This was an attempt to address problems with the Bug#12612184 fix.
Even with this follow-up fix, crash recovery can be broken.
Let us fix the bug later.
The fix of Bug#12612184 broke crash recovery. When a record that
contains off-page columns (BLOBs) is updated, we must first write redo
log about the BLOB page writes, and only after that write the redo log
about the B-tree changes. The buggy fix would log the B-tree changes
first, meaning that after recovery, we could end up having a record
that contains a null BLOB pointer.
Because we will be redo logging the writes off the off-page columns
before the B-tree changes, we must make sure that the pages chosen for
the off-page columns are free both before and after the B-tree
changes. In this way, the worst thing that can happen in crash
recovery is that the BLOBs are written to free pages, but the B-tree
changes are not applied. The BLOB pages would correctly remain free in
this case. To achieve this, we must allocate the BLOB pages in the
mini-transaction of the B-tree operation. A further quirk is that BLOB
pages are allocated from the same file segment as leaf pages. Because
of this, we must temporarily "hide" any leaf pages that were freed
during the B-tree operation by "fake allocating" them prior to writing
the BLOBs, and freeing them again before the mtr_commit() of the
B-tree operation, in btr_mark_freed_leaves().
btr_cur_mtr_commit_and_start(): Remove this faulty function that was
introduced in the Bug#12612184 fix. The problem that this function was
trying to address was that when we did mtr_commit() the BLOB writes
before the mtr_commit() of the update, the new BLOB pages could have
overwritten clustered index B-tree leaf pages that were freed during
the update. If recovery applied the redo log of the BLOB writes but
did not see the log of the record update, the index tree would be
corrupted. The correct solution is to make the freed clustered index
pages unavailable to the BLOB allocation. This function is also a
likely culprit of InnoDB hangs that were observed when testing the
Bug#12612184 fix.
btr_mark_freed_leaves(): Mark all freed clustered index leaf pages of
a mini-transaction allocated (nonfree=TRUE) before storing the BLOBs,
or freed (nonfree=FALSE) before committing the mini-transaction.
btr_freed_leaves_validate(): A debug function for checking that all
clustered index leaf pages that have been marked free in the
mini-transaction are consistent (have not been zeroed out).
btr_page_alloc_low(): Refactored from btr_page_alloc(). Return the
number of the allocated page, or FIL_NULL if out of space. Add the
parameter "mtr_t* init_mtr" for specifying the mini-transaction where
the page should be initialized, or if this is a "fake allocation"
(init_mtr=NULL) by btr_mark_freed_leaves(nonfree=TRUE).
btr_page_alloc(): Add the parameter init_mtr, allowing the page to be
initialized and X-latched in a different mini-transaction than the one
that is used for the allocation. Invoke btr_page_alloc_low(). If a
clustered index leaf page was previously freed in mtr, remove it from
the memo of previously freed pages.
btr_page_free(): Assert that the page is a B-tree page and it has been
X-latched by the mini-transaction. If the freed page was a leaf page
of a clustered index, link it by a MTR_MEMO_FREE_CLUST_LEAF marker to
the mini-transaction.
btr_store_big_rec_extern_fields_func(): Add the parameter alloc_mtr,
which is NULL (old behaviour in inserts) and the same as local_mtr in
updates. If alloc_mtr!=NULL, the BLOB pages will be allocated from it
instead of the mini-transaction that is used for writing the BLOBs.
fsp_alloc_from_free_frag(): Refactored from
fsp_alloc_free_page(). Allocate the specified page from a partially
free extent.
fseg_alloc_free_page_low(), fseg_alloc_free_page_general(): Add the
parameter "mtr_t* init_mtr" for specifying the mini-transaction where
the page should be initialized, or NULL if this is a "fake allocation"
that prevents the reuse of a previously freed B-tree page for BLOB
storage. If init_mtr==NULL, try harder to reallocate the specified page
and assert that it succeeded.
fsp_alloc_free_page(): Add the parameter "mtr_t* init_mtr" for
specifying the mini-transaction where the page should be initialized.
