vnn_map->map[i]));
284 /* Process all records we can delete (if any) */
285 if (vdata->delete_count > 0) {
286 struct delete_records_list *recs;
287 TDB_DATA indata, outdata;
290 recs = talloc_zero(vdata, struct delete_records_list);
292 DEBUG(DEBUG_ERR,(__location__ " Out of memory\n"));
295 recs->records = (struct ctdb_marshall_buffer *)
296 talloc_zero_size(vdata,
297 offsetof(struct ctdb_marshall_buffer, data));
298 if (recs->records == NULL) {
299 DEBUG(DEBUG_ERR,(__location__ " Out of memory\n"));
302 recs->records->db_id = ctdb_db->db_id;
305 * traverse the tree of all records we want to delete and
306 * create a blob we can send to the other nodes.
308 trbt_traversearray32(vdata->delete_tree, 1, delete_traverse, recs);
310 indata.dsize = talloc_get_size(recs->records);
311 indata.dptr = (void *)recs->records;
314 * now tell all the other nodes to delete all these records
317 for (i = 0; i < ctdb->vnn_map->size; i++) {
318 struct ctdb_marshall_buffer *records;
319 struct ctdb_rec_data *rec;
321 if (ctdb->vnn_map->map[i] == ctdb->pnn) {
322 /* we dont delete the records on the local node just yet */
326 ret = ctdb_control(ctdb, ctdb->vnn_map->map[i], 0,
327 CTDB_CONTROL_TRY_DELETE_RECORDS, 0,
328 indata, recs, &outdata, &res,
330 if (ret != 0 || res != 0) {
331 DEBUG(DEBUG_ERR,("Failed to delete records on node %u\n", ctdb->vnn_map->map[i]));
336 * outdata countains the list of records coming back
337 * from the node which the node could not delete
339 records = (struct ctdb_marshall_buffer *)outdata.dptr;
340 rec = (struct ctdb_rec_data *)&records->data[0];
341 while (records->count-- > 1) {
342 TDB_DATA reckey, recdata;
343 struct ctdb_ltdb_header *rechdr;
345 reckey.dptr = &rec->data[0];
346 reckey.dsize = rec->keylen;
347 recdata.dptr = &rec->data[reckey.dsize];
348 recdata.dsize = rec->datalen;
350 if (recdata.dsize < sizeof(struct ctdb_ltdb_header)) {
351 DEBUG(DEBUG_CRIT,(__location__ " bad ltdb record\n"));
354 rechdr = (struct ctdb_ltdb_header *)recdata.dptr;
355 recdata.dptr += sizeof(*rechdr);
356 recdata.dsize -= sizeof(*rechdr);
359 * that other node couldnt delete the record
360 * so we should delete it and thereby remove it from the tree
362 talloc_free(trbt_lookup32(vdata->delete_tree, ctdb_hash(&reckey)));
364 rec = (struct ctdb_rec_data *)(rec->length + (uint8_t *)rec);
369 * The only records remaining in the tree would be those
370 * records where all other nodes could successfully
371 * delete them, so we can safely delete them on the
372 * lmaster as well. Deletion implictely happens while
373 * we repack the database. The repack algorithm revisits
374 * the tree in order to find the records that don't need
375 * to be copied / repacked.
