Commit graph

702 commits

Author SHA1 Message Date
marko
f4c776f518 branches/zip: Initial steps towards disk-based storage of compressed pages.
dict_mem_table_create(): Account for DICT_TF_COMPRESSED in a debug assertion.

btr_store_big_rec_extern_fields(), btr_free_externally_stored_field(),
btr_copy_externally_stored_field(): Implement the disk format for
compressed BLOB pages.

btr_copy_externally_stored_field(): Improve error reporting and handling
when decompressing BLOB pages.

buf_flush_init_for_writing(), buf_page_is_corrupted(), buf_page_print():
Account for compressed BLOB pages (FIL_PAGE_TYPE_ZBLOB).

buf_calc_zblob_page_checksum(): New function.
2006-04-05 13:41:12 +00:00
marko
ac5b886f4f branches/zip: Fix some BLOB handling bugs.
btr_store_big_rec_extern_fields(): Assert that page_zip is non-NULL
if and only if dict_table_is_zip() holds.

btr_free_externally_stored_field(): Observe dict_table_is_zip().
Allow page_zip==NULL even if dict_table_is_zip().  Remove the
related TODO comment in row_purge_upd_exist_or_extern().

page_zip_available(): uncompressed_size already includes
PAGE_ZIP_DIR_SLOT_SIZE.

page_zip_decompress(): Remove bogus assertion d_stream.next_out == last.
Do not subtract BTR_EXTERN_FIELD_REF_SIZE from d_stream.avail_in when
decompressing records, because the records may be deleted later in
page_zip_apply_log(), and no BLOB pointers are allocated for deleted
records.
2006-04-03 20:33:31 +00:00
marko
e2b4592d6d branches/zip: Do not clear the records in page_delete_rec_list_end().
page_delete_rec_list_end(): Do not invoke page_zip_clear_rec().

page_zip_clear_rec(): Remove the global declaration.  Add an assertion
that the record must be on the free list.  Copy onl rec_offs_data_size()
bytes before attempting to clear the record and to compress the page.

page_zip_compress(), page_zip_write_rec(): Remove debug assertions that
trx_id, roll_ptr and node_ptr of deleted records be zero-filled.
2006-03-23 08:13:08 +00:00
marko
b03053a633 branches/zip: Minor fixes
Restore page_zip_clear_rec() to the global scope.
Invoke it in page_delete_rec_list_end().

Add debug assertions to page0zip.c to guard against overwriting data
on the compressed page.
2006-03-17 14:11:57 +00:00
marko
f3f76d249a branches/zip: Improve the clearing of deleted records. Try to support
operations on BLOB columns.  There are some bugs in the code, because
test-insert and a few other tests fail.

page_mem_free(): Add parameter index.  Decrement PAGE_N_RECS here.
Move some operations to page_zip_dir_delete().

page_zip_clear_rec(): Make this a static function.

page_zip_dir_delete(): Add parameters index and offsets.
Decrement PAGE_N_RECS and clear info_bits and n_owned.

page_zip_get_n_prev_extern(): Correct the synopsis and the algorithm.
Add parameter page_zip.  Search the records in heap_no order instead
of collation order.

page_zip_compress(), page_zip_decompress(): Only copy BLOB pointers
and increment n_blobs for records that have not been deleted.

page_zip_clear_rec(): Clear trx_id and roll_ptr on the compressed page.

page_zip_dir_delete(): Decrement PAGE_N_RECS.  Shift the array of
BLOB pointers.  Call page_zip_clear_rec().

page_zip_dir_add_slot(): Shift the array of BLOB pointers to make
space of roll_ptr and trx_id.

page_cur_delete_rec(): Do not decrement PAGE_N_RECS or call
page_zip_clear_rec(), as page_mem_free() already does it.
2006-03-16 14:02:22 +00:00
marko
df85f7287c branches/zip: Minor fixes
trx-undo_prev_version_build(): Pass offsets==NULL to
rec_set_field_extern_bits().

rec_set_field_extern_bits(), rec_set_field_extern_bits_new():
Accept offsets==NULL.