Do not allow init_mtr == NULL, because this function is never to be
used for "fake allocations".
mtr_t: Add the operation MTR_MEMO_FREE_CLUST_LEAF and the flag
mtr->freed_clust_leaf for quickly determining if any
MTR_MEMO_FREE_CLUST_LEAF operations have been posted.
row_ins_index_entry_low(): When columns are being made off-page in
insert-by-update, invoke btr_mark_freed_leaves(nonfree=TRUE) and pass
the mini-transaction as the alloc_mtr to
btr_store_big_rec_extern_fields(). Finally, invoke
btr_mark_freed_leaves(nonfree=FALSE) to avoid leaking pages.
row_build(): Correct a comment, and add a debug assertion that a
record that contains NULL BLOB pointers must be a fresh insert.
row_upd_clust_rec(): When columns are being moved off-page, invoke
btr_mark_freed_leaves(nonfree=TRUE) and pass the mini-transaction as
the alloc_mtr to btr_store_big_rec_extern_fields(). Finally, invoke
btr_mark_freed_leaves(nonfree=FALSE) to avoid leaking pages.
buf_reset_check_index_page_at_flush(): Remove. The function
fsp_init_file_page_low() already sets
bpage->check_index_page_at_flush=FALSE.
There is a known issue in tablespace extension. If the request to
allocate a BLOB page leads to the tablespace being extended, crash
recovery could see BLOB writes to pages that are off the tablespace
file bounds. This should trigger an assertion failure in fil_io() at
crash recovery. The safe thing would be to write redo log about the
tablespace extension to the mini-transaction of the BLOB write, not to
the mini-transaction of the record update. However, there is no redo
log record for file extension in the current redo log format.
rb:693 approved by Sunny Bains
not address the printouts issue
Detailed revision comments:
r6310 | marko | 2009-12-15 15:23:54 +0200 (Tue, 15 Dec 2009) | 30 lines
branches/5.1: Merge r4922 from branches/zip.
This the fix for the first part of Bug #41609 from InnoDB Plugin to
the built-in InnoDB in MySQL 5.1. This allows InnoDB Hot Backup to
back up a database while the built-in InnoDB in MySQL 5.1 is creating
temporary tables. (This fix does not address the printouts about
missing .ibd files for temporary tables at InnoDB startup, which was
committed to branches/zip in r6252.)
rb://219 approved by Sunny Bains.
branches/zip: Distinguish temporary tables in MLOG_FILE_CREATE.
This addresses Mantis Issue #23 in InnoDB Hot Backup and some
of MySQL Bug #41609.
In MLOG_FILE_CREATE, we need to distinguish temporary tables, so that
InnoDB Hot Backup can work correctly. It turns out that we can do this
easily, by using a bit of the previously unused parameter for page number.
(The page number parameter of MLOG_FILE_CREATE has been written as 0
ever since MySQL 4.1, which introduced MLOG_FILE_CREATE.)
MLOG_FILE_FLAG_TEMP: A flag for indicating a temporary table in
the page number parameter of MLOG_FILE_ operations.
fil_op_write_log(): Add the parameter log_flags.
fil_op_log_parse_or_replay(): Add the parameter log_flags.
Do not replay MLOG_FILE_CREATE when MLOG_FILE_FLAG_TEMP is set in log_flags.
This only affects ibbackup --apply-log. InnoDB itself never replays file
operations.
Bugs fixed:
- Bug #20791 valgrind errors in InnoDB
Remove Valgrind warning of Bug #20791 : in new database
creation, we read the doublewrite buffer magic number from
uninitialized memory; the code worked because it was extremely
unlikely that the memory would contain the magic number
- Bug #21784 DROP TABLE crashes 5.1.12-pre if concurrent
queries on the table
remove update_thd() in ::store_lock()
Also includes numerous coding style fixes, etc. See file-level
comments for details.
Fixed BUGS:
#3300: "UPDATE statement with no index column in where condition locks
all rows"
Implement semi-consistent read to reduce lock conflicts at the cost
of breaking serializability.
ha_innobase::unlock_row(): reset the "did semi consistent read" flag
ha_innobase::was_semi_consistent_read(),
ha_innobase::try_semi_consistent_read(): new methods
row_prebuilt_t, row_create_prebuilt(): add field row_read_type for
keeping track of semi-consistent reads
row_vers_build_for_semi_consistent_read(),
row_sel_build_committed_vers_for_mysql(): new functions
row_search_for_mysql(): implement semi-consistent reads
#9802: "Foreign key checks disallow alter table".
Added test cases.
