along with this program; if not, see <http://www.gnu.org/licenses/>.
*/
-#include "includes.h"
-#include "tdb.h"
+#include "replace.h"
#include "system/network.h"
#include "system/filesys.h"
-#include "system/dir.h"
-#include "../include/ctdb_private.h"
-#include "db_wrap.h"
+#include "system/time.h"
+
+#include <talloc.h>
+#include <tevent.h>
+
+#include "lib/tdb_wrap/tdb_wrap.h"
#include "lib/util/dlinklist.h"
-#include "../include/ctdb_private.h"
-#include "../common/rb_tree.h"
+#include "lib/util/debug.h"
+#include "lib/util/samba_util.h"
+#include "lib/util/util_process.h"
+
+#include "ctdb_private.h"
+#include "ctdb_client.h"
+
+#include "common/rb_tree.h"
+#include "common/system.h"
+#include "common/common.h"
+#include "common/logging.h"
#define TIMELIMIT() timeval_current_ofs(10, 0)
/* a list of records to possibly delete */
struct vacuum_data {
- uint32_t repack_limit;
struct ctdb_context *ctdb;
struct ctdb_db_context *ctdb_db;
struct tdb_context *dest_db;
trbt_tree_t *delete_list;
- uint32_t delete_count;
struct ctdb_marshall_buffer **vacuum_fetch_list;
struct timeval start;
bool traverse_error;
bool vacuum;
- uint32_t total;
- uint32_t vacuumed;
- uint32_t copied;
- uint32_t fast_added_to_vacuum_fetch_list;
- uint32_t fast_added_to_delete_list;
- uint32_t fast_deleted;
- uint32_t fast_skipped;
- uint32_t fast_error;
- uint32_t fast_total;
- uint32_t full_scheduled;
- uint32_t full_skipped;
- uint32_t full_error;
- uint32_t full_total;
- uint32_t delete_left;
- uint32_t delete_remote_error;
- uint32_t delete_local_error;
- uint32_t delete_deleted;
- uint32_t delete_skipped;
+ struct {
+ struct {
+ uint32_t added_to_vacuum_fetch_list;
+ uint32_t added_to_delete_list;
+ uint32_t deleted;
+ uint32_t skipped;
+ uint32_t error;
+ uint32_t total;
+ } delete_queue;
+ struct {
+ uint32_t scheduled;
+ uint32_t skipped;
+ uint32_t error;
+ uint32_t total;
+ } db_traverse;
+ struct {
+ uint32_t total;
+ uint32_t remote_error;
+ uint32_t local_error;
+ uint32_t deleted;
+ uint32_t skipped;
+ uint32_t left;
+ } delete_list;
+ struct {
+ uint32_t vacuumed;
+ uint32_t copied;
+ } repack;
+ } count;
};
/* this structure contains the information for one record to be deleted */
return -1;
}
- vdata->delete_count++;
+ vdata->count.delete_list.total++;
return 0;
}
TDB_DATA key)
{
struct ctdb_context *ctdb = vdata->ctdb;
- struct ctdb_rec_data *rec;
uint32_t lmaster;
- size_t old_size;
struct ctdb_marshall_buffer *vfl;
lmaster = ctdb_lmaster(ctdb, &key);
vfl = vdata->vacuum_fetch_list[lmaster];
- rec = ctdb_marshall_record(vfl, ctdb->pnn, key, NULL, tdb_null);
- if (rec == NULL) {
+ vfl = ctdb_marshall_add(ctdb, vfl, vfl->db_id, ctdb->pnn,
+ key, NULL, tdb_null);
+ if (vfl == NULL) {
DEBUG(DEBUG_ERR,(__location__ " Out of memory\n"));
vdata->traverse_error = true;
return -1;
}
- old_size = talloc_get_size(vfl);
- vfl = talloc_realloc_size(NULL, vfl, old_size + rec->length);
- if (vfl == NULL) {
- DEBUG(DEBUG_ERR,(__location__ " Failed to expand\n"));
- vdata->traverse_error = true;
- return -1;
- }
vdata->vacuum_fetch_list[lmaster] = vfl;
- vfl->count++;
- memcpy(old_size+(uint8_t *)vfl, rec, rec->length);
- talloc_free(rec);
-
- vdata->total++;
-
return 0;
}
-static void ctdb_vacuum_event(struct event_context *ev, struct timed_event *te,
+static void ctdb_vacuum_event(struct tevent_context *ev,
+ struct tevent_timer *te,
struct timeval t, void *private_data);
static int vacuum_record_parser(TDB_DATA key, TDB_DATA data, void *private_data)
struct ctdb_ltdb_header *hdr;
int res = 0;
- vdata->full_total++;
+ vdata->count.db_traverse.total++;
lmaster = ctdb_lmaster(ctdb, &key);
if (lmaster >= ctdb->num_nodes) {
- vdata->full_error++;
+ vdata->count.db_traverse.error++;
DEBUG(DEBUG_CRIT, (__location__
" lmaster[%u] >= ctdb->num_nodes[%u] for key"
" with hash[%u]!\n",
if (data.dsize != sizeof(struct ctdb_ltdb_header)) {
/* it is not a deleted record */
- vdata->full_skipped++;
+ vdata->count.db_traverse.skipped++;
return 0;
}
hdr = (struct ctdb_ltdb_header *)data.dptr;
if (hdr->dmaster != ctdb->pnn) {
- vdata->full_skipped++;
+ vdata->count.db_traverse.skipped++;
return 0;
}
*/
res = insert_record_into_delete_queue(ctdb_db, hdr, key);
if (res != 0) {
- vdata->full_error++;
+ vdata->count.db_traverse.error++;
} else {
- vdata->full_scheduled++;
+ vdata->count.db_traverse.scheduled++;
}
return 0;
{
struct delete_record_data *dd = talloc_get_type(data, struct delete_record_data);
struct delete_records_list *recs = talloc_get_type(param, struct delete_records_list);
- struct ctdb_rec_data *rec;
- size_t old_size;
+ struct ctdb_marshall_buffer *m;
- rec = ctdb_marshall_record(dd, recs->records->db_id, dd->key, &dd->hdr, tdb_null);
- if (rec == NULL) {
+ m = ctdb_marshall_add(recs, recs->records, recs->records->db_id,
+ recs->records->db_id,
+ dd->key, &dd->hdr, tdb_null);
+ if (m == NULL) {
DEBUG(DEBUG_ERR, (__location__ " failed to marshall record\n"));
- return 0;
+ return -1;
}
- old_size = talloc_get_size(recs->records);
- recs->records = talloc_realloc_size(NULL, recs->records, old_size + rec->length);
- if (recs->records == NULL) {
- DEBUG(DEBUG_ERR,(__location__ " Failed to expand\n"));
