Hive connector#
The Hive connector allows querying data stored in an Apache Hive data warehouse. Hive is a combination of three components:
Data files in varying formats, that are typically stored in the Hadoop Distributed File System (HDFS) or in object storage systems such as Amazon S3.
Metadata about how the data files are mapped to schemas and tables. This metadata is stored in a database, such as MySQL, and is accessed via the Hive metastore service.
A query language called HiveQL. This query language is executed on a distributed computing framework such as MapReduce or Tez.
Trino only uses the first two components: the data and the metadata. It does not use HiveQL or any part of Hive’s execution environment.
Requirements#
The Hive connector requires a Hive metastore service (HMS), or a compatible implementation of the Hive metastore, such as AWS Glue Data Catalog.
Apache Hadoop HDFS 2.x and 3.x are supported.
Many distributed storage systems including HDFS, Amazon S3 or S3-compatible systems, Google Cloud Storage, Azure Storage, and IBM Cloud Object Storage can be queried with the Hive connector.
The coordinator and all workers must have network access to the Hive metastore and the storage system. Hive metastore access with the Thrift protocol defaults to using port 9083.
General configuration#
Create etc/catalog/hive.properties with the following contents
to mount the hive connector as the hive catalog,
replacing example.net:9083 with the correct host and port
for your Hive metastore Thrift service:
connector.name=hive
hive.metastore.uri=thrift://example.net:9083
Multiple Hive clusters#
You can have as many catalogs as you need, so if you have additional
Hive clusters, simply add another properties file to etc/catalog
with a different name, making sure it ends in .properties. For
example, if you name the property file sales.properties, Trino
creates a catalog named sales using the configured connector.
HDFS configuration#
For basic setups, Trino configures the HDFS client automatically and
does not require any configuration files. In some cases, such as when using
federated HDFS or NameNode high availability, it is necessary to specify
additional HDFS client options in order to access your HDFS cluster. To do so,
add the hive.config.resources property to reference your HDFS config files:
hive.config.resources=/etc/hadoop/conf/core-site.xml,/etc/hadoop/conf/hdfs-site.xml
Only specify additional configuration files if necessary for your setup. We recommend reducing the configuration files to have the minimum set of required properties, as additional properties may cause problems.
The configuration files must exist on all Trino nodes. If you are referencing existing Hadoop config files, make sure to copy them to any Trino nodes that are not running Hadoop.
HDFS username and permissions#
Before running any CREATE TABLE or CREATE TABLE AS statements
for Hive tables in Trino, you must check that the user Trino is
using to access HDFS has access to the Hive warehouse directory. The Hive
warehouse directory is specified by the configuration variable
hive.metastore.warehouse.dir in hive-site.xml, and the default
value is /user/hive/warehouse.
When not using Kerberos with HDFS, Trino accesses HDFS using the
OS user of the Trino process. For example, if Trino is running as
nobody, it accesses HDFS as nobody. You can override this
username by setting the HADOOP_USER_NAME system property in the
Trino JVM config, replacing hdfs_user with the
appropriate username:
-DHADOOP_USER_NAME=hdfs_user
The hive user generally works, since Hive is often started with
the hive user and this user has access to the Hive warehouse.
Whenever you change the user Trino is using to access HDFS, remove
/tmp/presto-* on HDFS, as the new user may not have access to
the existing temporary directories.
Hive general configuration properties#
The following table lists general configuration properties for the Hive connector. There are additional sets of configuration properties throughout the Hive connector documentation.
Property Name |
Description |
Default |
|---|---|---|
|
An optional comma-separated list of HDFS configuration files. These
files must exist on the machines running Trino. Only specify this if
absolutely necessary to access HDFS. Example: |
|
|
Enable reading data from subdirectories of table or partition locations.
