Welcome to pgloader’s documentation!

The pgloader project is an Open Source Software project. The development happens at https://github.com/dimitri/pgloader and is public: everyone is welcome to participate by opening issues, pull requests, giving feedback, etc.

pgloader loads data from various sources into PostgreSQL. It can transform the data it reads on the fly and submit raw SQL before and after the loading. It uses the COPY PostgreSQL protocol to stream the data into the server, and manages errors by filling a pair of reject.dat and reject.log files.

Thanks to being able to load data directly from a database source, pgloader also supports from migrations from other productions to PostgreSQL. In this mode of operations, pgloader handles both the schema and data parts of the migration, in a single unmanned command, allowing to implement Continuous Migration.

Features Overview

pgloader has two modes of operation: loading from files, migrating databases. In both cases, pgloader uses the PostgreSQL COPY protocol which implements a streaming to send data in a very efficient way.

Loading file content in PostgreSQL

When loading from files, pgloader implements the following features:

Many source formats supported

Support for a wide variety of file based formats are included in pgloader: the CSV family, fixed columns formats, dBase files (db3), and IBM IXF files.

The SQLite database engine is accounted for in the next section: pgloader considers SQLite as a database source and implements schema discovery from SQLite catalogs.

On the fly data transformation

Often enough the data as read from a CSV file (or another format) needs some tweaking and clean-up before being sent to PostgreSQL.

For instance in the geolite example we can see that integer values are being rewritten as IP address ranges, allowing to target an ip4r column directly.

Full Field projections

pgloader supports loading data into less fields than found on file, or more, doing some computation on the data read before sending it to PostgreSQL.

Reading files from an archive

Archive formats zip, tar, and gzip are supported by pgloader: the archive is extracted in a temporary directly and expanded files are then loaded.

HTTP(S) support

pgloader knows how to download a source file or a source archive using HTTP directly. It might be better to use curl -O- http://... | pgloader and read the data from standard input, then allowing for streaming of the data from its source down to PostgreSQL.

Target schema discovery

When loading in an existing table, pgloader takes into account the existing columns and may automatically guess the CSV format for you.

On error stop / On error resume next

In some cases the source data is so damaged as to be impossible to migrate in full, and when loading from a file then the default for pgloader is to use on error resume next option, where the rows rejected by PostgreSQL are saved away and the migration continues with the other rows.

In other cases loading only a part of the input data might not be a great idea, and in such cases it’s possible to use the on error stop option.

Pre/Post SQL commands

This feature allows pgloader commands to include SQL commands to run before and after loading a file. It might be about creating a table first, then loading the data into it, and then doing more processing on-top of the data (implementing an ELT pipeline then), or creating specific indexes as soon as the data has been made ready.

One-command migration to PostgreSQL

When migrating a full database in a single command, pgloader implements the following features:

One-command migration

The whole migration is started with a single command line and then runs unattended. pgloader is meant to be integrated in a fully automated tooling that you can repeat as many times as needed.

Schema discovery

The source database is introspected using its SQL catalogs to get the list of tables, attributes (with data types, default values, not null constraints, etc), primary key constraints, foreign key constraints, indexes, comments, etc. This feeds an internal database catalog of all the objects to migrate from the source database to the target database.

User defined casting rules

Some source database have ideas about their data types that might not be compatible with PostgreSQL implementaion of equivalent data types.

For instance, SQLite since version 3 has a Dynamic Type System which of course isn’t compatible with the idea of a Relation. Or MySQL accepts datetime for year zero, which doesn’t exists in our calendar, and doesn’t have a boolean data type.

When migrating from another source database technology to PostgreSQL, data type casting choices must be made. pgloader implements solid defaults that you can rely upon, and a facility for user defined data type casting rules for specific cases. The idea is to allow users to specify the how the migration should be done, in order for it to be repeatable and included in a Continuous Migration process.

