samba-mirror/ctdb/server/ctdb_vacuum.c

/*
   ctdb vacuuming events

   Copyright (C) Ronnie Sahlberg  2009
   Copyright (C) Michael Adam 2010-2013
   Copyright (C) Stefan Metzmacher 2010-2011

   This program is free software; you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by
   the Free Software Foundation; either version 3 of the License, or
   (at your option) any later version.

   This program is distributed in the hope that it will be useful,
   but WITHOUT ANY WARRANTY; without even the implied warranty of
   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
   GNU General Public License for more details.

   You should have received a copy of the GNU General Public License
   along with this program; if not, see <http://www.gnu.org/licenses/>.
*/

#include "replace.h"
#include "system/network.h"
#include "system/filesys.h"
#include "system/time.h"

#include <talloc.h>
#include <tevent.h>

#include "lib/tdb_wrap/tdb_wrap.h"
#include "lib/util/dlinklist.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)

enum vacuum_child_status { VACUUM_RUNNING, VACUUM_OK, VACUUM_ERROR, VACUUM_TIMEOUT};

struct ctdb_vacuum_child_context {
	struct ctdb_vacuum_child_context *next, *prev;
	struct ctdb_vacuum_handle *vacuum_handle;
	/* fd child writes status to */
	int fd[2];
	pid_t child_pid;
	enum vacuum_child_status status;
	struct timeval start_time;
};

struct ctdb_vacuum_handle {
	struct ctdb_db_context *ctdb_db;
	struct ctdb_vacuum_child_context *child_ctx;
	uint32_t fast_path_count;
};


/*  a list of records to possibly delete */
struct vacuum_data {
	struct ctdb_context *ctdb;
	struct ctdb_db_context *ctdb_db;
	struct tdb_context *dest_db;
	trbt_tree_t *delete_list;
	struct ctdb_marshall_buffer **vacuum_fetch_list;
	struct timeval start;
	bool traverse_error;
	bool vacuum;
	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 */
struct delete_record_data {
	struct ctdb_context *ctdb;
	struct ctdb_db_context *ctdb_db;
	struct ctdb_ltdb_header hdr;
	TDB_DATA key;
	uint8_t keydata[1];
};

struct delete_records_list {
	struct ctdb_marshall_buffer *records;
	struct vacuum_data *vdata;
};

static int insert_record_into_delete_queue(struct ctdb_db_context *ctdb_db,
					   const struct ctdb_ltdb_header *hdr,
					   TDB_DATA key);

/**
 * Store key and header in a tree, indexed by the key hash.
 */
static int insert_delete_record_data_into_tree(struct ctdb_context *ctdb,
					       struct ctdb_db_context *ctdb_db,
					       trbt_tree_t *tree,
					       const struct ctdb_ltdb_header *hdr,
					       TDB_DATA key)
{
	struct delete_record_data *dd;
	uint32_t hash;
	size_t len;

	len = offsetof(struct delete_record_data, keydata) + key.dsize;

	dd = (struct delete_record_data *)talloc_size(tree, len);
	if (dd == NULL) {
		DEBUG(DEBUG_ERR,(__location__ " Out of memory\n"));
		return -1;
	}
	talloc_set_name_const(dd, "struct delete_record_data");

	dd->ctdb      = ctdb;
	dd->ctdb_db   = ctdb_db;
	dd->key.dsize = key.dsize;
	dd->key.dptr  = dd->keydata;
	memcpy(dd->keydata, key.dptr, key.dsize);

	dd->hdr = *hdr;

	hash = ctdb_hash(&key);

	trbt_insert32(tree, hash, dd);

	return 0;
}

static int add_record_to_delete_list(struct vacuum_data *vdata, TDB_DATA key,
				     struct ctdb_ltdb_header *hdr)
{
	struct ctdb_context *ctdb = vdata->ctdb;
	struct ctdb_db_context *ctdb_db = vdata->ctdb_db;
	uint32_t hash;
	int ret;

	hash = ctdb_hash(&key);

	if (trbt_lookup32(vdata->delete_list, hash)) {
		DEBUG(DEBUG_INFO, (__location__ " Hash collision when vacuuming, skipping this record.\n"));
		return 0;
	}

	ret = insert_delete_record_data_into_tree(ctdb, ctdb_db,
						  vdata->delete_list,
						  hdr, key);
	if (ret != 0) {
		return -1;
	}

	vdata->count.delete_list.total++;

	return 0;
}

/**
 * Add a record to the list of records to be sent
 * to their lmaster with VACUUM_FETCH.
 */
static int add_record_to_vacuum_fetch_list(struct vacuum_data *vdata,
					   TDB_DATA key)
{
	struct ctdb_context *ctdb = vdata->ctdb;
	uint32_t lmaster;
	struct ctdb_marshall_buffer *vfl;

	lmaster = ctdb_lmaster(ctdb, &key);

	vfl = vdata->vacuum_fetch_list[lmaster];

	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;
	}

	vdata->vacuum_fetch_list[lmaster] = vfl;

	return 0;
}


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 *header =
		(struct ctdb_ltdb_header *)private_data;

	if (data.dsize != sizeof(struct ctdb_ltdb_header)) {
		return -1;
	}

	*header = *(struct ctdb_ltdb_header *)data.dptr;

	return 0;
}

/*
 * traverse function for gathering the records that can be deleted
 */
static int vacuum_traverse(struct tdb_context *tdb, TDB_DATA key, TDB_DATA data,
			   void *private_data)
{
	struct vacuum_data *vdata = talloc_get_type(private_data,
						    struct vacuum_data);
	struct ctdb_context *ctdb = vdata->ctdb;
	struct ctdb_db_context *ctdb_db = vdata->ctdb_db;
	uint32_t lmaster;
	struct ctdb_ltdb_header *hdr;
	int res = 0;

	vdata->count.db_traverse.total++;

	lmaster = ctdb_lmaster(ctdb, &key);
	if (lmaster >= ctdb->num_nodes) {
		vdata->count.db_traverse.error++;
		DEBUG(DEBUG_CRIT, (__location__
				   " lmaster[%u] >= ctdb->num_nodes[%u] for key"
				   " with hash[%u]!\n",
				   (unsigned)lmaster,
				   (unsigned)ctdb->num_nodes,
				   (unsigned)ctdb_hash(&key)));
		return -1;
	}

	if (data.dsize != sizeof(struct ctdb_ltdb_header)) {
		/* it is not a deleted record */
		vdata->count.db_traverse.skipped++;
		return 0;
	}

	hdr = (struct ctdb_ltdb_header *)data.dptr;

	if (hdr->dmaster != ctdb->pnn) {
		vdata->count.db_traverse.skipped++;
		return 0;
	}