If disabled, subdirectories are ignored. This is equivalent to the
|
|
|
Ignore partitions when the file system location does not exist rather than failing the query. This skips data that may be expected to be part of the table. |
|
|
The default file format used when creating new tables. |
|
|
The compression codec to use when writing files. Possible values are
|
|
|
Force splits to be scheduled on the same node as the Hadoop DataNode process serving the split data. This is useful for installations where Trino is collocated with every DataNode. |
|
|
Should new partitions be written using the existing table format or the default Trino format? |
|
|
Can new data be inserted into existing partitions? If |
|
|
What happens when data is inserted into an existing partition? Possible values are
|
|
|
Best effort maximum size of new files. |
|
|
Should empty files be created for buckets that have no data? |
|
|
Specifies the number of partitions to analyze when computing table statistics. |
100 |
|
Maximum number of partitions per writer. |
100 |
|
The maximum number of partitions for a single table scan to load eagerly on the coordinator. Certain optimizations are not possible without eager loading. |
100,000 |
|
Maximum number of partitions for a single table scan. |
1,000,000 |
|
Hadoop file system replication factor. |
|
|
||
|
Path of config file to use when |
|
|
Enable writes to non-managed (external) Hive tables. |
|
|
Enable creating non-managed (external) Hive tables. |
|
|
Enables automatic column level statistics collection on write. See Table Statistics for details. |
|
|
Enable query pushdown to AWS S3 Select service. |
|
|
Maximum number of simultaneously open connections to S3 for S3 Select pushdown. |
500 |
|
Cache directory listing for specific tables. Examples:
|
|
|
Maximum total number of cached file status entries. |
1,000,000 |
|
How long a cached directory listing is considered valid. |
|
|
Adjusts binary encoded timestamp values to a specific time zone. For Hive 3.1+, this must be set to UTC. |
JVM default |
|
Specifies the precision to use for Hive columns of type |
|
|
Controls whether the temporary staging directory configured at
|
|
|
Controls the location of temporary staging directory that is used for
write operations. The |
|
|
Enable translation for Hive views. |
|
|
Use the legacy algorithm to translate Hive views.
You can use the |
|
|
Improve parallelism of partitioned and bucketed table writes. When disabled, the number of writing threads is limited to number of buckets. |
|
|
Controls the permissions set on new directories created for tables. It must be either ‘skip’ or an octal number, with a leading 0. If set to ‘skip’, permissions of newly created directories will not be set by Trino. |
|
|
Maximum number of cached file system objects. |
1000 |
|
Set to |
|
|
Enables Table statistics. The equivalent
catalog session property is
|
|
|
Set the default value for the auto_purge table property for managed tables. See the Table properties for more information on auto_purge. |
|
|
Enables Athena partition projection support |
|
|
Maximum number of partitions to drop in a single query. |
100,000 |
Metastores#
The Hive connector supports the use of the Hive Metastore Service (HMS) and AWS Glue data catalog.
Additionally, accessing tables with Athena partition projection metadata, as well as first class support for Avro tables, are available with additional configuration.
General metastore configuration properties#
The required Hive metastore can be configured with a number of properties. Specific properties can be used to further configure the Thrift or Glue metastore.
Property Name |
Description |
Default |
|---|---|---|
|
The type of Hive metastore to use. Trino currently supports the default
Hive Thrift metastore ( |
|
|
Enable caching for partition metadata. You can disable caching to avoid inconsistent behavior that results from it. |
|
|
Duration of how long cached metastore data is considered valid. |
|
|
Maximum number of metastore data objects in the Hive metastore cache. |
|
|
Asynchronously refresh cached metastore data after access if it is older than this but is not yet expired, allowing subsequent accesses to see fresh data. |
|
|
Maximum threads used to refresh cached metastore data. |
|
|
Timeout for Hive metastore requests. |
|
|
Controls whether to hide Delta Lake tables in table listings. Currently applies only when using the AWS Glue metastore. |
|
Thrift metastore configuration properties#
In order to use a Hive Thrisft metastore, you must configure the metastore with
hive.metastore=thrift and provide further details with the following
properties:
Property Name |
Description |
|---|---|
|
The URIs of the Hive metastore to connect to using the Thrift protocol.
