Create External Table

Creates a new external table, which is a local database object whose source data is located externally to the database. The source data can be located either in KiFS; on the cluster, accessible to the database; or remotely, accessible via a pre-defined external data source.

The external table can have its structure defined explicitly, via input parameter create_table_options, which contains many of the options from Create Table; or defined implicitly, inferred from the source data.

Input Parameter Description

Name Type Description
table_name string Name of the table to be created, in [schema_name.]table_name format, using standard name resolution rules and meeting table naming criteria.
filepaths array of strings A list of file paths from which data will be sourced; For paths in KiFS, use the uri prefix of kifs:// followed by the path to a file or directory. File matching by prefix is supported, e.g. kifs://dir/file would match dir/file_1 and dir/file_2. When prefix matching is used, the path must start with a full, valid KiFS directory name. If an external data source is specified in datasource_name, these file paths must resolve to accessible files at that data source location. Prefix matching is supported. If the data source is hdfs, prefixes must be aligned with directories, i.e. partial file names will not match. If no data source is specified, the files are assumed to be local to the database and must all be accessible to the gpudb user, residing on the path (or relative to the path) specified by the external files directory in the Kinetica configuration file. Wildcards (*) can be used to specify a group of files. Prefix matching is supported, the prefixes must be aligned with directories. If the first path ends in .tsv, the text delimiter will be defaulted to a tab character. If the first path ends in .psv, the text delimiter will be defaulted to a pipe character (|).
modify_columns map of string to maps of string to strings Not implemented yet. The default value is an empty map ( {} ).
create_table_options map of string to strings

Options from Create Table , allowing the structure of the table to be defined independently of the data source. The default value is an empty map ( {} ).

Supported Parameters (keys) Parameter Description
type_id ID of a currently registered type. The default value is ''.
no_error_if_exists

If true, prevents an error from occurring if the table already exists and is of the given type. If a table with the same name but a different type exists, it is still an error. The default value is false. The supported values are:

  • true
  • false
is_replicated

Affects the distribution scheme for the table's data. If true and the given table has no explicit shard key defined, the table will be replicated. If false, the table will be sharded according to the shard key specified in the given type_id, or randomly sharded, if no shard key is specified. Note that a type containing a shard key cannot be used to create a replicated table. The default value is false. The supported values are:

  • true
  • false
foreign_keys Semicolon-separated list of foreign keys, of the format '(source_column_name [, ...]) references target_table_name(primary_key_column_name [, ...]) [as foreign_key_name]'.
foreign_shard_key Foreign shard key of the format 'source_column references shard_by_column from target_table(primary_key_column)'.
partition_type

Partitioning scheme to use.

Supported Values Description
RANGE Use range partitioning.
INTERVAL Use interval partitioning.
LIST Use list partitioning.
HASH Use hash partitioning.
SERIES Use series partitioning.
partition_keys Comma-separated list of partition keys, which are the columns or column expressions by which records will be assigned to partitions defined by partition_definitions.
partition_definitions Comma-separated list of partition definitions, whose format depends on the choice of partition_type. See range partitioning, interval partitioning, list partitioning, or hash partitioning for example formats.
is_automatic_partition

If true, a new partition will be created for values which don't fall into an existing partition. Currently only supported for list partitions. The default value is false. The supported values are:

  • true
  • false
ttl Sets the TTL of the table specified in input parameter table_name.
chunk_size Indicates the number of records per chunk to be used for this table.
is_result_table

Indicates whether the table is a memory-only table. A result table cannot contain columns with store_only or text_search data-handling or that are non-charN strings, and it will not be retained if the server is restarted. The default value is false. The supported values are:

  • true
  • false
strategy_definition The tier strategy for the table and its columns.
options map of string to strings

Optional parameters. The default value is an empty map ( {} ).

