Specifying Where to Load Data From

Each COPY statement requires either a FROM clause to indicate the location of the file or files being loaded or a SOURCE clause when using a user-defined source. For more about the SOURCE clause, see COPY Parameters. This section covers use of the FROM clause.

Loading from a Specific Path

Use the path-to-data argument to indicate the location of the file to load. You can load data from the following locations:

  • The local file system.
  • NFS, through a mount point on the local file system.
  • HDFS, using a URL of the form [[s]web]hdfs://[nameservice]/path. For more information about HDFS URLs and configuration, see HDFS File System.

    If a data file you want to read resides on an HDFS cluster that uses Kerberos authentication, Vertica uses the current user's principal, session doAs user, or session delegation token. See Accessing Kerberized HDFS Data for more information about these options.

  • S3, Google Cloud Storage, or Azure Blob Storage, for data in text, delimited, Parquet, and ORC formats only. For more information about reading from these file systems, see S3 Object Store, Google Cloud Storage (GCS) Object Store, and Azure Blob Storage Object Store.

If the path is a URL, you must use URL encoding for the '%' character. Otherwise, URL encoding ('%NN' where NN is a two-digit hexadecimal number) is permitted but not required.

When copying from the local file system, the COPY statement expects to find files in the same location on every node that participates in the query. If you are using NFS, then you can create an NFS mount point on each node. Doing so allows all database nodes to participate in the load for better performance without requiring files to be copied to all nodes.

Treat NFS mount points as local files in paths:

=> COPY sales FROM '/mount/sales.dat' ON ANY NODE;

You can specify more than one path in the same COPY statement, as in the following example.

=> COPY myTable FROM 'webhdfs:///data/sales/01/*.dat', 'webhdfs:///data/sales/02/*.dat', 

For files in HDFS, you can specify a name service (hadoopNS). In this example, the COPY statement is part of the definition of an external table:

	AS COPY FROM 'webhdfs://hadoopNS/data/users.csv';

If path-to-data resolves to a storage location on a local file system, and the user invoking COPY is not a superuser, these permissions are required:

  • The storage location must have been created with the USER option (see CREATE LOCATION).
  • The user must already have been granted READ access to the storage location where the file or files exist, as described in GRANT (Storage Location)

Vertica prevents symbolic links from allowing unauthorized access.

Loading with Wildcards (glob)

You can invoke COPY for a large number of files in a shared directory with a single statement such as:

=> COPY myTable FROM '/data/manyfiles/*.dat' ON ANY NODE;

The glob (*) must indicate a set of files, not directories. The following statement fails if /data/manyfiles contains any subdirectories:

=> COPY myTable FROM '/data/manyfiles/*' ON ANY NODE;

Using a wildcard with the ON ANY NODE clause expands the file list on the initiator node. This command then distributes the individual files among all nodes, so that the COPY workload is evenly distributed across the entire cluster.

ON ANY NODE is the default for all file systems other than Linux, as in the following example:

=> COPY myTable FROM 'webhdfs:///data/manyfiles/*';

You can also distribute a file set across a subset of nodes, which you might do to balance concurrent loads. For example, this command distributes the loading of individual files among the three named nodes:

=> COPY myTable FROM '/mydirectory/ofmanyfiles/*.dat' 
   ON (v_vmart_node0001, v_vmart_node0002, v_vmart_node0003);

Distributing file loads across nodes depends on two configuration parameters, EnableApportionLoad and EnableApportionFileLoad. Both are enabled by default. See General Parameters for more information about these parameters.

Loading from a Vertica Client

Use COPY LOCAL to load files on a client system to the Vertica database. For example, to copy a GZIP file from your local client, use a command such as this:

=> COPY store.store_dimension FROM LOCAL '/usr/files/my_data/input_file' GZIP;

You can use a comma-separated list to load multiple files of the same compression type. COPY LOCAL then concatenates the files into a single file, so you cannot combine files with different compression types in the list. When listing multiple files, be sure to specify the type of every input file, such as BZIP, as shown:

=>COPY simple_table FROM LOCAL 'input_file.bz' BZIP, 'input_file.bz' BZIP;

You can load data from a local client from STDIN, as follows:

=> COPY simple_table FROM LOCAL STDIN;

Loading from Kafka or Spark

For information about streaming data from Kafka, see Integrating with Apache Kafka .

For information about using Vertica with Spark data, see Integrating with Apache Spark.

Loading Data from an IDOL CFS Client

The IDOL Connector Framework Server (CFS) VerticaIndexer feature lets CFS clients connect to your Vertica database using ODBC. After it is connected, CFS uses COPY...FROM LOCAL statements to load IDOL document metadata into an existing flex table. For more information, see the Using Flex Tables for IDOL Data section in Using Flex Tables.