If a comma-separated list of URIs is provided, the first URI is used by
default, and the rest of the URIs are fallback metastores. This property
is required. Example: |
|
The username Trino uses to access the Hive metastore. |
|
Hive metastore authentication type. Possible values are |
|
Enable Hive metastore end user impersonation. |
|
Enable usage of table statistics generated by Apache Spark when hive table statistics are not available |
|
Time to live delegation token cache for metastore. Default is |
|
Delegation token cache maximum size. Default is |
|
Use SSL when connecting to metastore. Default is |
|
Path to private key and client certification (key store). |
|
Password for the private key. |
|
Path to the server certificate chain (trust store). Required when SSL is enabled. |
|
Password for the trust store. |
|
The Kerberos principal of the Hive metastore service. |
|
The Kerberos principal that Trino uses when connecting to the Hive metastore service. |
|
Hive metastore client keytab location. |
|
Actively delete the files for drop table or partition operations, for cases when the
metastore does not delete the files. Default is |
AWS Glue catalog configuration properties#
In order to use a Glue catalog, you must configure the metastore with
hive.metastore=glue and provide further details with the following
properties:
Property Name |
Description |
Default |
|---|---|---|
|
AWS region of the Glue Catalog. This is required when not running in
EC2, or when the catalog is in a different region. Example:
|
|
|
Glue API endpoint URL (optional). Example:
|
|
|
AWS region of the STS service to authenticate with. This is required
when running in a GovCloud region. Example: |
|
|
The ID of the Glue Proxy API, when accessing Glue via an VPC endpoint in API Gateway. |
|
|
STS endpoint URL to use when authenticating to Glue (optional). Example:
|
|
|
Pin Glue requests to the same region as the EC2 instance where Trino is running. |
|
|
Max number of concurrent connections to Glue. |
|
|
Maximum number of error retries for the Glue client. |
|
|
Default warehouse directory for schemas created without an explicit
|
|
|
Fully qualified name of the Java class to use for obtaining AWS credentials. Can be used to supply a custom credentials provider. |
|
|
AWS access key to use to connect to the Glue Catalog. If specified along
with |
|
|
AWS secret key to use to connect to the Glue Catalog. If specified along
with |
|
|
The ID of the Glue Catalog in which the metadata database resides. |
|
|
ARN of an IAM role to assume when connecting to the Glue Catalog. |
|
|
External ID for the IAM role trust policy when connecting to the Glue Catalog. |
|
|
Number of segments for partitioned Glue tables. |
|
|
Number of threads for parallel partition fetches from Glue. |
|
|
Number of threads for parallel statistic fetches from Glue. |
|
|
Number of threads for parallel statistic writes to Glue. |
|
Accessing tables with Athena partition projection metadata#
Partition projection is a feature of AWS Athena often used to speed up query processing with highly partitioned tables.
Trino supports partition projection table properties stored in the metastore,
and it reimplements this functionality. Currently, there is a limitation in
comparison to AWS Athena for date projection, as it only supports intervals of
DAYS, HOURS, MINUTES, and SECONDS.
If there are any compatibility issues blocking access to a requested table when
you have partition projection enabled, you can set the
partition_projection_ignore table property to true for a table to bypass
any errors.
Refer to Table properties and Column properties for configuration of partition projection.
Metastore configuration for Avro#
In order to enable first-class support for Avro tables when using
Hive 3.x, you must add the following property definition to the Hive metastore
configuration file hive-site.xml and restart the metastore service:
<property>
<!-- https://community.hortonworks.com/content/supportkb/247055/errorjavalangunsupportedoperationexception-storage.html -->
<name>metastore.storage.schema.reader.impl</name>
<value>org.apache.hadoop.hive.metastore.SerDeStorageSchemaReader</value>
</property>
Storage#
The Hive connector supports the following storage options:
Google Cloud Storage
The Hive connector also supports storage caching.
Google Cloud Storage configuration#
The Hive connector can access data stored in GCS, using the gs:// URI prefix.
Please refer to the Hive connector GCS tutorial for step-by-step instructions.
GCS configuration properties#
Property Name |
Description |
|---|---|
|
JSON key file used to authenticate with Google Cloud Storage. |
|
Use client-provided OAuth token to access Google Cloud Storage. This is mutually exclusive with a global JSON key file. |
Security#
Please see the Hive connector security configuration section for information on the security options available for the Hive connector.
SQL support#
The connector provides read access and write access to data and metadata in the configured object storage system and metadata stores:
Globally available statements; see also Globally available statements
Security operations: see also SQL standard-based authorization for object storage
Refer to the migration guide for practical advice on migrating from Hive to Trino.
The following sections provide Hive-specific information regarding SQL support.
Basic usage examples#
The examples shown here work on Google Cloud Storage by replacing s3:// with
gs://.
