Transaction on the slave sql thread got blocked against a slave's mysqld local ta's
lock. Since the default, slave-transaction-retries=10, there was replaying of the
replicated ta. That failed because of a new started from 5.0.13 policy not to rollback
a timed-out transaction. Effectively the first round of a timed-out ta becomes committed
by the replaying's first "BEGIN".
It was decided to backport already existed method working in 5.1 implemented in
bug #16228 for handling symmetrical deadlock problem. That patch introduced end_trans
execution whenever a replicated ta deadlocks or timed-out.
Note, that this solution can be practically suboptimal - in the light of the changed behavior
due to timeout we still could replay only the last statement - only with a high rate of timeouting
replicated transactions.
OPTIMIZE TABLE with myisam_repair_threads > 1 performs a non-quick
parallel repair. This means that it does not only rebuild all
indexes, but also the data file.
Non-quick parallel repair works so that there is one thread per
index. The first of the threads rebuilds also the new data file.
The problem was that all threads shared the read io cache on the
old data file. If there were holes (deleted records) in the table,
the first thread skipped them, writing only contiguous, non-deleted
records to the new data file. Then it built the new index so that
its entries pointed to the correct record positions. But the other
threads didn't know the new record positions, but put the positions
from the old data file into the index.
The new design is so that there is a shared io cache which is filled
by the first thread (the data file writer) with the new contiguous
records and read by the other threads. Now they know the new record
positions.
Another problem was that for the parallel repair of compressed
tables a common bit_buff and rec_buff was used. I changed it so
that thread specific buffers are used for parallel repair.
A similar problem existed for checksum calculation. I made this
multi-thread safe too.
Currently SQL_BIG_RESULT is checked only at compile time.
However, additional optimizations may take place after
this check that change the sort method from 'filesort'
to sorting via index. As a result the actual plan
executed is not the one specified by the SQL_BIG_RESULT
hint. Similarly, there is no such test when executing
EXPLAIN, resulting in incorrect output.
The patch corrects the problem by testing for
SQL_BIG_RESULT both during the explain and execution
phases.
Though this is not storage engine specific problem, I was able to
repeat this problem with BDB and NDB engines only. That was the
reason to add a test case into ndb_update.test. As a result
different bad things could happen.
BDB has removed duplicate rows which is not expected.
NDB returns an error.
For multi table update notify storage engine about UPDATE IGNORE
as it is done in single table UPDATE.
---
BUG#20265 (Replication of CREATE-SELECT does not work correctly):
Fixing bug by making binary log handle statement transactions.
The binary log transaction cache can now be truncated to remove
events inserted during this statement or transaction. Also, the
binary log participate in XA transaction handling, although not
as a full 2pc resource.
