This was the last abuse of trx_sys.mutex, which is now exclusively
protecting trx_sys.trx_list.
This global acquisition was also potential scalability bottleneck for
oltp_read_write benchmark. Although it didn't expose itself as such due
to larger scalability issues.
Replaced trx_sys.mutex based synchronisation between ReadView creator
thread and purge coordinator thread performing latest view clone with
ReadView::m_mutex.
It also allowed to simplify tri-state view m_state down to boolean
m_open flag.
For performance reasons trx->read_view.close() is left as atomic relaxed
store, so that we don't have to waste resources for otherwise meaningless
mutex acquisition.
dict_stats_update_if_needed(): Replace the parameter THD*
with const trx_t& so that trx_t::is_wsrep() can be invoked
instead of the more expensive wsrep_on().
Replace also other occurrences of wsrep_on() with trx_t::is_wsrep().
The template parameter mtr_t::OPT refers to optional, not optimized.
Also the default parameter mtr_t::NORMAL refers to optimized writes.
The name MAYBE_NOP would be more descriptive, conveying the idea
that a write to a durable page might not actually have any effect.
The -Wconversion in GCC seems to be stricter than in clang.
GCC at least since version 4.4.7 issues truncation warnings for
assignments to bitfields, while clang 10 appears to only issue
warnings when the sizes in bytes rounded to the nearest integer
powers of 2 are different.
Before GCC 10.0.0, -Wconversion required more casts and would not
allow some operations, such as x<<=1 or x+=1 on a data type that
is narrower than int.
GCC 5 (but not GCC 4, GCC 6, or any later version) is complaining
about x|=y even when x and y are compatible types that are narrower
than int. Hence, we must rewrite some x|=y as
x=static_cast<byte>(x|y) or similar, or we must disable -Wconversion.
In GCC 6 and later, the warning for assigning wider to bitfields
that are narrower than 8, 16, or 32 bits can be suppressed by
applying a bitwise & with the exact bitmask of the bitfield.
For older GCC, we must disable -Wconversion for GCC 4 or 5 in such
cases.
The bitwise negation operator appears to promote short integers
to a wider type, and hence we must add explicit truncation casts
around them. Microsoft Visual C does not allow a static_cast to
truncate a constant, such as static_cast<byte>(1) truncating int.
Hence, we will use the constructor-style cast byte(~1) for such cases.
This has been tested at least with GCC 4.8.5, 5.4.0, 7.4.0, 9.2.1, 10.0.0,
clang 9.0.1, 10.0.0, and MSVC 14.22.27905 (Microsoft Visual Studio 2019)
on 64-bit and 32-bit targets (IA-32, AMD64, POWER 8, POWER 9, ARMv8).
log_t::FORMAT_10_5: physical redo log format tag
log_phys_t: Buffered records in the physical format.
The log record bytes will follow the last data field,
making use of alignment padding that would otherwise be wasted.
If there are multiple records for the same page, also those
may be appended to an existing log_phys_t object if the memory
is available.
In the physical format, the first byte of a record identifies the
record and its length (up to 15 bytes). For longer records, the
immediately following bytes will encode the remaining length
in a variable-length encoding. Usually, a variable-length-encoded
page identifier will follow, followed by optional payload, whose
length is included in the initially encoded total record length.
When a mini-transaction is updating multiple fields in a page,
it can avoid repeating the tablespace identifier and page number
by setting the same_page flag (most significant bit) in the first
byte of the log record. The byte offset of the record will be
relative to where the previous record for that page ended.
Until MDEV-14425 introduces a separate file-level log for
redo log checkpoints and file operations, we will write the
file-level records in the page-level redo log file.
The record FILE_CHECKPOINT (which replaces MLOG_CHECKPOINT)
will be removed in MDEV-14425, and one sequential scan of the
page recovery log will suffice.
Compared to MLOG_FILE_CREATE2, FILE_CREATE will not include any flags.
If the information is needed, it can be parsed from WRITE records that
modify FSP_SPACE_FLAGS.
MLOG_ZIP_WRITE_STRING: Remove. The record was only introduced temporarily
as part of this work, before being replaced with WRITE (along with
MLOG_WRITE_STRING, MLOG_1BYTE, MLOG_nBYTES).
mtr_buf_t::empty(): Check if the buffer is empty.
mtr_t::m_n_log_recs: Remove. It suffices to check if m_log is empty.
mtr_t::m_last, mtr_t::m_last_offset: End of the latest m_log record,
for the same_page encoding.
page_recv_t::last_offset: Reflects mtr_t::m_last_offset.
Valid values for last_offset during recovery should be 0 or above 8.
(The first 8 bytes of a page are the checksum and the page number,
and neither are ever updated directly by log records.)