Create a new Hive table named page_views in the web schema
that is stored using the ORC file format, partitioned by date and
country, and bucketed by user into 50 buckets. Note that Hive
requires the partition columns to be the last columns in the table:
CREATE TABLE hive.web.page_views (
view_time timestamp,
user_id bigint,
page_url varchar,
ds date,
country varchar
)
WITH (
format = 'ORC',
partitioned_by = ARRAY['ds', 'country'],
bucketed_by = ARRAY['user_id'],
bucket_count = 50
)
Create a new Hive schema named web that stores tables in an
S3 bucket named my-bucket:
CREATE SCHEMA hive.web
WITH (location = 's3://my-bucket/')
Drop a schema:
DROP SCHEMA hive.web
Drop a partition from the page_views table:
DELETE FROM hive.web.page_views
WHERE ds = DATE '2016-08-09'
AND country = 'US'
Query the page_views table:
SELECT * FROM hive.web.page_views
List the partitions of the page_views table:
SELECT * FROM hive.web."page_views$partitions"
Create an external Hive table named request_logs that points at
existing data in S3:
CREATE TABLE hive.web.request_logs (
request_time timestamp,
url varchar,
ip varchar,
user_agent varchar
)
WITH (
format = 'TEXTFILE',
external_location = 's3://my-bucket/data/logs/'
)
Collect statistics for the request_logs table:
ANALYZE hive.web.request_logs;
Drop the external table request_logs. This only drops the metadata
for the table. The referenced data directory is not deleted:
DROP TABLE hive.web.request_logs
CREATE TABLE AS can be used to create transactional tables in ORC format like this:
CREATE TABLE <name> WITH ( format='ORC', transactional=true ) AS <query>
Add an empty partition to the page_views table:
CALL system.create_empty_partition(
schema_name => 'web',
table_name => 'page_views',
partition_columns => ARRAY['ds', 'country'],
partition_values => ARRAY['2016-08-09', 'US']);
Drop stats for a partition of the page_views table:
CALL system.drop_stats(
schema_name => 'web',
table_name => 'page_views',
partition_values => ARRAY['2016-08-09', 'US']);
Procedures#
Use the CALL statement to perform data manipulation or
administrative tasks. Procedures must include a qualified catalog name, if your
Hive catalog is called web:
CALL web.system.example_procedure()
The following procedures are available:
system.create_empty_partition(schema_name, table_name, partition_columns, partition_values)Create an empty partition in the specified table.
system.sync_partition_metadata(schema_name, table_name, mode, case_sensitive)Check and update partitions list in metastore. There are three modes available:
ADD: add any partitions that exist on the file system, but not in the metastore.DROP: drop any partitions that exist in the metastore, but not on the file system.FULL: perform bothADDandDROP.
The
case_sensitiveargument is optional. The default value istruefor compatibility with Hive’sMSCK REPAIR TABLEbehavior, which expects the partition column names in file system paths to use lowercase (e.g.col_x=SomeValue). Partitions on the file system not conforming to this convention are ignored, unless the argument is set tofalse.system.drop_stats(schema_name, table_name, partition_values)Drops statistics for a subset of partitions or the entire table. The partitions are specified as an array whose elements are arrays of partition values (similar to the
partition_valuesargument increate_empty_partition). Ifpartition_valuesargument is omitted, stats are dropped for the entire table.
system.register_partition(schema_name, table_name, partition_columns, partition_values, location)Registers existing location as a new partition in the metastore for the specified table.
When the
locationargument is omitted, the partition location is constructed usingpartition_columnsandpartition_values.Due to security reasons, the procedure is enabled only when
hive.allow-register-partition-procedureis set totrue.
system.unregister_partition(schema_name, table_name, partition_columns, partition_values)Unregisters given, existing partition in the metastore for the specified table. The partition data is not deleted.
system.flush_metadata_cache()Flush all Hive metadata caches.
system.flush_metadata_cache(schema_name => ..., table_name => ...)Flush Hive metadata caches entries connected with selected table. Procedure requires named parameters to be passed
system.flush_metadata_cache(schema_name => ..., table_name => ..., partition_columns => ARRAY[...], partition_values => ARRAY[...])Flush Hive metadata cache entries connected with selected partition. Procedure requires named parameters to be passed.
Data management#
Some data management statements may be affected by
the Hive catalog’s authorization check policy. In the default legacy policy,
some statements are disabled by default. See Hive connector security configuration for more
information.
The Data management functionality includes support for INSERT,
UPDATE, DELETE, and MERGE statements, with the exact support
depending on the storage system, file format, and metastore.
When connecting to a Hive metastore version 3.x, the Hive connector supports reading from and writing to insert-only and ACID tables, with full support for partitioning and bucketing.
DELETE applied to non-transactional tables is only supported if the
table is partitioned and the WHERE clause matches entire partitions.
Transactional Hive tables with ORC format support “row-by-row” deletion, in
which the WHERE clause may match arbitrary sets of rows.
UPDATE is only supported for transactional Hive tables with format
ORC. UPDATE of partition or bucket columns is not supported.
MERGE is only supported for ACID tables.
ACID tables created with Hive Streaming Ingest are not supported.
Schema and table management#
The Hive connector supports querying and manipulating Hive tables and schemas (databases). While some uncommon operations must be performed using Hive directly, most operations can be performed using Trino.