- return 0;
- }
- recs->records->count++;
- memcpy(old_size+(uint8_t *)(recs->records), rec, rec->length);
+ recs->records = m;
return 0;
}
uint32_t hash = ctdb_hash(&(dd->key));
int res;
- res = tdb_chainlock(ctdb_db->ltdb->tdb, dd->key);
+ res = tdb_chainlock_nonblock(ctdb_db->ltdb->tdb, dd->key);
if (res != 0) {
- DEBUG(DEBUG_ERR,
- (__location__ " Error getting chainlock on record with "
- "key hash [0x%08x] on database db[%s].\n",
- hash, ctdb_db->db_name));
- recs->vdata->delete_skipped++;
+ recs->vdata->count.delete_list.skipped++;
+ recs->vdata->count.delete_list.left--;
talloc_free(dd);
return 0;
}
skip:
tdb_chainunlock(ctdb_db->ltdb->tdb, dd->key);
- recs->vdata->delete_skipped++;
+ recs->vdata->count.delete_list.skipped++;
+ recs->vdata->count.delete_list.left--;
talloc_free(dd);
dd = NULL;
uint32_t lmaster;
uint32_t hash = ctdb_hash(&(dd->key));
- vdata->fast_total++;
+ vdata->count.delete_queue.total++;
- res = tdb_chainlock(ctdb_db->ltdb->tdb, dd->key);
+ res = tdb_chainlock_nonblock(ctdb_db->ltdb->tdb, dd->key);
if (res != 0) {
- DEBUG(DEBUG_ERR,
- (__location__ " Error getting chainlock on record with "
- "key hash [0x%08x] on database db[%s].\n",
- hash, ctdb_db->db_name));
- vdata->fast_error++;
+ vdata->count.delete_queue.error++;
return 0;
}
DEBUG(DEBUG_ERR,
(__location__ " Error adding record to list "
"of records to send to lmaster.\n"));
- vdata->fast_error++;
+ vdata->count.delete_queue.error++;
} else {
- vdata->fast_added_to_vacuum_fetch_list++;
+ vdata->count.delete_queue.added_to_vacuum_fetch_list++;
}
goto done;
}
DEBUG(DEBUG_ERR,
(__location__ " Error adding record to list "
"of records for deletion on lmaster.\n"));
- vdata->fast_error++;
+ vdata->count.delete_queue.error++;
} else {
- vdata->fast_added_to_delete_list++;
+ vdata->count.delete_queue.added_to_delete_list++;
}
} else {
res = tdb_delete(ctdb_db->ltdb->tdb, dd->key);
(__location__ " Error deleting record with key "
"hash [0x%08x] from local data base db[%s].\n",
hash, ctdb_db->db_name));
- vdata->fast_error++;
+ vdata->count.delete_queue.error++;
goto done;
}
(__location__ " Deleted record with key hash "
"[0x%08x] from local data base db[%s].\n",
hash, ctdb_db->db_name));
- vdata->fast_deleted++;
+ vdata->count.delete_queue.deleted++;
}
goto done;
skipped:
- vdata->fast_skipped++;
+ vdata->count.delete_queue.skipped++;
done:
tdb_chainunlock(ctdb_db->ltdb->tdb, dd->key);
(__location__ " Error getting chainlock on record with "
"key hash [0x%08x] on database db[%s].\n",
hash, ctdb_db->db_name));
- vdata->delete_local_error++;
- vdata->delete_left--;
+ vdata->count.delete_list.local_error++;
+ vdata->count.delete_list.left--;
talloc_free(dd);
return 0;
}
(__location__ " Error deleting record with key hash "
"[0x%08x] from local data base db[%s].\n",
hash, ctdb_db->db_name));
- vdata->delete_local_error++;
+ vdata->count.delete_list.local_error++;
goto done;
}
(__location__ " Deleted record with key hash [0x%08x] from "
"local data base db[%s].\n", hash, ctdb_db->db_name));
- vdata->delete_deleted++;
+ vdata->count.delete_list.deleted++;
goto done;
skip:
- vdata->delete_skipped++;
+ vdata->count.delete_list.skipped++;
done:
tdb_chainunlock(ctdb_db->ltdb->tdb, dd->key);
talloc_free(dd);
- vdata->delete_left--;
+ vdata->count.delete_list.left--;
return 0;
}
struct vacuum_data *vdata)
{
uint32_t sum;
+ int ret;
+
+ ret = trbt_traversearray32(ctdb_db->delete_queue, 1,
+ delete_queue_traverse, vdata);
- trbt_traversearray32(ctdb_db->delete_queue, 1, delete_queue_traverse, vdata);
+ if (ret != 0) {
+ DEBUG(DEBUG_ERR, (__location__ " Error traversing "
+ "the delete queue.\n"));
+ }
- sum = vdata->fast_deleted
- + vdata->fast_skipped
- + vdata->fast_error
- + vdata->fast_added_to_delete_list
- + vdata->fast_added_to_vacuum_fetch_list;
+ sum = vdata->count.delete_queue.deleted
+ + vdata->count.delete_queue.skipped
+ + vdata->count.delete_queue.error
+ + vdata->count.delete_queue.added_to_delete_list
+ + vdata->count.delete_queue.added_to_vacuum_fetch_list;
- if (vdata->fast_total != sum) {
+ if (vdata->count.delete_queue.total != sum) {
DEBUG(DEBUG_ERR, (__location__ " Inconsistency in fast vacuum "
"counts for db[%s]: total[%u] != sum[%u]\n",
- ctdb_db->db_name, (unsigned)vdata->fast_total,
+ ctdb_db->db_name,
+ (unsigned)vdata->count.delete_queue.total,
(unsigned)sum));
}
- if (vdata->fast_total > 0) {
+ if (vdata->count.delete_queue.total > 0) {
DEBUG(DEBUG_INFO,
(__location__
" fast vacuuming delete_queue traverse statistics: "
"adl[%u] "
"avf[%u]\n",
ctdb_db->db_name,
- (unsigned)vdata->fast_total,
- (unsigned)vdata->fast_deleted,
- (unsigned)vdata->fast_skipped,
- (unsigned)vdata->fast_error,
- (unsigned)vdata->fast_added_to_delete_list,
- (unsigned)vdata->fast_added_to_vacuum_fetch_list));
+ (unsigned)vdata->count.delete_queue.total,
+ (unsigned)vdata->count.delete_queue.deleted,
+ (unsigned)vdata->count.delete_queue.skipped,
+ (unsigned)vdata->count.delete_queue.error,
+ (unsigned)vdata->count.delete_queue.added_to_delete_list,
+ (unsigned)vdata->count.delete_queue.added_to_vacuum_fetch_list));
}
return;
* This is not done each time but only every tunable
* VacuumFastPathCount times.