	/*
	 * Add the record to this process's delete_queue for processing
	 * in the subsequent traverse in the fast vacuum run.
	 */
	res = insert_record_into_delete_queue(ctdb_db, hdr, key);
	if (res != 0) {
		vdata->count.db_traverse.error++;
	} else {
		vdata->count.db_traverse.scheduled++;
	}

	return 0;
}

/*
 * traverse the tree of records to delete and marshall them into
 * a blob
 */
static int delete_marshall_traverse(void *param, void *data)
{
	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_marshall_buffer *m;

	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 -1;
	}

	recs->records = m;
	return 0;
}

/**
 * Variant of delete_marshall_traverse() that bumps the
 * RSN of each traversed record in the database.
 *
 * This is needed to ensure that when rolling out our
 * empty record copy before remote deletion, we as the
 * record's dmaster keep a higher RSN than the non-dmaster
 * nodes. This is needed to prevent old copies from
 * resurrection in recoveries.
 */
static int delete_marshall_traverse_first(void *param, void *data)
{
	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_db_context *ctdb_db = dd->ctdb_db;
	struct ctdb_context *ctdb = ctdb_db->ctdb;
	struct ctdb_ltdb_header header;
	uint32_t lmaster;
	uint32_t hash = ctdb_hash(&(dd->key));
	int res;

	res = tdb_chainlock_nonblock(ctdb_db->ltdb->tdb, dd->key);
	if (res != 0) {
		recs->vdata->count.delete_list.skipped++;
		recs->vdata->count.delete_list.left--;
		talloc_free(dd);
		return 0;
	}

	/*
	 * Verify that the record is still empty, its RSN has not
	 * changed and that we are still its lmaster and dmaster.
	 */

	res = tdb_parse_record(ctdb_db->ltdb->tdb, dd->key,
			       vacuum_record_parser, &header);
	if (res != 0) {
		goto skip;
	}

	if (header.flags & CTDB_REC_RO_FLAGS) {
		DEBUG(DEBUG_INFO, (__location__ ": record with hash [0x%08x] "
				   "on database db[%s] has read-only flags. "
				   "skipping.\n",
				   hash, ctdb_db->db_name));
		goto skip;
	}

	if (header.dmaster != ctdb->pnn) {
		DEBUG(DEBUG_INFO, (__location__ ": record with hash [0x%08x] "
				   "on database db[%s] has been migrated away. "
				   "skipping.\n",
				   hash, ctdb_db->db_name));
		goto skip;
	}

	if (header.rsn != dd->hdr.rsn) {
		DEBUG(DEBUG_INFO, (__location__ ": record with hash [0x%08x] "
				   "on database db[%s] seems to have been "
				   "migrated away and back again (with empty "
				   "data). skipping.\n",
				   hash, ctdb_db->db_name));
		goto skip;
	}

	lmaster = ctdb_lmaster(ctdb_db->ctdb, &dd->key);

	if (lmaster != ctdb->pnn) {
		DEBUG(DEBUG_INFO, (__location__ ": not lmaster for record in "
				   "delete list (key hash [0x%08x], db[%s]). "
				   "Strange! skipping.\n",
				   hash, ctdb_db->db_name));
		goto skip;
	}

	/*
	 * Increment the record's RSN to ensure the dmaster (i.e. the current
	 * node) has the highest RSN of the record in the cluster.
	 * This is to prevent old record copies from resurrecting in recoveries
	 * if something should fail during the deletion process.
	 * Note that ctdb_ltdb_store_server() increments the RSN if called
	 * on the record's dmaster.
	 */

	res = ctdb_ltdb_store(ctdb_db, dd->key, &header, tdb_null);
	if (res != 0) {
		DEBUG(DEBUG_ERR, (__location__ ": Failed to store record with "
				  "key hash [0x%08x] on database db[%s].\n",
				  hash, ctdb_db->db_name));
		goto skip;
	}

	tdb_chainunlock(ctdb_db->ltdb->tdb, dd->key);

	goto done;

skip:
	tdb_chainunlock(ctdb_db->ltdb->tdb, dd->key);

	recs->vdata->count.delete_list.skipped++;
	recs->vdata->count.delete_list.left--;
	talloc_free(dd);
	dd = NULL;

done:
	if (dd == NULL) {
		return 0;
	}

	return delete_marshall_traverse(param, data);
}

/**
 * traverse function for the traversal of the delete_queue,
 * the fast-path vacuuming list.
 *
 *  - If the record has been migrated off the node
 *    or has been revived (filled with data) on the node,
 *    then skip the record.
 *
 *  - If the current node is the record's lmaster and it is
 *    a record that has never been migrated with data, then
 *    delete the record from the local tdb.
 *
 *  - If the current node is the record's lmaster and it has
 *    been migrated with data, then schedule it for the normal
 *    vacuuming procedure (i.e. add it to the delete_list).
 *
 *  - If the current node is NOT the record's lmaster then
 *    add it to the list of records that are to be sent to
 *    the lmaster with the VACUUM_FETCH message.
 */
static int delete_queue_traverse(void *param, void *data)
{
	struct delete_record_data *dd =
		talloc_get_type(data, struct delete_record_data);
	struct vacuum_data *vdata = talloc_get_type(param, struct vacuum_data);
	struct ctdb_db_context *ctdb_db = dd->ctdb_db;
	struct ctdb_context *ctdb = ctdb_db->ctdb; /* or dd->ctdb ??? */
	int res;
	struct ctdb_ltdb_header header;
	uint32_t lmaster;
	uint32_t hash = ctdb_hash(&(dd->key));