Schema evolution#
Hive allows the partitions in a table to have a different schema than the table. This occurs when the column types of a table are changed after partitions already exist (that use the original column types). The Hive connector supports this by allowing the same conversions as Hive:
varcharto and fromtinyint,smallint,integerandbigintrealtodoubleWidening conversions for integers, such as
tinyinttosmallint
Any conversion failure results in null, which is the same behavior
as Hive. For example, converting the string 'foo' to a number,
or converting the string '1234' to a tinyint (which has a
maximum value of 127).
Avro schema evolution#
Trino supports querying and manipulating Hive tables with the Avro storage format, which has the schema set based on an Avro schema file/literal. Trino is also capable of creating the tables in Trino by infering the schema from a valid Avro schema file located locally, or remotely in HDFS/Web server.
To specify that the Avro schema must be used for interpreting table’s data one must use avro_schema_url table property.
The schema can be placed remotely in
HDFS (e.g. avro_schema_url = 'hdfs://user/avro/schema/avro_data.avsc'),
S3 (e.g. avro_schema_url = 's3n:///schema_bucket/schema/avro_data.avsc'),
a web server (e.g. avro_schema_url = 'http://example.org/schema/avro_data.avsc')
as well as local file system. This URL, where the schema is located, must be accessible from the
Hive metastore and Trino coordinator/worker nodes.
The table created in Trino using avro_schema_url behaves the same way as a Hive table with avro.schema.url or avro.schema.literal set.
Example:
CREATE TABLE hive.avro.avro_data (
id bigint
)
WITH (
format = 'AVRO',
avro_schema_url = '/usr/local/avro_data.avsc'
)
The columns listed in the DDL (id in the above example) is ignored if avro_schema_url is specified.
The table schema matches the schema in the Avro schema file. Before any read operation, the Avro schema is
accessed so the query result reflects any changes in schema. Thus Trino takes advantage of Avro’s backward compatibility abilities.
If the schema of the table changes in the Avro schema file, the new schema can still be used to read old data. Newly added/renamed fields must have a default value in the Avro schema file.
The schema evolution behavior is as follows:
Column added in new schema: Data created with an older schema produces a default value when table is using the new schema.
Column removed in new schema: Data created with an older schema no longer outputs the data from the column that was removed.
Column is renamed in the new schema: This is equivalent to removing the column and adding a new one, and data created with an older schema produces a default value when table is using the new schema.
Changing type of column in the new schema: If the type coercion is supported by Avro or the Hive connector, then the conversion happens. An error is thrown for incompatible types.
Limitations#
The following operations are not supported when avro_schema_url is set:
CREATE TABLE ASis not supported.Bucketing(
bucketed_by) columns are not supported inCREATE TABLE.ALTER TABLEcommands modifying columns are not supported.
ALTER TABLE EXECUTE#
The connector supports the optimize command for use with
ALTER TABLE EXECUTE.
The optimize command is used for rewriting the content
of the specified non-transactional table so that it is merged
into fewer but larger files.
In case that the table is partitioned, the data compaction
acts separately on each partition selected for optimization.
This operation improves read performance.
All files with a size below the optional file_size_threshold
parameter (default value for the threshold is 100MB) are
merged:
ALTER TABLE test_table EXECUTE optimize
The following statement merges files in a table that are under 10 megabytes in size:
ALTER TABLE test_table EXECUTE optimize(file_size_threshold => '10MB')
You can use a WHERE clause with the columns used to partition the table,
to filter which partitions are optimized:
ALTER TABLE test_partitioned_table EXECUTE optimize
WHERE partition_key = 1
The optimize command is disabled by default, and can be enabled for a
catalog with the <catalog-name>.non_transactional_optimize_enabled
session property:
SET SESSION <catalog_name>.non_transactional_optimize_enabled=true
Warning
Because Hive tables are non-transactional, take note of the following possible outcomes:
If queries are run against tables that are currently being optimized, duplicate rows may be read.
In rare cases where exceptions occur during the
optimizeoperation, a manual cleanup of the table directory is needed. In this situation, refer to the Trino logs and query failure messages to see which files must be deleted.
