/*
* __stat_tree_walk --
* Gather btree statistics that require traversing the tree.
*/
static int
__stat_tree_walk(WT_SESSION_IMPL *session)
{
WT_BTREE *btree;
WT_DECL_RET;
WT_DSRC_STATS **stats;
WT_REF *next_walk;
btree = S2BT(session);
stats = btree->dhandle->stats;
/*
* Clear the statistics we're about to count.
*/
WT_STAT_SET(session, stats, btree_column_deleted, 0);
WT_STAT_SET(session, stats, btree_column_fix, 0);
WT_STAT_SET(session, stats, btree_column_internal, 0);
WT_STAT_SET(session, stats, btree_column_rle, 0);
WT_STAT_SET(session, stats, btree_column_variable, 0);
WT_STAT_SET(session, stats, btree_entries, 0);
WT_STAT_SET(session, stats, btree_overflow, 0);
WT_STAT_SET(session, stats, btree_row_internal, 0);
WT_STAT_SET(session, stats, btree_row_leaf, 0);
next_walk = NULL;
while ((ret = __wt_tree_walk(
session, &next_walk, 0)) == 0 && next_walk != NULL) {
WT_WITH_PAGE_INDEX(session,
ret = __stat_page(session, next_walk->page, stats));
WT_RET(ret);
}
return (ret == WT_NOTFOUND ? 0 : ret);
}
/*
* __wt_btcur_next_random --
* Move to a random record in the tree.
*/
int
__wt_btcur_next_random(WT_CURSOR_BTREE *cbt)
{
WT_BTREE *btree;
WT_DECL_RET;
WT_SESSION_IMPL *session;
WT_UPDATE *upd;
session = (WT_SESSION_IMPL *)cbt->iface.session;
btree = cbt->btree;
/*
* Only supports row-store: applications can trivially select a random
* value from a column-store, if there were any reason to do so.
*/
if (btree->type != BTREE_ROW)
WT_RET(ENOTSUP);
WT_STAT_FAST_CONN_INCR(session, cursor_next);
WT_STAT_FAST_DATA_INCR(session, cursor_next);
WT_RET(__cursor_func_init(cbt, 1));
WT_WITH_PAGE_INDEX(session,
ret = __wt_row_random(session, cbt));
WT_ERR(ret);
if (__cursor_valid(cbt, &upd))
WT_ERR(__wt_kv_return(session, cbt, upd));
else
WT_ERR(__wt_btcur_search_near(cbt, NULL));
err: if (ret != 0)
WT_TRET(__cursor_reset(cbt));
return (ret);
}
/*
* __cursor_row_search --
* Row-store search from an application cursor.
*/
static inline int
__cursor_row_search(WT_SESSION_IMPL *session, WT_CURSOR_BTREE *cbt, int insert)
{
WT_DECL_RET;
WT_WITH_PAGE_INDEX(session,
ret = __wt_row_search(session, &cbt->iface.key, NULL, cbt, insert));
return (ret);
}
/*
* __cursor_col_search --
* Column-store search from an application cursor.
*/
static inline int
__cursor_col_search(WT_SESSION_IMPL *session, WT_CURSOR_BTREE *cbt)
{
WT_DECL_RET;
WT_WITH_PAGE_INDEX(session,
ret = __wt_col_search(session, cbt->iface.recno, NULL, cbt));
return (ret);
}
/*
* __wt_btree_stat_init --
* Initialize the Btree statistics.
*/
int
__wt_btree_stat_init(WT_SESSION_IMPL *session, WT_CURSOR_STAT *cst)
{
WT_BM *bm;
WT_BTREE *btree;
WT_DECL_RET;
WT_DSRC_STATS **stats;
WT_REF *next_walk;
btree = S2BT(session);
bm = btree->bm;
stats = btree->dhandle->stats;
WT_RET(bm->stat(bm, session, stats[0]));
WT_STAT_SET(session, stats, btree_fixed_len, btree->bitcnt);
WT_STAT_SET(session, stats, btree_maximum_depth, btree->maximum_depth);
WT_STAT_SET(session, stats, btree_maxintlkey, btree->maxintlkey);
WT_STAT_SET(session, stats, btree_maxintlpage, btree->maxintlpage);
WT_STAT_SET(session, stats, btree_maxleafkey, btree->maxleafkey);
WT_STAT_SET(session, stats, btree_maxleafpage, btree->maxleafpage);
WT_STAT_SET(session, stats, btree_maxleafvalue, btree->maxleafvalue);
/* Everything else is really, really expensive. */
if (!F_ISSET(cst, WT_CONN_STAT_ALL))
return (0);
/*
* Clear the statistics we're about to count.