Internally, the special value 1 indicates that the same_page form
will not be allowed for the subsequent record.
mtr_t::page_create(): Take the block descriptor as parameter,
so that it can be compared to mtr_t::m_last. The INIT_INDEX_PAGE
record will always followed by a subtype byte, because same_page
records must be longer than 1 byte.
trx_undo_page_init(): Combine the writes in WRITE record.
trx_undo_header_create(): Write 4 bytes using a special MEMSET
record that includes 1 bytes of length and 2 bytes of payload.
flst_write_addr(): Define as a static function. Combine the writes.
flst_zero_both(): Replaces two flst_zero_addr() calls.
flst_init(): Do not inline the function.
fsp_free_seg_inode(): Zerofill the whole inode.
fsp_apply_init_file_page(): Initialize FIL_PAGE_PREV,FIL_PAGE_NEXT
to FIL_NULL when using the physical format.
btr_create(): Assert !page_has_siblings() because fsp_apply_init_file_page()
must have been invoked.
fil_ibd_create(): Do not write FILE_MODIFY after FILE_CREATE.
fil_names_dirty_and_write(): Remove the parameter mtr.
Write the records using a separate mini-transaction object,
because any FILE_ records must be at the start of a mini-transaction log.
recv_recover_page(): Add a fil_space_t* parameter.
After applying log to the a ROW_FORMAT=COMPRESSED page,
invoke buf_zip_decompress() to restore the uncompressed page.
buf_page_io_complete(): Remove the temporary hack to discard the
uncompressed page of a ROW_FORMAT=COMPRESSED page.
page_zip_write_header(): Remove. Use mtr_t::write() or
mtr_t::memset() instead, and update the compressed page frame
separately.
trx_undo_header_add_space_for_xid(): Remove.
trx_undo_seg_create(): Perform the changes that were previously
made by trx_undo_header_add_space_for_xid().
btr_reset_instant(): New function: Reset the table to MariaDB 10.2
or 10.3 format when rolling back an instant ALTER TABLE operation.
page_rec_find_owner_rec(): Merge with the only callers.
page_cur_insert_rec_low(): Combine writes by using a local buffer.
MEMMOVE data from the preceding record whenever feasible
(copying at least 3 bytes).
page_cur_insert_rec_zip(): Combine writes to page header fields.
PageBulk::insertPage(): Issue MEMMOVE records to copy a matching
part from the preceding record.
PageBulk::finishPage(): Combine the writes to the page header
and to the sparse page directory slots.
mtr_t::write(): Only log the least significant (last) bytes
of multi-byte fields that actually differ.
For updating FSP_SIZE, we must always write all 4 bytes to the
redo log, so that the fil_space_set_recv_size() logic in
recv_sys_t::parse() will work.
mtr_t::memcpy(), mtr_t::zmemcpy(): Take a pointer argument
instead of a numeric offset to the page frame. Only log the
last bytes of multi-byte fields that actually differ.
In fil_space_crypt_t::write_page0(), we must log also any
unchanged bytes, so that recovery will recognize the record
and invoke fil_crypt_parse().
Future work:
MDEV-21724 Optimize page_cur_insert_rec_low() redo logging
MDEV-21725 Optimize btr_page_reorganize_low() redo logging
MDEV-21727 Optimize redo logging for ROW_FORMAT=COMPRESSED
Pass buf_block_t* to all functions that write redo log.
Specifically, replace the parameters page,page_zip
with buf_block_t* block in page_zip_ functions.
btr_cur_upd_rec_sys(): Replaces row_upd_rec_sys_fields() and
implements redo logging.
row_upd_rec_sys_fields_in_recovery(): Remove, and merge to the
only remaining caller btr_cur_parse_update_in_place().
btr_cur_del_mark_set_clust_rec_log(),
btr_cur_del_mark_set_sec_rec_log(),
btr_cur_set_deleted_flag_for_ibuf():
Remove, and replace with btr_rec_set_deleted<bool>().
page_zip_rec_set_deleted(): Add the parameter mtr, and write a
MLOG_ZIP_WRITE_STRING record to the log.
In commit 0e5a4ac253 (MDEV-15562)
we introduced was a bogus debug check failure that does not affect
the correctness of the release build.
With a fixed-length PRIMARY KEY, we do not have to recompute
the rec_get_offsets() after restarting the mini-transaction,
because the offsets of DB_TRX_ID,DB_ROLL_PTR are not going
to change.
row_undo_mod_clust(): Invoke rec_offs_make_valid() to keep the
debug check in page_zip_write_trx_id_and_roll_ptr() happy.