Table properties#
Table properties supply or set metadata for the underlying tables. This is key for CREATE TABLE AS statements. Table properties are passed to the connector using a WITH clause:
CREATE TABLE tablename
WITH (format='CSV',
csv_escape = '"')
Property name |
Description |
Default |
|---|---|---|
|
Indicates to the configured metastore to perform a purge when a table or partition is deleted instead of a soft deletion using the trash. |
|
|
The URI pointing to Avro schema evolution for the table. |
|
|
The number of buckets to group data into. Only valid if used with
|
0 |
|
The bucketing column for the storage table. Only valid if used with
|
|
|
Specifies which Hive bucketing version to use. Valid values are |
|
|
The CSV escape character. Requires CSV format. |
|
|
The CSV quote character. Requires CSV format. |
|
|
The CSV separator character. Requires CSV format. You can use other
separators such as |
|
|
The URI for an external Hive table on S3, Azure Blob Storage, etc. See the Basic usage examples for more information. |
|
|
The table file format. Valid values include |
|
|
The serialization format for |
|
|
Comma separated list of columns to use for ORC bloom filter. It improves the performance of queries using range predicates when reading ORC files. Requires ORC format. |
|
|
The ORC bloom filters false positive probability. Requires ORC format. |
0.05 |
|
The partitioning column for the storage table. The columns listed in the
|
|
|
The number of footer lines to ignore when parsing the file for data. Requires TextFile or CSV format tables. |
|
|
The number of header lines to ignore when parsing the file for data. Requires TextFile or CSV format tables. |
|
|
The column to sort by to determine bucketing for row. Only valid if
|
|
|
Allows the use of custom field separators, such as ‘|’, for TextFile formatted tables. |
|
|
Allows the use of a custom escape character for TextFile formatted tables. |
|
|
Set this property to |
|
|
Enables partition projection for selected table. Mapped from AWS Athena table property projection.enabled. |
|
|
Ignore any partition projection properties stored in the metastore for the selected table. This is a Trino-only property which allows you to work around compatibility issues on a specific table, and if enabled, Trino ignores all other configuration options related to partition projection. |
|
|
Projected partition location template, such as
|
|
Metadata tables#
The raw Hive table properties are available as a hidden table, containing a
separate column per table property, with a single row containing the property
values. The properties table name is the same as the table name with
$properties appended.
You can inspect the property names and values with a simple query:
SELECT * FROM hive.web."page_views$properties";
Column properties#
Property name |
Description |
Default |
|---|---|---|
|
Defines the type of partition projection to use on this column.
May be used only on partition columns. Available types:
|
|
|
Used with |
|
|
Used with |
|
|
Used with |
|
|
Used with |
|
|
Used with |
|
|
Used with |
Metadata columns#
In addition to the defined columns, the Hive connector automatically exposes metadata in a number of hidden columns in each table:
$bucket: Bucket number for this row$path: Full file system path name of the file for this row$file_modified_time: Date and time of the last modification of the file for this row$file_size: Size of the file for this row$partition: Partition name for this row
You can use these columns in your SQL statements like any other column. They can be selected directly, or used in conditional statements. For example, you can inspect the file size, location and partition for each record:
SELECT *, "$path", "$file_size", "$partition"
FROM hive.web.page_views;
Retrieve all records that belong to files stored in the partition
ds=2016-08-09/country=US:
SELECT *, "$path", "$file_size"
FROM hive.web.page_views
WHERE "$partition" = 'ds=2016-08-09/country=US'
View management#
Trino allows reading from Hive materialized views, and can be configured to support reading Hive views.
Materialized views#
The Hive connector supports reading from Hive materialized views. In Trino, these views are presented as regular, read-only tables.
Hive views#
Hive views are defined in HiveQL and stored in the Hive Metastore Service. They are analyzed to allow read access to the data.
The Hive connector includes support for reading Hive views with three different modes.
Disabled
Legacy
Experimental
You can configure the behavior in your catalog properties file.
By default, Hive views are executed with the RUN AS DEFINER security mode.
Set the hive.hive-views.run-as-invoker catalog configuration property to
true to use RUN AS INVOKER semantics.
Disabled
The default behavior is to ignore Hive views. This means that your business logic and data encoded in the views is not available in Trino.
Legacy
A very simple implementation to execute Hive views, and therefore allow read
access to the data in Trino, can be enabled with
hive.hive-views.enabled=true and
hive.hive-views.legacy-translation=true.
For temporary usage of the legacy behavior for a specific catalog, you can set
the hive_views_legacy_translation catalog session property to true.
This legacy behavior interprets any HiveQL query that defines a view as if it is written in SQL. It does not do any translation, but instead relies on the fact that HiveQL is very similar to SQL.
This works for very simple Hive views, but can lead to problems for more complex queries. For example, if a HiveQL function has an identical signature but different behaviors to the SQL version, the returned results may differ. In more extreme cases the queries might fail, or not even be able to be parsed and executed.