	vdata->count.delete_queue.total++;

	res = tdb_chainlock_nonblock(ctdb_db->ltdb->tdb, dd->key);
	if (res != 0) {
		vdata->count.delete_queue.error++;
		return 0;
	}

	res = tdb_parse_record(ctdb_db->ltdb->tdb, dd->key,
			       vacuum_record_parser, &header);
	if (res != 0) {
		goto skipped;
	}

	if (header.dmaster != ctdb->pnn) {
		/* The record has been migrated off the node. Skip. */
		goto skipped;
	}

	if (header.rsn != dd->hdr.rsn) {
		/*
		 * The record has been migrated off the node and back again.
		 * But not requeued for deletion. Skip it.
		 */
		goto skipped;
	}

	/*
	 * We are dmaster, and the record has no data, and it has
	 * not been migrated after it has been queued for deletion.
	 *
	 * At this stage, the record could still have been revived locally
	 * and last been written with empty data. This can only be
	 * fixed with the addition of an active or delete flag. (TODO)
	 */

	lmaster = ctdb_lmaster(ctdb_db->ctdb, &dd->key);

	if (lmaster != ctdb->pnn) {
		res = add_record_to_vacuum_fetch_list(vdata, dd->key);

		if (res != 0) {
			DEBUG(DEBUG_ERR,
			      (__location__ " Error adding record to list "
			       "of records to send to lmaster.\n"));
			vdata->count.delete_queue.error++;
		} else {
			vdata->count.delete_queue.added_to_vacuum_fetch_list++;
		}
		goto done;
	}

	/* use header->flags or dd->hdr.flags ?? */
	if (dd->hdr.flags & CTDB_REC_FLAG_MIGRATED_WITH_DATA) {
		res = add_record_to_delete_list(vdata, dd->key, &dd->hdr);

		if (res != 0) {
			DEBUG(DEBUG_ERR,
			      (__location__ " Error adding record to list "
			       "of records for deletion on lmaster.\n"));
			vdata->count.delete_queue.error++;
		} else {
			vdata->count.delete_queue.added_to_delete_list++;
		}
	} else {
		res = tdb_delete(ctdb_db->ltdb->tdb, dd->key);

		if (res != 0) {
			DEBUG(DEBUG_ERR,
			      (__location__ " Error deleting record with key "
			       "hash [0x%08x] from local data base db[%s].\n",
			       hash, ctdb_db->db_name));
			vdata->count.delete_queue.error++;
			goto done;
		}

		DEBUG(DEBUG_DEBUG,
		      (__location__ " Deleted record with key hash "
		       "[0x%08x] from local data base db[%s].\n",
		       hash, ctdb_db->db_name));
		vdata->count.delete_queue.deleted++;
	}

	goto done;

skipped:
	vdata->count.delete_queue.skipped++;

done:
	tdb_chainunlock(ctdb_db->ltdb->tdb, dd->key);

	return 0;
}

/**
 * Delete the records that we are lmaster and dmaster for and
 * that could be deleted on all other nodes via the TRY_DELETE_RECORDS
 * control.
 */
static int delete_record_traverse(void *param, void *data)
{
	struct delete_record_data *dd =
		talloc_get_type(data, struct delete_record_data);
	struct vacuum_data *vdata = talloc_get_type(param, struct vacuum_data);
	struct ctdb_db_context *ctdb_db = dd->ctdb_db;
	struct ctdb_context *ctdb = ctdb_db->ctdb;
	int res;
	struct ctdb_ltdb_header header;
	uint32_t lmaster;
	uint32_t hash = ctdb_hash(&(dd->key));

	res = tdb_chainlock(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->count.delete_list.local_error++;
		vdata->count.delete_list.left--;
		talloc_free(dd);
		return 0;
	}

	/*
	 * Verify that the record is still empty, its RSN has not
	 * changed and that we are still its lmaster and dmaster.
	 */

	res = tdb_parse_record(ctdb_db->ltdb->tdb, dd->key,
			       vacuum_record_parser, &header);
	if (res != 0) {
		goto skip;
	}

	if (header.flags & CTDB_REC_RO_FLAGS) {
		DEBUG(DEBUG_INFO, (__location__ ": record with hash [0x%08x] "
				   "on database db[%s] has read-only flags. "
				   "skipping.\n",
				   hash, ctdb_db->db_name));
		goto skip;
	}

	if (header.dmaster != ctdb->pnn) {
		DEBUG(DEBUG_INFO, (__location__ ": record with hash [0x%08x] "
				   "on database db[%s] has been migrated away. "
				   "skipping.\n",
				   hash, ctdb_db->db_name));
		goto skip;
	}

	if (header.rsn != dd->hdr.rsn + 1) {
		/*
		 * The record has been migrated off the node and back again.
		 * But not requeued for deletion. Skip it.
		 * (Note that the first marshall traverse has bumped the RSN
		 *  on disk.)
		 */
		DEBUG(DEBUG_INFO, (__location__ ": record with hash [0x%08x] "
				   "on database db[%s] seems to have been "
				   "migrated away and back again (with empty "
				   "data). skipping.\n",
				   hash, ctdb_db->db_name));
		goto skip;
	}

	lmaster = ctdb_lmaster(ctdb_db->ctdb, &dd->key);

	if (lmaster != ctdb->pnn) {
		DEBUG(DEBUG_INFO, (__location__ ": not lmaster for record in "
				   "delete list (key hash [0x%08x], db[%s]). "
				   "Strange! skipping.\n",
				   hash, ctdb_db->db_name));
		goto skip;
	}

	res = tdb_delete(ctdb_db->ltdb->tdb, dd->key);

	if (res != 0) {
		DEBUG(DEBUG_ERR,
		      (__location__ " Error deleting record with key hash "
		       "[0x%08x] from local data base db[%s].\n",
		       hash, ctdb_db->db_name));
		vdata->count.delete_list.local_error++;
		goto done;
	}