*/
-static int ctdb_vacuum_traverse_db(struct ctdb_db_context *ctdb_db,
- struct vacuum_data *vdata)
+static void ctdb_vacuum_traverse_db(struct ctdb_db_context *ctdb_db,
+ struct vacuum_data *vdata)
{
int ret;
if (ret == -1 || vdata->traverse_error) {
DEBUG(DEBUG_ERR, (__location__ " Traverse error in vacuuming "
"'%s'\n", ctdb_db->db_name));
- return -1;
+ return;
}
- if (vdata->full_total > 0) {
+ if (vdata->count.db_traverse.total > 0) {
DEBUG(DEBUG_INFO,
(__location__
" full vacuuming db traverse statistics: "
"err[%u] "
"sched[%u]\n",
ctdb_db->db_name,
- (unsigned)vdata->full_total,
- (unsigned)vdata->full_skipped,
- (unsigned)vdata->full_error,
- (unsigned)vdata->full_scheduled));
+ (unsigned)vdata->count.db_traverse.total,
+ (unsigned)vdata->count.db_traverse.skipped,
+ (unsigned)vdata->count.db_traverse.error,
+ (unsigned)vdata->count.db_traverse.scheduled));
}
- return 0;
+ return;
}
/**
* For records for which we are not the lmaster, tell the lmaster to
* fetch the record.
*/
-static int ctdb_process_vacuum_fetch_lists(struct ctdb_db_context *ctdb_db,
- struct vacuum_data *vdata)
+static void ctdb_process_vacuum_fetch_lists(struct ctdb_db_context *ctdb_db,
+ struct vacuum_data *vdata)
{
int i;
struct ctdb_context *ctdb = ctdb_db->ctdb;
vfl->count, ctdb->nodes[i]->pnn,
ctdb_db->db_name));
- data.dsize = talloc_get_size(vfl);
- data.dptr = (void *)vfl;
+ data = ctdb_marshall_finish(vfl);
if (ctdb_client_send_message(ctdb, ctdb->nodes[i]->pnn,
CTDB_SRVID_VACUUM_FETCH,
data) != 0)
DEBUG(DEBUG_ERR, (__location__ " Failed to send vacuum "
"fetch message to %u\n",
ctdb->nodes[i]->pnn));
- return -1;
}
}
- return 0;
+ return;
}
/**
* 3) The lmaster locally deletes its copies of all records that
* could successfully be deleted remotely in step #2.
*/
-static int ctdb_process_delete_list(struct ctdb_db_context *ctdb_db,
- struct vacuum_data *vdata)
+static void ctdb_process_delete_list(struct ctdb_db_context *ctdb_db,
+ struct vacuum_data *vdata)
{
int ret, i;
struct ctdb_context *ctdb = ctdb_db->ctdb;
struct delete_records_list *recs;
TDB_DATA indata;
- struct ctdb_node_map *nodemap;
+ struct ctdb_node_map_old *nodemap;
uint32_t *active_nodes;
int num_active_nodes;
TALLOC_CTX *tmp_ctx;
uint32_t sum;
- if (vdata->delete_count == 0) {
- return 0;
+ if (vdata->count.delete_list.total == 0) {
+ return;
}
tmp_ctx = talloc_new(vdata);
if (tmp_ctx == NULL) {
DEBUG(DEBUG_ERR,(__location__ " Out of memory\n"));
- return 0;
+ return;
}
- vdata->delete_left = vdata->delete_count;
+ vdata->count.delete_list.left = vdata->count.delete_list.total;
/*
* get the list of currently active nodes
&nodemap);
if (ret != 0) {
DEBUG(DEBUG_ERR,(__location__ " unable to get node map\n"));
- ret = -1;
goto done;
}
recs = talloc_zero(tmp_ctx, struct delete_records_list);
if (recs == NULL) {
DEBUG(DEBUG_ERR,(__location__ " Out of memory\n"));
- ret = -1;
goto done;
}
recs->records = (struct ctdb_marshall_buffer *)
offsetof(struct ctdb_marshall_buffer, data));
if (recs->records == NULL) {
DEBUG(DEBUG_ERR,(__location__ " Out of memory\n"));
- ret = -1;
goto done;
}
recs->records->db_id = ctdb_db->db_id;
* records' RSNs in the database, to ensure we (as dmaster)
* keep the highest RSN of the records in the cluster.
