Based on mysql/mysql-server@bc9c46bf28
but without sleeps.
The test was verified to hit the debug assertion if the change to
fts_add_doc_by_id() in commit 2d98b967e3
was reverted.
InnoDB commit fails when consecutive FTS_DOC_ID value
is greater than 4294967295.
Fix is that InnoDB should remove the delta FTS_DOC_ID
value limitations and fts should encode 8 byte value,
remove FTS_DOC_ID_MAX_STEP variable. Replaced the
fts0vlc.ic file with fts0vlc.h
fts_encode_int(): Should be able to encode 10 bytes value
fts_get_encoded_len(): Should get the length of the value
which has 10 bytes
fts_decode_vlc(): Add debug assertion to verify the maximum
length allowed is 10.
mach_read_uint64_little_endian(): Reads 64 bit stored in
little endian format
Added a unit test case which check for minimum and maximum
value to do the fts encoding
create_table_info_t::innobase_table_flags(): Refuse to create
a PAGE_COMPRESSED table with PAGE_COMPRESSION_LEVEL=0 if also
innodb_compression_level=0.
The parameter value innodb_compression_level=0 was only somewhat
meaningful for testing or debugging ROW_FORMAT=COMPRESSED tables.
For the page_compressed format, it never made any sense, and the
check in dict_tf_is_valid_not_redundant() that was added in
72378a2583 (MDEV-12873) would cause
the server to crash.
This is a duplicate of MDEV-18278 89936f11e9, but I will add an
additional assertion
Description:
The frm corruption should not be reported during CREATE TABLE. Normally
it doesn't, and the data to fill TABLE is taken by open_table_from_share
call. However, the vcol data is stored as SQL string in
table->s->vcol_defs.str and is anyway parsed on each table open.
It is impossible [or hard] to avoid, because it's hard to clone the
expression tree in general (it's easier to parse).
Normally parse_vcol_defs should only fail on semantic errors. If so,
error_reported is set to true. Any other failure is not expected during
table creation. There is either unhandled/unacknowledged error, or
something went really wrong, like memory reject. This all should be
asserted anyway.
Solution:
* Set *error_reported=true for the forward references check;
* Assert for every unacknowledged error during table creation.
Problem:
========
This patch addresses two issues.
First, if a CHANGE MASTER command is issued and an error happens
while locating the replica’s relay logs, the logs can be put into an
invalid state where future updates fail and future CHANGE MASTER
calls crash the server. More specifically, right before a replica
purges the relay logs (part of the `CHANGE MASTER TO` logic), the
relay log is temporarily closed with state LOG_TO_BE_OPENED. If the
server errors in-between the temporary log closure and purge, i.e.
during the function find_log_pos, the log should be closed.
MDEV-25284 reveals the log is not properly closed.
Second, upon issuing a RESET SLAVE ALL command, a slave’s GTID
filters are not cleared (DO_DOMAIN_IDS, IGNORE_DOMIAN_IDS,
IGNORE_SERVER_IDS). MySQL had a similar bug report, Bug #18816897,
which fixed this issue to clear IGNORE_SERVER_IDS after issuing
RESET SLAVE ALL in version 5.7.
Solution:
=========
To fix the first problem, the CHANGE MASTER error handling logic was
extended to transition the relay log state to LOG_CLOSED from
LOG_TO_BE_OPENED.
To fix the second problem, the RESET SLAVE ALL logic is extended to
clear the domain_id filter and ignore_server_ids.
Reviewed By:
============
Andrei Elkin <andrei.elkin@mariadb.com>
This happens upon CREATE USER and DROP ROLE.
The underlying problem is that our HASH implementation shuffles elements
around when performing an update or delete. This means that when doing a
scan through the HASH table by index, in search of elements to delete or
update one must restart the scan to make sure nothing is missed if at least
one delete / update happened.
More specifically, what happened in this case:
The hash has 131 element, DROP ROLE removes the element
[119]. Its [119]->next was element [129], so [129] is moved to [119].
Now we need to compact the hash, removing the last element [130]. It
gets one bit off its hash value and becomes element [2]. The existing
element [2] is moved to [129], and old [130] is moved to [2].
We cannot simply move [130] to [129] and make [2]->next=130, it won't
work if [2] is itself in the collision list and doesn't belong in [2].
The handle_grant_struct code assumed that it is safe to continue by only
reexamining the currently modified / deleted element index, but that is
not true.
Missing to delete an element in the hash triggered the assertion in
the test case. DROP ROLE would not clear all necessary role->role or
role->user mappings.