Supported Parameters (keys) Parameter Description
avro_header_bytes Optional number of bytes to skip when reading an avro record.
avro_num_records Optional number of avro records, if data includes only records.
avro_schema Optional string representing avro schema, for insert records in avro format, that does not include is schema.
avro_schemaless

When user provides 'avro_schema', avro data is assumed to be schemaless, unless specified. Default is 'true' when given avro_schema. Igonred when avro_schema is not given. The supported values are:

  • true
  • false
bad_record_table_name Optional name of a table to which records that were rejected are written. The bad-record-table has the following columns: line_number (long), line_rejected (string), error_message (string).
bad_record_table_limit A positive integer indicating the maximum number of records that can be written to the bad-record-table. Default value is 10000
bad_record_table_limit_per_input For subscriptions: A positive integer indicating the maximum number of records that can be written to the bad-record-table per file/payload. Default value will be 'bad_record_table_limit' and total size of the table per rank is limited to 'bad_record_table_limit'
batch_size Internal tuning parameter--number of records per batch when inserting data.
column_formats For each target column specified, applies the column-property-bound format to the source data loaded into that column. Each column format will contain a mapping of one or more of its column properties to an appropriate format for each property. Currently supported column properties include date, time, & datetime. The parameter value must be formatted as a JSON string of maps of column names to maps of column properties to their corresponding column formats, e.g., '{ "order_date" : { "date" : "%Y.%m.%d" }, "order_time" : { "time" : "%H:%M:%S" } }'. See default_column_formats for valid format syntax.
columns_to_load Specifies a comma-delimited list of columns from the source data to load. If more than one file is being loaded, this list applies to all files. Column numbers can be specified discretely or as a range. For example, a value of '5,7,1..3' will insert values from the fifth column in the source data into the first column in the target table, from the seventh column in the source data into the second column in the target table, and from the first through third columns in the source data into the third through fifth columns in the target table. If the source data contains a header, column names matching the file header names may be provided instead of column numbers. If the target table doesn't exist, the table will be created with the columns in this order. If the target table does exist with columns in a different order than the source data, this list can be used to match the order of the target table. For example, a value of 'C, B, A' will create a three column table with column C, followed by column B, followed by column A; or will insert those fields in that order into a table created with columns in that order. If the target table exists, the column names must match the source data field names for a name-mapping to be successful. Mutually exclusive with columns_to_skip.
columns_to_skip Specifies a comma-delimited list of columns from the source data to skip. Mutually exclusive with columns_to_load.
compression_type

Optional: compression type The default value is auto.

Supported Values Description
none Uncompressed
auto Default. Auto detect compression type
gzip gzip file compression.
bzip2 bzip2 file compression.
datasource_name Name of an existing external data source from which data file(s) specified in input parameter filepaths will be loaded
default_column_formats Specifies the default format to be applied to source data loaded into columns with the corresponding column property. Currently supported column properties include date, time, & datetime. This default column-property-bound format can be overridden by specifying a column property & format for a given target column in column_formats. For each specified annotation, the format will apply to all columns with that annotation unless a custom column_formats for that annotation is specified. The parameter value must be formatted as a JSON string that is a map of column properties to their respective column formats, e.g., '{ "date" : "%Y.%m.%d", "time" : "%H:%M:%S" }'. Column formats are specified as a string of control characters and plain text. The supported control characters are 'Y', 'm', 'd', 'H', 'M', 'S', and 's', which follow the Linux 'strptime()' specification, as well as 's', which specifies seconds and fractional seconds (though the fractional component will be truncated past milliseconds). Formats for the 'date' annotation must include the 'Y', 'm', and 'd' control characters. Formats for the 'time' annotation must include the 'H', 'M', and either 'S' or 's' (but not both) control characters. Formats for the 'datetime' annotation meet both the 'date' and 'time' control character requirements. For example, '{"datetime" : "%m/%d/%Y %H:%M:%S" }' would be used to interpret text as "05/04/2000 12:12:11"
error_handling

Specifies how errors should be handled upon insertion. The default value is abort.

Supported Values Description
permissive Records with missing columns are populated with nulls if possible; otherwise, the malformed records are skipped.
ignore_bad_records Malformed records are skipped.
abort Stops current insertion and aborts entire operation when an error is encountered. Primary key collisions are considered abortable errors in this mode.
external_table_type

Specifies whether the external table holds a local copy of the external data. The default value is materialized.

Supported Values Description
materialized Loads a copy of the external data into the database, refreshed on demand
logical External data will not be loaded into the database; the data will be retrieved from the source upon servicing each query against the external table
file_type

Specifies the type of the file(s) whose records will be inserted. The default value is delimited_text.