*/
- trbt_traversearray32(vdata->delete_list, 1,
- delete_marshall_traverse_first, recs);
+ ret = trbt_traversearray32(vdata->delete_list, 1,
+ delete_marshall_traverse_first, recs);
+ if (ret != 0) {
+ DEBUG(DEBUG_ERR, (__location__ " Error traversing the "
+ "delete list for first marshalling.\n"));
+ goto done;
+ }
- indata.dsize = talloc_get_size(recs->records);
- indata.dptr = (void *)recs->records;
+ indata = ctdb_marshall_finish(recs->records);
for (i = 0; i < num_active_nodes; i++) {
struct ctdb_marshall_buffer *records;
- struct ctdb_rec_data *rec;
+ struct ctdb_rec_data_old *rec;
int32_t res;
TDB_DATA outdata;
DEBUG(DEBUG_ERR, ("Error storing record copies on "
"node %u: ret[%d] res[%d]\n",
active_nodes[i], ret, res));
- ret = -1;
goto done;
}
* the list to process further.
*/
records = (struct ctdb_marshall_buffer *)outdata.dptr;
- rec = (struct ctdb_rec_data *)&records->data[0];
+ rec = (struct ctdb_rec_data_old *)&records->data[0];
while (records->count-- > 1) {
TDB_DATA reckey, recdata;
struct ctdb_ltdb_header *rechdr;
if (recdata.dsize < sizeof(struct ctdb_ltdb_header)) {
DEBUG(DEBUG_CRIT,(__location__ " bad ltdb record\n"));
- ret = -1;
goto done;
}
rechdr = (struct ctdb_ltdb_header *)recdata.dptr;
* update statistics.
*/
talloc_free(dd);
- vdata->delete_remote_error++;
- vdata->delete_left--;
+ vdata->count.delete_list.remote_error++;
+ vdata->count.delete_list.left--;
+ } else {
+ DEBUG(DEBUG_ERR, (__location__ " Failed to "
+ "find record with hash 0x%08x coming "
+ "back from RECEIVE_RECORDS "
+ "control in delete list.\n",
+ ctdb_hash(&reckey)));
+ vdata->count.delete_list.local_error++;
+ vdata->count.delete_list.left--;
}
- rec = (struct ctdb_rec_data *)(rec->length + (uint8_t *)rec);
+ rec = (struct ctdb_rec_data_old *)(rec->length + (uint8_t *)rec);
}
}
- if (vdata->delete_left == 0) {
+ if (vdata->count.delete_list.left == 0) {
goto success;
}
offsetof(struct ctdb_marshall_buffer, data));
if (recs->records == NULL) {
DEBUG(DEBUG_ERR,(__location__ " Out of memory\n"));
- ret = -1;
goto done;
}
recs->records->db_id = ctdb_db->db_id;
- trbt_traversearray32(vdata->delete_list, 1,
- delete_marshall_traverse, recs);
+ ret = trbt_traversearray32(vdata->delete_list, 1,
+ delete_marshall_traverse, recs);
+ if (ret != 0) {
+ DEBUG(DEBUG_ERR, (__location__ " Error traversing the "
+ "delete list for second marshalling.\n"));
+ goto done;
+ }
- indata.dsize = talloc_get_size(recs->records);
- indata.dptr = (void *)recs->records;
+ indata = ctdb_marshall_finish(recs->records);
for (i = 0; i < num_active_nodes; i++) {
struct ctdb_marshall_buffer *records;
- struct ctdb_rec_data *rec;
+ struct ctdb_rec_data_old *rec;
int32_t res;
TDB_DATA outdata;
DEBUG(DEBUG_ERR, ("Failed to delete records on "
"node %u: ret[%d] res[%d]\n",
active_nodes[i], ret, res));
- ret = -1;
goto done;
}
* the list to delete locally.
*/
records = (struct ctdb_marshall_buffer *)outdata.dptr;
- rec = (struct ctdb_rec_data *)&records->data[0];
+ rec = (struct ctdb_rec_data_old *)&records->data[0];
while (records->count-- > 1) {
TDB_DATA reckey, recdata;
struct ctdb_ltdb_header *rechdr;
if (recdata.dsize < sizeof(struct ctdb_ltdb_header)) {
DEBUG(DEBUG_CRIT,(__location__ " bad ltdb record\n"));
- ret = -1;
goto done;
}
rechdr = (struct ctdb_ltdb_header *)recdata.dptr;
* the tree and update statistics.
*/
talloc_free(dd);
- vdata->delete_remote_error++;
- vdata->delete_left--;
+ vdata->count.delete_list.remote_error++;
+ vdata->count.delete_list.left--;
+ } else {
+ DEBUG(DEBUG_ERR, (__location__ " Failed to "
+ "find record with hash 0x%08x coming "
+ "back from TRY_DELETE_RECORDS "
+ "control in delete list.\n",
+ ctdb_hash(&reckey)));
+ vdata->count.delete_list.local_error++;
+ vdata->count.delete_list.left--;
}
- rec = (struct ctdb_rec_data *)(rec->length + (uint8_t *)rec);
+ rec = (struct ctdb_rec_data_old *)(rec->length + (uint8_t *)rec);
}
}
- if (vdata->delete_left == 0) {
+ if (vdata->count.delete_list.left == 0) {
goto success;
}
* active remote nodes.