	DEBUG(DEBUG_DEBUG,
	      (__location__ " Deleted record with key hash [0x%08x] from "
	       "local data base db[%s].\n", hash, ctdb_db->db_name));

	vdata->count.delete_list.deleted++;
	goto done;

skip:
	vdata->count.delete_list.skipped++;

done:
	tdb_chainunlock(ctdb_db->ltdb->tdb, dd->key);

	talloc_free(dd);
	vdata->count.delete_list.left--;

	return 0;
}

/**
 * Traverse the delete_queue.
 * Records are either deleted directly or filled
 * into the delete list or the vacuum fetch lists
 * for further processing.
 */
static void ctdb_process_delete_queue(struct ctdb_db_context *ctdb_db,
				      struct vacuum_data *vdata)
{
	uint32_t sum;
	int ret;

	ret = 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->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->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->count.delete_queue.total,
		      (unsigned)sum));
	}

	if (vdata->count.delete_queue.total > 0) {
		DEBUG(DEBUG_INFO,
		      (__location__
		       " fast vacuuming delete_queue traverse statistics: "
		       "db[%s] "
		       "total[%u] "
		       "del[%u] "
		       "skp[%u] "
		       "err[%u] "
		       "adl[%u] "
		       "avf[%u]\n",
		       ctdb_db->db_name,
		       (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;
}

/**
 * read-only traverse of the database, looking for records that
 * might be able to be vacuumed.
 *
 * This is not done each time but only every tunable
 * VacuumFastPathCount times.
 */
static void ctdb_vacuum_traverse_db(struct ctdb_db_context *ctdb_db,
				    struct vacuum_data *vdata)
{
	int ret;

	ret = tdb_traverse_read(ctdb_db->ltdb->tdb, vacuum_traverse, vdata);
	if (ret == -1 || vdata->traverse_error) {
		DEBUG(DEBUG_ERR, (__location__ " Traverse error in vacuuming "
				  "'%s'\n", ctdb_db->db_name));
		return;
	}

	if (vdata->count.db_traverse.total > 0) {
		DEBUG(DEBUG_INFO,
		      (__location__
		       " full vacuuming db traverse statistics: "
		       "db[%s] "
		       "total[%u] "
		       "skp[%u] "
		       "err[%u] "
		       "sched[%u]\n",
		       ctdb_db->db_name,
		       (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;
}

/**
 * Process the vacuum fetch lists:
 * For records for which we are not the lmaster, tell the lmaster to
 * fetch the record.
 */
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;

	for (i = 0; i < ctdb->num_nodes; i++) {
		TDB_DATA data;
		struct ctdb_marshall_buffer *vfl = vdata->vacuum_fetch_list[i];

		if (ctdb->nodes[i]->pnn == ctdb->pnn) {
			continue;
		}

		if (vfl->count == 0) {
			continue;
		}

		DEBUG(DEBUG_INFO, ("Found %u records for lmaster %u in '%s'\n",
				   vfl->count, ctdb->nodes[i]->pnn,
				   ctdb_db->db_name));

		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;
}

/**
 * Process the delete list:
 *
 * This is the last step of vacuuming that consistently deletes
 * those records that have been migrated with data and can hence
 * not be deleted when leaving a node.
 *
 * In this step, the lmaster does the final deletion of those empty
 * records that it is also dmaster for. It has ususally received
 * at least some of these records previously from the former dmasters
 * with the vacuum fetch message.
 *
 * This last step is implemented as a 3-phase process to protect from
 * races leading to data corruption:
 *
 *  1) Send the lmaster's copy to all other active nodes with the
 *     RECEIVE_RECORDS control: The remote nodes store the lmaster's copy.
 *  2) Send the records that could successfully be stored remotely
 *     in step #1 to all active nodes with the TRY_DELETE_RECORDS
 *     control. The remote notes delete their local copy.
 *  3) The lmaster locally deletes its copies of all records that
 *     could successfully be deleted remotely in step #2.
 */
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_old *nodemap;
	uint32_t *active_nodes;
	int num_active_nodes;
	TALLOC_CTX *tmp_ctx;
	uint32_t sum;

	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;
	}

	vdata->count.delete_list.left = vdata->count.delete_list.total;

	/*
	 * get the list of currently active nodes
	 */

	ret = ctdb_ctrl_getnodemap(ctdb, TIMELIMIT(),
				   CTDB_CURRENT_NODE,
				   tmp_ctx,
				   &nodemap);
	if (ret != 0) {
		DEBUG(DEBUG_ERR,(__location__ " unable to get node map\n"));
		goto done;
	}

	active_nodes = list_of_active_nodes(ctdb, nodemap,
					    nodemap, /* talloc context */
					    false /* include self */);
	/* yuck! ;-) */
	num_active_nodes = talloc_get_size(active_nodes)/sizeof(*active_nodes);

	/*
	 * Now delete the records all active nodes in a three-phase process:
	 * 1) send all active remote nodes the current empty copy with this
	 *    node as DMASTER
	 * 2) if all nodes could store the new copy,
	 *    tell all the active remote nodes to delete all their copy
	 * 3) if all remote nodes deleted their record copy, delete it locally
	 */