row_upd_rec_in_place(): Remove the bogus comment that the function
would only be invoked on a clustered index.  Remove the related
debug assertion.
2006-03-16 09:44:52 +00:00
marko
2e08171ada branches/zip: Bug fixes for BLOB handling. At least one bug remains:
page_zip_dir_delete() will need to handle BLOBs.

rec_set_field_extern_bits(), rec_set_field_extern_bits_new():
Add parameter offsets.

rec_offs_set_nth_extern(): New function to set an extern bit in offsets.
This will be called when an extern bit is set in a record.

page_cur_rec_insert(), page_cur_insert_rec_low(): Document that the
parameter "offsets" is in/out.

page_zip_dir_delete(): Note that the array of BLOB pointers will need
to be shifted.

page0zip.ic: Document the entry type for clearing a record.

page_zip_available(): Add parameter "index".  Remove parameters
"is_leaf" and "is_clustered".

page_zip_get_trailer_len(): New function for computing the trailer length
of the compressed page.

page_zip_apply_log(): Implement the modification log entry type for
clearing the data bytes of a record.

page_zip_decompress(): Initialize n_blobs when actually copying the
BLOB pointers to place.

page_zip_validate(): Add diagnostic messages for failures.  Check
also m_start, m_end, and n_blobs.

page_zip_write_blob_ptr(): Add page_zip_validate() assertion.
2006-03-14 14:38:45 +00:00
marko
77aba5109d branches/zip: Implement in-place updates of BLOB pointers.
There are still some bugs in the code.

btr_store_big_rec_extern_fields(): Remove assertion on dict_table_is_zip()
to ease testing.

btr_free_externally_stored_field(): Test page_zip instead of
dict_table_is_zip().

page_zip_write_rec(): Add parameter "create".  Try to handle externally
stored columns.

rec_offs_any_extern(): Correct the function comment.

Add rec_offs_n_extern() and page_zip_get_n_prev_extern().

page_zip_dir_decode(): Replace assertion with if (...) return(FALSE).

page_zip_decompress(): Do not clear page_zip->n_blobs after counting the
BLOBs.

page_zip_write_blob_ptr(): Use page_zip_get_n_prev_extern().
Correct an off-by-one error in memcpy().
2006-03-13 15:02:16 +00:00
marko
9f51e0227f branches/zip: Remove unused parameter "index" of page_mem_free(). 2006-03-10 12:29:22 +00:00
marko
c6fd670dfe branches/zip: Replace buf_block_align(btr_cur_get_page()) with
buf_block_align(btr_cur_get_rec()).
2006-03-10 12:10:47 +00:00
marko
98eda4e949 branches/zip: Make test-insert --small-test pass.
btr_compress(): Pass the correct page_zip to btr_node_ptr_set_child_page_no().

page_mem_free(): Remove parameter mtr.  Do not call page_zip_clear_rec() here.

page_cur_delete_rec(): Call page_zip_clear_rec() here.  Decrement PAGE_N_RECS
only after calling page_mem_free().  Assert page_zip_validate().
2006-03-10 10:30:56 +00:00
marko
ee152b828f branches/zip: Note that TRX_ID and ROLL_PTR only exist on B-tree leaf pages
of clustered indexes.  Previously, parts of the code assumed that these
columns would exist on all leaf pages.  Simplify the update-in-place of
these columns.

Add inline function dict_index_is_clust() to replace all tests
index->type & DICT_CLUSTERED.

Remove the redo log entry types MLOG_ZIP_WRITE_TRX_ID and
MLOG_ZIP_WRITE_ROLL_PTR, because the modifications to these columns
are covered by logical logging.