379 /* this ensures we run our event queue */
380 ctdb_ctrl_getpnn(ctdb, TIMELIMIT(), CTDB_CURRENT_NODE);
387 * traverse function for repacking
389 static int repack_traverse(struct tdb_context *tdb, TDB_DATA key, TDB_DATA data, void *private)
391 struct vacuum_data *vdata = (struct vacuum_data *)private;
394 uint32_t hash = ctdb_hash(&key);
395 struct delete_record_data *kd;
397 * check if we can ignore this record because it's in the delete_tree
399 kd = (struct delete_record_data *)trbt_lookup32(vdata->delete_tree, hash);
401 * there might be hash collisions so we have to compare the keys here to be sure
403 if (kd && kd->key.dsize == key.dsize && memcmp(kd->key.dptr, key.dptr, key.dsize) == 0) {
408 if (tdb_store(vdata->dest_db, key, data, TDB_INSERT) != 0) {
409 vdata->traverse_error = true;
419 static int ctdb_repack_tdb(struct tdb_context *tdb, TALLOC_CTX *mem_ctx, struct vacuum_data *vdata)
421 struct tdb_context *tmp_db;
423 if (tdb_transaction_start(tdb) != 0) {
424 DEBUG(DEBUG_ERR,(__location__ " Failed to start transaction\n"));
428 tmp_db = tdb_open("tmpdb", tdb_hash_size(tdb), TDB_INTERNAL, O_RDWR|O_CREAT, 0);
429 if (tmp_db == NULL) {
430 DEBUG(DEBUG_ERR,(__location__ " Failed to create tmp_db\n"));
431 tdb_transaction_cancel(tdb);
435 vdata->traverse_error = false;
436 vdata->dest_db = tmp_db;
437 vdata->vacuum = true;
442 * repack and vacuum on-the-fly by not writing the records that are
445 if (tdb_traverse_read(tdb, repack_traverse, vdata) == -1) {
446 DEBUG(DEBUG_ERR,(__location__ " Failed to traverse copying out\n"));
447 tdb_transaction_cancel(tdb);
452 DEBUG(DEBUG_NOTICE,(__location__ " %u records vacuumed\n", vdata->vacuumed));
454 if (vdata->traverse_error) {
455 DEBUG(DEBUG_ERR,(__location__ " Error during traversal\n"));
456 tdb_transaction_cancel(tdb);
461 if (tdb_wipe_all(tdb) != 0) {
462 DEBUG(DEBUG_ERR,(__location__ " Failed to wipe database\n"));
463 tdb_transaction_cancel(tdb);
468 vdata->traverse_error = false;
469 vdata->dest_db = tdb;
470 vdata->vacuum = false;
473 if (tdb_traverse_read(tmp_db, repack_traverse, vdata) == -1) {
474 DEBUG(DEBUG_ERR,(__location__ " Failed to traverse copying back\n"));
475 tdb_transaction_cancel(tdb);
480 if (vdata->traverse_error) {
481 DEBUG(DEBUG_ERR,(__location__ " Error during second traversal\n"));
482 tdb_transaction_cancel(tdb);
490 if (tdb_transaction_commit(tdb) != 0) {
491 DEBUG(DEBUG_ERR,(__location__ " Failed to commit\n"));
494 DEBUG(DEBUG_NOTICE,(__location__ " %u records copied\n", vdata->copied));
499 static int update_tuning_db(struct ctdb_db_context *ctdb_db, struct vacuum_data *vdata, uint32_t freelist)
501 TALLOC_CTX *tmp_ctx = talloc_new(NULL);
502 TDB_CONTEXT *tune_tdb;
504 struct vacuum_tuning_data tdata;
505 struct vacuum_tuning_data *tptr;
508 vac_dbname = talloc_asprintf(tmp_ctx, "%s/%s.%u",
509 ctdb_db->ctdb->db_directory,
510 TUNINGDBNAME, ctdb_db->ctdb->pnn);
511 if (vac_dbname == NULL) {
512 DEBUG(DEBUG_CRIT,(__location__ " Out of memory error while allocating '%s'\n", vac_dbname));
513 talloc_free(tmp_ctx);
517 tune_tdb = tdb_open(vac_dbname, 0, 0, O_RDWR|O_CREAT, 0600);
518 if (tune_tdb == NULL) {
519 DEBUG(DEBUG_ERR,(__location__ " Failed to create/open %s\n", TUNINGDBNAME));
520 talloc_free(tmp_ctx);
524 if (tdb_transaction_start(tune_tdb) != 0) {
525 DEBUG(DEBUG_ERR,(__location__ " Failed to start transaction\n"));