*/
- trbt_traversearray32(vdata->delete_list, 1,
- delete_record_traverse, vdata);
+ ret = trbt_traversearray32(vdata->delete_list, 1,
+ delete_record_traverse, vdata);
+ if (ret != 0) {
+ DEBUG(DEBUG_ERR, (__location__ " Error traversing the "
+ "delete list for deletion.\n"));
+ }
success:
- if (vdata->delete_left != 0) {
+ if (vdata->count.delete_list.left != 0) {
DEBUG(DEBUG_ERR, (__location__ " Vaccum db[%s] error: "
"there are %u records left for deletion after "
"processing delete list\n",
ctdb_db->db_name,
- (unsigned)vdata->delete_left));
+ (unsigned)vdata->count.delete_list.left));
}
- sum = vdata->delete_deleted
- + vdata->delete_skipped
- + vdata->delete_remote_error
- + vdata->delete_local_error
- + vdata->delete_left;
+ sum = vdata->count.delete_list.deleted
+ + vdata->count.delete_list.skipped
+ + vdata->count.delete_list.remote_error
+ + vdata->count.delete_list.local_error
+ + vdata->count.delete_list.left;
- if (vdata->delete_count != sum) {
+ if (vdata->count.delete_list.total != sum) {
DEBUG(DEBUG_ERR, (__location__ " Inconsistency in vacuum "
"delete list counts for db[%s]: total[%u] != sum[%u]\n",
- ctdb_db->db_name, (unsigned)vdata->delete_count,
+ ctdb_db->db_name,
+ (unsigned)vdata->count.delete_list.total,
(unsigned)sum));
}
- if (vdata->delete_count > 0) {
+ if (vdata->count.delete_list.total > 0) {
DEBUG(DEBUG_INFO,
(__location__
" vacuum delete list statistics: "
"loc.err[%u] "
"left[%u]\n",
ctdb_db->db_name,
- (unsigned)vdata->delete_count,
- (unsigned)vdata->delete_deleted,
- (unsigned)vdata->delete_skipped,
- (unsigned)vdata->delete_remote_error,
- (unsigned)vdata->delete_local_error,
- (unsigned)vdata->delete_left));
+ (unsigned)vdata->count.delete_list.total,
+ (unsigned)vdata->count.delete_list.deleted,
+ (unsigned)vdata->count.delete_list.skipped,
+ (unsigned)vdata->count.delete_list.remote_error,
+ (unsigned)vdata->count.delete_list.local_error,
+ (unsigned)vdata->count.delete_list.left));
}
- ret = 0;
-
done:
talloc_free(tmp_ctx);
- return ret;
+ return;
}
/**
* initialize the vacuum_data
*/
-static int ctdb_vacuum_init_vacuum_data(struct ctdb_db_context *ctdb_db,
- struct vacuum_data *vdata)
+static struct vacuum_data *ctdb_vacuum_init_vacuum_data(
+ struct ctdb_db_context *ctdb_db,
+ TALLOC_CTX *mem_ctx)
{
int i;
struct ctdb_context *ctdb = ctdb_db->ctdb;
+ struct vacuum_data *vdata;
- vdata->fast_added_to_delete_list = 0;
- vdata->fast_added_to_vacuum_fetch_list = 0;
- vdata->fast_deleted = 0;
- vdata->fast_skipped = 0;
- vdata->fast_error = 0;
- vdata->fast_total = 0;
- vdata->full_scheduled = 0;
- vdata->full_skipped = 0;
- vdata->full_error = 0;
- vdata->full_total = 0;
- vdata->delete_count = 0;
- vdata->delete_left = 0;
- vdata->delete_remote_error = 0;
- vdata->delete_local_error = 0;
- vdata->delete_skipped = 0;
- vdata->delete_deleted = 0;
+ vdata = talloc_zero(mem_ctx, struct vacuum_data);
+ if (vdata == NULL) {
+ DEBUG(DEBUG_ERR,(__location__ " Out of memory\n"));
+ return NULL;
+ }
+
+ vdata->ctdb = ctdb_db->ctdb;
+ vdata->ctdb_db = ctdb_db;
+ vdata->delete_list = trbt_create(vdata, 0);
+ if (vdata->delete_list == NULL) {
+ DEBUG(DEBUG_ERR,(__location__ " Out of memory\n"));
+ goto fail;
+ }
+
+ vdata->start = timeval_current();
+
+ vdata->count.delete_queue.added_to_delete_list = 0;
+ vdata->count.delete_queue.added_to_vacuum_fetch_list = 0;
+ vdata->count.delete_queue.deleted = 0;
+ vdata->count.delete_queue.skipped = 0;
+ vdata->count.delete_queue.error = 0;
+ vdata->count.delete_queue.total = 0;
+ vdata->count.db_traverse.scheduled = 0;
+ vdata->count.db_traverse.skipped = 0;
+ vdata->count.db_traverse.error = 0;
+ vdata->count.db_traverse.total = 0;
+ vdata->count.delete_list.total = 0;
+ vdata->count.delete_list.left = 0;
+ vdata->count.delete_list.remote_error = 0;
+ vdata->count.delete_list.local_error = 0;
+ vdata->count.delete_list.skipped = 0;
+ vdata->count.delete_list.deleted = 0;
/* the list needs to be of length num_nodes */
vdata->vacuum_fetch_list = talloc_zero_array(vdata,
ctdb->num_nodes);
if (vdata->vacuum_fetch_list == NULL) {
DEBUG(DEBUG_ERR,(__location__ " Out of memory\n"));
- return -1;
+ goto fail;
}
for (i = 0; i < ctdb->num_nodes; i++) {
vdata->vacuum_fetch_list[i] = (struct ctdb_marshall_buffer *)
offsetof(struct ctdb_marshall_buffer, data));
if (vdata->vacuum_fetch_list[i] == NULL) {
DEBUG(DEBUG_ERR,(__location__ " Out of memory\n"));
- return -1;
+ talloc_free(vdata);
+ return NULL;
}
vdata->vacuum_fetch_list[i]->db_id = ctdb_db->db_id;
}
- return 0;
+ return vdata;
+
+fail:
+ talloc_free(vdata);
+ return NULL;
}
/**
* This executes in the child context.