*/
WT_STAT_SET(session, stats, btree_column_deleted, 0);
WT_STAT_SET(session, stats, btree_column_fix, 0);
WT_STAT_SET(session, stats, btree_column_internal, 0);
WT_STAT_SET(session, stats, btree_column_rle, 0);
WT_STAT_SET(session, stats, btree_column_variable, 0);
WT_STAT_SET(session, stats, btree_entries, 0);
WT_STAT_SET(session, stats, btree_overflow, 0);
WT_STAT_SET(session, stats, btree_row_internal, 0);
WT_STAT_SET(session, stats, btree_row_leaf, 0);
next_walk = NULL;
while ((ret = __wt_tree_walk(
session, &next_walk, 0)) == 0 && next_walk != NULL) {
WT_WITH_PAGE_INDEX(session,
ret = __stat_page(session, next_walk->page, stats));
WT_RET(ret);
}
return (ret == WT_NOTFOUND ? 0 : ret);
}
/*
* __wt_btcur_next_random --
* Move to a random record in the tree. There are two algorithms, one
* where we select a record at random from the whole tree on each
* retrieval and one where we first select a record at random from the
* whole tree, and then subsequently sample forward from that location.
* The sampling approach allows us to select reasonably uniform random
* points from unbalanced trees.
*/
int
__wt_btcur_next_random(WT_CURSOR_BTREE *cbt)
{
WT_BTREE *btree;
WT_DECL_RET;
WT_SESSION_IMPL *session;
WT_UPDATE *upd;
wt_off_t size;
uint64_t skip;
session = (WT_SESSION_IMPL *)cbt->iface.session;
btree = cbt->btree;
/*
* Only supports row-store: applications can trivially select a random
* value from a column-store, if there were any reason to do so.
*/
if (btree->type != BTREE_ROW)
WT_RET_MSG(session, ENOTSUP,
"WT_CURSOR.next_random only supported by row-store tables");
WT_STAT_CONN_INCR(session, cursor_next);
WT_STAT_DATA_INCR(session, cursor_next);
/*
* If retrieving random values without sampling, or we don't have a
* page reference, pick a roughly random leaf page in the tree.
*/
if (cbt->ref == NULL || cbt->next_random_sample_size == 0) {
/*
* Skip past the sample size of the leaf pages in the tree
* between each random key return to compensate for unbalanced
* trees.
*
* Use the underlying file size divided by its block allocation
* size as our guess of leaf pages in the file (this can be
* entirely wrong, as it depends on how many pages are in this
* particular checkpoint, how large the leaf and internal pages
* really are, and other factors). Then, divide that value by
* the configured sample size and increment the final result to
* make sure tiny files don't leave us with a skip value of 0.
*
* !!!
* Ideally, the number would be prime to avoid restart issues.
*/
if (cbt->next_random_sample_size != 0) {
WT_ERR(btree->bm->size(btree->bm, session, &size));
cbt->next_random_leaf_skip = (uint64_t)
((size / btree->allocsize) /
cbt->next_random_sample_size) + 1;
}
/*
* Choose a leaf page from the tree.
*/
WT_ERR(__cursor_func_init(cbt, true));
WT_WITH_PAGE_INDEX(
session, ret = __wt_row_random_descent(session, cbt));
WT_ERR(ret);
} else {
/*
* Read through the tree, skipping leaf pages. Be cautious about
* the skip count: if the last leaf page skipped was also the
* last leaf page in the tree, it may be set to zero on return
* with the end-of-walk condition.
*
* Pages read for data sampling aren't "useful"; don't update
* the read generation of pages already in memory, and if a page
* is read, set its generation to a low value so it is evicted
* quickly.