Experimental
The new behavior is better engineered and has the potential to become a lot more powerful than the legacy implementation. It can analyze, process, and rewrite Hive views and contained expressions and statements.
It supports the following Hive view functionality:
UNION [DISTINCT]andUNION ALLagainst Hive viewsNested
GROUP BYclausescurrent_user()LATERAL VIEW OUTER EXPLODELATERAL VIEW [OUTER] EXPLODEon array of structLATERAL VIEW json_tuple
You can enable the experimental behavior with
hive.hive-views.enabled=true. Remove the
hive.hive-views.legacy-translation property or set it to false to make
sure legacy is not enabled.
Keep in mind that numerous features are not yet implemented when experimenting with this feature. The following is an incomplete list of missing functionality:
HiveQL
current_date,current_timestamp, and othersHive function calls including
translate(), window functions, and othersCommon table expressions and simple case expressions
Honor timestamp precision setting
Support all Hive data types and correct mapping to Trino types
Ability to process custom UDFs
Performance#
The connector includes a number of performance improvements, detailed in the following sections.
Table statistics#
The Hive connector supports collecting and managing table statistics to improve query processing performance.
When writing data, the Hive connector always collects basic statistics
(numFiles, numRows, rawDataSize, totalSize)
and by default will also collect column level statistics:
Column type |
Collectible statistics |
|---|---|
|
Number of nulls, number of distinct values, min/max values |
|
Number of nulls, number of distinct values, min/max values |
|
Number of nulls, number of distinct values, min/max values |
|
Number of nulls, number of distinct values, min/max values |
|
Number of nulls, number of distinct values, min/max values |
|
Number of nulls, number of distinct values, min/max values |
|
Number of nulls, number of distinct values, min/max values |
|
Number of nulls, number of distinct values, min/max values |
|
Number of nulls, number of distinct values, min/max values |
|
Number of nulls, number of distinct values |
|
Number of nulls, number of distinct values |
|
Number of nulls |
|
Number of nulls, number of true/false values |
Updating table and partition statistics#
If your queries are complex and include joining large data sets, running ANALYZE on tables/partitions may improve query performance by collecting statistical information about the data.
When analyzing a partitioned table, the partitions to analyze can be specified
via the optional partitions property, which is an array containing
the values of the partition keys in the order they are declared in the table schema:
ANALYZE table_name WITH (
partitions = ARRAY[
ARRAY['p1_value1', 'p1_value2'],
ARRAY['p2_value1', 'p2_value2']])
This query will collect statistics for two partitions with keys
p1_value1, p1_value2 and p2_value1, p2_value2.
On wide tables, collecting statistics for all columns can be expensive and can have a
detrimental effect on query planning. It is also typically unnecessary - statistics are
only useful on specific columns, like join keys, predicates, grouping keys. One can
specify a subset of columns to be analyzed via the optional columns property:
ANALYZE table_name WITH (
partitions = ARRAY[ARRAY['p2_value1', 'p2_value2']],
columns = ARRAY['col_1', 'col_2'])
This query collects statistics for columns col_1 and col_2 for the partition
with keys p2_value1, p2_value2.
Note that if statistics were previously collected for all columns, they must be dropped before re-analyzing just a subset:
CALL system.drop_stats('schema_name', 'table_name')
You can also drop statistics for selected partitions only:
CALL system.drop_stats(
schema_name => 'schema',
table_name => 'table',
partition_values => ARRAY[ARRAY['p2_value1', 'p2_value2']])
Dynamic filtering#
The Hive connector supports the dynamic filtering optimization. Dynamic partition pruning is supported for partitioned tables stored in any file format for broadcast as well as partitioned joins. Dynamic bucket pruning is supported for bucketed tables stored in any file format for broadcast joins only.
For tables stored in ORC or Parquet file format, dynamic filters are also pushed into local table scan on worker nodes for broadcast joins. Dynamic filter predicates pushed into the ORC and Parquet readers are used to perform stripe or row-group pruning and save on disk I/O. Sorting the data within ORC or Parquet files by the columns used in join criteria significantly improves the effectiveness of stripe or row-group pruning. This is because grouping similar data within the same stripe or row-group greatly improves the selectivity of the min/max indexes maintained at stripe or row-group level.
Delaying execution for dynamic filters#
It can often be beneficial to wait for the collection of dynamic filters before starting a table scan. This extra wait time can potentially result in significant overall savings in query and CPU time, if dynamic filtering is able to reduce the amount of scanned data.