*/
static int ctdb_vacuum_db(struct ctdb_db_context *ctdb_db,
- struct vacuum_data *vdata,
bool full_vacuum_run)
{
struct ctdb_context *ctdb = ctdb_db->ctdb;
int ret, pnn;
+ struct vacuum_data *vdata;
+ TALLOC_CTX *tmp_ctx;
DEBUG(DEBUG_INFO, (__location__ " Entering %s vacuum run for db "
"%s db_id[0x%08x]\n",
ctdb->pnn = pnn;
- ret = ctdb_vacuum_init_vacuum_data(ctdb_db, vdata);
- if (ret != 0) {
- return ret;
- }
-
- if (full_vacuum_run) {
- ret = ctdb_vacuum_traverse_db(ctdb_db, vdata);
- if (ret != 0) {
- return ret;
- }
- }
-
- ctdb_process_delete_queue(ctdb_db, vdata);
-
- ret = ctdb_process_vacuum_fetch_lists(ctdb_db, vdata);
- if (ret != 0) {
- return ret;
- }
-
- ret = ctdb_process_delete_list(ctdb_db, vdata);
- if (ret != 0) {
- return ret;
- }
-
- /* this ensures we run our event queue */
- ctdb_ctrl_getpnn(ctdb, TIMELIMIT(), CTDB_CURRENT_NODE);
-
- return 0;
-}
-
-
-/*
- * traverse function for repacking
- */
-static int repack_traverse(struct tdb_context *tdb, TDB_DATA key, TDB_DATA data,
- void *private_data)
-{
- struct vacuum_data *vdata = (struct vacuum_data *)private_data;
-
- if (vdata->vacuum) {
- uint32_t hash = ctdb_hash(&key);
- struct delete_record_data *kd;
- /*
- * check if we can ignore this record because it's in the delete_list
- */
- kd = (struct delete_record_data *)trbt_lookup32(vdata->delete_list, hash);
- /*
- * there might be hash collisions so we have to compare the keys here to be sure
- */
- if (kd && kd->key.dsize == key.dsize && memcmp(kd->key.dptr, key.dptr, key.dsize) == 0) {
- struct ctdb_ltdb_header *hdr = (struct ctdb_ltdb_header *)data.dptr;
- /*
- * we have to check if the record hasn't changed in the meantime in order to
- * savely remove it from the database
- */
- if (data.dsize == sizeof(struct ctdb_ltdb_header) &&
- hdr->dmaster == kd->ctdb->pnn &&
- ctdb_lmaster(kd->ctdb, &(kd->key)) == kd->ctdb->pnn &&
- kd->hdr.rsn == hdr->rsn) {
- vdata->vacuumed++;
- return 0;
- }
- }
- }
- if (tdb_store(vdata->dest_db, key, data, TDB_INSERT) != 0) {
- vdata->traverse_error = true;
- return -1;
- }
- vdata->copied++;
- return 0;
-}
-
-/*
- * repack a tdb
- */
-static int ctdb_repack_tdb(struct tdb_context *tdb, TALLOC_CTX *mem_ctx, struct vacuum_data *vdata)
-{
- struct tdb_context *tmp_db;
-
- if (tdb_transaction_start(tdb) != 0) {
- DEBUG(DEBUG_ERR,(__location__ " Failed to start transaction\n"));
- return -1;
- }
-
- tmp_db = tdb_open("tmpdb", tdb_hash_size(tdb),
- TDB_INTERNAL|TDB_DISALLOW_NESTING,
- O_RDWR|O_CREAT, 0);
- if (tmp_db == NULL) {
- DEBUG(DEBUG_ERR,(__location__ " Failed to create tmp_db\n"));
- tdb_transaction_cancel(tdb);
+ tmp_ctx = talloc_new(ctdb_db);
+ if (tmp_ctx == NULL) {
+ DEBUG(DEBUG_ERR, ("Out of memory!\n"));
return -1;
}
- vdata->traverse_error = false;
- vdata->dest_db = tmp_db;
- vdata->vacuum = true;
- vdata->vacuumed = 0;
- vdata->copied = 0;
-
- /*
- * repack and vacuum on-the-fly by not writing the records that are
- * no longer needed
- */
- if (tdb_traverse_read(tdb, repack_traverse, vdata) == -1) {
- DEBUG(DEBUG_ERR,(__location__ " Failed to traverse copying out\n"));
- tdb_transaction_cancel(tdb);
- tdb_close(tmp_db);
- return -1;
- }
-
- DEBUG(DEBUG_INFO,(__location__ " %u records vacuumed\n", vdata->vacuumed));
-
- if (vdata->traverse_error) {
- DEBUG(DEBUG_ERR,(__location__ " Error during traversal\n"));
- tdb_transaction_cancel(tdb);
- tdb_close(tmp_db);
+ vdata = ctdb_vacuum_init_vacuum_data(ctdb_db, tmp_ctx);
+ if (vdata == NULL) {
+ talloc_free(tmp_ctx);
return -1;
}
- if (tdb_wipe_all(tdb) != 0) {
- DEBUG(DEBUG_ERR,(__location__ " Failed to wipe database\n"));
- tdb_transaction_cancel(tdb);
- tdb_close(tmp_db);
- return -1;
+ if (full_vacuum_run) {
+ ctdb_vacuum_traverse_db(ctdb_db, vdata);
}
- vdata->traverse_error = false;
- vdata->dest_db = tdb;
- vdata->vacuum = false;
- vdata->copied = 0;
-
- if (tdb_traverse_read(tmp_db, repack_traverse, vdata) == -1) {
- DEBUG(DEBUG_ERR,(__location__ " Failed to traverse copying back\n"));
- tdb_transaction_cancel(tdb);
- tdb_close(tmp_db);
- return -1;
- }
+ ctdb_process_delete_queue(ctdb_db, vdata);
- if (vdata->traverse_error) {
- DEBUG(DEBUG_ERR,(__location__ " Error during second traversal\n"));
- tdb_transaction_cancel(tdb);
- tdb_close(tmp_db);
- return -1;
- }
+ ctdb_process_vacuum_fetch_lists(ctdb_db, vdata);
- tdb_close(tmp_db);
+ ctdb_process_delete_list(ctdb_db, vdata);
+ talloc_free(tmp_ctx);
- if (tdb_transaction_commit(tdb) != 0) {
- DEBUG(DEBUG_ERR,(__location__ " Failed to commit\n"));
- return -1;
- }
- DEBUG(DEBUG_INFO,(__location__ " %u records copied\n", vdata->copied));
+ /* this ensures we run our event queue */
+ ctdb_ctrl_getpnn(ctdb, TIMELIMIT(), CTDB_CURRENT_NODE);
return 0;
}
* called from the child context
*/
static int ctdb_vacuum_and_repack_db(struct ctdb_db_context *ctdb_db,
- TALLOC_CTX *mem_ctx,
bool full_vacuum_run)
{
uint32_t repack_limit = ctdb_db->ctdb->tunable.repack_limit;
const char *name = ctdb_db->db_name;
int freelist_size = 0;
- struct vacuum_data *vdata;
-
- vdata = talloc_zero(mem_ctx, struct vacuum_data);
