2 ctdb control tool - database vacuum
4 Copyright (C) Andrew Tridgell 2008
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/network.h"
24 #include "../include/ctdb.h"
25 #include "../include/ctdb_private.h"
28 /* should be tunable */
29 #define TIMELIMIT() timeval_current_ofs(10, 0)
36 static void async_callback(struct ctdb_client_control_state *state)
38 struct async_data *data = talloc_get_type(state->async.private_data, struct async_data);
42 /* one more node has responded with recmode data */
45 /* if we failed to push the db, then return an error and let
46 the main loop try again.
48 if (state->state != CTDB_CONTROL_DONE) {
53 state->async.fn = NULL;
55 ret = ctdb_control_recv(state->ctdb, state, data, NULL, &res, NULL);
56 if ((ret != 0) || (res != 0)) {
61 static void async_add(struct async_data *data, struct ctdb_client_control_state *state)
63 /* set up the callback functions */
64 state->async.fn = async_callback;
65 state->async.private_data = data;
67 /* one more control to wait for to complete */
72 /* wait for up to the maximum number of seconds allowed
73 or until all nodes we expect a response from has replied
75 static int async_wait(struct ctdb_context *ctdb, struct async_data *data)
77 while (data->count > 0) {
78 event_loop_once(ctdb->ev);
80 if (data->fail_count != 0) {
87 perform a simple control on nodes in the vnn map except ourselves.
88 The control cannot return data
90 static int async_control_on_vnnmap(struct ctdb_context *ctdb, enum ctdb_controls opcode,
93 struct async_data *async_data;
94 struct ctdb_client_control_state *state;
96 struct timeval timeout = TIMELIMIT();
98 async_data = talloc_zero(ctdb, struct async_data);
99 CTDB_NO_MEMORY_FATAL(ctdb, async_data);
101 /* loop over all active nodes and send an async control to each of them */
102 for (j=0; j<ctdb->vnn_map->size; j++) {
103 uint32_t pnn = ctdb->vnn_map->map[j];
104 if (pnn == ctdb->pnn) {
107 state = ctdb_control_send(ctdb, pnn, 0, opcode,
108 0, data, async_data, NULL, &timeout, NULL);
110 DEBUG(0,(__location__ " Failed to call async control %u\n", (unsigned)opcode));
111 talloc_free(async_data);
115 async_add(async_data, state);
118 if (async_wait(ctdb, async_data) != 0) {
119 talloc_free(async_data);
123 talloc_free(async_data);
131 static int ctdb_vacuum_one(struct ctdb_context *ctdb, TDB_DATA key,
132 struct ctdb_db_context *ctdb_db, uint32_t *count)
135 struct ctdb_ltdb_header *hdr;
136 struct ctdb_rec_data *rec;
139 if (tdb_chainlock_nonblock(ctdb_db->ltdb->tdb, key) != 0) {
140 /* the chain is busy - come back later */
144 data = tdb_fetch(ctdb_db->ltdb->tdb, key);
145 tdb_chainunlock(ctdb_db->ltdb->tdb, key);
146 if (data.dptr == NULL) {
149 if (data.dsize != sizeof(struct ctdb_ltdb_header)) {
155 hdr = (struct ctdb_ltdb_header *)data.dptr;
158 /* if we are not the lmaster and the dmaster then skip the record */
159 if (hdr->dmaster != ctdb->pnn ||
160 ctdb_lmaster(ctdb, &key) != ctdb->pnn) {
165 rec = ctdb_marshall_record(ctdb, ctdb_db->db_id, key, hdr, tdb_null);
168 /* try it again later */
172 data.dptr = (void *)rec;
173 data.dsize = rec->length;