Supported Values Description
avro Avro file format
delimited_text Delimited text file format; e.g., CSV, TSV, PSV, etc.
gdb Esri/GDB file format
json Json file format
parquet Apache Parquet file format
shapefile ShapeFile file format
gdal_configuration_options Comma separated list of gdal conf options, for the specific requets: key=value. The default value is ''.
ignore_existing_pk

Specifies the record collision error-suppression policy for inserting into a table with a primary key, only used when not in upsert mode (upsert mode is disabled when update_on_existing_pk is false). If set to true, any record being inserted that is rejected for having primary key values that match those of an existing table record will be ignored with no error generated. If false, the rejection of any record for having primary key values matching an existing record will result in an error being reported, as determined by error_handling. If the specified table does not have a primary key or if upsert mode is in effect (update_on_existing_pk is true), then this option has no effect. The default value is false.

Supported Values Description
true Ignore new records whose primary key values collide with those of existing records
false Treat as errors any new records whose primary key values collide with those of existing records
ingestion_mode

Whether to do a full load, dry run, or perform a type inference on the source data. The default value is full.

Supported Values Description
full Run a type inference on the source data (if needed) and ingest
dry_run Does not load data, but walks through the source data and determines the number of valid records, taking into account the current mode of error_handling.
type_inference_only Infer the type of the source data and return, without ingesting any data. The inferred type is returned in the response.
jdbc_fetch_size The JDBC fetch size, which determines how many rows to fetch per round trip.
kafka_group_id The group id to be used consuming data from a kakfa topic (valid only for kafka datasource subscriptions).
kafka_offset_reset_policy

Policy to determine whether the data consumption starts either at earliest offset or latest offset. The default value is earliest. The supported values are:

  • earliest
  • latest
kafka_subscription_cancel_after Sets the subscription lifespan (in minutes). Expired subscription will be cancelled automatically.
layer Optional: geo files layer(s) name(s): comma separated. The default value is ''.
loading_mode

Scheme for distributing the extraction and loading of data from the source data file(s). This option applies only when loading files that are local to the database The default value is head.

Supported Values Description
head The head node loads all data. All files must be available to the head node.
distributed_shared The head node coordinates loading data by worker processes across all nodes from shared files available to all workers. NOTE: Instead of existing on a shared source, the files can be duplicated on a source local to each host to improve performance, though the files must appear as the same data set from the perspective of all hosts performing the load.
distributed_local A single worker process on each node loads all files that are available to it. This option works best when each worker loads files from its own file system, to maximize performance. In order to avoid data duplication, either each worker performing the load needs to have visibility to a set of files unique to it (no file is visible to more than one node) or the target table needs to have a primary key (which will allow the worker to automatically deduplicate data). NOTE: If the target table doesn't exist, the table structure will be determined by the head node. If the head node has no files local to it, it will be unable to determine the structure and the request will fail. If the head node is configured to have no worker processes, no data strictly accessible to the head node will be loaded.
local_time_offset For Avro local timestamp columns
max_records_to_load Limit the number of records to load in this request: If this number is larger than a batch_size, then the number of records loaded will be limited to the next whole number of batch_size (per working thread). The default value is ''.
num_tasks_per_rank Optional: number of tasks for reading file per rank. Default will be external_file_reader_num_tasks
poll_interval If true, the number of seconds between attempts to load external files into the table. If zero, polling will be continuous as long as data is found. If no data is found, the interval will steadily increase to a maximum of 60 seconds.
primary_keys Optional: comma separated list of column names, to set as primary keys, when not specified in the type. The default value is ''.
refresh_method

Method by which the table can be refreshed from its source data. The default value is manual.