	/*
	 * Step 1:
	 * Send currently empty record copy to all active nodes for storing.
	 */

	recs = talloc_zero(tmp_ctx, struct delete_records_list);
	if (recs == NULL) {
		DEBUG(DEBUG_ERR,(__location__ " Out of memory\n"));
		goto done;
	}
	recs->records = (struct ctdb_marshall_buffer *)
		talloc_zero_size(recs,
				 offsetof(struct ctdb_marshall_buffer, data));
	if (recs->records == NULL) {
		DEBUG(DEBUG_ERR,(__location__ " Out of memory\n"));
		goto done;
	}
	recs->records->db_id = ctdb_db->db_id;
	recs->vdata = vdata;

	/*
	 * traverse the tree of all records we want to delete and
	 * create a blob we can send to the other nodes.
	 *
	 * We call delete_marshall_traverse_first() to bump the
	 * records' RSNs in the database, to ensure we (as dmaster)
	 * keep the highest RSN of the records in the cluster.
	 */
	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 = ctdb_marshall_finish(recs->records);

	for (i = 0; i < num_active_nodes; i++) {
		struct ctdb_marshall_buffer *records;
		struct ctdb_rec_data_old *rec;
		int32_t res;
		TDB_DATA outdata;

		ret = ctdb_control(ctdb, active_nodes[i], 0,
				CTDB_CONTROL_RECEIVE_RECORDS, 0,
				indata, recs, &outdata, &res,
				NULL, NULL);
		if (ret != 0 || res != 0) {
			DEBUG(DEBUG_ERR, ("Error storing record copies on "
					  "node %u: ret[%d] res[%d]\n",
					  active_nodes[i], ret, res));
			goto done;
		}

		/*
		 * outdata contains the list of records coming back
		 * from the node: These are the records that the
		 * remote node could not store. We remove these from
		 * the list to process further.
		 */
		records = (struct ctdb_marshall_buffer *)outdata.dptr;
		rec = (struct ctdb_rec_data_old *)&records->data[0];
		while (records->count-- > 1) {
			TDB_DATA reckey, recdata;
			struct ctdb_ltdb_header *rechdr;
			struct delete_record_data *dd;

			reckey.dptr = &rec->data[0];
			reckey.dsize = rec->keylen;
			recdata.dptr = &rec->data[reckey.dsize];
			recdata.dsize = rec->datalen;

			if (recdata.dsize < sizeof(struct ctdb_ltdb_header)) {
				DEBUG(DEBUG_CRIT,(__location__ " bad ltdb record\n"));
				goto done;
			}
			rechdr = (struct ctdb_ltdb_header *)recdata.dptr;
			recdata.dptr += sizeof(*rechdr);
			recdata.dsize -= sizeof(*rechdr);

			dd = (struct delete_record_data *)trbt_lookup32(
					vdata->delete_list,
					ctdb_hash(&reckey));
			if (dd != NULL) {
				/*
				 * The other node could not store the record
				 * copy and it is the first node that failed.
				 * So we should remove it from the tree and
				 * update statistics.
				 */
				talloc_free(dd);
				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_old *)(rec->length + (uint8_t *)rec);
		}
	}

	if (vdata->count.delete_list.left == 0) {
		goto success;
	}

	/*
	 * Step 2:
	 * Send the remaining records to all active nodes for deletion.
	 *
	 * The lmaster's (i.e. our) copies of these records have been stored
	 * successfully on the other nodes.
	 */

	/*
	 * Create a marshall blob from the remaining list of records to delete.
	 */

	talloc_free(recs->records);

	recs->records = (struct ctdb_marshall_buffer *)
		talloc_zero_size(recs,
				 offsetof(struct ctdb_marshall_buffer, data));
	if (recs->records == NULL) {
		DEBUG(DEBUG_ERR,(__location__ " Out of memory\n"));
		goto done;
	}
	recs->records->db_id = ctdb_db->db_id;

	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 = ctdb_marshall_finish(recs->records);

	for (i = 0; i < num_active_nodes; i++) {
		struct ctdb_marshall_buffer *records;
		struct ctdb_rec_data_old *rec;
		int32_t res;
		TDB_DATA outdata;

		ret = ctdb_control(ctdb, active_nodes[i], 0,
				CTDB_CONTROL_TRY_DELETE_RECORDS, 0,
				indata, recs, &outdata, &res,
				NULL, NULL);
		if (ret != 0 || res != 0) {
			DEBUG(DEBUG_ERR, ("Failed to delete records on "
					  "node %u: ret[%d] res[%d]\n",
					  active_nodes[i], ret, res));
			goto done;
		}

		/*
		 * outdata contains the list of records coming back
		 * from the node: These are the records that the
		 * remote node could not delete. We remove these from
		 * the list to delete locally.
		 */
		records = (struct ctdb_marshall_buffer *)outdata.dptr;
		rec = (struct ctdb_rec_data_old *)&records->data[0];
		while (records->count-- > 1) {
			TDB_DATA reckey, recdata;
			struct ctdb_ltdb_header *rechdr;
			struct delete_record_data *dd;

			reckey.dptr = &rec->data[0];
			reckey.dsize = rec->keylen;
			recdata.dptr = &rec->data[reckey.dsize];
			recdata.dsize = rec->datalen;

			if (recdata.dsize < sizeof(struct ctdb_ltdb_header)) {
				DEBUG(DEBUG_CRIT,(__location__ " bad ltdb record\n"));
				goto done;
			}
			rechdr = (struct ctdb_ltdb_header *)recdata.dptr;
			recdata.dptr += sizeof(*rechdr);
			recdata.dsize -= sizeof(*rechdr);

			dd = (struct delete_record_data *)trbt_lookup32(
					vdata->delete_list,
					ctdb_hash(&reckey));
			if (dd != NULL) {
				/*
				 * The other node could not delete the
				 * record and it is the first node that
				 * failed. So we should remove it from
				 * the tree and update statistics.
				 */
				talloc_free(dd);
				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_old *)(rec->length + (uint8_t *)rec);
		}
	}

	if (vdata->count.delete_list.left == 0) {
		goto success;
	}

	/*
	 * Step 3:
	 * Delete the remaining records locally.
	 *
	 * These records have successfully been deleted on all
	 * active remote nodes.
	 */