529 key.dptr = discard_const(ctdb_db->db_name);
530 key.dsize = strlen(ctdb_db->db_name);
531 value = tdb_fetch(tune_tdb, key);
533 if (value.dptr != NULL && value.dsize == sizeof(struct vacuum_tuning_data)) {
534 tptr = (struct vacuum_tuning_data *)value.dptr;
538 * re-calc new vacuum interval:
539 * in case no limit was reached we continously increase the interval
540 * until vacuum_max_interval is reached
541 * in case a limit was reached we divide the current interval by 2
542 * unless vacuum_min_interval is reached
544 if (freelist < vdata->repack_limit &&
545 vdata->delete_count < vdata->vacuum_limit) {
546 if (tdata.last_interval < ctdb_db->ctdb->tunable.vacuum_max_interval) {
547 tdata.new_interval = tdata.last_interval * 110 / 100;
548 DEBUG(DEBUG_NOTICE,("Increasing vacuum interval %u -> %u for %s\n",
549 tdata.last_interval, tdata.new_interval, ctdb_db->db_name));
552 tdata.new_interval = tdata.last_interval / 2;
553 if (tdata.new_interval < ctdb_db->ctdb->tunable.vacuum_min_interval ||
554 tdata.new_interval > ctdb_db->ctdb->tunable.vacuum_max_interval) {
555 tdata.new_interval = ctdb_db->ctdb->tunable.vacuum_min_interval;
557 DEBUG(DEBUG_ERR,("Decreasing vacuum interval %u -> %u for %s\n",
558 tdata.last_interval, tdata.new_interval, ctdb_db->db_name));
560 tdata.last_interval = tdata.new_interval;
562 DEBUG(DEBUG_ERR,(__location__ " Cannot find tunedb record for %s. Using default interval\n", ctdb_db->db_name));
563 tdata.last_num_repack = freelist;
564 tdata.last_num_empty = vdata->delete_count;
565 tdata.last_interval = ctdb_db->ctdb->tunable.vacuum_default_interval;
568 if (value.dptr != NULL) {
572 tdata.last_start = vdata->start;
573 tdata.last_duration = timeval_elapsed(&vdata->start);
575 value.dptr = (unsigned char *)&tdata;
576 value.dsize = sizeof(tdata);
578 if (tdb_store(tune_tdb, key, value, 0) != 0) {
579 DEBUG(DEBUG_ERR,(__location__ " Unable to store tundb record for %s\n", ctdb_db->db_name));
580 tdb_transaction_cancel(tune_tdb);
582 talloc_free(tmp_ctx);
585 tdb_transaction_commit(tune_tdb);
587 talloc_free(tmp_ctx);
593 * repack and vaccum a db
594 * called from the child context
596 static int ctdb_repack_db(struct ctdb_db_context *ctdb_db, TALLOC_CTX *mem_ctx)
598 uint32_t repack_limit = ctdb_db->ctdb->tunable.repack_limit;
599 uint32_t vacuum_limit = ctdb_db->ctdb->tunable.vacuum_limit;
600 const char *name = ctdb_db->db_name;
602 struct vacuum_data *vdata;
604 size = tdb_freelist_size(ctdb_db->ltdb->tdb);
606 DEBUG(DEBUG_ERR,(__location__ " Failed to get freelist size for '%s'\n", name));
610 vdata = talloc_zero(mem_ctx, struct vacuum_data);
612 DEBUG(DEBUG_ERR,(__location__ " Out of memory\n"));
616 vdata->ctdb = ctdb_db->ctdb;
617 vdata->vacuum_limit = vacuum_limit;
618 vdata->repack_limit = repack_limit;
619 vdata->delete_tree = trbt_create(vdata, 0);
620 if (vdata->delete_tree == NULL) {
621 DEBUG(DEBUG_ERR,(__location__ " Out of memory\n"));
626 vdata->start = timeval_current();
629 * gather all records that can be deleted in vdata
631 if (ctdb_vacuum_db(ctdb_db, vdata) != 0) {
632 DEBUG(DEBUG_ERR,(__location__ " Failed to vacuum '%s'\n", name));
636 * decide if a repack is necessary
638 if (size < repack_limit && vdata->delete_count < vacuum_limit) {
640 update_tuning_db(ctdb_db, vdata, size);
644 DEBUG(DEBUG_NOTICE,("Repacking %s with %u freelist entries and %u records to delete\n",