The scenario to reproduce this bug should be rather unlikely:
In the time frame when row_undo_mod_clust() has committed its
first mini-transaction and has not yet started the next one,
another mini-transaction must do something that causes the page
to be reorganized, split or merged.
WL#6326 in MariaDB 10.2.2 introduced a potential hang on purge or rollback
when an index tree is being shrunk by multiple levels.
This fix is based on
mysql/mysql-server@f2c5852630
with the main difference that our version of the test case uses
DEBUG_SYNC instrumentation on ROLLBACK, not on purge.
btr_cur_will_modify_tree(): Simplify the check further.
This is the actual bug fix.
row_undo_mod_remove_clust_low(), row_undo_mod_clust(): Add DEBUG_SYNC
instrumentation for the test case.
offset_t: this is a type which represents one record offset.
It's unsigned short int.
a lot of functions: replace ulint with offset_t
btr_pcur_restore_position_func(),
page_validate(),
row_ins_scan_sec_index_for_duplicate(),
row_upd_clust_rec_by_insert_inherit_func(),
row_vers_impl_x_locked_low(),
trx_undo_prev_version_build():
allocate record offsets on the stack instead of waiting for rec_get_offsets()
to allocate it from mem_heap_t. So, reducing memory allocations.
RECORD_OFFSET, INDEX_OFFSET:
now it's less convenient to store pointers in offset_t*
array. One pointer occupies now several offset_t. And those constant are start
indexes into array to places where to store pointer values
REC_OFFS_HEADER_SIZE: adjusted for the new reality
REC_OFFS_NORMAL_SIZE:
increase size from 100 to 300 which means less heap allocations.
And sizeof(offset_t[REC_OFFS_NORMAL_SIZE]) now is 600 bytes which
is smaller than previous 800 bytes.
REC_OFFS_SEC_INDEX_SIZE: adjusted for the new reality
rem0rec.h, rem0rec.ic, rem0rec.cc:
various arguments, return values and local variables types were changed to
fix numerous integer conversions issues.
enum field_type_t:
offset types concept was introduces which replaces old offset flags stuff.
Like in earlier version, 2 upper bits are used to store offset type.
And this enum represents those types.
REC_OFFS_SQL_NULL, REC_OFFS_MASK: removed
get_type(), set_type(), get_value(), combine():
these are convenience functions to work with offsets and it's types
rec_offs_base()[0]:
still uses an old scheme with flags REC_OFFS_COMPACT and REC_OFFS_EXTERNAL
rec_offs_base()[i]:
these have type offset_t now. Two upper bits contains type.
mtr_t::memcpy(): Replaces mlog_write_string(), mlog_log_string().
The buf_block_t is passed a parameter, so that
mlog_write_initial_log_record_low() can be used instead of
mlog_write_initial_log_record_fast().
fil_space_crypt_t::write_page0(): Remove the fil_space_t* parameter.
Apart from page latches (buf_block_t::lock), mini-transactions
are keeping track of at most one dict_index_t::lock and
fil_space_t::latch at a time, and in a rare case, purge_sys.latch.
Let us introduce interfaces for acquiring an index latch
or a tablespace latch.
In a later version, we may want to introduce mtr_t members
for holding a latched dict_index_t* and fil_space_t*,
and replace the remaining use of mtr_t::m_memo
with std::set<buf_block_t*> or with a map<buf_block_t*,byte*>
pointing to log records.
dict_sys.lock(), dict_sys_lock(): Acquire both mutex and latch.
dict_sys.unlock(), dict_sys_unlock(): Release both mutex and latch.
dict_sys.assert_locked(): Assert that both mutex and latch are held.
dict_sys_t::create(): Renamed from dict_init().
dict_sys_t::close(): Renamed from dict_close().
dict_sys_t::add(): Sliced from dict_table_t::add_to_cache().
dict_sys_t::remove(): Renamed from dict_table_remove_from_cache().
dict_sys_t::prevent_eviction(): Renamed from
dict_table_move_from_lru_to_non_lru().
dict_sys_t::acquire(): Replaces dict_move_to_mru() and some more logic.
dict_sys_t::resize(): Renamed from dict_resize().
dict_sys_t::find(): Replaces dict_lru_find_table() and
dict_non_lru_find_table().
In MySQL 5.7.8 an extra level of pointer indirection was added to
dict_operation_lock and some other rw_lock_t without solid justification,
in mysql/mysql-server@52720f1772.
Let us revert that change and remove the rather useless rw_lock_t
constructor and destructor and the magic_n field. In this way,
some unnecessary pointer dereferences and heap allocation will be avoided
and debugging might be a little easier.
This is fixing a regression that was introduced in MDEV-15562
commit 0e5a4ac253.
