It appears that rollback errors are possible outside of
transactions. Since this was not something we expected to see, logging it
as an error allows us to see why this happens in production deployments.
The hangup and error handlers now have unique messages. Although the
behavior in the handlers is practically the same in both cases, the cause
of the error is not the same.
If a socket error is present, it is added to the error message. If an
error is present, it should clearly show the reason why the TCP socket was
closed.
The is_fake_event boolean helps distinguish fake events from real
ones. This makes figuring out the real source of hangup events easier.
By checking whether the users have changed whenever they are reloaded, we
improve the visibility of the user reloading process. Using a checksum
allows us to easily compress the information with acceptable loss of
accuracy. Using a CAS loop prevents duplicate messages without losing any
updates even if multiple user reloads result in different outcomes.
The use of a regular expression allows multiple rewrite rules to be
combined into one. This allows more versatile conversions but, given the
simple nature of regular expressions, also makes accidental changes more
likely.
Addd mxs::pcre2_substitute that is a more C++-friendly version of
mxs_pcre2_substitute to make. This makes string replacement a lot easier
to do when the source and destination are not C strings.
When rewrite_src and rewrite_dest have different lengths, the slave must
use GTID based replication. This removes the need for one-to-one matching
between the slave's relay log and the master's binlog which gets broken
when event lengths are modified due to event rewriting.
The replication events use a redundant format that has both the length of
the event and the position of the next event. The length can be modified
so that the next event position of the previous event and the length of
the curren event can be different. This includes overlap of the events
where the next event position of an event is "inside" the current event.
The next event position must retain its original value as that allows
replication slaves to reconnect with the correct position when file and
position based replication is used. For GTID replication, the slave asks
for the coordinates from the master and uses those.
When a slave receives a heartbeat event from a master, it checks that the
binlog name matches and that the next event position in the event is not
behind the slave's relay log position. These events must be modified to
contain a fake next event position that will never be reached by the
slave. This makes sure that the simple sanity checks never fail even if
we've caused the slave's relay log to be ahead of the master's binlog.
The parameters allow rudimentary database rewriting in the replication
stream. This is still very limited as the replacement must have the same
length as the original. In theory it could be shorter without causing
problems but making it longer is not easy.
The binlogfilter needs to read results one packet at a time but it needs
resultsets to be collected into a single buffer. This behavior is
guaranteed implicitly when the binlogrouter is used but is not present
when it is used without it. To support the use of the binlogfilter with
readconnroute, the filter must properly declare the capabilities.
The SQL for the second recursive CTE table can be optimized by adding a
where condition on the recursive part that rules out users that are not
roles. The functionality remains the same as only roles can be granted to
users.
If an existing cache-entry should be updated, but the new value
is larger that the maximum size of the cache, then the cache can
not be updated, but the old value must be removed.
Whether or not we succeed in removing the entry, an error result
must be returned. Earlier OK was returned, but no node was
allocated, which then caused a crash.
The `global` parameter causes the time window defined by the `time`
parameter to be applied at the instance level instead of the session
level. This means that a write from one connection will cause all other
connections to use the master for a certain period of time.
Using a configurable time window for consistency is not good as it is not
absolute and cannot adjust to how servers behave.
One example that demonstrates this is when a slave is normally lagging
behind by less than a second but some event causes the lag to spike up to
several seconds. In this case the configured time window would no longer
guarantee consistency.
Another reason to avoid a "static" time window is the fact taht it
prevents load balancing in the cases where slaves catch up to the master
within time window. This happens when time is configured to a higher value
to avoid inconsistencies at all costs.
Added a test case that verified the feature works.
The number of sessions wasn't always incremented but it was always
decremented. This happened primarily when authentication failed. By making
the management of the counters a part of the object lifecycle, this
problem goes away.
The number of sessions wasn't always incremented but it was always
decremented. This happened primarily when authentication failed. By making
the management of the counters a part of the object lifecycle, this
problem goes away.
The client count was incremented before authentication was complete, and
should be decremented if it fails. Otherwise service connection limit can
be easily reached.
RENAME TABLE is now fully supported and works as expected. With the fix to
table versioning, the new table name will receive the latest version
number.
The table versions are now stored in memory and are only resolved on
startup. This simplifies things and removes the need to know where the
data is stored as that information is not available to the Rpl class.
This adds preliminary support for renaming tables. There is still a
problem where the table version will always be set to 1 on a rename. This
should not be done and the version should be set to the largest value that
ever was for that table.
When lazy_connect is enabled and the first query is `SET autocommit=0`, a
master connection can be created. If it is, then the m_current_master
pointer must also be updated.
Also fixed the case where a failure to connect to one slave would cause
the connection attempts to stop too early.
TIMESTAMP2, DATETIME2 and TIME2 values with decimal parts are now
correctly converted into their string forms. Previously the decimal part
was ignored but most of the code required to extract it was in place.
The unsigned integers that would previously be interpreted as negative
values are now correctly converted into their corresponding avro
values. Due to a limitation in the Avro file format, 64-bit unsigned
integers cannot be represented in their unsigned form.
Since both the signed and unsigned versions of a 32-bit integer cannot fit
into a single Avro int, the type for these was changed to long. This is a
backwards incompatible change which means files generated with older
versions will not convert unsigned values correctly.
The unsignedness of a column is now retained in the Column type as well as
the JSON schema. This allows correct conversion of unsigned integer types
which will be done in a later commit.