To fix the problem we ensure that the scan is restarted, only if an
element was deleted / updated, similar to how bubble-sort keeps sorting
until it finds no more elements to swap.
On deadlock transaction is rolled back (and trx->state is cleared) but
SELECT continued the loop because evaluate_join_record() ignored the
error status returned from lower join evaluation. val_int() does not
return error status so it is checked by thd->is_error().
Test case was created by Thirunarayanan Balathandayuthapani
<thiru@mariadb.com>
If a table has no unique indexes, write set key information will be collected on all columns in the table.
The write set key information has space only for max 3500 bytes for individual column, and if a varchar colummn of such non-primary key table is longer than
this limit, currently a crash follows.
The fix in this commit, is to truncate key values extracted from such long varhar columns to max 3500 bytes.
This may potentially lead to false positive certification failures for transactions, which operate on separate cluster nodes, and update/insert/delete table rows, which differ only in the part of such long columns after 3500 bytes border.
Reviewed-by: Jan Lindström <jan.lindstrom@mariadb.com>
Let us mask the actual values of the defragmentation-related fields,
because they may vary. Also, remove the dependency on purge,
and instead delete records by a ROLLBACK of INSERT.
Use in_sum_func (and so nest_level) only in LEX to which SELECT lex belong to
Reduce usage of current_select (because it does not always point on the correct
SELECT_LEX, for example with prepare.
Change context for all classes inherited from Item_ident (was only for Item_field) in case of pushing down it to HAVING.
Now name resolution context have to have SELECT_LEX reference if the context is present.
Fixed feedback plugin stack usage.
This patch is the plan D variant for fixing potetial mutex locking
order exercised by BF aborting and KILL command execution.
In this approach, KILL command is replicated as TOI operation.
This guarantees total isolation for the KILL command execution
in the first node: there is no concurrent replication applying
and no concurrent DDL executing. Therefore there is no risk of
BF aborting to happen in parallel with KILL command execution
either. Potential mutex deadlocks between the different mutex
access paths with KILL command execution and BF aborting cannot
therefore happen.
TOI replication is used, in this approach, purely as means
to provide isolated KILL command execution in the first node.
KILL command should not (and must not) be applied in secondary
nodes. In this patch, we make this sure by skipping KILL
execution in secondary nodes, in applying phase, where we
bail out if applier thread is trying to execute KILL command.
This is effective, but skipping the applying of KILL command
could happen much earlier as well.
This patch also fixes mutex locking order and unprotected
THD member accesses on bf aborting case. We try to hold
THD::LOCK_thd_data during bf aborting. Only case where it
is not possible is at wsrep_abort_transaction before
call wsrep_innobase_kill_one_trx where we take InnoDB
mutexes first and then THD::LOCK_thd_data.
This will also fix possible race condition during
close_connection and while wsrep is disconnecting
connections.
Added wsrep_bf_kill_debug test case
Reviewed-by: Jan Lindström <jan.lindstrom@mariadb.com>
At least since commit 055a3334ad
(MDEV-13564) the undo log truncation in InnoDB did not work correctly.
The main issue is that during the execution of
trx_purge_truncate_history() some pages of the newly truncated
undo tablespace could be discarded.
fsp_try_extend_data_file(): Apply the peculiar rounding of
fil_space_t::size_in_header only to the system tablespace,
whose size can be expressed in megabytes in a configuration parameter.
Other files may freely grow by a number of pages.
fseg_alloc_free_page_low(): Do allow the extension of undo tablespaces,
and mention the file name in the error message.
mtr_t::commit_shrink(): Implement crash-safe shrinking of a tablespace
file. First, durably write the log, then shrink the file, and finally
release the page latches of the rebuilt tablespace. Refactored from
trx_purge_truncate_history().
log_write_and_flush_prepare(), log_write_and_flush(): New functions
to durably write log during mtr_t::commit_shrink().
btr_defragment_save_defrag_stats_if_needed(): Do not save
defragmentation statistics for temporary tables.
They are exempt of defragmentation anyway
(ha_innobase::optimize() never invokes defragmentation for them),
and the user-visible names are not available inside InnoDB.
Furthermore, InnoDB assumes that temporary tables are never accessed
by other threads than the one that handles the session with which
the temporary table is associated with.
Furthermore, we simplify the test innodb.innodb_defrag_stats
and include a test case that demonstrates that defragmentation
statistics are no longer being saved for temporary tables.
The st_blksize returned by fstat(2) is not documented to be
a power of 2, like we assumed in
commit 58252fff15 (MDEV-26040).