On rollback of UPDATE or DELETE of a ROW_FORMAT=COMPRESSED table,
page_zip_write_trx_id_and_roll_ptr() was accessing
offsets=offsets_ after offsets_ went out of scope.
InnoDB in MySQL 5.7 introduced two new parameters to the function
dict_hdr_get_new_id(), to allow redo logging to be disabled when
assigning identifiers to temporary tables or during the
backup-unfriendly TRUNCATE TABLE that was replaced in MariaDB
by MDEV-13564.
Now that MariaDB 10.4.0 removed the crash recovery code for the
backup-unfriendly TRUNCATE, we can revert dict_hdr_get_new_id()
to be used only for persistent data structures.
dict_table_assign_new_id(): Remove. This was a simple 2-line function
that was called from few places.
dict_table_open_on_id_low(): Declare in the only file where it
is called.
dict_sys_t::temp_id_hash: A separate lookup table for temporary tables.
Table names will be in the common dict_sys_t::table_hash.
dict_sys_t::get_temporary_table_id(): Assign a temporary table ID.
dict_sys_t::get_table(): Look up a persistent table.
dict_sys_t::get_temporary_table(): Look up a temporary table.
dict_sys_t::temp_table_id: The sequence of temporary table identifiers.
Starts from DICT_HDR_FIRST_ID, so that we can continue to simply compare
dict_table_t::id to a few constants for the persistent hard-coded
data dictionary tables.
undo_node_t::state: Distinguish temporary and persistent tables.
lock_check_dict_lock(), lock_get_table_id(): Assert that there cannot
be locks on temporary tables.
row_rec_to_index_entry_impl(): Assert that there cannot be metadata
records on temporary tables.
row_undo_ins_parse_undo_rec(): Distinguish temporary and persistent tables.
Move some assertions from the only caller. Return whether the table was
found.
row_undo_ins(): Add some assertions.
row_undo_mod_clust(), row_undo_mod(): Do not assign node->state.
Let row_undo() do that.
row_undo_mod_parse_undo_rec(): Distinguish temporary and persistent tables.
Move some assertions from the only caller. Return whether the table was
found.
row_undo_try_truncate(): Renamed and simplified from trx_roll_try_truncate().
row_undo_rec_get(): Replaces trx_roll_pop_top_rec_of_trx() and
trx_roll_pop_top_rec(). Fetch an undo log record, and assign undo->state
accordingly.
trx_undo_truncate_end(): Acquire the rseg->mutex only for the minimum
required duration, and release it between mini-transactions.
Allow ADD COLUMN anywhere in a table, not only adding as the
last column.
Allow instant DROP COLUMN and instant changing the order of columns.
The added columns will always be added last in clustered index records.
In new records, instantly dropped columns will be stored as NULL or
empty when possible.
Information about dropped and reordered columns will be written in
a metadata BLOB (mblob), which is stored before the first 'user' field
in the hidden metadata record at the start of the clustered index.
The presence of mblob is indicated by setting the delete-mark flag in
the metadata record.
The metadata BLOB stores the number of clustered index fields,
followed by an array of column information for each field.
For dropped columns, we store the NOT NULL flag, the fixed length,
and for variable-length columns, whether the maximum length exceeded
255 bytes. For non-dropped columns, we store the column position.
Unlike with MDEV-11369, when a table becomes empty, it cannot
be converted back to the canonical format. The reason for this is
that other threads may hold cached objects such as
row_prebuilt_t::ins_node that could refer to dropped or reordered
index fields.
For instant DROP COLUMN and ROW_FORMAT=COMPACT or ROW_FORMAT=DYNAMIC,
we must store the n_core_null_bytes in the root page, so that the
chain of node pointer records can be followed in order to reach the
leftmost leaf page where the metadata record is located.
If the mblob is present, we will zero-initialize the strings
"infimum" and "supremum" in the root page, and use the last byte of
"supremum" for storing the number of null bytes (which are allocated
but useless on node pointer pages). This is necessary for
btr_cur_instant_init_metadata() to be able to navigate to the mblob.
If the PRIMARY KEY contains any variable-length column and some
nullable columns were instantly dropped, the dict_index_t::n_nullable
in the data dictionary could be smaller than it actually is in the
non-leaf pages. Because of this, the non-leaf pages could use more
bytes for the null flags than the data dictionary expects, and we
could be reading the lengths of the variable-length columns from the
wrong offset, and thus reading the child page number from wrong place.
This is the result of two design mistakes that involve unnecessary
storage of data: First, it is nonsense to store any data fields for
the leftmost node pointer records, because the comparisons would be
resolved by the MIN_REC_FLAG alone. Second, there cannot be any null
fields in the clustered index node pointer fields, but we nevertheless
reserve space for all the null flags.
