/**
* flush the backend
*
* @remark this function is called during ham_flush
@note This is a B+-tree 'backend' method.
*/
static ham_status_t
my_fun_flush(ham_btree_t *be)
{
ham_db_t *db=be_get_db(be);
db_indexdata_t *indexdata=env_get_indexdata_ptr(db_get_env(db),
db_get_indexdata_offset(db));
/*
* nothing to do if the backend was not touched
*/
if (!be_is_dirty(be))
return (0);
index_set_max_keys(indexdata, btree_get_maxkeys(be));
index_set_keysize(indexdata, be_get_keysize(be));
index_set_self(indexdata, btree_get_rootpage(be));
index_set_flags(indexdata, be_get_flags(be));
index_set_recno(indexdata, be_get_recno(be));
index_clear_reserved(indexdata);
env_set_dirty(db_get_env(db));
be_set_dirty(be, HAM_FALSE);
return (0);
}
static void
print_database(ham_db_t *db, ham_u16_t dbname, int full)
{
ham_btree_t *be;
ham_cursor_t *cursor;
ham_status_t st;
ham_key_t key;
ham_record_t rec;
unsigned num_items=0, ext_keys=0, min_key_size=0xffffffff,
max_key_size=0, min_rec_size=0xffffffff, max_rec_size=0,
total_key_size=0, total_rec_size=0;
be=(ham_btree_t *)db_get_backend(db);
memset(&key, 0, sizeof(key));
memset(&rec, 0, sizeof(rec));
printf("\n");
printf(" database %d (0x%x)\n", (int)dbname, (int)dbname);
printf(" max key size: %u\n",
be_get_keysize(be));
printf(" max keys per page: %u\n",
btree_get_maxkeys(be));
printf(" address of root page: %llu\n",
(long long unsigned int)btree_get_rootpage(be));
printf(" flags: 0x%04x\n",
db_get_rt_flags(db));
if (!full)
return;
st=ham_cursor_create(db, 0, 0, &cursor);
if (st!=HAM_SUCCESS)
error("ham_cursor_create", st);
while (1) {
st=ham_cursor_move(cursor, &key, &rec, HAM_CURSOR_NEXT);
if (st!=HAM_SUCCESS) {
/* reached end of the database? */
if (st==HAM_KEY_NOT_FOUND)
break;
else
error("ham_cursor_next", st);
}
num_items++;
if (key.size<min_key_size)
min_key_size=key.size;
if (key.size>max_key_size)
max_key_size=key.size;
if (rec.size<min_rec_size)
min_rec_size=rec.size;
if (rec.size>max_rec_size)
max_rec_size=rec.size;
if (key.size>db_get_keysize(db))
ext_keys++;
total_key_size+=key.size;
total_rec_size+=rec.size;
}
ham_cursor_close(cursor);
printf(" number of items: %u\n", num_items);
if (num_items==0)
return;
printf(" average key size: %u\n", total_key_size/num_items);
printf(" minimum key size: %u\n", min_key_size);
printf(" maximum key size: %u\n", max_key_size);
printf(" number of extended keys:%u\n", ext_keys);
printf(" total keys (bytes): %u\n", total_key_size);
printf(" average record size: %u\n", total_rec_size/num_items);
printf(" minimum record size: %u\n", min_rec_size);
printf(" maximum record size: %u\n", min_rec_size);
printf(" total records (bytes): %u\n", total_rec_size);
}
ham_status_t
btree_find_cursor(ham_btree_t *be, ham_bt_cursor_t *cursor,
ham_key_t *key, ham_record_t *record, ham_u32_t flags)
{
ham_status_t st;
ham_page_t *page = NULL;
btree_node_t *node = NULL;
int_key_t *entry;
ham_s32_t idx = -1;
ham_db_t *db=be_get_db(be);
find_hints_t hints = {flags, flags, 0, HAM_FALSE, HAM_FALSE, 1};
btree_find_get_hints(&hints, db, key);
if (hints.key_is_out_of_bounds) {
stats_update_find_fail_oob(db, &hints);
return HAM_KEY_NOT_FOUND;
}
if (hints.try_fast_track) {
/*
* see if we get a sure hit within this btree leaf; if not, revert to
* regular scan
*
* As this is a speed-improvement hint re-using recent material, the
* page should still sit in the cache, or we're using old info, which
* should be discarded.