645 name, size, vdata->delete_count));
648 * repack and implicitely get rid of the records we can delete
650 if (ctdb_repack_tdb(ctdb_db->ltdb->tdb, mem_ctx, vdata) != 0) {
651 DEBUG(DEBUG_ERR,(__location__ " Failed to repack '%s'\n", name));
652 update_tuning_db(ctdb_db, vdata, size);
656 update_tuning_db(ctdb_db, vdata, size);
662 static int get_vacuum_interval(struct ctdb_db_context *ctdb_db)
664 TALLOC_CTX *tmp_ctx = talloc_new(NULL);
668 uint interval = ctdb_db->ctdb->tunable.vacuum_default_interval;
669 struct ctdb_context *ctdb = ctdb_db->ctdb;
671 vac_dbname = talloc_asprintf(tmp_ctx, "%s/%s.%u", ctdb->db_directory, TUNINGDBNAME, ctdb->pnn);
672 if (vac_dbname == NULL) {
673 DEBUG(DEBUG_CRIT,(__location__ " Out of memory error while allocating '%s'\n", vac_dbname));
674 talloc_free(tmp_ctx);
678 tdb = tdb_open(vac_dbname, 0, 0, O_RDONLY, 0600);
680 DEBUG(DEBUG_ERR,("Unable to open database %s using default interval\n", vac_dbname));
681 talloc_free(tmp_ctx);
685 key.dptr = discard_const(ctdb_db->db_name);
686 key.dsize = strlen(ctdb_db->db_name);
688 value = tdb_fetch(tdb, key);
690 if (value.dptr != NULL) {
691 if (value.dsize == sizeof(struct vacuum_tuning_data)) {
692 struct vacuum_tuning_data *tptr = (struct vacuum_tuning_data *)value.dptr;
694 interval = tptr->new_interval;
696 if (interval < ctdb->tunable.vacuum_min_interval) {
697 interval = ctdb->tunable.vacuum_min_interval;
699 if (interval > ctdb->tunable.vacuum_max_interval) {
700 interval = ctdb->tunable.vacuum_max_interval;
705 DEBUG(DEBUG_NOTICE,("Using new interval %u for database %s\n", interval, ctdb_db->db_name));
709 talloc_free(tmp_ctx);
714 static int vacuum_child_destructor(struct ctdb_vacuum_child_context *child_ctx)
716 double l = timeval_elapsed(&child_ctx->start_time);
717 struct ctdb_db_context *ctdb_db = child_ctx->vacuum_handle->ctdb_db;
718 struct ctdb_context *ctdb = ctdb_db->ctdb;
720 DEBUG(DEBUG_ERR,("Vacuuming took %.3f seconds for database %s\n", l, ctdb_db->db_name));
722 if (child_ctx->child_pid != -1) {
723 kill(child_ctx->child_pid, SIGKILL);
726 event_add_timed(ctdb->ev, child_ctx->vacuum_handle,
727 timeval_current_ofs(get_vacuum_interval(ctdb_db), 0),
728 ctdb_vacuum_event, child_ctx->vacuum_handle);
734 * this event is generated when a vacuum child process times out
736 static void vacuum_child_timeout(struct event_context *ev, struct timed_event *te,
737 struct timeval t, void *private_data)
739 struct ctdb_vacuum_child_context *child_ctx = talloc_get_type(private_data, struct ctdb_vacuum_child_context);
741 DEBUG(DEBUG_ERR,("Vacuuming child process timed out for db %s\n", child_ctx->vacuum_handle->ctdb_db->db_name));
743 child_ctx->status = VACUUM_TIMEOUT;
745 talloc_free(child_ctx);
750 * this event is generated when a vacuum child process has completed
752 static void vacuum_child_handler(struct event_context *ev, struct fd_event *fde,
753 uint16_t flags, void *private_data)
755 struct ctdb_vacuum_child_context *child_ctx = talloc_get_type(private_data, struct ctdb_vacuum_child_context);
759 DEBUG(DEBUG_NOTICE,("Vacuuming child finished for db %s\n", child_ctx->vacuum_handle->ctdb_db->db_name));
760 child_ctx->child_pid = -1;
762 ret = read(child_ctx->fd[0], &c, 1);
763 if (ret != 1 || c != 0) {
764 child_ctx->status = VACUUM_ERROR;
