btr_search_drop_page_hash_when_freed(): Check if buf_page_get_gen()
returns NULL. The page may have been evicted from the buffer pool
between buf_page_peek_if_search_hashed() and buf_page_get_gen(),
because the buffer pool mutex will be released between these two calls.
(Bug #42279)
rb://82 approved by Heikki Tuuri
Fixed a timing hole where a thread dropping an index can free the
in-memory index struct while another thread is still using
that structure to remove entries from adaptive hash index belonging
to one of the pages that belongs to the index being dropped.
The fix is to have a reference counter in the index struct and to
wait for this counter to drop to zero beforing freeing the struct.
Reviewed by: Heikki
There always was a one-to-one mapping between dict_tree_t and dict_index_t.
This saves 6 machine words per B-tree index in the data dictionary
cache plus the memory allocation overhead. We save one mem_heap_t
object per index (15 machine words). Considering the internal
fragmentation of the buddy allocator in mem_area_alloc(), this should
save 32 machine words per index (128 bytes on 32-bit systems and 256
bytes on 64-bit systems). (Bug #20877)
struct dict_tree_struct, dict_tree_t: Remove.
struct dict_index_struct: Add page and lock.
dict_tree_create(): Remove. Replace the invocation with
assignment to index->page and a call to rw_lock_create(&index->lock).
dict_tree_free(): Remove. Replace the invocation wtih
a call to rw_lock_free(&index->lock).
dict_index_get_tree(): Remove.
dict_tree_get_space_reserve(): Rename to dict_index_get_space_reserve()
and remove the parameter, which was unused.
btr_level_list_remove(): Remove the unused parameter "tree".
Replace the occurrences of "tree" with "index" in names of variables,
functions and data types, e.g. "dict_tree_t tree" becomes
"dict_index_t index". Remove local variables "tree" or "index" of
functions that needed both "tree" and "index".
structure is reserved for every index in the data dictionary cache.
(Bug #20877)
We could shrink the structure further to three 32-bit words or two
64-bit words by turning the remaining fields to bit-fields.
Unfortunately, the fields are not protected by any mutex, and thus we
would better keep each field aligned to a machine word.
btr_search_t, buf_block_t: Rename "ulint side" to "ibool left_side".
Remove BTR_SEARCH_LEFT_SIDE and BTR_SEARCH_RIGHT_SIDE, and also rename
some local variables and function parameters from "ulint side" to
"ibool left_side".
btr_search_t: Remove the unused fields last_search, n_direction, direction,
and modify_clock. Remove the unused constants BTR_SEA_NO_DIRECTION,
BTR_SEA_LEFT, BTR_SEA_RIGHT, and BTR_SEA_SAME_REC.
btr_search_t: Remove magic_n unless #defined UNIV_DEBUG. Turn an assertion
about the magic number into a debug assertion.
dtype_t: Remove unused field "prec", which was supposed to be used for
the precision of decimal columns in stand-alone InnoDB.
dtype_get_prec(): Remove.
dtype_set(), dict_mem_table_add_col(): Remove parameter "prec".
dtype_t: Turn all fields (mtype, prtype, len, mbminlen, mbmaxlen, len)
into bit-fields.
dict_table_t, dict_index_t, dict_tree_t: Omit magic_n from non-debug builds.
dict_col_t: Turn ind, clust_pos, and ord_part into bit-fields.
Replace the default clust_pos value ULINT_UNDEFINED with
REC_MAX_N_FIELDS and replace all references to clust_pos with calls to
the accessor function dict_col_get_clust_pos().
dict_field_t: Turn prefix_len and fixed_len into bit-fields.
dict_tree_t: Remove pad[64].
dict_table_t: Turn the fields ibd_file_missing, tablespace_discarded,
cached, flags, stat_initialized, and autoinc_inited into bit-fields.
Remove does_not_fit_in_memory from non-debug builds.
dict_index_t: Turn the fields trx_id_offset, n_user_defined_cols,
n_uniq, n_def, n_fields, n_nullable, and cached into bit-fields.
dict_foreign_struct: Turn n_fields and type into bit-fields.
rw_lock_t: Turn cline, last_s_line, and last_x_line into bit-fields.
Omit level unless #defined UNIV_SYNC_DEBUG.
Move REC_MAX_N_FIELDS (and REC_MAX_HEAP_NO and REC_MAX_N_OWNED)
from rem0rec.c to rem0types.h, as they are needed in dict0dict.ic.
dict_col_get_clust_pos(): Map REC_MAX_N_FIELDS to ULINT_UNDEFINED.
and lexer files). From now on, the following Emacs cc-mode settings apply
when indenting C function bodies in InnoDB:
(setq c-basic-offset 8)
(setq c-label-minimum-indentation 0)
(add-to-list 'c-offsets-alist '(c . 0))
(add-to-list 'c-offsets-alist '(label . [0]))
The indentation rules for function declarations still have not been
formalized, and they must be formatted manually.
Try to limit all lines to at most 79 characters (assuming TAB stops every
8 characters) by splitting lines before opening parenthesis, or at
string constants.
Fix some grammar mistakes in diagnostic output:
match to, match with -> match
found from -> found in
trying rename -> trying to rename
Fix an error in page_check_dir(): it said "supremum not pointed to"
when the infimum was not pointed to.
Enclose commented-out code snippets in #if 0 ... #endif instead of /* ... */.
Add (void*) casts to some %p parameters in fprintf() calls. Try to
split lines before a binary operator, not after one. (These three fixes
were not made everywhere.)
adaptive hash index for a B-tree page. (Bug #16582)
btr_search_drop_page_hash_index(): Retry the operation if a hash index
with different parameters was built meanwhile. Add diagnostics for the
case that hash node pointers to the page remain. This fix is from Heikki.
btr_search_info_update_hash(), btr_search_info_update_slow():
Document the parameter "info" as in/out.
Note that buf_block_t::index should be protected by btr_search_latch
or an s-latch or x-latch on the index page.
btr_search_drop_page_hash_index(): Read block->index while holding
btr_search_latch and use the cached value in the loop. Remove some
redundant assertions.