- if (vdata == NULL) {
- DEBUG(DEBUG_ERR,(__location__ " Out of memory\n"));
- return -1;
- }
-
- vdata->ctdb = ctdb_db->ctdb;
- vdata->repack_limit = repack_limit;
- vdata->delete_list = trbt_create(vdata, 0);
- vdata->ctdb_db = ctdb_db;
- if (vdata->delete_list == NULL) {
- DEBUG(DEBUG_ERR,(__location__ " Out of memory\n"));
- talloc_free(vdata);
- return -1;
- }
+ int ret;
- vdata->start = timeval_current();
-
- /*
- * gather all records that can be deleted in vdata
- */
- if (ctdb_vacuum_db(ctdb_db, vdata, full_vacuum_run) != 0) {
+ if (ctdb_vacuum_db(ctdb_db, full_vacuum_run) != 0) {
DEBUG(DEBUG_ERR,(__location__ " Failed to vacuum '%s'\n", name));
}
- if (repack_limit != 0) {
- freelist_size = tdb_freelist_size(ctdb_db->ltdb->tdb);
- if (freelist_size == -1) {
- DEBUG(DEBUG_ERR,(__location__ " Failed to get freelist size for '%s'\n", name));
- talloc_free(vdata);
- return -1;
- }
+ freelist_size = tdb_freelist_size(ctdb_db->ltdb->tdb);
+ if (freelist_size == -1) {
+ DEBUG(DEBUG_ERR,(__location__ " Failed to get freelist size for '%s'\n", name));
+ return -1;
}
/*
*/
if ((repack_limit == 0 || (uint32_t)freelist_size < repack_limit))
{
- talloc_free(vdata);
return 0;
}
- DEBUG(DEBUG_INFO,("Repacking %s with %u freelist entries and %u records to delete\n",
- name, freelist_size, vdata->delete_left));
+ DEBUG(DEBUG_INFO, ("Repacking %s with %u freelist entries\n",
+ name, freelist_size));
- /*
- * repack and implicitely get rid of the records we can delete
- */
- if (ctdb_repack_tdb(ctdb_db->ltdb->tdb, mem_ctx, vdata) != 0) {
+ ret = tdb_repack(ctdb_db->ltdb->tdb);
+ if (ret != 0) {
DEBUG(DEBUG_ERR,(__location__ " Failed to repack '%s'\n", name));
- talloc_free(vdata);
return -1;
}
- talloc_free(vdata);
return 0;
}
struct ctdb_db_context *ctdb_db = child_ctx->vacuum_handle->ctdb_db;
struct ctdb_context *ctdb = ctdb_db->ctdb;
+ CTDB_UPDATE_DB_LATENCY(ctdb_db, "vacuum", vacuum.latency, l);
DEBUG(DEBUG_INFO,("Vacuuming took %.3f seconds for database %s\n", l, ctdb_db->db_name));
if (child_ctx->child_pid != -1) {
DLIST_REMOVE(ctdb->vacuumers, child_ctx);
- event_add_timed(ctdb->ev, child_ctx->vacuum_handle,
- timeval_current_ofs(get_vacuum_interval(ctdb_db), 0),
- ctdb_vacuum_event, child_ctx->vacuum_handle);
+ tevent_add_timer(ctdb->ev, child_ctx->vacuum_handle,
+ timeval_current_ofs(get_vacuum_interval(ctdb_db), 0),
+ ctdb_vacuum_event, child_ctx->vacuum_handle);
return 0;
}
/*
* this event is generated when a vacuum child process times out
*/
-static void vacuum_child_timeout(struct event_context *ev, struct timed_event *te,
- struct timeval t, void *private_data)
+static void vacuum_child_timeout(struct tevent_context *ev,
+ struct tevent_timer *te,
+ struct timeval t, void *private_data)
{
struct ctdb_vacuum_child_context *child_ctx = talloc_get_type(private_data, struct ctdb_vacuum_child_context);
/*
* this event is generated when a vacuum child process has completed
*/
-static void vacuum_child_handler(struct event_context *ev, struct fd_event *fde,
- uint16_t flags, void *private_data)
+static void vacuum_child_handler(struct tevent_context *ev,
+ struct tevent_fd *fde,
+ uint16_t flags, void *private_data)
{
struct ctdb_vacuum_child_context *child_ctx = talloc_get_type(private_data, struct ctdb_vacuum_child_context);
char c = 0;
DEBUG(DEBUG_INFO,("Vacuuming child process %d finished for db %s\n", child_ctx->child_pid, child_ctx->vacuum_handle->ctdb_db->db_name));
child_ctx->child_pid = -1;
- ret = read(child_ctx->fd[0], &c, 1);
+ ret = sys_read(child_ctx->fd[0], &c, 1);
if (ret != 1 || c != 0) {
child_ctx->status = VACUUM_ERROR;
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));
/*
* this event is called every time we need to start a new vacuum process
*/
-static void
-ctdb_vacuum_event(struct event_context *ev, struct timed_event *te,
- struct timeval t, void *private_data)
+static void ctdb_vacuum_event(struct tevent_context *ev,
+ struct tevent_timer *te,
+ struct timeval t, void *private_data)
{
struct ctdb_vacuum_handle *vacuum_handle = talloc_get_type(private_data, struct ctdb_vacuum_handle);
struct ctdb_db_context *ctdb_db = vacuum_handle->ctdb_db;
struct tevent_fd *fde;
int ret;
- /* we dont vacuum if we are in recovery mode, or db frozen */
+ /* we don't vacuum if we are in recovery mode, or db frozen */
if (ctdb->recovery_mode == CTDB_RECOVERY_ACTIVE ||
ctdb->freeze_mode[ctdb_db->priority] != CTDB_FREEZE_NONE) {
DEBUG(DEBUG_INFO, ("Not vacuuming %s (%s)\n", ctdb_db->db_name,
: ctdb->freeze_mode[ctdb_db->priority] == CTDB_FREEZE_PENDING
? "freeze pending"
: "frozen"));
- event_add_timed(ctdb->ev, vacuum_handle,
- timeval_current_ofs(get_vacuum_interval(ctdb_db), 0),
- ctdb_vacuum_event, vacuum_handle);
+ tevent_add_timer(ctdb->ev, vacuum_handle,
+ timeval_current_ofs(get_vacuum_interval(ctdb_db), 0),
+ ctdb_vacuum_event, vacuum_handle);
+ return;
+ }
+
+ /* Do not allow multiple vacuuming child processes to be active at the
+ * same time. If there is vacuuming child process active, delay
+ * new vacuuming event to stagger vacuuming events.