Fuse page_zip_write_trx_id() and page_zip_write_roll_ptr() into
page_zip_write_trx_id_and_roll_ptr().

page_zip_dir_add_slot(), page_zip_available(): Add flag "is_clustered",
so that no space will be reserved for TRX_ID and ROLL_PTR on leaf pages
of secondary indexes.

page_zip_apply_log(): Flag an error when val==0 is encoded with two bytes.

page_zip_write_rec(): Add debug assertions that there is enough space
available for the entry before copying the data bytes of the record.
2006-03-09 17:26:02 +00:00
marko
16f3d27113 branches/zip: Fix various bugs.
btr_root_raise_and_insert(): Because btr_page_set_level() must not
change level from non-zero to zero on compressed pages, invoke
btr_page_set_level() with page_zip==NULL and compress the entire
root page after creating it from the scratch.

btr_attach_half_pages(): Simplify the computation of lower_page_zip
and upper_page_zip.  Invoke btr_node_ptr_set_child_page_no() with
the correct page_zip.

page0page.h: Add __attribute__((const)) to functions testing for
infimum or supremum.

page_zip_dir_delete(): Note that the third parameter may be NULL.
Correct offset errors.

page_zip_available(): Use n_heap instead of n_recs.

page_zip_dir_find(), page_zip_dir_find_free(): Fix off-by-one error.

page_zip_fields_encode(), page_zip_fields_decode(): Encode and decode
index->n_nullable for non-leaf pages.

page_zip_apply_log(): Write REC_NEW_HEAP_NO before calling
rec_offs_make_valid().

page_zip_write_node_ptr(): Correct off-by-one error.

page_cur_search_with_match(): Make use of page_is_leaf().

page_dir_add_slots(): Replaced with page_dir_add_slot().  Use memmove().
2006-03-08 22:19:20 +00:00
marko
63e9515826 branches/zip: Fix some bugs in the insertion of records.
row_upd_rec_in_place(), page_zip_write_rec(): Add parameter "index".

page_dir_set_n_heap(): Add a debug assertion that on compressed
pages, n_heap will always be incremented by one.  Improve code formatting.

page_zip_dir_add_slot(): New function, called from
page_cur_insert_rec_low() after page_mem_alloc_heap().

rec_set_n_owned_new(): Do not call page_zip_rec_set_owned()
on the supremum record.

rec_offs_make_valid(): Add debug assertions.

page_zip_dir_user_size(): Correct an off-by-one error in the debug assertion.

page_zip_apply_log(): Add parameter trx_id_col.  Skip trx_id and roll_ptr.

page_zip_decompress(): Simplify the handling of "storage" in the loop that
copies the uncompressed fields.

page_zip_write_rec(): Store trx_id and roll_ptr separately.

page_zip_write_trx_id(), page_zip_write_roll_ptr(): Fix off-by-one errors.

page_cur_insert_rec_low(): Call page_zip_dir_add_slot() after
page_mem_alloc_heap().  Remove some redundant assertions.
Pass page_zip to page_dir_split_slot().
2006-03-06 21:00:05 +00:00
marko
682d39515e branches/zip: Merge revisions 247:265 from trunk. 2006-03-03 14:43:05 +00:00
marko
9d38a29b09 branches/zip: Merge table->zip to table->flags. Add dict_table_is_zip(). 2006-03-03 08:39:20 +00:00
marko
9a601a1c8b branches/zip: Fix some bugs in incremental compression.
btr_create(): page_zip_compress() returns FALSE on failure.

page_zip_write_header(): Write to page_zip->data[] instead of page_zip[].

buf_flush_init_for_writing(): Add parameter page_zip and set the fields
also in the header of the compressed page.

btr_cur_search_to_nth_level(): Add ut_ad() on page_zip_validate().
2006-03-02 14:05:32 +00:00
marko
c86de5b837 branches/zip: Move some private functions from page0zip.ic and page0zip.h
to page0zip.c.
2006-03-01 11:38:25 +00:00
marko
c1f7cc85eb branches/zip: Remove remaining references to relocated records in
function prototypes and comments.  These changes were accidentally
forgotten from revision 253.
2006-03-01 11:29:18 +00:00
marko
fb00d066bb branches/zip: In page_zip_compress(), do not touch page_zip unless the
operation succeeds.

page_zip_dir_set(): Remove, as the slots will be written to a temporary
buffer.

page_zip_dir_encode(): Replace page_zip with buf.  Make some assertions
more accurate.  Write the dense page directory to buf.

page_zip_compress(): Allocate a temporary buffer for the compressed page
except the page header.  Make some assertions more accurate.