	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->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->count.delete_list.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->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->count.delete_list.total,
		      (unsigned)sum));
	}

	if (vdata->count.delete_list.total > 0) {
		DEBUG(DEBUG_INFO,
		      (__location__
		       " vacuum delete list statistics: "
		       "db[%s] "
		       "total[%u] "
		       "del[%u] "
		       "skip[%u] "
		       "rem.err[%u] "
		       "loc.err[%u] "
		       "left[%u]\n",
		       ctdb_db->db_name,
		       (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));
	}

done:
	talloc_free(tmp_ctx);

	return;
}

/**
 * initialize the vacuum_data
 */
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 = 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,
						struct ctdb_marshall_buffer *,
						ctdb->num_nodes);
	if (vdata->vacuum_fetch_list == NULL) {
		DEBUG(DEBUG_ERR,(__location__ " Out of memory\n"));
		goto fail;
	}
	for (i = 0; i < ctdb->num_nodes; i++) {
		vdata->vacuum_fetch_list[i] = (struct ctdb_marshall_buffer *)
			talloc_zero_size(vdata->vacuum_fetch_list,
					 offsetof(struct ctdb_marshall_buffer, data));
		if (vdata->vacuum_fetch_list[i] == NULL) {
			DEBUG(DEBUG_ERR,(__location__ " Out of memory\n"));
			talloc_free(vdata);
			return NULL;
		}
		vdata->vacuum_fetch_list[i]->db_id = ctdb_db->db_id;
	}

	return vdata;

fail:
	talloc_free(vdata);
	return NULL;
}

/**
 * Vacuum a DB:
 *  - Always do the fast vacuuming run, which traverses
 *    the in-memory delete queue: these records have been
 *    scheduled for deletion.
 *  - Only if explicitly requested, the database is traversed
 *    in order to use the traditional heuristics on empty records
 *    to trigger deletion.
 *    This is done only every VacuumFastPathCount'th vacuuming run.
 *
 * The traverse runs fill two lists:
 *
 * - The delete_list:
 *   This is the list of empty records the current
 *   node is lmaster and dmaster for. These records are later
 *   deleted first on other nodes and then locally.
 *
 *   The fast vacuuming run has a short cut for those records
 *   that have never been migrated with data: these records
 *   are immediately deleted locally, since they have left
 *   no trace on other nodes.
 *
 * - The vacuum_fetch lists
 *   (one for each other lmaster node):
 *   The records in this list are sent for deletion to
 *   their lmaster in a bulk VACUUM_FETCH message.
 *
 *   The lmaster then migrates all these records to itelf
 *   so that they can be vacuumed there.
 *
 * This executes in the child context.
 */
static int ctdb_vacuum_db(struct ctdb_db_context *ctdb_db,
			  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",
			   full_vacuum_run ? "full" : "fast",
			   ctdb_db->db_name, ctdb_db->db_id));

	ret = ctdb_ctrl_getvnnmap(ctdb, TIMELIMIT(), CTDB_CURRENT_NODE, ctdb, &ctdb->vnn_map);
	if (ret != 0) {
		DEBUG(DEBUG_ERR, ("Unable to get vnnmap from local node\n"));
		return ret;
	}

	pnn = ctdb_ctrl_getpnn(ctdb, TIMELIMIT(), CTDB_CURRENT_NODE);
	if (pnn == -1) {
		DEBUG(DEBUG_ERR, ("Unable to get pnn from local node\n"));
		return -1;
	}

	ctdb->pnn = pnn;

	tmp_ctx = talloc_new(ctdb_db);
	if (tmp_ctx == NULL) {
		DEBUG(DEBUG_ERR, ("Out of memory!\n"));
		return -1;
	}

	vdata = ctdb_vacuum_init_vacuum_data(ctdb_db, tmp_ctx);
	if (vdata == NULL) {
		talloc_free(tmp_ctx);
		return -1;
	}

	if (full_vacuum_run) {
		ctdb_vacuum_traverse_db(ctdb_db, vdata);
	}

	ctdb_process_delete_queue(ctdb_db, vdata);

	ctdb_process_vacuum_fetch_lists(ctdb_db, vdata);

	ctdb_process_delete_list(ctdb_db, vdata);

	talloc_free(tmp_ctx);

	/* this ensures we run our event queue */
	ctdb_ctrl_getpnn(ctdb, TIMELIMIT(), CTDB_CURRENT_NODE);

	return 0;
}

/*
 * repack and vaccum a db
 * called from the child context
 */
static int ctdb_vacuum_and_repack_db(struct ctdb_db_context *ctdb_db,
				     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;
	int ret;

	if (ctdb_vacuum_db(ctdb_db, full_vacuum_run) != 0) {
		DEBUG(DEBUG_ERR,(__location__ " Failed to vacuum '%s'\n", name));
	}

	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;
	}

	/*
	 * decide if a repack is necessary
	 */
	if ((repack_limit == 0 || (uint32_t)freelist_size < repack_limit))
	{
		return 0;
	}

	DEBUG(DEBUG_INFO, ("Repacking %s with %u freelist entries\n",
			   name, freelist_size));

	ret = tdb_repack(ctdb_db->ltdb->tdb);
	if (ret != 0) {
		DEBUG(DEBUG_ERR,(__location__ " Failed to repack '%s'\n", name));
		return -1;
	}

	return 0;
}

static uint32_t get_vacuum_interval(struct ctdb_db_context *ctdb_db)
{
	uint32_t interval = ctdb_db->ctdb->tunable.vacuum_interval;

	return interval;
}

static int vacuum_child_destructor(struct ctdb_vacuum_child_context *child_ctx)
{
	double l = timeval_elapsed(&child_ctx->start_time);
	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) {
		ctdb_kill(ctdb, child_ctx->child_pid, SIGKILL);
	} else {
		/* Bump the number of successful fast-path runs. */
		child_ctx->vacuum_handle->fast_path_count++;
	}