Supported Values Description
manual Refresh only occurs when manually requested by invoking the refresh action of Alter Table on this table.
on_start Refresh table on database startup and when manually requested by invoking the refresh action of Alter Table on this table.
schema_registry_schema_id  
schema_registry_schema_name  
schema_registry_schema_version  
shard_keys Optional: comma separated list of column names, to set as primary keys, when not specified in the type. The default value is ''.
skip_lines Skip number of lines from begining of file.
subscribe

Continuously poll the data source to check for new data and load it into the table. The default value is false. The supported values are:

  • true
  • false
table_insert_mode

Optional: table_insert_mode. When inserting records from multiple files: if table_per_file then insert from each file into a new table. Currently supported only for shapefiles. The default value is single. The supported values are:

  • single
  • table_per_file
text_comment_string Specifies the character string that should be interpreted as a comment line prefix in the source data. All lines in the data starting with the provided string are ignored. For delimited_text file_type only. The default value is '#'.
text_delimiter Specifies the character delimiting field values in the source data and field names in the header (if present). For delimited_text file_type only. The default value is ','.
text_escape_character Specifies the character that is used to escape other characters in the source data. An 'a', 'b', 'f', 'n', 'r', 't', or 'v' preceded by an escape character will be interpreted as the ASCII bell, backspace, form feed, line feed, carriage return, horizontal tab, & vertical tab, respectively. For example, the escape character followed by an 'n' will be interpreted as a newline within a field value. The escape character can also be used to escape the quoting character, and will be treated as an escape character whether it is within a quoted field value or not. For delimited_text file_type only.
text_has_header

Indicates whether the source data contains a header row. For delimited_text file_type only. The default value is true. The supported values are:

  • true
  • false
text_header_property_delimiter Specifies the delimiter for column properties in the header row (if present). Cannot be set to same value as text_delimiter. For delimited_text file_type only. The default value is '|'.
text_null_string Specifies the character string that should be interpreted as a null value in the source data. For delimited_text file_type only. The default value is '\N'.
text_quote_character Specifies the character that should be interpreted as a field value quoting character in the source data. The character must appear at beginning and end of field value to take effect. Delimiters within quoted fields are treated as literals and not delimiters. Within a quoted field, two consecutive quote characters will be interpreted as a single literal quote character, effectively escaping it. To not have a quote character, specify an empty string. For delimited_text file_type only. The default value is '"'.
text_search_columns Add 'text_search' property to internally inferenced string columns. Comma seperated list of column names or '*' for all columns. To add text_search property only to string columns of minimum size, set also the option 'text_search_min_column_length'
text_search_min_column_length Set minimum column size. Used only when 'text_search_columns' has a value.
truncate_strings

If set to true, truncate string values that are longer than the column's type size. The default value is false. The supported values are:

  • true
  • false
truncate_table

If set to true, truncates the table specified by input parameter table_name prior to loading the file(s). The default value is false. The supported values are:

  • true
  • false
type_inference_mode

optimize type inference for: The default value is speed.

Supported Values Description
accuracy Scans data to get exactly-typed & sized columns for all data scanned.
speed Scans data and picks the widest possible column types so that 'all' values will fit with minimum data scanned
remote_query Remote SQL query from which data will be sourced
remote_query_filter_column Name of column to be used for splitting the query into multiple sub-queries using the data distribution of given column. The default value is ''.
remote_query_increasing_column Column on subscribed remote query result that will increase for new records (e.g., TIMESTAMP). The default value is ''.
remote_query_partition_column Alias name for remote_query_filter_column. The default value is ''.
update_on_existing_pk

Specifies the record collision policy for inserting into a table with a primary key. If set to true, any existing table record with primary key values that match those of a record being inserted will be replaced by that new record (the new data will be "upserted"). If set to false, any existing table record with primary key values that match those of a record being inserted will remain unchanged, while the new record will be rejected and the error handled as determined by ignore_existing_pk & error_handling. If the specified table does not have a primary key, then this option has no effect. The default value is false.

Supported Values Description
true Upsert new records when primary keys match existing records
false Reject new records when primary keys match existing records

Output Parameter Description

Name Type Description
table_name string Value of input parameter table_name.
type_id string ID of the currently registered table structure type for this external table
type_definition string A JSON string describing the columns of the created external table
type_label string The user-defined description associated with the table's structure
type_properties map of string to arrays of strings A mapping of each external table column name to an array of column properties associated with that column
count_inserted long Number of records inserted into the external table.
count_skipped long Number of records skipped, when not running in abort error handling mode.
count_updated long [Not yet implemented] Number of records updated within the external table.
info map of string to strings Additional information.
files array of strings