Limitations (future work):
MDEV-17459 Allow instant ALTER TABLE even if FULLTEXT INDEX exists
MDEV-17468 Avoid table rebuild on operations on generated columns
MDEV-17494 Refuse ALGORITHM=INSTANT when the row size is too large
btr_page_reorganize_low(): Preserve any metadata in the root page.
Call lock_move_reorganize_page() only after restoring the "infimum"
and "supremum" records, to avoid a memcmp() assertion failure.
dict_col_t::DROPPED: Magic value for dict_col_t::ind.
dict_col_t::clear_instant(): Renamed from dict_col_t::remove_instant().
Do not assert that the column was instantly added, because we
sometimes call this unconditionally for all columns.
Convert an instantly added column to a "core column". The old name
remove_instant() could be mistaken to refer to "instant DROP COLUMN".
dict_col_t::is_added(): Rename from dict_col_t::is_instant().
dtype_t::metadata_blob_init(): Initialize the mblob data type.
dtuple_t::is_metadata(), dtuple_t::is_alter_metadata(),
upd_t::is_metadata(), upd_t::is_alter_metadata(): Check if info_bits
refer to a metadata record.
dict_table_t::instant: Metadata about dropped or reordered columns.
dict_table_t::prepare_instant(): Prepare
ha_innobase_inplace_ctx::instant_table for instant ALTER TABLE.
innobase_instant_try() will pass this to dict_table_t::instant_column().
On rollback, dict_table_t::rollback_instant() will be called.
dict_table_t::instant_column(): Renamed from instant_add_column().
Add the parameter col_map so that columns can be reordered.
Copy and adjust v_cols[] as well.
dict_table_t::find(): Find an old column based on a new column number.
dict_table_t::serialise_columns(), dict_table_t::deserialise_columns():
Convert the mblob.
dict_index_t::instant_metadata(): Create the metadata record
for instant ALTER TABLE. Invoke dict_table_t::serialise_columns().
dict_index_t::reconstruct_fields(): Invoked by
dict_table_t::deserialise_columns().
dict_index_t::clear_instant_alter(): Move the fields for the
dropped columns to the end, and sort the surviving index fields
in ascending order of column position.
ha_innobase::check_if_supported_inplace_alter(): Do not allow
adding a FTS_DOC_ID column if a hidden FTS_DOC_ID column exists
due to FULLTEXT INDEX. (This always required ALGORITHM=COPY.)
instant_alter_column_possible(): Add a parameter for InnoDB table,
to check for additional conditions, such as the maximum number of
index fields.
ha_innobase_inplace_ctx::first_alter_pos: The first column whose position
is affected by instant ADD, DROP, or changing the order of columns.
innobase_build_col_map(): Skip added virtual columns.
prepare_inplace_add_virtual(): Correctly compute num_to_add_vcol.
Remove some unnecessary code. Note that the call to
innodb_base_col_setup() should be executed later.
commit_try_norebuild(): If ctx->is_instant(), let the virtual
columns be added or dropped by innobase_instant_try().
innobase_instant_try(): Fill in a zero default value for the
hidden column FTS_DOC_ID (to reduce the work needed in MDEV-17459).
If any columns were dropped or reordered (or added not last),
delete any SYS_COLUMNS records for the following columns, and
insert SYS_COLUMNS records for all subsequent stored columns as well
as for all virtual columns. If any virtual column is dropped, rewrite
all virtual column metadata. Use a shortcut only for adding
virtual columns. This is because innobase_drop_virtual_try()
assumes that the dropped virtual columns still exist in ctx->old_table.
innodb_update_cols(): Renamed from innodb_update_n_cols().
innobase_add_one_virtual(), innobase_insert_sys_virtual(): Change
the return type to bool, and invoke my_error() when detecting an error.
innodb_insert_sys_columns(): Insert a record into SYS_COLUMNS.
Refactored from innobase_add_one_virtual() and innobase_instant_add_col().
innobase_instant_add_col(): Replace the parameter dfield with type.
innobase_instant_drop_cols(): Drop matching columns from SYS_COLUMNS
and all columns from SYS_VIRTUAL.
innobase_add_virtual_try(), innobase_drop_virtual_try(): Let
the caller invoke innodb_update_cols().
innobase_rename_column_try(): Skip dropped columns.
commit_cache_norebuild(): Update table->fts->doc_col.
dict_mem_table_col_rename_low(): Skip dropped columns.
trx_undo_rec_get_partial_row(): Skip dropped columns.
trx_undo_update_rec_get_update(): Handle the metadata BLOB correctly.
trx_undo_page_report_modify(): Avoid out-of-bounds access to record fields.
Log metadata records consistently.