	DLIST_REMOVE(ctdb->vacuumers, child_ctx);

	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 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);

	DEBUG(DEBUG_ERR,("Vacuuming child process timed out for db %s\n", child_ctx->vacuum_handle->ctdb_db->db_name));

	child_ctx->status = VACUUM_TIMEOUT;

	talloc_free(child_ctx);
}


/*
 * this event is generated when a vacuum child process has completed
 */
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;
	int ret;

	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 = 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));
	} else {
		child_ctx->status = VACUUM_OK;
	}

	talloc_free(child_ctx);
}

/*
 * this event is called every time we need to start a new vacuum process
 */
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 ctdb_context *ctdb = ctdb_db->ctdb;
	struct ctdb_vacuum_child_context *child_ctx;
	struct tevent_fd *fde;
	int ret;

	/* 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->recovery_mode == CTDB_RECOVERY_ACTIVE ? "in recovery"
				   : ctdb->freeze_mode[ctdb_db->priority] == CTDB_FREEZE_PENDING
				   ? "freeze pending"
				   : "frozen"));
		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;
	}

	child_ctx = talloc(vacuum_handle, struct ctdb_vacuum_child_context);
	if (child_ctx == NULL) {
		DEBUG(DEBUG_CRIT, (__location__ " Failed to allocate child context for vacuuming of %s\n", ctdb_db->db_name));
		ctdb_fatal(ctdb, "Out of memory when crating vacuum child context. Shutting down\n");
	}


	ret = pipe(child_ctx->fd);
	if (ret != 0) {
		talloc_free(child_ctx);
		DEBUG(DEBUG_ERR, ("Failed to create pipe for vacuum child process.\n"));
		tevent_add_timer(ctdb->ev, vacuum_handle,
				 timeval_current_ofs(get_vacuum_interval(ctdb_db), 0),
				 ctdb_vacuum_event, vacuum_handle);
		return;
	}

	if (vacuum_handle->fast_path_count > ctdb->tunable.vacuum_fast_path_count) {
		vacuum_handle->fast_path_count = 0;
	}

	child_ctx->child_pid = ctdb_fork(ctdb);
	if (child_ctx->child_pid == (pid_t)-1) {
		close(child_ctx->fd[0]);
		close(child_ctx->fd[1]);
		talloc_free(child_ctx);
		DEBUG(DEBUG_ERR, ("Failed to fork vacuum child process.\n"));
		tevent_add_timer(ctdb->ev, vacuum_handle,
				 timeval_current_ofs(get_vacuum_interval(ctdb_db), 0),
				 ctdb_vacuum_event, vacuum_handle);
		return;
	}


	if (child_ctx->child_pid == 0) {
		char cc = 0;
		bool full_vacuum_run = false;
		close(child_ctx->fd[0]);

		DEBUG(DEBUG_INFO,("Vacuuming child process %d for db %s started\n", getpid(), ctdb_db->db_name));
		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);
		}

		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, full_vacuum_run);

		sys_write(child_ctx->fd[1], &cc, 1);
		_exit(0);
	}

	set_close_on_exec(child_ctx->fd[0]);
	close(child_ctx->fd[1]);

	child_ctx->status = VACUUM_RUNNING;
	child_ctx->start_time = timeval_current();

	DLIST_ADD(ctdb->vacuumers, child_ctx);
	talloc_set_destructor(child_ctx, vacuum_child_destructor);

	/*
	 * Clear the fastpath vacuuming list in the parent.
	 */
	talloc_free(ctdb_db->delete_queue);
	ctdb_db->delete_queue = trbt_create(ctdb_db, 0);
	if (ctdb_db->delete_queue == NULL) {
		/* fatal here? ... */
		ctdb_fatal(ctdb, "Out of memory when re-creating vacuum tree "
				 "in parent context. Shutting down\n");
	}

	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 = 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;
	child_ctx->vacuum_handle = vacuum_handle;
}

void ctdb_stop_vacuuming(struct ctdb_context *ctdb)
{
	/* Simply free them all. */
	while (ctdb->vacuumers) {
		DEBUG(DEBUG_INFO, ("Aborting vacuuming for %s (%i)\n",
			   ctdb->vacuumers->vacuum_handle->ctdb_db->db_name,
			   (int)ctdb->vacuumers->child_pid));
		/* vacuum_child_destructor kills it, removes from list */
		talloc_free(ctdb->vacuumers);
	}
}

/* this function initializes the vacuuming context for a database
 * starts the vacuuming events
 */
int ctdb_vacuum_init(struct ctdb_db_context *ctdb_db)
{
	if (ctdb_db->persistent != 0) {
		DEBUG(DEBUG_ERR,("Vacuuming is disabled for persistent database %s\n", ctdb_db->db_name));
		return 0;
	}

	ctdb_db->vacuum_handle = talloc(ctdb_db, struct ctdb_vacuum_handle);
	CTDB_NO_MEMORY(ctdb_db->ctdb, ctdb_db->vacuum_handle);

	ctdb_db->vacuum_handle->ctdb_db         = ctdb_db;
	ctdb_db->vacuum_handle->fast_path_count = 0;

	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;
}

static void remove_record_from_delete_queue(struct ctdb_db_context *ctdb_db,
					    const struct ctdb_ltdb_header *hdr,
					    const TDB_DATA key)
{
	struct delete_record_data *kd;
	uint32_t hash;

	hash = (uint32_t)ctdb_hash(&key);

