Lifted long standing limitation to the XA of rolling it back at the
transaction's
connection close even if the XA is prepared.
Prepared XA-transaction is made to sustain connection close or server
restart.
The patch consists of
- binary logging extension to write prepared XA part of
transaction signified with
its XID in a new XA_prepare_log_event. The concusion part -
with Commit or Rollback decision - is logged separately as
Query_log_event.
That is in the binlog the XA consists of two separate group of
events.
That makes the whole XA possibly interweaving in binlog with
other XA:s or regular transaction but with no harm to
replication and data consistency.
Gtid_log_event receives two more flags to identify which of the
two XA phases of the transaction it represents. With either flag
set also XID info is added to the event.
When binlog is ON on the server XID::formatID is
constrained to 4 bytes.
- engines are made aware of the server policy to keep up user
prepared XA:s so they (Innodb, rocksdb) don't roll them back
anymore at their disconnect methods.
- slave applier is refined to cope with two phase logged XA:s
including parallel modes of execution.
This patch does not address crash-safe logging of the new events which
is being addressed by MDEV-21469.
CORNER CASES: read-only, pure myisam, binlog-*, @@skip_log_bin, etc
Are addressed along the following policies.
1. The read-only at reconnect marks XID to fail for future
completion with ER_XA_RBROLLBACK.
2. binlog-* filtered XA when it changes engine data is regarded as
loggable even when nothing got cached for binlog. An empty
XA-prepare group is recorded. Consequent Commit-or-Rollback
succeeds in the Engine(s) as well as recorded into binlog.
3. The same applies to the non-transactional engine XA.
4. @@skip_log_bin=OFF does not record anything at XA-prepare
(obviously), but the completion event is recorded into binlog to
admit inconsistency with slave.
The following actions are taken by the patch.
At XA-prepare:
when empty binlog cache - don't do anything to binlog if RO,
otherwise write empty XA_prepare (assert(binlog-filter case)).
At Disconnect:
when Prepared && RO (=> no binlogging was done)
set Xid_cache_element::error := ER_XA_RBROLLBACK
*keep* XID in the cache, and rollback the transaction.
At XA-"complete":
Discover the error, if any don't binlog the "complete",
return the error to the user.
Kudos
-----
Alexey Botchkov took to drive this work initially.
Sergei Golubchik, Sergei Petrunja, Marko Mäkelä provided a number of
good recommendations.
Sergei Voitovich made a magnificent review and improvements to the code.
They all deserve a bunch of thanks for making this work done!
Some test are still disabled as they provide different results,
fail on different error or crash.
Fixes also MDEV-14060 InnoDB: Foreign key constraint to
geometry type does not work where test intentionally tried
incorrect cases.
namely, restart_mysqld_with_option.inc and kill_and_restart_mysqld.inc -
use restart_mysqld.inc instead.
Also remove innodb_wl6501_crash_stripped.inc that wasn't used anywhere.
This should be functionally equivalent to WL#6204 in MySQL 8.0.0, with
the notable difference that the file format changes are limited to
repurposing a previously unused data field in B-tree pages.
For persistent InnoDB tables, write the last used AUTO_INCREMENT
value to the root page of the clustered index, in the previously
unused (0) PAGE_MAX_TRX_ID field, now aliased as PAGE_ROOT_AUTO_INC.
Unlike some other previously unused InnoDB data fields, this one was
actually always zero-initialized, at least since MySQL 3.23.49.
The writes to PAGE_ROOT_AUTO_INC are protected by SX or X latch on the
root page. The SX latch will allow concurrent read access to the root
page. (The field PAGE_ROOT_AUTO_INC will only be read on the
first-time call to ha_innobase::open() from the SQL layer. The
PAGE_ROOT_AUTO_INC can only be updated when executing SQL, so
read/write races are not possible.)
During INSERT, the PAGE_ROOT_AUTO_INC is updated by the low-level
function btr_cur_search_to_nth_level(), adding no extra page
access. [Adaptive hash index lookup will be disabled during INSERT.]
If some rare UPDATE modifies an AUTO_INCREMENT column, the
PAGE_ROOT_AUTO_INC will be adjusted in a separate mini-transaction in
ha_innobase::update_row().
When a page is reorganized, we have to preserve the PAGE_ROOT_AUTO_INC
field.
During ALTER TABLE, the initial AUTO_INCREMENT value will be copied
from the table. ALGORITHM=COPY and online log apply in LOCK=NONE will
update PAGE_ROOT_AUTO_INC in real time.
innodb_col_no(): Determine the dict_table_t::cols[] element index
corresponding to a Field of a non-virtual column.
(The MySQL 5.7 implementation of virtual columns breaks the 1:1
relationship between Field::field_index and dict_table_t::cols[].
Virtual columns are omitted from dict_table_t::cols[]. Therefore,
we must translate the field_index of AUTO_INCREMENT columns into
an index of dict_table_t::cols[].)
Upgrade from old data files:
By default, the AUTO_INCREMENT sequence in old data files would appear
to be reset, because PAGE_MAX_TRX_ID or PAGE_ROOT_AUTO_INC would contain
the value 0 in each clustered index page. In new data files,
PAGE_ROOT_AUTO_INC can only be 0 if the table is empty or does not contain
any AUTO_INCREMENT column.
For backward compatibility, we use the old method of
SELECT MAX(auto_increment_column) for initializing the sequence.
btr_read_autoinc(): Read the AUTO_INCREMENT sequence from a new-format
data file.
btr_read_autoinc_with_fallback(): A variant of btr_read_autoinc()
that will resort to reading MAX(auto_increment_column) for data files
that did not use AUTO_INCREMENT yet. It was manually tested that during
the execution of innodb.autoinc_persist the compatibility logic is
not activated (for new files, PAGE_ROOT_AUTO_INC is never 0 in nonempty
clustered index root pages).
initialize_auto_increment(): Replaces
ha_innobase::innobase_initialize_autoinc(). This initializes
the AUTO_INCREMENT metadata. Only called from ha_innobase::open().
ha_innobase::info_low(): Do not try to lazily initialize
dict_table_t::autoinc. It must already have been initialized by
ha_innobase::open() or ha_innobase::create().
Note: The adjustments to class ha_innopart were not tested, because
the source code (native InnoDB partitioning) is not being compiled.
