ALTER TABLE ... ADD FOREIGN KEY may trigger assertion failure when
it has LOCK=EXCLUSIVE clause or concurrent FLUSH TABLES is being
executed.
In both cases being altered table is marked as flushed, which forces
subsequent attempt to open parent table to re-open. Which in turn is
not allowed while transaction is running.
Rather than opening parent table, just take appropriate MDL lock.
Also removed table_already_fk_prelocked() check: MDL itself has much
better methods to handle duplicate locks. E.g. the former won't acquire
MDL_SHARED_NO_WRITE if it already has MDL_SHARED_READ.
In commit d30f17af49 the change of
the loop iteration broke another error handling path that did
"goto error_handling_drop_uncached". Cover this code path with
fault injection, and revert to the correct iteration.
There are two fault injection labels innodb_OOM_prepare_inplace_alter.
Their order was swapped in MDEV-11369, so that the label that used
to be covered in an ADD INDEX code path would become unreachable
because the label that is executed for any ALTER TABLE was executed
first. Let us introduce the label innodb_OOM_prepare_add_index
for the more specific case.
row_merge_create_index_graph(): Relay the internal state
from dict_create_index_step(). Our caller should free the index
only if it was not copied, added to the cache, and freed.
row_merge_create_index(): Free the index template if it was
not added to the cache. This is a safer variant of the logic
that was introduced in 65070beffd in 10.2.
prepare_inplace_alter_table_dict(): Add additional fault injection
to exercise a code path where we have already added an index
to the cache.
Only starting with MariaDB 10.3.8 (MDEV-16365), InnoDB can actually
handle ALTER IGNORE TABLE correctly when introducing a NOT NULL
attribute to a column that contains a NULL value. Between
MariaDB Server 10.0 and 10.2, we would incorrectly return an error
for ALTER IGNORE TABLE when the column contains a NULL value.
On an error (such as when an index cannot be dropped due to
FOREIGN KEY constraints), the field dict_index_t::to_be_dropped
was only being cleared in debug builds, even though the field
is available and being used also in non-debug builds.
This was a regression that was introduced by myself originally
in MySQL 5.7.6 and later merged to MariaDB 10.2.2, in
d39898de8e
An error manifested itself in the MariaDB Server 10.4 non-debug build,
involving instant ADD or DROP column. Because an earlier failed
ALTER TABLE operation incorrectly left the dict_index_t::to_be_dropped
flag set, the column pointers of the index fields would fail to be
adjusted for instant ADD or DROP column (MDEV-15562). The instant
ADD COLUMN in MariaDB Server 10.3 is unlikely to be affected by a
similar scenario, because dict_table_t::instant_add_column() in 10.3
is applying the transformations to all indexes, not skipping
to-be-dropped ones.
The prtype & DATA_LONG_TRUE_VARCHAR flag only plays a role when
converting between InnoDB internal format and the MariaDB SQL layer
row format. Ideally this flag would never have been persisted in the
InnoDB data dictionary.
There were bogus assertion failures when an instant ADD, DROP, or
column reordering was combined with a change of extending a VARCHAR
from less than 256 bytes to more than 255 bytes. Such changes are
allowed starting with MDEV-15563 in MariaDB 10.4.3.
dict_table_t::instant_column(), dict_col_t::same_format(): Ignore
the DATA_LONG_TRUE_VARCHAR flag, because it does not affect the
persistent storage format.
This is follow-up to MDEV-18048: Relax a too strict debug assertion
This assertion should have been relaxed when implementing the first part of
MDEV-15563: instant removal of NOT NULL attribute for ROW_FORMAT=REDUNDANT
tables.
For ROW_FORMAT=REDUNDANT, there is no bitmap of null columns;
the null flags are encoded in the end offset of each field.
We do not really care about the number of fields that can be NULL.
MariaDB data-at-rest encryption (innodb_encrypt_tables)
had repurposed the same unused data field that was repurposed
in MySQL 5.7 (and MariaDB 10.2) for the Split Sequence Number (SSN)
field of SPATIAL INDEX. Because of this, MariaDB was unable to
support encryption on SPATIAL INDEX pages.
Furthermore, InnoDB page checksums skipped some bytes, and there
are multiple variations and checksum algorithms. By default,
InnoDB accepts all variations of all algorithms that ever existed.
This unnecessarily weakens the page checksums.
We hereby introduce two more innodb_checksum_algorithm variants
(full_crc32, strict_full_crc32) that are special in a way:
When either setting is active, newly created data files will
carry a flag (fil_space_t::full_crc32()) that indicates that
all pages of the file will use a full CRC-32C checksum over the
entire page contents (excluding the bytes where the checksum
is stored, at the very end of the page). Such files will always
use that checksum, no matter what the parameter
innodb_checksum_algorithm is assigned to.
For old files, the old checksum algorithms will continue to be
used. The value strict_full_crc32 will be equivalent to strict_crc32
and the value full_crc32 will be equivalent to crc32.
ROW_FORMAT=COMPRESSED tables will only use the old format.