175 if (async_control_on_vnnmap(ctdb, CTDB_CONTROL_DELETE_RECORD, data) != 0) {
176 /* one or more nodes failed to delete a record - no problem! */
183 /* its deleted on all other nodes - refetch, check and delete */
184 if (tdb_chainlock_nonblock(ctdb_db->ltdb->tdb, key) != 0) {
185 /* the chain is busy - come back later */
189 data = tdb_fetch(ctdb_db->ltdb->tdb, key);
190 if (data.dptr == NULL) {
191 tdb_chainunlock(ctdb_db->ltdb->tdb, key);
194 if (data.dsize != sizeof(struct ctdb_ltdb_header)) {
196 tdb_chainunlock(ctdb_db->ltdb->tdb, key);
200 hdr = (struct ctdb_ltdb_header *)data.dptr;
202 /* if we are not the lmaster and the dmaster then skip the record */
203 if (hdr->dmaster != ctdb->pnn ||
204 ctdb_lmaster(ctdb, &key) != ctdb->pnn ||
206 tdb_chainunlock(ctdb_db->ltdb->tdb, key);
211 ctdb_block_signal(SIGALRM);
212 tdb_delete(ctdb_db->ltdb->tdb, key);
213 ctdb_unblock_signal(SIGALRM);
214 tdb_chainunlock(ctdb_db->ltdb->tdb, key);
224 vacuum records for which we are the lmaster
226 static int ctdb_vacuum_local(struct ctdb_context *ctdb, struct ctdb_control_pulldb_reply *list,
227 struct ctdb_db_context *ctdb_db, uint32_t *count)
229 struct ctdb_rec_data *r;
232 r = (struct ctdb_rec_data *)&list->data[0];
236 r = (struct ctdb_rec_data *)(r->length + (uint8_t *)r), i++) {
238 key.dptr = &r->data[0];
239 key.dsize = r->keylen;
240 if (ctdb_vacuum_one(ctdb, key, ctdb_db, count) != 0) {
249 a list of records to possibly delete
252 uint32_t vacuum_limit;
253 struct ctdb_context *ctdb;
254 struct ctdb_control_pulldb_reply **list;
260 traverse function for vacuuming
262 static int vacuum_traverse(struct tdb_context *tdb, TDB_DATA key, TDB_DATA data, void *private)
264 struct vacuum_data *vdata = talloc_get_type(private, struct vacuum_data);
266 struct ctdb_ltdb_header *hdr;
267 struct ctdb_rec_data *rec;
270 lmaster = ctdb_lmaster(vdata->ctdb, &key);
271 if (lmaster >= vdata->ctdb->vnn_map->size) {
275 if (data.dsize != sizeof(struct ctdb_ltdb_header)) {
276 /* its not a deleted record */
280 hdr = (struct ctdb_ltdb_header *)data.dptr;
282 if (hdr->dmaster != vdata->ctdb->pnn) {
287 /* add the record to the blob ready to send to the nodes */
288 rec = ctdb_marshall_record(vdata->list[lmaster], vdata->ctdb->pnn, key, NULL, tdb_null);
290 DEBUG(0,(__location__ " Out of memory\n"));
291 vdata->traverse_error = true;
294 old_size = talloc_get_size(vdata->list[lmaster]);
295 vdata->list[lmaster] = talloc_realloc_size(NULL, vdata->list[lmaster],
296 old_size + rec->length);
297 if (vdata->list[lmaster] == NULL) {
298 DEBUG(0,(__location__ " Failed to expand\n"));
299 vdata->traverse_error = true;
302 vdata->list[lmaster]->count++;
303 memcpy(old_size+(uint8_t *)vdata->list[lmaster], rec, rec->length);
308 /* don't gather too many records */
309 if (vdata->vacuum_limit != 0 &&
310 vdata->total == vdata->vacuum_limit) {
318 /* vacuum one database */
319 static int ctdb_vacuum_db(struct ctdb_context *ctdb, uint32_t db_id, struct ctdb_node_map *map,
320 bool persistent, uint32_t vacuum_limit)
322 struct ctdb_db_context *ctdb_db;
324 struct vacuum_data *vdata;
327 vdata = talloc_zero(ctdb, struct vacuum_data);
