Version 2.3.0
Released on 2017/12/22.
Note
If you are upgrading a cluster, you must be running CrateDB Version 1.1.3 or higher before you upgrade to 2.3.0.
You cannot perform a rolling upgrade to this version. Any upgrade to this version will require a full restart upgrade.
Warning
Before upgrading, you should back up your data.
Changelog
Breaking Changes
Certain metrics in the
sys.nodes
table have been deprecated and now always return-1
, except network metrics with still return0
as previously when Sigar was not available.The affected metrics are: network metrics, read/write filesystem metrics for individual disks, system/user/idle/stolen values for CPU, and system/user values for process CPU.
This is due to the removal of the “Sigar” library dependency.
The default value of the setting
auth.host_based.enabled
(false
) is overwritten withtrue
in thecrate.yml
that is shipped with the tarball and Linux distributions of CrateDB and also contains a sane default configuration forauth.host_based.config
.This is done in order to provide a better default installation for new users without breaking exising HBA configurations.
An implication of these settings is that whenever Enterprise is enabled (default behaviour) connections from remote hosts always require password authentication. Note, that when running CrateDB in Docker, the host of the Docker container is also considered a remote host.
Columns aren’t implicitly cast to a type anymore. Whenever columns are compared to Literals (e.g. ‘string’, 1, 1.2), these literals will be converted to the column type but not vice-versa. The column can still be manually cast to a type by using a cast function.
Table
information_schema.table_constraints
is now returningconstraint_name
as type string instead of type array. Constraint typePRIMARY_KEY
has been changed toPRIMARY KEY
. AlsoPRIMARY KEY
constraint is not returned when not explicitly defined.Scalar functions are resolved more strictly based on the argument types. This means that built-in functions with the same name as user-defined functions will always “hide” the latter, even if the UDF has a different set of arguments. Using the same name as a built-in function for a user defined function is considered bad practice.
Changes
Added the
hyperloglog_distinct
aggregation function. This feature is only available in the Enterprise Edition.Added a
os['cpu']['used']
column to thesys.nodes
table. This replaces the deprecated system/user/idle/stolen values.Added a
cgroup
object column to thesys.nodes
table. This column contains cgroup information which is relevant when running CrateDB inside a containers.Subqueries which filter on primary key columns now have the same realtime semantics as the equivalent top-level queries.
Added support for disabling the column store for
STRING
columns on table creation and when adding columns. In conjunction with disabling indexing on that columns, this will support storing strings greater than 32kb. Be aware, that the performance of aggregations, groupings and sorting on such columns will decrease.Added support for
ORDER BY
,GROUP BY
and global aggregates on columns with disabled indexing (INDEX OFF
).Added support for scalar subqueries in
DELETE
andUPDATE
statements.Added
UNION ALL
operator to produce the combined result of two or more queries, e.g.:SELECT id FROM t1
UNION ALL
SELECT id FROM t2
Added the “password” authentication method which is available for connections via the PostgreSQL wire protocol and HTTP. This method allows clients to authenticate using a valid database user and its password. For HTTP, the
X-User
header, used to provide a username, has been deprecated in favour of the standard HTTPAuthorization
header with the Basic Authentication Scheme.Added a
WITH
clause toCREATE USER
statement to specify user properties upon creation. The single property available right now is thepassword
property which can be used for “password” authentication.The passwords of existing users can be changed using the
ALTER USER
statement.Note that user passwords are never stored in clear-text inside CrateDB!
The “address” field of the
auth.host_based.config
setting allows the special_local_
identifier additionally to IP and CIDR notation._local_
matches both IPv4 and IPv6 connections from localhost.Table
information_schema.key_column_usage
is now populated with primary key information of user generated tables.Table
information_schema.table_constraints
is now also returning theNOT_NULL
constraint.Added new cluster setting
routing.rebalance.enable
that allows to enable or disable shard rebalancing on the cluster.Added support to manually control the allocation of shards using
ALTER TABLE REROUTE
. Supported reroute-options are:MOVE
,ALLOCATE REPLICA
, andCANCEL
.Added support to manually retry the allocation of shards that failed to allocate using
ALTER CLUSTER REROUTE RETRY FAILED
.Added new table setting
allocation.max_retries
that defines the number of attempts to allocate a shard before giving up and leaving it unallocated.Added new system table
sys.allocations
which lists shards and their allocation state including the reasoning why they are in a certain state.Function arguments are now linked to each other, where possible. This enables type inference between arguments such that arguments can be converted to match a function’s signature. For example,
coalesce(integer, long)
would have resulted in an “unknown function” message. We now convert this call intocoalesce(long, long)
. The conversion is possible through a type precedence list and convertibility checks on the data types.Functions which accept regular expression flags now throw an error when invalid flags are provided.
Clients using the PostgreSQL wire protocol will now receive an additional
crate_version
ParameterStatus message when establishing a connection. This can be used to identify the server asCrateDB
.Added the
typtype
column topg_catalog.pg_type
for better compatibility with certain PostgreSQL client libraries.Added the
pg_backend_pid()
function for enhanced PostgreSQL compatibility.Added support for the PSQL ParameterDescription message which allows to get the parameter types in prepared statements up front without specifying the actual arguments first. This fixes compatibility issues with some drivers. This works for the most common use cases except for DDL statements.
Upgraded Elasticsearch to version 5.6.3.
Updated the CrateDB command line shell (Crash) to version 0.23.0, which added support for password authentication and pasting multiple statements at once.
Update the Admin UI to use new CPU metrics for its graphs.
Hadoop2 dependencies for the HDFS repository plugin have been upgraded to version 2.8.1.