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
for compressed InnoDB tables
ha_innodb::info_low(): For calculating data_length or index_length,
use the compressed page size for compressed tables instead of UNIV_PAGE_SIZE.
rb:714 approved by Sunny Bains
The server crashes if it processes table map events that are
corrupted, especially if they map different tables to the same
identifier. This could happen, for instance, due to BUG 56226.
We fix this by checking whether the table map has already been
mapped before actually applying the event. If it has been mapped
with different settings an error is raised and the slave SQL
thread stops. If it has been mapped with same settings the event
is skipped. If the table is set to be ignored by the filtering
rules, there is no change in behavior: the event is skipped and
ids are not checked.
bug. It added this assert;
ut_ad(ind_field->prefix_len);
before a section of code that assumes there is a prefix_len.
The patch replaced code that explicitly avoided this with a check for
prefix_len. It turns out that the purge thread can get to that assert
without a prefix_len because it does not use a row_ext_t* .
When UNIV_DEBUG is not defined, the affect of this is that the purge thread
sets the dfield->len to zero and then cannot find the entry in the index to
purge. So secondary index entries remain unpurged.
This patch does not do the assert. Instead, it uses
'if (ind_field->prefix_len) {...}'
around the section of code that assumes a prefix_len. This is the way the
patch I provided to Marko did it.
The test case is simply modified to do a sleep(10) in order to give the
purge thread a chance to run. Without the code change to row0row.c, this
modified testcase will assert if InnoDB was compiled with UNIV_DEBUG.
I tried to sleep(5), but it did not always assert.
row_build_index_entry(): In innodb_file_format=Barracuda
(ROW_FORMAT=DYNAMIC or ROW_FORMAT=COMPRESSED), a secondary index on a
full column can refer to a field that is stored off-page in the
clustered index record. Take that into account.
rb:692 approved by Jimmy Yang
HA_INNOBASE::UPDATE_ROW, TEMPORARY TABLE, TABLE LOCK".
Attempt to update an InnoDB temporary table under LOCK TABLES
led to assertion failure in both debug and production builds
if this temporary table was explicitly locked for READ. The
same scenario works fine for MyISAM temporary tables.
The assertion failure was caused by discrepancy between lock
that was requested on the rows of temporary table at LOCK TABLES
time and by update operation. Since SQL-layer requested a
read-lock at LOCK TABLES time InnoDB engine assumed that upcoming
statements which are going to be executed under LOCK TABLES will
only read table and therefore should acquire only S-lock.
An update operation broken this assumption by requesting X-lock.
Possible approaches to fixing this problem are:
1) Skip locking of temporary tables as locking doesn't make any
sense for connection-local objects.
2) Prohibit changing of temporary table locked by LOCK TABLES ...
READ.
Unfortunately both of these approaches have drawbacks which make
them unviable for stable versions of server.
So this patch takes another approach and changes code in such way
that LOCK TABLES for a temporary table will always request write
lock. In 5.1 version of this patch switch from read lock to write
lock is done inside of InnoDBs handler methods as doing it on
SQL-layer causes compatibility troubles with FLUSH TABLES WITH
READ LOCK.
Problem: MYSQL_BIN_LOG::reset_logs acquires mutexes in wrong order.
The correct order is first LOCK_thread_count and then LOCK_log. This function
does it the other way around. This leads to deadlock when run in parallel
with a thread that takes the two locks in correct order. For example, a thread
that disconnects will take the locks in the correct order.
Fix: change order of the locks in MYSQL_BIN_LOG::reset_logs:
first LOCK_thread_count and then LOCK_log.
VM-WIN2003-32-A, SLES10-IA64-A
The test case waits for master_pos_wait not to timeout, which
means that the deadlock between SQL and IO threads was
succesfully and automatically dealt with.
However, very rarely, master_pos_wait reports a timeout. This
happens because the time set for master_pos_wait to wait was
too small (6 seconds). On slow test env this could be a
problem.
We fix this by setting the timeout inline with the one used
in sync_slave_with_master (300 seconds). In addition we
refactored the test case and refined some comments.
causes future shutdown hang
InnoDB would hang on shutdown if any XA transactions exist in the
system in the PREPARED state. This has been masked by the fact that
MySQL would roll back any PREPARED transaction on shutdown, in the
spirit of Bug #12161 Xa recovery and client disconnection.
[mysql-test-run] do_shutdown_server: Interpret --shutdown_server 0 as
a request to kill the server immediately without initiating a
shutdown procedure.
xid_cache_insert(): Initialize XID_STATE::rm_error in order to avoid a
bogus error message on XA ROLLBACK of a recovered PREPARED transaction.
innobase_commit_by_xid(), innobase_rollback_by_xid(): Free the InnoDB
transaction object after rolling back a PREPARED transaction.
trx_get_trx_by_xid(): Only consider transactions whose
trx->is_prepared flag is set. The MySQL layer seems to prevent
attempts to roll back connected transactions that are in the PREPARED
state from another connection, but it is better to play it safe. The
is_prepared flag was introduced in the InnoDB Plugin.
trx_n_prepared: A new counter, counting the number of InnoDB
transactions in the PREPARED state.
logs_empty_and_mark_files_at_shutdown(): On shutdown, allow
trx_n_prepared transactions to exist in the system.
trx_undo_free_prepared(), trx_free_prepared(): New functions, to free
the memory objects of PREPARED transactions on shutdown. This is not
needed in the built-in InnoDB, because it would collect all allocated
memory on shutdown. The InnoDB Plugin needs this because of
innodb_use_sys_malloc.
trx_sys_close(): Invoke trx_free_prepared() on all remaining
transactions.
Bug#59410 read uncommitted: unlock row could not find a 3 mode lock
on the record
This bug is present only in 5.6 but I am adding the test case to earlier
versions to ensure it never appears in earlier versions too.
Bug#59410 read uncommitted: unlock row could not find a 3 mode lock
on the record
This bug is present only in 5.6 but I am adding the test case to earlier
versions to ensure it never appears in earlier versions too.
ARE NOT BEING HONORED
max_allowed_packet works in conjunction with net_buffer_length.
max_allowed_packet is an upper bound of net_buffer_length.
So it doesn't make sense to set the upper limit lower than the value.
Added a warning (using ER_UNKNOWN_ERRROR and a specific message)
when this is done (in the log at startup and when setting either
max_allowed_packet or the net_buffer_length variables)
Added a test case.
Fixed several tests that broke the above rule.
The slave was not able to find the correct row in the innodb
table, because the row fetched from the innodb table would not
match the before image. This happened because the (don't care)
bytes in the NULLed fields would change once the row was stored
in the storage engine (from zero to the default value). This
would make bulk memory comparison (using memcmp) to fail.
We fix this by taking a preventing measure and avoiding memcmp
for tables that contain nullable fields. Therefore, we protect
the slave search routine from engines that return arbitrary
values for don't care bytes (in the nulled fields). Instead, the
slave thread will only check null_bits and those fields that are
not set to NULL when comparing the before image against the
storage engine row.
There is a race between two threads: user thread and the dump
thread. The former sets a debug instruction that makes the latter wait
before processing an Xid event. There can be cases that the dump
thread has not yet processed the previous Xid event, causing it to
wait one Xid event too soon, thus causing sync_slave_with_master never
to resume.
We fix this by moving the instructions that set the debug variable
after calling sync_slave_with_master.