329 DEBUG(0,(__location__ " Out of memory\n"));
334 vdata->vacuum_limit = vacuum_limit;
336 if (ctdb_ctrl_getdbname(ctdb, TIMELIMIT(), CTDB_CURRENT_NODE, db_id, vdata, &name) != 0) {
337 DEBUG(0,(__location__ " Failed to get name of db 0x%x\n", db_id));
342 ctdb_db = ctdb_attach(ctdb, name, persistent);
343 if (ctdb_db == NULL) {
344 DEBUG(0,(__location__ " Failed to attach to database '%s'\n", name));
349 /* the list needs to be of length num_nodes */
350 vdata->list = talloc_array(vdata, struct ctdb_control_pulldb_reply *, ctdb->vnn_map->size);
351 if (vdata->list == NULL) {
352 DEBUG(0,(__location__ " Out of memory\n"));
356 for (i=0;i<ctdb->vnn_map->size;i++) {
357 vdata->list[i] = (struct ctdb_control_pulldb_reply *)
358 talloc_zero_size(vdata->list,
359 offsetof(struct ctdb_control_pulldb_reply, data));
360 if (vdata->list[i] == NULL) {
361 DEBUG(0,(__location__ " Out of memory\n"));
365 vdata->list[i]->db_id = db_id;
368 /* traverse, looking for records that might be able to be vacuumed */
369 if (tdb_traverse_read(ctdb_db->ltdb->tdb, vacuum_traverse, vdata) == -1 ||
370 vdata->traverse_error) {
371 DEBUG(0,(__location__ " Traverse error in vacuuming '%s'\n", name));
377 for (i=0;i<ctdb->vnn_map->size;i++) {
378 if (vdata->list[i]->count == 0) {
382 /* for records where we are not the lmaster, tell the lmaster to fetch the record */
383 if (ctdb->vnn_map->map[i] != ctdb->pnn) {
385 printf("Found %u records for lmaster %u in '%s'\n", vdata->list[i]->count, i, name);
387 data.dsize = talloc_get_size(vdata->list[i]);
388 data.dptr = (void *)vdata->list[i];
389 if (ctdb_send_message(ctdb, ctdb->vnn_map->map[i], CTDB_SRVID_VACUUM_FETCH, data) != 0) {
390 DEBUG(0,(__location__ " Failed to send vacuum fetch message to %u\n",
391 ctdb->vnn_map->map[i]));
399 for (i=0;i<ctdb->vnn_map->size;i++) {
402 if (vdata->list[i]->count == 0) {
406 /* for records where we are the lmaster, we can try to delete them */
407 if (ctdb_vacuum_local(ctdb, vdata->list[i], ctdb_db, &count) != 0) {
408 DEBUG(0,(__location__ " Deletion error in vacuuming '%s'\n", name));
413 printf("Deleted %u records on this node from '%s'\n", count, name);
417 /* this ensures we run our event queue */
418 ctdb_ctrl_getpnn(ctdb, TIMELIMIT(), CTDB_CURRENT_NODE);
427 vacuum all our databases
429 int ctdb_vacuum(struct ctdb_context *ctdb, int argc, const char **argv)
431 struct ctdb_dbid_map *dbmap=NULL;
432 struct ctdb_node_map *nodemap=NULL;
434 uint32_t vacuum_limit = 0;
437 vacuum_limit = atoi(argv[0]);
440 ret = ctdb_ctrl_getdbmap(ctdb, TIMELIMIT(), CTDB_CURRENT_NODE, ctdb, &dbmap);
442 DEBUG(0, ("Unable to get dbids from local node\n"));
446 ret = ctdb_ctrl_getnodemap(ctdb, TIMELIMIT(), CTDB_CURRENT_NODE, ctdb, &nodemap);
448 DEBUG(0, ("Unable to get nodemap from local node\n"));
452 ret = ctdb_ctrl_getvnnmap(ctdb, TIMELIMIT(), CTDB_CURRENT_NODE, ctdb, &ctdb->vnn_map);
454 DEBUG(0, ("Unable to get vnnmap from local node\n"));
458 pnn = ctdb_ctrl_getpnn(ctdb, TIMELIMIT(), CTDB_CURRENT_NODE);
460 DEBUG(0, ("Unable to get pnn from local node\n"));
465 for (i=0;i<dbmap->num;i++) {