These tables do not support new features, such as larger
innodb_page_size or instant ADD/DROP COLUMN. They may be
deprecated in the future. We do not want an unnecessary
file format change for them.
The new full_crc32() format also cleans up the MariaDB tablespace
flags. We will reserve flags to store the page_compressed
compression algorithm, and to store the compressed payload length,
so that checksum can be computed over the compressed (and
possibly encrypted) stream and can be validated without
decrypting or decompressing the page.
In the full_crc32 format, there no longer are separate before-encryption
and after-encryption checksums for pages. The single checksum is
computed on the page contents that is written to the file.
We do not make the new algorithm the default for two reasons.
First, MariaDB 10.4.2 was a beta release, and the default values
of parameters should not change after beta. Second, we did not
yet implement the full_crc32 format for page_compressed pages.
This will be fixed in MDEV-18644.
This is joint work with Marko Mäkelä.
dict_create_foreign_constraints_low(): Clean up the way in
which the error messages are initialized, and ensure that
the table name is always initialized.
If we instantly change the size of a fixed-length field
and treat it as kind-of variable-length, then we will need
conversions between old column values and new ones.
I tried adding such a conversion to row_build(), but then I
noticed that more conversions would be needed, because
old values still appeared in a freshly rebuilt secondary index,
causing a mismatch when trying to search with the correct
longer value that was converted in my provisional fix to row_build().
So, we will revert the essential part of
MDEV-15563: Instant ROW_FORMAT=REDUNDANT column extension
(commit 22feb179ae), but not
remove any tests.
innobase_build_col_map_add(): Do not assume that old_field->pack_length()
equals to field->pack_length(). Fix submitted by Aleksey Midenkov.
innobase_instant_try(): Assert that the column length of fixed-length
NOT NULL columns is only changing for ROW_FORMAT=REDUNDANT.
The Create_field::charset can contain garbage for columns
that the SQL layer does not consider as being string columns.
InnoDB considers BIT a string column for historical reasons
(and backward compatibility with old persistent InnoDB metadata),
and therefore it checked the charset.
The Field::charset() consistently is my_charset_bin for BIT,
so we can trust that one.
Field_str::is_equal(): Do not allow instant conversions between
BIT (which is stored big-endian) and integer types (which can
be stored big-endian or little-endian, depending on storage engine).
row_sel_field_store_in_mysql_format_func(): Properly extend
narrower integer and DATA_FIXBINARY values to the current format.
DATA_FIXBINARY was incorrectly padded with 0x20 instead of 0.
In tests that directly write InnoDB data file pages,
compute the innodb_checksum_algorithm=crc32 checksums,
instead of writing the 0xdeadbeef value used by
innodb_checksum_algorithm=none. In this way, these tests
will not cause failures when executing
./mtr --mysqld=--loose-innodb-checksum-algorithm=strict_crc32
instant_alter_column_possible(): Add the other MDEV-17459 work-around
condition. The existence of fulltext indexes only prevents instant
DROP COLUMN or changing the order of columns. Other forms of instant
ALTER TABLE are no problem.
Before commit 4e7ee166a9 that merged
the MDEV-18295 fix from 10.3, the work-around of MDEV-17459 in
instant_alter_column_possible() was categorically refusing any
ALGORITHM=INSTANT if any FULLTEXT INDEX was present. After that commit,
a related condition was only present in prepare_inplace_alter_table_dict()
but not in the other callers of instant_alter_column_possible().
Allow ALGORITHM=INSTANT (or avoid touching any data)
when changing the collation, or in some cases, the character set,
of a non-indexed CHAR or VARCHAR column. There is no penalty
for subsequent DDL or DML operations, and compatibility with
older MariaDB versions will be unaffected.
Character sets may be changed when the old encoding is compatible
with the new one. For example, changing from ASCII to anything
ASCII-based, or from 3-byte to 4-byte UTF-8 can sometimes be
performed instantly.
This is joint work with Eugene Kosov.
The test cases as well as ALTER_CONVERT_TO, charsets_are_compatible(),
Type_handler::Charsets_are_compatible() are his work.
The Field_str::is_equal(), Field_varstring::is_equal() and
the InnoDB changes were mostly rewritten by me due to conflicts
with MDEV-15563.
Limitations:
Changes of indexed columns will still require
ALGORITHM=COPY. We should allow ALGORITHM=NOCOPY and allow
the indexes to be rebuilt inside the storage engine,
without copying the entire table.
Instant column size changes (in bytes) are not supported by
all storage engines.
Instant CHAR column changes are only allowed for InnoDB
ROW_FORMAT=REDUNDANT. We could allow this for InnoDB
when the CHAR internally uses a variable-length encoding,
say, when converting from 3-byte UTF-8 to 4-byte UTF-8.
Instant VARCHAR column changes are allowed for InnoDB
ROW_FORMAT=REDUNDANT, and for others only if the size
in bytes does not change from 128..255 bytes to more
than 256 bytes.
Inside InnoDB, this slightly changes the way how MDEV-15563
works and fixes the result of the innodb.instant_alter_extend test.