+ */
+ if (ctdb->vacuumers != NULL) {
+ tevent_add_timer(ctdb->ev, vacuum_handle,
+ timeval_current_ofs(0, 500*1000),
+ ctdb_vacuum_event, vacuum_handle);
return;
}
if (ret != 0) {
talloc_free(child_ctx);
DEBUG(DEBUG_ERR, ("Failed to create pipe for vacuum child process.\n"));
- event_add_timed(ctdb->ev, vacuum_handle,
- timeval_current_ofs(get_vacuum_interval(ctdb_db), 0),
- ctdb_vacuum_event, vacuum_handle);
+ tevent_add_timer(ctdb->ev, vacuum_handle,
+ timeval_current_ofs(get_vacuum_interval(ctdb_db), 0),
+ ctdb_vacuum_event, vacuum_handle);
return;
}
close(child_ctx->fd[1]);
talloc_free(child_ctx);
DEBUG(DEBUG_ERR, ("Failed to fork vacuum child process.\n"));
- event_add_timed(ctdb->ev, vacuum_handle,
- timeval_current_ofs(get_vacuum_interval(ctdb_db), 0),
- ctdb_vacuum_event, vacuum_handle);
+ tevent_add_timer(ctdb->ev, vacuum_handle,
+ timeval_current_ofs(get_vacuum_interval(ctdb_db), 0),
+ ctdb_vacuum_event, vacuum_handle);
return;
}
close(child_ctx->fd[0]);
DEBUG(DEBUG_INFO,("Vacuuming child process %d for db %s started\n", getpid(), ctdb_db->db_name));
- ctdb_set_process_name("ctdb_vacuum");
+ prctl_set_comment("ctdb_vacuum");
if (switch_from_server_to_client(ctdb, "vacuum-%s", ctdb_db->db_name) != 0) {
DEBUG(DEBUG_CRIT, (__location__ "ERROR: failed to switch vacuum daemon into client mode. Shutting down.\n"));
_exit(1);
}
- /*
- * repack the db
- */
if ((ctdb->tunable.vacuum_fast_path_count > 0) &&
(vacuum_handle->fast_path_count == 0))
{
full_vacuum_run = true;
}
- cc = ctdb_vacuum_and_repack_db(ctdb_db, child_ctx,
- full_vacuum_run);
+ cc = ctdb_vacuum_and_repack_db(ctdb_db, full_vacuum_run);
- write(child_ctx->fd[1], &cc, 1);
+ sys_write(child_ctx->fd[1], &cc, 1);
_exit(0);
}
"in parent context. Shutting down\n");
}
- event_add_timed(ctdb->ev, child_ctx,
- timeval_current_ofs(ctdb->tunable.vacuum_max_run_time, 0),
- vacuum_child_timeout, child_ctx);
+ tevent_add_timer(ctdb->ev, child_ctx,
+ timeval_current_ofs(ctdb->tunable.vacuum_max_run_time, 0),
+ vacuum_child_timeout, child_ctx);
DEBUG(DEBUG_DEBUG, (__location__ " Created PIPE FD:%d to child vacuum process\n", child_ctx->fd[0]));
- fde = event_add_fd(ctdb->ev, child_ctx, child_ctx->fd[0],
- EVENT_FD_READ, vacuum_child_handler, child_ctx);
+ fde = tevent_add_fd(ctdb->ev, child_ctx, child_ctx->fd[0],
+ TEVENT_FD_READ, vacuum_child_handler, child_ctx);
tevent_fd_set_auto_close(fde);
vacuum_handle->child_ctx = child_ctx;
ctdb_db->vacuum_handle->ctdb_db = ctdb_db;
ctdb_db->vacuum_handle->fast_path_count = 0;
- event_add_timed(ctdb_db->ctdb->ev, ctdb_db->vacuum_handle,
- timeval_current_ofs(get_vacuum_interval(ctdb_db), 0),
- ctdb_vacuum_event, ctdb_db->vacuum_handle);
+ tevent_add_timer(ctdb_db->ctdb->ev, ctdb_db->vacuum_handle,
+ timeval_current_ofs(get_vacuum_interval(ctdb_db), 0),
+ ctdb_vacuum_event, ctdb_db->vacuum_handle);
return 0;
}
hash = (uint32_t)ctdb_hash(&key);
- DEBUG(DEBUG_INFO, (__location__ " schedule for deletion: db[%s] "
- "db_id[0x%08x] "
- "key_hash[0x%08x] "
- "lmaster[%u] "
- "migrated_with_data[%s]\n",
+ DEBUG(DEBUG_DEBUG, (__location__ " schedule for deletion: db[%s] "
+ "db_id[0x%08x] "
+ "key_hash[0x%08x] "
+ "lmaster[%u] "
+ "migrated_with_data[%s]\n",
ctdb_db->db_name, ctdb_db->db_id,
hash,
ctdb_lmaster(ctdb_db->ctdb, &key),
return ret;
}
- /* if we dont have a connection to the daemon we can not send
+ /* if we don't have a connection to the daemon we can not send
a control. For example sometimes from update_record control child
process.
*/