*/
st = db_fetch_page(&page, db, hints.leaf_page_addr, DB_ONLY_FROM_CACHE);
ham_assert(st ? !page : 1, (0));
if (st)
return st;
if (page) {
node=ham_page_get_btree_node(page);
ham_assert(btree_node_is_leaf(node), (0));
ham_assert(btree_node_get_count(node) >= 3, (0)); /* edges + middle match */
idx = btree_node_search_by_key(db, page, key, hints.flags);
/*
* if we didn't hit a match OR a match at either edge, FAIL.
* A match at one of the edges is very risky, as this can also
* signal a match far away from the current node, so we need
* the full tree traversal then.
*/
if (idx <= 0 || idx >= btree_node_get_count(node) - 1) {
idx = -1;
}
/*
* else: we landed in the middle of the node, so we don't need to
* traverse the entire tree now.
*/
}
/* Reset any errors which may have been collected during the hinting
* phase -- this is done by setting 'idx = -1' above as that effectively
* clears the possible error code stored in there when (idx < -1)
*/
}
if (idx == -1) {
/* get the address of the root page */
if (!btree_get_rootpage(be)) {
stats_update_find_fail(db, &hints);
return HAM_KEY_NOT_FOUND;
}
/* load the root page */
st=db_fetch_page(&page, db, btree_get_rootpage(be), 0);
ham_assert(st ? !page : 1, (0));
if (!page) {
ham_assert(st, (0));
stats_update_find_fail(db, &hints);
return st ? st : HAM_INTERNAL_ERROR;
}
/* now traverse the root to the leaf nodes, till we find a leaf */
node=ham_page_get_btree_node(page);
if (!btree_node_is_leaf(node)) {
/* signal 'don't care' when we have multiple pages; we resolve
this once we've got a hit further down */
if (hints.flags & (HAM_FIND_LT_MATCH | HAM_FIND_GT_MATCH))
hints.flags |= (HAM_FIND_LT_MATCH | HAM_FIND_GT_MATCH);
for (;;) {
hints.cost++;
st=btree_traverse_tree(&page, 0, db, page, key);
if (!page) {
stats_update_find_fail(db, &hints);
return st ? st : HAM_KEY_NOT_FOUND;
}
node=ham_page_get_btree_node(page);
if (btree_node_is_leaf(node))
break;
}
}
/* check the leaf page for the key */
idx=btree_node_search_by_key(db, page, key, hints.flags);
if (idx < -1) {
stats_update_find_fail(db, &hints);
return (ham_status_t)idx;
}
} /* end of regular search */
/*
* When we are performing an approximate match, the worst case
* scenario is where we've picked the wrong side of the fence
* while sitting at a page/node boundary: that's what this
* next piece of code resolves:
*
* essentially it moves one record forwards or backward when
* the flags tell us this is mandatory and we're not yet in the proper
* position yet.
*
* The whole trick works, because the code above detects when
* we need to traverse a multi-page btree -- where this worst-case
* scenario can happen -- and adjusted the flags to accept
* both LT and GT approximate matches so that btree_node_search_by_key()
* will be hard pressed to return a 'key not found' signal (idx==-1),
* instead delivering the nearest LT or GT match; all we need to
* do now is ensure we've got the right one and if not,
* shift by one.