On the fly data transformations

The user defined casting rules come with on the fly rewrite of the data. For instance zero dates (it’s not just the year, MySQL accepts 0000-00-00 as a valid datetime) are rewritten to NULL values by default.

Partial Migrations

It is possible to include only a partial list of the source database tables in the migration, or to exclude some of the tables on the source database.

Schema only, Data only

This is the ORM compatibility feature of pgloader, where it is possible to create the schema using your ORM and then have pgloader migrate the data targeting this already created schema.

When doing this, it is possible for pgloader to reindex the target schema: before loading the data from the source database into PostgreSQL using COPY, pgloader DROPs the indexes and constraints, and reinstalls the exact same definitions of them once the data has been loaded.

The reason for operating that way is of course data load performance.

Repeatable (DROP+CREATE)

By default, pgloader issues DROP statements in the target PostgreSQL database before issuing any CREATE statement, so that you can repeat the migration as many times as necessary until migration specifications and rules are bug free.

The schedule the data migration to run every night (or even more often!) for the whole duration of the code migration project. See the Continuous Migration methodology for more details about the approach.

On error stop / On error resume next

The default behavior of pgloader when migrating from a database is on error stop. The idea is to let the user fix either the migration specifications or the source data, and run the process again, until it works.

In some cases the source data is so damaged as to be impossible to migrate in full, and it might be necessary to then resort to the on error resume next option, where the rows rejected by PostgreSQL are saved away and the migration continues with the other rows.

Pre/Post SQL commands, Post-Schema SQL commands

While pgloader takes care of rewriting the schema to PostgreSQL expectations, and even provides user-defined data type casting rules support to that end, sometimes it is necessary to add some specific SQL commands around the migration. It’s of course supported right from pgloader itself, without having to script around it.

Online ALTER schema

At times migrating to PostgreSQL is also a good opportunity to review and fix bad decisions that were made in the past, or simply that are not relevant to PostgreSQL.

The pgloader command syntax allows to ALTER pgloader’s internal representation of the target catalogs so that the target schema can be created a little different from the source one. Changes supported include target a different schema or table name.

Materialized Views, or schema rewrite on-the-fly

In some cases the schema rewriting goes deeper than just renaming the SQL objects to being a full normalization exercise. Because PostgreSQL is great at running a normalized schema in production under most workloads.

pgloader implements full flexibility in on-the-fly schema rewriting, by making it possible to migrate from a view definition. The view attribute list becomes a table definition in PostgreSQL, and the data is fetched by querying the view on the source system.

A SQL view allows to implement both content filtering at the column level using the SELECT projection clause, and at the row level using the WHERE restriction clause. And backfilling from reference tables thanks to JOINs.

Distribute to Citus

When migrating from PostgreSQL to Citus, a important part of the process consists of adjusting the schema to the distribution key. Read Preparing Tables and Ingesting Data in the Citus documentation for a complete example showing how to do that.

When using pgloader it’s possible to specify the distribution keys and reference tables and let pgloader take care of adjusting the table, indexes, primary keys and foreign key definitions all by itself.

Encoding Overrides

MySQL doesn’t actually enforce the encoding of the data in the database to match the encoding known in the metadata, defined at the database, table, or attribute level. Sometimes, it’s necessary to override the metadata in order to make sense of the text, and pgloader makes it easy to do so.

Continuous Migration

pgloader is meant to migrate a whole database in a single command line and without any manual intervention. The goal is to be able to setup a Continuous Integration environment as described in the Project Methodology document of the MySQL to PostgreSQL webpage.

  1. Setup your target PostgreSQL Architecture

  2. Fork a Continuous Integration environment that uses PostgreSQL

  3. Migrate the data over and over again every night, from production

  4. As soon as the CI is all green using PostgreSQL, schedule the D-Day

  5. Migrate without suprise and enjoy!

In order to be able to follow this great methodology, you need tooling to implement the third step in a fully automated way. That’s pgloader.

Indices and tables