	DEBUG(DEBUG_DEBUG, (__location__
			    " remove_record_from_delete_queue: "
			    "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),
			     hdr->flags & CTDB_REC_FLAG_MIGRATED_WITH_DATA ? "yes" : "no"));

	kd = (struct delete_record_data *)trbt_lookup32(ctdb_db->delete_queue, hash);
	if (kd == NULL) {
		DEBUG(DEBUG_DEBUG, (__location__
				    " remove_record_from_delete_queue: "
				    "record not in queue (hash[0x%08x])\n.",
				    hash));
		return;
	}

	if ((kd->key.dsize != key.dsize) ||
	    (memcmp(kd->key.dptr, key.dptr, key.dsize) != 0))
	{
		DEBUG(DEBUG_DEBUG, (__location__
				    " remove_record_from_delete_queue: "
				    "hash collision for key with hash[0x%08x] "
				    "in db[%s] - skipping\n",
				    hash, ctdb_db->db_name));
		return;
	}

	DEBUG(DEBUG_DEBUG, (__location__
			    " remove_record_from_delete_queue: "
			    "removing key with hash[0x%08x]\n",
			     hash));

	talloc_free(kd);

	return;
}

/**
 * Insert a record into the ctdb_db context's delete queue,
 * handling hash collisions.
 */
static int insert_record_into_delete_queue(struct ctdb_db_context *ctdb_db,
					   const struct ctdb_ltdb_header *hdr,
					   TDB_DATA key)
{
	struct delete_record_data *kd;
	uint32_t hash;
	int ret;

	hash = (uint32_t)ctdb_hash(&key);

	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),
			    hdr->flags & CTDB_REC_FLAG_MIGRATED_WITH_DATA ? "yes" : "no"));

	kd = (struct delete_record_data *)trbt_lookup32(ctdb_db->delete_queue, hash);
	if (kd != NULL) {
		if ((kd->key.dsize != key.dsize) ||
		    (memcmp(kd->key.dptr, key.dptr, key.dsize) != 0))
		{
			DEBUG(DEBUG_INFO,
			      (__location__ " schedule for deletion: "
			       "hash collision for key hash [0x%08x]. "
			       "Skipping the record.\n", hash));
			return 0;
		} else {
			DEBUG(DEBUG_DEBUG,
			      (__location__ " schedule for deletion: "
			       "updating entry for key with hash [0x%08x].\n",
			       hash));
		}
	}

	ret = insert_delete_record_data_into_tree(ctdb_db->ctdb, ctdb_db,
						  ctdb_db->delete_queue,
						  hdr, key);
	if (ret != 0) {
		DEBUG(DEBUG_INFO,
		      (__location__ " schedule for deletion: error "
		       "inserting key with hash [0x%08x] into delete queue\n",
		       hash));
		return -1;
	}

	return 0;
}

/**
 * Schedule a record for deletetion.
 * Called from the parent context.
 */
int32_t ctdb_control_schedule_for_deletion(struct ctdb_context *ctdb,
					   TDB_DATA indata)
{
	struct ctdb_control_schedule_for_deletion *dd;
	struct ctdb_db_context *ctdb_db;
	int ret;
	TDB_DATA key;

	dd = (struct ctdb_control_schedule_for_deletion *)indata.dptr;

	ctdb_db = find_ctdb_db(ctdb, dd->db_id);
	if (ctdb_db == NULL) {
		DEBUG(DEBUG_ERR, (__location__ " Unknown db id 0x%08x\n",
				  dd->db_id));
		return -1;
	}

	key.dsize = dd->keylen;
	key.dptr = dd->key;

	ret = insert_record_into_delete_queue(ctdb_db, &dd->hdr, key);

	return ret;
}

int32_t ctdb_local_schedule_for_deletion(struct ctdb_db_context *ctdb_db,
					 const struct ctdb_ltdb_header *hdr,
					 TDB_DATA key)
{
	int ret;
	struct ctdb_control_schedule_for_deletion *dd;
	TDB_DATA indata;
	int32_t status;

	if (ctdb_db->ctdb->ctdbd_pid == getpid()) {
		/* main daemon - directly queue */
		ret = insert_record_into_delete_queue(ctdb_db, hdr, key);

		return ret;
	}

	/* 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.
	*/
	if (!ctdb_db->ctdb->can_send_controls) {
		return -1;
	}


	/* child process: send the main daemon a control */
	indata.dsize = offsetof(struct ctdb_control_schedule_for_deletion, key) + key.dsize;
	indata.dptr = talloc_zero_array(ctdb_db, uint8_t, indata.dsize);
	if (indata.dptr == NULL) {
		DEBUG(DEBUG_ERR, (__location__ " out of memory\n"));
		return -1;
	}
	dd = (struct ctdb_control_schedule_for_deletion *)(void *)indata.dptr;
	dd->db_id = ctdb_db->db_id;
	dd->hdr = *hdr;
	dd->keylen = key.dsize;
	memcpy(dd->key, key.dptr, key.dsize);

	ret = ctdb_control(ctdb_db->ctdb,
			   CTDB_CURRENT_NODE,
			   ctdb_db->db_id,
			   CTDB_CONTROL_SCHEDULE_FOR_DELETION,
			   CTDB_CTRL_FLAG_NOREPLY, /* flags */
			   indata,
			   NULL, /* mem_ctx */
			   NULL, /* outdata */
			   &status,
			   NULL, /* timeout : NULL == wait forever */
			   NULL); /* error message */

	talloc_free(indata.dptr);

	if (ret != 0 || status != 0) {
		DEBUG(DEBUG_ERR, (__location__ " Error sending "
				  "SCHEDULE_FOR_DELETION "
				  "control.\n"));
		if (status != 0) {
			ret = -1;
		}
	}

	return ret;
}

void ctdb_local_remove_from_delete_queue(struct ctdb_db_context *ctdb_db,
					 const struct ctdb_ltdb_header *hdr,
					 const TDB_DATA key)
{
	if (ctdb_db->ctdb->ctdbd_pid != getpid()) {
		/*
		 * Only remove the record from the delete queue if called
		 * in the main daemon.
		 */
		return;
	}

	remove_record_from_delete_queue(ctdb_db, hdr, key);

	return;
}