*/
if (idx >= 0) {
if ((ham_key_get_intflags(key) & KEY_IS_APPROXIMATE)
&& (hints.original_flags
& (HAM_FIND_LT_MATCH | HAM_FIND_GT_MATCH))
!= (HAM_FIND_LT_MATCH | HAM_FIND_GT_MATCH)) {
if ((ham_key_get_intflags(key) & KEY_IS_GT)
&& (hints.original_flags & HAM_FIND_LT_MATCH)) {
/*
* if the index-1 is still in the page, just decrement the
* index
*/
if (idx > 0) {
idx--;
}
else {
/*
* otherwise load the left sibling page
*/
if (!btree_node_get_left(node)) {
stats_update_find_fail(db, &hints);
ham_assert(node == ham_page_get_btree_node(page), (0));
stats_update_any_bound(db, page, key, hints.original_flags, -1);
return HAM_KEY_NOT_FOUND;
}
hints.cost++;
st = db_fetch_page(&page, db, btree_node_get_left(node), 0);
ham_assert(st ? !page : 1, (0));
if (!page) {
ham_assert(st, (0));
stats_update_find_fail(db, &hints);
return st ? st : HAM_INTERNAL_ERROR;
}
node = ham_page_get_btree_node(page);
idx = btree_node_get_count(node) - 1;
}
ham_key_set_intflags(key, (ham_key_get_intflags(key)
& ~KEY_IS_APPROXIMATE) | KEY_IS_LT);
}
else if ((ham_key_get_intflags(key) & KEY_IS_LT)
&& (hints.original_flags & HAM_FIND_GT_MATCH)) {
/*
* if the index+1 is still in the page, just increment the
* index
*/
if (idx + 1 < btree_node_get_count(node)) {
idx++;
}
else {
/*
* otherwise load the right sibling page
*/
if (!btree_node_get_right(node))
{
stats_update_find_fail(db, &hints);
ham_assert(node == ham_page_get_btree_node(page), (0));
stats_update_any_bound(db, page, key, hints.original_flags, -1);
return HAM_KEY_NOT_FOUND;
}
hints.cost++;
st = db_fetch_page(&page, db,
btree_node_get_right(node), 0);
if (!page) {
ham_assert(st, (0));
stats_update_find_fail(db, &hints);
return st ? st : HAM_INTERNAL_ERROR;
}
node = ham_page_get_btree_node(page);
idx = 0;
}
ham_key_set_intflags(key, (ham_key_get_intflags(key)
& ~KEY_IS_APPROXIMATE) | KEY_IS_GT);
}
}
else if (!(ham_key_get_intflags(key) & KEY_IS_APPROXIMATE)
&& !(hints.original_flags & HAM_FIND_EXACT_MATCH)
&& (hints.original_flags != 0)) {
/*
* 'true GT/LT' has been added @ 2009/07/18 to complete
* the EQ/LEQ/GEQ/LT/GT functionality;
*
* 'true LT/GT' is simply an extension upon the already existing
* LEQ/GEQ logic just above; all we do here is move one record
* up/down as it just happens that we get an exact ('equal')
* match here.
*
* The fact that the LT/GT constants share their bits with the
* LEQ/GEQ flags so that LEQ==(LT|EXACT) and GEQ==(GT|EXACT)
* ensures that we can restrict our work to a simple adjustment
* right here; everything else has already been taken of by the
* LEQ/GEQ logic in the section above when the key has been
* flagged with the KEY_IS_APPROXIMATE flag.
*/
if (hints.original_flags & HAM_FIND_LT_MATCH)
{
/*
* if the index-1 is still in the page, just decrement the
* index
*/
if (idx > 0)
{
idx--;
ham_key_set_intflags(key, (ham_key_get_intflags(key)
& ~KEY_IS_APPROXIMATE) | KEY_IS_LT);
}
else
{
/*
* otherwise load the left sibling page
*/
if (!btree_node_get_left(node))
{
/* when an error is otherwise unavoidable, see if
we have an escape route through GT? */
if (hints.original_flags & HAM_FIND_GT_MATCH)
{
/*
* if the index+1 is still in the page, just
* increment the index
*/
if (idx + 1 < btree_node_get_count(node))
{
idx++;
}
else
{
/*
* otherwise load the right sibling page
*/
if (!btree_node_get_right(node))
{
stats_update_find_fail(db, &hints);
ham_assert(node == ham_page_get_btree_node(page), (0));
stats_update_any_bound(db, page, key, hints.original_flags, -1);
return HAM_KEY_NOT_FOUND;
}
hints.cost++;
st = db_fetch_page(&page, db,
btree_node_get_right(node), 0);
if (!page)
{
ham_assert(st, (0));
stats_update_find_fail(db, &hints);
return st ? st : HAM_INTERNAL_ERROR;
}
node = ham_page_get_btree_node(page);
idx = 0;
}
ham_key_set_intflags(key, (ham_key_get_intflags(key) &
~KEY_IS_APPROXIMATE) | KEY_IS_GT);