For the Hive connector, a table scan can be delayed for a configured amount of
time until the collection of dynamic filters by using the configuration property
hive.dynamic-filtering.wait-timeout in the catalog file or the catalog
session property <hive-catalog>.dynamic_filtering_wait_timeout.
Table redirection#
Trino offers the possibility to transparently redirect operations on an existing table to the appropriate catalog based on the format of the table and catalog configuration.
In the context of connectors which depend on a metastore service (for example, Hive connector, Iceberg connector and Delta Lake connector), the metastore (Hive metastore service, AWS Glue Data Catalog) can be used to accustom tables with different table formats. Therefore, a metastore database can hold a variety of tables with different table formats.
As a concrete example, let’s use the following simple scenario which makes use of table redirection:
USE example.example_schema;
EXPLAIN SELECT * FROM example_table;
Query Plan
-------------------------------------------------------------------------
Fragment 0 [SOURCE]
...
Output[columnNames = [...]]
│ ...
└─ TableScan[table = another_catalog:example_schema:example_table]
...
The output of the EXPLAIN statement points out the actual
catalog which is handling the SELECT query over the table example_table.
The table redirection functionality works also when using fully qualified names for the tables:
EXPLAIN SELECT * FROM example.example_schema.example_table;
Query Plan
-------------------------------------------------------------------------
Fragment 0 [SOURCE]
...
Output[columnNames = [...]]
│ ...
└─ TableScan[table = another_catalog:example_schema:example_table]
...
Trino offers table redirection support for the following operations:
Table read operations
Table write operations
Table management operations
Trino does not offer view redirection support.
The connector supports redirection from Hive tables to Iceberg and Delta Lake tables with the following catalog configuration properties:
hive.iceberg-catalog-namefor redirecting the query to Iceberg connectorhive.delta-lake-catalog-namefor redirecting the query to Delta Lake connector
Performance tuning configuration properties#
The following table describes performance tuning properties for the Hive connector.
Warning
Performance tuning configuration properties are considered expert-level features. Altering these properties from their default values is likely to cause instability and performance degradation.
Property name |
Description |
Default value |
|---|---|---|
|
The target number of buffered splits for each table scan in a query, before the scheduler tries to pause. |
|
|
The maximum number of splits generated per second per table scan. This can be used to reduce the load on the storage system. By default, there is no limit, which results in Trino maximizing the parallelization of data access. |
|
|
For each table scan, the coordinator first assigns file sections of up
to |
|
|
The size of a single file section assigned to a worker until
|
|
|
The largest size of a single file section assigned to a worker. Smaller splits result in more parallelism and thus can decrease latency, but also have more overhead and increase load on the system. |
|
File formats#
The following file types and formats are supported for the Hive connector:
ORC
Parquet
Avro
RCText (RCFile using
ColumnarSerDe)RCBinary (RCFile using
LazyBinaryColumnarSerDe)SequenceFile
JSON (using
org.apache.hive.hcatalog.data.JsonSerDe)CSV (using
org.apache.hadoop.hive.serde2.OpenCSVSerde)TextFile
ORC format configuration properties#
The following properties are used to configure the read and write operations with ORC files performed by the Hive connector.
Property Name |
Description |
Default |
|---|---|---|
|
Sets the default time zone for legacy ORC files that did not declare a time zone. |
JVM default |
|
Access ORC columns by name. By default, columns in ORC files are
accessed by their ordinal position in the Hive table definition. The
equivalent catalog session property is |
|
|
Enable bloom filters for predicate pushdown. |
|
Parquet format configuration properties#
The following properties are used to configure the read and write operations with Parquet files performed by the Hive connector.
Property Name |
Description |
Default |
|---|---|---|
|
Adjusts timestamp values to a specific time zone. For Hive 3.1+, set this to UTC. |
JVM default |
|
Access Parquet columns by name by default. Set this property to
|
|
|
Whether batched column readers are used when reading Parquet files
for improved performance. Set this property to |
|
|
Whether the optimized writer is used when writing Parquet files.
Set this property to |
|
|
Percentage of parquet files to validate after write by re-reading the whole file
when |
|
|
Maximum page size for the Parquet writer. |
|
|
Maximum row group size for the Parquet writer. |
|
|
Maximum number of rows processed by the parquet writer in a batch. |
|
|
Whether bloom filters are used for predicate pushdown when reading
Parquet files. Set this property to |
|
|
Sets the maximum number of rows read in a batch. |
|
|
Whether batched column readers should be used when reading ARRAY, MAP
and ROW types from Parquet files for improved performance. Set this
property to |
|