765 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));
767 child_ctx->status = VACUUM_OK;
770 talloc_free(child_ctx);
774 * this event is called every time we need to start a new vacuum process
777 ctdb_vacuum_event(struct event_context *ev, struct timed_event *te,
778 struct timeval t, void *private_data)
780 struct ctdb_vacuum_handle *vacuum_handle = talloc_get_type(private_data, struct ctdb_vacuum_handle);
781 struct ctdb_db_context *ctdb_db = vacuum_handle->ctdb_db;
782 struct ctdb_context *ctdb = ctdb_db->ctdb;
783 struct ctdb_vacuum_child_context *child_ctx;
786 /* we dont vacuum if we are in recovery mode */
787 if (ctdb->recovery_mode == CTDB_RECOVERY_ACTIVE) {
788 event_add_timed(ctdb->ev, vacuum_handle, timeval_current_ofs(ctdb->tunable.vacuum_default_interval, 0), ctdb_vacuum_event, vacuum_handle);
792 DEBUG(DEBUG_NOTICE,("Start a vacuuming child process for db %s\n", ctdb_db->db_name));
794 child_ctx = talloc(vacuum_handle, struct ctdb_vacuum_child_context);
795 if (child_ctx == NULL) {
796 DEBUG(DEBUG_CRIT, (__location__ " Failed to allocate child context for vacuuming of %s\n", ctdb_db->db_name));
797 ctdb_fatal(ctdb, "Out of memory when crating vacuum child context. Shutting down\n");
801 ret = pipe(child_ctx->fd);
803 talloc_free(child_ctx);
804 DEBUG(DEBUG_ERR, ("Failed to create pipe for vacuum child process.\n"));
805 event_add_timed(ctdb->ev, vacuum_handle, timeval_current_ofs(ctdb->tunable.vacuum_default_interval, 0), ctdb_vacuum_event, vacuum_handle);
809 child_ctx->child_pid = fork();
810 if (child_ctx->child_pid == (pid_t)-1) {
811 close(child_ctx->fd[0]);
812 close(child_ctx->fd[1]);
813 talloc_free(child_ctx);
814 DEBUG(DEBUG_ERR, ("Failed to fork vacuum child process.\n"));
815 event_add_timed(ctdb->ev, vacuum_handle, timeval_current_ofs(ctdb->tunable.vacuum_default_interval, 0), ctdb_vacuum_event, vacuum_handle);
820 if (child_ctx->child_pid == 0) {
822 close(child_ctx->fd[0]);
824 if (switch_from_server_to_client(ctdb) != 0) {
825 DEBUG(DEBUG_CRIT, (__location__ "ERROR: failed to switch vacuum daemon into client mode. Shutting down.\n"));
832 cc = ctdb_repack_db(ctdb_db, child_ctx);
834 write(child_ctx->fd[1], &cc, 1);
838 set_close_on_exec(child_ctx->fd[0]);
839 close(child_ctx->fd[1]);
841 child_ctx->status = VACUUM_RUNNING;
842 child_ctx->start_time = timeval_current();
844 talloc_set_destructor(child_ctx, vacuum_child_destructor);
846 event_add_timed(ctdb->ev, child_ctx,
847 timeval_current_ofs(ctdb->tunable.vacuum_max_run_time, 0),
848 vacuum_child_timeout, child_ctx);
850 DEBUG(DEBUG_NOTICE, (__location__ " Created PIPE FD:%d to child vacuum process\n", child_ctx->fd[0]));
852 event_add_fd(ctdb->ev, child_ctx, child_ctx->fd[0],
853 EVENT_FD_READ|EVENT_FD_AUTOCLOSE,
854 vacuum_child_handler,
857 vacuum_handle->child_ctx = child_ctx;
858 child_ctx->vacuum_handle = vacuum_handle;
862 /* this function initializes the vacuuming context for a database
863 * starts the vacuuming events
865 int ctdb_vacuum_init(struct ctdb_db_context *ctdb_db)
867 ctdb_db->vacuum_handle = talloc(ctdb_db, struct ctdb_vacuum_handle);
868 CTDB_NO_MEMORY(ctdb_db->ctdb, ctdb_db->vacuum_handle);
870 ctdb_db->vacuum_handle->ctdb_db = ctdb_db;
872 event_add_timed(ctdb_db->ctdb->ev, ctdb_db->vacuum_handle,
873 timeval_current_ofs(get_vacuum_interval(ctdb_db), 0),
874 ctdb_vacuum_event, ctdb_db->vacuum_handle);