}
else
{
stats_update_find_fail(db, &hints);
ham_assert(node == ham_page_get_btree_node(page), (0));
stats_update_any_bound(db, page, key, hints.original_flags, -1);
return HAM_KEY_NOT_FOUND;
}
}
else
{
hints.cost++;
st = db_fetch_page(&page, db,
btree_node_get_left(node), 0);
if (!page)
{
ham_assert(st, (0));
stats_update_find_fail(db, &hints);
return st ? st : HAM_INTERNAL_ERROR;
}
node = ham_page_get_btree_node(page);
idx = btree_node_get_count(node) - 1;
ham_key_set_intflags(key, (ham_key_get_intflags(key)
& ~KEY_IS_APPROXIMATE) | KEY_IS_LT);
}
}
}
else if (hints.original_flags & HAM_FIND_GT_MATCH)
{
/*
* if the index+1 is still in the page, just increment the
* index
*/
if (idx + 1 < btree_node_get_count(node))
{
idx++;
}
else
{
/*
* otherwise load the right sibling page
*/
if (!btree_node_get_right(node))
{
stats_update_find_fail(db, &hints);
ham_assert(node == ham_page_get_btree_node(page), (0));
stats_update_any_bound(db, page, key, hints.original_flags, -1);
return HAM_KEY_NOT_FOUND;
}
hints.cost++;
st = db_fetch_page(&page, db,
btree_node_get_right(node), 0);
if (!page)
{
ham_assert(st, (0));
stats_update_find_fail(db, &hints);
return st ? st : HAM_INTERNAL_ERROR;
}
node = ham_page_get_btree_node(page);
idx = 0;
}
ham_key_set_intflags(key, (ham_key_get_intflags(key)
& ~KEY_IS_APPROXIMATE) | KEY_IS_GT);
}
}
}
if (idx<0) {
stats_update_find_fail(db, &hints);
ham_assert(node, (0));
ham_assert(page, (0));
ham_assert(node == ham_page_get_btree_node(page), (0));
stats_update_any_bound(db, page, key, hints.original_flags, -1);
return HAM_KEY_NOT_FOUND;
}
/* load the entry, and store record ID and key flags */
entry=btree_node_get_key(db, node, idx);
/* set the cursor-position to this key */
if (cursor) {
ham_assert(!(bt_cursor_get_flags(cursor)&BT_CURSOR_FLAG_UNCOUPLED),
("coupling an uncoupled cursor, but need a nil-cursor"));
ham_assert(!(bt_cursor_get_flags(cursor)&BT_CURSOR_FLAG_COUPLED),
("coupling a coupled cursor, but need a nil-cursor"));
page_add_cursor(page, (ham_cursor_t *)cursor);
bt_cursor_set_flags(cursor,
bt_cursor_get_flags(cursor)|BT_CURSOR_FLAG_COUPLED);
bt_cursor_set_coupled_page(cursor, page);
bt_cursor_set_coupled_index(cursor, idx);
}
/*
* during util_read_key and util_read_record, new pages might be needed,
* and the page at which we're pointing could be moved out of memory;
* that would mean that the cursor would be uncoupled, and we're losing
* the 'entry'-pointer. therefore we 'lock' the page by incrementing
* the reference counter
*/
page_add_ref(page);
ham_assert(btree_node_is_leaf(node), ("iterator points to internal node"));
/* no need to load the key if we have an exact match: */
if (key && (ham_key_get_intflags(key) & KEY_IS_APPROXIMATE))
{
ham_status_t st=util_read_key(db, entry, key);
if (st)
{
page_release_ref(page);
stats_update_find_fail(db, &hints);
return (st);
}
}
if (record)
{
ham_status_t st;
record->_intflags=key_get_flags(entry);
record->_rid=key_get_ptr(entry);
st=util_read_record(db, record, flags);
if (st)
{
page_release_ref(page);
stats_update_find_fail(db, &hints);
return (st);
}
}
page_release_ref(page);
stats_update_find(db, page, &hints);
ham_assert(node == ham_page_get_btree_node(page), (0));
stats_update_any_bound(db, page, key, hints.original_flags, idx);
return (0);
}
static ham_status_t
__insert_cursor(ham_btree_t *be, ham_key_t *key, ham_record_t *record,
ham_bt_cursor_t *cursor, insert_hints_t *hints)
{
ham_status_t st;
ham_page_t *root;
ham_db_t *db=be_get_db(be);
ham_env_t *env = db_get_env(db);
insert_scratchpad_t scratchpad;
ham_assert(hints->force_append == HAM_FALSE, (0));
ham_assert(hints->force_prepend == HAM_FALSE, (0));
/*
* initialize the scratchpad
*/
memset(&scratchpad, 0, sizeof(scratchpad));
scratchpad.be=be;
scratchpad.record=record;
scratchpad.cursor=cursor;
/*
* get the root-page...