Apparently, the first fields of a clustered index may be updated
in an update_undo vector when the index is ID_IND of SYS_FOREIGN,
as part of renaming the table during ALTER TABLE. Normally, updates of
the PRIMARY KEY should be logged as delete-mark and an insert.
row_undo_mod_parse_undo_rec(), row_purge_parse_undo_rec():
Use trx_undo_metadata.
row_undo_mod_clust_low(): On metadata rollback, roll back the root page too.
row_undo_mod_clust(): Relax an assertion. The delete-mark flag was
repurposed for ALTER TABLE metadata records.
row_rec_to_index_entry_impl(): Add the template parameter mblob
and the optional parameter info_bits for specifying the desired new
info bits. For the metadata tuple, allow conversion between the original
format (ADD COLUMN only) and the generic format (with hidden BLOB).
Add the optional parameter "pad" to determine whether the tuple should
be padded to the index fields (on ALTER TABLE it should), or whether
it should remain at its original size (on rollback).
row_build_index_entry_low(): Clean up the code, removing
redundant variables and conditions. For instantly dropped columns,
generate a dummy value that is NULL, the empty string, or a
fixed length of NUL bytes, depending on the type of the dropped column.
row_upd_clust_rec_by_insert_inherit_func(): On the update of PRIMARY KEY
of a record that contained a dropped column whose value was stored
externally, we will be inserting a dummy NULL or empty string value
to the field of the dropped column. The externally stored column would
eventually be dropped when purge removes the delete-marked record for
the old PRIMARY KEY value.
btr_index_rec_validate(): Recognize the metadata record.
btr_discard_only_page_on_level(): Preserve the generic instant
ALTER TABLE metadata.
btr_set_instant(): Replaces page_set_instant(). This sets a clustered
index root page to the appropriate format, or upgrades from
the MDEV-11369 instant ADD COLUMN to generic ALTER TABLE format.
btr_cur_instant_init_low(): Read and validate the metadata BLOB page
before reconstructing the dictionary information based on it.
btr_cur_instant_init_metadata(): Do not read any lengths from the
metadata record header before reading the BLOB. At this point, we
would not actually know how many nullable fields the metadata record
contains.
btr_cur_instant_root_init(): Initialize n_core_null_bytes in one
of two possible ways.
btr_cur_trim(): Handle the mblob record.
row_metadata_to_tuple(): Convert a metadata record to a data tuple,
based on the new info_bits of the metadata record.
btr_cur_pessimistic_update(): Invoke row_metadata_to_tuple() if needed.
Invoke dtuple_convert_big_rec() for metadata records if the record is
too large, or if the mblob is not yet marked as externally stored.
btr_cur_optimistic_delete_func(), btr_cur_pessimistic_delete():
When the last user record is deleted, do not delete the
generic instant ALTER TABLE metadata record. Only delete
MDEV-11369 instant ADD COLUMN metadata records.
btr_cur_optimistic_insert(): Avoid unnecessary computation of rec_size.
btr_pcur_store_position(): Allow a logically empty page to contain
a metadata record for generic ALTER TABLE.
REC_INFO_DEFAULT_ROW_ADD: Renamed from REC_INFO_DEFAULT_ROW.
This is for the old instant ADD COLUMN (MDEV-11369) only.
REC_INFO_DEFAULT_ROW_ALTER: The more generic metadata record,
with additional information for dropped or reordered columns.
rec_info_bits_valid(): Remove. The only case when this would fail
is when the record is the generic ALTER TABLE metadata record.
rec_is_alter_metadata(): Check if a record is the metadata record
for instant ALTER TABLE (other than ADD COLUMN). NOTE: This function
must not be invoked on node pointer records, because the delete-mark
flag in those records may be set (it is garbage), and then a debug
assertion could fail because index->is_instant() does not necessarily
hold.
rec_is_add_metadata(): Check if a record is MDEV-11369 ADD COLUMN metadata
record (not more generic instant ALTER TABLE).
rec_get_converted_size_comp_prefix_low(): Assume that the metadata
field will be stored externally. In dtuple_convert_big_rec() during
the rec_get_converted_size() call, it would not be there yet.
rec_get_converted_size_comp(): Replace status,fields,n_fields with tuple.
rec_init_offsets_comp_ordinary(), rec_get_converted_size_comp_prefix_low(),
rec_convert_dtuple_to_rec_comp(): Add template<bool mblob = false>.
With mblob=true, process a record with a metadata BLOB.
rec_copy_prefix_to_buf(): Assert that no fields beyond the key and
system columns are being copied. Exclude the metadata BLOB field.
rec_convert_dtuple_to_metadata_comp(): Convert an alter metadata tuple
into a record.
row_upd_index_replace_metadata(): Apply an update vector to an
alter_metadata tuple.
row_log_allocate(): Replace dict_index_t::is_instant()
with a more appropriate condition that ignores dict_table_t::instant.
