3.8.0.txt

New Features

Dataset#each_server was added, allowing you to run the same query (most likely insert/update/delete) on all shards. This is useful if you have a sharded database but have lookup tables that should be identical on all shards. It works by yielding copies of the current dataset that are tied to each server/shard:
```
DB[:table].filter(:id=>1).each_server do |ds|
  ds.update(:name=>'foo')
end
```
Database#each_server was added, allowing you to run schema modification methods on all shards. It works by yielding a new Sequel::Database object for each shard, that will connect to only that shard:
```
DB.each_server do |db|
  db.create_table(:t){Integer :num}
end
```

You can now add and remove servers/shards from the connection pool while Sequel is running:

DB.add_servers(:shard1=>{:host=>'s1'}, :shard2=>{:host=>'s2'})
DB.remove_servers(:shard1, :shard2)

When you attempt to disconnect from a server that has connections currently in use, Sequel will now schedule those connections to be disconnected when they are returned to the pool. Previously, Sequel disconnected available connections, but ignored connections currently in use, so it wasn’t possible to guarantee complete disconnection from the server. Even with this new feature, you can only guarantee eventual disconnection, since disconnection of connections in use happens asynchronously.
Database#disconnect now accepts a :servers option specifying the server(s) from which to disconnect. This should be a symbol or array of symbols representing servers/shards. Only those specified will be disconnected:
```
DB.disconnect(:servers=>[:shard1, :shard2])
```
A validates_type validation was added to the validation_helpers plugin. It allows you to check that a given column contains the correct type. I can be helpful if you are also using the serialization plugin to store serialized ruby objects, by making sure that the objects are of the correct type (e.g. Hash):
```
def validate
  validates_type(Hash, :options)
end
```

Sequel::SQL::Expression#== is now supported for all expressions:

:column.qualify(:table).cast(:type) == \
  :column.qualify(:table).cast(:type)
# => true
:column.qualify(:table).cast(:type) == \
  :other_column.qualify(:table).cast(:type)
# => false

When using the generic File type to create blob columns on MySQL, you can specify the specific database type by using the :size option (with :tiny, :medium, and :long values recognized):
```
DB.create_table(:docs){File :body, :size=>:long} # longblob
```
The mysql adapter will now default to using mysqlplus, falling back to use mysql. mysqlplus is significantly better for threaded code because queries do not block the entire interpreter.
The JDBC adapter is now able to detect certain types of disconnect errors.
ConnectionPool.servers and Database.servers were added, which return an array of symbols specifying the servers/shards in use.

Other Improvements

The single-threaded connection pool now raises DatabaseConnectionErrors if unable to connect, so it now operates more similarly to the default connection pool.
The single-threaded connection pool now operates more similar to the default connection pool when given a nonexistent server.
PGErrors are now correctly converted to DatabaseErrors in the postgres adapter when preparing statements or executing prepared statements.
DatabaseDisconnectErrors are now raised correctly in the postgres adapter if the connection status is not OK after a query raises an error.
In the mysql adapter, multiple statements in a single query should now be handled correctly in the all cases, not just when using Dataset#each. So you can now submit multiple queries in a single string to Database#run.
Model object creation on Microsoft SQL Server 2000 once again works correctly. Previously, an optimization was used that was only supported on 2005+.
Backslashes are no longer doubled inside string literals when connecting to Microsoft SQL Server.
The ORDER clause now correctly comes after the HAVING clause on Microsoft SQL Server.
Sequel now checks that there is an active transaction before rolling back transactions on Microsoft SQL Server, since there are cases where Microsoft SQL Server will roll back transactions implicitly.
Blobs are now handled correctly when connecting to H2.
64-bit integers are now handled correctly in JDBC prepared statements.
In the boolean_readers plugin, correctly handle columns not in the db_schema, and don’t raise an error if the model’s columns can’t be determined.
In the identity_map plugin, remove instances from the cache if they are deleted or destroyed.

Backwards Compatibility

Dataset::FROM_SELF_KEEP_OPTS was merged into Dataset::NON_SQL_OPTIONS. While used in different places, they were used for the same purpose, and entries missing from one should have been included in the other.
The connection pool internals changed substantially. Now, ConnectionPool allocated and available_connections will return nil instead of an array or hash if they are called with a nonexistent server. These are generally only used internally, though they are part of the public API. created_count and size still return the size of the :default server when called with a nonexistent server, though.
The meta_eval and metaclass private methods were removed from Sequel::MetaProgramming (only the meta_def public method remains). If you want these methods, use the metaid gem.
The irregular ox->oxen pluralization rule was removed from the default inflections, as it screws up the more common box->boxes.