While on Linux, the st_blksize appears to report the file system
block size (which hopefully is not smaller than the sector size
of the underlying block device), on FreeBSD we observed
st_blksize values that might have been something similar to st_size.
Also IBM AIX was affected by this. A simple test case would
lead to a crash when using the minimum innodb_buffer_pool_size=5m
on both FreeBSD and AIX:
seq -f 'create table t%g engine=innodb select * from seq_1_to_200000;' \
1 100|mysql test&
seq -f 'create table u%g engine=innodb select * from seq_1_to_200000;' \
1 100|mysql test&
We will fix this by not trusting st_blksize at all, and assuming that
the smallest allowed write size (for O_DIRECT) is 4096 bytes. We hope
that no storage systems with larger block size exist. Anything larger
than 4096 bytes should be unlikely, given that it is the minimum
virtual memory page size of many contemporary processors.
MariaDB Server on Microsoft Windows was not affected by this.
While the 512-byte sector size of the venerable Seagate ST-225 is still
in widespread use, the minimum innodb_page_size is 4096 bytes, and
innodb_log_file_size can be set in integer multiples of 65536 bytes.
The only occasion where InnoDB uses smaller data file block sizes than
4096 bytes is with ROW_FORMAT=COMPRESSED tables with KEY_BLOCK_SIZE=1
or KEY_BLOCK_SIZE=2 (or innodb_page_size=4096). For such tables,
we will from now on preallocate space in integer multiples of 4096 bytes
and let regular writes extend the file by 1024, 2048, or 3072 bytes.
The view INFORMATION_SCHEMA.INNODB_SYS_TABLESPACES.FS_BLOCK_SIZE
should report the raw st_blksize.
For page_compressed tables, the function fil_space_get_block_size()
will map to 512 any st_blksize value that is larger than 4096.
os_file_set_size(): Assume that the file system block size is 4096 bytes,
and only support extending files to integer multiples of 4096 bytes.
fil_space_extend_must_retry(): Round down the preallocation size to
an integer multiple of 4096 bytes.
Improve documentation of performance_schema tables by appending COLUMN
comments to tables. Additionally improve test coverage and update corresponding
tests.
Import operation without .cfg file fails when there is mismatch of index
between metadata table and .ibd file. Moreover, MDEV-19022 shows
that InnoDB can end up with index tree where non-leaf page has only
one child page. So it is unsafe to find the secondary index root page.
This patch does the following when importing the table without .cfg file:
1) If the metadata contains more than one index then InnoDB stops
the import operation and report the user to drop all secondary
indexes before doing import operation.
2) When the metadata contain only clustered index then InnoDB finds the
index id by reading page 0 & page 3 instead of traversing the
whole tablespace.
ha_partition stores records in array of m_ordered_rec_buffer and uses
it for prio queue in ordered index scan. When the records are restored
from the array the blob buffers may be already freed or rewritten.
The solution is to take temporary ownership of cached blob buffers via
String::swap(). When the record is restored from m_ordered_rec_buffer
the ownership is returned to table fields.
Cleanups:
init_record_priority_queue(): removed needless !m_ordered_rec_buffer
check as there is same assertion few lines before.
dbug_print_row() for arbitrary row pointer
Server crashes in Field::register_field_in_read_map upon select from
partitioned table with indexed by prefix virtual column.
After several read-mark fixes a problem has surfaced:
Since KEY (c(10),a) uses only a prefix of c, a new field is created,
duplicated from table->field[3], with a new length. However,
vcol_inco->expr is not copied.
Therefore, (*key_info)->key_part[i].field->vcol_info->expr was left NULL
in ha_partition::index_init().
Solution: copy vcol_info from table field when it's set up.
len was containing garbage, since vctempl->mysql_col_offset was
containing old value while calling row_mysql_store_col_in_innobase_format
from innobase_get_computed_value().
It was not updated after the first ALTER TABLE call, because it's INPLACE
logic considered there's nothing to update, and exited immediately from
ha_innobase::inplace_alter_table().
However, vcol metadata needs an update, since vcols structure is changed
in mysql record.
The regression was introduced by 12614af1fe. There, refcount==1 condition
was removed, which turned out to be crucial, though racy. The idea was to
update vc_templ after each (sequencing) ALTER TABLE.
We should do the same another way, and there may be a plenty of solutions,
but the simplest one is to add a following condition:
if vcol structure is changed, drop vc_templ; it will be recreated on next
ha_innobase::open() call.
in prepare_inplace_alter_table. It is safe, since innodb inplace changes
require at least HA_ALTER_INPLACE_SHARED_LOCK_AFTER_PREPARE, which
guarantee MDL_EXCLUSIVE on this stage.
alter_templ_needs_rebuild() also has to track the columns not indexed, to
keep vc_templ correct.