page_zip_compress(), page_zip_decompress(): Remove 2 bytes of empty space
adjacent to the area reserved for storing uncompressed columns.
2006-02-28 20:37:29 +00:00
marko
7c8b6bda4b branches/zip: Make page_delete_rec_list_start() work on compressed pages,
now that deletion of records on compressed pages is guaranteed to succeed.
2006-02-27 12:38:49 +00:00
marko
86ae15b86a branches/zip: Remove the possibility to relocate reallocated records
on a compressed page.

page_zip_get_n_relocated(), page_zip_set_n_relocated(),
page_zip_get_relocated(): Remove.

page0zip.c, page0zip.ic: Remove all references to n_relocated.

page_cur_insert_rec_low(): On pages that are compressed,
never relocate records that are allocated from the free list.
2006-02-27 10:26:59 +00:00
marko
82d38b63e3 branches/zip: Merge revisions 236:247 from trunk. 2006-02-27 09:33:26 +00:00
marko
f40121415d branches/zip: Merge revisions 200:236 from trunk. 2006-02-23 19:25:29 +00:00
marko
c952dc5b49 branches/zip: When allocating records from the free list,
do not allow extra_size to decrease on compressed pages.

Split page_mem_alloc() to page_mem_alloc_free() and page_mem_alloc_heap().

page_cur_insert_rec_low(): Remove parameter "tuple".  Implement some of the
logic from page_mem_alloc().

page_cur_tuple_insert(): Convert the tuple to a record and calculate offsets.

page_zip_validate(): Assert that the page is in compact format.
2006-02-23 14:45:12 +00:00
marko
4167b06c33 branches/zip: Minor cleanup.
Remove the declaration of the undefined function
page_zip_available_noninline().

Make page_zip_available() private to page0zip.c and page0zip.ic.
2006-02-22 13:49:05 +00:00
marko
c04ddfef05 branches/zip: Try to synchronize the updates of uncompressed and
compressed pages.

btr_root_raise_and_insert(): Distinguish root_page_zip and new_page_zip.

btr_cur_set_ownership_of_extern_field(): Do not log the write on the
uncompressed page if it will be logged for page_zip.

lock_rec_insert_check_and_lock(), lock_sec_rec_modify_check_and_lock():
Update the max_trx_id field also on the compressed page.

mlog_write_ulint(): Add UNIV_UNLIKELY hints.  Remove trailing white space.

mlog_log_string(): Remove trailing white space.

rec_set_field_extern_bits(): Remove parameter mtr, as the write will either
occur in the heap, or it will be logged at a higher level.

recv_parse_or_apply_log_rec_body(),
page_zip_write_header(): Add log record type MLOG_ZIP_WRITE_HEADER.

page_header_set_field(): Pass mtr=NULL to page_zip_write_header().

page_header_reset_last_insert(): Pass mtr to page_zip_write_header().

btr_page_set_index_id(), btr_page_set_level(),
btr_page_set_next(), btr_page_set_prev(): Pass mtr to page_zip_write_header().

row_upd_rec_sys_fields(): Pass mtr=NULL to page_zip_write_trx_id() and
page_zip_write_roll_ptr(), since the write will be logged at a higher level.

page_zip_write_header(): Add parameter mtr.
page_zip_write_header_log(): New function.

Remove rec_set_nth_field_extern_bit().
Make rec_set_nth_field_extern_bit_old() static.
Rename rec_set_nth_field_extern_bit_new()
to rec_set_field_extern_bits_new() and make it static.

row_ins_index_entry_low(): Remove bogus TODO comment.
2006-02-22 13:02:40 +00:00
marko
69f236c29f branches/zip:
Fix the way how btr_free_externally_stored_field() is called in purge.

btr_free_externally_stored_field(): Add parameter field_ref that points
directly to the BLOB reference.  Use rec, offsets, page_zip, and i
only for the page_zip_write_blob_ptr() call.

