Help Center/ GaussDB/ Developer Guide(Distributed_2.x)/ Importing Data/ Importing Data in Parallel Using Foreign Tables/ Parallel Data Import
Updated on 2024-10-14 GMT+08:00
View PDF
Share

Parallel Data Import

The INSERT (Running the INSERT Statement to Insert Data) and COPY (Running the COPY FROM STDIN Statement to Import Data) statements can be used only for serially importing a small volume of data. To import a large volume of data to GaussDB, you can import data in parallel through a foreign table. To import foreign tables in parallel, you must enable the stream operator (controlled by the GUC parameter enable_stream_operator).

Overview

GDS imports data in parallel from the common file system of a server to GaussDB.

Source data files to import are specified based on the import policy and data formats set in a foreign table. Data is imported in parallel through multiple DNs from source data files to the database, which improves overall data import performance. Figure 1 shows an example.

  • The CN only plans data import tasks and delivers the tasks to DNs. Then the CN is released to process other tasks.
  • In this way, the computing capacities and bandwidths of all the DNs are fully leveraged to import data, improving import efficiency.
You can preprocess data (such as invalid character replacement and fault tolerance processing) by setting parameters in a foreign table. For details, see CREATE FOREIGN TABLE (for Import and Export).
Figure 1 Importing data in parallel

The concepts mentioned in this figure are as follows:

  • CN: coordinator node of GaussDB. After receiving import SQL requests from an application or client, the CN plans import tasks and delivers the tasks to DNs.
  • DN: data node of GaussDB. After receiving import tasks delivered by the CN, DNs import data from the source data file to the target table in the database through a foreign table.
  • Source data file: a file that stores data to import.
  • Data server: a server that stores source data files. For security purposes, it is recommended that the data server and GaussDB cluster be on the same intranet.
  • Foreign table: a table that stores information, such as the current location, format, destination location, encoding format, and data delimiter of a source data file. It is used to associate source data files with the target table.
  • Target table: a table in the database. It can be a row-store table or column-store table. Data in the source data files will be imported to this table.

Loading Policies

To fully use computing resources for parallel import, import tasks are pushed down to DNs. The CN delivers tasks but does not import data. In this case, user data managers, correct and unique data allocation to DNs, and the uniqueness of data imported into databases must be first considered. GaussDB provides the following import policies:
  • Normal: GDS is used to load source data into DNs. This policy is used for importing data to a cluster from hosts outside the cluster.
  • Shared: The network file system (NFS) is used to load source data into DNs. After you configure NFS on a data server, mount the data server to the same directory of each DN. This policy is used for importing data to a cluster from hosts outside the cluster. In this policy, the CN scans all data files during task planning and evenly allocates source data files to each DN.
  • Private: Users upload source data files to each DN. Before uploading data files, create a directory named after the DN name on each DN. Each DN searches for unloaded data files in its corresponding directory and loads them, until all data has been loaded.

As shown in Table 1, the Normal policy is recommended because it is scalable, easy to prepare, and has no limit on the size of a single row of data to be imported. The foreign tables in Private and Shared modes require the initial user permissions or the O&M administrator permissions in operation_mode. This section describes how to use GDS to import data in parallel. For details on the other policies, see Example 2: Importing Data in Shared Mode and Example 3: Importing Data in Private Mode.

Table 1 Import policies

Import Policy

Preparation for Data Import

Data Format

Normal

Deploy GDS on data servers. Multi-GDS concurrent import is supported.

CSV, TEXT, and FIXED

Shared

Configure NFS on data servers and mount the data servers to the same directory of each DN. The mount directory on each DN must be the same.

TEXT

The size of data in a single row must be less than 1 GB.

Private

Evenly allocate source data files to DN servers.

Data files in each folder named after a DN name under the same directory of hosts should not be duplicate.

CSV, TEXT, and FIXED

The size of data in a single row must be less than 1 GB.

Parallel Import Using GDS

  • If a large volume of data is stored on multiple servers, deploy, configure, and start GDS on each server. Then, data on all the servers can be imported in parallel, as shown in Figure 2.
    Figure 2 Parallel import from multiple data servers

    The number of GDS processes cannot exceed that of DNs. If the number of connections exceeds the maximum, one DN may be connected to multiple GDS processes. As a result, some GDS processes may run abnormally, and the error message "Session doesn't exist" may be displayed. If the number of GDSs involved in an import is greater than the number of DNs, a warning "It is recommended that the number of GDS should not be greater than the number of datanode" is returned when the import starts.

  • If data is stored on one data server, and both GaussDB and the data server have available I/O resources, you can use GDS for multi-thread parallel import.

    GDS determines the number of threads based on the number of parallel import transactions. Even if multi-thread import is configured before GDS startup, the import of a single transaction will not be accelerated. By default, an INSERT statement is an import transaction.

    Multi-thread parallel import enables you to:

    • Make full use of resources and improve the parallel import efficiency when you import multiple tables to the database.
    • Speed up the import of a table with a large volume of data.

      Table data is split into multiple data files, and multi-thread parallel import is implemented by importing data using multiple foreign tables at the same time. Ensure that a data file is read only by one foreign table.

Import Process

Figure 3 Parallel import process
Table 2 Process description

Process

Description

Prepare source ata

Prepare source data files to import to the database and upload the files to the data server.

For details, see Preparing Source Data.

Start GDS

Install, configure, and enable GDS on the data server.

For details, see Installing, Configuring, and Starting GDS.

Create a foreign table

A foreign table is used to identify data in a source data file. Foreign tables store the location, file format, encoding format, and delimiters information about the data source files.

For details, see Creating a GDS Foreign Table.

Import data

After creating the foreign table, execute the INSERT statement to efficiently import data to the target table. For details, see Importing Data.

Handle import errors

If there are errors during parallel data import, handle errors based on the error information to ensure data integrity.

For details, see Handling Import Errors.

Improve query efficiency

After data is imported, run the ANALYZE statement to generate table statistics. The ANALYZE statement stores the statistics in the system catalog PG_STATISTIC. The statistics data is useful when you run the planner, which provides you with an efficient query execution plan.

For details, see Analyzing Tables.

Stop GDS

After data import is complete, log in to each data server and stop GDS.

For details, see Stopping GDS.