We change the way how ALTER_COLUMN_EQUAL_PACK_LENGTH_EXT
is handled. All column extension, type changes and renaming
now go through a common route, except when ctx->is_instant()
is in effect, for example, instant ADD or DROP COLUMN has
been initiated. Only in that case we will go through
innobase_instant_try() and rewrite all column metadata.
get_type(field, prtype, mtype, len): Convert a SQL data type into
InnoDB column metadata.
innobase_rename_column_try(): Remove the update of SYS_COLUMNS.
innobase_rename_or_enlarge_column_try(): New function,
replacing part of innobase_rename_column_try() and all of
innobase_enlarge_column_try(). Also changes column types.
innobase_rename_or_enlarge_columns_cache(): Also change
the column type.
This was developed by Aleksey Midenkov based on my design.
In the original InnoDB storage format (that was retroactively named
ROW_FORMAT=REDUNDANT in MySQL 5.0.3), the length of each index field
is stored explicitly.
Because of this, we can and now will allow instant conversion from
VARCHAR to CHAR or VARBINARY to BINARY of equal or greater size,
as well as instant conversion of TINYINT to SMALLINT to MEDIUMINT
to INT to BIGINT (while not changing between signed and unsigned).
Theoretically, we could allow changing from an unsigned integer to
a bigger unsigned integer, as well as changing CHAR to VARCHAR, but
that would require additional metadata and conversions whenever
reading old records.
Field_str::is_equal(), Field_varstring::is_equal(), Field_num::is_equal():
Return the new result IS_EQUAL_PACK_LENGTH_EXT if the table advertises
HA_EXTENDED_TYPES_CONVERSION capability and we are considering the
above-mentioned conversions.
ALTER_COLUMN_EQUAL_PACK_LENGTH_EXT: A new ALTER TABLE flag, similar
to ALTER_COLUMN_EQUAL_PACK_LENGTH but requiring conversions when
reading the data. The Field::is_equal() result IS_EQUAL_PACK_LENGTH_EXT
will map to this flag.
dtype_get_fixed_size_low(): For BINARY, CHAR and integer columns
in ROW_FORMAT=REDUNDANT, return 0 (variable length) from now on.
dtype_get_sql_null_size(): Keep returning the current size for
BINARY, CHAR and integer columns, so that in ROW_FORMAT=REDUNDANT
it will remain possible to update in place between NULL and NOT NULL
values.
btr_index_rec_validate(): Relax a CHECK TABLE length check for
ROW_FORMAT=REDUNDANT tables.
btr_cur_instant_init_low(): No longer trust fixed_len
for ROW_FORMAT=REDUNDANT tables.
We cannot rely on fixed_len anymore because the record can have shorter
length from before instant extension. Note that importing such tablespace
into earlier MariaDB versions produces ER_TABLE_SCHEMA_MISMATCH when
using a .cfg file.
In the original InnoDB storage format (which was retroactively named
ROW_FORMAT=REDUNDANT in MySQL 5.0.3), the length of each index field
is stored explicitly. Thus, we can and from now on will allow arbitrary
extension of VARBINARY and VARCHAR columns when the table is in
ROW_FORMAT=REDUNDANT.
ha_innobase::open(): Advertise a new HA_EXTENDED_TYPES_CONVERSION
capability for ROW_FORMAT=REDUNDANT tables.
Field_varstring::is_equal(): If the HA_EXTENDED_TYPES_CONVERSION
capability is advertised for the table, return IS_EQUAL_PACK_LENGTH
for any length extension.
For up to 127 bytes length, InnoDB would use 1 byte for length, and
that byte would always be less than 128. If the maximum length is
longer than 255 bytes, InnoDB would use a variable-length encoding
for the length, using 1 byte for lengths up to 127 bytes, and
2 bytes for longer lengths.
Thus, 1-byte lengths are always compatible when the maximum size
changes from less than 128 bytes to anything longer.
Field_varstring::is_equal(): Return IS_EQUAL_PACK_LENGTH also when
converting from VARCHAR less than 128 bytes to any longer VARCHAR.
When importing a tablespace, we must initialize dummy DEFAULT NULL
values for any instantly added columns in order to avoid a debug
assertion failure when PageConverter::update_records() invokes
rec_get_offsets(). Finally, when the operation completes, we must
evict and reload the table definition, so that the correct
default values for instantly added columns will be loaded.
ha_innobase::discard_or_import_tablespace(): On successful
IMPORT TABLESPACE, evict and reload the table definition,
so that btr_cur_instant_init() will load the correct metadata.
PageConverter::update_index_page(): Fill in dummy DEFAULT NULL values
for instantly added columns. These will be replaced upon the
completion of the operation by evicting and reloading the metadata.
row_discard_tablespace(): Invoke dict_table_t::remove_instant().
After DISCARD TABLESPACE, the table is no longer in "instant ALTER"
format, because there is no data file attached.
The code path where the table was not being rebuilt during ALTER TABLE
was not covered by the test. Add coverage, and remove the debug assertion
that could fail in this case.