row_purge_upd_exist_or_extern(): Do not assume that the undo log contains
the entire record.  Only pass the BLOB reference to
btr_free_externally_stored_field().
2006-02-21 14:15:11 +00:00
marko
6ce8427cb8 branches/zip: Merge revisions 79:200 from trunk. 2006-02-17 14:19:39 +00:00
marko
ee5a92aaef branches/zip: Implement the compression of BLOB columns.
This has not been extensively tested yet, because some other part of the
code breaks in "ibtestblob".

btr_free_page_low(): Add parameters "space" and "page_no", because they
are omitted from compressed BLOB pages.

btr0cur.c: Implement the compression and decompression of BLOB columns,
enabled at compile-time (#define ZIP_BLOB TRUE) for now.

btr_rec_free_externally_stored_fields(),
btr_copy_externally_stored_field(): Made static

mlog_log_string(): New function, split from mlog_write_string(), allows
to avoid a dummy memcpy() of compressed BLOB pages.
2006-02-16 12:58:18 +00:00
marko
d8e3b8114c branches/zip: Remove in-place updates of the "external storage"
flag of records.  The flags may only be updated in heap-allocated
copies of records.

btr_root_raise_and_insert(),
btr_page_split_and_insert(),
btr_cur_insert_if_possible(),
btr_cur_optimistic_insert(),
btr_cur_pessimistic_insert(),
page_cur_tuple_insert(),
page_cur_insert_rec_low(): Add parameters "ext" and "n_ext".

dtuple_convert_big_rec(): Make parameter "ext" const.
2006-02-13 14:28:00 +00:00
marko
062c64e87e branches/zip: Prepare for in-place updates of B-tree node pointers,
BLOB pointers, trx_id, and roll_ptr.

btr_empty(), btr_create(), page_create(): Add parameter "index", as some
index information will be encoded on the compressed page.

Define REC_NODE_PTR_SIZE as 4.

Allow btr_page_reorganize() and btr_page_reorganize_low() to fail.

Define the error code DB_ZIP_OVERFLOW.

Make row_ins_index_entry_low() static.

page0zip: Encode the index, log reorganized records, and store uncompressed
fields separately from the compressed data stream.
2006-02-10 15:06:17 +00:00
marko
c0ac96d9b4 branches/zip: Prepare for upcoming implementation of page_zip_compress()
that will require complete index information.

dict_create_index_step(): invoke dict_index_add_to_cache() before btr_create()

dict_index_remove_from_cache(): make public

dict_index_get_if_in_cache_low(): new function, for holding dict_sys->mutex

buf_flush_init_for_writing(): remove the temporary hook to page_zip_compress()

page_create(): add temporary hook to page_zip_compress()
2006-02-03 14:44:29 +00:00
marko
77c570dc4d branches/zip: Minor improvements for debugging and testing purposes.
buf_flush_init_for_writing(): Improve the diagnostics and make the
condition for skipping pages accurate.

univ.i: Introduce UNIV_ZIP_DEBUG for enabling some page_zip_validate() tests.

page0zip.h, page0zip.c: Define and use page_zip_validate() in
page_zip_compress() and page_zip_write() if UNIV_ZIP_DEBUG or UNIV_DEBUG
is defined.
2005-11-30 13:57:54 +00:00
marko
e12d8a0f10 branches/zip: Merge revisions 9:72 from trunk. 2005-11-29 12:30:46 +00:00
marko
4fc27c2723 branches/zip: Minor fixes. Now one non-trivial page compresses and
decompresses properly.

page_mem_free(): Adjust the temporary condition for zeroing data.

page_zip_compress(): Do not accept Z_STREAM_END from non-final deflate().
Tighten the assertion on c_stream.avail_in.

page_zip_dir_sort(): Note that the high bound is not inclusive.

page_zip_dir_decode(): Remove off-by-one error in page_zip_dir_sort() call.

page_zip_set_extra_bytes(): Add the termination condition that was forgotten
from the second for loop.  Simplify the code a little.

page_zip_decompress(): Do not accept Z_STREAM_END from non-final inflate().
Tighten the assertion on d_stream.avail_out.  Increment heap_status correctly.
2005-11-25 14:17:40 +00:00
marko
49ef38ef28 branches/zip: Minor fixes to get a page with one record compress/decompress.
A page with multiple records or deleted records still does not compress
or decompress properly.