Only a table on which the MDEV-11369 ADD COLUMN was performed
can "lose its instantness" when it becomes empty. After
instant DROP COLUMN or reordering columns, we cannot simply
convert the table to the canonical format, because the data
dictionary cache and all possibly existing references to it
from other client connection threads would have to be adjusted.
row_quiesce_write_index_fields(): Do not crash when the table contains
an instantly dropped column.
Thanks to Thirunarayanan Balathandayuthapani for discussing the design
and implementing an initial prototype of this.
Thanks to Matthias Leich for testing.
On the rollback of changes to SYS_COLUMNS, MDEV-15562 will
break the assumption that the only instantaneous changes to columns
are the addition to the end of the column list.
The function dict_table_t::rollback_instant(unsigned n)
is inherently incompatible with instantly dropping or reordering
columns.
When a change to SYS_COLUMNS is rolled back, we must simply evict
the affected table definition, at the end of the rollback. We cannot
free the table object immediately, because the current transaction
that is being rolled back may be holding a lock on the table and
its metadata record.
dict_table_remove_from_cache_low(): Replaced
by dict_table_remove_from_cache().
dict_table_remove_from_cache(): Add a third parameter keep=false,
so that the table can be freed by the caller.
trx_lock_t::evicted_tables: List of tables on which trx_t::evict_table()
was invoked.
trx_t::evict_table(): Evict a table definition during rollback.
trx_commit_in_memory(): Empty the trx->lock.evicted_tables list
after the locks were released, by freeing the table objects.
row_undo_ins_remove_clust_rec(), row_undo_mod_clust_low():
Invoke trx_t::evict_table() on the affected table if a change to
SYS_COLUMNS is being rolled back.
For instant ALTER TABLE, we store a hidden metadata record at the
start of the clustered index, to indicate how the format of the
records differs from the latest table definition.
The term 'default row' is too specific, because it applies to
instant ADD COLUMN only, and we will be supporting more classes
of instant ALTER TABLE later on. For instant ADD COLUMN, we
store the initial default values in the metadata record.
This is a merge from 10.2, but the 10.2 version of this will not
be pushed into 10.2 yet, because the 10.2 version would include
backports of MDEV-14717 and MDEV-14585, which would introduce
a crash recovery regression: Tables could be lost on
table-rebuilding DDL operations, such as ALTER TABLE,
OPTIMIZE TABLE or this new backup-friendly TRUNCATE TABLE.
The test innodb.truncate_crash occasionally loses the table due to
the following bug:
MDEV-17158 log_write_up_to() sometimes fails
Implement undo tablespace truncation via normal redo logging.
Implement TRUNCATE TABLE as a combination of RENAME to #sql-ib name,
CREATE, and DROP.
Note: Orphan #sql-ib*.ibd may be left behind if MariaDB Server 10.2
is killed before the DROP operation is committed. If MariaDB Server 10.2
is killed during TRUNCATE, it is also possible that the old table
was renamed to #sql-ib*.ibd but the data dictionary will refer to the
table using the original name.
In MariaDB Server 10.3, RENAME inside InnoDB is transactional,
and #sql-* tables will be dropped on startup. So, this new TRUNCATE
will be fully crash-safe in 10.3.
ha_mroonga::wrapper_truncate(): Pass table options to the underlying
storage engine, now that ha_innobase::truncate() will need them.
rpl_slave_state::truncate_state_table(): Before truncating
mysql.gtid_slave_pos, evict any cached table handles from
the table definition cache, so that there will be no stale
references to the old table after truncating.
== TRUNCATE TABLE ==
WL#6501 in MySQL 5.7 introduced separate log files for implementing
atomic and crash-safe TRUNCATE TABLE, instead of using the InnoDB
undo and redo log. Some convoluted logic was added to the InnoDB
crash recovery, and some extra synchronization (including a redo log
checkpoint) was introduced to make this work. This synchronization
has caused performance problems and race conditions, and the extra
log files cannot be copied or applied by external backup programs.
In order to support crash-upgrade from MariaDB 10.2, we will keep
the logic for parsing and applying the extra log files, but we will
no longer generate those files in TRUNCATE TABLE.
A prerequisite for crash-safe TRUNCATE is a crash-safe RENAME TABLE
(with full redo and undo logging and proper rollback). This will
be implemented in MDEV-14717.
ha_innobase::truncate(): Invoke RENAME, create(), delete_table().