466 if (ctdb_vacuum_db(ctdb, dbmap->dbs[i].dbid, nodemap,
467 dbmap->dbs[i].persistent, vacuum_limit) != 0) {
468 DEBUG(0,("Failed to vacuum db 0x%x\n", dbmap->dbs[i].dbid));
476 struct traverse_state {
478 struct tdb_context *dest_db;
482 traverse function for repacking
484 static int repack_traverse(struct tdb_context *tdb, TDB_DATA key, TDB_DATA data, void *private)
486 struct traverse_state *state = (struct traverse_state *)private;
487 if (tdb_store(state->dest_db, key, data, TDB_INSERT) != 0) {
497 static int ctdb_repack_tdb(struct tdb_context *tdb)
499 struct tdb_context *tmp_db;
500 struct traverse_state state;
502 if (tdb_transaction_start(tdb) != 0) {
503 DEBUG(0,(__location__ " Failed to start transaction\n"));
507 tmp_db = tdb_open("tmpdb", tdb_hash_size(tdb), TDB_INTERNAL, O_RDWR|O_CREAT, 0);
508 if (tmp_db == NULL) {
509 DEBUG(0,(__location__ " Failed to create tmp_db\n"));
510 tdb_transaction_cancel(tdb);
515 state.dest_db = tmp_db;
517 if (tdb_traverse_read(tdb, repack_traverse, &state) == -1) {
518 DEBUG(0,(__location__ " Failed to traverse copying out\n"));
519 tdb_transaction_cancel(tdb);
525 DEBUG(0,(__location__ " Error during traversal\n"));
526 tdb_transaction_cancel(tdb);
531 if (tdb_wipe_all(tdb) != 0) {
532 DEBUG(0,(__location__ " Failed to wipe database\n"));
533 tdb_transaction_cancel(tdb);
541 if (tdb_traverse_read(tmp_db, repack_traverse, &state) == -1) {
542 DEBUG(0,(__location__ " Failed to traverse copying back\n"));
543 tdb_transaction_cancel(tdb);
549 DEBUG(0,(__location__ " Error during second traversal\n"));
550 tdb_transaction_cancel(tdb);
557 if (tdb_transaction_commit(tdb) != 0) {
558 DEBUG(0,(__location__ " Failed to commit\n"));
566 /* repack one database */
567 static int ctdb_repack_db(struct ctdb_context *ctdb, uint32_t db_id,
568 bool persistent, uint32_t repack_limit)
570 struct ctdb_db_context *ctdb_db;
574 if (ctdb_ctrl_getdbname(ctdb, TIMELIMIT(), CTDB_CURRENT_NODE, db_id, ctdb, &name) != 0) {
575 DEBUG(0,(__location__ " Failed to get name of db 0x%x\n", db_id));
579 ctdb_db = ctdb_attach(ctdb, name, persistent);
580 if (ctdb_db == NULL) {
581 DEBUG(0,(__location__ " Failed to attach to database '%s'\n", name));
585 size = tdb_freelist_size(ctdb_db->ltdb->tdb);
587 DEBUG(0,(__location__ " Failed to get freelist size for '%s'\n", name));
591 if (size <= repack_limit) {
595 printf("Repacking %s with %u freelist entries\n", name, size);
597 if (ctdb_repack_tdb(ctdb_db->ltdb->tdb) != 0) {
598 DEBUG(0,(__location__ " Failed to repack '%s'\n", name));
607 repack all our databases
609 int ctdb_repack(struct ctdb_context *ctdb, int argc, const char **argv)
611 struct ctdb_dbid_map *dbmap=NULL;
613 /* a reasonable default limit to prevent us using too much memory */
614 uint32_t repack_limit = 10000;
617 repack_limit = atoi(argv[0]);
620 ret = ctdb_ctrl_getdbmap(ctdb, TIMELIMIT(), CTDB_CURRENT_NODE, ctdb, &dbmap);
622 DEBUG(0, ("Unable to get dbids from local node\n"));
626 for (i=0;i<dbmap->num;i++) {
627 if (ctdb_repack_db(ctdb, dbmap->dbs[i].dbid,
628 dbmap->dbs[i].persistent, repack_limit) != 0) {
629 DEBUG(0,("Failed to repack db 0x%x\n", dbmap->dbs[i].dbid));