buf_flush_init_for_writing(): Initialize block->page_zip properly so that all
assertions in page0zip can be enabled.

page_zip_decompress(): Note that corrupt data should not lead to assertions.

page_zip_dir_set(): Correct the interface.  Fix off-by-one error.
page_zip_dir_get(): Fix off-by-one error.

page0zip.c: Replace n_heap with n_dense and add comments about
the infimum and supremum records whenever we subtract 2 from heap_no.
Fix some programming errors.
2005-11-25 12:34:38 +00:00
marko
5ef5378f2e branches/zip: Minor fixes. Compression/decompression still fails.
buf0flu.c: Allocate the temporary buffer from buf_frame_alloc().

page_zip_simple_validate(): Do not assert page_zip->m_start >= PAGE_DATA.

page_zip_compress(): Replace some assertions with page_zip_simple_validate(),
and do not assert anything about page_zip->data contents.

page_zip_validate(): Do not compare the page trailer bytes.

page_zip_write(): Assert that the entire page headers match and
that page_zip->m_start >= PAGE_DATA.
2005-11-25 07:51:28 +00:00
marko
b348fcafa5 branches/zip: Implement a more compact page format.
Add a hook to buf0flu.c for testing compression and decompression.
TODO: adapt page_zip_write() calls.
2005-11-24 14:13:10 +00:00
marko
42875604eb branches/zip: Remember to update page_zip when updating "extern" flags.
Add function parameters and TODO comments regarding that.
2005-11-18 07:40:34 +00:00
marko
8f9d2a767c branches/zip: Minor improvements.
page_dir_slot_set_n_owned(): Check page_zip_available().
page_dir_add_slots(): Add parameter page_zip.
btr_attach_half_pages(): Add parameter page_zip.
btr_node_ptr_set_child_page_no(): Add parameter page_zip.
Make comments about page_zip modification log usage more accurate.
2005-11-14 15:46:12 +00:00
marko
7a696ebf52 branches/zip: Minor improvements.
page_dir_set_n_slots(): Add debug assertion that there will be enough space
available in the modification log of the compressed page.
page_trailer_get_len(): Simplify the formula.  Make more use of the function.
2005-11-14 10:40:02 +00:00
marko
a6fe621ca1 branches/zip: Minor improvements.
page_set_max_trx_id(), page_update_max_trx_id(): Add parameter page_zip.

Check that page_zip_decompress() is never called after modifying the
header or trailer of the compressed page, i.e., that page_zip_decompress()
will restore the uncompressed page as it was before the failed operation.
2005-11-11 21:18:42 +00:00
marko
d4231b11c5 branches/zip: Restore left b-tree splits.
page_move_rec_list_start(): Restored.  Reorganize old page if compressed.
btr_page_split_and_insert(): Do left page splits.
2005-11-11 13:59:40 +00:00
marko
6ff00e1dbb branches/zip: Bug fixes and improvements.
page_header_reset_last_insert(): Add page_zip parameter.
rec_get_next_ptr(): Return a pointer to the same page, or NULL.
page_dir_delete_slot(): Fix off-by-one errors.
2005-11-09 15:43:36 +00:00
marko
1fe2afd36f branches/zip:
Fix bugs found during testing.
2005-11-04 12:49:39 +00:00
marko
edff1526a6 branches/zip:
Add page_zip parameter to page_dir_slot_set_rec().
Improve the documentation of the page_zip parameters in page0page.
Add function page_trailer_get_len().
2005-11-04 08:50:25 +00:00
marko
3ba94cbe8e branches/zip:
Implement page_zip_write_header().
Add page_zip parameter to btr_page_set_ functions.
Allow page_copy_rec_list_end() to be called with invalid new_page_zip
if the page has been newly created.
2005-10-28 10:45:20 +00:00
marko
88e6d6863d branches/zip:
Initial import of the zip-like compression of B-tree pages.
2005-10-27 11:48:10 +00:00
osku
9bb1601227 include/Makefile.am: Remove reference to removed makefilewin.i. 2005-10-27 09:37:17 +00:00