Because RENAME cannot be fully rolled back before MariaDB 10.3
due to missing undo logging, add some explicit rename-back in
case the operation fails.
ha_innobase::delete(): Introduce a variant that takes sqlcom as
a parameter. In TRUNCATE TABLE, we do not want to touch any
FOREIGN KEY constraints.
ha_innobase::create(): Add the parameters file_per_table, trx.
In TRUNCATE, the new table must be created in the same transaction
that renames the old table.
create_table_info_t::create_table_info_t(): Add the parameters
file_per_table, trx.
row_drop_table_for_mysql(): Replace a bool parameter with sqlcom.
row_drop_table_after_create_fail(): New function, wrapping
row_drop_table_for_mysql().
dict_truncate_index_tree_in_mem(), fil_truncate_tablespace(),
fil_prepare_for_truncate(), fil_reinit_space_header_for_table(),
row_truncate_table_for_mysql(), TruncateLogger,
row_truncate_prepare(), row_truncate_rollback(),
row_truncate_complete(), row_truncate_fts(),
row_truncate_update_system_tables(),
row_truncate_foreign_key_checks(), row_truncate_sanity_checks():
Remove.
row_upd_check_references_constraints(): Remove a check for
TRUNCATE, now that the table is no longer truncated in place.
The new test innodb.truncate_foreign uses DEBUG_SYNC to cover some
race-condition like scenarios. The test innodb-innodb.truncate does
not use any synchronization.
We add a redo log subformat to indicate backup-friendly format.
MariaDB 10.4 will remove support for the old TRUNCATE logging,
so crash-upgrade from old 10.2 or 10.3 to 10.4 will involve
limitations.
== Undo tablespace truncation ==
MySQL 5.7 implements undo tablespace truncation. It is only
possible when innodb_undo_tablespaces is set to at least 2.
The logging is implemented similar to the WL#6501 TRUNCATE,
that is, using separate log files and a redo log checkpoint.
We can simply implement undo tablespace truncation within
a single mini-transaction that reinitializes the undo log
tablespace file. Unfortunately, due to the redo log format
of some operations, currently, the total redo log written by
undo tablespace truncation will be more than the combined size
of the truncated undo tablespace. It should be acceptable
to have a little more than 1 megabyte of log in a single
mini-transaction. This will be fixed in MDEV-17138 in
MariaDB Server 10.4.
recv_sys_t: Add truncated_undo_spaces[] to remember for which undo
tablespaces a MLOG_FILE_CREATE2 record was seen.
namespace undo: Remove some unnecessary declarations.
fil_space_t::is_being_truncated: Document that this flag now
only applies to undo tablespaces. Remove some references.
fil_space_t::is_stopping(): Do not refer to is_being_truncated.
This check is for tablespaces of tables. Potentially used
tablespaces are never truncated any more.
buf_dblwr_process(): Suppress the out-of-bounds warning
for undo tablespaces.
fil_truncate_log(): Write a MLOG_FILE_CREATE2 with a nonzero
page number (new size of the tablespace in pages) to inform
crash recovery that the undo tablespace size has been reduced.
fil_op_write_log(): Relax assertions, so that MLOG_FILE_CREATE2
can be written for undo tablespaces (without .ibd file suffix)
for a nonzero page number.
os_file_truncate(): Add the parameter allow_shrink=false
so that undo tablespaces can actually be shrunk using this function.
fil_name_parse(): For undo tablespace truncation,
buffer MLOG_FILE_CREATE2 in truncated_undo_spaces[].
recv_read_in_area(): Avoid reading pages for which no redo log
records remain buffered, after recv_addr_trim() removed them.
trx_rseg_header_create(): Add a FIXME comment that we could write
much less redo log.
trx_undo_truncate_tablespace(): Reinitialize the undo tablespace
in a single mini-transaction, which will be flushed to the redo log
before the file size is trimmed.
recv_addr_trim(): Discard any redo logs for pages that were
logged after the new end of a file, before the truncation LSN.
If the rec_list becomes empty, reduce n_addrs. After removing
any affected records, actually truncate the file.
recv_apply_hashed_log_recs(): Invoke recv_addr_trim() right before
applying any log records. The undo tablespace files must be open
at this point.
buf_flush_or_remove_pages(), buf_flush_dirty_pages(),
buf_LRU_flush_or_remove_pages(): Add a parameter for specifying
the number of the first page to flush or remove (default 0).
trx_purge_initiate_truncate(): Remove the log checkpoints, the
extra logging, and some unnecessary crash points. Merge the code
from trx_undo_truncate_tablespace(). First, flush all to-be-discarded
pages (beyond the new end of the file), then trim the space->size
to make the page allocation deterministic. At the only remaining
crash injection point, flush the redo log, so that the recovery
can be tested.