#12456: "Cursor shows incorrect data - DML does not affect,
probably caching"
This patch implements a high-granularity read view to be used with
cursors. In this high-granularity consistent read view modifications
done by the creating transaction after the cursor is created or
future transactions are not visible. But those modifications that
transaction did before the cursor was created are visible.
#12701: "Support >4GB buffer pool and log files on 64-bit Windows"
Do not call os_file_create_tmpfile() at runtime. Instead, create all
tempfiles at startup and guard access to them with mutexes.
#13778: "If FOREIGN_KEY_CHECKS=0, one can create inconsistent FOREIGN KEYs".
When FOREIGN_KEY_CHECKS=0 we still need to check that datatypes between
foreign key references are compatible.
#14189: "VARBINARY and BINARY variables: trailing space ignored with InnoDB"
innobase_init(): Assert that
DATA_MYSQL_BINARY_CHARSET_COLL == my_charset_bin.number.
dtype_get_pad_char(): Do not pad VARBINARY or BINARY columns.
row_ins_cascade_calc_update_vec(): Refuse ON UPDATE CASCADE when trying
to change the length of a VARBINARY column that refers to or is referenced
by a BINARY column. BINARY columns are no longer padded on comparison,
and thus they cannot be padded on storage either.
#14747: "Race condition can cause btr_search_drop_page_hash_index() to crash"
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.
#15108: "mysqld crashes when innodb_log_file_size is set > 4G"
#15308: "Problem of Order with Enum Column in Primary Key"
#15550: "mysqld crashes in printing a FOREIGN KEY error in InnoDB"
row_ins_foreign_report_add_err(): When printing the parent record,
use the index in the parent table rather than the index in the child table.
#15653: "Slow inserts to InnoDB if many thousands of .ibd files"
Keep track on unflushed modifications to file spaces. When there are tens
of thousands of file spaces, flushing all files in fil_flush_file_spaces()
would be very slow.
fil_flush_file_spaces(): Only flush unflushed file spaces.
fil_space_t, fil_system_t: Add a list of unflushed spaces.
#15991: "innodb-file-per-table + symlink database + rename = cr"
os_file_handle_error(): Map the error codes EXDEV, ENOTDIR, and EISDIR
to the new code OS_FILE_PATH_ERROR. Treat this code as OS_FILE_PATH_ERROR.
This fixes the crash on RENAME TABLE when the .ibd file is a symbolic link
to a different file system.
#16157: "InnoDB crashes when main location settings are empty"
This patch is from Heikki.
#16298: "InnoDB segfaults in INSERTs in upgrade of 4.0 -> 5.0 tables
with VARCHAR BINARY"
dict_load_columns(): Set the charset-collation code
DATA_MYSQL_BINARY_CHARSET_COLL for those binary string columns
that lack a charset-collation code, i.e., the tables were created
with an older version of MySQL/InnoDB than 4.1.2.
#16229: "MySQL/InnoDB uses full explicit table locks in trigger processing"
Take a InnoDB table lock only if user has explicitly requested a table
lock. Added some additional comments to store_lock() and external_lock().
#16387: "InnoDB crash when dropping a foreign key <table>_ibfk_0"
Do not mistake TABLENAME_ibfk_0 for auto-generated id.
dict_table_get_highest_foreign_id(): Ignore foreign constraint
identifiers starting with the pattern TABLENAME_ibfk_0.
#16582: "InnoDB: Error in an adaptive hash index pointer to page"
Account for a race condition when dropping the adaptive hash index
for a B-tree page.
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.
btr_search_info_update_hash(), btr_search_info_update_slow():
Document the parameter "info" as in/out.
#16814: "SHOW INNODB STATUS format error in LATEST FOREIGN KEY ERROR
section"
Add a missing newline to the LAST FOREIGN KEY ERROR section in SHOW
INNODB STATUS output.
dict_foreign_error_report(): Always print a newline after invoking
dict_print_info_on_foreign_key_in_create_format().
#16827: "Better InnoDB error message if ibdata files omitted from my.cnf"
#17126: "CHECK TABLE on InnoDB causes a short hang during check of adaptive
hash"
CHECK TABLE blocking other queries, by releasing the btr_search_latch
periodically during the adaptive hash table validation.
#17405: "Valgrind: conditional jump or move depends on unititialised values"
buf_block_init(): Reset magic_n, buf_fix_count and io_fix to avoid
testing uninitialized variables.
