Tungsten Clustering 4.0.3 is a bugfix release that contains critical fixes and improvements to the Tungsten Clustering 4.0.2 release.
Due to an internal bug identified shortly before release, Tungsten Clustering 4.0.3 was never released to customers.
The following issues may affect the operation of Tungsten Clustering and should be taken into account when deploying or updating to this release.
Due to a bug within the Drizzle JDBC driver when communicating with MySQL, using the
optimizeRowEventsoptions could lead to significant memory usage and subsequent failure. To alleviate the problem. For more information, see Drizzle JDBC Issue 38.
A slave replicator could fail to come
ONLINE if the last THL
file is empty.
The replicator incorrectly assigns
LOAD DATA statements to
#UNKNOWN shard. This
can happen when the entire length is above 200 characters.
CSV files generated during batch loading into datawarehouses
would be created within a directory structure within the
/tmp. On long-running
replictors, automated processes that would clean up the
/tmp directory could
delete the files causing replication to fail temporarily due to
the missing directory.
The location where staging CSV files are created has now been
updated. Files are now stored within the
directory, following the same naming structure. For example, if
Tungsten Clustering has been installed in
/opt/continuent, then CSV files for the
alpha, CSV files
for the first active applier channel will be stored in
pkey filter could
force table metadata to be updated when the update was not
When using the
filter in combination with the
colnames, an issue could
arise where differences in the incoming Schema and target schema
could result in incorrect SQL statements. The solution is to
on the slave not to extract the schema information from the
database but instead to use the incoming data from the source
database and the translated THL.