*/
for (skip =
cbt->next_random_leaf_skip; cbt->ref == NULL || skip > 0;)
WT_ERR(__wt_tree_walk_skip(session, &cbt->ref, &skip,
WT_READ_NO_GEN |
WT_READ_SKIP_INTL | WT_READ_WONT_NEED));
}
/*
* Select a random entry from the leaf page. If it's not valid, move to
* the next entry, if that doesn't work, move to the previous entry.
*/
WT_ERR(__wt_row_random_leaf(session, cbt));
if (__cursor_valid(cbt, &upd))
WT_ERR(__wt_kv_return(session, cbt, upd));
else {
if ((ret = __wt_btcur_next(cbt, false)) == WT_NOTFOUND)
ret = __wt_btcur_prev(cbt, false);
WT_ERR(ret);
}
return (0);
err: WT_TRET(__cursor_reset(cbt));
return (ret);
}
/*
* __wt_verify --
* Verify a file.
*/
int
__wt_verify(WT_SESSION_IMPL *session, const char *cfg[])
{
WT_BM *bm;
WT_BTREE *btree;
WT_CKPT *ckptbase, *ckpt;
WT_DECL_RET;
WT_VSTUFF *vs, _vstuff;
size_t root_addr_size;
uint8_t root_addr[WT_BTREE_MAX_ADDR_COOKIE];
bool bm_start, quit;
btree = S2BT(session);
bm = btree->bm;
ckptbase = NULL;
bm_start = false;
WT_CLEAR(_vstuff);
vs = &_vstuff;
WT_ERR(__wt_scr_alloc(session, 0, &vs->max_key));
WT_ERR(__wt_scr_alloc(session, 0, &vs->max_addr));
WT_ERR(__wt_scr_alloc(session, 0, &vs->tmp1));
WT_ERR(__wt_scr_alloc(session, 0, &vs->tmp2));
WT_ERR(__wt_scr_alloc(session, 0, &vs->tmp3));
WT_ERR(__wt_scr_alloc(session, 0, &vs->tmp4));
/* Check configuration strings. */
WT_ERR(__verify_config(session, cfg, vs));
/* Optionally dump specific block offsets. */
WT_ERR(__verify_config_offsets(session, cfg, &quit));
if (quit)
goto done;
/* Get a list of the checkpoints for this file. */
WT_ERR(
__wt_meta_ckptlist_get(session, btree->dhandle->name, &ckptbase));
/* Inform the underlying block manager we're verifying. */
WT_ERR(bm->verify_start(bm, session, ckptbase, cfg));
bm_start = true;
/* Loop through the file's checkpoints, verifying each one. */
WT_CKPT_FOREACH(ckptbase, ckpt) {
WT_ERR(__wt_verbose(session, WT_VERB_VERIFY,
"%s: checkpoint %s", btree->dhandle->name, ckpt->name));
/* Fake checkpoints require no work. */
if (F_ISSET(ckpt, WT_CKPT_FAKE))
continue;
/* House-keeping between checkpoints. */
__verify_checkpoint_reset(vs);
if (WT_VRFY_DUMP(vs))
WT_ERR(__wt_msg(session, "%s: checkpoint %s",
btree->dhandle->name, ckpt->name));
/* Load the checkpoint. */
WT_ERR(bm->checkpoint_load(bm, session,
ckpt->raw.data, ckpt->raw.size,
root_addr, &root_addr_size, true));
/*
* Ignore trees with no root page.
* Verify, then discard the checkpoint from the cache.
*/
if (root_addr_size != 0 &&
(ret = __wt_btree_tree_open(
session, root_addr, root_addr_size)) == 0) {
if (WT_VRFY_DUMP(vs))
WT_ERR(__wt_msg(session, "Root: %s %s",
__wt_addr_string(session,
root_addr, root_addr_size, vs->tmp1),
__wt_page_type_string(
btree->root.page->type)));
WT_WITH_PAGE_INDEX(session,
ret = __verify_tree(session, &btree->root, vs));
WT_TRET(__wt_cache_op(session, WT_SYNC_DISCARD));
}
/* Unload the checkpoint. */
WT_TRET(bm->checkpoint_unload(bm, session));
WT_ERR(ret);
/* Display the tree shape. */
if (vs->dump_shape)
WT_ERR(__verify_tree_shape(session, vs));
}
