- Don't connect right away in ha_cassandra::open. If we do this, it becomes 
  impossible to do SHOW CREATE TABLE when the server is not present. 
- Note: CREATE TABLE still requires that connection is present, as it needs 
  to check whether the specified DDL can be used with Cassandra.  We could
  delay that check also, but then one would not be able to find out about 
  errors in table DDL until they do a SELECT.

- Support UPDATE statements
- Follow what CQL does: don't show deleted rows (they show up as rows without any columns in reads)

- Better error messages.

- Add a test for ALTER TABLE

- Add capability to retry calls that have failed with UnavailableException or
  [Cassandra's] TimedOutException. 
- We don't retry for Thrift errors yet, although could easily do, now.

- Support mapping Cassandra's timestamp to INT64
- Support mapping Cassadnra's decimal to VARBINARY.

- allow only VARBINARY(n), all other types can get meaningless data after conversions
- more comments

- Add support for Cassandra's 'varint' datatype, mappable to VARBINARY.

- Added @@cassandra_thrift_host global variable.

- added option thrift_port which allows to specify which port to connect to
- not adding username/password - it turns out, there are no authentication
  schemes in stock cassandra distribution.

- Use more permissive locking.

- Also provide handling for generic Thrift exceptions. These are not listed in the 'throws' clause
  of API definition but still can happen.

- Catch all kinds of exceptions when calling Thrift code.

Cassandra SE: small optimization: StringCopyConverter::mariadb_to_cassandra doesn't need to make NULL-terminated strings.

- Make cassandra.test drop and re-crate the test keyspace.

- Make an attempt at fixing.

- add support for Cassandra's UUID datatype. We map it to CHAR(36).

- Add mapping for INT datatype
- Primary key column should now be named like CQL's primary key, 
  or 'rowkey' if CF has key_alias.

- Full table scan internally uses LIMIT n, and re-starts the scan from
  the last seen rowkey value.  rowkey ranges are inclusive, so we will
  see the same rowkey again. We should ignore it.

- Remove HTON_CAN_RECREATE flag, re-create won't delete rows in cassandra.

- We use HA_MRR_NO_ASSOC ("optimizer_switch=join_cache_hashed") mode
- Not able to use BKA's buffers yet.
- There is a variable to control batch size
- There are status counters.
- Nedeed to make some fixes in BKA code (to be checked with Igor)

- bulk inserts themselves
- control variable and counters.

- preparations for support of bulk INSERT.

- Fix typo in ha_cassandra::rnd_pos().
- in ::index_read_map(), do not assume that pk column is part of table->read_set.

- Partial support for DELETE ... WHERE.

- Descriptive error messages
- Unpack PK column on range scans

- Support "DELETE FROM cassandra_table"

- Got range reads to work (except for unpacking of the rowkey value)

- Introduce type converters (so far rather trivial)
- switch INSERT to using batch_mutate()