*/
ham_assert(btree_get_rootpage(be)!=0, ("btree has no root page"));
st=db_fetch_page(&root, db, btree_get_rootpage(be), 0);
ham_assert(st ? root == NULL : 1, (0));
if (st)
return st;
/*
* ... and start the recursion
*/
st=__insert_recursive(root, key, 0, &scratchpad, hints);
/*
* if the root page was split, we have to create a new
* root page.
*/
if (st==SPLIT) {
ham_page_t *newroot;
btree_node_t *node;
/*
* the root-page will be changed...
*/
st=ham_log_add_page_before(root);
if (st)
return (st);
/*
* allocate a new root page
*/
st=db_alloc_page(&newroot, db, PAGE_TYPE_B_ROOT, 0);
ham_assert(st ? newroot == NULL : 1, (0));
if (st)
return (st);
ham_assert(page_get_owner(newroot), (""));
/* clear the node header */
memset(page_get_payload(newroot), 0, sizeof(btree_node_t));
stats_page_is_nuked(db, root, HAM_TRUE);
/*
* insert the pivot element and the ptr_left
*/
node=ham_page_get_btree_node(newroot);
btree_node_set_ptr_left(node, btree_get_rootpage(be));
st=__insert_nosplit(newroot, &scratchpad.key,
scratchpad.rid, scratchpad.record, scratchpad.cursor,
hints);
ham_assert(!(scratchpad.key.flags & HAM_KEY_USER_ALLOC), (0));
scratchpad.cursor=0; /* don't overwrite cursor if __insert_nosplit
is called again */
if (st) {
ham_assert(!(scratchpad.key.flags & HAM_KEY_USER_ALLOC), (0));
if (scratchpad.key.data)
allocator_free(env_get_allocator(env), scratchpad.key.data);
return (st);
}
/*
* set the new root page
*
* !!
* do NOT delete the old root page - it's still in use!
*
* also don't forget to flush the backend - otherwise the header
* page of the database will not contain the updated information.
* The backend is flushed when the database is closed, but if
* recovery is enabled then the flush here is critical.
*/
btree_set_rootpage(be, page_get_self(newroot));
be_set_dirty(be, HAM_TRUE);
be->_fun_flush(be);
/*
* As we re-purpose a page, we will reset its pagecounter
* as well to signal its first use as the new type assigned
* here.
*/
if (env_get_cache(env) && (page_get_type(root)!=PAGE_TYPE_B_INDEX))
cache_update_page_access_counter(root, env_get_cache(env), 0);
page_set_type(root, PAGE_TYPE_B_INDEX);
page_set_dirty(root, env);
page_set_dirty(newroot, env);
/* the root page was modified (btree_set_rootpage) - make sure that
* it's logged */
if (env_get_rt_flags(env)&HAM_ENABLE_RECOVERY) {
st=txn_add_page(env_get_txn(env), env_get_header_page(env),
HAM_TRUE);
if (st)
return (st);
}
}
/*
* release the scratchpad-memory and return to caller
*/
ham_assert(!(scratchpad.key.flags & HAM_KEY_USER_ALLOC), (0));
if (scratchpad.key.data)
allocator_free(env_get_allocator(env), scratchpad.key.data);
return (st);
}