Note that vc_templ is always kept constructed and available after
ha_innobase::open() call, even on INSERT, though no virtual columns are
evaluated during that statement
inside innodb.
In the test case suplied, it will be recreated on the second ALTER TABLE.
Server crashes in Field::register_field_in_read_map upon select from
partitioned table with indexed by prefix virtual column.
After several read-mark fixes a problem has surfaced:
Since KEY (c(10),a) uses only a prefix of c, a new field is created,
duplicated from table->field[3], with a new length. However,
vcol_inco->expr is not copied.
Therefore, (*key_info)->key_part[i].field->vcol_info->expr was left NULL
in ha_partition::index_init().
Solution: initialize vcols before key initialization
Also key initialization is moved to a function.
Problem was that not all normal error codes where not handled
after wsrep_row_upd_check_foreign_constraints() call. Furhermore,
debug assertion did not contain all normal error cases. Changed
ib:: calls to WSREP_ calls to use wsrep instrumentation.
Problem:
=========
As a part of MDEV-14398 patch, InnoDB added and removed
the tablespace from default encrypt list. But InnoDB removes
the tablespace from the default encrypt list too early due to
i) other encryption thread working on the tablespace
ii) When tablespace is being flushed at the end of
key rotation
InnoDB fails to decrypt/encrypt the tablespace since
the tablespace removed too early and it leads to
test case failure.
Solution:
=========
Avoid the removal of tablespace from default_encrypt_list
only when
1) Another active encryption thread working on tablespace
2) Eligible for tablespace key rotation
3) Tablespace is in flushing phase
Removed the workaround in encryption.innodb_encryption_filekeys test case.
Set tests to non-valgrind:
oqgraph.social
encryption.innodb-page_encryption
binlog_encryption.encrypted_master
innodb.innodb-page_compression_lz4
main.lock_multi_bug38499
main.lock_multi_bug38691
In commit 83d2e0841e (MDEV-24041)
we failed to notice that in addition to the bug with
DELETE and ON DELETE CASCADE, there is another bug with
UPDATE and ON UPDATE CASCADE.
row_ins_foreign_fill_virtual(): Use the correct memory heap
for everything that will be reachable from the cascade->update
that we return to the caller.
Note: It is correct to use the shorter-lived cascade->heap for
rec_get_offsets(), because that memory will be abandoned when
row_ins_foreign_fill_virtual() returns.
ha_innobase::prepare_inplace_alter_table(): Unless the table is
being rebuilt, determine the maximum column length based on the
current ROW_FORMAT of the table. When TABLE_SHARE (and the .frm file)
contains no explicit ROW_FORMAT, InnoDB table creation or rebuild
will use innodb_default_row_format.
Based on mysql/mysql-server@3287d33acd
InnoDB tablespace identifiers and page numbers are 32-bit numbers.
Let us use a 32-bit type for them in innochecksum.
The changes in commit 1918bdf32c
broke the build on 32-bit Windows.
Thanks to Vicențiu Ciorbaru for an initial version of this fixup.
The columns that are part of DEFAULT expression were not read-marked
in statements like UPDATE...SET b=DEFAULT.
The problem is `F(DEFAULT)` expression depends of the left-hand side of an
assignment. However, setup_fields accepts only right-hand side value.
Neither Item::fix_fields does.
Suchwise, b=DEFAULT(b) works fine, because Item_default_field has
information on what field it is default of:
if (thd->mark_used_columns != MARK_COLUMNS_NONE)
def_field->default_value->expr->update_used_tables();
in Item_default_value::fix_fields().
It is not reasonable to pass a left-hand side to Item:fix_fields, because
the case is rare, so the rewrite
b= F(DEFAULT) -> b= F(DEFAULT(b))
is made instead.
Both UPDATE and multi-UPDATE are affected, however any form of INSERT
is not: it marks all the fields in DEFAULT expressions for read in
TABLE::mark_default_fields_for_write().
The problem is the same as in MDEV-18166: columns in virtual field
expression are not marked for read, while the field itself does.
field->register_field_in_read_map() should be called for read-marking all
fields.
The test is reproduced only in 10.4+, however the fix is applicable to
10.2+.
Several different test cases were failing under the same reason: the
fields in a vcol expression were not marked during marking columns of a key
contatining virtual column for read.
Fix: make marking columns of a key for read a special case where
register_field_in_read_map() is done instead of plain bitmap_set_bit().
Some test cases are only reproducible in 10.4+, but the fix is applicable
to 10.2+
