PGLoader

pgloader is a data loading tool for PostgreSQL, using the COPY command.

Its main avantage over just using COPY or \copy and over using a Foreign Data Wrapper is the transaction behaviour, where pgloader will keep a separate file of rejected data and continue trying to copy good data in your database.

The default PostgreSQL behaviour is transactional, which means that any erroneous line in the input data (file or remote database) will stop the bulk load for the whole table.

pgloader also implements data reformating, the main example of that being a transformation from MySQL dates 0000-00-00 and 0000-00-00 00:00:00 to PostgreSQL NULL value (because our calendar never had a year zero).

INSTALL

pgloader is now a Common Lisp program, tested using the SBCL and CCL implementation with Quicklisp.

apt-get install sbcl
apt-get install libmysqlclient-dev libsqlite3-dev
wget http://beta.quicklisp.org/quicklisp.lisp
sbcl --load quicklisp.lisp
* (quicklisp-quickstart:install)
* (ql:add-to-init-file)

The current version of the code depends on a recent version of Postmodern not found in Quicklisp yet at the time of this writing:

cd ~/quicklisp/local-projects/
git clone https://github.com/marijnh/Postmodern.git
git clone -b empty-strings-and-nil  https://github.com/dimitri/cl-csv.git
git clone http://git.tapoueh.org/git/pgloader.git

Now you can use the #! script or build a self-contained binary executable file, as shown below. You might have to modify it the pgloader.lisp script because it's now hard coded to use /usr/local/bin/sbcl and you probably want to change that part then:

./pgloader.lisp --help

Each time you run the pgloader command line, it will check that all its dependencies are installed and compiled and if that's not the case fetch them from the internet and prepare them (thanks to Quicklisp). So please be patient while that happens and make sure we can actually connect and download the dependencies.

Compile into a self-contained binary file

First, make sure you have downloaded all the required Common Lisp dependencies that pgloader uses, and install the buildapp application:

$ sbcl
* (ql:quickload "pgloader")
* (ql:quickload "buildapp")
* (buildapp:build-buildapp "./buildapp")

If you just installed SBCL and Quicklisp to use pgloader, that command should do it:

./buildapp --logfile /tmp/build.log        \
		   --asdf-tree ~/quicklisp/dists   \
		   --load-system pgloader          \
		   --entry pgloader:main           \
		   --dynamic-space-size 4096       \
		   --output pgloader.exe

You can also use the option --compress-core if your platform supports it, so has to reduce the size of the generated binary.

When you're a Common Lisp developper or otherwise already using Quicklisp with some local-projects and a local source registry setup for asdf, use a command line like this:

./buildapp --logfile /tmp/build.log                \
           --asdf-tree ~/quicklisp/local-projects  \
		   --manifest-file ./manifest.ql           \
		   --asdf-tree ~/quicklisp/dists           \
		   --load-system pgloader                  \
		   --entry pgloader:main                   \
		   --dynamic-space-size 4096               \
		   --output pgloader.exe

That command requires a manifest.ql file that you can obtain with the lisp command:

(ql:write-asdf-manifest-file "path/to/manifest.ql")

Usage

Give as many command files that you need to pgloader:

./pgloader.lisp <file.load>

See the documentation file pgloader.1.md for details. You can compile that file into a manual page or an HTML page thanks to the pandoc application:

$ apt-get install pandoc
$ pandoc pgloader.1.md -o pgloader.1
$ pandoc pgloader.1.md -o pgloader.html

TODO

Some notes about what I intend to be working on next.

binary distribution

• prepare an all-included binary for several platforms

internals & refactoring

• review pgloader.pgsql:reformat-row date-columns arguments
• review connection string handling for both PostgreSQL and MySQL
• provide a better toplevel API
• implement tests

command & control

• commands: LOAD and INI formats
• compat with SQL*Loader format
• see pgloader.1.md for details.

docs

• host a proper website for the tool, with use cases and a tutorial

error management

• error management with a local buffer (done)
• error reporting (done)
• add input line number to log file?

data input

• import directly from MySQL, file based export/import (done)
• import directly from MySQL streaming (done)
• general CSV and Flexible Text source formats
• fixed cols input data format
• compressed input (gzip, other algos)
• fetch data from S3

transformation and casts

• experiment with perfs and inlining the transformation functions
• add typemod expression to cast rules in the command language
• add per-column support for cast rules in the system

data output

• PostgreSQL COPY Text format output for any supported input

convenience

• automatic creation of schema (from MySQL schema, or from CSV header)
• pre-fetch some rows to guesstimate data types?

performances

• some more parallelizing options
• support for partitionning in pgloader itself

reformating

Data reformating is now going to have to happen in Common Lisp mostly, maybe offer some other languages (cl-awk etc).

• raw reformating, before rows are split
• per column reformating
  • date (zero dates)
  • integer and "" that should be NULL
• user-defined columns (constants, functions of other rows)
• column re-ordering

Have a try at something approaching:

WITH data AS (
	COPY FROM ...
	RETURNING x, y
)
SELECT foo(x), bar(y)
  FROM data
 WHERE ...

A part of that needs to happen client-side, another part server-side, and the grammar has to make it clear what happens where. Maybe add a WHERE clause to the COPY or LOAD grammar for the client.

UI

• add a web controler with pretty monitoring
• launch new jobs from the web controler

crazy ideas

• MySQL replication, reading from the binlog directly
• plproxy (re-)sharding support
• partitioning support
• remote archiving support (with (delete returning *) insert into)